AuroraMiddleware/SPIRV-Tools
1
0
Fork 0
mirror of https://github.com/KhronosGroup/SPIRV-Tools synced 2024-11-25 21:10:04 +00:00
Code Issues Packages Projects Releases Wiki Activity
a5e766b2b4
SPIRV-Tools/test/opt/aggressive_dead_code_elim_test.cpp

7783 lines
253 KiB
C++
Raw Normal View History

Add AggressiveDCEPass Create aggressive dead code elimination pass This pass eliminates unused code from functions. In addition, it detects and eliminates code which may have spurious uses but which do not contribute to the output of the function. The most common cause of such code sequences is summations in loops whose result is no longer used due to dead code elimination. This optimization has additional compile time cost over standard dead code elimination. This pass only processes entry point functions. It also only processes shaders with logical addressing. It currently will not process functions with function calls. It currently only supports the GLSL.std.450 extended instruction set. It currently does not support any extensions. This pass will be made more effective by first running passes that remove dead control flow and inlines function calls. This pass can be especially useful after running Local Access Chain Conversion, which tends to cause cycles of dead code to be left after Store/Load elimination passes are completed. These cycles cannot be eliminated with standard dead code elimination. Additionally: This transform uses a whitelist of instructions that it knows do have side effects, (a.k.a. combinators). It assumes other instructions have side effects: it will not remove them, and assumes they have side effects via their ID operands.
2017-06-08 16:37:21 +00:00
// Copyright (c) 2017 Valve Corporation
// Copyright (c) 2017 LunarG Inc.
//
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
//
// http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS,
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
// See the License for the specific language governing permissions and
// limitations under the License.
Cleanup includes. (#1795) This Cl cleans up the include paths to be relative to the top level directory. Various include-what-you-use fixes have been added.
2018-08-03 19:06:09 +00:00
#include <string>
#include <vector>
#include "test/opt/assembly_builder.h"
#include "test/opt/pass_fixture.h"
#include "test/opt/pass_utils.h"
Add AggressiveDCEPass Create aggressive dead code elimination pass This pass eliminates unused code from functions. In addition, it detects and eliminates code which may have spurious uses but which do not contribute to the output of the function. The most common cause of such code sequences is summations in loops whose result is no longer used due to dead code elimination. This optimization has additional compile time cost over standard dead code elimination. This pass only processes entry point functions. It also only processes shaders with logical addressing. It currently will not process functions with function calls. It currently only supports the GLSL.std.450 extended instruction set. It currently does not support any extensions. This pass will be made more effective by first running passes that remove dead control flow and inlines function calls. This pass can be especially useful after running Local Access Chain Conversion, which tends to cause cycles of dead code to be left after Store/Load elimination passes are completed. These cycles cannot be eliminated with standard dead code elimination. Additionally: This transform uses a whitelist of instructions that it knows do have side effects, (a.k.a. combinators). It assumes other instructions have side effects: it will not remove them, and assumes they have side effects via their ID operands.
2017-06-08 16:37:21 +00:00
Move tests into namespaces (#1689) This CL moves the test into namespaces based on their directories.
2018-07-11 13:24:49 +00:00
namespace spvtools {
namespace opt {
Add AggressiveDCEPass Create aggressive dead code elimination pass This pass eliminates unused code from functions. In addition, it detects and eliminates code which may have spurious uses but which do not contribute to the output of the function. The most common cause of such code sequences is summations in loops whose result is no longer used due to dead code elimination. This optimization has additional compile time cost over standard dead code elimination. This pass only processes entry point functions. It also only processes shaders with logical addressing. It currently will not process functions with function calls. It currently only supports the GLSL.std.450 extended instruction set. It currently does not support any extensions. This pass will be made more effective by first running passes that remove dead control flow and inlines function calls. This pass can be especially useful after running Local Access Chain Conversion, which tends to cause cycles of dead code to be left after Store/Load elimination passes are completed. These cycles cannot be eliminated with standard dead code elimination. Additionally: This transform uses a whitelist of instructions that it knows do have side effects, (a.k.a. combinators). It assumes other instructions have side effects: it will not remove them, and assumes they have side effects via their ID operands.
2017-06-08 16:37:21 +00:00
namespace {
using AggressiveDCETest = PassTest<::testing::Test>;
TEST_F(AggressiveDCETest, EliminateExtendedInst) {
// #version 140
Re-format source tree - NFC. Re-formatted the source tree with the command: $ /usr/bin/clang-format -style=file -i \ $(find include source tools test utils -name '*.cpp' -or -name '*.h') This required a fix to source/val/decoration.h. It was not including spirv.h, which broke builds when the #include headers were re-ordered by clang-format.
2017-11-27 15:16:41 +00:00
//
Add AggressiveDCEPass Create aggressive dead code elimination pass This pass eliminates unused code from functions. In addition, it detects and eliminates code which may have spurious uses but which do not contribute to the output of the function. The most common cause of such code sequences is summations in loops whose result is no longer used due to dead code elimination. This optimization has additional compile time cost over standard dead code elimination. This pass only processes entry point functions. It also only processes shaders with logical addressing. It currently will not process functions with function calls. It currently only supports the GLSL.std.450 extended instruction set. It currently does not support any extensions. This pass will be made more effective by first running passes that remove dead control flow and inlines function calls. This pass can be especially useful after running Local Access Chain Conversion, which tends to cause cycles of dead code to be left after Store/Load elimination passes are completed. These cycles cannot be eliminated with standard dead code elimination. Additionally: This transform uses a whitelist of instructions that it knows do have side effects, (a.k.a. combinators). It assumes other instructions have side effects: it will not remove them, and assumes they have side effects via their ID operands.
2017-06-08 16:37:21 +00:00
// in vec4 BaseColor;
// in vec4 Dead;
Re-format source tree - NFC. Re-formatted the source tree with the command: $ /usr/bin/clang-format -style=file -i \ $(find include source tools test utils -name '*.cpp' -or -name '*.h') This required a fix to source/val/decoration.h. It was not including spirv.h, which broke builds when the #include headers were re-ordered by clang-format.
2017-11-27 15:16:41 +00:00
//
Add AggressiveDCEPass Create aggressive dead code elimination pass This pass eliminates unused code from functions. In addition, it detects and eliminates code which may have spurious uses but which do not contribute to the output of the function. The most common cause of such code sequences is summations in loops whose result is no longer used due to dead code elimination. This optimization has additional compile time cost over standard dead code elimination. This pass only processes entry point functions. It also only processes shaders with logical addressing. It currently will not process functions with function calls. It currently only supports the GLSL.std.450 extended instruction set. It currently does not support any extensions. This pass will be made more effective by first running passes that remove dead control flow and inlines function calls. This pass can be especially useful after running Local Access Chain Conversion, which tends to cause cycles of dead code to be left after Store/Load elimination passes are completed. These cycles cannot be eliminated with standard dead code elimination. Additionally: This transform uses a whitelist of instructions that it knows do have side effects, (a.k.a. combinators). It assumes other instructions have side effects: it will not remove them, and assumes they have side effects via their ID operands.
2017-06-08 16:37:21 +00:00
// void main()
// {
// vec4 v = BaseColor;
// vec4 dv = sqrt(Dead);
// gl_FragColor = v;
// }
Allow ADCE to remove dead inputs (#4629) * https://github.com/KhronosGroup/Vulkan-Docs/issues/666 clearly specified that interfaces do not require an input if there is an associated output * ADCE can now remove unused input variables (though they are kept if the preserve interfaces option is used)
2021-11-16 20:54:17 +00:00
const std::string spirv = R"(
OpCapability Shader
Add AggressiveDCEPass Create aggressive dead code elimination pass This pass eliminates unused code from functions. In addition, it detects and eliminates code which may have spurious uses but which do not contribute to the output of the function. The most common cause of such code sequences is summations in loops whose result is no longer used due to dead code elimination. This optimization has additional compile time cost over standard dead code elimination. This pass only processes entry point functions. It also only processes shaders with logical addressing. It currently will not process functions with function calls. It currently only supports the GLSL.std.450 extended instruction set. It currently does not support any extensions. This pass will be made more effective by first running passes that remove dead control flow and inlines function calls. This pass can be especially useful after running Local Access Chain Conversion, which tends to cause cycles of dead code to be left after Store/Load elimination passes are completed. These cycles cannot be eliminated with standard dead code elimination. Additionally: This transform uses a whitelist of instructions that it knows do have side effects, (a.k.a. combinators). It assumes other instructions have side effects: it will not remove them, and assumes they have side effects via their ID operands.
2017-06-08 16:37:21 +00:00
%1 = OpExtInstImport "GLSL.std.450"
OpMemoryModel Logical GLSL450
Allow ADCE to remove dead inputs (#4629) * https://github.com/KhronosGroup/Vulkan-Docs/issues/666 clearly specified that interfaces do not require an input if there is an associated output * ADCE can now remove unused input variables (though they are kept if the preserve interfaces option is used)
2021-11-16 20:54:17 +00:00
; CHECK: OpEntryPoint Fragment %main "main" %BaseColor %gl_FragColor
Add AggressiveDCEPass Create aggressive dead code elimination pass This pass eliminates unused code from functions. In addition, it detects and eliminates code which may have spurious uses but which do not contribute to the output of the function. The most common cause of such code sequences is summations in loops whose result is no longer used due to dead code elimination. This optimization has additional compile time cost over standard dead code elimination. This pass only processes entry point functions. It also only processes shaders with logical addressing. It currently will not process functions with function calls. It currently only supports the GLSL.std.450 extended instruction set. It currently does not support any extensions. This pass will be made more effective by first running passes that remove dead control flow and inlines function calls. This pass can be especially useful after running Local Access Chain Conversion, which tends to cause cycles of dead code to be left after Store/Load elimination passes are completed. These cycles cannot be eliminated with standard dead code elimination. Additionally: This transform uses a whitelist of instructions that it knows do have side effects, (a.k.a. combinators). It assumes other instructions have side effects: it will not remove them, and assumes they have side effects via their ID operands.
2017-06-08 16:37:21 +00:00
OpEntryPoint Fragment %main "main" %BaseColor %Dead %gl_FragColor
OpExecutionMode %main OriginUpperLeft
OpSource GLSL 140
Allow ADCE to remove dead inputs (#4629) * https://github.com/KhronosGroup/Vulkan-Docs/issues/666 clearly specified that interfaces do not require an input if there is an associated output * ADCE can now remove unused input variables (though they are kept if the preserve interfaces option is used)
2021-11-16 20:54:17 +00:00
OpName %main "main"
Add AggressiveDCEPass Create aggressive dead code elimination pass This pass eliminates unused code from functions. In addition, it detects and eliminates code which may have spurious uses but which do not contribute to the output of the function. The most common cause of such code sequences is summations in loops whose result is no longer used due to dead code elimination. This optimization has additional compile time cost over standard dead code elimination. This pass only processes entry point functions. It also only processes shaders with logical addressing. It currently will not process functions with function calls. It currently only supports the GLSL.std.450 extended instruction set. It currently does not support any extensions. This pass will be made more effective by first running passes that remove dead control flow and inlines function calls. This pass can be especially useful after running Local Access Chain Conversion, which tends to cause cycles of dead code to be left after Store/Load elimination passes are completed. These cycles cannot be eliminated with standard dead code elimination. Additionally: This transform uses a whitelist of instructions that it knows do have side effects, (a.k.a. combinators). It assumes other instructions have side effects: it will not remove them, and assumes they have side effects via their ID operands.
2017-06-08 16:37:21 +00:00
OpName %v "v"
OpName %BaseColor "BaseColor"
Allow ADCE to remove dead inputs (#4629) * https://github.com/KhronosGroup/Vulkan-Docs/issues/666 clearly specified that interfaces do not require an input if there is an associated output * ADCE can now remove unused input variables (though they are kept if the preserve interfaces option is used)
2021-11-16 20:54:17 +00:00
; CHECK-NOT: OpName %dv "dv"
Add AggressiveDCEPass Create aggressive dead code elimination pass This pass eliminates unused code from functions. In addition, it detects and eliminates code which may have spurious uses but which do not contribute to the output of the function. The most common cause of such code sequences is summations in loops whose result is no longer used due to dead code elimination. This optimization has additional compile time cost over standard dead code elimination. This pass only processes entry point functions. It also only processes shaders with logical addressing. It currently will not process functions with function calls. It currently only supports the GLSL.std.450 extended instruction set. It currently does not support any extensions. This pass will be made more effective by first running passes that remove dead control flow and inlines function calls. This pass can be especially useful after running Local Access Chain Conversion, which tends to cause cycles of dead code to be left after Store/Load elimination passes are completed. These cycles cannot be eliminated with standard dead code elimination. Additionally: This transform uses a whitelist of instructions that it knows do have side effects, (a.k.a. combinators). It assumes other instructions have side effects: it will not remove them, and assumes they have side effects via their ID operands.
2017-06-08 16:37:21 +00:00
OpName %dv "dv"
Allow ADCE to remove dead inputs (#4629) * https://github.com/KhronosGroup/Vulkan-Docs/issues/666 clearly specified that interfaces do not require an input if there is an associated output * ADCE can now remove unused input variables (though they are kept if the preserve interfaces option is used)
2021-11-16 20:54:17 +00:00
; CHECK-NOT: OpName %Dead "Dead"
Add AggressiveDCEPass Create aggressive dead code elimination pass This pass eliminates unused code from functions. In addition, it detects and eliminates code which may have spurious uses but which do not contribute to the output of the function. The most common cause of such code sequences is summations in loops whose result is no longer used due to dead code elimination. This optimization has additional compile time cost over standard dead code elimination. This pass only processes entry point functions. It also only processes shaders with logical addressing. It currently will not process functions with function calls. It currently only supports the GLSL.std.450 extended instruction set. It currently does not support any extensions. This pass will be made more effective by first running passes that remove dead control flow and inlines function calls. This pass can be especially useful after running Local Access Chain Conversion, which tends to cause cycles of dead code to be left after Store/Load elimination passes are completed. These cycles cannot be eliminated with standard dead code elimination. Additionally: This transform uses a whitelist of instructions that it knows do have side effects, (a.k.a. combinators). It assumes other instructions have side effects: it will not remove them, and assumes they have side effects via their ID operands.
2017-06-08 16:37:21 +00:00
OpName %Dead "Dead"
OpName %gl_FragColor "gl_FragColor"
Allow ADCE to remove dead inputs (#4629) * https://github.com/KhronosGroup/Vulkan-Docs/issues/666 clearly specified that interfaces do not require an input if there is an associated output * ADCE can now remove unused input variables (though they are kept if the preserve interfaces option is used)
2021-11-16 20:54:17 +00:00
%void = OpTypeVoid
Add AggressiveDCEPass Create aggressive dead code elimination pass This pass eliminates unused code from functions. In addition, it detects and eliminates code which may have spurious uses but which do not contribute to the output of the function. The most common cause of such code sequences is summations in loops whose result is no longer used due to dead code elimination. This optimization has additional compile time cost over standard dead code elimination. This pass only processes entry point functions. It also only processes shaders with logical addressing. It currently will not process functions with function calls. It currently only supports the GLSL.std.450 extended instruction set. It currently does not support any extensions. This pass will be made more effective by first running passes that remove dead control flow and inlines function calls. This pass can be especially useful after running Local Access Chain Conversion, which tends to cause cycles of dead code to be left after Store/Load elimination passes are completed. These cycles cannot be eliminated with standard dead code elimination. Additionally: This transform uses a whitelist of instructions that it knows do have side effects, (a.k.a. combinators). It assumes other instructions have side effects: it will not remove them, and assumes they have side effects via their ID operands.
2017-06-08 16:37:21 +00:00
%9 = OpTypeFunction %void
%float = OpTypeFloat 32
%v4float = OpTypeVector %float 4
%_ptr_Function_v4float = OpTypePointer Function %v4float
%_ptr_Input_v4float = OpTypePointer Input %v4float
%BaseColor = OpVariable %_ptr_Input_v4float Input
Allow ADCE to remove dead inputs (#4629) * https://github.com/KhronosGroup/Vulkan-Docs/issues/666 clearly specified that interfaces do not require an input if there is an associated output * ADCE can now remove unused input variables (though they are kept if the preserve interfaces option is used)
2021-11-16 20:54:17 +00:00
; CHECK-NOT: %Dead = OpVariable
Add AggressiveDCEPass Create aggressive dead code elimination pass This pass eliminates unused code from functions. In addition, it detects and eliminates code which may have spurious uses but which do not contribute to the output of the function. The most common cause of such code sequences is summations in loops whose result is no longer used due to dead code elimination. This optimization has additional compile time cost over standard dead code elimination. This pass only processes entry point functions. It also only processes shaders with logical addressing. It currently will not process functions with function calls. It currently only supports the GLSL.std.450 extended instruction set. It currently does not support any extensions. This pass will be made more effective by first running passes that remove dead control flow and inlines function calls. This pass can be especially useful after running Local Access Chain Conversion, which tends to cause cycles of dead code to be left after Store/Load elimination passes are completed. These cycles cannot be eliminated with standard dead code elimination. Additionally: This transform uses a whitelist of instructions that it knows do have side effects, (a.k.a. combinators). It assumes other instructions have side effects: it will not remove them, and assumes they have side effects via their ID operands.
2017-06-08 16:37:21 +00:00
%Dead = OpVariable %_ptr_Input_v4float Input
%_ptr_Output_v4float = OpTypePointer Output %v4float
%gl_FragColor = OpVariable %_ptr_Output_v4float Output
Allow ADCE to remove dead inputs (#4629) * https://github.com/KhronosGroup/Vulkan-Docs/issues/666 clearly specified that interfaces do not require an input if there is an associated output * ADCE can now remove unused input variables (though they are kept if the preserve interfaces option is used)
2021-11-16 20:54:17 +00:00
%main = OpFunction %void None %9
Add AggressiveDCEPass Create aggressive dead code elimination pass This pass eliminates unused code from functions. In addition, it detects and eliminates code which may have spurious uses but which do not contribute to the output of the function. The most common cause of such code sequences is summations in loops whose result is no longer used due to dead code elimination. This optimization has additional compile time cost over standard dead code elimination. This pass only processes entry point functions. It also only processes shaders with logical addressing. It currently will not process functions with function calls. It currently only supports the GLSL.std.450 extended instruction set. It currently does not support any extensions. This pass will be made more effective by first running passes that remove dead control flow and inlines function calls. This pass can be especially useful after running Local Access Chain Conversion, which tends to cause cycles of dead code to be left after Store/Load elimination passes are completed. These cycles cannot be eliminated with standard dead code elimination. Additionally: This transform uses a whitelist of instructions that it knows do have side effects, (a.k.a. combinators). It assumes other instructions have side effects: it will not remove them, and assumes they have side effects via their ID operands.
2017-06-08 16:37:21 +00:00
%15 = OpLabel
%v = OpVariable %_ptr_Function_v4float Function
Allow ADCE to remove dead inputs (#4629) * https://github.com/KhronosGroup/Vulkan-Docs/issues/666 clearly specified that interfaces do not require an input if there is an associated output * ADCE can now remove unused input variables (though they are kept if the preserve interfaces option is used)
2021-11-16 20:54:17 +00:00
; CHECK-NOT: %dv = OpVariable
ADCE: Empty Loop Elimination This entirely eliminates loops which do not contain live code.
2017-10-25 00:58:48 +00:00
%dv = OpVariable %_ptr_Function_v4float Function
Add AggressiveDCEPass Create aggressive dead code elimination pass This pass eliminates unused code from functions. In addition, it detects and eliminates code which may have spurious uses but which do not contribute to the output of the function. The most common cause of such code sequences is summations in loops whose result is no longer used due to dead code elimination. This optimization has additional compile time cost over standard dead code elimination. This pass only processes entry point functions. It also only processes shaders with logical addressing. It currently will not process functions with function calls. It currently only supports the GLSL.std.450 extended instruction set. It currently does not support any extensions. This pass will be made more effective by first running passes that remove dead control flow and inlines function calls. This pass can be especially useful after running Local Access Chain Conversion, which tends to cause cycles of dead code to be left after Store/Load elimination passes are completed. These cycles cannot be eliminated with standard dead code elimination. Additionally: This transform uses a whitelist of instructions that it knows do have side effects, (a.k.a. combinators). It assumes other instructions have side effects: it will not remove them, and assumes they have side effects via their ID operands.
2017-06-08 16:37:21 +00:00
%16 = OpLoad %v4float %BaseColor
OpStore %v %16
Allow ADCE to remove dead inputs (#4629) * https://github.com/KhronosGroup/Vulkan-Docs/issues/666 clearly specified that interfaces do not require an input if there is an associated output * ADCE can now remove unused input variables (though they are kept if the preserve interfaces option is used)
2021-11-16 20:54:17 +00:00
; CHECK-NOT: OpLoad %v4float %Dead
Add AggressiveDCEPass Create aggressive dead code elimination pass This pass eliminates unused code from functions. In addition, it detects and eliminates code which may have spurious uses but which do not contribute to the output of the function. The most common cause of such code sequences is summations in loops whose result is no longer used due to dead code elimination. This optimization has additional compile time cost over standard dead code elimination. This pass only processes entry point functions. It also only processes shaders with logical addressing. It currently will not process functions with function calls. It currently only supports the GLSL.std.450 extended instruction set. It currently does not support any extensions. This pass will be made more effective by first running passes that remove dead control flow and inlines function calls. This pass can be especially useful after running Local Access Chain Conversion, which tends to cause cycles of dead code to be left after Store/Load elimination passes are completed. These cycles cannot be eliminated with standard dead code elimination. Additionally: This transform uses a whitelist of instructions that it knows do have side effects, (a.k.a. combinators). It assumes other instructions have side effects: it will not remove them, and assumes they have side effects via their ID operands.
2017-06-08 16:37:21 +00:00
%17 = OpLoad %v4float %Dead
Allow ADCE to remove dead inputs (#4629) * https://github.com/KhronosGroup/Vulkan-Docs/issues/666 clearly specified that interfaces do not require an input if there is an associated output * ADCE can now remove unused input variables (though they are kept if the preserve interfaces option is used)
2021-11-16 20:54:17 +00:00
; CHECK-NOT: OpExtInst %v4float %1 Sqrt
ADCE: Empty Loop Elimination This entirely eliminates loops which do not contain live code.
2017-10-25 00:58:48 +00:00
%18 = OpExtInst %v4float %1 Sqrt %17
Allow ADCE to remove dead inputs (#4629) * https://github.com/KhronosGroup/Vulkan-Docs/issues/666 clearly specified that interfaces do not require an input if there is an associated output * ADCE can now remove unused input variables (though they are kept if the preserve interfaces option is used)
2021-11-16 20:54:17 +00:00
; CHECK-NOT: OpStore %dv
Add AggressiveDCEPass Create aggressive dead code elimination pass This pass eliminates unused code from functions. In addition, it detects and eliminates code which may have spurious uses but which do not contribute to the output of the function. The most common cause of such code sequences is summations in loops whose result is no longer used due to dead code elimination. This optimization has additional compile time cost over standard dead code elimination. This pass only processes entry point functions. It also only processes shaders with logical addressing. It currently will not process functions with function calls. It currently only supports the GLSL.std.450 extended instruction set. It currently does not support any extensions. This pass will be made more effective by first running passes that remove dead control flow and inlines function calls. This pass can be especially useful after running Local Access Chain Conversion, which tends to cause cycles of dead code to be left after Store/Load elimination passes are completed. These cycles cannot be eliminated with standard dead code elimination. Additionally: This transform uses a whitelist of instructions that it knows do have side effects, (a.k.a. combinators). It assumes other instructions have side effects: it will not remove them, and assumes they have side effects via their ID operands.
2017-06-08 16:37:21 +00:00
OpStore %dv %18
%19 = OpLoad %v4float %v
OpStore %gl_FragColor %19
OpReturn
OpFunctionEnd
)";
Allow ADCE to remove dead inputs (#4629) * https://github.com/KhronosGroup/Vulkan-Docs/issues/666 clearly specified that interfaces do not require an input if there is an associated output * ADCE can now remove unused input variables (though they are kept if the preserve interfaces option is used)
2021-11-16 20:54:17 +00:00
SinglePassRunAndMatch<AggressiveDCEPass>(spirv, true);
Add AggressiveDCEPass Create aggressive dead code elimination pass This pass eliminates unused code from functions. In addition, it detects and eliminates code which may have spurious uses but which do not contribute to the output of the function. The most common cause of such code sequences is summations in loops whose result is no longer used due to dead code elimination. This optimization has additional compile time cost over standard dead code elimination. This pass only processes entry point functions. It also only processes shaders with logical addressing. It currently will not process functions with function calls. It currently only supports the GLSL.std.450 extended instruction set. It currently does not support any extensions. This pass will be made more effective by first running passes that remove dead control flow and inlines function calls. This pass can be especially useful after running Local Access Chain Conversion, which tends to cause cycles of dead code to be left after Store/Load elimination passes are completed. These cycles cannot be eliminated with standard dead code elimination. Additionally: This transform uses a whitelist of instructions that it knows do have side effects, (a.k.a. combinators). It assumes other instructions have side effects: it will not remove them, and assumes they have side effects via their ID operands.
2017-06-08 16:37:21 +00:00
}
TEST_F(AggressiveDCETest, NoEliminateFrexp) {
// Note: SPIR-V hand-edited to utilize Frexp
//
// #version 450
Re-format source tree - NFC. Re-formatted the source tree with the command: $ /usr/bin/clang-format -style=file -i \ $(find include source tools test utils -name '*.cpp' -or -name '*.h') This required a fix to source/val/decoration.h. It was not including spirv.h, which broke builds when the #include headers were re-ordered by clang-format.
2017-11-27 15:16:41 +00:00
//
Add AggressiveDCEPass Create aggressive dead code elimination pass This pass eliminates unused code from functions. In addition, it detects and eliminates code which may have spurious uses but which do not contribute to the output of the function. The most common cause of such code sequences is summations in loops whose result is no longer used due to dead code elimination. This optimization has additional compile time cost over standard dead code elimination. This pass only processes entry point functions. It also only processes shaders with logical addressing. It currently will not process functions with function calls. It currently only supports the GLSL.std.450 extended instruction set. It currently does not support any extensions. This pass will be made more effective by first running passes that remove dead control flow and inlines function calls. This pass can be especially useful after running Local Access Chain Conversion, which tends to cause cycles of dead code to be left after Store/Load elimination passes are completed. These cycles cannot be eliminated with standard dead code elimination. Additionally: This transform uses a whitelist of instructions that it knows do have side effects, (a.k.a. combinators). It assumes other instructions have side effects: it will not remove them, and assumes they have side effects via their ID operands.
2017-06-08 16:37:21 +00:00
// in vec4 BaseColor;
// in vec4 Dead;
// out vec4 Color;
// out ivec4 iv2;
Re-format source tree - NFC. Re-formatted the source tree with the command: $ /usr/bin/clang-format -style=file -i \ $(find include source tools test utils -name '*.cpp' -or -name '*.h') This required a fix to source/val/decoration.h. It was not including spirv.h, which broke builds when the #include headers were re-ordered by clang-format.
2017-11-27 15:16:41 +00:00
//
Add AggressiveDCEPass Create aggressive dead code elimination pass This pass eliminates unused code from functions. In addition, it detects and eliminates code which may have spurious uses but which do not contribute to the output of the function. The most common cause of such code sequences is summations in loops whose result is no longer used due to dead code elimination. This optimization has additional compile time cost over standard dead code elimination. This pass only processes entry point functions. It also only processes shaders with logical addressing. It currently will not process functions with function calls. It currently only supports the GLSL.std.450 extended instruction set. It currently does not support any extensions. This pass will be made more effective by first running passes that remove dead control flow and inlines function calls. This pass can be especially useful after running Local Access Chain Conversion, which tends to cause cycles of dead code to be left after Store/Load elimination passes are completed. These cycles cannot be eliminated with standard dead code elimination. Additionally: This transform uses a whitelist of instructions that it knows do have side effects, (a.k.a. combinators). It assumes other instructions have side effects: it will not remove them, and assumes they have side effects via their ID operands.
2017-06-08 16:37:21 +00:00
// void main()
// {
// vec4 v = BaseColor;
// vec4 dv = frexp(Dead, iv2);
// Color = v;
// }
const std::string predefs1 =
R"(OpCapability Shader
%1 = OpExtInstImport "GLSL.std.450"
OpMemoryModel Logical GLSL450
OpEntryPoint Fragment %main "main" %BaseColor %Dead %iv2 %Color
OpExecutionMode %main OriginUpperLeft
OpSource GLSL 450
)";
const std::string names_before =
R"(OpName %main "main"
OpName %v "v"
OpName %BaseColor "BaseColor"
OpName %dv "dv"
OpName %Dead "Dead"
OpName %iv2 "iv2"
OpName %ResType "ResType"
OpName %Color "Color"
)";
const std::string names_after =
R"(OpName %main "main"
OpName %v "v"
OpName %BaseColor "BaseColor"
OpName %Dead "Dead"
OpName %iv2 "iv2"
OpName %Color "Color"
)";
Adding additional functionality to ADCE. Modified ADCE to remove dead globals. * Entry point and execution mode instructions are marked as alive * Reachable functions and their parameters are marked as alive * Instruction deletion now deferred until the end of the pass * Eliminated dead insts set, added IsDead to calculate that value instead * Ported applicable dead variable elimination tests * Ported dead constant elim tests Added dead function elimination to ADCE * ported dead function elim tests Added handling of decoration groups in ADCE * Uses a custom sorter to traverse decorations in a specific order * Simplifies necessary checks Updated -O and -Os pass lists.
2017-12-18 17:13:10 +00:00
const std::string predefs2_before =
Add AggressiveDCEPass Create aggressive dead code elimination pass This pass eliminates unused code from functions. In addition, it detects and eliminates code which may have spurious uses but which do not contribute to the output of the function. The most common cause of such code sequences is summations in loops whose result is no longer used due to dead code elimination. This optimization has additional compile time cost over standard dead code elimination. This pass only processes entry point functions. It also only processes shaders with logical addressing. It currently will not process functions with function calls. It currently only supports the GLSL.std.450 extended instruction set. It currently does not support any extensions. This pass will be made more effective by first running passes that remove dead control flow and inlines function calls. This pass can be especially useful after running Local Access Chain Conversion, which tends to cause cycles of dead code to be left after Store/Load elimination passes are completed. These cycles cannot be eliminated with standard dead code elimination. Additionally: This transform uses a whitelist of instructions that it knows do have side effects, (a.k.a. combinators). It assumes other instructions have side effects: it will not remove them, and assumes they have side effects via their ID operands.
2017-06-08 16:37:21 +00:00
R"(%void = OpTypeVoid
%11 = OpTypeFunction %void
%float = OpTypeFloat 32
%v4float = OpTypeVector %float 4
%_ptr_Function_v4float = OpTypePointer Function %v4float
%_ptr_Input_v4float = OpTypePointer Input %v4float
%BaseColor = OpVariable %_ptr_Input_v4float Input
%Dead = OpVariable %_ptr_Input_v4float Input
%int = OpTypeInt 32 1
%v4int = OpTypeVector %int 4
%_ptr_Output_v4int = OpTypePointer Output %v4int
%iv2 = OpVariable %_ptr_Output_v4int Output
%ResType = OpTypeStruct %v4float %v4int
%_ptr_Output_v4float = OpTypePointer Output %v4float
%Color = OpVariable %_ptr_Output_v4float Output
Adding additional functionality to ADCE. Modified ADCE to remove dead globals. * Entry point and execution mode instructions are marked as alive * Reachable functions and their parameters are marked as alive * Instruction deletion now deferred until the end of the pass * Eliminated dead insts set, added IsDead to calculate that value instead * Ported applicable dead variable elimination tests * Ported dead constant elim tests Added dead function elimination to ADCE * ported dead function elim tests Added handling of decoration groups in ADCE * Uses a custom sorter to traverse decorations in a specific order * Simplifies necessary checks Updated -O and -Os pass lists.
2017-12-18 17:13:10 +00:00
)";
const std::string predefs2_after =
R"(%void = OpTypeVoid
%11 = OpTypeFunction %void
%float = OpTypeFloat 32
%v4float = OpTypeVector %float 4
%_ptr_Function_v4float = OpTypePointer Function %v4float
%_ptr_Input_v4float = OpTypePointer Input %v4float
%BaseColor = OpVariable %_ptr_Input_v4float Input
%Dead = OpVariable %_ptr_Input_v4float Input
%int = OpTypeInt 32 1
%v4int = OpTypeVector %int 4
%_ptr_Output_v4int = OpTypePointer Output %v4int
%iv2 = OpVariable %_ptr_Output_v4int Output
%_ptr_Output_v4float = OpTypePointer Output %v4float
%Color = OpVariable %_ptr_Output_v4float Output
Add AggressiveDCEPass Create aggressive dead code elimination pass This pass eliminates unused code from functions. In addition, it detects and eliminates code which may have spurious uses but which do not contribute to the output of the function. The most common cause of such code sequences is summations in loops whose result is no longer used due to dead code elimination. This optimization has additional compile time cost over standard dead code elimination. This pass only processes entry point functions. It also only processes shaders with logical addressing. It currently will not process functions with function calls. It currently only supports the GLSL.std.450 extended instruction set. It currently does not support any extensions. This pass will be made more effective by first running passes that remove dead control flow and inlines function calls. This pass can be especially useful after running Local Access Chain Conversion, which tends to cause cycles of dead code to be left after Store/Load elimination passes are completed. These cycles cannot be eliminated with standard dead code elimination. Additionally: This transform uses a whitelist of instructions that it knows do have side effects, (a.k.a. combinators). It assumes other instructions have side effects: it will not remove them, and assumes they have side effects via their ID operands.
2017-06-08 16:37:21 +00:00
)";
const std::string func_before =
R"(%main = OpFunction %void None %11
%20 = OpLabel
%v = OpVariable %_ptr_Function_v4float Function
ADCE: Empty Loop Elimination This entirely eliminates loops which do not contain live code.
2017-10-25 00:58:48 +00:00
%dv = OpVariable %_ptr_Function_v4float Function
Add AggressiveDCEPass Create aggressive dead code elimination pass This pass eliminates unused code from functions. In addition, it detects and eliminates code which may have spurious uses but which do not contribute to the output of the function. The most common cause of such code sequences is summations in loops whose result is no longer used due to dead code elimination. This optimization has additional compile time cost over standard dead code elimination. This pass only processes entry point functions. It also only processes shaders with logical addressing. It currently will not process functions with function calls. It currently only supports the GLSL.std.450 extended instruction set. It currently does not support any extensions. This pass will be made more effective by first running passes that remove dead control flow and inlines function calls. This pass can be especially useful after running Local Access Chain Conversion, which tends to cause cycles of dead code to be left after Store/Load elimination passes are completed. These cycles cannot be eliminated with standard dead code elimination. Additionally: This transform uses a whitelist of instructions that it knows do have side effects, (a.k.a. combinators). It assumes other instructions have side effects: it will not remove them, and assumes they have side effects via their ID operands.
2017-06-08 16:37:21 +00:00
%21 = OpLoad %v4float %BaseColor
OpStore %v %21
%22 = OpLoad %v4float %Dead
%23 = OpExtInst %v4float %1 Frexp %22 %iv2
OpStore %dv %23
%24 = OpLoad %v4float %v
OpStore %Color %24
OpReturn
OpFunctionEnd
)";
const std::string func_after =
R"(%main = OpFunction %void None %11
%20 = OpLabel
%v = OpVariable %_ptr_Function_v4float Function
%21 = OpLoad %v4float %BaseColor
OpStore %v %21
%22 = OpLoad %v4float %Dead
%23 = OpExtInst %v4float %1 Frexp %22 %iv2
%24 = OpLoad %v4float %v
OpStore %Color %24
OpReturn
OpFunctionEnd
)";
Move tests into namespaces (#1689) This CL moves the test into namespaces based on their directories.
2018-07-11 13:24:49 +00:00
SinglePassRunAndCheck<AggressiveDCEPass>(
Adding additional functionality to ADCE. Modified ADCE to remove dead globals. * Entry point and execution mode instructions are marked as alive * Reachable functions and their parameters are marked as alive * Instruction deletion now deferred until the end of the pass * Eliminated dead insts set, added IsDead to calculate that value instead * Ported applicable dead variable elimination tests * Ported dead constant elim tests Added dead function elimination to ADCE * ported dead function elim tests Added handling of decoration groups in ADCE * Uses a custom sorter to traverse decorations in a specific order * Simplifies necessary checks Updated -O and -Os pass lists.
2017-12-18 17:13:10 +00:00
predefs1 + names_before + predefs2_before + func_before,
predefs1 + names_after + predefs2_after + func_after, true, true);
Add AggressiveDCEPass Create aggressive dead code elimination pass This pass eliminates unused code from functions. In addition, it detects and eliminates code which may have spurious uses but which do not contribute to the output of the function. The most common cause of such code sequences is summations in loops whose result is no longer used due to dead code elimination. This optimization has additional compile time cost over standard dead code elimination. This pass only processes entry point functions. It also only processes shaders with logical addressing. It currently will not process functions with function calls. It currently only supports the GLSL.std.450 extended instruction set. It currently does not support any extensions. This pass will be made more effective by first running passes that remove dead control flow and inlines function calls. This pass can be especially useful after running Local Access Chain Conversion, which tends to cause cycles of dead code to be left after Store/Load elimination passes are completed. These cycles cannot be eliminated with standard dead code elimination. Additionally: This transform uses a whitelist of instructions that it knows do have side effects, (a.k.a. combinators). It assumes other instructions have side effects: it will not remove them, and assumes they have side effects via their ID operands.
2017-06-08 16:37:21 +00:00
}
TEST_F(AggressiveDCETest, EliminateDecorate) {
// Note: The SPIR-V was hand-edited to add the OpDecorate
//
// #version 140
Re-format source tree - NFC. Re-formatted the source tree with the command: $ /usr/bin/clang-format -style=file -i \ $(find include source tools test utils -name '*.cpp' -or -name '*.h') This required a fix to source/val/decoration.h. It was not including spirv.h, which broke builds when the #include headers were re-ordered by clang-format.
2017-11-27 15:16:41 +00:00
//
Add AggressiveDCEPass Create aggressive dead code elimination pass This pass eliminates unused code from functions. In addition, it detects and eliminates code which may have spurious uses but which do not contribute to the output of the function. The most common cause of such code sequences is summations in loops whose result is no longer used due to dead code elimination. This optimization has additional compile time cost over standard dead code elimination. This pass only processes entry point functions. It also only processes shaders with logical addressing. It currently will not process functions with function calls. It currently only supports the GLSL.std.450 extended instruction set. It currently does not support any extensions. This pass will be made more effective by first running passes that remove dead control flow and inlines function calls. This pass can be especially useful after running Local Access Chain Conversion, which tends to cause cycles of dead code to be left after Store/Load elimination passes are completed. These cycles cannot be eliminated with standard dead code elimination. Additionally: This transform uses a whitelist of instructions that it knows do have side effects, (a.k.a. combinators). It assumes other instructions have side effects: it will not remove them, and assumes they have side effects via their ID operands.
2017-06-08 16:37:21 +00:00
// in vec4 BaseColor;
// in vec4 Dead;
Re-format source tree - NFC. Re-formatted the source tree with the command: $ /usr/bin/clang-format -style=file -i \ $(find include source tools test utils -name '*.cpp' -or -name '*.h') This required a fix to source/val/decoration.h. It was not including spirv.h, which broke builds when the #include headers were re-ordered by clang-format.
2017-11-27 15:16:41 +00:00
//
Add AggressiveDCEPass Create aggressive dead code elimination pass This pass eliminates unused code from functions. In addition, it detects and eliminates code which may have spurious uses but which do not contribute to the output of the function. The most common cause of such code sequences is summations in loops whose result is no longer used due to dead code elimination. This optimization has additional compile time cost over standard dead code elimination. This pass only processes entry point functions. It also only processes shaders with logical addressing. It currently will not process functions with function calls. It currently only supports the GLSL.std.450 extended instruction set. It currently does not support any extensions. This pass will be made more effective by first running passes that remove dead control flow and inlines function calls. This pass can be especially useful after running Local Access Chain Conversion, which tends to cause cycles of dead code to be left after Store/Load elimination passes are completed. These cycles cannot be eliminated with standard dead code elimination. Additionally: This transform uses a whitelist of instructions that it knows do have side effects, (a.k.a. combinators). It assumes other instructions have side effects: it will not remove them, and assumes they have side effects via their ID operands.
2017-06-08 16:37:21 +00:00
// void main()
// {
// vec4 v = BaseColor;
// vec4 dv = Dead * 0.5;
// gl_FragColor = v;
// }
Allow ADCE to remove dead inputs (#4629) * https://github.com/KhronosGroup/Vulkan-Docs/issues/666 clearly specified that interfaces do not require an input if there is an associated output * ADCE can now remove unused input variables (though they are kept if the preserve interfaces option is used)
2021-11-16 20:54:17 +00:00
const std::string spirv =
R"(
OpCapability Shader
Add AggressiveDCEPass Create aggressive dead code elimination pass This pass eliminates unused code from functions. In addition, it detects and eliminates code which may have spurious uses but which do not contribute to the output of the function. The most common cause of such code sequences is summations in loops whose result is no longer used due to dead code elimination. This optimization has additional compile time cost over standard dead code elimination. This pass only processes entry point functions. It also only processes shaders with logical addressing. It currently will not process functions with function calls. It currently only supports the GLSL.std.450 extended instruction set. It currently does not support any extensions. This pass will be made more effective by first running passes that remove dead control flow and inlines function calls. This pass can be especially useful after running Local Access Chain Conversion, which tends to cause cycles of dead code to be left after Store/Load elimination passes are completed. These cycles cannot be eliminated with standard dead code elimination. Additionally: This transform uses a whitelist of instructions that it knows do have side effects, (a.k.a. combinators). It assumes other instructions have side effects: it will not remove them, and assumes they have side effects via their ID operands.
2017-06-08 16:37:21 +00:00
%1 = OpExtInstImport "GLSL.std.450"
OpMemoryModel Logical GLSL450
OpEntryPoint Fragment %main "main" %BaseColor %Dead %gl_FragColor
OpExecutionMode %main OriginUpperLeft
OpSource GLSL 140
Allow ADCE to remove dead inputs (#4629) * https://github.com/KhronosGroup/Vulkan-Docs/issues/666 clearly specified that interfaces do not require an input if there is an associated output * ADCE can now remove unused input variables (though they are kept if the preserve interfaces option is used)
2021-11-16 20:54:17 +00:00
OpName %main "main"
Add AggressiveDCEPass Create aggressive dead code elimination pass This pass eliminates unused code from functions. In addition, it detects and eliminates code which may have spurious uses but which do not contribute to the output of the function. The most common cause of such code sequences is summations in loops whose result is no longer used due to dead code elimination. This optimization has additional compile time cost over standard dead code elimination. This pass only processes entry point functions. It also only processes shaders with logical addressing. It currently will not process functions with function calls. It currently only supports the GLSL.std.450 extended instruction set. It currently does not support any extensions. This pass will be made more effective by first running passes that remove dead control flow and inlines function calls. This pass can be especially useful after running Local Access Chain Conversion, which tends to cause cycles of dead code to be left after Store/Load elimination passes are completed. These cycles cannot be eliminated with standard dead code elimination. Additionally: This transform uses a whitelist of instructions that it knows do have side effects, (a.k.a. combinators). It assumes other instructions have side effects: it will not remove them, and assumes they have side effects via their ID operands.
2017-06-08 16:37:21 +00:00
OpName %v "v"
OpName %BaseColor "BaseColor"
OpName %dv "dv"
OpName %Dead "Dead"
OpName %gl_FragColor "gl_FragColor"
Allow ADCE to remove dead inputs (#4629) * https://github.com/KhronosGroup/Vulkan-Docs/issues/666 clearly specified that interfaces do not require an input if there is an associated output * ADCE can now remove unused input variables (though they are kept if the preserve interfaces option is used)
2021-11-16 20:54:17 +00:00
; CHECK-NOT: OpDecorate
Add AggressiveDCEPass Create aggressive dead code elimination pass This pass eliminates unused code from functions. In addition, it detects and eliminates code which may have spurious uses but which do not contribute to the output of the function. The most common cause of such code sequences is summations in loops whose result is no longer used due to dead code elimination. This optimization has additional compile time cost over standard dead code elimination. This pass only processes entry point functions. It also only processes shaders with logical addressing. It currently will not process functions with function calls. It currently only supports the GLSL.std.450 extended instruction set. It currently does not support any extensions. This pass will be made more effective by first running passes that remove dead control flow and inlines function calls. This pass can be especially useful after running Local Access Chain Conversion, which tends to cause cycles of dead code to be left after Store/Load elimination passes are completed. These cycles cannot be eliminated with standard dead code elimination. Additionally: This transform uses a whitelist of instructions that it knows do have side effects, (a.k.a. combinators). It assumes other instructions have side effects: it will not remove them, and assumes they have side effects via their ID operands.
2017-06-08 16:37:21 +00:00
OpDecorate %8 RelaxedPrecision
Allow ADCE to remove dead inputs (#4629) * https://github.com/KhronosGroup/Vulkan-Docs/issues/666 clearly specified that interfaces do not require an input if there is an associated output * ADCE can now remove unused input variables (though they are kept if the preserve interfaces option is used)
2021-11-16 20:54:17 +00:00
%void = OpTypeVoid
Add AggressiveDCEPass Create aggressive dead code elimination pass This pass eliminates unused code from functions. In addition, it detects and eliminates code which may have spurious uses but which do not contribute to the output of the function. The most common cause of such code sequences is summations in loops whose result is no longer used due to dead code elimination. This optimization has additional compile time cost over standard dead code elimination. This pass only processes entry point functions. It also only processes shaders with logical addressing. It currently will not process functions with function calls. It currently only supports the GLSL.std.450 extended instruction set. It currently does not support any extensions. This pass will be made more effective by first running passes that remove dead control flow and inlines function calls. This pass can be especially useful after running Local Access Chain Conversion, which tends to cause cycles of dead code to be left after Store/Load elimination passes are completed. These cycles cannot be eliminated with standard dead code elimination. Additionally: This transform uses a whitelist of instructions that it knows do have side effects, (a.k.a. combinators). It assumes other instructions have side effects: it will not remove them, and assumes they have side effects via their ID operands.
2017-06-08 16:37:21 +00:00
%10 = OpTypeFunction %void
%float = OpTypeFloat 32
%v4float = OpTypeVector %float 4
%_ptr_Function_v4float = OpTypePointer Function %v4float
%_ptr_Input_v4float = OpTypePointer Input %v4float
%BaseColor = OpVariable %_ptr_Input_v4float Input
%Dead = OpVariable %_ptr_Input_v4float Input
%float_0_5 = OpConstant %float 0.5
%_ptr_Output_v4float = OpTypePointer Output %v4float
%gl_FragColor = OpVariable %_ptr_Output_v4float Output
Allow ADCE to remove dead inputs (#4629) * https://github.com/KhronosGroup/Vulkan-Docs/issues/666 clearly specified that interfaces do not require an input if there is an associated output * ADCE can now remove unused input variables (though they are kept if the preserve interfaces option is used)
2021-11-16 20:54:17 +00:00
%main = OpFunction %void None %10
Add AggressiveDCEPass Create aggressive dead code elimination pass This pass eliminates unused code from functions. In addition, it detects and eliminates code which may have spurious uses but which do not contribute to the output of the function. The most common cause of such code sequences is summations in loops whose result is no longer used due to dead code elimination. This optimization has additional compile time cost over standard dead code elimination. This pass only processes entry point functions. It also only processes shaders with logical addressing. It currently will not process functions with function calls. It currently only supports the GLSL.std.450 extended instruction set. It currently does not support any extensions. This pass will be made more effective by first running passes that remove dead control flow and inlines function calls. This pass can be especially useful after running Local Access Chain Conversion, which tends to cause cycles of dead code to be left after Store/Load elimination passes are completed. These cycles cannot be eliminated with standard dead code elimination. Additionally: This transform uses a whitelist of instructions that it knows do have side effects, (a.k.a. combinators). It assumes other instructions have side effects: it will not remove them, and assumes they have side effects via their ID operands.
2017-06-08 16:37:21 +00:00
%17 = OpLabel
%v = OpVariable %_ptr_Function_v4float Function
ADCE: Empty Loop Elimination This entirely eliminates loops which do not contain live code.
2017-10-25 00:58:48 +00:00
%dv = OpVariable %_ptr_Function_v4float Function
Add AggressiveDCEPass Create aggressive dead code elimination pass This pass eliminates unused code from functions. In addition, it detects and eliminates code which may have spurious uses but which do not contribute to the output of the function. The most common cause of such code sequences is summations in loops whose result is no longer used due to dead code elimination. This optimization has additional compile time cost over standard dead code elimination. This pass only processes entry point functions. It also only processes shaders with logical addressing. It currently will not process functions with function calls. It currently only supports the GLSL.std.450 extended instruction set. It currently does not support any extensions. This pass will be made more effective by first running passes that remove dead control flow and inlines function calls. This pass can be especially useful after running Local Access Chain Conversion, which tends to cause cycles of dead code to be left after Store/Load elimination passes are completed. These cycles cannot be eliminated with standard dead code elimination. Additionally: This transform uses a whitelist of instructions that it knows do have side effects, (a.k.a. combinators). It assumes other instructions have side effects: it will not remove them, and assumes they have side effects via their ID operands.
2017-06-08 16:37:21 +00:00
%18 = OpLoad %v4float %BaseColor
OpStore %v %18
%19 = OpLoad %v4float %Dead
Allow ADCE to remove dead inputs (#4629) * https://github.com/KhronosGroup/Vulkan-Docs/issues/666 clearly specified that interfaces do not require an input if there is an associated output * ADCE can now remove unused input variables (though they are kept if the preserve interfaces option is used)
2021-11-16 20:54:17 +00:00
; CHECK-NOT: OpVectorTimesScalar
Add AggressiveDCEPass Create aggressive dead code elimination pass This pass eliminates unused code from functions. In addition, it detects and eliminates code which may have spurious uses but which do not contribute to the output of the function. The most common cause of such code sequences is summations in loops whose result is no longer used due to dead code elimination. This optimization has additional compile time cost over standard dead code elimination. This pass only processes entry point functions. It also only processes shaders with logical addressing. It currently will not process functions with function calls. It currently only supports the GLSL.std.450 extended instruction set. It currently does not support any extensions. This pass will be made more effective by first running passes that remove dead control flow and inlines function calls. This pass can be especially useful after running Local Access Chain Conversion, which tends to cause cycles of dead code to be left after Store/Load elimination passes are completed. These cycles cannot be eliminated with standard dead code elimination. Additionally: This transform uses a whitelist of instructions that it knows do have side effects, (a.k.a. combinators). It assumes other instructions have side effects: it will not remove them, and assumes they have side effects via their ID operands.
2017-06-08 16:37:21 +00:00
%8 = OpVectorTimesScalar %v4float %19 %float_0_5
OpStore %dv %8
%20 = OpLoad %v4float %v
OpStore %gl_FragColor %20
OpReturn
OpFunctionEnd
)";
Allow ADCE to remove dead inputs (#4629) * https://github.com/KhronosGroup/Vulkan-Docs/issues/666 clearly specified that interfaces do not require an input if there is an associated output * ADCE can now remove unused input variables (though they are kept if the preserve interfaces option is used)
2021-11-16 20:54:17 +00:00
SinglePassRunAndMatch<AggressiveDCEPass>(spirv, true);
Add AggressiveDCEPass Create aggressive dead code elimination pass This pass eliminates unused code from functions. In addition, it detects and eliminates code which may have spurious uses but which do not contribute to the output of the function. The most common cause of such code sequences is summations in loops whose result is no longer used due to dead code elimination. This optimization has additional compile time cost over standard dead code elimination. This pass only processes entry point functions. It also only processes shaders with logical addressing. It currently will not process functions with function calls. It currently only supports the GLSL.std.450 extended instruction set. It currently does not support any extensions. This pass will be made more effective by first running passes that remove dead control flow and inlines function calls. This pass can be especially useful after running Local Access Chain Conversion, which tends to cause cycles of dead code to be left after Store/Load elimination passes are completed. These cycles cannot be eliminated with standard dead code elimination. Additionally: This transform uses a whitelist of instructions that it knows do have side effects, (a.k.a. combinators). It assumes other instructions have side effects: it will not remove them, and assumes they have side effects via their ID operands.
2017-06-08 16:37:21 +00:00
}
TEST_F(AggressiveDCETest, Simple) {
// #version 140
Re-format source tree - NFC. Re-formatted the source tree with the command: $ /usr/bin/clang-format -style=file -i \ $(find include source tools test utils -name '*.cpp' -or -name '*.h') This required a fix to source/val/decoration.h. It was not including spirv.h, which broke builds when the #include headers were re-ordered by clang-format.
2017-11-27 15:16:41 +00:00
//
Add AggressiveDCEPass Create aggressive dead code elimination pass This pass eliminates unused code from functions. In addition, it detects and eliminates code which may have spurious uses but which do not contribute to the output of the function. The most common cause of such code sequences is summations in loops whose result is no longer used due to dead code elimination. This optimization has additional compile time cost over standard dead code elimination. This pass only processes entry point functions. It also only processes shaders with logical addressing. It currently will not process functions with function calls. It currently only supports the GLSL.std.450 extended instruction set. It currently does not support any extensions. This pass will be made more effective by first running passes that remove dead control flow and inlines function calls. This pass can be especially useful after running Local Access Chain Conversion, which tends to cause cycles of dead code to be left after Store/Load elimination passes are completed. These cycles cannot be eliminated with standard dead code elimination. Additionally: This transform uses a whitelist of instructions that it knows do have side effects, (a.k.a. combinators). It assumes other instructions have side effects: it will not remove them, and assumes they have side effects via their ID operands.
2017-06-08 16:37:21 +00:00
// in vec4 BaseColor;
// in vec4 Dead;
Re-format source tree - NFC. Re-formatted the source tree with the command: $ /usr/bin/clang-format -style=file -i \ $(find include source tools test utils -name '*.cpp' -or -name '*.h') This required a fix to source/val/decoration.h. It was not including spirv.h, which broke builds when the #include headers were re-ordered by clang-format.
2017-11-27 15:16:41 +00:00
//
Add AggressiveDCEPass Create aggressive dead code elimination pass This pass eliminates unused code from functions. In addition, it detects and eliminates code which may have spurious uses but which do not contribute to the output of the function. The most common cause of such code sequences is summations in loops whose result is no longer used due to dead code elimination. This optimization has additional compile time cost over standard dead code elimination. This pass only processes entry point functions. It also only processes shaders with logical addressing. It currently will not process functions with function calls. It currently only supports the GLSL.std.450 extended instruction set. It currently does not support any extensions. This pass will be made more effective by first running passes that remove dead control flow and inlines function calls. This pass can be especially useful after running Local Access Chain Conversion, which tends to cause cycles of dead code to be left after Store/Load elimination passes are completed. These cycles cannot be eliminated with standard dead code elimination. Additionally: This transform uses a whitelist of instructions that it knows do have side effects, (a.k.a. combinators). It assumes other instructions have side effects: it will not remove them, and assumes they have side effects via their ID operands.
2017-06-08 16:37:21 +00:00
// void main()
// {
// vec4 v = BaseColor;
// vec4 dv = Dead;
// gl_FragColor = v;
// }
Allow ADCE to remove dead inputs (#4629) * https://github.com/KhronosGroup/Vulkan-Docs/issues/666 clearly specified that interfaces do not require an input if there is an associated output * ADCE can now remove unused input variables (though they are kept if the preserve interfaces option is used)
2021-11-16 20:54:17 +00:00
const std::string spirv =
R"(
OpCapability Shader
Add AggressiveDCEPass Create aggressive dead code elimination pass This pass eliminates unused code from functions. In addition, it detects and eliminates code which may have spurious uses but which do not contribute to the output of the function. The most common cause of such code sequences is summations in loops whose result is no longer used due to dead code elimination. This optimization has additional compile time cost over standard dead code elimination. This pass only processes entry point functions. It also only processes shaders with logical addressing. It currently will not process functions with function calls. It currently only supports the GLSL.std.450 extended instruction set. It currently does not support any extensions. This pass will be made more effective by first running passes that remove dead control flow and inlines function calls. This pass can be especially useful after running Local Access Chain Conversion, which tends to cause cycles of dead code to be left after Store/Load elimination passes are completed. These cycles cannot be eliminated with standard dead code elimination. Additionally: This transform uses a whitelist of instructions that it knows do have side effects, (a.k.a. combinators). It assumes other instructions have side effects: it will not remove them, and assumes they have side effects via their ID operands.
2017-06-08 16:37:21 +00:00
%1 = OpExtInstImport "GLSL.std.450"
OpMemoryModel Logical GLSL450
Allow ADCE to remove dead inputs (#4629) * https://github.com/KhronosGroup/Vulkan-Docs/issues/666 clearly specified that interfaces do not require an input if there is an associated output * ADCE can now remove unused input variables (though they are kept if the preserve interfaces option is used)
2021-11-16 20:54:17 +00:00
; CHECK: OpEntryPoint Fragment %main "main" %BaseColor %gl_FragColor
Add AggressiveDCEPass Create aggressive dead code elimination pass This pass eliminates unused code from functions. In addition, it detects and eliminates code which may have spurious uses but which do not contribute to the output of the function. The most common cause of such code sequences is summations in loops whose result is no longer used due to dead code elimination. This optimization has additional compile time cost over standard dead code elimination. This pass only processes entry point functions. It also only processes shaders with logical addressing. It currently will not process functions with function calls. It currently only supports the GLSL.std.450 extended instruction set. It currently does not support any extensions. This pass will be made more effective by first running passes that remove dead control flow and inlines function calls. This pass can be especially useful after running Local Access Chain Conversion, which tends to cause cycles of dead code to be left after Store/Load elimination passes are completed. These cycles cannot be eliminated with standard dead code elimination. Additionally: This transform uses a whitelist of instructions that it knows do have side effects, (a.k.a. combinators). It assumes other instructions have side effects: it will not remove them, and assumes they have side effects via their ID operands.
2017-06-08 16:37:21 +00:00
OpEntryPoint Fragment %main "main" %BaseColor %Dead %gl_FragColor
OpExecutionMode %main OriginUpperLeft
OpSource GLSL 140
Allow ADCE to remove dead inputs (#4629) * https://github.com/KhronosGroup/Vulkan-Docs/issues/666 clearly specified that interfaces do not require an input if there is an associated output * ADCE can now remove unused input variables (though they are kept if the preserve interfaces option is used)
2021-11-16 20:54:17 +00:00
OpName %main "main"
Add AggressiveDCEPass Create aggressive dead code elimination pass This pass eliminates unused code from functions. In addition, it detects and eliminates code which may have spurious uses but which do not contribute to the output of the function. The most common cause of such code sequences is summations in loops whose result is no longer used due to dead code elimination. This optimization has additional compile time cost over standard dead code elimination. This pass only processes entry point functions. It also only processes shaders with logical addressing. It currently will not process functions with function calls. It currently only supports the GLSL.std.450 extended instruction set. It currently does not support any extensions. This pass will be made more effective by first running passes that remove dead control flow and inlines function calls. This pass can be especially useful after running Local Access Chain Conversion, which tends to cause cycles of dead code to be left after Store/Load elimination passes are completed. These cycles cannot be eliminated with standard dead code elimination. Additionally: This transform uses a whitelist of instructions that it knows do have side effects, (a.k.a. combinators). It assumes other instructions have side effects: it will not remove them, and assumes they have side effects via their ID operands.
2017-06-08 16:37:21 +00:00
OpName %v "v"
OpName %BaseColor "BaseColor"
Allow ADCE to remove dead inputs (#4629) * https://github.com/KhronosGroup/Vulkan-Docs/issues/666 clearly specified that interfaces do not require an input if there is an associated output * ADCE can now remove unused input variables (though they are kept if the preserve interfaces option is used)
2021-11-16 20:54:17 +00:00
; CHECK-NOT: OpName %dv "dv"
Add AggressiveDCEPass Create aggressive dead code elimination pass This pass eliminates unused code from functions. In addition, it detects and eliminates code which may have spurious uses but which do not contribute to the output of the function. The most common cause of such code sequences is summations in loops whose result is no longer used due to dead code elimination. This optimization has additional compile time cost over standard dead code elimination. This pass only processes entry point functions. It also only processes shaders with logical addressing. It currently will not process functions with function calls. It currently only supports the GLSL.std.450 extended instruction set. It currently does not support any extensions. This pass will be made more effective by first running passes that remove dead control flow and inlines function calls. This pass can be especially useful after running Local Access Chain Conversion, which tends to cause cycles of dead code to be left after Store/Load elimination passes are completed. These cycles cannot be eliminated with standard dead code elimination. Additionally: This transform uses a whitelist of instructions that it knows do have side effects, (a.k.a. combinators). It assumes other instructions have side effects: it will not remove them, and assumes they have side effects via their ID operands.
2017-06-08 16:37:21 +00:00
OpName %dv "dv"
Allow ADCE to remove dead inputs (#4629) * https://github.com/KhronosGroup/Vulkan-Docs/issues/666 clearly specified that interfaces do not require an input if there is an associated output * ADCE can now remove unused input variables (though they are kept if the preserve interfaces option is used)
2021-11-16 20:54:17 +00:00
; CHECK-NOT: OpName %Dead "Dead"
Add AggressiveDCEPass Create aggressive dead code elimination pass This pass eliminates unused code from functions. In addition, it detects and eliminates code which may have spurious uses but which do not contribute to the output of the function. The most common cause of such code sequences is summations in loops whose result is no longer used due to dead code elimination. This optimization has additional compile time cost over standard dead code elimination. This pass only processes entry point functions. It also only processes shaders with logical addressing. It currently will not process functions with function calls. It currently only supports the GLSL.std.450 extended instruction set. It currently does not support any extensions. This pass will be made more effective by first running passes that remove dead control flow and inlines function calls. This pass can be especially useful after running Local Access Chain Conversion, which tends to cause cycles of dead code to be left after Store/Load elimination passes are completed. These cycles cannot be eliminated with standard dead code elimination. Additionally: This transform uses a whitelist of instructions that it knows do have side effects, (a.k.a. combinators). It assumes other instructions have side effects: it will not remove them, and assumes they have side effects via their ID operands.
2017-06-08 16:37:21 +00:00
OpName %Dead "Dead"
OpName %gl_FragColor "gl_FragColor"
Allow ADCE to remove dead inputs (#4629) * https://github.com/KhronosGroup/Vulkan-Docs/issues/666 clearly specified that interfaces do not require an input if there is an associated output * ADCE can now remove unused input variables (though they are kept if the preserve interfaces option is used)
2021-11-16 20:54:17 +00:00
%void = OpTypeVoid
Add AggressiveDCEPass Create aggressive dead code elimination pass This pass eliminates unused code from functions. In addition, it detects and eliminates code which may have spurious uses but which do not contribute to the output of the function. The most common cause of such code sequences is summations in loops whose result is no longer used due to dead code elimination. This optimization has additional compile time cost over standard dead code elimination. This pass only processes entry point functions. It also only processes shaders with logical addressing. It currently will not process functions with function calls. It currently only supports the GLSL.std.450 extended instruction set. It currently does not support any extensions. This pass will be made more effective by first running passes that remove dead control flow and inlines function calls. This pass can be especially useful after running Local Access Chain Conversion, which tends to cause cycles of dead code to be left after Store/Load elimination passes are completed. These cycles cannot be eliminated with standard dead code elimination. Additionally: This transform uses a whitelist of instructions that it knows do have side effects, (a.k.a. combinators). It assumes other instructions have side effects: it will not remove them, and assumes they have side effects via their ID operands.
2017-06-08 16:37:21 +00:00
%9 = OpTypeFunction %void
%float = OpTypeFloat 32
%v4float = OpTypeVector %float 4
%_ptr_Function_v4float = OpTypePointer Function %v4float
%_ptr_Input_v4float = OpTypePointer Input %v4float
%BaseColor = OpVariable %_ptr_Input_v4float Input
Allow ADCE to remove dead inputs (#4629) * https://github.com/KhronosGroup/Vulkan-Docs/issues/666 clearly specified that interfaces do not require an input if there is an associated output * ADCE can now remove unused input variables (though they are kept if the preserve interfaces option is used)
2021-11-16 20:54:17 +00:00
; CHECK-NOT: %Dead = OpVariable
Add AggressiveDCEPass Create aggressive dead code elimination pass This pass eliminates unused code from functions. In addition, it detects and eliminates code which may have spurious uses but which do not contribute to the output of the function. The most common cause of such code sequences is summations in loops whose result is no longer used due to dead code elimination. This optimization has additional compile time cost over standard dead code elimination. This pass only processes entry point functions. It also only processes shaders with logical addressing. It currently will not process functions with function calls. It currently only supports the GLSL.std.450 extended instruction set. It currently does not support any extensions. This pass will be made more effective by first running passes that remove dead control flow and inlines function calls. This pass can be especially useful after running Local Access Chain Conversion, which tends to cause cycles of dead code to be left after Store/Load elimination passes are completed. These cycles cannot be eliminated with standard dead code elimination. Additionally: This transform uses a whitelist of instructions that it knows do have side effects, (a.k.a. combinators). It assumes other instructions have side effects: it will not remove them, and assumes they have side effects via their ID operands.
2017-06-08 16:37:21 +00:00
%Dead = OpVariable %_ptr_Input_v4float Input
%_ptr_Output_v4float = OpTypePointer Output %v4float
%gl_FragColor = OpVariable %_ptr_Output_v4float Output
Allow ADCE to remove dead inputs (#4629) * https://github.com/KhronosGroup/Vulkan-Docs/issues/666 clearly specified that interfaces do not require an input if there is an associated output * ADCE can now remove unused input variables (though they are kept if the preserve interfaces option is used)
2021-11-16 20:54:17 +00:00
%main = OpFunction %void None %9
Add AggressiveDCEPass Create aggressive dead code elimination pass This pass eliminates unused code from functions. In addition, it detects and eliminates code which may have spurious uses but which do not contribute to the output of the function. The most common cause of such code sequences is summations in loops whose result is no longer used due to dead code elimination. This optimization has additional compile time cost over standard dead code elimination. This pass only processes entry point functions. It also only processes shaders with logical addressing. It currently will not process functions with function calls. It currently only supports the GLSL.std.450 extended instruction set. It currently does not support any extensions. This pass will be made more effective by first running passes that remove dead control flow and inlines function calls. This pass can be especially useful after running Local Access Chain Conversion, which tends to cause cycles of dead code to be left after Store/Load elimination passes are completed. These cycles cannot be eliminated with standard dead code elimination. Additionally: This transform uses a whitelist of instructions that it knows do have side effects, (a.k.a. combinators). It assumes other instructions have side effects: it will not remove them, and assumes they have side effects via their ID operands.
2017-06-08 16:37:21 +00:00
%15 = OpLabel
%v = OpVariable %_ptr_Function_v4float Function
Allow ADCE to remove dead inputs (#4629) * https://github.com/KhronosGroup/Vulkan-Docs/issues/666 clearly specified that interfaces do not require an input if there is an associated output * ADCE can now remove unused input variables (though they are kept if the preserve interfaces option is used)
2021-11-16 20:54:17 +00:00
; CHECK-NOT: %dv = OpVariable
ADCE: Empty Loop Elimination This entirely eliminates loops which do not contain live code.
2017-10-25 00:58:48 +00:00
%dv = OpVariable %_ptr_Function_v4float Function
Add AggressiveDCEPass Create aggressive dead code elimination pass This pass eliminates unused code from functions. In addition, it detects and eliminates code which may have spurious uses but which do not contribute to the output of the function. The most common cause of such code sequences is summations in loops whose result is no longer used due to dead code elimination. This optimization has additional compile time cost over standard dead code elimination. This pass only processes entry point functions. It also only processes shaders with logical addressing. It currently will not process functions with function calls. It currently only supports the GLSL.std.450 extended instruction set. It currently does not support any extensions. This pass will be made more effective by first running passes that remove dead control flow and inlines function calls. This pass can be especially useful after running Local Access Chain Conversion, which tends to cause cycles of dead code to be left after Store/Load elimination passes are completed. These cycles cannot be eliminated with standard dead code elimination. Additionally: This transform uses a whitelist of instructions that it knows do have side effects, (a.k.a. combinators). It assumes other instructions have side effects: it will not remove them, and assumes they have side effects via their ID operands.
2017-06-08 16:37:21 +00:00
%16 = OpLoad %v4float %BaseColor
OpStore %v %16
Allow ADCE to remove dead inputs (#4629) * https://github.com/KhronosGroup/Vulkan-Docs/issues/666 clearly specified that interfaces do not require an input if there is an associated output * ADCE can now remove unused input variables (though they are kept if the preserve interfaces option is used)
2021-11-16 20:54:17 +00:00
; CHECK-NOT: OpLoad %v4float %Dead
Add AggressiveDCEPass Create aggressive dead code elimination pass This pass eliminates unused code from functions. In addition, it detects and eliminates code which may have spurious uses but which do not contribute to the output of the function. The most common cause of such code sequences is summations in loops whose result is no longer used due to dead code elimination. This optimization has additional compile time cost over standard dead code elimination. This pass only processes entry point functions. It also only processes shaders with logical addressing. It currently will not process functions with function calls. It currently only supports the GLSL.std.450 extended instruction set. It currently does not support any extensions. This pass will be made more effective by first running passes that remove dead control flow and inlines function calls. This pass can be especially useful after running Local Access Chain Conversion, which tends to cause cycles of dead code to be left after Store/Load elimination passes are completed. These cycles cannot be eliminated with standard dead code elimination. Additionally: This transform uses a whitelist of instructions that it knows do have side effects, (a.k.a. combinators). It assumes other instructions have side effects: it will not remove them, and assumes they have side effects via their ID operands.
2017-06-08 16:37:21 +00:00
%17 = OpLoad %v4float %Dead
Allow ADCE to remove dead inputs (#4629) * https://github.com/KhronosGroup/Vulkan-Docs/issues/666 clearly specified that interfaces do not require an input if there is an associated output * ADCE can now remove unused input variables (though they are kept if the preserve interfaces option is used)
2021-11-16 20:54:17 +00:00
; CHECK-NOT: OpStore %dv
Add AggressiveDCEPass Create aggressive dead code elimination pass This pass eliminates unused code from functions. In addition, it detects and eliminates code which may have spurious uses but which do not contribute to the output of the function. The most common cause of such code sequences is summations in loops whose result is no longer used due to dead code elimination. This optimization has additional compile time cost over standard dead code elimination. This pass only processes entry point functions. It also only processes shaders with logical addressing. It currently will not process functions with function calls. It currently only supports the GLSL.std.450 extended instruction set. It currently does not support any extensions. This pass will be made more effective by first running passes that remove dead control flow and inlines function calls. This pass can be especially useful after running Local Access Chain Conversion, which tends to cause cycles of dead code to be left after Store/Load elimination passes are completed. These cycles cannot be eliminated with standard dead code elimination. Additionally: This transform uses a whitelist of instructions that it knows do have side effects, (a.k.a. combinators). It assumes other instructions have side effects: it will not remove them, and assumes they have side effects via their ID operands.
2017-06-08 16:37:21 +00:00
OpStore %dv %17
%18 = OpLoad %v4float %v
OpStore %gl_FragColor %18
OpReturn
OpFunctionEnd
)";
Allow ADCE to remove dead inputs (#4629) * https://github.com/KhronosGroup/Vulkan-Docs/issues/666 clearly specified that interfaces do not require an input if there is an associated output * ADCE can now remove unused input variables (though they are kept if the preserve interfaces option is used)
2021-11-16 20:54:17 +00:00
SinglePassRunAndMatch<AggressiveDCEPass>(spirv, true);
Add AggressiveDCEPass Create aggressive dead code elimination pass This pass eliminates unused code from functions. In addition, it detects and eliminates code which may have spurious uses but which do not contribute to the output of the function. The most common cause of such code sequences is summations in loops whose result is no longer used due to dead code elimination. This optimization has additional compile time cost over standard dead code elimination. This pass only processes entry point functions. It also only processes shaders with logical addressing. It currently will not process functions with function calls. It currently only supports the GLSL.std.450 extended instruction set. It currently does not support any extensions. This pass will be made more effective by first running passes that remove dead control flow and inlines function calls. This pass can be especially useful after running Local Access Chain Conversion, which tends to cause cycles of dead code to be left after Store/Load elimination passes are completed. These cycles cannot be eliminated with standard dead code elimination. Additionally: This transform uses a whitelist of instructions that it knows do have side effects, (a.k.a. combinators). It assumes other instructions have side effects: it will not remove them, and assumes they have side effects via their ID operands.
2017-06-08 16:37:21 +00:00
}
Update access control lists. (#3433) This CL updates the access control lists used in SPIRV-Tools to the more descriptive allow/deny naming.
2020-06-15 17:20:40 +00:00
TEST_F(AggressiveDCETest, OptAllowListExtension) {
Add extension whitelists to size-reduction passes. Currently only SPV_KHR_variable_pointers is disallowed in passes which do pointer analysis. Positive and negative tests of the general extensions mechanism were added to aggressive_dce but cover all passes.
2017-07-19 00:57:26 +00:00
// #version 140
Re-format source tree - NFC. Re-formatted the source tree with the command: $ /usr/bin/clang-format -style=file -i \ $(find include source tools test utils -name '*.cpp' -or -name '*.h') This required a fix to source/val/decoration.h. It was not including spirv.h, which broke builds when the #include headers were re-ordered by clang-format.
2017-11-27 15:16:41 +00:00
//
Add extension whitelists to size-reduction passes. Currently only SPV_KHR_variable_pointers is disallowed in passes which do pointer analysis. Positive and negative tests of the general extensions mechanism were added to aggressive_dce but cover all passes.
2017-07-19 00:57:26 +00:00
// in vec4 BaseColor;
// in vec4 Dead;
Re-format source tree - NFC. Re-formatted the source tree with the command: $ /usr/bin/clang-format -style=file -i \ $(find include source tools test utils -name '*.cpp' -or -name '*.h') This required a fix to source/val/decoration.h. It was not including spirv.h, which broke builds when the #include headers were re-ordered by clang-format.
2017-11-27 15:16:41 +00:00
//
Add extension whitelists to size-reduction passes. Currently only SPV_KHR_variable_pointers is disallowed in passes which do pointer analysis. Positive and negative tests of the general extensions mechanism were added to aggressive_dce but cover all passes.
2017-07-19 00:57:26 +00:00
// void main()
// {
// vec4 v = BaseColor;
// vec4 dv = Dead;
// gl_FragColor = v;
// }
Allow ADCE to remove dead inputs (#4629) * https://github.com/KhronosGroup/Vulkan-Docs/issues/666 clearly specified that interfaces do not require an input if there is an associated output * ADCE can now remove unused input variables (though they are kept if the preserve interfaces option is used)
2021-11-16 20:54:17 +00:00
const std::string spirv =
Add extension whitelists to size-reduction passes. Currently only SPV_KHR_variable_pointers is disallowed in passes which do pointer analysis. Positive and negative tests of the general extensions mechanism were added to aggressive_dce but cover all passes.
2017-07-19 00:57:26 +00:00
R"(OpCapability Shader
OpExtension "SPV_AMD_gpu_shader_int16"
%1 = OpExtInstImport "GLSL.std.450"
OpMemoryModel Logical GLSL450
Allow ADCE to remove dead inputs (#4629) * https://github.com/KhronosGroup/Vulkan-Docs/issues/666 clearly specified that interfaces do not require an input if there is an associated output * ADCE can now remove unused input variables (though they are kept if the preserve interfaces option is used)
2021-11-16 20:54:17 +00:00
; CHECK: OpEntryPoint Fragment %main "main" %BaseColor %gl_FragColor
Add extension whitelists to size-reduction passes. Currently only SPV_KHR_variable_pointers is disallowed in passes which do pointer analysis. Positive and negative tests of the general extensions mechanism were added to aggressive_dce but cover all passes.
2017-07-19 00:57:26 +00:00
OpEntryPoint Fragment %main "main" %BaseColor %Dead %gl_FragColor
OpExecutionMode %main OriginUpperLeft
OpSource GLSL 140
Allow ADCE to remove dead inputs (#4629) * https://github.com/KhronosGroup/Vulkan-Docs/issues/666 clearly specified that interfaces do not require an input if there is an associated output * ADCE can now remove unused input variables (though they are kept if the preserve interfaces option is used)
2021-11-16 20:54:17 +00:00
OpName %main "main"
Add extension whitelists to size-reduction passes. Currently only SPV_KHR_variable_pointers is disallowed in passes which do pointer analysis. Positive and negative tests of the general extensions mechanism were added to aggressive_dce but cover all passes.
2017-07-19 00:57:26 +00:00
OpName %v "v"
OpName %BaseColor "BaseColor"
OpName %dv "dv"
OpName %Dead "Dead"
OpName %gl_FragColor "gl_FragColor"
Allow ADCE to remove dead inputs (#4629) * https://github.com/KhronosGroup/Vulkan-Docs/issues/666 clearly specified that interfaces do not require an input if there is an associated output * ADCE can now remove unused input variables (though they are kept if the preserve interfaces option is used)
2021-11-16 20:54:17 +00:00
%void = OpTypeVoid
Add extension whitelists to size-reduction passes. Currently only SPV_KHR_variable_pointers is disallowed in passes which do pointer analysis. Positive and negative tests of the general extensions mechanism were added to aggressive_dce but cover all passes.
2017-07-19 00:57:26 +00:00
%9 = OpTypeFunction %void
%float = OpTypeFloat 32
%v4float = OpTypeVector %float 4
%_ptr_Function_v4float = OpTypePointer Function %v4float
%_ptr_Input_v4float = OpTypePointer Input %v4float
%BaseColor = OpVariable %_ptr_Input_v4float Input
%Dead = OpVariable %_ptr_Input_v4float Input
%_ptr_Output_v4float = OpTypePointer Output %v4float
%gl_FragColor = OpVariable %_ptr_Output_v4float Output
Allow ADCE to remove dead inputs (#4629) * https://github.com/KhronosGroup/Vulkan-Docs/issues/666 clearly specified that interfaces do not require an input if there is an associated output * ADCE can now remove unused input variables (though they are kept if the preserve interfaces option is used)
2021-11-16 20:54:17 +00:00
%main = OpFunction %void None %9
Add extension whitelists to size-reduction passes. Currently only SPV_KHR_variable_pointers is disallowed in passes which do pointer analysis. Positive and negative tests of the general extensions mechanism were added to aggressive_dce but cover all passes.
2017-07-19 00:57:26 +00:00
%15 = OpLabel
%v = OpVariable %_ptr_Function_v4float Function
ADCE: Empty Loop Elimination This entirely eliminates loops which do not contain live code.
2017-10-25 00:58:48 +00:00
%dv = OpVariable %_ptr_Function_v4float Function
Add extension whitelists to size-reduction passes. Currently only SPV_KHR_variable_pointers is disallowed in passes which do pointer analysis. Positive and negative tests of the general extensions mechanism were added to aggressive_dce but cover all passes.
2017-07-19 00:57:26 +00:00
%16 = OpLoad %v4float %BaseColor
OpStore %v %16
%17 = OpLoad %v4float %Dead
OpStore %dv %17
%18 = OpLoad %v4float %v
OpStore %gl_FragColor %18
OpReturn
OpFunctionEnd
)";
Allow ADCE to remove dead inputs (#4629) * https://github.com/KhronosGroup/Vulkan-Docs/issues/666 clearly specified that interfaces do not require an input if there is an associated output * ADCE can now remove unused input variables (though they are kept if the preserve interfaces option is used)
2021-11-16 20:54:17 +00:00
SinglePassRunAndMatch<AggressiveDCEPass>(spirv, true);
Add extension whitelists to size-reduction passes. Currently only SPV_KHR_variable_pointers is disallowed in passes which do pointer analysis. Positive and negative tests of the general extensions mechanism were added to aggressive_dce but cover all passes.
2017-07-19 00:57:26 +00:00
}
Update access control lists. (#3433) This CL updates the access control lists used in SPIRV-Tools to the more descriptive allow/deny naming.
2020-06-15 17:20:40 +00:00
TEST_F(AggressiveDCETest, NoOptDenyListExtension) {
Add extension whitelists to size-reduction passes. Currently only SPV_KHR_variable_pointers is disallowed in passes which do pointer analysis. Positive and negative tests of the general extensions mechanism were added to aggressive_dce but cover all passes.
2017-07-19 00:57:26 +00:00
// #version 140
Re-format source tree - NFC. Re-formatted the source tree with the command: $ /usr/bin/clang-format -style=file -i \ $(find include source tools test utils -name '*.cpp' -or -name '*.h') This required a fix to source/val/decoration.h. It was not including spirv.h, which broke builds when the #include headers were re-ordered by clang-format.
2017-11-27 15:16:41 +00:00
//
Add extension whitelists to size-reduction passes. Currently only SPV_KHR_variable_pointers is disallowed in passes which do pointer analysis. Positive and negative tests of the general extensions mechanism were added to aggressive_dce but cover all passes.
2017-07-19 00:57:26 +00:00
// in vec4 BaseColor;
// in vec4 Dead;
Re-format source tree - NFC. Re-formatted the source tree with the command: $ /usr/bin/clang-format -style=file -i \ $(find include source tools test utils -name '*.cpp' -or -name '*.h') This required a fix to source/val/decoration.h. It was not including spirv.h, which broke builds when the #include headers were re-ordered by clang-format.
2017-11-27 15:16:41 +00:00
//
Add extension whitelists to size-reduction passes. Currently only SPV_KHR_variable_pointers is disallowed in passes which do pointer analysis. Positive and negative tests of the general extensions mechanism were added to aggressive_dce but cover all passes.
2017-07-19 00:57:26 +00:00
// void main()
// {
// vec4 v = BaseColor;
// vec4 dv = Dead;
// gl_FragColor = v;
// }
const std::string assembly =
R"(OpCapability Shader
OpExtension "SPV_KHR_variable_pointers"
%1 = OpExtInstImport "GLSL.std.450"
OpMemoryModel Logical GLSL450
OpEntryPoint Fragment %main "main" %BaseColor %Dead %gl_FragColor
OpExecutionMode %main OriginUpperLeft
OpSource GLSL 140
OpName %main "main"
OpName %v "v"
OpName %BaseColor "BaseColor"
OpName %dv "dv"
OpName %Dead "Dead"
OpName %gl_FragColor "gl_FragColor"
%void = OpTypeVoid
%9 = OpTypeFunction %void
%float = OpTypeFloat 32
%v4float = OpTypeVector %float 4
%_ptr_Function_v4float = OpTypePointer Function %v4float
%_ptr_Input_v4float = OpTypePointer Input %v4float
%BaseColor = OpVariable %_ptr_Input_v4float Input
%Dead = OpVariable %_ptr_Input_v4float Input
%_ptr_Output_v4float = OpTypePointer Output %v4float
%gl_FragColor = OpVariable %_ptr_Output_v4float Output
%main = OpFunction %void None %9
%15 = OpLabel
%v = OpVariable %_ptr_Function_v4float Function
%dv = OpVariable %_ptr_Function_v4float Function
%16 = OpLoad %v4float %BaseColor
OpStore %v %16
%17 = OpLoad %v4float %Dead
OpStore %dv %17
%18 = OpLoad %v4float %v
OpStore %gl_FragColor %18
OpReturn
OpFunctionEnd
)";
Move tests into namespaces (#1689) This CL moves the test into namespaces based on their directories.
2018-07-11 13:24:49 +00:00
SinglePassRunAndCheck<AggressiveDCEPass>(assembly, assembly, true, true);
Add extension whitelists to size-reduction passes. Currently only SPV_KHR_variable_pointers is disallowed in passes which do pointer analysis. Positive and negative tests of the general extensions mechanism were added to aggressive_dce but cover all passes.
2017-07-19 00:57:26 +00:00
}
ADCE: Add support for function calls ADCE will now generate correct code in the presence of function calls. This is needed for opaque type optimization needed by glslang. Currently all function calls are marked as live. TODO: mark calls live only if they write a non-local.
2017-08-04 21:04:37 +00:00
TEST_F(AggressiveDCETest, ElimWithCall) {
// This demonstrates that "dead" function calls are not eliminated.
// Also demonstrates that DCE will happen in presence of function call.
// #version 140
// in vec4 i1;
// in vec4 i2;
Re-format source tree - NFC. Re-formatted the source tree with the command: $ /usr/bin/clang-format -style=file -i \ $(find include source tools test utils -name '*.cpp' -or -name '*.h') This required a fix to source/val/decoration.h. It was not including spirv.h, which broke builds when the #include headers were re-ordered by clang-format.
2017-11-27 15:16:41 +00:00
//
ADCE: Add support for function calls ADCE will now generate correct code in the presence of function calls. This is needed for opaque type optimization needed by glslang. Currently all function calls are marked as live. TODO: mark calls live only if they write a non-local.
2017-08-04 21:04:37 +00:00
// void nothing(vec4 v)
// {
// }
Re-format source tree - NFC. Re-formatted the source tree with the command: $ /usr/bin/clang-format -style=file -i \ $(find include source tools test utils -name '*.cpp' -or -name '*.h') This required a fix to source/val/decoration.h. It was not including spirv.h, which broke builds when the #include headers were re-ordered by clang-format.
2017-11-27 15:16:41 +00:00
//
ADCE: Add support for function calls ADCE will now generate correct code in the presence of function calls. This is needed for opaque type optimization needed by glslang. Currently all function calls are marked as live. TODO: mark calls live only if they write a non-local.
2017-08-04 21:04:37 +00:00
// void main()
// {
// vec4 v1 = i1;
// vec4 v2 = i2;
// nothing(v1);
// gl_FragColor = vec4(0.0);
// }
Allow ADCE to remove dead inputs (#4629) * https://github.com/KhronosGroup/Vulkan-Docs/issues/666 clearly specified that interfaces do not require an input if there is an associated output * ADCE can now remove unused input variables (though they are kept if the preserve interfaces option is used)
2021-11-16 20:54:17 +00:00
const std::string text =
ADCE: Add support for function calls ADCE will now generate correct code in the presence of function calls. This is needed for opaque type optimization needed by glslang. Currently all function calls are marked as live. TODO: mark calls live only if they write a non-local.
2017-08-04 21:04:37 +00:00
R"( OpCapability Shader
%1 = OpExtInstImport "GLSL.std.450"
OpMemoryModel Logical GLSL450
OpEntryPoint Fragment %main "main" %i1 %i2 %gl_FragColor
OpExecutionMode %main OriginUpperLeft
OpSource GLSL 140
OpName %main "main"
ADCE: Empty Loop Elimination This entirely eliminates loops which do not contain live code.
2017-10-25 00:58:48 +00:00
OpName %nothing_vf4_ "nothing(vf4;"
ADCE: Add support for function calls ADCE will now generate correct code in the presence of function calls. This is needed for opaque type optimization needed by glslang. Currently all function calls are marked as live. TODO: mark calls live only if they write a non-local.
2017-08-04 21:04:37 +00:00
OpName %v "v"
OpName %v1 "v1"
OpName %i1 "i1"
OpName %v2 "v2"
OpName %i2 "i2"
ADCE: Empty Loop Elimination This entirely eliminates loops which do not contain live code.
2017-10-25 00:58:48 +00:00
OpName %param "param"
OpName %gl_FragColor "gl_FragColor"
ADCE: Add support for function calls ADCE will now generate correct code in the presence of function calls. This is needed for opaque type optimization needed by glslang. Currently all function calls are marked as live. TODO: mark calls live only if they write a non-local.
2017-08-04 21:04:37 +00:00
%void = OpTypeVoid
%12 = OpTypeFunction %void
%float = OpTypeFloat 32
%v4float = OpTypeVector %float 4
%_ptr_Function_v4float = OpTypePointer Function %v4float
%16 = OpTypeFunction %void %_ptr_Function_v4float
%_ptr_Input_v4float = OpTypePointer Input %v4float
%i1 = OpVariable %_ptr_Input_v4float Input
%i2 = OpVariable %_ptr_Input_v4float Input
%_ptr_Output_v4float = OpTypePointer Output %v4float
%gl_FragColor = OpVariable %_ptr_Output_v4float Output
%float_0 = OpConstant %float 0
ADCE: Empty Loop Elimination This entirely eliminates loops which do not contain live code.
2017-10-25 00:58:48 +00:00
%20 = OpConstantComposite %v4float %float_0 %float_0 %float_0 %float_0
Allow ADCE to remove dead inputs (#4629) * https://github.com/KhronosGroup/Vulkan-Docs/issues/666 clearly specified that interfaces do not require an input if there is an associated output * ADCE can now remove unused input variables (though they are kept if the preserve interfaces option is used)
2021-11-16 20:54:17 +00:00
%main = OpFunction %void None %12
ADCE: Add support for function calls ADCE will now generate correct code in the presence of function calls. This is needed for opaque type optimization needed by glslang. Currently all function calls are marked as live. TODO: mark calls live only if they write a non-local.
2017-08-04 21:04:37 +00:00
%21 = OpLabel
ADCE: Empty Loop Elimination This entirely eliminates loops which do not contain live code.
2017-10-25 00:58:48 +00:00
%v1 = OpVariable %_ptr_Function_v4float Function
%v2 = OpVariable %_ptr_Function_v4float Function
ADCE: Add support for function calls ADCE will now generate correct code in the presence of function calls. This is needed for opaque type optimization needed by glslang. Currently all function calls are marked as live. TODO: mark calls live only if they write a non-local.
2017-08-04 21:04:37 +00:00
%param = OpVariable %_ptr_Function_v4float Function
%22 = OpLoad %v4float %i1
OpStore %v1 %22
Allow ADCE to remove dead inputs (#4629) * https://github.com/KhronosGroup/Vulkan-Docs/issues/666 clearly specified that interfaces do not require an input if there is an associated output * ADCE can now remove unused input variables (though they are kept if the preserve interfaces option is used)
2021-11-16 20:54:17 +00:00
; CHECK-NOT: OpLoad %v4float %i2
ADCE: Add support for function calls ADCE will now generate correct code in the presence of function calls. This is needed for opaque type optimization needed by glslang. Currently all function calls are marked as live. TODO: mark calls live only if they write a non-local.
2017-08-04 21:04:37 +00:00
%23 = OpLoad %v4float %i2
Allow ADCE to remove dead inputs (#4629) * https://github.com/KhronosGroup/Vulkan-Docs/issues/666 clearly specified that interfaces do not require an input if there is an associated output * ADCE can now remove unused input variables (though they are kept if the preserve interfaces option is used)
2021-11-16 20:54:17 +00:00
; CHECK-NOT: OpStore %v2
ADCE: Add support for function calls ADCE will now generate correct code in the presence of function calls. This is needed for opaque type optimization needed by glslang. Currently all function calls are marked as live. TODO: mark calls live only if they write a non-local.
2017-08-04 21:04:37 +00:00
OpStore %v2 %23
%24 = OpLoad %v4float %v1
OpStore %param %24
Allow ADCE to remove dead inputs (#4629) * https://github.com/KhronosGroup/Vulkan-Docs/issues/666 clearly specified that interfaces do not require an input if there is an associated output * ADCE can now remove unused input variables (though they are kept if the preserve interfaces option is used)
2021-11-16 20:54:17 +00:00
; CHECK: OpFunctionCall %void %nothing_vf4_
ADCE: Add support for function calls ADCE will now generate correct code in the presence of function calls. This is needed for opaque type optimization needed by glslang. Currently all function calls are marked as live. TODO: mark calls live only if they write a non-local.
2017-08-04 21:04:37 +00:00
%25 = OpFunctionCall %void %nothing_vf4_ %param
OpStore %gl_FragColor %20
OpReturn
OpFunctionEnd
Allow ADCE to remove dead inputs (#4629) * https://github.com/KhronosGroup/Vulkan-Docs/issues/666 clearly specified that interfaces do not require an input if there is an associated output * ADCE can now remove unused input variables (though they are kept if the preserve interfaces option is used)
2021-11-16 20:54:17 +00:00
; CHECK: %nothing_vf4_ = OpFunction
ADCE: Add support for function calls ADCE will now generate correct code in the presence of function calls. This is needed for opaque type optimization needed by glslang. Currently all function calls are marked as live. TODO: mark calls live only if they write a non-local.
2017-08-04 21:04:37 +00:00
%nothing_vf4_ = OpFunction %void None %16
%v = OpFunctionParameter %_ptr_Function_v4float
%26 = OpLabel
OpReturn
OpFunctionEnd
)";
Allow ADCE to remove dead inputs (#4629) * https://github.com/KhronosGroup/Vulkan-Docs/issues/666 clearly specified that interfaces do not require an input if there is an associated output * ADCE can now remove unused input variables (though they are kept if the preserve interfaces option is used)
2021-11-16 20:54:17 +00:00
SinglePassRunAndMatch<AggressiveDCEPass>(text, true);
ADCE: Add support for function calls ADCE will now generate correct code in the presence of function calls. This is needed for opaque type optimization needed by glslang. Currently all function calls are marked as live. TODO: mark calls live only if they write a non-local.
2017-08-04 21:04:37 +00:00
}
Opt: Have "size" passes process full entry point call tree. Includes code to deal correctly with OpFunctionParameter. This is needed by opaque propagation which may not exhaustively inline entry point functions. Adds ProcessEntryPointCallTree: a method to do work on the functions in the entry point call trees in a deterministic order.
2017-08-10 22:42:16 +00:00
TEST_F(AggressiveDCETest, NoParamElim) {
// This demonstrates that unused parameters are not eliminated, but
// dead uses of them are.
// #version 140
Re-format source tree - NFC. Re-formatted the source tree with the command: $ /usr/bin/clang-format -style=file -i \ $(find include source tools test utils -name '*.cpp' -or -name '*.h') This required a fix to source/val/decoration.h. It was not including spirv.h, which broke builds when the #include headers were re-ordered by clang-format.
2017-11-27 15:16:41 +00:00
//
Opt: Have "size" passes process full entry point call tree. Includes code to deal correctly with OpFunctionParameter. This is needed by opaque propagation which may not exhaustively inline entry point functions. Adds ProcessEntryPointCallTree: a method to do work on the functions in the entry point call trees in a deterministic order.
2017-08-10 22:42:16 +00:00
// in vec4 BaseColor;
Re-format source tree - NFC. Re-formatted the source tree with the command: $ /usr/bin/clang-format -style=file -i \ $(find include source tools test utils -name '*.cpp' -or -name '*.h') This required a fix to source/val/decoration.h. It was not including spirv.h, which broke builds when the #include headers were re-ordered by clang-format.
2017-11-27 15:16:41 +00:00
//
Opt: Have "size" passes process full entry point call tree. Includes code to deal correctly with OpFunctionParameter. This is needed by opaque propagation which may not exhaustively inline entry point functions. Adds ProcessEntryPointCallTree: a method to do work on the functions in the entry point call trees in a deterministic order.
2017-08-10 22:42:16 +00:00
// vec4 foo(vec4 v1, vec4 v2)
// {
// vec4 t = -v1;
// return v2;
// }
Re-format source tree - NFC. Re-formatted the source tree with the command: $ /usr/bin/clang-format -style=file -i \ $(find include source tools test utils -name '*.cpp' -or -name '*.h') This required a fix to source/val/decoration.h. It was not including spirv.h, which broke builds when the #include headers were re-ordered by clang-format.
2017-11-27 15:16:41 +00:00
//
Opt: Have "size" passes process full entry point call tree. Includes code to deal correctly with OpFunctionParameter. This is needed by opaque propagation which may not exhaustively inline entry point functions. Adds ProcessEntryPointCallTree: a method to do work on the functions in the entry point call trees in a deterministic order.
2017-08-10 22:42:16 +00:00
// void main()
// {
// vec4 dead;
// gl_FragColor = foo(dead, BaseColor);
// }
const std::string defs_before =
R"(OpCapability Shader
%1 = OpExtInstImport "GLSL.std.450"
OpMemoryModel Logical GLSL450
OpEntryPoint Fragment %main "main" %gl_FragColor %BaseColor
OpExecutionMode %main OriginUpperLeft
OpSource GLSL 140
OpName %main "main"
ADCE: Empty Loop Elimination This entirely eliminates loops which do not contain live code.
2017-10-25 00:58:48 +00:00
OpName %foo_vf4_vf4_ "foo(vf4;vf4;"
Opt: Have "size" passes process full entry point call tree. Includes code to deal correctly with OpFunctionParameter. This is needed by opaque propagation which may not exhaustively inline entry point functions. Adds ProcessEntryPointCallTree: a method to do work on the functions in the entry point call trees in a deterministic order.
2017-08-10 22:42:16 +00:00
OpName %v1 "v1"
OpName %v2 "v2"
OpName %t "t"
ADCE: Empty Loop Elimination This entirely eliminates loops which do not contain live code.
2017-10-25 00:58:48 +00:00
OpName %gl_FragColor "gl_FragColor"
Opt: Have "size" passes process full entry point call tree. Includes code to deal correctly with OpFunctionParameter. This is needed by opaque propagation which may not exhaustively inline entry point functions. Adds ProcessEntryPointCallTree: a method to do work on the functions in the entry point call trees in a deterministic order.
2017-08-10 22:42:16 +00:00
OpName %dead "dead"
OpName %BaseColor "BaseColor"
ADCE: Empty Loop Elimination This entirely eliminates loops which do not contain live code.
2017-10-25 00:58:48 +00:00
OpName %param "param"
Opt: Have "size" passes process full entry point call tree. Includes code to deal correctly with OpFunctionParameter. This is needed by opaque propagation which may not exhaustively inline entry point functions. Adds ProcessEntryPointCallTree: a method to do work on the functions in the entry point call trees in a deterministic order.
2017-08-10 22:42:16 +00:00
OpName %param_0 "param"
%void = OpTypeVoid
%13 = OpTypeFunction %void
%float = OpTypeFloat 32
%v4float = OpTypeVector %float 4
%_ptr_Function_v4float = OpTypePointer Function %v4float
%17 = OpTypeFunction %v4float %_ptr_Function_v4float %_ptr_Function_v4float
%_ptr_Output_v4float = OpTypePointer Output %v4float
%gl_FragColor = OpVariable %_ptr_Output_v4float Output
%_ptr_Input_v4float = OpTypePointer Input %v4float
%BaseColor = OpVariable %_ptr_Input_v4float Input
%main = OpFunction %void None %13
%20 = OpLabel
%dead = OpVariable %_ptr_Function_v4float Function
%param = OpVariable %_ptr_Function_v4float Function
%param_0 = OpVariable %_ptr_Function_v4float Function
%21 = OpLoad %v4float %dead
OpStore %param %21
%22 = OpLoad %v4float %BaseColor
OpStore %param_0 %22
%23 = OpFunctionCall %v4float %foo_vf4_vf4_ %param %param_0
OpStore %gl_FragColor %23
OpReturn
OpFunctionEnd
)";
const std::string defs_after =
R"(OpCapability Shader
%1 = OpExtInstImport "GLSL.std.450"
OpMemoryModel Logical GLSL450
OpEntryPoint Fragment %main "main" %gl_FragColor %BaseColor
OpExecutionMode %main OriginUpperLeft
OpSource GLSL 140
OpName %main "main"
OpName %foo_vf4_vf4_ "foo(vf4;vf4;"
OpName %v1 "v1"
OpName %v2 "v2"
OpName %gl_FragColor "gl_FragColor"
OpName %dead "dead"
OpName %BaseColor "BaseColor"
OpName %param "param"
OpName %param_0 "param"
%void = OpTypeVoid
%13 = OpTypeFunction %void
%float = OpTypeFloat 32
%v4float = OpTypeVector %float 4
%_ptr_Function_v4float = OpTypePointer Function %v4float
%17 = OpTypeFunction %v4float %_ptr_Function_v4float %_ptr_Function_v4float
%_ptr_Output_v4float = OpTypePointer Output %v4float
%gl_FragColor = OpVariable %_ptr_Output_v4float Output
%_ptr_Input_v4float = OpTypePointer Input %v4float
%BaseColor = OpVariable %_ptr_Input_v4float Input
%main = OpFunction %void None %13
%20 = OpLabel
%dead = OpVariable %_ptr_Function_v4float Function
%param = OpVariable %_ptr_Function_v4float Function
%param_0 = OpVariable %_ptr_Function_v4float Function
%21 = OpLoad %v4float %dead
OpStore %param %21
%22 = OpLoad %v4float %BaseColor
OpStore %param_0 %22
%23 = OpFunctionCall %v4float %foo_vf4_vf4_ %param %param_0
OpStore %gl_FragColor %23
OpReturn
OpFunctionEnd
)";
const std::string func_before =
R"(%foo_vf4_vf4_ = OpFunction %v4float None %17
%v1 = OpFunctionParameter %_ptr_Function_v4float
%v2 = OpFunctionParameter %_ptr_Function_v4float
%24 = OpLabel
%t = OpVariable %_ptr_Function_v4float Function
ADCE: Empty Loop Elimination This entirely eliminates loops which do not contain live code.
2017-10-25 00:58:48 +00:00
%25 = OpLoad %v4float %v1
Opt: Have "size" passes process full entry point call tree. Includes code to deal correctly with OpFunctionParameter. This is needed by opaque propagation which may not exhaustively inline entry point functions. Adds ProcessEntryPointCallTree: a method to do work on the functions in the entry point call trees in a deterministic order.
2017-08-10 22:42:16 +00:00
%26 = OpFNegate %v4float %25
ADCE: Empty Loop Elimination This entirely eliminates loops which do not contain live code.
2017-10-25 00:58:48 +00:00
OpStore %t %26
%27 = OpLoad %v4float %v2
Opt: Have "size" passes process full entry point call tree. Includes code to deal correctly with OpFunctionParameter. This is needed by opaque propagation which may not exhaustively inline entry point functions. Adds ProcessEntryPointCallTree: a method to do work on the functions in the entry point call trees in a deterministic order.
2017-08-10 22:42:16 +00:00
OpReturnValue %27
OpFunctionEnd
)";
const std::string func_after =
R"(%foo_vf4_vf4_ = OpFunction %v4float None %17
%v1 = OpFunctionParameter %_ptr_Function_v4float
%v2 = OpFunctionParameter %_ptr_Function_v4float
%24 = OpLabel
%27 = OpLoad %v4float %v2
OpReturnValue %27
OpFunctionEnd
)";
Move tests into namespaces (#1689) This CL moves the test into namespaces based on their directories.
2018-07-11 13:24:49 +00:00
SinglePassRunAndCheck<AggressiveDCEPass>(defs_before + func_before,
defs_after + func_after, true, true);
Opt: Have "size" passes process full entry point call tree. Includes code to deal correctly with OpFunctionParameter. This is needed by opaque propagation which may not exhaustively inline entry point functions. Adds ProcessEntryPointCallTree: a method to do work on the functions in the entry point call trees in a deterministic order.
2017-08-10 22:42:16 +00:00
}
Opt: Add opaque tests
2017-08-15 21:54:41 +00:00
TEST_F(AggressiveDCETest, ElimOpaque) {
// SPIR-V not representable from GLSL; not generatable from HLSL
// for the moment.
const std::string defs_before =
R"(OpCapability Shader
%1 = OpExtInstImport "GLSL.std.450"
OpMemoryModel Logical GLSL450
OpEntryPoint Fragment %main "main" %outColor %texCoords
OpExecutionMode %main OriginUpperLeft
OpSource GLSL 140
OpName %main "main"
OpName %S_t "S_t"
OpMemberName %S_t 0 "v0"
OpMemberName %S_t 1 "v1"
OpMemberName %S_t 2 "smp"
OpName %outColor "outColor"
OpName %sampler15 "sampler15"
OpName %s0 "s0"
OpName %texCoords "texCoords"
OpDecorate %sampler15 DescriptorSet 0
%void = OpTypeVoid
%9 = OpTypeFunction %void
%float = OpTypeFloat 32
%v2float = OpTypeVector %float 2
%v4float = OpTypeVector %float 4
%_ptr_Output_v4float = OpTypePointer Output %v4float
%outColor = OpVariable %_ptr_Output_v4float Output
%14 = OpTypeImage %float 2D 0 0 0 1 Unknown
%15 = OpTypeSampledImage %14
%S_t = OpTypeStruct %v2float %v2float %15
%_ptr_Function_S_t = OpTypePointer Function %S_t
%17 = OpTypeFunction %void %_ptr_Function_S_t
%_ptr_UniformConstant_15 = OpTypePointer UniformConstant %15
%_ptr_Function_15 = OpTypePointer Function %15
%sampler15 = OpVariable %_ptr_UniformConstant_15 UniformConstant
%int = OpTypeInt 32 1
%int_0 = OpConstant %int 0
%int_2 = OpConstant %int 2
%_ptr_Function_v2float = OpTypePointer Function %v2float
%_ptr_Input_v2float = OpTypePointer Input %v2float
%texCoords = OpVariable %_ptr_Input_v2float Input
)";
const std::string defs_after =
R"(OpCapability Shader
%1 = OpExtInstImport "GLSL.std.450"
OpMemoryModel Logical GLSL450
OpEntryPoint Fragment %main "main" %outColor %texCoords
OpExecutionMode %main OriginUpperLeft
OpSource GLSL 140
OpName %main "main"
OpName %outColor "outColor"
OpName %sampler15 "sampler15"
OpName %texCoords "texCoords"
OpDecorate %sampler15 DescriptorSet 0
%void = OpTypeVoid
%9 = OpTypeFunction %void
%float = OpTypeFloat 32
%v2float = OpTypeVector %float 2
%v4float = OpTypeVector %float 4
%_ptr_Output_v4float = OpTypePointer Output %v4float
%outColor = OpVariable %_ptr_Output_v4float Output
%14 = OpTypeImage %float 2D 0 0 0 1 Unknown
%15 = OpTypeSampledImage %14
%_ptr_UniformConstant_15 = OpTypePointer UniformConstant %15
%sampler15 = OpVariable %_ptr_UniformConstant_15 UniformConstant
%_ptr_Input_v2float = OpTypePointer Input %v2float
%texCoords = OpVariable %_ptr_Input_v2float Input
)";
const std::string func_before =
R"(%main = OpFunction %void None %9
%25 = OpLabel
ADCE: Empty Loop Elimination This entirely eliminates loops which do not contain live code.
2017-10-25 00:58:48 +00:00
%s0 = OpVariable %_ptr_Function_S_t Function
Opt: Add opaque tests
2017-08-15 21:54:41 +00:00
%26 = OpLoad %v2float %texCoords
ADCE: Empty Loop Elimination This entirely eliminates loops which do not contain live code.
2017-10-25 00:58:48 +00:00
%27 = OpLoad %S_t %s0
Opt: Add opaque tests
2017-08-15 21:54:41 +00:00
%28 = OpCompositeInsert %S_t %26 %27 0
%29 = OpLoad %15 %sampler15
%30 = OpCompositeInsert %S_t %29 %28 2
OpStore %s0 %30
%31 = OpImageSampleImplicitLod %v4float %29 %26
OpStore %outColor %31
OpReturn
OpFunctionEnd
)";
const std::string func_after =
Re-format source tree - NFC. Re-formatted the source tree with the command: $ /usr/bin/clang-format -style=file -i \ $(find include source tools test utils -name '*.cpp' -or -name '*.h') This required a fix to source/val/decoration.h. It was not including spirv.h, which broke builds when the #include headers were re-ordered by clang-format.
2017-11-27 15:16:41 +00:00
R"(%main = OpFunction %void None %9
Opt: Add opaque tests
2017-08-15 21:54:41 +00:00
%25 = OpLabel
%26 = OpLoad %v2float %texCoords
%29 = OpLoad %15 %sampler15
%31 = OpImageSampleImplicitLod %v4float %29 %26
OpStore %outColor %31
OpReturn
OpFunctionEnd
)";
Move tests into namespaces (#1689) This CL moves the test into namespaces based on their directories.
2018-07-11 13:24:49 +00:00
SinglePassRunAndCheck<AggressiveDCEPass>(defs_before + func_before,
defs_after + func_after, true, true);
Opt: Add opaque tests
2017-08-15 21:54:41 +00:00
}
AggressiveDCE: Fix to not treat parameter memory refs as local This fixes a bug that incorrectly deletes stores to parameters, which can be used to return values from functions.
2017-09-29 22:22:39 +00:00
TEST_F(AggressiveDCETest, NoParamStoreElim) {
// Should not eliminate stores to params
//
// #version 450
Re-format source tree - NFC. Re-formatted the source tree with the command: $ /usr/bin/clang-format -style=file -i \ $(find include source tools test utils -name '*.cpp' -or -name '*.h') This required a fix to source/val/decoration.h. It was not including spirv.h, which broke builds when the #include headers were re-ordered by clang-format.
2017-11-27 15:16:41 +00:00
//
AggressiveDCE: Fix to not treat parameter memory refs as local This fixes a bug that incorrectly deletes stores to parameters, which can be used to return values from functions.
2017-09-29 22:22:39 +00:00
// layout(location = 0) in vec4 BaseColor;
// layout(location = 0) out vec4 OutColor;
Re-format source tree - NFC. Re-formatted the source tree with the command: $ /usr/bin/clang-format -style=file -i \ $(find include source tools test utils -name '*.cpp' -or -name '*.h') This required a fix to source/val/decoration.h. It was not including spirv.h, which broke builds when the #include headers were re-ordered by clang-format.
2017-11-27 15:16:41 +00:00
//
AggressiveDCE: Fix to not treat parameter memory refs as local This fixes a bug that incorrectly deletes stores to parameters, which can be used to return values from functions.
2017-09-29 22:22:39 +00:00
// void foo(in vec4 v1, out vec4 v2)
// {
// v2 = -v1;
// }
Re-format source tree - NFC. Re-formatted the source tree with the command: $ /usr/bin/clang-format -style=file -i \ $(find include source tools test utils -name '*.cpp' -or -name '*.h') This required a fix to source/val/decoration.h. It was not including spirv.h, which broke builds when the #include headers were re-ordered by clang-format.
2017-11-27 15:16:41 +00:00
//
AggressiveDCE: Fix to not treat parameter memory refs as local This fixes a bug that incorrectly deletes stores to parameters, which can be used to return values from functions.
2017-09-29 22:22:39 +00:00
// void main()
// {
// foo(BaseColor, OutColor);
// }
const std::string assembly =
R"(OpCapability Shader
%1 = OpExtInstImport "GLSL.std.450"
OpMemoryModel Logical GLSL450
OpEntryPoint Fragment %main "main" %BaseColor %OutColor
OpExecutionMode %main OriginUpperLeft
OpSource GLSL 450
OpName %main "main"
OpName %foo_vf4_vf4_ "foo(vf4;vf4;"
OpName %v1 "v1"
OpName %v2 "v2"
OpName %BaseColor "BaseColor"
OpName %OutColor "OutColor"
OpName %param "param"
OpName %param_0 "param"
OpDecorate %BaseColor Location 0
OpDecorate %OutColor Location 0
%void = OpTypeVoid
%11 = OpTypeFunction %void
%float = OpTypeFloat 32
%v4float = OpTypeVector %float 4
%_ptr_Function_v4float = OpTypePointer Function %v4float
%15 = OpTypeFunction %void %_ptr_Function_v4float %_ptr_Function_v4float
%_ptr_Input_v4float = OpTypePointer Input %v4float
%BaseColor = OpVariable %_ptr_Input_v4float Input
%_ptr_Output_v4float = OpTypePointer Output %v4float
%OutColor = OpVariable %_ptr_Output_v4float Output
%main = OpFunction %void None %11
%18 = OpLabel
%param = OpVariable %_ptr_Function_v4float Function
%param_0 = OpVariable %_ptr_Function_v4float Function
%19 = OpLoad %v4float %BaseColor
OpStore %param %19
%20 = OpFunctionCall %void %foo_vf4_vf4_ %param %param_0
%21 = OpLoad %v4float %param_0
OpStore %OutColor %21
OpReturn
OpFunctionEnd
%foo_vf4_vf4_ = OpFunction %void None %15
%v1 = OpFunctionParameter %_ptr_Function_v4float
%v2 = OpFunctionParameter %_ptr_Function_v4float
%22 = OpLabel
%23 = OpLoad %v4float %v1
%24 = OpFNegate %v4float %23
OpStore %v2 %24
OpReturn
OpFunctionEnd
)";
Move tests into namespaces (#1689) This CL moves the test into namespaces based on their directories.
2018-07-11 13:24:49 +00:00
SinglePassRunAndCheck<AggressiveDCEPass>(assembly, assembly, true, true);
AggressiveDCE: Fix to not treat parameter memory refs as local This fixes a bug that incorrectly deletes stores to parameters, which can be used to return values from functions.
2017-09-29 22:22:39 +00:00
}
ADCE: Treat privates like locals in entry point with no calls This is needed for ongoing legalization of HLSL. It allows removal of accesses to textures/buffers that are not used.
2017-10-10 20:35:53 +00:00
TEST_F(AggressiveDCETest, PrivateStoreElimInEntryNoCalls) {
// Eliminate stores to private in entry point with no calls
// Note: Not legal GLSL
//
// layout(location = 0) in vec4 BaseColor;
// layout(location = 1) in vec4 Dead;
// layout(location = 0) out vec4 OutColor;
//
// private vec4 dv;
Re-format source tree - NFC. Re-formatted the source tree with the command: $ /usr/bin/clang-format -style=file -i \ $(find include source tools test utils -name '*.cpp' -or -name '*.h') This required a fix to source/val/decoration.h. It was not including spirv.h, which broke builds when the #include headers were re-ordered by clang-format.
2017-11-27 15:16:41 +00:00
//
ADCE: Treat privates like locals in entry point with no calls This is needed for ongoing legalization of HLSL. It allows removal of accesses to textures/buffers that are not used.
2017-10-10 20:35:53 +00:00
// void main()
// {
// vec4 v = BaseColor;
// dv = Dead;
// OutColor = v;
// }
Allow ADCE to remove dead inputs (#4629) * https://github.com/KhronosGroup/Vulkan-Docs/issues/666 clearly specified that interfaces do not require an input if there is an associated output * ADCE can now remove unused input variables (though they are kept if the preserve interfaces option is used)
2021-11-16 20:54:17 +00:00
const std::string spirv =
ADCE: Treat privates like locals in entry point with no calls This is needed for ongoing legalization of HLSL. It allows removal of accesses to textures/buffers that are not used.
2017-10-10 20:35:53 +00:00
R"(OpCapability Shader
%1 = OpExtInstImport "GLSL.std.450"
OpMemoryModel Logical GLSL450
OpEntryPoint Fragment %main "main" %BaseColor %Dead %OutColor
OpExecutionMode %main OriginUpperLeft
OpSource GLSL 450
OpName %main "main"
OpName %v "v"
OpName %BaseColor "BaseColor"
Allow ADCE to remove dead inputs (#4629) * https://github.com/KhronosGroup/Vulkan-Docs/issues/666 clearly specified that interfaces do not require an input if there is an associated output * ADCE can now remove unused input variables (though they are kept if the preserve interfaces option is used)
2021-11-16 20:54:17 +00:00
; CHECK-NOT: OpName %dv "dv"
ADCE: Treat privates like locals in entry point with no calls This is needed for ongoing legalization of HLSL. It allows removal of accesses to textures/buffers that are not used.
2017-10-10 20:35:53 +00:00
OpName %dv "dv"
OpName %Dead "Dead"
OpName %OutColor "OutColor"
OpDecorate %BaseColor Location 0
OpDecorate %Dead Location 1
OpDecorate %OutColor Location 0
%void = OpTypeVoid
%9 = OpTypeFunction %void
%float = OpTypeFloat 32
%v4float = OpTypeVector %float 4
%_ptr_Function_v4float = OpTypePointer Function %v4float
Allow ADCE to remove dead inputs (#4629) * https://github.com/KhronosGroup/Vulkan-Docs/issues/666 clearly specified that interfaces do not require an input if there is an associated output * ADCE can now remove unused input variables (though they are kept if the preserve interfaces option is used)
2021-11-16 20:54:17 +00:00
; CHECK-NOT: OpTypePointer Private
ADCE: Treat privates like locals in entry point with no calls This is needed for ongoing legalization of HLSL. It allows removal of accesses to textures/buffers that are not used.
2017-10-10 20:35:53 +00:00
%_ptr_Private_v4float = OpTypePointer Private %v4float
%_ptr_Input_v4float = OpTypePointer Input %v4float
%BaseColor = OpVariable %_ptr_Input_v4float Input
%Dead = OpVariable %_ptr_Input_v4float Input
%_ptr_Output_v4float = OpTypePointer Output %v4float
Allow ADCE to remove dead inputs (#4629) * https://github.com/KhronosGroup/Vulkan-Docs/issues/666 clearly specified that interfaces do not require an input if there is an associated output * ADCE can now remove unused input variables (though they are kept if the preserve interfaces option is used)
2021-11-16 20:54:17 +00:00
; CHECK-NOT: %dv = OpVariable
%dv = OpVariable %_ptr_Private_v4float Private
Adding additional functionality to ADCE. Modified ADCE to remove dead globals. * Entry point and execution mode instructions are marked as alive * Reachable functions and their parameters are marked as alive * Instruction deletion now deferred until the end of the pass * Eliminated dead insts set, added IsDead to calculate that value instead * Ported applicable dead variable elimination tests * Ported dead constant elim tests Added dead function elimination to ADCE * ported dead function elim tests Added handling of decoration groups in ADCE * Uses a custom sorter to traverse decorations in a specific order * Simplifies necessary checks Updated -O and -Os pass lists.
2017-12-18 17:13:10 +00:00
%OutColor = OpVariable %_ptr_Output_v4float Output
Allow ADCE to remove dead inputs (#4629) * https://github.com/KhronosGroup/Vulkan-Docs/issues/666 clearly specified that interfaces do not require an input if there is an associated output * ADCE can now remove unused input variables (though they are kept if the preserve interfaces option is used)
2021-11-16 20:54:17 +00:00
%main = OpFunction %void None %9
ADCE: Treat privates like locals in entry point with no calls This is needed for ongoing legalization of HLSL. It allows removal of accesses to textures/buffers that are not used.
2017-10-10 20:35:53 +00:00
%16 = OpLabel
%v = OpVariable %_ptr_Function_v4float Function
%17 = OpLoad %v4float %BaseColor
OpStore %v %17
%18 = OpLoad %v4float %Dead
Allow ADCE to remove dead inputs (#4629) * https://github.com/KhronosGroup/Vulkan-Docs/issues/666 clearly specified that interfaces do not require an input if there is an associated output * ADCE can now remove unused input variables (though they are kept if the preserve interfaces option is used)
2021-11-16 20:54:17 +00:00
; CHECK-NOT: OpStore %dv
ADCE: Treat privates like locals in entry point with no calls This is needed for ongoing legalization of HLSL. It allows removal of accesses to textures/buffers that are not used.
2017-10-10 20:35:53 +00:00
OpStore %dv %18
%19 = OpLoad %v4float %v
%20 = OpFNegate %v4float %19
OpStore %OutColor %20
OpReturn
OpFunctionEnd
)";
Allow ADCE to remove dead inputs (#4629) * https://github.com/KhronosGroup/Vulkan-Docs/issues/666 clearly specified that interfaces do not require an input if there is an associated output * ADCE can now remove unused input variables (though they are kept if the preserve interfaces option is used)
2021-11-16 20:54:17 +00:00
SinglePassRunAndMatch<AggressiveDCEPass>(spirv, true);
ADCE: Treat privates like locals in entry point with no calls This is needed for ongoing legalization of HLSL. It allows removal of accesses to textures/buffers that are not used.
2017-10-10 20:35:53 +00:00
}
TEST_F(AggressiveDCETest, NoPrivateStoreElimIfLoad) {
// Should not eliminate stores to private when there is a load
// Note: Not legal GLSL
//
// #version 450
Re-format source tree - NFC. Re-formatted the source tree with the command: $ /usr/bin/clang-format -style=file -i \ $(find include source tools test utils -name '*.cpp' -or -name '*.h') This required a fix to source/val/decoration.h. It was not including spirv.h, which broke builds when the #include headers were re-ordered by clang-format.
2017-11-27 15:16:41 +00:00
//
ADCE: Treat privates like locals in entry point with no calls This is needed for ongoing legalization of HLSL. It allows removal of accesses to textures/buffers that are not used.
2017-10-10 20:35:53 +00:00
// layout(location = 0) in vec4 BaseColor;
// layout(location = 0) out vec4 OutColor;
//
// private vec4 pv;
Re-format source tree - NFC. Re-formatted the source tree with the command: $ /usr/bin/clang-format -style=file -i \ $(find include source tools test utils -name '*.cpp' -or -name '*.h') This required a fix to source/val/decoration.h. It was not including spirv.h, which broke builds when the #include headers were re-ordered by clang-format.
2017-11-27 15:16:41 +00:00
//
ADCE: Treat privates like locals in entry point with no calls This is needed for ongoing legalization of HLSL. It allows removal of accesses to textures/buffers that are not used.
2017-10-10 20:35:53 +00:00
// void main()
// {
// pv = BaseColor;
// OutColor = pv;
// }
const std::string assembly =
R"(OpCapability Shader
%1 = OpExtInstImport "GLSL.std.450"
OpMemoryModel Logical GLSL450
OpEntryPoint Fragment %main "main" %BaseColor %OutColor
OpExecutionMode %main OriginUpperLeft
OpSource GLSL 450
OpName %main "main"
OpName %pv "pv"
OpName %BaseColor "BaseColor"
OpName %OutColor "OutColor"
OpDecorate %BaseColor Location 0
OpDecorate %OutColor Location 0
%void = OpTypeVoid
%7 = OpTypeFunction %void
%float = OpTypeFloat 32
%v4float = OpTypeVector %float 4
%_ptr_Private_v4float = OpTypePointer Private %v4float
%_ptr_Input_v4float = OpTypePointer Input %v4float
%BaseColor = OpVariable %_ptr_Input_v4float Input
%_ptr_Output_v4float = OpTypePointer Output %v4float
%OutColor = OpVariable %_ptr_Output_v4float Output
%pv = OpVariable %_ptr_Private_v4float Private
%main = OpFunction %void None %7
%13 = OpLabel
%14 = OpLoad %v4float %BaseColor
OpStore %pv %14
%15 = OpLoad %v4float %pv
%16 = OpFNegate %v4float %15
OpStore %OutColor %16
OpReturn
OpFunctionEnd
)";
Move tests into namespaces (#1689) This CL moves the test into namespaces based on their directories.
2018-07-11 13:24:49 +00:00
SinglePassRunAndCheck<AggressiveDCEPass>(assembly, assembly, true, true);
ADCE: Treat privates like locals in entry point with no calls This is needed for ongoing legalization of HLSL. It allows removal of accesses to textures/buffers that are not used.
2017-10-10 20:35:53 +00:00
}
TEST_F(AggressiveDCETest, NoPrivateStoreElimWithCall) {
// Should not eliminate stores to private when function contains call
// Note: Not legal GLSL
//
// #version 450
Re-format source tree - NFC. Re-formatted the source tree with the command: $ /usr/bin/clang-format -style=file -i \ $(find include source tools test utils -name '*.cpp' -or -name '*.h') This required a fix to source/val/decoration.h. It was not including spirv.h, which broke builds when the #include headers were re-ordered by clang-format.
2017-11-27 15:16:41 +00:00
//
ADCE: Treat privates like locals in entry point with no calls This is needed for ongoing legalization of HLSL. It allows removal of accesses to textures/buffers that are not used.
2017-10-10 20:35:53 +00:00
// layout(location = 0) in vec4 BaseColor;
// layout(location = 0) out vec4 OutColor;
//
// private vec4 v1;
Re-format source tree - NFC. Re-formatted the source tree with the command: $ /usr/bin/clang-format -style=file -i \ $(find include source tools test utils -name '*.cpp' -or -name '*.h') This required a fix to source/val/decoration.h. It was not including spirv.h, which broke builds when the #include headers were re-ordered by clang-format.
2017-11-27 15:16:41 +00:00
//
ADCE: Treat privates like locals in entry point with no calls This is needed for ongoing legalization of HLSL. It allows removal of accesses to textures/buffers that are not used.
2017-10-10 20:35:53 +00:00
// void foo()
// {
// OutColor = -v1;
// }
Re-format source tree - NFC. Re-formatted the source tree with the command: $ /usr/bin/clang-format -style=file -i \ $(find include source tools test utils -name '*.cpp' -or -name '*.h') This required a fix to source/val/decoration.h. It was not including spirv.h, which broke builds when the #include headers were re-ordered by clang-format.
2017-11-27 15:16:41 +00:00
//
ADCE: Treat privates like locals in entry point with no calls This is needed for ongoing legalization of HLSL. It allows removal of accesses to textures/buffers that are not used.
2017-10-10 20:35:53 +00:00
// void main()
// {
// v1 = BaseColor;
// foo();
// }
const std::string assembly =
R"(OpCapability Shader
%1 = OpExtInstImport "GLSL.std.450"
OpMemoryModel Logical GLSL450
OpEntryPoint Fragment %main "main" %OutColor %BaseColor
OpExecutionMode %main OriginUpperLeft
OpSource GLSL 450
OpName %main "main"
OpName %foo_ "foo("
OpName %OutColor "OutColor"
OpName %v1 "v1"
OpName %BaseColor "BaseColor"
OpDecorate %OutColor Location 0
OpDecorate %BaseColor Location 0
%void = OpTypeVoid
%8 = OpTypeFunction %void
%float = OpTypeFloat 32
%v4float = OpTypeVector %float 4
%_ptr_Output_v4float = OpTypePointer Output %v4float
%OutColor = OpVariable %_ptr_Output_v4float Output
%_ptr_Private_v4float = OpTypePointer Private %v4float
%_ptr_Input_v4float = OpTypePointer Input %v4float
%v1 = OpVariable %_ptr_Private_v4float Private
%BaseColor = OpVariable %_ptr_Input_v4float Input
%main = OpFunction %void None %8
%14 = OpLabel
%15 = OpLoad %v4float %BaseColor
OpStore %v1 %15
%16 = OpFunctionCall %void %foo_
OpReturn
OpFunctionEnd
%foo_ = OpFunction %void None %8
%17 = OpLabel
%18 = OpLoad %v4float %v1
%19 = OpFNegate %v4float %18
OpStore %OutColor %19
OpReturn
OpFunctionEnd
)";
Move tests into namespaces (#1689) This CL moves the test into namespaces based on their directories.
2018-07-11 13:24:49 +00:00
SinglePassRunAndCheck<AggressiveDCEPass>(assembly, assembly, true, true);
ADCE: Treat privates like locals in entry point with no calls This is needed for ongoing legalization of HLSL. It allows removal of accesses to textures/buffers that are not used.
2017-10-10 20:35:53 +00:00
}
TEST_F(AggressiveDCETest, NoPrivateStoreElimInNonEntry) {
// Should not eliminate stores to private when function is not entry point
// Note: Not legal GLSL
//
// #version 450
Re-format source tree - NFC. Re-formatted the source tree with the command: $ /usr/bin/clang-format -style=file -i \ $(find include source tools test utils -name '*.cpp' -or -name '*.h') This required a fix to source/val/decoration.h. It was not including spirv.h, which broke builds when the #include headers were re-ordered by clang-format.
2017-11-27 15:16:41 +00:00
//
ADCE: Treat privates like locals in entry point with no calls This is needed for ongoing legalization of HLSL. It allows removal of accesses to textures/buffers that are not used.
2017-10-10 20:35:53 +00:00
// layout(location = 0) in vec4 BaseColor;
// layout(location = 0) out vec4 OutColor;
//
// private vec4 v1;
Re-format source tree - NFC. Re-formatted the source tree with the command: $ /usr/bin/clang-format -style=file -i \ $(find include source tools test utils -name '*.cpp' -or -name '*.h') This required a fix to source/val/decoration.h. It was not including spirv.h, which broke builds when the #include headers were re-ordered by clang-format.
2017-11-27 15:16:41 +00:00
//
ADCE: Treat privates like locals in entry point with no calls This is needed for ongoing legalization of HLSL. It allows removal of accesses to textures/buffers that are not used.
2017-10-10 20:35:53 +00:00
// void foo()
// {
// v1 = BaseColor;
// }
Re-format source tree - NFC. Re-formatted the source tree with the command: $ /usr/bin/clang-format -style=file -i \ $(find include source tools test utils -name '*.cpp' -or -name '*.h') This required a fix to source/val/decoration.h. It was not including spirv.h, which broke builds when the #include headers were re-ordered by clang-format.
2017-11-27 15:16:41 +00:00
//
ADCE: Treat privates like locals in entry point with no calls This is needed for ongoing legalization of HLSL. It allows removal of accesses to textures/buffers that are not used.
2017-10-10 20:35:53 +00:00
// void main()
// {
// foo();
// OutColor = -v1;
// }
const std::string assembly =
R"(OpCapability Shader
%1 = OpExtInstImport "GLSL.std.450"
OpMemoryModel Logical GLSL450
OpEntryPoint Fragment %main "main" %BaseColor %OutColor
OpExecutionMode %main OriginUpperLeft
OpSource GLSL 450
OpName %main "main"
OpName %foo_ "foo("
OpName %v1 "v1"
OpName %BaseColor "BaseColor"
OpName %OutColor "OutColor"
OpDecorate %BaseColor Location 0
OpDecorate %OutColor Location 0
%void = OpTypeVoid
%8 = OpTypeFunction %void
%float = OpTypeFloat 32
%v4float = OpTypeVector %float 4
%_ptr_Private_v4float = OpTypePointer Private %v4float
%_ptr_Input_v4float = OpTypePointer Input %v4float
%BaseColor = OpVariable %_ptr_Input_v4float Input
%_ptr_Output_v4float = OpTypePointer Output %v4float
%v1 = OpVariable %_ptr_Private_v4float Private
%OutColor = OpVariable %_ptr_Output_v4float Output
%main = OpFunction %void None %8
%14 = OpLabel
%15 = OpFunctionCall %void %foo_
%16 = OpLoad %v4float %v1
%17 = OpFNegate %v4float %16
OpStore %OutColor %17
OpReturn
OpFunctionEnd
%foo_ = OpFunction %void None %8
%18 = OpLabel
%19 = OpLoad %v4float %BaseColor
OpStore %v1 %19
OpReturn
OpFunctionEnd
)";
Move tests into namespaces (#1689) This CL moves the test into namespaces based on their directories.
2018-07-11 13:24:49 +00:00
SinglePassRunAndCheck<AggressiveDCEPass>(assembly, assembly, true, true);
ADCE: Treat privates like locals in entry point with no calls This is needed for ongoing legalization of HLSL. It allows removal of accesses to textures/buffers that are not used.
2017-10-10 20:35:53 +00:00
}
Remove dead Workgroup variables in ADCE. If there is a shader with a variable in the workgroup storage class that is stored to, but not loadeds, then we know nothing will read those loads. It should be safe to remove them. This is implemented in ADCE by treating workgroup variables the same way that private variables are treated. Fixes https://github.com/KhronosGroup/SPIRV-Tools/issues/1550.
2018-05-08 18:02:03 +00:00
TEST_F(AggressiveDCETest, WorkgroupStoreElimInEntryNoCalls) {
// Eliminate stores to private in entry point with no calls
// Note: Not legal GLSL
//
// layout(location = 0) in vec4 BaseColor;
// layout(location = 1) in vec4 Dead;
// layout(location = 0) out vec4 OutColor;
//
// workgroup vec4 dv;
//
// void main()
// {
// vec4 v = BaseColor;
// dv = Dead;
// OutColor = v;
// }
Allow ADCE to remove dead inputs (#4629) * https://github.com/KhronosGroup/Vulkan-Docs/issues/666 clearly specified that interfaces do not require an input if there is an associated output * ADCE can now remove unused input variables (though they are kept if the preserve interfaces option is used)
2021-11-16 20:54:17 +00:00
const std::string spirv =
Remove dead Workgroup variables in ADCE. If there is a shader with a variable in the workgroup storage class that is stored to, but not loadeds, then we know nothing will read those loads. It should be safe to remove them. This is implemented in ADCE by treating workgroup variables the same way that private variables are treated. Fixes https://github.com/KhronosGroup/SPIRV-Tools/issues/1550.
2018-05-08 18:02:03 +00:00
R"(OpCapability Shader
%1 = OpExtInstImport "GLSL.std.450"
OpMemoryModel Logical GLSL450
OpEntryPoint Fragment %main "main" %BaseColor %Dead %OutColor
OpExecutionMode %main OriginUpperLeft
OpSource GLSL 450
OpName %main "main"
OpName %v "v"
OpName %BaseColor "BaseColor"
Allow ADCE to remove dead inputs (#4629) * https://github.com/KhronosGroup/Vulkan-Docs/issues/666 clearly specified that interfaces do not require an input if there is an associated output * ADCE can now remove unused input variables (though they are kept if the preserve interfaces option is used)
2021-11-16 20:54:17 +00:00
; CHECK-NOT: OpName %dv "dv"
Remove dead Workgroup variables in ADCE. If there is a shader with a variable in the workgroup storage class that is stored to, but not loadeds, then we know nothing will read those loads. It should be safe to remove them. This is implemented in ADCE by treating workgroup variables the same way that private variables are treated. Fixes https://github.com/KhronosGroup/SPIRV-Tools/issues/1550.
2018-05-08 18:02:03 +00:00
OpName %dv "dv"
OpName %Dead "Dead"
OpName %OutColor "OutColor"
OpDecorate %BaseColor Location 0
OpDecorate %Dead Location 1
OpDecorate %OutColor Location 0
%void = OpTypeVoid
%9 = OpTypeFunction %void
%float = OpTypeFloat 32
%v4float = OpTypeVector %float 4
%_ptr_Function_v4float = OpTypePointer Function %v4float
Allow ADCE to remove dead inputs (#4629) * https://github.com/KhronosGroup/Vulkan-Docs/issues/666 clearly specified that interfaces do not require an input if there is an associated output * ADCE can now remove unused input variables (though they are kept if the preserve interfaces option is used)
2021-11-16 20:54:17 +00:00
; CHECK-NOT: OpTypePointer Workgroup
Remove dead Workgroup variables in ADCE. If there is a shader with a variable in the workgroup storage class that is stored to, but not loadeds, then we know nothing will read those loads. It should be safe to remove them. This is implemented in ADCE by treating workgroup variables the same way that private variables are treated. Fixes https://github.com/KhronosGroup/SPIRV-Tools/issues/1550.
2018-05-08 18:02:03 +00:00
%_ptr_Workgroup_v4float = OpTypePointer Workgroup %v4float
%_ptr_Input_v4float = OpTypePointer Input %v4float
%BaseColor = OpVariable %_ptr_Input_v4float Input
%Dead = OpVariable %_ptr_Input_v4float Input
%_ptr_Output_v4float = OpTypePointer Output %v4float
Allow ADCE to remove dead inputs (#4629) * https://github.com/KhronosGroup/Vulkan-Docs/issues/666 clearly specified that interfaces do not require an input if there is an associated output * ADCE can now remove unused input variables (though they are kept if the preserve interfaces option is used)
2021-11-16 20:54:17 +00:00
; CHECK-NOT: %dv = OpVariable
Remove dead Workgroup variables in ADCE. If there is a shader with a variable in the workgroup storage class that is stored to, but not loadeds, then we know nothing will read those loads. It should be safe to remove them. This is implemented in ADCE by treating workgroup variables the same way that private variables are treated. Fixes https://github.com/KhronosGroup/SPIRV-Tools/issues/1550.
2018-05-08 18:02:03 +00:00
%dv = OpVariable %_ptr_Workgroup_v4float Workgroup
%OutColor = OpVariable %_ptr_Output_v4float Output
Allow ADCE to remove dead inputs (#4629) * https://github.com/KhronosGroup/Vulkan-Docs/issues/666 clearly specified that interfaces do not require an input if there is an associated output * ADCE can now remove unused input variables (though they are kept if the preserve interfaces option is used)
2021-11-16 20:54:17 +00:00
%main = OpFunction %void None %9
Remove dead Workgroup variables in ADCE. If there is a shader with a variable in the workgroup storage class that is stored to, but not loadeds, then we know nothing will read those loads. It should be safe to remove them. This is implemented in ADCE by treating workgroup variables the same way that private variables are treated. Fixes https://github.com/KhronosGroup/SPIRV-Tools/issues/1550.
2018-05-08 18:02:03 +00:00
%16 = OpLabel
%v = OpVariable %_ptr_Function_v4float Function
%17 = OpLoad %v4float %BaseColor
OpStore %v %17
%18 = OpLoad %v4float %Dead
Allow ADCE to remove dead inputs (#4629) * https://github.com/KhronosGroup/Vulkan-Docs/issues/666 clearly specified that interfaces do not require an input if there is an associated output * ADCE can now remove unused input variables (though they are kept if the preserve interfaces option is used)
2021-11-16 20:54:17 +00:00
; CHECK-NOT: OpStore %dv
Remove dead Workgroup variables in ADCE. If there is a shader with a variable in the workgroup storage class that is stored to, but not loadeds, then we know nothing will read those loads. It should be safe to remove them. This is implemented in ADCE by treating workgroup variables the same way that private variables are treated. Fixes https://github.com/KhronosGroup/SPIRV-Tools/issues/1550.
2018-05-08 18:02:03 +00:00
OpStore %dv %18
%19 = OpLoad %v4float %v
%20 = OpFNegate %v4float %19
OpStore %OutColor %20
OpReturn
OpFunctionEnd
)";
Allow ADCE to remove dead inputs (#4629) * https://github.com/KhronosGroup/Vulkan-Docs/issues/666 clearly specified that interfaces do not require an input if there is an associated output * ADCE can now remove unused input variables (though they are kept if the preserve interfaces option is used)
2021-11-16 20:54:17 +00:00
SinglePassRunAndMatch<AggressiveDCEPass>(spirv, true);
Remove dead Workgroup variables in ADCE. If there is a shader with a variable in the workgroup storage class that is stored to, but not loadeds, then we know nothing will read those loads. It should be safe to remove them. This is implemented in ADCE by treating workgroup variables the same way that private variables are treated. Fixes https://github.com/KhronosGroup/SPIRV-Tools/issues/1550.
2018-05-08 18:02:03 +00:00
}
ADCE: Dead if elimination Mark structured conditional branches live only if one or more instructions in their associated construct is marked live. After closure, replace dead structured conditional branches with a branch to its merge and remove dead blocks. ADCE: Dead If Elim: Remove duplicate StructuredOrder code Also generalize ComputeStructuredOrder so that the caller can specify the root block for the order. Phi insertion uses pseudo_entry_block and adce and dead branch elim use the first block of the function. ADCE: Dead If Elim: Pull redundant code out of InsertPhiInstructions ADCE: Dead If Elim: Encapsulate CFG Cleanup Initialization ADCE: Dead If Elim: Remove redundant code from ADCE initialization ADCE: Dead If: Use CFGCleanup to eliminate newly dead blocks Moved bulk of CFG Cleanup code into MemPass.
2017-10-17 22:33:43 +00:00
TEST_F(AggressiveDCETest, EliminateDeadIfThenElse) {
// #version 450
Re-format source tree - NFC. Re-formatted the source tree with the command: $ /usr/bin/clang-format -style=file -i \ $(find include source tools test utils -name '*.cpp' -or -name '*.h') This required a fix to source/val/decoration.h. It was not including spirv.h, which broke builds when the #include headers were re-ordered by clang-format.
2017-11-27 15:16:41 +00:00
//
ADCE: Dead if elimination Mark structured conditional branches live only if one or more instructions in their associated construct is marked live. After closure, replace dead structured conditional branches with a branch to its merge and remove dead blocks. ADCE: Dead If Elim: Remove duplicate StructuredOrder code Also generalize ComputeStructuredOrder so that the caller can specify the root block for the order. Phi insertion uses pseudo_entry_block and adce and dead branch elim use the first block of the function. ADCE: Dead If Elim: Pull redundant code out of InsertPhiInstructions ADCE: Dead If Elim: Encapsulate CFG Cleanup Initialization ADCE: Dead If Elim: Remove redundant code from ADCE initialization ADCE: Dead If: Use CFGCleanup to eliminate newly dead blocks Moved bulk of CFG Cleanup code into MemPass.
2017-10-17 22:33:43 +00:00
// layout(location = 0) in vec4 BaseColor;
// layout(location = 0) out vec4 OutColor;
Re-format source tree - NFC. Re-formatted the source tree with the command: $ /usr/bin/clang-format -style=file -i \ $(find include source tools test utils -name '*.cpp' -or -name '*.h') This required a fix to source/val/decoration.h. It was not including spirv.h, which broke builds when the #include headers were re-ordered by clang-format.
2017-11-27 15:16:41 +00:00
//
ADCE: Dead if elimination Mark structured conditional branches live only if one or more instructions in their associated construct is marked live. After closure, replace dead structured conditional branches with a branch to its merge and remove dead blocks. ADCE: Dead If Elim: Remove duplicate StructuredOrder code Also generalize ComputeStructuredOrder so that the caller can specify the root block for the order. Phi insertion uses pseudo_entry_block and adce and dead branch elim use the first block of the function. ADCE: Dead If Elim: Pull redundant code out of InsertPhiInstructions ADCE: Dead If Elim: Encapsulate CFG Cleanup Initialization ADCE: Dead If Elim: Remove redundant code from ADCE initialization ADCE: Dead If: Use CFGCleanup to eliminate newly dead blocks Moved bulk of CFG Cleanup code into MemPass.
2017-10-17 22:33:43 +00:00
// void main()
// {
// float d;
// if (BaseColor.x == 0)
// d = BaseColor.y;
// else
// d = BaseColor.z;
// OutColor = vec4(1.0,1.0,1.0,1.0);
// }
Allow ADCE to remove dead inputs (#4629) * https://github.com/KhronosGroup/Vulkan-Docs/issues/666 clearly specified that interfaces do not require an input if there is an associated output * ADCE can now remove unused input variables (though they are kept if the preserve interfaces option is used)
2021-11-16 20:54:17 +00:00
const std::string spirv =
ADCE: Dead if elimination Mark structured conditional branches live only if one or more instructions in their associated construct is marked live. After closure, replace dead structured conditional branches with a branch to its merge and remove dead blocks. ADCE: Dead If Elim: Remove duplicate StructuredOrder code Also generalize ComputeStructuredOrder so that the caller can specify the root block for the order. Phi insertion uses pseudo_entry_block and adce and dead branch elim use the first block of the function. ADCE: Dead If Elim: Pull redundant code out of InsertPhiInstructions ADCE: Dead If Elim: Encapsulate CFG Cleanup Initialization ADCE: Dead If Elim: Remove redundant code from ADCE initialization ADCE: Dead If: Use CFGCleanup to eliminate newly dead blocks Moved bulk of CFG Cleanup code into MemPass.
2017-10-17 22:33:43 +00:00
R"(OpCapability Shader
%1 = OpExtInstImport "GLSL.std.450"
OpMemoryModel Logical GLSL450
OpEntryPoint Fragment %main "main" %BaseColor %OutColor
OpExecutionMode %main OriginUpperLeft
OpSource GLSL 450
OpName %main "main"
OpName %BaseColor "BaseColor"
OpName %d "d"
OpName %OutColor "OutColor"
OpDecorate %BaseColor Location 0
OpDecorate %OutColor Location 0
%void = OpTypeVoid
%7 = OpTypeFunction %void
%float = OpTypeFloat 32
%v4float = OpTypeVector %float 4
%_ptr_Input_v4float = OpTypePointer Input %v4float
%BaseColor = OpVariable %_ptr_Input_v4float Input
%uint = OpTypeInt 32 0
%uint_0 = OpConstant %uint 0
%_ptr_Input_float = OpTypePointer Input %float
%float_0 = OpConstant %float 0
%bool = OpTypeBool
%_ptr_Function_float = OpTypePointer Function %float
%uint_1 = OpConstant %uint 1
%uint_2 = OpConstant %uint 2
%_ptr_Output_v4float = OpTypePointer Output %v4float
%OutColor = OpVariable %_ptr_Output_v4float Output
%float_1 = OpConstant %float 1
%21 = OpConstantComposite %v4float %float_1 %float_1 %float_1 %float_1
Allow ADCE to remove dead inputs (#4629) * https://github.com/KhronosGroup/Vulkan-Docs/issues/666 clearly specified that interfaces do not require an input if there is an associated output * ADCE can now remove unused input variables (though they are kept if the preserve interfaces option is used)
2021-11-16 20:54:17 +00:00
; CHECK: = OpFunction %void
; CHECK-NEXT: %22 = OpLabel
; CHECK-NEXT: OpBranch %26
; CHECK-NEXT: %26 = OpLabel
%main = OpFunction %void None %7
ADCE: Dead if elimination Mark structured conditional branches live only if one or more instructions in their associated construct is marked live. After closure, replace dead structured conditional branches with a branch to its merge and remove dead blocks. ADCE: Dead If Elim: Remove duplicate StructuredOrder code Also generalize ComputeStructuredOrder so that the caller can specify the root block for the order. Phi insertion uses pseudo_entry_block and adce and dead branch elim use the first block of the function. ADCE: Dead If Elim: Pull redundant code out of InsertPhiInstructions ADCE: Dead If Elim: Encapsulate CFG Cleanup Initialization ADCE: Dead If Elim: Remove redundant code from ADCE initialization ADCE: Dead If: Use CFGCleanup to eliminate newly dead blocks Moved bulk of CFG Cleanup code into MemPass.
2017-10-17 22:33:43 +00:00
%22 = OpLabel
%d = OpVariable %_ptr_Function_float Function
%23 = OpAccessChain %_ptr_Input_float %BaseColor %uint_0
%24 = OpLoad %float %23
ADCE: Empty Loop Elimination This entirely eliminates loops which do not contain live code.
2017-10-25 00:58:48 +00:00
%25 = OpFOrdEqual %bool %24 %float_0
ADCE: Dead if elimination Mark structured conditional branches live only if one or more instructions in their associated construct is marked live. After closure, replace dead structured conditional branches with a branch to its merge and remove dead blocks. ADCE: Dead If Elim: Remove duplicate StructuredOrder code Also generalize ComputeStructuredOrder so that the caller can specify the root block for the order. Phi insertion uses pseudo_entry_block and adce and dead branch elim use the first block of the function. ADCE: Dead If Elim: Pull redundant code out of InsertPhiInstructions ADCE: Dead If Elim: Encapsulate CFG Cleanup Initialization ADCE: Dead If Elim: Remove redundant code from ADCE initialization ADCE: Dead If: Use CFGCleanup to eliminate newly dead blocks Moved bulk of CFG Cleanup code into MemPass.
2017-10-17 22:33:43 +00:00
OpSelectionMerge %26 None
OpBranchConditional %25 %27 %28
%27 = OpLabel
%29 = OpAccessChain %_ptr_Input_float %BaseColor %uint_1
%30 = OpLoad %float %29
OpStore %d %30
OpBranch %26
%28 = OpLabel
%31 = OpAccessChain %_ptr_Input_float %BaseColor %uint_2
%32 = OpLoad %float %31
OpStore %d %32
OpBranch %26
%26 = OpLabel
OpStore %OutColor %21
OpReturn
OpFunctionEnd
)";
Allow ADCE to remove dead inputs (#4629) * https://github.com/KhronosGroup/Vulkan-Docs/issues/666 clearly specified that interfaces do not require an input if there is an associated output * ADCE can now remove unused input variables (though they are kept if the preserve interfaces option is used)
2021-11-16 20:54:17 +00:00
SinglePassRunAndMatch<AggressiveDCEPass>(spirv, true);
ADCE: Dead if elimination Mark structured conditional branches live only if one or more instructions in their associated construct is marked live. After closure, replace dead structured conditional branches with a branch to its merge and remove dead blocks. ADCE: Dead If Elim: Remove duplicate StructuredOrder code Also generalize ComputeStructuredOrder so that the caller can specify the root block for the order. Phi insertion uses pseudo_entry_block and adce and dead branch elim use the first block of the function. ADCE: Dead If Elim: Pull redundant code out of InsertPhiInstructions ADCE: Dead If Elim: Encapsulate CFG Cleanup Initialization ADCE: Dead If Elim: Remove redundant code from ADCE initialization ADCE: Dead If: Use CFGCleanup to eliminate newly dead blocks Moved bulk of CFG Cleanup code into MemPass.
2017-10-17 22:33:43 +00:00
}
TEST_F(AggressiveDCETest, EliminateDeadIfThen) {
// #version 450
Re-format source tree - NFC. Re-formatted the source tree with the command: $ /usr/bin/clang-format -style=file -i \ $(find include source tools test utils -name '*.cpp' -or -name '*.h') This required a fix to source/val/decoration.h. It was not including spirv.h, which broke builds when the #include headers were re-ordered by clang-format.
2017-11-27 15:16:41 +00:00
//
ADCE: Dead if elimination Mark structured conditional branches live only if one or more instructions in their associated construct is marked live. After closure, replace dead structured conditional branches with a branch to its merge and remove dead blocks. ADCE: Dead If Elim: Remove duplicate StructuredOrder code Also generalize ComputeStructuredOrder so that the caller can specify the root block for the order. Phi insertion uses pseudo_entry_block and adce and dead branch elim use the first block of the function. ADCE: Dead If Elim: Pull redundant code out of InsertPhiInstructions ADCE: Dead If Elim: Encapsulate CFG Cleanup Initialization ADCE: Dead If Elim: Remove redundant code from ADCE initialization ADCE: Dead If: Use CFGCleanup to eliminate newly dead blocks Moved bulk of CFG Cleanup code into MemPass.
2017-10-17 22:33:43 +00:00
// layout(location = 0) in vec4 BaseColor;
// layout(location = 0) out vec4 OutColor;
Re-format source tree - NFC. Re-formatted the source tree with the command: $ /usr/bin/clang-format -style=file -i \ $(find include source tools test utils -name '*.cpp' -or -name '*.h') This required a fix to source/val/decoration.h. It was not including spirv.h, which broke builds when the #include headers were re-ordered by clang-format.
2017-11-27 15:16:41 +00:00
//
ADCE: Dead if elimination Mark structured conditional branches live only if one or more instructions in their associated construct is marked live. After closure, replace dead structured conditional branches with a branch to its merge and remove dead blocks. ADCE: Dead If Elim: Remove duplicate StructuredOrder code Also generalize ComputeStructuredOrder so that the caller can specify the root block for the order. Phi insertion uses pseudo_entry_block and adce and dead branch elim use the first block of the function. ADCE: Dead If Elim: Pull redundant code out of InsertPhiInstructions ADCE: Dead If Elim: Encapsulate CFG Cleanup Initialization ADCE: Dead If Elim: Remove redundant code from ADCE initialization ADCE: Dead If: Use CFGCleanup to eliminate newly dead blocks Moved bulk of CFG Cleanup code into MemPass.
2017-10-17 22:33:43 +00:00
// void main()
// {
// float d;
// if (BaseColor.x == 0)
// d = BaseColor.y;
// OutColor = vec4(1.0,1.0,1.0,1.0);
// }
Allow ADCE to remove dead inputs (#4629) * https://github.com/KhronosGroup/Vulkan-Docs/issues/666 clearly specified that interfaces do not require an input if there is an associated output * ADCE can now remove unused input variables (though they are kept if the preserve interfaces option is used)
2021-11-16 20:54:17 +00:00
const std::string spirv =
ADCE: Dead if elimination Mark structured conditional branches live only if one or more instructions in their associated construct is marked live. After closure, replace dead structured conditional branches with a branch to its merge and remove dead blocks. ADCE: Dead If Elim: Remove duplicate StructuredOrder code Also generalize ComputeStructuredOrder so that the caller can specify the root block for the order. Phi insertion uses pseudo_entry_block and adce and dead branch elim use the first block of the function. ADCE: Dead If Elim: Pull redundant code out of InsertPhiInstructions ADCE: Dead If Elim: Encapsulate CFG Cleanup Initialization ADCE: Dead If Elim: Remove redundant code from ADCE initialization ADCE: Dead If: Use CFGCleanup to eliminate newly dead blocks Moved bulk of CFG Cleanup code into MemPass.
2017-10-17 22:33:43 +00:00
R"(OpCapability Shader
%1 = OpExtInstImport "GLSL.std.450"
OpMemoryModel Logical GLSL450
OpEntryPoint Fragment %main "main" %BaseColor %OutColor
OpExecutionMode %main OriginUpperLeft
OpSource GLSL 450
OpName %main "main"
OpName %BaseColor "BaseColor"
OpName %d "d"
OpName %OutColor "OutColor"
OpDecorate %BaseColor Location 0
OpDecorate %OutColor Location 0
%void = OpTypeVoid
%7 = OpTypeFunction %void
%float = OpTypeFloat 32
%v4float = OpTypeVector %float 4
%_ptr_Input_v4float = OpTypePointer Input %v4float
%BaseColor = OpVariable %_ptr_Input_v4float Input
%uint = OpTypeInt 32 0
%uint_0 = OpConstant %uint 0
%_ptr_Input_float = OpTypePointer Input %float
%float_0 = OpConstant %float 0
%bool = OpTypeBool
%_ptr_Function_float = OpTypePointer Function %float
%uint_1 = OpConstant %uint 1
%_ptr_Output_v4float = OpTypePointer Output %v4float
%OutColor = OpVariable %_ptr_Output_v4float Output
%float_1 = OpConstant %float 1
%20 = OpConstantComposite %v4float %float_1 %float_1 %float_1 %float_1
Allow ADCE to remove dead inputs (#4629) * https://github.com/KhronosGroup/Vulkan-Docs/issues/666 clearly specified that interfaces do not require an input if there is an associated output * ADCE can now remove unused input variables (though they are kept if the preserve interfaces option is used)
2021-11-16 20:54:17 +00:00
; CHECK: = OpFunction
; CHECK-NEXT: %21 = OpLabel
; CHECK-NEXT: OpBranch [[target:%\w+]]
; CHECK-NEXT: [[target]] = OpLabel
%main = OpFunction %void None %7
ADCE: Dead if elimination Mark structured conditional branches live only if one or more instructions in their associated construct is marked live. After closure, replace dead structured conditional branches with a branch to its merge and remove dead blocks. ADCE: Dead If Elim: Remove duplicate StructuredOrder code Also generalize ComputeStructuredOrder so that the caller can specify the root block for the order. Phi insertion uses pseudo_entry_block and adce and dead branch elim use the first block of the function. ADCE: Dead If Elim: Pull redundant code out of InsertPhiInstructions ADCE: Dead If Elim: Encapsulate CFG Cleanup Initialization ADCE: Dead If Elim: Remove redundant code from ADCE initialization ADCE: Dead If: Use CFGCleanup to eliminate newly dead blocks Moved bulk of CFG Cleanup code into MemPass.
2017-10-17 22:33:43 +00:00
%21 = OpLabel
%d = OpVariable %_ptr_Function_float Function
%22 = OpAccessChain %_ptr_Input_float %BaseColor %uint_0
%23 = OpLoad %float %22
ADCE: Empty Loop Elimination This entirely eliminates loops which do not contain live code.
2017-10-25 00:58:48 +00:00
%24 = OpFOrdEqual %bool %23 %float_0
ADCE: Dead if elimination Mark structured conditional branches live only if one or more instructions in their associated construct is marked live. After closure, replace dead structured conditional branches with a branch to its merge and remove dead blocks. ADCE: Dead If Elim: Remove duplicate StructuredOrder code Also generalize ComputeStructuredOrder so that the caller can specify the root block for the order. Phi insertion uses pseudo_entry_block and adce and dead branch elim use the first block of the function. ADCE: Dead If Elim: Pull redundant code out of InsertPhiInstructions ADCE: Dead If Elim: Encapsulate CFG Cleanup Initialization ADCE: Dead If Elim: Remove redundant code from ADCE initialization ADCE: Dead If: Use CFGCleanup to eliminate newly dead blocks Moved bulk of CFG Cleanup code into MemPass.
2017-10-17 22:33:43 +00:00
OpSelectionMerge %25 None
OpBranchConditional %24 %26 %25
%26 = OpLabel
%27 = OpAccessChain %_ptr_Input_float %BaseColor %uint_1
%28 = OpLoad %float %27
OpStore %d %28
OpBranch %25
%25 = OpLabel
OpStore %OutColor %20
OpReturn
OpFunctionEnd
)";
Allow ADCE to remove dead inputs (#4629) * https://github.com/KhronosGroup/Vulkan-Docs/issues/666 clearly specified that interfaces do not require an input if there is an associated output * ADCE can now remove unused input variables (though they are kept if the preserve interfaces option is used)
2021-11-16 20:54:17 +00:00
SinglePassRunAndMatch<AggressiveDCEPass>(spirv, true);
ADCE: Dead if elimination Mark structured conditional branches live only if one or more instructions in their associated construct is marked live. After closure, replace dead structured conditional branches with a branch to its merge and remove dead blocks. ADCE: Dead If Elim: Remove duplicate StructuredOrder code Also generalize ComputeStructuredOrder so that the caller can specify the root block for the order. Phi insertion uses pseudo_entry_block and adce and dead branch elim use the first block of the function. ADCE: Dead If Elim: Pull redundant code out of InsertPhiInstructions ADCE: Dead If Elim: Encapsulate CFG Cleanup Initialization ADCE: Dead If Elim: Remove redundant code from ADCE initialization ADCE: Dead If: Use CFGCleanup to eliminate newly dead blocks Moved bulk of CFG Cleanup code into MemPass.
2017-10-17 22:33:43 +00:00
}
Adding support for switch removal in ADCE * Updated code to handle switches * Enabled disabled test and added a couple new ones
2018-01-15 18:25:45 +00:00
TEST_F(AggressiveDCETest, EliminateDeadSwitch) {
Generic value propagation engine. This class implements a generic value propagation algorithm based on the conditional constant propagation algorithm proposed in Constant propagation with conditional branches, Wegman and Zadeck, ACM TOPLAS 13(2):181-210. The implementation is based on A Propagation Engine for GCC Diego Novillo, GCC Summit 2005 http://ols.fedoraproject.org/GCC/Reprints-2005/novillo-Reprint.pdf The purpose of this implementation is to act as a common framework for any transformation that needs to propagate values from statements producing new values to statements using those values.
2017-11-17 13:59:25 +00:00
// #version 450
//
// layout(location = 0) in vec4 BaseColor;
// layout(location = 1) in flat int x;
// layout(location = 0) out vec4 OutColor;
//
// void main()
// {
// float d;
// switch (x) {
// case 0:
// d = BaseColor.y;
// }
// OutColor = vec4(1.0,1.0,1.0,1.0);
// }
Allow ADCE to remove dead inputs (#4629) * https://github.com/KhronosGroup/Vulkan-Docs/issues/666 clearly specified that interfaces do not require an input if there is an associated output * ADCE can now remove unused input variables (though they are kept if the preserve interfaces option is used)
2021-11-16 20:54:17 +00:00
const std::string spirv =
Generic value propagation engine. This class implements a generic value propagation algorithm based on the conditional constant propagation algorithm proposed in Constant propagation with conditional branches, Wegman and Zadeck, ACM TOPLAS 13(2):181-210. The implementation is based on A Propagation Engine for GCC Diego Novillo, GCC Summit 2005 http://ols.fedoraproject.org/GCC/Reprints-2005/novillo-Reprint.pdf The purpose of this implementation is to act as a common framework for any transformation that needs to propagate values from statements producing new values to statements using those values.
2017-11-17 13:59:25 +00:00
R"(OpCapability Shader
%1 = OpExtInstImport "GLSL.std.450"
OpMemoryModel Logical GLSL450
OpEntryPoint Fragment %main "main" %x %BaseColor %OutColor
OpExecutionMode %main OriginUpperLeft
OpSource GLSL 450
OpName %main "main"
OpName %x "x"
OpName %d "d"
OpName %BaseColor "BaseColor"
OpName %OutColor "OutColor"
OpDecorate %x Flat
OpDecorate %x Location 1
OpDecorate %BaseColor Location 0
OpDecorate %OutColor Location 0
%void = OpTypeVoid
%3 = OpTypeFunction %void
%int = OpTypeInt 32 1
%_ptr_Input_int = OpTypePointer Input %int
%x = OpVariable %_ptr_Input_int Input
%float = OpTypeFloat 32
%_ptr_Function_float = OpTypePointer Function %float
%v4float = OpTypeVector %float 4
%_ptr_Input_v4float = OpTypePointer Input %v4float
%BaseColor = OpVariable %_ptr_Input_v4float Input
%uint = OpTypeInt 32 0
%uint_1 = OpConstant %uint 1
%_ptr_Input_float = OpTypePointer Input %float
%_ptr_Output_v4float = OpTypePointer Output %v4float
%OutColor = OpVariable %_ptr_Output_v4float Output
%float_1 = OpConstant %float 1
%27 = OpConstantComposite %v4float %float_1 %float_1 %float_1 %float_1
Allow ADCE to remove dead inputs (#4629) * https://github.com/KhronosGroup/Vulkan-Docs/issues/666 clearly specified that interfaces do not require an input if there is an associated output * ADCE can now remove unused input variables (though they are kept if the preserve interfaces option is used)
2021-11-16 20:54:17 +00:00
; CHECK: = OpFunction
; CHECK-NEXT: = OpLabel
; CHECK-NEXT: OpBranch [[target:%\w+]]
; CHECK-NEXT: [[target]] = OpLabel
Generic value propagation engine. This class implements a generic value propagation algorithm based on the conditional constant propagation algorithm proposed in Constant propagation with conditional branches, Wegman and Zadeck, ACM TOPLAS 13(2):181-210. The implementation is based on A Propagation Engine for GCC Diego Novillo, GCC Summit 2005 http://ols.fedoraproject.org/GCC/Reprints-2005/novillo-Reprint.pdf The purpose of this implementation is to act as a common framework for any transformation that needs to propagate values from statements producing new values to statements using those values.
2017-11-17 13:59:25 +00:00
%main = OpFunction %void None %3
%5 = OpLabel
%d = OpVariable %_ptr_Function_float Function
%9 = OpLoad %int %x
OpSelectionMerge %11 None
OpSwitch %9 %11 0 %10
%10 = OpLabel
%21 = OpAccessChain %_ptr_Input_float %BaseColor %uint_1
%22 = OpLoad %float %21
OpStore %d %22
OpBranch %11
%11 = OpLabel
OpStore %OutColor %27
OpReturn
OpFunctionEnd)";
Allow ADCE to remove dead inputs (#4629) * https://github.com/KhronosGroup/Vulkan-Docs/issues/666 clearly specified that interfaces do not require an input if there is an associated output * ADCE can now remove unused input variables (though they are kept if the preserve interfaces option is used)
2021-11-16 20:54:17 +00:00
SinglePassRunAndMatch<AggressiveDCEPass>(spirv, true);
Generic value propagation engine. This class implements a generic value propagation algorithm based on the conditional constant propagation algorithm proposed in Constant propagation with conditional branches, Wegman and Zadeck, ACM TOPLAS 13(2):181-210. The implementation is based on A Propagation Engine for GCC Diego Novillo, GCC Summit 2005 http://ols.fedoraproject.org/GCC/Reprints-2005/novillo-Reprint.pdf The purpose of this implementation is to act as a common framework for any transformation that needs to propagate values from statements producing new values to statements using those values.
2017-11-17 13:59:25 +00:00
}
ADCE: Dead if elimination Mark structured conditional branches live only if one or more instructions in their associated construct is marked live. After closure, replace dead structured conditional branches with a branch to its merge and remove dead blocks. ADCE: Dead If Elim: Remove duplicate StructuredOrder code Also generalize ComputeStructuredOrder so that the caller can specify the root block for the order. Phi insertion uses pseudo_entry_block and adce and dead branch elim use the first block of the function. ADCE: Dead If Elim: Pull redundant code out of InsertPhiInstructions ADCE: Dead If Elim: Encapsulate CFG Cleanup Initialization ADCE: Dead If Elim: Remove redundant code from ADCE initialization ADCE: Dead If: Use CFGCleanup to eliminate newly dead blocks Moved bulk of CFG Cleanup code into MemPass.
2017-10-17 22:33:43 +00:00
TEST_F(AggressiveDCETest, EliminateDeadIfThenElseNested) {
// #version 450
Re-format source tree - NFC. Re-formatted the source tree with the command: $ /usr/bin/clang-format -style=file -i \ $(find include source tools test utils -name '*.cpp' -or -name '*.h') This required a fix to source/val/decoration.h. It was not including spirv.h, which broke builds when the #include headers were re-ordered by clang-format.
2017-11-27 15:16:41 +00:00
//
ADCE: Dead if elimination Mark structured conditional branches live only if one or more instructions in their associated construct is marked live. After closure, replace dead structured conditional branches with a branch to its merge and remove dead blocks. ADCE: Dead If Elim: Remove duplicate StructuredOrder code Also generalize ComputeStructuredOrder so that the caller can specify the root block for the order. Phi insertion uses pseudo_entry_block and adce and dead branch elim use the first block of the function. ADCE: Dead If Elim: Pull redundant code out of InsertPhiInstructions ADCE: Dead If Elim: Encapsulate CFG Cleanup Initialization ADCE: Dead If Elim: Remove redundant code from ADCE initialization ADCE: Dead If: Use CFGCleanup to eliminate newly dead blocks Moved bulk of CFG Cleanup code into MemPass.
2017-10-17 22:33:43 +00:00
// layout(location = 0) in vec4 BaseColor;
// layout(location = 0) out vec4 OutColor;
Re-format source tree - NFC. Re-formatted the source tree with the command: $ /usr/bin/clang-format -style=file -i \ $(find include source tools test utils -name '*.cpp' -or -name '*.h') This required a fix to source/val/decoration.h. It was not including spirv.h, which broke builds when the #include headers were re-ordered by clang-format.
2017-11-27 15:16:41 +00:00
//
ADCE: Dead if elimination Mark structured conditional branches live only if one or more instructions in their associated construct is marked live. After closure, replace dead structured conditional branches with a branch to its merge and remove dead blocks. ADCE: Dead If Elim: Remove duplicate StructuredOrder code Also generalize ComputeStructuredOrder so that the caller can specify the root block for the order. Phi insertion uses pseudo_entry_block and adce and dead branch elim use the first block of the function. ADCE: Dead If Elim: Pull redundant code out of InsertPhiInstructions ADCE: Dead If Elim: Encapsulate CFG Cleanup Initialization ADCE: Dead If Elim: Remove redundant code from ADCE initialization ADCE: Dead If: Use CFGCleanup to eliminate newly dead blocks Moved bulk of CFG Cleanup code into MemPass.
2017-10-17 22:33:43 +00:00
// void main()
// {
// float d;
// if (BaseColor.x == 0)
// if (BaseColor.y == 0)
// d = 0.0;
// else
// d = 0.25;
// else
// if (BaseColor.y == 0)
// d = 0.5;
// else
// d = 0.75;
// OutColor = vec4(1.0,1.0,1.0,1.0);
// }
Allow ADCE to remove dead inputs (#4629) * https://github.com/KhronosGroup/Vulkan-Docs/issues/666 clearly specified that interfaces do not require an input if there is an associated output * ADCE can now remove unused input variables (though they are kept if the preserve interfaces option is used)
2021-11-16 20:54:17 +00:00
const std::string spirv =
ADCE: Dead if elimination Mark structured conditional branches live only if one or more instructions in their associated construct is marked live. After closure, replace dead structured conditional branches with a branch to its merge and remove dead blocks. ADCE: Dead If Elim: Remove duplicate StructuredOrder code Also generalize ComputeStructuredOrder so that the caller can specify the root block for the order. Phi insertion uses pseudo_entry_block and adce and dead branch elim use the first block of the function. ADCE: Dead If Elim: Pull redundant code out of InsertPhiInstructions ADCE: Dead If Elim: Encapsulate CFG Cleanup Initialization ADCE: Dead If Elim: Remove redundant code from ADCE initialization ADCE: Dead If: Use CFGCleanup to eliminate newly dead blocks Moved bulk of CFG Cleanup code into MemPass.
2017-10-17 22:33:43 +00:00
R"(OpCapability Shader
%1 = OpExtInstImport "GLSL.std.450"
OpMemoryModel Logical GLSL450
OpEntryPoint Fragment %main "main" %BaseColor %OutColor
OpExecutionMode %main OriginUpperLeft
OpSource GLSL 450
OpName %main "main"
OpName %BaseColor "BaseColor"
OpName %d "d"
OpName %OutColor "OutColor"
OpDecorate %BaseColor Location 0
OpDecorate %OutColor Location 0
%void = OpTypeVoid
%7 = OpTypeFunction %void
%float = OpTypeFloat 32
%v4float = OpTypeVector %float 4
%_ptr_Input_v4float = OpTypePointer Input %v4float
%BaseColor = OpVariable %_ptr_Input_v4float Input
%uint = OpTypeInt 32 0
%uint_0 = OpConstant %uint 0
%_ptr_Input_float = OpTypePointer Input %float
%float_0 = OpConstant %float 0
%bool = OpTypeBool
%uint_1 = OpConstant %uint 1
%_ptr_Function_float = OpTypePointer Function %float
%float_0_25 = OpConstant %float 0.25
%float_0_5 = OpConstant %float 0.5
%float_0_75 = OpConstant %float 0.75
%_ptr_Output_v4float = OpTypePointer Output %v4float
%OutColor = OpVariable %_ptr_Output_v4float Output
%float_1 = OpConstant %float 1
%23 = OpConstantComposite %v4float %float_1 %float_1 %float_1 %float_1
Allow ADCE to remove dead inputs (#4629) * https://github.com/KhronosGroup/Vulkan-Docs/issues/666 clearly specified that interfaces do not require an input if there is an associated output * ADCE can now remove unused input variables (though they are kept if the preserve interfaces option is used)
2021-11-16 20:54:17 +00:00
; CHECK: = OpFunction
; CHECK-NEXT: = OpLabel
; CHECK-NEXT: OpBranch [[target:%\w+]]
; CHECK-NEXT: [[target]] = OpLabel
; CHECK-NOT: OpLabel
ADCE: Dead if elimination Mark structured conditional branches live only if one or more instructions in their associated construct is marked live. After closure, replace dead structured conditional branches with a branch to its merge and remove dead blocks. ADCE: Dead If Elim: Remove duplicate StructuredOrder code Also generalize ComputeStructuredOrder so that the caller can specify the root block for the order. Phi insertion uses pseudo_entry_block and adce and dead branch elim use the first block of the function. ADCE: Dead If Elim: Pull redundant code out of InsertPhiInstructions ADCE: Dead If Elim: Encapsulate CFG Cleanup Initialization ADCE: Dead If Elim: Remove redundant code from ADCE initialization ADCE: Dead If: Use CFGCleanup to eliminate newly dead blocks Moved bulk of CFG Cleanup code into MemPass.
2017-10-17 22:33:43 +00:00
Allow ADCE to remove dead inputs (#4629) * https://github.com/KhronosGroup/Vulkan-Docs/issues/666 clearly specified that interfaces do not require an input if there is an associated output * ADCE can now remove unused input variables (though they are kept if the preserve interfaces option is used)
2021-11-16 20:54:17 +00:00
%main = OpFunction %void None %7
ADCE: Dead if elimination Mark structured conditional branches live only if one or more instructions in their associated construct is marked live. After closure, replace dead structured conditional branches with a branch to its merge and remove dead blocks. ADCE: Dead If Elim: Remove duplicate StructuredOrder code Also generalize ComputeStructuredOrder so that the caller can specify the root block for the order. Phi insertion uses pseudo_entry_block and adce and dead branch elim use the first block of the function. ADCE: Dead If Elim: Pull redundant code out of InsertPhiInstructions ADCE: Dead If Elim: Encapsulate CFG Cleanup Initialization ADCE: Dead If Elim: Remove redundant code from ADCE initialization ADCE: Dead If: Use CFGCleanup to eliminate newly dead blocks Moved bulk of CFG Cleanup code into MemPass.
2017-10-17 22:33:43 +00:00
%24 = OpLabel
%d = OpVariable %_ptr_Function_float Function
%25 = OpAccessChain %_ptr_Input_float %BaseColor %uint_0
%26 = OpLoad %float %25
ADCE: Empty Loop Elimination This entirely eliminates loops which do not contain live code.
2017-10-25 00:58:48 +00:00
%27 = OpFOrdEqual %bool %26 %float_0
ADCE: Dead if elimination Mark structured conditional branches live only if one or more instructions in their associated construct is marked live. After closure, replace dead structured conditional branches with a branch to its merge and remove dead blocks. ADCE: Dead If Elim: Remove duplicate StructuredOrder code Also generalize ComputeStructuredOrder so that the caller can specify the root block for the order. Phi insertion uses pseudo_entry_block and adce and dead branch elim use the first block of the function. ADCE: Dead If Elim: Pull redundant code out of InsertPhiInstructions ADCE: Dead If Elim: Encapsulate CFG Cleanup Initialization ADCE: Dead If Elim: Remove redundant code from ADCE initialization ADCE: Dead If: Use CFGCleanup to eliminate newly dead blocks Moved bulk of CFG Cleanup code into MemPass.
2017-10-17 22:33:43 +00:00
OpSelectionMerge %28 None
OpBranchConditional %27 %29 %30
%29 = OpLabel
%31 = OpAccessChain %_ptr_Input_float %BaseColor %uint_1
%32 = OpLoad %float %31
ADCE: Empty Loop Elimination This entirely eliminates loops which do not contain live code.
2017-10-25 00:58:48 +00:00
%33 = OpFOrdEqual %bool %32 %float_0
ADCE: Dead if elimination Mark structured conditional branches live only if one or more instructions in their associated construct is marked live. After closure, replace dead structured conditional branches with a branch to its merge and remove dead blocks. ADCE: Dead If Elim: Remove duplicate StructuredOrder code Also generalize ComputeStructuredOrder so that the caller can specify the root block for the order. Phi insertion uses pseudo_entry_block and adce and dead branch elim use the first block of the function. ADCE: Dead If Elim: Pull redundant code out of InsertPhiInstructions ADCE: Dead If Elim: Encapsulate CFG Cleanup Initialization ADCE: Dead If Elim: Remove redundant code from ADCE initialization ADCE: Dead If: Use CFGCleanup to eliminate newly dead blocks Moved bulk of CFG Cleanup code into MemPass.
2017-10-17 22:33:43 +00:00
OpSelectionMerge %34 None
OpBranchConditional %33 %35 %36
%35 = OpLabel
OpStore %d %float_0
OpBranch %34
%36 = OpLabel
ADCE: Empty Loop Elimination This entirely eliminates loops which do not contain live code.
2017-10-25 00:58:48 +00:00
OpStore %d %float_0_25
ADCE: Dead if elimination Mark structured conditional branches live only if one or more instructions in their associated construct is marked live. After closure, replace dead structured conditional branches with a branch to its merge and remove dead blocks. ADCE: Dead If Elim: Remove duplicate StructuredOrder code Also generalize ComputeStructuredOrder so that the caller can specify the root block for the order. Phi insertion uses pseudo_entry_block and adce and dead branch elim use the first block of the function. ADCE: Dead If Elim: Pull redundant code out of InsertPhiInstructions ADCE: Dead If Elim: Encapsulate CFG Cleanup Initialization ADCE: Dead If Elim: Remove redundant code from ADCE initialization ADCE: Dead If: Use CFGCleanup to eliminate newly dead blocks Moved bulk of CFG Cleanup code into MemPass.
2017-10-17 22:33:43 +00:00
OpBranch %34
%34 = OpLabel
OpBranch %28
%30 = OpLabel
%37 = OpAccessChain %_ptr_Input_float %BaseColor %uint_1
%38 = OpLoad %float %37
ADCE: Empty Loop Elimination This entirely eliminates loops which do not contain live code.
2017-10-25 00:58:48 +00:00
%39 = OpFOrdEqual %bool %38 %float_0
ADCE: Dead if elimination Mark structured conditional branches live only if one or more instructions in their associated construct is marked live. After closure, replace dead structured conditional branches with a branch to its merge and remove dead blocks. ADCE: Dead If Elim: Remove duplicate StructuredOrder code Also generalize ComputeStructuredOrder so that the caller can specify the root block for the order. Phi insertion uses pseudo_entry_block and adce and dead branch elim use the first block of the function. ADCE: Dead If Elim: Pull redundant code out of InsertPhiInstructions ADCE: Dead If Elim: Encapsulate CFG Cleanup Initialization ADCE: Dead If Elim: Remove redundant code from ADCE initialization ADCE: Dead If: Use CFGCleanup to eliminate newly dead blocks Moved bulk of CFG Cleanup code into MemPass.
2017-10-17 22:33:43 +00:00
OpSelectionMerge %40 None
OpBranchConditional %39 %41 %42
%41 = OpLabel
OpStore %d %float_0_5
OpBranch %40
%42 = OpLabel
ADCE: Empty Loop Elimination This entirely eliminates loops which do not contain live code.
2017-10-25 00:58:48 +00:00
OpStore %d %float_0_75
ADCE: Dead if elimination Mark structured conditional branches live only if one or more instructions in their associated construct is marked live. After closure, replace dead structured conditional branches with a branch to its merge and remove dead blocks. ADCE: Dead If Elim: Remove duplicate StructuredOrder code Also generalize ComputeStructuredOrder so that the caller can specify the root block for the order. Phi insertion uses pseudo_entry_block and adce and dead branch elim use the first block of the function. ADCE: Dead If Elim: Pull redundant code out of InsertPhiInstructions ADCE: Dead If Elim: Encapsulate CFG Cleanup Initialization ADCE: Dead If Elim: Remove redundant code from ADCE initialization ADCE: Dead If: Use CFGCleanup to eliminate newly dead blocks Moved bulk of CFG Cleanup code into MemPass.
2017-10-17 22:33:43 +00:00
OpBranch %40
%40 = OpLabel
OpBranch %28
%28 = OpLabel
OpStore %OutColor %23
OpReturn
OpFunctionEnd
)";
Allow ADCE to remove dead inputs (#4629) * https://github.com/KhronosGroup/Vulkan-Docs/issues/666 clearly specified that interfaces do not require an input if there is an associated output * ADCE can now remove unused input variables (though they are kept if the preserve interfaces option is used)
2021-11-16 20:54:17 +00:00
SinglePassRunAndMatch<AggressiveDCEPass>(spirv, true);
ADCE: Dead if elimination Mark structured conditional branches live only if one or more instructions in their associated construct is marked live. After closure, replace dead structured conditional branches with a branch to its merge and remove dead blocks. ADCE: Dead If Elim: Remove duplicate StructuredOrder code Also generalize ComputeStructuredOrder so that the caller can specify the root block for the order. Phi insertion uses pseudo_entry_block and adce and dead branch elim use the first block of the function. ADCE: Dead If Elim: Pull redundant code out of InsertPhiInstructions ADCE: Dead If Elim: Encapsulate CFG Cleanup Initialization ADCE: Dead If Elim: Remove redundant code from ADCE initialization ADCE: Dead If: Use CFGCleanup to eliminate newly dead blocks Moved bulk of CFG Cleanup code into MemPass.
2017-10-17 22:33:43 +00:00
}
TEST_F(AggressiveDCETest, NoEliminateLiveIfThenElse) {
// #version 450
Re-format source tree - NFC. Re-formatted the source tree with the command: $ /usr/bin/clang-format -style=file -i \ $(find include source tools test utils -name '*.cpp' -or -name '*.h') This required a fix to source/val/decoration.h. It was not including spirv.h, which broke builds when the #include headers were re-ordered by clang-format.
2017-11-27 15:16:41 +00:00
//
ADCE: Dead if elimination Mark structured conditional branches live only if one or more instructions in their associated construct is marked live. After closure, replace dead structured conditional branches with a branch to its merge and remove dead blocks. ADCE: Dead If Elim: Remove duplicate StructuredOrder code Also generalize ComputeStructuredOrder so that the caller can specify the root block for the order. Phi insertion uses pseudo_entry_block and adce and dead branch elim use the first block of the function. ADCE: Dead If Elim: Pull redundant code out of InsertPhiInstructions ADCE: Dead If Elim: Encapsulate CFG Cleanup Initialization ADCE: Dead If Elim: Remove redundant code from ADCE initialization ADCE: Dead If: Use CFGCleanup to eliminate newly dead blocks Moved bulk of CFG Cleanup code into MemPass.
2017-10-17 22:33:43 +00:00
// layout(location = 0) in vec4 BaseColor;
// layout(location = 0) out vec4 OutColor;
Re-format source tree - NFC. Re-formatted the source tree with the command: $ /usr/bin/clang-format -style=file -i \ $(find include source tools test utils -name '*.cpp' -or -name '*.h') This required a fix to source/val/decoration.h. It was not including spirv.h, which broke builds when the #include headers were re-ordered by clang-format.
2017-11-27 15:16:41 +00:00
//
ADCE: Dead if elimination Mark structured conditional branches live only if one or more instructions in their associated construct is marked live. After closure, replace dead structured conditional branches with a branch to its merge and remove dead blocks. ADCE: Dead If Elim: Remove duplicate StructuredOrder code Also generalize ComputeStructuredOrder so that the caller can specify the root block for the order. Phi insertion uses pseudo_entry_block and adce and dead branch elim use the first block of the function. ADCE: Dead If Elim: Pull redundant code out of InsertPhiInstructions ADCE: Dead If Elim: Encapsulate CFG Cleanup Initialization ADCE: Dead If Elim: Remove redundant code from ADCE initialization ADCE: Dead If: Use CFGCleanup to eliminate newly dead blocks Moved bulk of CFG Cleanup code into MemPass.
2017-10-17 22:33:43 +00:00
// void main()
// {
// float t;
// if (BaseColor.x == 0)
// t = BaseColor.y;
// else
// t = BaseColor.z;
// OutColor = vec4(t);
// }
const std::string assembly =
R"(OpCapability Shader
%1 = OpExtInstImport "GLSL.std.450"
OpMemoryModel Logical GLSL450
OpEntryPoint Fragment %main "main" %BaseColor %OutColor
OpExecutionMode %main OriginUpperLeft
OpSource GLSL 450
OpName %main "main"
OpName %BaseColor "BaseColor"
OpName %t "t"
OpName %OutColor "OutColor"
OpDecorate %BaseColor Location 0
OpDecorate %OutColor Location 0
%void = OpTypeVoid
%7 = OpTypeFunction %void
%float = OpTypeFloat 32
%v4float = OpTypeVector %float 4
%_ptr_Input_v4float = OpTypePointer Input %v4float
%BaseColor = OpVariable %_ptr_Input_v4float Input
%uint = OpTypeInt 32 0
%uint_0 = OpConstant %uint 0
%_ptr_Input_float = OpTypePointer Input %float
%float_0 = OpConstant %float 0
%bool = OpTypeBool
%_ptr_Function_float = OpTypePointer Function %float
%uint_1 = OpConstant %uint 1
%uint_2 = OpConstant %uint 2
%_ptr_Output_v4float = OpTypePointer Output %v4float
%OutColor = OpVariable %_ptr_Output_v4float Output
%main = OpFunction %void None %7
%20 = OpLabel
%t = OpVariable %_ptr_Function_float Function
%21 = OpAccessChain %_ptr_Input_float %BaseColor %uint_0
%22 = OpLoad %float %21
%23 = OpFOrdEqual %bool %22 %float_0
OpSelectionMerge %24 None
OpBranchConditional %23 %25 %26
%25 = OpLabel
%27 = OpAccessChain %_ptr_Input_float %BaseColor %uint_1
%28 = OpLoad %float %27
OpStore %t %28
OpBranch %24
%26 = OpLabel
%29 = OpAccessChain %_ptr_Input_float %BaseColor %uint_2
%30 = OpLoad %float %29
OpStore %t %30
OpBranch %24
%24 = OpLabel
%31 = OpLoad %float %t
%32 = OpCompositeConstruct %v4float %31 %31 %31 %31
OpStore %OutColor %32
OpReturn
OpFunctionEnd
)";
Move tests into namespaces (#1689) This CL moves the test into namespaces based on their directories.
2018-07-11 13:24:49 +00:00
SinglePassRunAndCheck<AggressiveDCEPass>(assembly, assembly, true, true);
ADCE: Dead if elimination Mark structured conditional branches live only if one or more instructions in their associated construct is marked live. After closure, replace dead structured conditional branches with a branch to its merge and remove dead blocks. ADCE: Dead If Elim: Remove duplicate StructuredOrder code Also generalize ComputeStructuredOrder so that the caller can specify the root block for the order. Phi insertion uses pseudo_entry_block and adce and dead branch elim use the first block of the function. ADCE: Dead If Elim: Pull redundant code out of InsertPhiInstructions ADCE: Dead If Elim: Encapsulate CFG Cleanup Initialization ADCE: Dead If Elim: Remove redundant code from ADCE initialization ADCE: Dead If: Use CFGCleanup to eliminate newly dead blocks Moved bulk of CFG Cleanup code into MemPass.
2017-10-17 22:33:43 +00:00
}
TEST_F(AggressiveDCETest, NoEliminateLiveIfThenElseNested) {
// #version 450
Re-format source tree - NFC. Re-formatted the source tree with the command: $ /usr/bin/clang-format -style=file -i \ $(find include source tools test utils -name '*.cpp' -or -name '*.h') This required a fix to source/val/decoration.h. It was not including spirv.h, which broke builds when the #include headers were re-ordered by clang-format.
2017-11-27 15:16:41 +00:00
//
ADCE: Dead if elimination Mark structured conditional branches live only if one or more instructions in their associated construct is marked live. After closure, replace dead structured conditional branches with a branch to its merge and remove dead blocks. ADCE: Dead If Elim: Remove duplicate StructuredOrder code Also generalize ComputeStructuredOrder so that the caller can specify the root block for the order. Phi insertion uses pseudo_entry_block and adce and dead branch elim use the first block of the function. ADCE: Dead If Elim: Pull redundant code out of InsertPhiInstructions ADCE: Dead If Elim: Encapsulate CFG Cleanup Initialization ADCE: Dead If Elim: Remove redundant code from ADCE initialization ADCE: Dead If: Use CFGCleanup to eliminate newly dead blocks Moved bulk of CFG Cleanup code into MemPass.
2017-10-17 22:33:43 +00:00
// layout(location = 0) in vec4 BaseColor;
// layout(location = 0) out vec4 OutColor;
Re-format source tree - NFC. Re-formatted the source tree with the command: $ /usr/bin/clang-format -style=file -i \ $(find include source tools test utils -name '*.cpp' -or -name '*.h') This required a fix to source/val/decoration.h. It was not including spirv.h, which broke builds when the #include headers were re-ordered by clang-format.
2017-11-27 15:16:41 +00:00
//
ADCE: Dead if elimination Mark structured conditional branches live only if one or more instructions in their associated construct is marked live. After closure, replace dead structured conditional branches with a branch to its merge and remove dead blocks. ADCE: Dead If Elim: Remove duplicate StructuredOrder code Also generalize ComputeStructuredOrder so that the caller can specify the root block for the order. Phi insertion uses pseudo_entry_block and adce and dead branch elim use the first block of the function. ADCE: Dead If Elim: Pull redundant code out of InsertPhiInstructions ADCE: Dead If Elim: Encapsulate CFG Cleanup Initialization ADCE: Dead If Elim: Remove redundant code from ADCE initialization ADCE: Dead If: Use CFGCleanup to eliminate newly dead blocks Moved bulk of CFG Cleanup code into MemPass.
2017-10-17 22:33:43 +00:00
// void main()
// {
// float t;
// if (BaseColor.x == 0)
// if (BaseColor.y == 0)
// t = 0.0;
// else
// t = 0.25;
// else
// if (BaseColor.y == 0)
// t = 0.5;
// else
// t = 0.75;
// OutColor = vec4(t);
// }
const std::string assembly =
R"(OpCapability Shader
%1 = OpExtInstImport "GLSL.std.450"
OpMemoryModel Logical GLSL450
OpEntryPoint Fragment %main "main" %BaseColor %OutColor
OpExecutionMode %main OriginUpperLeft
OpSource GLSL 450
OpName %main "main"
OpName %BaseColor "BaseColor"
OpName %t "t"
OpName %OutColor "OutColor"
OpDecorate %BaseColor Location 0
OpDecorate %OutColor Location 0
%void = OpTypeVoid
%7 = OpTypeFunction %void
%float = OpTypeFloat 32
%v4float = OpTypeVector %float 4
%_ptr_Input_v4float = OpTypePointer Input %v4float
%BaseColor = OpVariable %_ptr_Input_v4float Input
%uint = OpTypeInt 32 0
%uint_0 = OpConstant %uint 0
%_ptr_Input_float = OpTypePointer Input %float
%float_0 = OpConstant %float 0
%bool = OpTypeBool
%uint_1 = OpConstant %uint 1
%_ptr_Function_float = OpTypePointer Function %float
%float_0_25 = OpConstant %float 0.25
%float_0_5 = OpConstant %float 0.5
%float_0_75 = OpConstant %float 0.75
%_ptr_Output_v4float = OpTypePointer Output %v4float
%OutColor = OpVariable %_ptr_Output_v4float Output
%main = OpFunction %void None %7
%22 = OpLabel
%t = OpVariable %_ptr_Function_float Function
%23 = OpAccessChain %_ptr_Input_float %BaseColor %uint_0
%24 = OpLoad %float %23
%25 = OpFOrdEqual %bool %24 %float_0
OpSelectionMerge %26 None
OpBranchConditional %25 %27 %28
%27 = OpLabel
%29 = OpAccessChain %_ptr_Input_float %BaseColor %uint_1
%30 = OpLoad %float %29
%31 = OpFOrdEqual %bool %30 %float_0
OpSelectionMerge %32 None
OpBranchConditional %31 %33 %34
%33 = OpLabel
OpStore %t %float_0
OpBranch %32
%34 = OpLabel
OpStore %t %float_0_25
OpBranch %32
%32 = OpLabel
OpBranch %26
%28 = OpLabel
%35 = OpAccessChain %_ptr_Input_float %BaseColor %uint_1
%36 = OpLoad %float %35
%37 = OpFOrdEqual %bool %36 %float_0
OpSelectionMerge %38 None
OpBranchConditional %37 %39 %40
%39 = OpLabel
OpStore %t %float_0_5
OpBranch %38
%40 = OpLabel
OpStore %t %float_0_75
OpBranch %38
%38 = OpLabel
OpBranch %26
%26 = OpLabel
%41 = OpLoad %float %t
%42 = OpCompositeConstruct %v4float %41 %41 %41 %41
OpStore %OutColor %42
OpReturn
OpFunctionEnd
)";
Move tests into namespaces (#1689) This CL moves the test into namespaces based on their directories.
2018-07-11 13:24:49 +00:00
SinglePassRunAndCheck<AggressiveDCEPass>(assembly, assembly, true, true);
ADCE: Dead if elimination Mark structured conditional branches live only if one or more instructions in their associated construct is marked live. After closure, replace dead structured conditional branches with a branch to its merge and remove dead blocks. ADCE: Dead If Elim: Remove duplicate StructuredOrder code Also generalize ComputeStructuredOrder so that the caller can specify the root block for the order. Phi insertion uses pseudo_entry_block and adce and dead branch elim use the first block of the function. ADCE: Dead If Elim: Pull redundant code out of InsertPhiInstructions ADCE: Dead If Elim: Encapsulate CFG Cleanup Initialization ADCE: Dead If Elim: Remove redundant code from ADCE initialization ADCE: Dead If: Use CFGCleanup to eliminate newly dead blocks Moved bulk of CFG Cleanup code into MemPass.
2017-10-17 22:33:43 +00:00
}
TEST_F(AggressiveDCETest, NoEliminateIfWithPhi) {
// Note: Assembly hand-optimized from GLSL
//
// #version 450
Re-format source tree - NFC. Re-formatted the source tree with the command: $ /usr/bin/clang-format -style=file -i \ $(find include source tools test utils -name '*.cpp' -or -name '*.h') This required a fix to source/val/decoration.h. It was not including spirv.h, which broke builds when the #include headers were re-ordered by clang-format.
2017-11-27 15:16:41 +00:00
//
ADCE: Dead if elimination Mark structured conditional branches live only if one or more instructions in their associated construct is marked live. After closure, replace dead structured conditional branches with a branch to its merge and remove dead blocks. ADCE: Dead If Elim: Remove duplicate StructuredOrder code Also generalize ComputeStructuredOrder so that the caller can specify the root block for the order. Phi insertion uses pseudo_entry_block and adce and dead branch elim use the first block of the function. ADCE: Dead If Elim: Pull redundant code out of InsertPhiInstructions ADCE: Dead If Elim: Encapsulate CFG Cleanup Initialization ADCE: Dead If Elim: Remove redundant code from ADCE initialization ADCE: Dead If: Use CFGCleanup to eliminate newly dead blocks Moved bulk of CFG Cleanup code into MemPass.
2017-10-17 22:33:43 +00:00
// layout(location = 0) in vec4 BaseColor;
// layout(location = 0) out vec4 OutColor;
Re-format source tree - NFC. Re-formatted the source tree with the command: $ /usr/bin/clang-format -style=file -i \ $(find include source tools test utils -name '*.cpp' -or -name '*.h') This required a fix to source/val/decoration.h. It was not including spirv.h, which broke builds when the #include headers were re-ordered by clang-format.
2017-11-27 15:16:41 +00:00
//
ADCE: Dead if elimination Mark structured conditional branches live only if one or more instructions in their associated construct is marked live. After closure, replace dead structured conditional branches with a branch to its merge and remove dead blocks. ADCE: Dead If Elim: Remove duplicate StructuredOrder code Also generalize ComputeStructuredOrder so that the caller can specify the root block for the order. Phi insertion uses pseudo_entry_block and adce and dead branch elim use the first block of the function. ADCE: Dead If Elim: Pull redundant code out of InsertPhiInstructions ADCE: Dead If Elim: Encapsulate CFG Cleanup Initialization ADCE: Dead If Elim: Remove redundant code from ADCE initialization ADCE: Dead If: Use CFGCleanup to eliminate newly dead blocks Moved bulk of CFG Cleanup code into MemPass.
2017-10-17 22:33:43 +00:00
// void main()
// {
// float t;
// if (BaseColor.x == 0)
// t = 0.0;
// else
// t = 1.0;
// OutColor = vec4(t);
// }
const std::string assembly =
R"(OpCapability Shader
%1 = OpExtInstImport "GLSL.std.450"
OpMemoryModel Logical GLSL450
OpEntryPoint Fragment %main "main" %BaseColor %OutColor
OpExecutionMode %main OriginUpperLeft
OpSource GLSL 450
OpName %main "main"
OpName %BaseColor "BaseColor"
OpName %OutColor "OutColor"
OpDecorate %BaseColor Location 0
OpDecorate %OutColor Location 0
%void = OpTypeVoid
%6 = OpTypeFunction %void
%float = OpTypeFloat 32
%v4float = OpTypeVector %float 4
%_ptr_Input_v4float = OpTypePointer Input %v4float
%BaseColor = OpVariable %_ptr_Input_v4float Input
%uint = OpTypeInt 32 0
%uint_0 = OpConstant %uint 0
%_ptr_Input_float = OpTypePointer Input %float
%float_0 = OpConstant %float 0
%bool = OpTypeBool
%float_1 = OpConstant %float 1
%_ptr_Output_v4float = OpTypePointer Output %v4float
%OutColor = OpVariable %_ptr_Output_v4float Output
%main = OpFunction %void None %6
Adding additional functionality to ADCE. Modified ADCE to remove dead globals. * Entry point and execution mode instructions are marked as alive * Reachable functions and their parameters are marked as alive * Instruction deletion now deferred until the end of the pass * Eliminated dead insts set, added IsDead to calculate that value instead * Ported applicable dead variable elimination tests * Ported dead constant elim tests Added dead function elimination to ADCE * ported dead function elim tests Added handling of decoration groups in ADCE * Uses a custom sorter to traverse decorations in a specific order * Simplifies necessary checks Updated -O and -Os pass lists.
2017-12-18 17:13:10 +00:00
%17 = OpLabel
%18 = OpAccessChain %_ptr_Input_float %BaseColor %uint_0
%19 = OpLoad %float %18
%20 = OpFOrdEqual %bool %19 %float_0
OpSelectionMerge %21 None
OpBranchConditional %20 %22 %23
ADCE: Dead if elimination Mark structured conditional branches live only if one or more instructions in their associated construct is marked live. After closure, replace dead structured conditional branches with a branch to its merge and remove dead blocks. ADCE: Dead If Elim: Remove duplicate StructuredOrder code Also generalize ComputeStructuredOrder so that the caller can specify the root block for the order. Phi insertion uses pseudo_entry_block and adce and dead branch elim use the first block of the function. ADCE: Dead If Elim: Pull redundant code out of InsertPhiInstructions ADCE: Dead If Elim: Encapsulate CFG Cleanup Initialization ADCE: Dead If Elim: Remove redundant code from ADCE initialization ADCE: Dead If: Use CFGCleanup to eliminate newly dead blocks Moved bulk of CFG Cleanup code into MemPass.
2017-10-17 22:33:43 +00:00
%22 = OpLabel
Adding additional functionality to ADCE. Modified ADCE to remove dead globals. * Entry point and execution mode instructions are marked as alive * Reachable functions and their parameters are marked as alive * Instruction deletion now deferred until the end of the pass * Eliminated dead insts set, added IsDead to calculate that value instead * Ported applicable dead variable elimination tests * Ported dead constant elim tests Added dead function elimination to ADCE * ported dead function elim tests Added handling of decoration groups in ADCE * Uses a custom sorter to traverse decorations in a specific order * Simplifies necessary checks Updated -O and -Os pass lists.
2017-12-18 17:13:10 +00:00
OpBranch %21
%23 = OpLabel
OpBranch %21
%21 = OpLabel
%24 = OpPhi %float %float_0 %22 %float_1 %23
%25 = OpCompositeConstruct %v4float %24 %24 %24 %24
OpStore %OutColor %25
ADCE: Dead if elimination Mark structured conditional branches live only if one or more instructions in their associated construct is marked live. After closure, replace dead structured conditional branches with a branch to its merge and remove dead blocks. ADCE: Dead If Elim: Remove duplicate StructuredOrder code Also generalize ComputeStructuredOrder so that the caller can specify the root block for the order. Phi insertion uses pseudo_entry_block and adce and dead branch elim use the first block of the function. ADCE: Dead If Elim: Pull redundant code out of InsertPhiInstructions ADCE: Dead If Elim: Encapsulate CFG Cleanup Initialization ADCE: Dead If Elim: Remove redundant code from ADCE initialization ADCE: Dead If: Use CFGCleanup to eliminate newly dead blocks Moved bulk of CFG Cleanup code into MemPass.
2017-10-17 22:33:43 +00:00
OpReturn
OpFunctionEnd
)";
Move tests into namespaces (#1689) This CL moves the test into namespaces based on their directories.
2018-07-11 13:24:49 +00:00
SinglePassRunAndCheck<AggressiveDCEPass>(assembly, assembly, true, true);
ADCE: Dead if elimination Mark structured conditional branches live only if one or more instructions in their associated construct is marked live. After closure, replace dead structured conditional branches with a branch to its merge and remove dead blocks. ADCE: Dead If Elim: Remove duplicate StructuredOrder code Also generalize ComputeStructuredOrder so that the caller can specify the root block for the order. Phi insertion uses pseudo_entry_block and adce and dead branch elim use the first block of the function. ADCE: Dead If Elim: Pull redundant code out of InsertPhiInstructions ADCE: Dead If Elim: Encapsulate CFG Cleanup Initialization ADCE: Dead If Elim: Remove redundant code from ADCE initialization ADCE: Dead If: Use CFGCleanup to eliminate newly dead blocks Moved bulk of CFG Cleanup code into MemPass.
2017-10-17 22:33:43 +00:00
}
AggressiveDCE: Add merge and continue branches for live loop. This ensures that an if-break is not eliminated from a loop. This fixes issue #989
2017-11-28 23:18:05 +00:00
TEST_F(AggressiveDCETest, NoEliminateIfBreak) {
// Note: Assembly optimized from GLSL
//
ADCE: Empty Loop Elimination This entirely eliminates loops which do not contain live code.
2017-10-25 00:58:48 +00:00
// #version 450
//
AggressiveDCE: Add merge and continue branches for live loop. This ensures that an if-break is not eliminated from a loop. This fixes issue #989
2017-11-28 23:18:05 +00:00
// layout(location=0) in vec4 InColor;
// layout(location=0) out vec4 OutColor;
ADCE: Empty Loop Elimination This entirely eliminates loops which do not contain live code.
2017-10-25 00:58:48 +00:00
//
AggressiveDCE: Add merge and continue branches for live loop. This ensures that an if-break is not eliminated from a loop. This fixes issue #989
2017-11-28 23:18:05 +00:00
// void main()
// {
// float f = 0.0;
// for (;;) {
// f += 2.0;
// if (f > 20.0)
// break;
// }
ADCE: Empty Loop Elimination This entirely eliminates loops which do not contain live code.
2017-10-25 00:58:48 +00:00
//
AggressiveDCE: Add merge and continue branches for live loop. This ensures that an if-break is not eliminated from a loop. This fixes issue #989
2017-11-28 23:18:05 +00:00
// OutColor = InColor / f;
// }
const std::string assembly =
R"(OpCapability Shader
%1 = OpExtInstImport "GLSL.std.450"
OpMemoryModel Logical GLSL450
OpEntryPoint Fragment %main "main" %OutColor %InColor
OpExecutionMode %main OriginUpperLeft
OpSource GLSL 450
OpName %main "main"
OpName %f "f"
OpName %OutColor "OutColor"
OpName %InColor "InColor"
OpDecorate %OutColor Location 0
OpDecorate %InColor Location 0
%void = OpTypeVoid
%7 = OpTypeFunction %void
%float = OpTypeFloat 32
%_ptr_Function_float = OpTypePointer Function %float
%float_0 = OpConstant %float 0
%float_2 = OpConstant %float 2
%float_20 = OpConstant %float 20
%bool = OpTypeBool
%v4float = OpTypeVector %float 4
%_ptr_Output_v4float = OpTypePointer Output %v4float
%OutColor = OpVariable %_ptr_Output_v4float Output
%_ptr_Input_v4float = OpTypePointer Input %v4float
%InColor = OpVariable %_ptr_Input_v4float Input
%main = OpFunction %void None %7
%17 = OpLabel
%f = OpVariable %_ptr_Function_float Function
OpStore %f %float_0
OpBranch %18
%18 = OpLabel
OpLoopMerge %19 %20 None
OpBranch %21
%21 = OpLabel
%22 = OpLoad %float %f
%23 = OpFAdd %float %22 %float_2
OpStore %f %23
%24 = OpLoad %float %f
%25 = OpFOrdGreaterThan %bool %24 %float_20
OpSelectionMerge %26 None
OpBranchConditional %25 %27 %26
%27 = OpLabel
OpBranch %19
%26 = OpLabel
OpBranch %20
%20 = OpLabel
OpBranch %18
%19 = OpLabel
%28 = OpLoad %v4float %InColor
%29 = OpLoad %float %f
%30 = OpCompositeConstruct %v4float %29 %29 %29 %29
%31 = OpFDiv %v4float %28 %30
OpStore %OutColor %31
OpReturn
OpFunctionEnd
)";
Move tests into namespaces (#1689) This CL moves the test into namespaces based on their directories.
2018-07-11 13:24:49 +00:00
SinglePassRunAndCheck<AggressiveDCEPass>(assembly, assembly, true, true);
AggressiveDCE: Add merge and continue branches for live loop. This ensures that an if-break is not eliminated from a loop. This fixes issue #989
2017-11-28 23:18:05 +00:00
}
ADCE: Only mark true breaks and continues of live loops This fixes issue #1075 - Mark continue when conditional branch with merge block. Only mark if merge block is not continue block. - Handle conditional branch break with preceding merge
2017-12-12 21:27:46 +00:00
TEST_F(AggressiveDCETest, NoEliminateIfBreak2) {
// Do not eliminate break as conditional branch with merge instruction
// Note: SPIR-V edited to add merge instruction before break.
//
// #version 430
//
// layout(std430) buffer U_t
// {
// float g_F[10];
// };
//
// layout(location = 0)out float o;
//
// void main(void)
// {
// float s = 0.0;
// for (int i=0; i<10; i++)
// s += g_F[i];
// o = s;
// }
const std::string assembly =
R"(OpCapability Shader
%1 = OpExtInstImport "GLSL.std.450"
OpMemoryModel Logical GLSL450
OpEntryPoint Fragment %main "main" %o
OpExecutionMode %main OriginUpperLeft
OpSource GLSL 430
OpName %main "main"
OpName %s "s"
OpName %i "i"
OpName %U_t "U_t"
OpMemberName %U_t 0 "g_F"
OpName %_ ""
OpName %o "o"
OpDecorate %_arr_float_uint_10 ArrayStride 4
OpMemberDecorate %U_t 0 Offset 0
OpDecorate %U_t BufferBlock
OpDecorate %_ DescriptorSet 0
OpDecorate %o Location 0
%void = OpTypeVoid
%10 = OpTypeFunction %void
%float = OpTypeFloat 32
%_ptr_Function_float = OpTypePointer Function %float
%float_0 = OpConstant %float 0
%int = OpTypeInt 32 1
%_ptr_Function_int = OpTypePointer Function %int
%int_0 = OpConstant %int 0
%int_10 = OpConstant %int 10
%bool = OpTypeBool
%uint = OpTypeInt 32 0
%uint_10 = OpConstant %uint 10
%_arr_float_uint_10 = OpTypeArray %float %uint_10
%U_t = OpTypeStruct %_arr_float_uint_10
%_ptr_Uniform_U_t = OpTypePointer Uniform %U_t
%_ = OpVariable %_ptr_Uniform_U_t Uniform
%_ptr_Uniform_float = OpTypePointer Uniform %float
%int_1 = OpConstant %int 1
%_ptr_Output_float = OpTypePointer Output %float
%o = OpVariable %_ptr_Output_float Output
%main = OpFunction %void None %10
Adding additional functionality to ADCE. Modified ADCE to remove dead globals. * Entry point and execution mode instructions are marked as alive * Reachable functions and their parameters are marked as alive * Instruction deletion now deferred until the end of the pass * Eliminated dead insts set, added IsDead to calculate that value instead * Ported applicable dead variable elimination tests * Ported dead constant elim tests Added dead function elimination to ADCE * ported dead function elim tests Added handling of decoration groups in ADCE * Uses a custom sorter to traverse decorations in a specific order * Simplifies necessary checks Updated -O and -Os pass lists.
2017-12-18 17:13:10 +00:00
%25 = OpLabel
ADCE: Only mark true breaks and continues of live loops This fixes issue #1075 - Mark continue when conditional branch with merge block. Only mark if merge block is not continue block. - Handle conditional branch break with preceding merge
2017-12-12 21:27:46 +00:00
%s = OpVariable %_ptr_Function_float Function
%i = OpVariable %_ptr_Function_int Function
OpStore %s %float_0
OpStore %i %int_0
Adding additional functionality to ADCE. Modified ADCE to remove dead globals. * Entry point and execution mode instructions are marked as alive * Reachable functions and their parameters are marked as alive * Instruction deletion now deferred until the end of the pass * Eliminated dead insts set, added IsDead to calculate that value instead * Ported applicable dead variable elimination tests * Ported dead constant elim tests Added dead function elimination to ADCE * ported dead function elim tests Added handling of decoration groups in ADCE * Uses a custom sorter to traverse decorations in a specific order * Simplifies necessary checks Updated -O and -Os pass lists.
2017-12-18 17:13:10 +00:00
OpBranch %26
%26 = OpLabel
OpLoopMerge %27 %28 None
ADCE: Only mark true breaks and continues of live loops This fixes issue #1075 - Mark continue when conditional branch with merge block. Only mark if merge block is not continue block. - Handle conditional branch break with preceding merge
2017-12-12 21:27:46 +00:00
OpBranch %29
%29 = OpLabel
Adding additional functionality to ADCE. Modified ADCE to remove dead globals. * Entry point and execution mode instructions are marked as alive * Reachable functions and their parameters are marked as alive * Instruction deletion now deferred until the end of the pass * Eliminated dead insts set, added IsDead to calculate that value instead * Ported applicable dead variable elimination tests * Ported dead constant elim tests Added dead function elimination to ADCE * ported dead function elim tests Added handling of decoration groups in ADCE * Uses a custom sorter to traverse decorations in a specific order * Simplifies necessary checks Updated -O and -Os pass lists.
2017-12-18 17:13:10 +00:00
%30 = OpLoad %int %i
%31 = OpSLessThan %bool %30 %int_10
OpSelectionMerge %32 None
OpBranchConditional %31 %32 %27
%32 = OpLabel
%33 = OpLoad %int %i
%34 = OpAccessChain %_ptr_Uniform_float %_ %int_0 %33
%35 = OpLoad %float %34
%36 = OpLoad %float %s
%37 = OpFAdd %float %36 %35
OpStore %s %37
OpBranch %28
ADCE: Only mark true breaks and continues of live loops This fixes issue #1075 - Mark continue when conditional branch with merge block. Only mark if merge block is not continue block. - Handle conditional branch break with preceding merge
2017-12-12 21:27:46 +00:00
%28 = OpLabel
Adding additional functionality to ADCE. Modified ADCE to remove dead globals. * Entry point and execution mode instructions are marked as alive * Reachable functions and their parameters are marked as alive * Instruction deletion now deferred until the end of the pass * Eliminated dead insts set, added IsDead to calculate that value instead * Ported applicable dead variable elimination tests * Ported dead constant elim tests Added dead function elimination to ADCE * ported dead function elim tests Added handling of decoration groups in ADCE * Uses a custom sorter to traverse decorations in a specific order * Simplifies necessary checks Updated -O and -Os pass lists.
2017-12-18 17:13:10 +00:00
%38 = OpLoad %int %i
%39 = OpIAdd %int %38 %int_1
OpStore %i %39
OpBranch %26
%27 = OpLabel
%40 = OpLoad %float %s
OpStore %o %40
ADCE: Only mark true breaks and continues of live loops This fixes issue #1075 - Mark continue when conditional branch with merge block. Only mark if merge block is not continue block. - Handle conditional branch break with preceding merge
2017-12-12 21:27:46 +00:00
OpReturn
OpFunctionEnd
)";
Move tests into namespaces (#1689) This CL moves the test into namespaces based on their directories.
2018-07-11 13:24:49 +00:00
SinglePassRunAndCheck<AggressiveDCEPass>(assembly, assembly, true, true);
ADCE: Only mark true breaks and continues of live loops This fixes issue #1075 - Mark continue when conditional branch with merge block. Only mark if merge block is not continue block. - Handle conditional branch break with preceding merge
2017-12-12 21:27:46 +00:00
}
ADCE: Empty Loop Elimination This entirely eliminates loops which do not contain live code.
2017-10-25 00:58:48 +00:00
TEST_F(AggressiveDCETest, EliminateEntireUselessLoop) {
// #version 140
// in vec4 BaseColor;
Re-format source tree - NFC. Re-formatted the source tree with the command: $ /usr/bin/clang-format -style=file -i \ $(find include source tools test utils -name '*.cpp' -or -name '*.h') This required a fix to source/val/decoration.h. It was not including spirv.h, which broke builds when the #include headers were re-ordered by clang-format.
2017-11-27 15:16:41 +00:00
//
ADCE: Empty Loop Elimination This entirely eliminates loops which do not contain live code.
2017-10-25 00:58:48 +00:00
// layout(std140) uniform U_t
// {
// int g_I ;
// } ;
Re-format source tree - NFC. Re-formatted the source tree with the command: $ /usr/bin/clang-format -style=file -i \ $(find include source tools test utils -name '*.cpp' -or -name '*.h') This required a fix to source/val/decoration.h. It was not including spirv.h, which broke builds when the #include headers were re-ordered by clang-format.
2017-11-27 15:16:41 +00:00
//
ADCE: Dead if elimination Mark structured conditional branches live only if one or more instructions in their associated construct is marked live. After closure, replace dead structured conditional branches with a branch to its merge and remove dead blocks. ADCE: Dead If Elim: Remove duplicate StructuredOrder code Also generalize ComputeStructuredOrder so that the caller can specify the root block for the order. Phi insertion uses pseudo_entry_block and adce and dead branch elim use the first block of the function. ADCE: Dead If Elim: Pull redundant code out of InsertPhiInstructions ADCE: Dead If Elim: Encapsulate CFG Cleanup Initialization ADCE: Dead If Elim: Remove redundant code from ADCE initialization ADCE: Dead If: Use CFGCleanup to eliminate newly dead blocks Moved bulk of CFG Cleanup code into MemPass.
2017-10-17 22:33:43 +00:00
// void main()
// {
ADCE: Empty Loop Elimination This entirely eliminates loops which do not contain live code.
2017-10-25 00:58:48 +00:00
// vec4 v = BaseColor;
// float df = 0.0;
// int i = 0;
// while (i < g_I) {
// df = df * 0.5;
// i = i + 1;
// }
// gl_FragColor = v;
ADCE: Dead if elimination Mark structured conditional branches live only if one or more instructions in their associated construct is marked live. After closure, replace dead structured conditional branches with a branch to its merge and remove dead blocks. ADCE: Dead If Elim: Remove duplicate StructuredOrder code Also generalize ComputeStructuredOrder so that the caller can specify the root block for the order. Phi insertion uses pseudo_entry_block and adce and dead branch elim use the first block of the function. ADCE: Dead If Elim: Pull redundant code out of InsertPhiInstructions ADCE: Dead If Elim: Encapsulate CFG Cleanup Initialization ADCE: Dead If Elim: Remove redundant code from ADCE initialization ADCE: Dead If: Use CFGCleanup to eliminate newly dead blocks Moved bulk of CFG Cleanup code into MemPass.
2017-10-17 22:33:43 +00:00
// }
ADCE: Empty Loop Elimination This entirely eliminates loops which do not contain live code.
2017-10-25 00:58:48 +00:00
const std::string predefs1 =
ADCE: Dead if elimination Mark structured conditional branches live only if one or more instructions in their associated construct is marked live. After closure, replace dead structured conditional branches with a branch to its merge and remove dead blocks. ADCE: Dead If Elim: Remove duplicate StructuredOrder code Also generalize ComputeStructuredOrder so that the caller can specify the root block for the order. Phi insertion uses pseudo_entry_block and adce and dead branch elim use the first block of the function. ADCE: Dead If Elim: Pull redundant code out of InsertPhiInstructions ADCE: Dead If Elim: Encapsulate CFG Cleanup Initialization ADCE: Dead If Elim: Remove redundant code from ADCE initialization ADCE: Dead If: Use CFGCleanup to eliminate newly dead blocks Moved bulk of CFG Cleanup code into MemPass.
2017-10-17 22:33:43 +00:00
R"(OpCapability Shader
%1 = OpExtInstImport "GLSL.std.450"
OpMemoryModel Logical GLSL450
ADCE: Empty Loop Elimination This entirely eliminates loops which do not contain live code.
2017-10-25 00:58:48 +00:00
OpEntryPoint Fragment %main "main" %BaseColor %gl_FragColor
ADCE: Dead if elimination Mark structured conditional branches live only if one or more instructions in their associated construct is marked live. After closure, replace dead structured conditional branches with a branch to its merge and remove dead blocks. ADCE: Dead If Elim: Remove duplicate StructuredOrder code Also generalize ComputeStructuredOrder so that the caller can specify the root block for the order. Phi insertion uses pseudo_entry_block and adce and dead branch elim use the first block of the function. ADCE: Dead If Elim: Pull redundant code out of InsertPhiInstructions ADCE: Dead If Elim: Encapsulate CFG Cleanup Initialization ADCE: Dead If Elim: Remove redundant code from ADCE initialization ADCE: Dead If: Use CFGCleanup to eliminate newly dead blocks Moved bulk of CFG Cleanup code into MemPass.
2017-10-17 22:33:43 +00:00
OpExecutionMode %main OriginUpperLeft
ADCE: Empty Loop Elimination This entirely eliminates loops which do not contain live code.
2017-10-25 00:58:48 +00:00
OpSource GLSL 140
)";
const std::string names_before =
R"(OpName %main "main"
OpName %v "v"
ADCE: Dead if elimination Mark structured conditional branches live only if one or more instructions in their associated construct is marked live. After closure, replace dead structured conditional branches with a branch to its merge and remove dead blocks. ADCE: Dead If Elim: Remove duplicate StructuredOrder code Also generalize ComputeStructuredOrder so that the caller can specify the root block for the order. Phi insertion uses pseudo_entry_block and adce and dead branch elim use the first block of the function. ADCE: Dead If Elim: Pull redundant code out of InsertPhiInstructions ADCE: Dead If Elim: Encapsulate CFG Cleanup Initialization ADCE: Dead If Elim: Remove redundant code from ADCE initialization ADCE: Dead If: Use CFGCleanup to eliminate newly dead blocks Moved bulk of CFG Cleanup code into MemPass.
2017-10-17 22:33:43 +00:00
OpName %BaseColor "BaseColor"
ADCE: Empty Loop Elimination This entirely eliminates loops which do not contain live code.
2017-10-25 00:58:48 +00:00
OpName %df "df"
OpName %i "i"
OpName %U_t "U_t"
OpMemberName %U_t 0 "g_I"
OpName %_ ""
OpName %gl_FragColor "gl_FragColor"
)";
const std::string names_after =
R"(OpName %main "main"
OpName %v "v"
OpName %BaseColor "BaseColor"
OpName %gl_FragColor "gl_FragColor"
)";
Adding additional functionality to ADCE. Modified ADCE to remove dead globals. * Entry point and execution mode instructions are marked as alive * Reachable functions and their parameters are marked as alive * Instruction deletion now deferred until the end of the pass * Eliminated dead insts set, added IsDead to calculate that value instead * Ported applicable dead variable elimination tests * Ported dead constant elim tests Added dead function elimination to ADCE * ported dead function elim tests Added handling of decoration groups in ADCE * Uses a custom sorter to traverse decorations in a specific order * Simplifies necessary checks Updated -O and -Os pass lists.
2017-12-18 17:13:10 +00:00
const std::string predefs2_before =
ADCE: Empty Loop Elimination This entirely eliminates loops which do not contain live code.
2017-10-25 00:58:48 +00:00
R"(OpMemberDecorate %U_t 0 Offset 0
OpDecorate %U_t Block
OpDecorate %_ DescriptorSet 0
ADCE: Dead if elimination Mark structured conditional branches live only if one or more instructions in their associated construct is marked live. After closure, replace dead structured conditional branches with a branch to its merge and remove dead blocks. ADCE: Dead If Elim: Remove duplicate StructuredOrder code Also generalize ComputeStructuredOrder so that the caller can specify the root block for the order. Phi insertion uses pseudo_entry_block and adce and dead branch elim use the first block of the function. ADCE: Dead If Elim: Pull redundant code out of InsertPhiInstructions ADCE: Dead If Elim: Encapsulate CFG Cleanup Initialization ADCE: Dead If Elim: Remove redundant code from ADCE initialization ADCE: Dead If: Use CFGCleanup to eliminate newly dead blocks Moved bulk of CFG Cleanup code into MemPass.
2017-10-17 22:33:43 +00:00
%void = OpTypeVoid
ADCE: Empty Loop Elimination This entirely eliminates loops which do not contain live code.
2017-10-25 00:58:48 +00:00
%11 = OpTypeFunction %void
ADCE: Dead if elimination Mark structured conditional branches live only if one or more instructions in their associated construct is marked live. After closure, replace dead structured conditional branches with a branch to its merge and remove dead blocks. ADCE: Dead If Elim: Remove duplicate StructuredOrder code Also generalize ComputeStructuredOrder so that the caller can specify the root block for the order. Phi insertion uses pseudo_entry_block and adce and dead branch elim use the first block of the function. ADCE: Dead If Elim: Pull redundant code out of InsertPhiInstructions ADCE: Dead If Elim: Encapsulate CFG Cleanup Initialization ADCE: Dead If Elim: Remove redundant code from ADCE initialization ADCE: Dead If: Use CFGCleanup to eliminate newly dead blocks Moved bulk of CFG Cleanup code into MemPass.
2017-10-17 22:33:43 +00:00
%float = OpTypeFloat 32
%v4float = OpTypeVector %float 4
ADCE: Empty Loop Elimination This entirely eliminates loops which do not contain live code.
2017-10-25 00:58:48 +00:00
%_ptr_Function_v4float = OpTypePointer Function %v4float
ADCE: Dead if elimination Mark structured conditional branches live only if one or more instructions in their associated construct is marked live. After closure, replace dead structured conditional branches with a branch to its merge and remove dead blocks. ADCE: Dead If Elim: Remove duplicate StructuredOrder code Also generalize ComputeStructuredOrder so that the caller can specify the root block for the order. Phi insertion uses pseudo_entry_block and adce and dead branch elim use the first block of the function. ADCE: Dead If Elim: Pull redundant code out of InsertPhiInstructions ADCE: Dead If Elim: Encapsulate CFG Cleanup Initialization ADCE: Dead If Elim: Remove redundant code from ADCE initialization ADCE: Dead If: Use CFGCleanup to eliminate newly dead blocks Moved bulk of CFG Cleanup code into MemPass.
2017-10-17 22:33:43 +00:00
%_ptr_Input_v4float = OpTypePointer Input %v4float
%BaseColor = OpVariable %_ptr_Input_v4float Input
ADCE: Empty Loop Elimination This entirely eliminates loops which do not contain live code.
2017-10-25 00:58:48 +00:00
%_ptr_Function_float = OpTypePointer Function %float
ADCE: Dead if elimination Mark structured conditional branches live only if one or more instructions in their associated construct is marked live. After closure, replace dead structured conditional branches with a branch to its merge and remove dead blocks. ADCE: Dead If Elim: Remove duplicate StructuredOrder code Also generalize ComputeStructuredOrder so that the caller can specify the root block for the order. Phi insertion uses pseudo_entry_block and adce and dead branch elim use the first block of the function. ADCE: Dead If Elim: Pull redundant code out of InsertPhiInstructions ADCE: Dead If Elim: Encapsulate CFG Cleanup Initialization ADCE: Dead If Elim: Remove redundant code from ADCE initialization ADCE: Dead If: Use CFGCleanup to eliminate newly dead blocks Moved bulk of CFG Cleanup code into MemPass.
2017-10-17 22:33:43 +00:00
%float_0 = OpConstant %float 0
ADCE: Empty Loop Elimination This entirely eliminates loops which do not contain live code.
2017-10-25 00:58:48 +00:00
%int = OpTypeInt 32 1
%_ptr_Function_int = OpTypePointer Function %int
%int_0 = OpConstant %int 0
%U_t = OpTypeStruct %int
%_ptr_Uniform_U_t = OpTypePointer Uniform %U_t
%_ = OpVariable %_ptr_Uniform_U_t Uniform
%_ptr_Uniform_int = OpTypePointer Uniform %int
ADCE: Dead if elimination Mark structured conditional branches live only if one or more instructions in their associated construct is marked live. After closure, replace dead structured conditional branches with a branch to its merge and remove dead blocks. ADCE: Dead If Elim: Remove duplicate StructuredOrder code Also generalize ComputeStructuredOrder so that the caller can specify the root block for the order. Phi insertion uses pseudo_entry_block and adce and dead branch elim use the first block of the function. ADCE: Dead If Elim: Pull redundant code out of InsertPhiInstructions ADCE: Dead If Elim: Encapsulate CFG Cleanup Initialization ADCE: Dead If Elim: Remove redundant code from ADCE initialization ADCE: Dead If: Use CFGCleanup to eliminate newly dead blocks Moved bulk of CFG Cleanup code into MemPass.
2017-10-17 22:33:43 +00:00
%bool = OpTypeBool
ADCE: Empty Loop Elimination This entirely eliminates loops which do not contain live code.
2017-10-25 00:58:48 +00:00
%float_0_5 = OpConstant %float 0.5
%int_1 = OpConstant %int 1
ADCE: Dead if elimination Mark structured conditional branches live only if one or more instructions in their associated construct is marked live. After closure, replace dead structured conditional branches with a branch to its merge and remove dead blocks. ADCE: Dead If Elim: Remove duplicate StructuredOrder code Also generalize ComputeStructuredOrder so that the caller can specify the root block for the order. Phi insertion uses pseudo_entry_block and adce and dead branch elim use the first block of the function. ADCE: Dead If Elim: Pull redundant code out of InsertPhiInstructions ADCE: Dead If Elim: Encapsulate CFG Cleanup Initialization ADCE: Dead If Elim: Remove redundant code from ADCE initialization ADCE: Dead If: Use CFGCleanup to eliminate newly dead blocks Moved bulk of CFG Cleanup code into MemPass.
2017-10-17 22:33:43 +00:00
%_ptr_Output_v4float = OpTypePointer Output %v4float
ADCE: Empty Loop Elimination This entirely eliminates loops which do not contain live code.
2017-10-25 00:58:48 +00:00
%gl_FragColor = OpVariable %_ptr_Output_v4float Output
Adding additional functionality to ADCE. Modified ADCE to remove dead globals. * Entry point and execution mode instructions are marked as alive * Reachable functions and their parameters are marked as alive * Instruction deletion now deferred until the end of the pass * Eliminated dead insts set, added IsDead to calculate that value instead * Ported applicable dead variable elimination tests * Ported dead constant elim tests Added dead function elimination to ADCE * ported dead function elim tests Added handling of decoration groups in ADCE * Uses a custom sorter to traverse decorations in a specific order * Simplifies necessary checks Updated -O and -Os pass lists.
2017-12-18 17:13:10 +00:00
)";
const std::string predefs2_after =
R"(%void = OpTypeVoid
%11 = OpTypeFunction %void
%float = OpTypeFloat 32
%v4float = OpTypeVector %float 4
%_ptr_Function_v4float = OpTypePointer Function %v4float
%_ptr_Input_v4float = OpTypePointer Input %v4float
%BaseColor = OpVariable %_ptr_Input_v4float Input
%_ptr_Output_v4float = OpTypePointer Output %v4float
%gl_FragColor = OpVariable %_ptr_Output_v4float Output
ADCE: Dead if elimination Mark structured conditional branches live only if one or more instructions in their associated construct is marked live. After closure, replace dead structured conditional branches with a branch to its merge and remove dead blocks. ADCE: Dead If Elim: Remove duplicate StructuredOrder code Also generalize ComputeStructuredOrder so that the caller can specify the root block for the order. Phi insertion uses pseudo_entry_block and adce and dead branch elim use the first block of the function. ADCE: Dead If Elim: Pull redundant code out of InsertPhiInstructions ADCE: Dead If Elim: Encapsulate CFG Cleanup Initialization ADCE: Dead If Elim: Remove redundant code from ADCE initialization ADCE: Dead If: Use CFGCleanup to eliminate newly dead blocks Moved bulk of CFG Cleanup code into MemPass.
2017-10-17 22:33:43 +00:00
)";
ADCE: Empty Loop Elimination This entirely eliminates loops which do not contain live code.
2017-10-25 00:58:48 +00:00
const std::string func_before =
R"(%main = OpFunction %void None %11
%27 = OpLabel
%v = OpVariable %_ptr_Function_v4float Function
%df = OpVariable %_ptr_Function_float Function
%i = OpVariable %_ptr_Function_int Function
%28 = OpLoad %v4float %BaseColor
OpStore %v %28
OpStore %df %float_0
OpStore %i %int_0
OpBranch %29
%29 = OpLabel
OpLoopMerge %30 %31 None
OpBranch %32
%32 = OpLabel
%33 = OpLoad %int %i
%34 = OpAccessChain %_ptr_Uniform_int %_ %int_0
%35 = OpLoad %int %34
%36 = OpSLessThan %bool %33 %35
OpBranchConditional %36 %37 %30
%37 = OpLabel
%38 = OpLoad %float %df
%39 = OpFMul %float %38 %float_0_5
OpStore %df %39
%40 = OpLoad %int %i
%41 = OpIAdd %int %40 %int_1
OpStore %i %41
OpBranch %31
%31 = OpLabel
OpBranch %29
%30 = OpLabel
%42 = OpLoad %v4float %v
OpStore %gl_FragColor %42
OpReturn
OpFunctionEnd
)";
const std::string func_after =
R"(%main = OpFunction %void None %11
%27 = OpLabel
%v = OpVariable %_ptr_Function_v4float Function
%28 = OpLoad %v4float %BaseColor
OpStore %v %28
OpBranch %29
%29 = OpLabel
OpBranch %30
%30 = OpLabel
%42 = OpLoad %v4float %v
OpStore %gl_FragColor %42
OpReturn
OpFunctionEnd
)";
Move tests into namespaces (#1689) This CL moves the test into namespaces based on their directories.
2018-07-11 13:24:49 +00:00
SinglePassRunAndCheck<AggressiveDCEPass>(
Adding additional functionality to ADCE. Modified ADCE to remove dead globals. * Entry point and execution mode instructions are marked as alive * Reachable functions and their parameters are marked as alive * Instruction deletion now deferred until the end of the pass * Eliminated dead insts set, added IsDead to calculate that value instead * Ported applicable dead variable elimination tests * Ported dead constant elim tests Added dead function elimination to ADCE * ported dead function elim tests Added handling of decoration groups in ADCE * Uses a custom sorter to traverse decorations in a specific order * Simplifies necessary checks Updated -O and -Os pass lists.
2017-12-18 17:13:10 +00:00
predefs1 + names_before + predefs2_before + func_before,
predefs1 + names_after + predefs2_after + func_after, true, true);
ADCE: Empty Loop Elimination This entirely eliminates loops which do not contain live code.
2017-10-25 00:58:48 +00:00
}
TEST_F(AggressiveDCETest, NoEliminateBusyLoop) {
// Note: SPIR-V edited to replace AtomicAdd(i,0) with AtomicLoad(i)
//
// #version 450
//
// layout(std430) buffer I_t
// {
// int g_I;
// int g_I2;
// };
//
// layout(location = 0) out int o;
//
// void main(void)
// {
// while (atomicAdd(g_I, 0) == 0) {}
// o = g_I2;
// }
const std::string assembly =
R"(OpCapability Shader
%1 = OpExtInstImport "GLSL.std.450"
OpMemoryModel Logical GLSL450
OpEntryPoint Fragment %main "main" %o
OpExecutionMode %main OriginUpperLeft
OpSource GLSL 450
OpName %main "main"
OpName %I_t "I_t"
OpMemberName %I_t 0 "g_I"
OpMemberName %I_t 1 "g_I2"
OpName %_ ""
OpName %o "o"
OpMemberDecorate %I_t 0 Offset 0
OpMemberDecorate %I_t 1 Offset 4
OpDecorate %I_t BufferBlock
OpDecorate %_ DescriptorSet 0
OpDecorate %o Location 0
%void = OpTypeVoid
%7 = OpTypeFunction %void
%int = OpTypeInt 32 1
%I_t = OpTypeStruct %int %int
%_ptr_Uniform_I_t = OpTypePointer Uniform %I_t
%_ = OpVariable %_ptr_Uniform_I_t Uniform
%int_0 = OpConstant %int 0
%int_1 = OpConstant %int 1
%_ptr_Uniform_int = OpTypePointer Uniform %int
%uint = OpTypeInt 32 0
%uint_1 = OpConstant %uint 1
%uint_0 = OpConstant %uint 0
%bool = OpTypeBool
%_ptr_Output_int = OpTypePointer Output %int
%o = OpVariable %_ptr_Output_int Output
%main = OpFunction %void None %7
%18 = OpLabel
OpBranch %19
%19 = OpLabel
OpLoopMerge %20 %21 None
OpBranch %22
%22 = OpLabel
%23 = OpAccessChain %_ptr_Uniform_int %_ %int_0
%24 = OpAtomicLoad %int %23 %uint_1 %uint_0
%25 = OpIEqual %bool %24 %int_0
OpBranchConditional %25 %26 %20
%26 = OpLabel
OpBranch %21
%21 = OpLabel
OpBranch %19
%20 = OpLabel
%27 = OpAccessChain %_ptr_Uniform_int %_ %int_1
%28 = OpLoad %int %27
OpStore %o %28
OpReturn
OpFunctionEnd
)";
Move tests into namespaces (#1689) This CL moves the test into namespaces based on their directories.
2018-07-11 13:24:49 +00:00
SinglePassRunAndCheck<AggressiveDCEPass>(assembly, assembly, true, true);
ADCE: Empty Loop Elimination This entirely eliminates loops which do not contain live code.
2017-10-25 00:58:48 +00:00
}
TEST_F(AggressiveDCETest, NoEliminateLiveLoop) {
// Note: SPIR-V optimized
//
// #version 430
//
// layout(std430) buffer U_t
// {
// float g_F[10];
// };
//
// layout(location = 0)out float o;
//
// void main(void)
// {
// float s = 0.0;
// for (int i=0; i<10; i++)
// s += g_F[i];
// o = s;
// }
const std::string assembly =
R"(OpCapability Shader
%1 = OpExtInstImport "GLSL.std.450"
OpMemoryModel Logical GLSL450
OpEntryPoint Fragment %main "main" %o
OpExecutionMode %main OriginUpperLeft
OpSource GLSL 430
OpName %main "main"
OpName %U_t "U_t"
OpMemberName %U_t 0 "g_F"
OpName %_ ""
OpName %o "o"
OpDecorate %_arr_float_uint_10 ArrayStride 4
OpMemberDecorate %U_t 0 Offset 0
OpDecorate %U_t BufferBlock
OpDecorate %_ DescriptorSet 0
OpDecorate %o Location 0
%void = OpTypeVoid
%8 = OpTypeFunction %void
%float = OpTypeFloat 32
%float_0 = OpConstant %float 0
%int = OpTypeInt 32 1
%int_0 = OpConstant %int 0
%int_10 = OpConstant %int 10
%bool = OpTypeBool
%uint = OpTypeInt 32 0
%uint_10 = OpConstant %uint 10
%_arr_float_uint_10 = OpTypeArray %float %uint_10
%U_t = OpTypeStruct %_arr_float_uint_10
%_ptr_Uniform_U_t = OpTypePointer Uniform %U_t
%_ = OpVariable %_ptr_Uniform_U_t Uniform
%_ptr_Uniform_float = OpTypePointer Uniform %float
%int_1 = OpConstant %int 1
%_ptr_Output_float = OpTypePointer Output %float
%o = OpVariable %_ptr_Output_float Output
%main = OpFunction %void None %8
Adding additional functionality to ADCE. Modified ADCE to remove dead globals. * Entry point and execution mode instructions are marked as alive * Reachable functions and their parameters are marked as alive * Instruction deletion now deferred until the end of the pass * Eliminated dead insts set, added IsDead to calculate that value instead * Ported applicable dead variable elimination tests * Ported dead constant elim tests Added dead function elimination to ADCE * ported dead function elim tests Added handling of decoration groups in ADCE * Uses a custom sorter to traverse decorations in a specific order * Simplifies necessary checks Updated -O and -Os pass lists.
2017-12-18 17:13:10 +00:00
%21 = OpLabel
OpBranch %22
%22 = OpLabel
%23 = OpPhi %float %float_0 %21 %24 %25
%26 = OpPhi %int %int_0 %21 %27 %25
OpLoopMerge %28 %25 None
OpBranch %29
%29 = OpLabel
%30 = OpSLessThan %bool %26 %int_10
OpBranchConditional %30 %31 %28
ADCE: Empty Loop Elimination This entirely eliminates loops which do not contain live code.
2017-10-25 00:58:48 +00:00
%31 = OpLabel
Adding additional functionality to ADCE. Modified ADCE to remove dead globals. * Entry point and execution mode instructions are marked as alive * Reachable functions and their parameters are marked as alive * Instruction deletion now deferred until the end of the pass * Eliminated dead insts set, added IsDead to calculate that value instead * Ported applicable dead variable elimination tests * Ported dead constant elim tests Added dead function elimination to ADCE * ported dead function elim tests Added handling of decoration groups in ADCE * Uses a custom sorter to traverse decorations in a specific order * Simplifies necessary checks Updated -O and -Os pass lists.
2017-12-18 17:13:10 +00:00
%32 = OpAccessChain %_ptr_Uniform_float %_ %int_0 %26
%33 = OpLoad %float %32
%24 = OpFAdd %float %23 %33
OpBranch %25
%25 = OpLabel
%27 = OpIAdd %int %26 %int_1
OpBranch %22
%28 = OpLabel
OpStore %o %23
ADCE: Empty Loop Elimination This entirely eliminates loops which do not contain live code.
2017-10-25 00:58:48 +00:00
OpReturn
OpFunctionEnd
)";
Move tests into namespaces (#1689) This CL moves the test into namespaces based on their directories.
2018-07-11 13:24:49 +00:00
SinglePassRunAndCheck<AggressiveDCEPass>(assembly, assembly, true, true);
ADCE: Empty Loop Elimination This entirely eliminates loops which do not contain live code.
2017-10-25 00:58:48 +00:00
}
TEST_F(AggressiveDCETest, EliminateEntireFunctionBody) {
// #version 450
//
// layout(location = 0) in vec4 BaseColor;
// layout(location = 0) out vec4 OutColor;
//
// void main()
// {
// float d;
// if (BaseColor.x == 0)
// d = BaseColor.y;
// else
// d = BaseColor.z;
// }
Allow ADCE to remove dead inputs (#4629) * https://github.com/KhronosGroup/Vulkan-Docs/issues/666 clearly specified that interfaces do not require an input if there is an associated output * ADCE can now remove unused input variables (though they are kept if the preserve interfaces option is used)
2021-11-16 20:54:17 +00:00
const std::string spirv =
ADCE: Empty Loop Elimination This entirely eliminates loops which do not contain live code.
2017-10-25 00:58:48 +00:00
R"(OpCapability Shader
%1 = OpExtInstImport "GLSL.std.450"
OpMemoryModel Logical GLSL450
OpEntryPoint Fragment %main "main" %BaseColor %OutColor
OpExecutionMode %main OriginUpperLeft
OpSource GLSL 450
OpName %main "main"
OpName %BaseColor "BaseColor"
OpName %d "d"
OpName %OutColor "OutColor"
OpDecorate %BaseColor Location 0
OpDecorate %OutColor Location 0
%void = OpTypeVoid
%7 = OpTypeFunction %void
%float = OpTypeFloat 32
%v4float = OpTypeVector %float 4
%_ptr_Input_v4float = OpTypePointer Input %v4float
%BaseColor = OpVariable %_ptr_Input_v4float Input
%uint = OpTypeInt 32 0
%uint_0 = OpConstant %uint 0
%_ptr_Input_float = OpTypePointer Input %float
%float_0 = OpConstant %float 0
%bool = OpTypeBool
%_ptr_Function_float = OpTypePointer Function %float
%uint_1 = OpConstant %uint 1
%uint_2 = OpConstant %uint 2
%_ptr_Output_v4float = OpTypePointer Output %v4float
%OutColor = OpVariable %_ptr_Output_v4float Output
Allow ADCE to remove dead inputs (#4629) * https://github.com/KhronosGroup/Vulkan-Docs/issues/666 clearly specified that interfaces do not require an input if there is an associated output * ADCE can now remove unused input variables (though they are kept if the preserve interfaces option is used)
2021-11-16 20:54:17 +00:00
; CHECK: = OpFunction
; CHECK-NEXT: = OpLabel
; CHECK-NEXT: OpBranch [[target:%\w+]]
; CHECK-NEXT: [[target]] = OpLabel
; CHECK-NEXT: OpReturn
; CHECK-NEXT: OpFunctionEnd
ADCE: Dead if elimination Mark structured conditional branches live only if one or more instructions in their associated construct is marked live. After closure, replace dead structured conditional branches with a branch to its merge and remove dead blocks. ADCE: Dead If Elim: Remove duplicate StructuredOrder code Also generalize ComputeStructuredOrder so that the caller can specify the root block for the order. Phi insertion uses pseudo_entry_block and adce and dead branch elim use the first block of the function. ADCE: Dead If Elim: Pull redundant code out of InsertPhiInstructions ADCE: Dead If Elim: Encapsulate CFG Cleanup Initialization ADCE: Dead If Elim: Remove redundant code from ADCE initialization ADCE: Dead If: Use CFGCleanup to eliminate newly dead blocks Moved bulk of CFG Cleanup code into MemPass.
2017-10-17 22:33:43 +00:00
Allow ADCE to remove dead inputs (#4629) * https://github.com/KhronosGroup/Vulkan-Docs/issues/666 clearly specified that interfaces do not require an input if there is an associated output * ADCE can now remove unused input variables (though they are kept if the preserve interfaces option is used)
2021-11-16 20:54:17 +00:00
%main = OpFunction %void None %7
ADCE: Dead if elimination Mark structured conditional branches live only if one or more instructions in their associated construct is marked live. After closure, replace dead structured conditional branches with a branch to its merge and remove dead blocks. ADCE: Dead If Elim: Remove duplicate StructuredOrder code Also generalize ComputeStructuredOrder so that the caller can specify the root block for the order. Phi insertion uses pseudo_entry_block and adce and dead branch elim use the first block of the function. ADCE: Dead If Elim: Pull redundant code out of InsertPhiInstructions ADCE: Dead If Elim: Encapsulate CFG Cleanup Initialization ADCE: Dead If Elim: Remove redundant code from ADCE initialization ADCE: Dead If: Use CFGCleanup to eliminate newly dead blocks Moved bulk of CFG Cleanup code into MemPass.
2017-10-17 22:33:43 +00:00
%20 = OpLabel
%d = OpVariable %_ptr_Function_float Function
%21 = OpAccessChain %_ptr_Input_float %BaseColor %uint_0
%22 = OpLoad %float %21
ADCE: Empty Loop Elimination This entirely eliminates loops which do not contain live code.
2017-10-25 00:58:48 +00:00
%23 = OpFOrdEqual %bool %22 %float_0
ADCE: Dead if elimination Mark structured conditional branches live only if one or more instructions in their associated construct is marked live. After closure, replace dead structured conditional branches with a branch to its merge and remove dead blocks. ADCE: Dead If Elim: Remove duplicate StructuredOrder code Also generalize ComputeStructuredOrder so that the caller can specify the root block for the order. Phi insertion uses pseudo_entry_block and adce and dead branch elim use the first block of the function. ADCE: Dead If Elim: Pull redundant code out of InsertPhiInstructions ADCE: Dead If Elim: Encapsulate CFG Cleanup Initialization ADCE: Dead If Elim: Remove redundant code from ADCE initialization ADCE: Dead If: Use CFGCleanup to eliminate newly dead blocks Moved bulk of CFG Cleanup code into MemPass.
2017-10-17 22:33:43 +00:00
OpSelectionMerge %24 None
OpBranchConditional %23 %25 %26
%25 = OpLabel
%27 = OpAccessChain %_ptr_Input_float %BaseColor %uint_1
%28 = OpLoad %float %27
OpStore %d %28
OpBranch %24
%26 = OpLabel
%29 = OpAccessChain %_ptr_Input_float %BaseColor %uint_2
%30 = OpLoad %float %29
OpStore %d %30
OpBranch %24
%24 = OpLabel
OpReturn
OpFunctionEnd
)";
Allow ADCE to remove dead inputs (#4629) * https://github.com/KhronosGroup/Vulkan-Docs/issues/666 clearly specified that interfaces do not require an input if there is an associated output * ADCE can now remove unused input variables (though they are kept if the preserve interfaces option is used)
2021-11-16 20:54:17 +00:00
SinglePassRunAndMatch<AggressiveDCEPass>(spirv, true);
ADCE: Dead if elimination Mark structured conditional branches live only if one or more instructions in their associated construct is marked live. After closure, replace dead structured conditional branches with a branch to its merge and remove dead blocks. ADCE: Dead If Elim: Remove duplicate StructuredOrder code Also generalize ComputeStructuredOrder so that the caller can specify the root block for the order. Phi insertion uses pseudo_entry_block and adce and dead branch elim use the first block of the function. ADCE: Dead If Elim: Pull redundant code out of InsertPhiInstructions ADCE: Dead If Elim: Encapsulate CFG Cleanup Initialization ADCE: Dead If Elim: Remove redundant code from ADCE initialization ADCE: Dead If: Use CFGCleanup to eliminate newly dead blocks Moved bulk of CFG Cleanup code into MemPass.
2017-10-17 22:33:43 +00:00
}
ADCE: Empty Loop Elimination This entirely eliminates loops which do not contain live code.
2017-10-25 00:58:48 +00:00
TEST_F(AggressiveDCETest, EliminateUselessInnerLoop) {
// #version 430
//
// layout(std430) buffer U_t
// {
// float g_F[10];
// };
//
// layout(location = 0)out float o;
//
// void main(void)
// {
// float s = 0.0;
// for (int i=0; i<10; i++) {
// for (int j=0; j<10; j++) {
// }
// s += g_F[i];
// }
// o = s;
// }
const std::string predefs_before =
R"(OpCapability Shader
%1 = OpExtInstImport "GLSL.std.450"
OpMemoryModel Logical GLSL450
OpEntryPoint Fragment %main "main" %o
OpExecutionMode %main OriginUpperLeft
OpSource GLSL 430
OpName %main "main"
OpName %s "s"
OpName %i "i"
OpName %j "j"
OpName %U_t "U_t"
OpMemberName %U_t 0 "g_F"
OpName %_ ""
OpName %o "o"
OpDecorate %_arr_float_uint_10 ArrayStride 4
OpMemberDecorate %U_t 0 Offset 0
OpDecorate %U_t BufferBlock
OpDecorate %_ DescriptorSet 0
OpDecorate %o Location 0
%void = OpTypeVoid
%11 = OpTypeFunction %void
%float = OpTypeFloat 32
%_ptr_Function_float = OpTypePointer Function %float
%float_0 = OpConstant %float 0
%int = OpTypeInt 32 1
%_ptr_Function_int = OpTypePointer Function %int
%int_0 = OpConstant %int 0
%int_10 = OpConstant %int 10
%bool = OpTypeBool
%int_1 = OpConstant %int 1
%uint = OpTypeInt 32 0
%uint_10 = OpConstant %uint 10
%_arr_float_uint_10 = OpTypeArray %float %uint_10
%U_t = OpTypeStruct %_arr_float_uint_10
%_ptr_Uniform_U_t = OpTypePointer Uniform %U_t
%_ = OpVariable %_ptr_Uniform_U_t Uniform
%_ptr_Uniform_float = OpTypePointer Uniform %float
%_ptr_Output_float = OpTypePointer Output %float
%o = OpVariable %_ptr_Output_float Output
)";
const std::string predefs_after =
R"(OpCapability Shader
%1 = OpExtInstImport "GLSL.std.450"
OpMemoryModel Logical GLSL450
OpEntryPoint Fragment %main "main" %o
OpExecutionMode %main OriginUpperLeft
OpSource GLSL 430
OpName %main "main"
OpName %s "s"
OpName %i "i"
OpName %U_t "U_t"
OpMemberName %U_t 0 "g_F"
OpName %_ ""
OpName %o "o"
OpDecorate %_arr_float_uint_10 ArrayStride 4
OpMemberDecorate %U_t 0 Offset 0
OpDecorate %U_t BufferBlock
OpDecorate %_ DescriptorSet 0
OpDecorate %o Location 0
%void = OpTypeVoid
%11 = OpTypeFunction %void
%float = OpTypeFloat 32
%_ptr_Function_float = OpTypePointer Function %float
%float_0 = OpConstant %float 0
%int = OpTypeInt 32 1
%_ptr_Function_int = OpTypePointer Function %int
%int_0 = OpConstant %int 0
%int_10 = OpConstant %int 10
%bool = OpTypeBool
%int_1 = OpConstant %int 1
%uint = OpTypeInt 32 0
%uint_10 = OpConstant %uint 10
%_arr_float_uint_10 = OpTypeArray %float %uint_10
%U_t = OpTypeStruct %_arr_float_uint_10
%_ptr_Uniform_U_t = OpTypePointer Uniform %U_t
%_ = OpVariable %_ptr_Uniform_U_t Uniform
%_ptr_Uniform_float = OpTypePointer Uniform %float
%_ptr_Output_float = OpTypePointer Output %float
%o = OpVariable %_ptr_Output_float Output
)";
const std::string func_before =
R"(%main = OpFunction %void None %11
%26 = OpLabel
%s = OpVariable %_ptr_Function_float Function
%i = OpVariable %_ptr_Function_int Function
%j = OpVariable %_ptr_Function_int Function
OpStore %s %float_0
OpStore %i %int_0
OpBranch %27
%27 = OpLabel
OpLoopMerge %28 %29 None
OpBranch %30
%30 = OpLabel
%31 = OpLoad %int %i
%32 = OpSLessThan %bool %31 %int_10
OpBranchConditional %32 %33 %28
%33 = OpLabel
OpStore %j %int_0
OpBranch %34
%34 = OpLabel
OpLoopMerge %35 %36 None
OpBranch %37
%37 = OpLabel
%38 = OpLoad %int %j
%39 = OpSLessThan %bool %38 %int_10
OpBranchConditional %39 %40 %35
%40 = OpLabel
OpBranch %36
%36 = OpLabel
%41 = OpLoad %int %j
%42 = OpIAdd %int %41 %int_1
OpStore %j %42
OpBranch %34
%35 = OpLabel
%43 = OpLoad %int %i
%44 = OpAccessChain %_ptr_Uniform_float %_ %int_0 %43
%45 = OpLoad %float %44
%46 = OpLoad %float %s
%47 = OpFAdd %float %46 %45
OpStore %s %47
OpBranch %29
%29 = OpLabel
%48 = OpLoad %int %i
%49 = OpIAdd %int %48 %int_1
OpStore %i %49
OpBranch %27
%28 = OpLabel
%50 = OpLoad %float %s
OpStore %o %50
OpReturn
OpFunctionEnd
)";
const std::string func_after =
R"(%main = OpFunction %void None %11
%26 = OpLabel
%s = OpVariable %_ptr_Function_float Function
%i = OpVariable %_ptr_Function_int Function
OpStore %s %float_0
OpStore %i %int_0
OpBranch %27
%27 = OpLabel
OpLoopMerge %28 %29 None
OpBranch %30
%30 = OpLabel
%31 = OpLoad %int %i
%32 = OpSLessThan %bool %31 %int_10
OpBranchConditional %32 %33 %28
%33 = OpLabel
OpBranch %34
%34 = OpLabel
OpBranch %35
%35 = OpLabel
%43 = OpLoad %int %i
%44 = OpAccessChain %_ptr_Uniform_float %_ %int_0 %43
%45 = OpLoad %float %44
%46 = OpLoad %float %s
%47 = OpFAdd %float %46 %45
OpStore %s %47
OpBranch %29
%29 = OpLabel
%48 = OpLoad %int %i
%49 = OpIAdd %int %48 %int_1
OpStore %i %49
OpBranch %27
%28 = OpLabel
%50 = OpLoad %float %s
OpStore %o %50
OpReturn
OpFunctionEnd
)";
Move tests into namespaces (#1689) This CL moves the test into namespaces based on their directories.
2018-07-11 13:24:49 +00:00
SinglePassRunAndCheck<AggressiveDCEPass>(
ADCE: Empty Loop Elimination This entirely eliminates loops which do not contain live code.
2017-10-25 00:58:48 +00:00
predefs_before + func_before, predefs_after + func_after, true, true);
}
TEST_F(AggressiveDCETest, EliminateUselessNestedLoopWithIf) {
// #version 430
//
// layout(std430) buffer U_t
// {
// float g_F[10][10];
// };
//
// layout(location = 0)out float o;
//
// void main(void)
// {
// float s = 0.0;
// for (int i=0; i<10; i++) {
// for (int j=0; j<10; j++) {
// float t = g_F[i][j];
// if (t > 0.0)
// s += t;
// }
// }
// o = 0.0;
// }
const std::string predefs_before =
R"(OpCapability Shader
%1 = OpExtInstImport "GLSL.std.450"
OpMemoryModel Logical GLSL450
OpEntryPoint Fragment %main "main" %o
OpExecutionMode %main OriginUpperLeft
OpSource GLSL 430
OpName %main "main"
OpName %s "s"
OpName %i "i"
OpName %j "j"
OpName %U_t "U_t"
OpMemberName %U_t 0 "g_F"
OpName %_ ""
OpName %o "o"
OpDecorate %_arr_float_uint_10 ArrayStride 4
OpDecorate %_arr__arr_float_uint_10_uint_10 ArrayStride 40
OpMemberDecorate %U_t 0 Offset 0
OpDecorate %U_t BufferBlock
OpDecorate %_ DescriptorSet 0
OpDecorate %o Location 0
%void = OpTypeVoid
%12 = OpTypeFunction %void
%float = OpTypeFloat 32
%_ptr_Function_float = OpTypePointer Function %float
%float_0 = OpConstant %float 0
%int = OpTypeInt 32 1
%_ptr_Function_int = OpTypePointer Function %int
%int_0 = OpConstant %int 0
%int_10 = OpConstant %int 10
%bool = OpTypeBool
%uint = OpTypeInt 32 0
%uint_10 = OpConstant %uint 10
%_arr_float_uint_10 = OpTypeArray %float %uint_10
%_arr__arr_float_uint_10_uint_10 = OpTypeArray %_arr_float_uint_10 %uint_10
%U_t = OpTypeStruct %_arr__arr_float_uint_10_uint_10
%_ptr_Uniform_U_t = OpTypePointer Uniform %U_t
%_ = OpVariable %_ptr_Uniform_U_t Uniform
%_ptr_Uniform_float = OpTypePointer Uniform %float
%int_1 = OpConstant %int 1
%_ptr_Output_float = OpTypePointer Output %float
%o = OpVariable %_ptr_Output_float Output
)";
const std::string predefs_after =
R"(OpCapability Shader
%1 = OpExtInstImport "GLSL.std.450"
OpMemoryModel Logical GLSL450
OpEntryPoint Fragment %main "main" %o
OpExecutionMode %main OriginUpperLeft
OpSource GLSL 430
OpName %main "main"
OpName %o "o"
OpDecorate %o Location 0
%void = OpTypeVoid
%12 = OpTypeFunction %void
%float = OpTypeFloat 32
%float_0 = OpConstant %float 0
%_ptr_Output_float = OpTypePointer Output %float
%o = OpVariable %_ptr_Output_float Output
)";
const std::string func_before =
R"(%main = OpFunction %void None %12
%27 = OpLabel
%s = OpVariable %_ptr_Function_float Function
%i = OpVariable %_ptr_Function_int Function
%j = OpVariable %_ptr_Function_int Function
OpStore %s %float_0
OpStore %i %int_0
OpBranch %28
%28 = OpLabel
OpLoopMerge %29 %30 None
OpBranch %31
%31 = OpLabel
%32 = OpLoad %int %i
%33 = OpSLessThan %bool %32 %int_10
OpBranchConditional %33 %34 %29
%34 = OpLabel
OpStore %j %int_0
OpBranch %35
%35 = OpLabel
OpLoopMerge %36 %37 None
OpBranch %38
%38 = OpLabel
%39 = OpLoad %int %j
%40 = OpSLessThan %bool %39 %int_10
OpBranchConditional %40 %41 %36
%41 = OpLabel
%42 = OpLoad %int %i
%43 = OpLoad %int %j
%44 = OpAccessChain %_ptr_Uniform_float %_ %int_0 %42 %43
%45 = OpLoad %float %44
%46 = OpFOrdGreaterThan %bool %45 %float_0
OpSelectionMerge %47 None
OpBranchConditional %46 %48 %47
%48 = OpLabel
%49 = OpLoad %float %s
%50 = OpFAdd %float %49 %45
OpStore %s %50
OpBranch %47
%47 = OpLabel
OpBranch %37
%37 = OpLabel
%51 = OpLoad %int %j
%52 = OpIAdd %int %51 %int_1
OpStore %j %52
OpBranch %35
%36 = OpLabel
OpBranch %30
%30 = OpLabel
%53 = OpLoad %int %i
%54 = OpIAdd %int %53 %int_1
OpStore %i %54
OpBranch %28
%29 = OpLabel
OpStore %o %float_0
OpReturn
OpFunctionEnd
)";
const std::string func_after =
R"(%main = OpFunction %void None %12
%27 = OpLabel
OpBranch %28
%28 = OpLabel
OpBranch %29
%29 = OpLabel
OpStore %o %float_0
OpReturn
OpFunctionEnd
)";
Move tests into namespaces (#1689) This CL moves the test into namespaces based on their directories.
2018-07-11 13:24:49 +00:00
SinglePassRunAndCheck<AggressiveDCEPass>(
ADCE: Empty Loop Elimination This entirely eliminates loops which do not contain live code.
2017-10-25 00:58:48 +00:00
predefs_before + func_before, predefs_after + func_after, true, true);
}
ADCE: Only mark true breaks and continues of live loops This fixes issue #1075 - Mark continue when conditional branch with merge block. Only mark if merge block is not continue block. - Handle conditional branch break with preceding merge
2017-12-12 21:27:46 +00:00
TEST_F(AggressiveDCETest, EliminateEmptyIfBeforeContinue) {
// #version 430
//
// layout(location = 0)out float o;
//
// void main(void)
// {
// float s = 0.0;
// for (int i=0; i<10; i++) {
// s += 1.0;
// if (i > s) {}
// }
// o = s;
// }
const std::string predefs_before =
R"(OpCapability Shader
%1 = OpExtInstImport "GLSL.std.450"
OpMemoryModel Logical GLSL450
OpEntryPoint Fragment %main "main" %3
OpExecutionMode %main OriginUpperLeft
OpSource GLSL 430
OpSourceExtension "GL_GOOGLE_cpp_style_line_directive"
OpSourceExtension "GL_GOOGLE_include_directive"
OpName %main "main"
OpDecorate %3 Location 0
%void = OpTypeVoid
%5 = OpTypeFunction %void
%float = OpTypeFloat 32
%float_0 = OpConstant %float 0
%int = OpTypeInt 32 1
%_ptr_Function_int = OpTypePointer Function %int
%int_0 = OpConstant %int 0
%int_10 = OpConstant %int 10
%bool = OpTypeBool
%float_1 = OpConstant %float 1
%int_1 = OpConstant %int 1
%_ptr_Output_float = OpTypePointer Output %float
%3 = OpVariable %_ptr_Output_float Output
)";
const std::string predefs_after =
R"(OpCapability Shader
%1 = OpExtInstImport "GLSL.std.450"
OpMemoryModel Logical GLSL450
OpEntryPoint Fragment %main "main" %3
OpExecutionMode %main OriginUpperLeft
OpSource GLSL 430
OpSourceExtension "GL_GOOGLE_cpp_style_line_directive"
OpSourceExtension "GL_GOOGLE_include_directive"
OpName %main "main"
OpDecorate %3 Location 0
%void = OpTypeVoid
%5 = OpTypeFunction %void
%float = OpTypeFloat 32
%float_0 = OpConstant %float 0
%int = OpTypeInt 32 1
%int_0 = OpConstant %int 0
%int_10 = OpConstant %int 10
%bool = OpTypeBool
%float_1 = OpConstant %float 1
%int_1 = OpConstant %int 1
%_ptr_Output_float = OpTypePointer Output %float
%3 = OpVariable %_ptr_Output_float Output
)";
const std::string func_before =
R"(%main = OpFunction %void None %5
%16 = OpLabel
OpBranch %17
%17 = OpLabel
%18 = OpPhi %float %float_0 %16 %19 %20
%21 = OpPhi %int %int_0 %16 %22 %20
OpLoopMerge %23 %20 None
OpBranch %24
%24 = OpLabel
%25 = OpSLessThan %bool %21 %int_10
OpBranchConditional %25 %26 %23
%26 = OpLabel
%19 = OpFAdd %float %18 %float_1
%27 = OpConvertFToS %int %19
%28 = OpSGreaterThan %bool %21 %27
OpSelectionMerge %20 None
OpBranchConditional %28 %29 %20
%29 = OpLabel
OpBranch %20
%20 = OpLabel
%22 = OpIAdd %int %21 %int_1
OpBranch %17
%23 = OpLabel
OpStore %3 %18
OpReturn
OpFunctionEnd
)";
const std::string func_after =
R"(%main = OpFunction %void None %5
%16 = OpLabel
OpBranch %17
%17 = OpLabel
%18 = OpPhi %float %float_0 %16 %19 %20
%21 = OpPhi %int %int_0 %16 %22 %20
OpLoopMerge %23 %20 None
OpBranch %24
%24 = OpLabel
%25 = OpSLessThan %bool %21 %int_10
OpBranchConditional %25 %26 %23
%26 = OpLabel
%19 = OpFAdd %float %18 %float_1
OpBranch %20
%20 = OpLabel
%22 = OpIAdd %int %21 %int_1
OpBranch %17
%23 = OpLabel
OpStore %3 %18
OpReturn
OpFunctionEnd
)";
Move tests into namespaces (#1689) This CL moves the test into namespaces based on their directories.
2018-07-11 13:24:49 +00:00
SinglePassRunAndCheck<AggressiveDCEPass>(
ADCE: Only mark true breaks and continues of live loops This fixes issue #1075 - Mark continue when conditional branch with merge block. Only mark if merge block is not continue block. - Handle conditional branch break with preceding merge
2017-12-12 21:27:46 +00:00
predefs_before + func_before, predefs_after + func_after, true, true);
}
ADCE: Empty Loop Elimination This entirely eliminates loops which do not contain live code.
2017-10-25 00:58:48 +00:00
TEST_F(AggressiveDCETest, NoEliminateLiveNestedLoopWithIf) {
// Note: SPIR-V optimized
//
// #version 430
//
// layout(std430) buffer U_t
// {
// float g_F[10][10];
// };
//
// layout(location = 0)out float o;
//
// void main(void)
// {
// float s = 0.0;
// for (int i=0; i<10; i++) {
// for (int j=0; j<10; j++) {
// float t = g_F[i][j];
// if (t > 0.0)
// s += t;
// }
// }
// o = s;
// }
const std::string assembly =
R"(OpCapability Shader
%1 = OpExtInstImport "GLSL.std.450"
OpMemoryModel Logical GLSL450
OpEntryPoint Fragment %main "main" %o
OpExecutionMode %main OriginUpperLeft
OpSource GLSL 430
OpName %main "main"
OpName %s "s"
OpName %i "i"
OpName %j "j"
OpName %U_t "U_t"
OpMemberName %U_t 0 "g_F"
OpName %_ ""
OpName %o "o"
OpDecorate %_arr_float_uint_10 ArrayStride 4
OpDecorate %_arr__arr_float_uint_10_uint_10 ArrayStride 40
OpMemberDecorate %U_t 0 Offset 0
OpDecorate %U_t BufferBlock
OpDecorate %_ DescriptorSet 0
OpDecorate %o Location 0
%void = OpTypeVoid
%12 = OpTypeFunction %void
%float = OpTypeFloat 32
%_ptr_Function_float = OpTypePointer Function %float
%float_0 = OpConstant %float 0
%int = OpTypeInt 32 1
%_ptr_Function_int = OpTypePointer Function %int
%int_0 = OpConstant %int 0
%int_10 = OpConstant %int 10
%bool = OpTypeBool
%uint = OpTypeInt 32 0
%uint_10 = OpConstant %uint 10
%_arr_float_uint_10 = OpTypeArray %float %uint_10
%_arr__arr_float_uint_10_uint_10 = OpTypeArray %_arr_float_uint_10 %uint_10
%U_t = OpTypeStruct %_arr__arr_float_uint_10_uint_10
%_ptr_Uniform_U_t = OpTypePointer Uniform %U_t
%_ = OpVariable %_ptr_Uniform_U_t Uniform
%_ptr_Uniform_float = OpTypePointer Uniform %float
%int_1 = OpConstant %int 1
%_ptr_Output_float = OpTypePointer Output %float
%o = OpVariable %_ptr_Output_float Output
%main = OpFunction %void None %12
%27 = OpLabel
%s = OpVariable %_ptr_Function_float Function
%i = OpVariable %_ptr_Function_int Function
%j = OpVariable %_ptr_Function_int Function
OpStore %s %float_0
OpStore %i %int_0
OpBranch %28
%28 = OpLabel
OpLoopMerge %29 %30 None
OpBranch %31
%31 = OpLabel
%32 = OpLoad %int %i
%33 = OpSLessThan %bool %32 %int_10
OpBranchConditional %33 %34 %29
%34 = OpLabel
OpStore %j %int_0
OpBranch %35
%35 = OpLabel
OpLoopMerge %36 %37 None
OpBranch %38
%38 = OpLabel
%39 = OpLoad %int %j
%40 = OpSLessThan %bool %39 %int_10
OpBranchConditional %40 %41 %36
%41 = OpLabel
%42 = OpLoad %int %i
%43 = OpLoad %int %j
%44 = OpAccessChain %_ptr_Uniform_float %_ %int_0 %42 %43
%45 = OpLoad %float %44
%46 = OpFOrdGreaterThan %bool %45 %float_0
OpSelectionMerge %47 None
OpBranchConditional %46 %48 %47
%48 = OpLabel
%49 = OpLoad %float %s
%50 = OpFAdd %float %49 %45
OpStore %s %50
OpBranch %47
%47 = OpLabel
OpBranch %37
%37 = OpLabel
%51 = OpLoad %int %j
%52 = OpIAdd %int %51 %int_1
OpStore %j %52
OpBranch %35
%36 = OpLabel
OpBranch %30
%30 = OpLabel
%53 = OpLoad %int %i
%54 = OpIAdd %int %53 %int_1
OpStore %i %54
OpBranch %28
%29 = OpLabel
%55 = OpLoad %float %s
OpStore %o %55
OpReturn
OpFunctionEnd
)";
Move tests into namespaces (#1689) This CL moves the test into namespaces based on their directories.
2018-07-11 13:24:49 +00:00
SinglePassRunAndCheck<AggressiveDCEPass>(assembly, assembly, true, true);
ADCE: Empty Loop Elimination This entirely eliminates loops which do not contain live code.
2017-10-25 00:58:48 +00:00
}
ADCE: Only mark true breaks and continues of live loops This fixes issue #1075 - Mark continue when conditional branch with merge block. Only mark if merge block is not continue block. - Handle conditional branch break with preceding merge
2017-12-12 21:27:46 +00:00
TEST_F(AggressiveDCETest, NoEliminateIfContinue) {
// Do not eliminate continue embedded in if construct
//
// #version 430
//
// layout(std430) buffer U_t
// {
// float g_F[10];
// };
//
// layout(location = 0)out float o;
//
// void main(void)
// {
// float s = 0.0;
// for (int i=0; i<10; i++) {
// if (i % 2 == 0) continue;
// s += g_F[i];
// }
// o = s;
// }
const std::string assembly =
R"(OpCapability Shader
%1 = OpExtInstImport "GLSL.std.450"
OpMemoryModel Logical GLSL450
OpEntryPoint Fragment %main "main" %o
OpExecutionMode %main OriginUpperLeft
OpSource GLSL 430
OpName %main "main"
OpName %s "s"
OpName %i "i"
OpName %U_t "U_t"
OpMemberName %U_t 0 "g_F"
OpName %_ ""
OpName %o "o"
OpDecorate %_arr_float_uint_10 ArrayStride 4
OpMemberDecorate %U_t 0 Offset 0
OpDecorate %U_t BufferBlock
OpDecorate %_ DescriptorSet 0
OpDecorate %o Location 0
%void = OpTypeVoid
%10 = OpTypeFunction %void
%float = OpTypeFloat 32
%_ptr_Function_float = OpTypePointer Function %float
%float_0 = OpConstant %float 0
%int = OpTypeInt 32 1
%_ptr_Function_int = OpTypePointer Function %int
%int_0 = OpConstant %int 0
%int_10 = OpConstant %int 10
%bool = OpTypeBool
%int_2 = OpConstant %int 2
%uint = OpTypeInt 32 0
%uint_10 = OpConstant %uint 10
%_arr_float_uint_10 = OpTypeArray %float %uint_10
%U_t = OpTypeStruct %_arr_float_uint_10
%_ptr_Uniform_U_t = OpTypePointer Uniform %U_t
%_ = OpVariable %_ptr_Uniform_U_t Uniform
%_ptr_Uniform_float = OpTypePointer Uniform %float
%int_1 = OpConstant %int 1
%_ptr_Output_float = OpTypePointer Output %float
%o = OpVariable %_ptr_Output_float Output
%main = OpFunction %void None %10
%26 = OpLabel
%s = OpVariable %_ptr_Function_float Function
%i = OpVariable %_ptr_Function_int Function
OpStore %s %float_0
OpStore %i %int_0
OpBranch %27
%27 = OpLabel
OpLoopMerge %28 %29 None
OpBranch %30
%30 = OpLabel
%31 = OpLoad %int %i
%32 = OpSLessThan %bool %31 %int_10
OpBranchConditional %32 %33 %28
%33 = OpLabel
%34 = OpLoad %int %i
%35 = OpSMod %int %34 %int_2
%36 = OpIEqual %bool %35 %int_0
OpSelectionMerge %37 None
OpBranchConditional %36 %38 %37
%38 = OpLabel
OpBranch %29
%37 = OpLabel
%39 = OpLoad %int %i
%40 = OpAccessChain %_ptr_Uniform_float %_ %int_0 %39
%41 = OpLoad %float %40
%42 = OpLoad %float %s
%43 = OpFAdd %float %42 %41
OpStore %s %43
OpBranch %29
%29 = OpLabel
%44 = OpLoad %int %i
%45 = OpIAdd %int %44 %int_1
OpStore %i %45
OpBranch %27
%28 = OpLabel
%46 = OpLoad %float %s
OpStore %o %46
OpReturn
OpFunctionEnd
)";
Move tests into namespaces (#1689) This CL moves the test into namespaces based on their directories.
2018-07-11 13:24:49 +00:00
SinglePassRunAndCheck<AggressiveDCEPass>(assembly, assembly, true, true);
ADCE: Only mark true breaks and continues of live loops This fixes issue #1075 - Mark continue when conditional branch with merge block. Only mark if merge block is not continue block. - Handle conditional branch break with preceding merge
2017-12-12 21:27:46 +00:00
}
TEST_F(AggressiveDCETest, NoEliminateIfContinue2) {
// Do not eliminate continue not embedded in if construct
//
// #version 430
//
// layout(std430) buffer U_t
// {
// float g_F[10];
// };
//
// layout(location = 0)out float o;
//
// void main(void)
// {
// float s = 0.0;
// for (int i=0; i<10; i++) {
// if (i % 2 == 0) continue;
// s += g_F[i];
// }
// o = s;
// }
const std::string assembly =
R"(OpCapability Shader
%1 = OpExtInstImport "GLSL.std.450"
OpMemoryModel Logical GLSL450
OpEntryPoint Fragment %main "main" %o
OpExecutionMode %main OriginUpperLeft
OpSource GLSL 430
OpName %main "main"
OpName %s "s"
OpName %i "i"
OpName %U_t "U_t"
OpMemberName %U_t 0 "g_F"
OpName %_ ""
OpName %o "o"
OpDecorate %_arr_float_uint_10 ArrayStride 4
OpMemberDecorate %U_t 0 Offset 0
OpDecorate %U_t BufferBlock
OpDecorate %_ DescriptorSet 0
OpDecorate %o Location 0
%void = OpTypeVoid
%10 = OpTypeFunction %void
%float = OpTypeFloat 32
%_ptr_Function_float = OpTypePointer Function %float
%float_0 = OpConstant %float 0
%int = OpTypeInt 32 1
%_ptr_Function_int = OpTypePointer Function %int
%int_0 = OpConstant %int 0
%int_10 = OpConstant %int 10
%bool = OpTypeBool
%int_2 = OpConstant %int 2
%uint = OpTypeInt 32 0
%uint_10 = OpConstant %uint 10
%_arr_float_uint_10 = OpTypeArray %float %uint_10
%U_t = OpTypeStruct %_arr_float_uint_10
%_ptr_Uniform_U_t = OpTypePointer Uniform %U_t
%_ = OpVariable %_ptr_Uniform_U_t Uniform
%_ptr_Uniform_float = OpTypePointer Uniform %float
%int_1 = OpConstant %int 1
%_ptr_Output_float = OpTypePointer Output %float
%o = OpVariable %_ptr_Output_float Output
%main = OpFunction %void None %10
%26 = OpLabel
%s = OpVariable %_ptr_Function_float Function
%i = OpVariable %_ptr_Function_int Function
OpStore %s %float_0
OpStore %i %int_0
OpBranch %27
%27 = OpLabel
OpLoopMerge %28 %29 None
OpBranch %30
%30 = OpLabel
%31 = OpLoad %int %i
%32 = OpSLessThan %bool %31 %int_10
OpBranchConditional %32 %33 %28
%33 = OpLabel
%34 = OpLoad %int %i
%35 = OpSMod %int %34 %int_2
%36 = OpIEqual %bool %35 %int_0
OpBranchConditional %36 %29 %37
%37 = OpLabel
%38 = OpLoad %int %i
%39 = OpAccessChain %_ptr_Uniform_float %_ %int_0 %38
%40 = OpLoad %float %39
%41 = OpLoad %float %s
%42 = OpFAdd %float %41 %40
OpStore %s %42
OpBranch %29
%29 = OpLabel
%43 = OpLoad %int %i
%44 = OpIAdd %int %43 %int_1
OpStore %i %44
OpBranch %27
%28 = OpLabel
%45 = OpLoad %float %s
OpStore %o %45
OpReturn
OpFunctionEnd
)";
Move tests into namespaces (#1689) This CL moves the test into namespaces based on their directories.
2018-07-11 13:24:49 +00:00
SinglePassRunAndCheck<AggressiveDCEPass>(assembly, assembly, true, true);
ADCE: Only mark true breaks and continues of live loops This fixes issue #1075 - Mark continue when conditional branch with merge block. Only mark if merge block is not continue block. - Handle conditional branch break with preceding merge
2017-12-12 21:27:46 +00:00
}
TEST_F(AggressiveDCETest, NoEliminateIfContinue3) {
// Do not eliminate continue as conditional branch with merge instruction
// Note: SPIR-V edited to add merge instruction before continue.
//
// #version 430
//
// layout(std430) buffer U_t
// {
// float g_F[10];
// };
//
// layout(location = 0)out float o;
//
// void main(void)
// {
// float s = 0.0;
// for (int i=0; i<10; i++) {
// if (i % 2 == 0) continue;
// s += g_F[i];
// }
// o = s;
// }
const std::string assembly =
R"(OpCapability Shader
%1 = OpExtInstImport "GLSL.std.450"
OpMemoryModel Logical GLSL450
OpEntryPoint Fragment %main "main" %o
OpExecutionMode %main OriginUpperLeft
OpSource GLSL 430
OpName %main "main"
OpName %s "s"
OpName %i "i"
OpName %U_t "U_t"
OpMemberName %U_t 0 "g_F"
OpName %_ ""
OpName %o "o"
OpDecorate %_arr_float_uint_10 ArrayStride 4
OpMemberDecorate %U_t 0 Offset 0
OpDecorate %U_t BufferBlock
OpDecorate %_ DescriptorSet 0
OpDecorate %o Location 0
%void = OpTypeVoid
%10 = OpTypeFunction %void
%float = OpTypeFloat 32
%_ptr_Function_float = OpTypePointer Function %float
%float_0 = OpConstant %float 0
%int = OpTypeInt 32 1
%_ptr_Function_int = OpTypePointer Function %int
%int_0 = OpConstant %int 0
%int_10 = OpConstant %int 10
%bool = OpTypeBool
%int_2 = OpConstant %int 2
%uint = OpTypeInt 32 0
%uint_10 = OpConstant %uint 10
%_arr_float_uint_10 = OpTypeArray %float %uint_10
%U_t = OpTypeStruct %_arr_float_uint_10
%_ptr_Uniform_U_t = OpTypePointer Uniform %U_t
%_ = OpVariable %_ptr_Uniform_U_t Uniform
%_ptr_Uniform_float = OpTypePointer Uniform %float
%int_1 = OpConstant %int 1
%_ptr_Output_float = OpTypePointer Output %float
%o = OpVariable %_ptr_Output_float Output
%main = OpFunction %void None %10
%26 = OpLabel
%s = OpVariable %_ptr_Function_float Function
%i = OpVariable %_ptr_Function_int Function
OpStore %s %float_0
OpStore %i %int_0
OpBranch %27
%27 = OpLabel
OpLoopMerge %28 %29 None
OpBranch %30
%30 = OpLabel
%31 = OpLoad %int %i
%32 = OpSLessThan %bool %31 %int_10
OpBranchConditional %32 %33 %28
%33 = OpLabel
%34 = OpLoad %int %i
%35 = OpSMod %int %34 %int_2
%36 = OpIEqual %bool %35 %int_0
OpSelectionMerge %37 None
OpBranchConditional %36 %29 %37
%37 = OpLabel
%38 = OpLoad %int %i
%39 = OpAccessChain %_ptr_Uniform_float %_ %int_0 %38
%40 = OpLoad %float %39
%41 = OpLoad %float %s
%42 = OpFAdd %float %41 %40
OpStore %s %42
OpBranch %29
%29 = OpLabel
%43 = OpLoad %int %i
%44 = OpIAdd %int %43 %int_1
OpStore %i %44
OpBranch %27
%28 = OpLabel
%45 = OpLoad %float %s
OpStore %o %45
OpReturn
OpFunctionEnd
)";
Move tests into namespaces (#1689) This CL moves the test into namespaces based on their directories.
2018-07-11 13:24:49 +00:00
SinglePassRunAndCheck<AggressiveDCEPass>(assembly, assembly, true, true);
ADCE: Only mark true breaks and continues of live loops This fixes issue #1075 - Mark continue when conditional branch with merge block. Only mark if merge block is not continue block. - Handle conditional branch break with preceding merge
2017-12-12 21:27:46 +00:00
}
Validate pointer variables (#2111) Fixes #2104 * Checks the rules for logical addressing and variable pointers * Has an out for relaxed logical pointers * Updated PassFixture to expose validator options * enabled relaxed logical pointers for some tests * New validator tests
2018-11-27 21:47:10 +00:00
// This is not valid input and ADCE does not support variable pointers and only
// supports shaders.
Allow pointers to pointers in logical addressing mode. A few optimizations are updates to handle code that is suppose to be using the logical addressing mode, but still has variables that contain pointers as long as the pointer are to opaque objects. This is called "relaxed logical addressing". |Instruction::GetBaseAddress| will check that pointers that are use meet the relaxed logical addressing rules. Optimization that now handle relaxed logical addressing instead of logical addressing are: - aggressive dead-code elimination - local access chain convert - local store elimination passes.
2017-12-11 18:10:24 +00:00
TEST_F(AggressiveDCETest, PointerVariable) {
// ADCE is able to handle code that contains a load whose base address
// comes from a load and not an OpVariable. I want to see an instruction
// removed to be sure that ADCE is not exiting early.
const std::string before =
R"(OpCapability Shader
OpMemoryModel Logical GLSL450
OpEntryPoint Fragment %1 "main" %2
OpExecutionMode %1 OriginUpperLeft
OpMemberDecorate %_struct_3 0 Offset 0
OpDecorate %_runtimearr__struct_3 ArrayStride 16
OpMemberDecorate %_struct_5 0 Offset 0
OpDecorate %_struct_5 BufferBlock
OpMemberDecorate %_struct_6 0 Offset 0
OpDecorate %_struct_6 BufferBlock
OpDecorate %2 Location 0
OpDecorate %7 DescriptorSet 0
OpDecorate %7 Binding 0
OpDecorate %8 DescriptorSet 0
OpDecorate %8 Binding 1
%void = OpTypeVoid
%10 = OpTypeFunction %void
%int = OpTypeInt 32 1
%uint = OpTypeInt 32 0
%float = OpTypeFloat 32
%v4float = OpTypeVector %float 4
%_ptr_Output_v4float = OpTypePointer Output %v4float
%_ptr_Uniform_v4float = OpTypePointer Uniform %v4float
%_struct_3 = OpTypeStruct %v4float
%_runtimearr__struct_3 = OpTypeRuntimeArray %_struct_3
%_struct_5 = OpTypeStruct %_runtimearr__struct_3
%_ptr_Uniform__struct_5 = OpTypePointer Uniform %_struct_5
%_struct_6 = OpTypeStruct %int
%_ptr_Uniform__struct_6 = OpTypePointer Uniform %_struct_6
%_ptr_Function__ptr_Uniform__struct_5 = OpTypePointer Function %_ptr_Uniform__struct_5
%_ptr_Function__ptr_Uniform__struct_6 = OpTypePointer Function %_ptr_Uniform__struct_6
%int_0 = OpConstant %int 0
%uint_0 = OpConstant %uint 0
%2 = OpVariable %_ptr_Output_v4float Output
%7 = OpVariable %_ptr_Uniform__struct_5 Uniform
%8 = OpVariable %_ptr_Uniform__struct_6 Uniform
%1 = OpFunction %void None %10
%23 = OpLabel
%24 = OpVariable %_ptr_Function__ptr_Uniform__struct_5 Function
OpStore %24 %7
%26 = OpLoad %_ptr_Uniform__struct_5 %24
%27 = OpAccessChain %_ptr_Uniform_v4float %26 %int_0 %uint_0 %int_0
%28 = OpLoad %v4float %27
%29 = OpCopyObject %v4float %28
OpStore %2 %28
OpReturn
OpFunctionEnd
)";
const std::string after =
R"(OpCapability Shader
OpMemoryModel Logical GLSL450
OpEntryPoint Fragment %1 "main" %2
OpExecutionMode %1 OriginUpperLeft
OpMemberDecorate %_struct_3 0 Offset 0
OpDecorate %_runtimearr__struct_3 ArrayStride 16
OpMemberDecorate %_struct_5 0 Offset 0
OpDecorate %_struct_5 BufferBlock
OpDecorate %2 Location 0
OpDecorate %7 DescriptorSet 0
OpDecorate %7 Binding 0
%void = OpTypeVoid
%10 = OpTypeFunction %void
%int = OpTypeInt 32 1
%uint = OpTypeInt 32 0
%float = OpTypeFloat 32
%v4float = OpTypeVector %float 4
%_ptr_Output_v4float = OpTypePointer Output %v4float
%_ptr_Uniform_v4float = OpTypePointer Uniform %v4float
%_struct_3 = OpTypeStruct %v4float
%_runtimearr__struct_3 = OpTypeRuntimeArray %_struct_3
%_struct_5 = OpTypeStruct %_runtimearr__struct_3
%_ptr_Uniform__struct_5 = OpTypePointer Uniform %_struct_5
%_ptr_Function__ptr_Uniform__struct_5 = OpTypePointer Function %_ptr_Uniform__struct_5
%int_0 = OpConstant %int 0
%uint_0 = OpConstant %uint 0
%2 = OpVariable %_ptr_Output_v4float Output
%7 = OpVariable %_ptr_Uniform__struct_5 Uniform
%1 = OpFunction %void None %10
%23 = OpLabel
%24 = OpVariable %_ptr_Function__ptr_Uniform__struct_5 Function
OpStore %24 %7
%25 = OpLoad %_ptr_Uniform__struct_5 %24
%26 = OpAccessChain %_ptr_Uniform_v4float %25 %int_0 %uint_0 %int_0
%27 = OpLoad %v4float %26
OpStore %2 %27
OpReturn
OpFunctionEnd
)";
Validate pointer variables (#2111) Fixes #2104 * Checks the rules for logical addressing and variable pointers * Has an out for relaxed logical pointers * Updated PassFixture to expose validator options * enabled relaxed logical pointers for some tests * New validator tests
2018-11-27 21:47:10 +00:00
// The input is not valid and ADCE only supports shaders, but not variable
// pointers. Workaround this by enabling relaxed logical pointers in the
// validator.
ValidatorOptions()->relax_logical_pointer = true;
Move tests into namespaces (#1689) This CL moves the test into namespaces based on their directories.
2018-07-11 13:24:49 +00:00
SinglePassRunAndCheck<AggressiveDCEPass>(before, after, true, true);
Allow pointers to pointers in logical addressing mode. A few optimizations are updates to handle code that is suppose to be using the logical addressing mode, but still has variables that contain pointers as long as the pointer are to opaque objects. This is called "relaxed logical addressing". |Instruction::GetBaseAddress| will check that pointers that are use meet the relaxed logical addressing rules. Optimization that now handle relaxed logical addressing instead of logical addressing are: - aggressive dead-code elimination - local access chain convert - local store elimination passes.
2017-12-11 18:10:24 +00:00
}
Adding additional functionality to ADCE. Modified ADCE to remove dead globals. * Entry point and execution mode instructions are marked as alive * Reachable functions and their parameters are marked as alive * Instruction deletion now deferred until the end of the pass * Eliminated dead insts set, added IsDead to calculate that value instead * Ported applicable dead variable elimination tests * Ported dead constant elim tests Added dead function elimination to ADCE * ported dead function elim tests Added handling of decoration groups in ADCE * Uses a custom sorter to traverse decorations in a specific order * Simplifies necessary checks Updated -O and -Os pass lists.
2017-12-18 17:13:10 +00:00
// %dead is unused. Make sure we remove it along with its name.
TEST_F(AggressiveDCETest, RemoveUnreferenced) {
const std::string before =
R"(OpCapability Shader
OpCapability Linkage
%1 = OpExtInstImport "GLSL.std.450"
OpMemoryModel Logical GLSL450
OpEntryPoint Fragment %main "main"
OpExecutionMode %main OriginUpperLeft
OpSource GLSL 150
OpName %main "main"
OpName %dead "dead"
%void = OpTypeVoid
%5 = OpTypeFunction %void
%float = OpTypeFloat 32
%_ptr_Private_float = OpTypePointer Private %float
%dead = OpVariable %_ptr_Private_float Private
%main = OpFunction %void None %5
%8 = OpLabel
OpReturn
OpFunctionEnd
)";
const std::string after =
R"(OpCapability Shader
OpCapability Linkage
%1 = OpExtInstImport "GLSL.std.450"
OpMemoryModel Logical GLSL450
OpEntryPoint Fragment %main "main"
OpExecutionMode %main OriginUpperLeft
OpSource GLSL 150
OpName %main "main"
%void = OpTypeVoid
%5 = OpTypeFunction %void
%main = OpFunction %void None %5
%8 = OpLabel
OpReturn
OpFunctionEnd
)";
SetAssembleOptions(SPV_TEXT_TO_BINARY_OPTION_PRESERVE_NUMERIC_IDS);
Move tests into namespaces (#1689) This CL moves the test into namespaces based on their directories.
2018-07-11 13:24:49 +00:00
SinglePassRunAndCheck<AggressiveDCEPass>(before, after, true, true);
Adding additional functionality to ADCE. Modified ADCE to remove dead globals. * Entry point and execution mode instructions are marked as alive * Reachable functions and their parameters are marked as alive * Instruction deletion now deferred until the end of the pass * Eliminated dead insts set, added IsDead to calculate that value instead * Ported applicable dead variable elimination tests * Ported dead constant elim tests Added dead function elimination to ADCE * ported dead function elim tests Added handling of decoration groups in ADCE * Uses a custom sorter to traverse decorations in a specific order * Simplifies necessary checks Updated -O and -Os pass lists.
2017-12-18 17:13:10 +00:00
}
// Delete %dead because it is unreferenced. Then %initializer becomes
// unreferenced, so remove it as well.
TEST_F(AggressiveDCETest, RemoveUnreferencedWithInit1) {
const std::string before =
R"(OpCapability Shader
OpCapability Linkage
%1 = OpExtInstImport "GLSL.std.450"
OpMemoryModel Logical GLSL450
OpEntryPoint Fragment %main "main"
OpExecutionMode %main OriginUpperLeft
OpSource GLSL 150
OpName %main "main"
OpName %dead "dead"
OpName %initializer "initializer"
%void = OpTypeVoid
%6 = OpTypeFunction %void
%float = OpTypeFloat 32
%_ptr_Private_float = OpTypePointer Private %float
%initializer = OpVariable %_ptr_Private_float Private
%dead = OpVariable %_ptr_Private_float Private %initializer
%main = OpFunction %void None %6
%9 = OpLabel
OpReturn
OpFunctionEnd
)";
const std::string after =
R"(OpCapability Shader
OpCapability Linkage
%1 = OpExtInstImport "GLSL.std.450"
OpMemoryModel Logical GLSL450
OpEntryPoint Fragment %main "main"
OpExecutionMode %main OriginUpperLeft
OpSource GLSL 150
OpName %main "main"
%void = OpTypeVoid
%6 = OpTypeFunction %void
%main = OpFunction %void None %6
%9 = OpLabel
OpReturn
OpFunctionEnd
)";
SetAssembleOptions(SPV_TEXT_TO_BINARY_OPTION_PRESERVE_NUMERIC_IDS);
Move tests into namespaces (#1689) This CL moves the test into namespaces based on their directories.
2018-07-11 13:24:49 +00:00
SinglePassRunAndCheck<AggressiveDCEPass>(before, after, true, true);
Adding additional functionality to ADCE. Modified ADCE to remove dead globals. * Entry point and execution mode instructions are marked as alive * Reachable functions and their parameters are marked as alive * Instruction deletion now deferred until the end of the pass * Eliminated dead insts set, added IsDead to calculate that value instead * Ported applicable dead variable elimination tests * Ported dead constant elim tests Added dead function elimination to ADCE * ported dead function elim tests Added handling of decoration groups in ADCE * Uses a custom sorter to traverse decorations in a specific order * Simplifies necessary checks Updated -O and -Os pass lists.
2017-12-18 17:13:10 +00:00
}
// Keep %live because it is used, and its initializer.
TEST_F(AggressiveDCETest, KeepReferenced) {
const std::string before =
R"(OpCapability Shader
OpCapability Linkage
%1 = OpExtInstImport "GLSL.std.450"
OpMemoryModel Logical GLSL450
OpEntryPoint Fragment %main "main" %output
OpExecutionMode %main OriginUpperLeft
OpSource GLSL 150
OpName %main "main"
OpName %live "live"
OpName %initializer "initializer"
OpName %output "output"
%void = OpTypeVoid
%6 = OpTypeFunction %void
%float = OpTypeFloat 32
%_ptr_Private_float = OpTypePointer Private %float
Validate variable initializer type (#2668) Fixes #249 * The pointed to type of Result Type must match the initializer type * Had to update some opt tests to be valid
2019-06-15 04:34:18 +00:00
%initializer = OpConstant %float 0
Adding additional functionality to ADCE. Modified ADCE to remove dead globals. * Entry point and execution mode instructions are marked as alive * Reachable functions and their parameters are marked as alive * Instruction deletion now deferred until the end of the pass * Eliminated dead insts set, added IsDead to calculate that value instead * Ported applicable dead variable elimination tests * Ported dead constant elim tests Added dead function elimination to ADCE * ported dead function elim tests Added handling of decoration groups in ADCE * Uses a custom sorter to traverse decorations in a specific order * Simplifies necessary checks Updated -O and -Os pass lists.
2017-12-18 17:13:10 +00:00
%live = OpVariable %_ptr_Private_float Private %initializer
%_ptr_Output_float = OpTypePointer Output %float
%output = OpVariable %_ptr_Output_float Output
%main = OpFunction %void None %6
%9 = OpLabel
%10 = OpLoad %float %live
OpStore %output %10
OpReturn
OpFunctionEnd
)";
SetAssembleOptions(SPV_TEXT_TO_BINARY_OPTION_PRESERVE_NUMERIC_IDS);
Move tests into namespaces (#1689) This CL moves the test into namespaces based on their directories.
2018-07-11 13:24:49 +00:00
SinglePassRunAndCheck<AggressiveDCEPass>(before, before, true, true);
Adding additional functionality to ADCE. Modified ADCE to remove dead globals. * Entry point and execution mode instructions are marked as alive * Reachable functions and their parameters are marked as alive * Instruction deletion now deferred until the end of the pass * Eliminated dead insts set, added IsDead to calculate that value instead * Ported applicable dead variable elimination tests * Ported dead constant elim tests Added dead function elimination to ADCE * ported dead function elim tests Added handling of decoration groups in ADCE * Uses a custom sorter to traverse decorations in a specific order * Simplifies necessary checks Updated -O and -Os pass lists.
2017-12-18 17:13:10 +00:00
}
// This test that the decoration associated with a variable are removed when the
// variable is removed.
TEST_F(AggressiveDCETest, RemoveVariableAndDecorations) {
const std::string before =
R"(OpCapability Shader
%1 = OpExtInstImport "GLSL.std.450"
OpMemoryModel Logical GLSL450
OpEntryPoint Vertex %main "main"
OpSource GLSL 450
OpName %main "main"
OpName %B "B"
OpMemberName %B 0 "a"
OpName %Bdat "Bdat"
OpMemberDecorate %B 0 Offset 0
OpDecorate %B BufferBlock
OpDecorate %Bdat DescriptorSet 0
OpDecorate %Bdat Binding 0
%void = OpTypeVoid
%6 = OpTypeFunction %void
%uint = OpTypeInt 32 0
%B = OpTypeStruct %uint
%_ptr_Uniform_B = OpTypePointer Uniform %B
%Bdat = OpVariable %_ptr_Uniform_B Uniform
%int = OpTypeInt 32 1
%int_0 = OpConstant %int 0
%uint_1 = OpConstant %uint 1
%_ptr_Uniform_uint = OpTypePointer Uniform %uint
%main = OpFunction %void None %6
%13 = OpLabel
OpReturn
OpFunctionEnd
)";
const std::string after =
R"(OpCapability Shader
%1 = OpExtInstImport "GLSL.std.450"
OpMemoryModel Logical GLSL450
OpEntryPoint Vertex %main "main"
OpSource GLSL 450
OpName %main "main"
%void = OpTypeVoid
%6 = OpTypeFunction %void
%main = OpFunction %void None %6
%13 = OpLabel
OpReturn
OpFunctionEnd
)";
SetAssembleOptions(SPV_TEXT_TO_BINARY_OPTION_PRESERVE_NUMERIC_IDS);
Move tests into namespaces (#1689) This CL moves the test into namespaces based on their directories.
2018-07-11 13:24:49 +00:00
SinglePassRunAndCheck<AggressiveDCEPass>(before, after, true, true);
Adding additional functionality to ADCE. Modified ADCE to remove dead globals. * Entry point and execution mode instructions are marked as alive * Reachable functions and their parameters are marked as alive * Instruction deletion now deferred until the end of the pass * Eliminated dead insts set, added IsDead to calculate that value instead * Ported applicable dead variable elimination tests * Ported dead constant elim tests Added dead function elimination to ADCE * ported dead function elim tests Added handling of decoration groups in ADCE * Uses a custom sorter to traverse decorations in a specific order * Simplifies necessary checks Updated -O and -Os pass lists.
2017-12-18 17:13:10 +00:00
}
Adding support for switch removal in ADCE * Updated code to handle switches * Enabled disabled test and added a couple new ones
2018-01-15 18:25:45 +00:00
TEST_F(AggressiveDCETest, DeadNestedSwitch) {
const std::string text = R"(
; CHECK: OpLabel
; CHECK: OpBranch [[block:%\w+]]
; CHECK-NOT: OpSwitch
; CHECK-NEXT: [[block]] = OpLabel
; CHECK: OpBranch [[block:%\w+]]
; CHECK-NOT: OpSwitch
; CHECK-NEXT: [[block]] = OpLabel
; CHECK-NEXT: OpStore
OpCapability Shader
OpMemoryModel Logical GLSL450
OpEntryPoint Fragment %func "func" %x
OpExecutionMode %func OriginUpperLeft
OpName %func "func"
%void = OpTypeVoid
%1 = OpTypeFunction %void
%uint = OpTypeInt 32 0
%uint_0 = OpConstant %uint 0
%uint_ptr_Output = OpTypePointer Output %uint
%uint_ptr_Input = OpTypePointer Input %uint
%x = OpVariable %uint_ptr_Output Output
%a = OpVariable %uint_ptr_Input Input
%func = OpFunction %void None %1
%entry = OpLabel
OpBranch %header
%header = OpLabel
%ld = OpLoad %uint %a
OpLoopMerge %merge %continue None
OpBranch %postheader
%postheader = OpLabel
; This switch doesn't require an OpSelectionMerge and is nested in the dead loop.
OpSwitch %ld %merge 0 %extra 1 %continue
%extra = OpLabel
OpBranch %continue
%continue = OpLabel
OpBranch %header
%merge = OpLabel
OpStore %x %uint_0
OpReturn
OpFunctionEnd
)";
Move tests into namespaces (#1689) This CL moves the test into namespaces based on their directories.
2018-07-11 13:24:49 +00:00
SinglePassRunAndMatch<AggressiveDCEPass>(text, true);
Adding support for switch removal in ADCE * Updated code to handle switches * Enabled disabled test and added a couple new ones
2018-01-15 18:25:45 +00:00
}
TEST_F(AggressiveDCETest, LiveNestedSwitch) {
const std::string text = R"(OpCapability Shader
OpMemoryModel Logical GLSL450
Validate static uses of interfaces Fixes #1120 Checks that all static uses of the Input and Output variables are listed as interfaces in each corresponding entry point declaration. * Changed validation state to track interface lists * updated many tests * Modified validation state to store entry point names * Combined with interface list and called EntryPointDescription * Updated uses * Changed interface validation error messages to output entry point name in addtion to ID
2018-05-31 13:07:09 +00:00
OpEntryPoint Fragment %func "func" %3 %10
Adding support for switch removal in ADCE * Updated code to handle switches * Enabled disabled test and added a couple new ones
2018-01-15 18:25:45 +00:00
OpExecutionMode %func OriginUpperLeft
OpName %func "func"
%void = OpTypeVoid
%1 = OpTypeFunction %void
%uint = OpTypeInt 32 0
%uint_0 = OpConstant %uint 0
%uint_1 = OpConstant %uint 1
%_ptr_Output_uint = OpTypePointer Output %uint
%_ptr_Input_uint = OpTypePointer Input %uint
%3 = OpVariable %_ptr_Output_uint Output
%10 = OpVariable %_ptr_Input_uint Input
%func = OpFunction %void None %1
%11 = OpLabel
OpBranch %12
%12 = OpLabel
%13 = OpLoad %uint %10
OpLoopMerge %14 %15 None
OpBranch %16
%16 = OpLabel
Require an OpSelectionMerge before an OpSwitch (#4154) Fixes #4153 * OpSwitch is required to be preceeded by OpSelectionMerge * Update newly invalid tests
2021-03-02 20:40:56 +00:00
OpSelectionMerge %18 None
Use structural dominance to validate cfg (#4832) * Structural dominance introduced in SPIR-V 1.6 rev2 * Changes the structured cfg validation to use structural dominance * structural dominance is based on a cfg where merge and continue declarations are counted as graph edges * Basic blocks now track structural predecessors and structural successors * Add validation for entry into a loop * Fixed an issue with inlining a single block loop * The continue target needs to be moved to the latch block * Simplify the calculation of structured exits * no longer requires block depth * Update many invalid tests
2022-06-30 03:32:20 +00:00
OpSwitch %13 %18 0 %17 1 %19
Adding support for switch removal in ADCE * Updated code to handle switches * Enabled disabled test and added a couple new ones
2018-01-15 18:25:45 +00:00
%17 = OpLabel
OpStore %3 %uint_1
Use structural dominance to validate cfg (#4832) * Structural dominance introduced in SPIR-V 1.6 rev2 * Changes the structured cfg validation to use structural dominance * structural dominance is based on a cfg where merge and continue declarations are counted as graph edges * Basic blocks now track structural predecessors and structural successors * Add validation for entry into a loop * Fixed an issue with inlining a single block loop * The continue target needs to be moved to the latch block * Simplify the calculation of structured exits * no longer requires block depth * Update many invalid tests
2022-06-30 03:32:20 +00:00
OpBranch %19
%19 = OpLabel
Adding support for switch removal in ADCE * Updated code to handle switches * Enabled disabled test and added a couple new ones
2018-01-15 18:25:45 +00:00
OpBranch %15
%15 = OpLabel
OpBranch %12
Require an OpSelectionMerge before an OpSwitch (#4154) Fixes #4153 * OpSwitch is required to be preceeded by OpSelectionMerge * Update newly invalid tests
2021-03-02 20:40:56 +00:00
%18 = OpLabel
OpBranch %14
Adding support for switch removal in ADCE * Updated code to handle switches * Enabled disabled test and added a couple new ones
2018-01-15 18:25:45 +00:00
%14 = OpLabel
OpStore %3 %uint_0
OpReturn
OpFunctionEnd
)";
SetAssembleOptions(SPV_TEXT_TO_BINARY_OPTION_PRESERVE_NUMERIC_IDS);
Move tests into namespaces (#1689) This CL moves the test into namespaces based on their directories.
2018-07-11 13:24:49 +00:00
SinglePassRunAndCheck<AggressiveDCEPass>(text, text, false, true);
Adding support for switch removal in ADCE * Updated code to handle switches * Enabled disabled test and added a couple new ones
2018-01-15 18:25:45 +00:00
}
Adding additional functionality to ADCE. Modified ADCE to remove dead globals. * Entry point and execution mode instructions are marked as alive * Reachable functions and their parameters are marked as alive * Instruction deletion now deferred until the end of the pass * Eliminated dead insts set, added IsDead to calculate that value instead * Ported applicable dead variable elimination tests * Ported dead constant elim tests Added dead function elimination to ADCE * ported dead function elim tests Added handling of decoration groups in ADCE * Uses a custom sorter to traverse decorations in a specific order * Simplifies necessary checks Updated -O and -Os pass lists.
2017-12-18 17:13:10 +00:00
TEST_F(AggressiveDCETest, BasicDeleteDeadFunction) {
// The function Dead should be removed because it is never called.
const std::vector<const char*> common_code = {
// clang-format off
"OpCapability Shader",
"OpMemoryModel Logical GLSL450",
"OpEntryPoint Fragment %main \"main\"",
"OpName %main \"main\"",
"OpName %Live \"Live\"",
"%void = OpTypeVoid",
"%7 = OpTypeFunction %void",
"%main = OpFunction %void None %7",
"%15 = OpLabel",
"%16 = OpFunctionCall %void %Live",
"%17 = OpFunctionCall %void %Live",
"OpReturn",
"OpFunctionEnd",
"%Live = OpFunction %void None %7",
"%20 = OpLabel",
"OpReturn",
"OpFunctionEnd"
// clang-format on
};
const std::vector<const char*> dead_function = {
// clang-format off
"%Dead = OpFunction %void None %7",
"%19 = OpLabel",
"OpReturn",
"OpFunctionEnd",
// clang-format on
};
SetAssembleOptions(SPV_TEXT_TO_BINARY_OPTION_PRESERVE_NUMERIC_IDS);
Move tests into namespaces (#1689) This CL moves the test into namespaces based on their directories.
2018-07-11 13:24:49 +00:00
SinglePassRunAndCheck<AggressiveDCEPass>(
Adding additional functionality to ADCE. Modified ADCE to remove dead globals. * Entry point and execution mode instructions are marked as alive * Reachable functions and their parameters are marked as alive * Instruction deletion now deferred until the end of the pass * Eliminated dead insts set, added IsDead to calculate that value instead * Ported applicable dead variable elimination tests * Ported dead constant elim tests Added dead function elimination to ADCE * ported dead function elim tests Added handling of decoration groups in ADCE * Uses a custom sorter to traverse decorations in a specific order * Simplifies necessary checks Updated -O and -Os pass lists.
2017-12-18 17:13:10 +00:00
JoinAllInsts(Concat(common_code, dead_function)),
JoinAllInsts(common_code), /* skip_nop = */ true);
}
TEST_F(AggressiveDCETest, BasicKeepLiveFunction) {
// Everything is reachable from an entry point, so no functions should be
// deleted.
const std::vector<const char*> text = {
// clang-format off
"OpCapability Shader",
"OpMemoryModel Logical GLSL450",
"OpEntryPoint Fragment %main \"main\"",
"OpName %main \"main\"",
"OpName %Live1 \"Live1\"",
"OpName %Live2 \"Live2\"",
"%void = OpTypeVoid",
"%7 = OpTypeFunction %void",
"%main = OpFunction %void None %7",
"%15 = OpLabel",
"%16 = OpFunctionCall %void %Live2",
"%17 = OpFunctionCall %void %Live1",
"OpReturn",
"OpFunctionEnd",
"%Live1 = OpFunction %void None %7",
"%19 = OpLabel",
"OpReturn",
"OpFunctionEnd",
"%Live2 = OpFunction %void None %7",
"%20 = OpLabel",
"OpReturn",
"OpFunctionEnd"
// clang-format on
};
SetAssembleOptions(SPV_TEXT_TO_BINARY_OPTION_PRESERVE_NUMERIC_IDS);
std::string assembly = JoinAllInsts(text);
Move tests into namespaces (#1689) This CL moves the test into namespaces based on their directories.
2018-07-11 13:24:49 +00:00
auto result = SinglePassRunAndDisassemble<AggressiveDCEPass>(
Adding additional functionality to ADCE. Modified ADCE to remove dead globals. * Entry point and execution mode instructions are marked as alive * Reachable functions and their parameters are marked as alive * Instruction deletion now deferred until the end of the pass * Eliminated dead insts set, added IsDead to calculate that value instead * Ported applicable dead variable elimination tests * Ported dead constant elim tests Added dead function elimination to ADCE * ported dead function elim tests Added handling of decoration groups in ADCE * Uses a custom sorter to traverse decorations in a specific order * Simplifies necessary checks Updated -O and -Os pass lists.
2017-12-18 17:13:10 +00:00
assembly, /* skip_nop = */ true, /* do_validation = */ false);
Move tests into namespaces (#1689) This CL moves the test into namespaces based on their directories.
2018-07-11 13:24:49 +00:00
EXPECT_EQ(Pass::Status::SuccessWithoutChange, std::get<1>(result));
Adding additional functionality to ADCE. Modified ADCE to remove dead globals. * Entry point and execution mode instructions are marked as alive * Reachable functions and their parameters are marked as alive * Instruction deletion now deferred until the end of the pass * Eliminated dead insts set, added IsDead to calculate that value instead * Ported applicable dead variable elimination tests * Ported dead constant elim tests Added dead function elimination to ADCE * ported dead function elim tests Added handling of decoration groups in ADCE * Uses a custom sorter to traverse decorations in a specific order * Simplifies necessary checks Updated -O and -Os pass lists.
2017-12-18 17:13:10 +00:00
EXPECT_EQ(assembly, std::get<0>(result));
}
TEST_F(AggressiveDCETest, BasicRemoveDecorationsAndNames) {
// We want to remove the names and decorations associated with results that
// are removed. This test will check for that.
const std::string text = R"(
OpCapability Shader
OpMemoryModel Logical GLSL450
OpEntryPoint Vertex %main "main"
OpName %main "main"
OpName %Dead "Dead"
OpName %x "x"
OpName %y "y"
OpName %z "z"
OpDecorate %x RelaxedPrecision
OpDecorate %y RelaxedPrecision
OpDecorate %z RelaxedPrecision
OpDecorate %6 RelaxedPrecision
OpDecorate %7 RelaxedPrecision
OpDecorate %8 RelaxedPrecision
%void = OpTypeVoid
%10 = OpTypeFunction %void
%float = OpTypeFloat 32
%_ptr_Function_float = OpTypePointer Function %float
%float_1 = OpConstant %float 1
%main = OpFunction %void None %10
%14 = OpLabel
OpReturn
OpFunctionEnd
%Dead = OpFunction %void None %10
%15 = OpLabel
%x = OpVariable %_ptr_Function_float Function
%y = OpVariable %_ptr_Function_float Function
%z = OpVariable %_ptr_Function_float Function
OpStore %x %float_1
OpStore %y %float_1
%6 = OpLoad %float %x
%7 = OpLoad %float %y
%8 = OpFAdd %float %6 %7
OpStore %z %8
OpReturn
OpFunctionEnd)";
const std::string expected_output = R"(OpCapability Shader
OpMemoryModel Logical GLSL450
OpEntryPoint Vertex %main "main"
OpName %main "main"
%void = OpTypeVoid
%10 = OpTypeFunction %void
%main = OpFunction %void None %10
%14 = OpLabel
OpReturn
OpFunctionEnd
)";
SetAssembleOptions(SPV_TEXT_TO_BINARY_OPTION_PRESERVE_NUMERIC_IDS);
Move tests into namespaces (#1689) This CL moves the test into namespaces based on their directories.
2018-07-11 13:24:49 +00:00
SinglePassRunAndCheck<AggressiveDCEPass>(text, expected_output,
/* skip_nop = */ true);
Adding additional functionality to ADCE. Modified ADCE to remove dead globals. * Entry point and execution mode instructions are marked as alive * Reachable functions and their parameters are marked as alive * Instruction deletion now deferred until the end of the pass * Eliminated dead insts set, added IsDead to calculate that value instead * Ported applicable dead variable elimination tests * Ported dead constant elim tests Added dead function elimination to ADCE * ported dead function elim tests Added handling of decoration groups in ADCE * Uses a custom sorter to traverse decorations in a specific order * Simplifies necessary checks Updated -O and -Os pass lists.
2017-12-18 17:13:10 +00:00
}
TEST_F(AggressiveDCETest, BasicAllDeadConstants) {
const std::string text = R"(
; CHECK-NOT: OpConstant
OpCapability Shader
OpCapability Float64
%1 = OpExtInstImport "GLSL.std.450"
OpMemoryModel Logical GLSL450
OpEntryPoint Vertex %main "main"
OpName %main "main"
%void = OpTypeVoid
%4 = OpTypeFunction %void
%bool = OpTypeBool
%true = OpConstantTrue %bool
%false = OpConstantFalse %bool
%int = OpTypeInt 32 1
%9 = OpConstant %int 1
%uint = OpTypeInt 32 0
%11 = OpConstant %uint 2
%float = OpTypeFloat 32
hex_float: Use max_digits10 for the float precision CPPreference.com has this description of digits10: “The value of std::numeric_limits<T>::digits10 is the number of base-10 digits that can be represented by the type T without change, that is, any number with this many significant decimal digits can be converted to a value of type T and back to decimal form, without change due to rounding or overflow.” This means that any number with this many digits can be represented accurately in the corresponding type. A change in any digit in a number after that may or may not cause it a different bitwise representation. Therefore this isn’t necessarily enough precision to accurately represent the value in text. Instead we need max_digits10 which has the following description: “The value of std::numeric_limits<T>::max_digits10 is the number of base-10 digits that are necessary to uniquely represent all distinct values of the type T, such as necessary for serialization/deserialization to text.” The patch includes a test case in hex_float_test which tries to do a round-robin conversion of a number that requires more than 6 decimal places to be accurately represented. This would fail without the patch. Sadly this also breaks a bunch of other tests. Some of the tests in hex_float_test use ldexp and then compare it with a value which is not the same as the one returned by ldexp but instead is the value rounded to 6 decimals. Others use values that are not evenly representable as a binary floating fraction but then happened to generate the same value when rounded to 6 decimals. Where the actual value didn’t seem to matter these have been changed with different values that can be represented as a binary fraction.
2018-03-30 23:35:45 +00:00
%13 = OpConstant %float 3.1415
Adding additional functionality to ADCE. Modified ADCE to remove dead globals. * Entry point and execution mode instructions are marked as alive * Reachable functions and their parameters are marked as alive * Instruction deletion now deferred until the end of the pass * Eliminated dead insts set, added IsDead to calculate that value instead * Ported applicable dead variable elimination tests * Ported dead constant elim tests Added dead function elimination to ADCE * ported dead function elim tests Added handling of decoration groups in ADCE * Uses a custom sorter to traverse decorations in a specific order * Simplifies necessary checks Updated -O and -Os pass lists.
2017-12-18 17:13:10 +00:00
%double = OpTypeFloat 64
%15 = OpConstant %double 3.14159265358979
%main = OpFunction %void None %4
%16 = OpLabel
OpReturn
OpFunctionEnd
)";
Move tests into namespaces (#1689) This CL moves the test into namespaces based on their directories.
2018-07-11 13:24:49 +00:00
SinglePassRunAndMatch<AggressiveDCEPass>(text, true);
Adding additional functionality to ADCE. Modified ADCE to remove dead globals. * Entry point and execution mode instructions are marked as alive * Reachable functions and their parameters are marked as alive * Instruction deletion now deferred until the end of the pass * Eliminated dead insts set, added IsDead to calculate that value instead * Ported applicable dead variable elimination tests * Ported dead constant elim tests Added dead function elimination to ADCE * ported dead function elim tests Added handling of decoration groups in ADCE * Uses a custom sorter to traverse decorations in a specific order * Simplifies necessary checks Updated -O and -Os pass lists.
2017-12-18 17:13:10 +00:00
}
TEST_F(AggressiveDCETest, BasicNoneDeadConstants) {
const std::vector<const char*> text = {
// clang-format off
"OpCapability Shader",
"OpCapability Float64",
"%1 = OpExtInstImport \"GLSL.std.450\"",
"OpMemoryModel Logical GLSL450",
"OpEntryPoint Vertex %main \"main\" %btv %bfv %iv %uv %fv %dv",
"OpName %main \"main\"",
"OpName %btv \"btv\"",
"OpName %bfv \"bfv\"",
"OpName %iv \"iv\"",
"OpName %uv \"uv\"",
"OpName %fv \"fv\"",
"OpName %dv \"dv\"",
"%void = OpTypeVoid",
"%10 = OpTypeFunction %void",
"%bool = OpTypeBool",
"%_ptr_Output_bool = OpTypePointer Output %bool",
"%true = OpConstantTrue %bool",
"%false = OpConstantFalse %bool",
"%int = OpTypeInt 32 1",
"%_ptr_Output_int = OpTypePointer Output %int",
"%int_1 = OpConstant %int 1",
"%uint = OpTypeInt 32 0",
"%_ptr_Output_uint = OpTypePointer Output %uint",
"%uint_2 = OpConstant %uint 2",
"%float = OpTypeFloat 32",
"%_ptr_Output_float = OpTypePointer Output %float",
hex_float: Use max_digits10 for the float precision CPPreference.com has this description of digits10: “The value of std::numeric_limits<T>::digits10 is the number of base-10 digits that can be represented by the type T without change, that is, any number with this many significant decimal digits can be converted to a value of type T and back to decimal form, without change due to rounding or overflow.” This means that any number with this many digits can be represented accurately in the corresponding type. A change in any digit in a number after that may or may not cause it a different bitwise representation. Therefore this isn’t necessarily enough precision to accurately represent the value in text. Instead we need max_digits10 which has the following description: “The value of std::numeric_limits<T>::max_digits10 is the number of base-10 digits that are necessary to uniquely represent all distinct values of the type T, such as necessary for serialization/deserialization to text.” The patch includes a test case in hex_float_test which tries to do a round-robin conversion of a number that requires more than 6 decimal places to be accurately represented. This would fail without the patch. Sadly this also breaks a bunch of other tests. Some of the tests in hex_float_test use ldexp and then compare it with a value which is not the same as the one returned by ldexp but instead is the value rounded to 6 decimals. Others use values that are not evenly representable as a binary floating fraction but then happened to generate the same value when rounded to 6 decimals. Where the actual value didn’t seem to matter these have been changed with different values that can be represented as a binary fraction.
2018-03-30 23:35:45 +00:00
"%float_3_1415 = OpConstant %float 3.1415",
Adding additional functionality to ADCE. Modified ADCE to remove dead globals. * Entry point and execution mode instructions are marked as alive * Reachable functions and their parameters are marked as alive * Instruction deletion now deferred until the end of the pass * Eliminated dead insts set, added IsDead to calculate that value instead * Ported applicable dead variable elimination tests * Ported dead constant elim tests Added dead function elimination to ADCE * ported dead function elim tests Added handling of decoration groups in ADCE * Uses a custom sorter to traverse decorations in a specific order * Simplifies necessary checks Updated -O and -Os pass lists.
2017-12-18 17:13:10 +00:00
"%double = OpTypeFloat 64",
"%_ptr_Output_double = OpTypePointer Output %double",
"%double_3_14159265358979 = OpConstant %double 3.14159265358979",
"%btv = OpVariable %_ptr_Output_bool Output",
"%bfv = OpVariable %_ptr_Output_bool Output",
"%iv = OpVariable %_ptr_Output_int Output",
"%uv = OpVariable %_ptr_Output_uint Output",
"%fv = OpVariable %_ptr_Output_float Output",
"%dv = OpVariable %_ptr_Output_double Output",
"%main = OpFunction %void None %10",
"%27 = OpLabel",
"OpStore %btv %true",
"OpStore %bfv %false",
"OpStore %iv %int_1",
"OpStore %uv %uint_2",
hex_float: Use max_digits10 for the float precision CPPreference.com has this description of digits10: “The value of std::numeric_limits<T>::digits10 is the number of base-10 digits that can be represented by the type T without change, that is, any number with this many significant decimal digits can be converted to a value of type T and back to decimal form, without change due to rounding or overflow.” This means that any number with this many digits can be represented accurately in the corresponding type. A change in any digit in a number after that may or may not cause it a different bitwise representation. Therefore this isn’t necessarily enough precision to accurately represent the value in text. Instead we need max_digits10 which has the following description: “The value of std::numeric_limits<T>::max_digits10 is the number of base-10 digits that are necessary to uniquely represent all distinct values of the type T, such as necessary for serialization/deserialization to text.” The patch includes a test case in hex_float_test which tries to do a round-robin conversion of a number that requires more than 6 decimal places to be accurately represented. This would fail without the patch. Sadly this also breaks a bunch of other tests. Some of the tests in hex_float_test use ldexp and then compare it with a value which is not the same as the one returned by ldexp but instead is the value rounded to 6 decimals. Others use values that are not evenly representable as a binary floating fraction but then happened to generate the same value when rounded to 6 decimals. Where the actual value didn’t seem to matter these have been changed with different values that can be represented as a binary fraction.
2018-03-30 23:35:45 +00:00
"OpStore %fv %float_3_1415",
Adding additional functionality to ADCE. Modified ADCE to remove dead globals. * Entry point and execution mode instructions are marked as alive * Reachable functions and their parameters are marked as alive * Instruction deletion now deferred until the end of the pass * Eliminated dead insts set, added IsDead to calculate that value instead * Ported applicable dead variable elimination tests * Ported dead constant elim tests Added dead function elimination to ADCE * ported dead function elim tests Added handling of decoration groups in ADCE * Uses a custom sorter to traverse decorations in a specific order * Simplifies necessary checks Updated -O and -Os pass lists.
2017-12-18 17:13:10 +00:00
"OpStore %dv %double_3_14159265358979",
"OpReturn",
"OpFunctionEnd",
// clang-format on
};
// All constants are used, so none of them should be eliminated.
Move tests into namespaces (#1689) This CL moves the test into namespaces based on their directories.
2018-07-11 13:24:49 +00:00
SinglePassRunAndCheck<AggressiveDCEPass>(
Adding additional functionality to ADCE. Modified ADCE to remove dead globals. * Entry point and execution mode instructions are marked as alive * Reachable functions and their parameters are marked as alive * Instruction deletion now deferred until the end of the pass * Eliminated dead insts set, added IsDead to calculate that value instead * Ported applicable dead variable elimination tests * Ported dead constant elim tests Added dead function elimination to ADCE * ported dead function elim tests Added handling of decoration groups in ADCE * Uses a custom sorter to traverse decorations in a specific order * Simplifies necessary checks Updated -O and -Os pass lists.
2017-12-18 17:13:10 +00:00
JoinAllInsts(text), JoinAllInsts(text), /* skip_nop = */ true);
}
Reduce number of test targets (#2024) This CL takes the various opt unit tests and makes a single executable instead of one per test. This reduces the number of build targets by ~125 when building with ninja.
2018-11-01 14:19:37 +00:00
struct AggressiveEliminateDeadConstantTestCase {
Adding additional functionality to ADCE. Modified ADCE to remove dead globals. * Entry point and execution mode instructions are marked as alive * Reachable functions and their parameters are marked as alive * Instruction deletion now deferred until the end of the pass * Eliminated dead insts set, added IsDead to calculate that value instead * Ported applicable dead variable elimination tests * Ported dead constant elim tests Added dead function elimination to ADCE * ported dead function elim tests Added handling of decoration groups in ADCE * Uses a custom sorter to traverse decorations in a specific order * Simplifies necessary checks Updated -O and -Os pass lists.
2017-12-18 17:13:10 +00:00
// Type declarations and constants that should be kept.
std::vector<std::string> used_consts;
// Instructions that refer to constants, this is added to create uses for
// some constants so they won't be treated as dead constants.
std::vector<std::string> main_insts;
// Dead constants that should be removed.
std::vector<std::string> dead_consts;
// Expectations
std::vector<std::string> checks;
};
Reduce number of test targets (#2024) This CL takes the various opt unit tests and makes a single executable instead of one per test. This reduces the number of build targets by ~125 when building with ninja.
2018-11-01 14:19:37 +00:00
// All types that are potentially required in
// AggressiveEliminateDeadConstantTest.
Adding additional functionality to ADCE. Modified ADCE to remove dead globals. * Entry point and execution mode instructions are marked as alive * Reachable functions and their parameters are marked as alive * Instruction deletion now deferred until the end of the pass * Eliminated dead insts set, added IsDead to calculate that value instead * Ported applicable dead variable elimination tests * Ported dead constant elim tests Added dead function elimination to ADCE * ported dead function elim tests Added handling of decoration groups in ADCE * Uses a custom sorter to traverse decorations in a specific order * Simplifies necessary checks Updated -O and -Os pass lists.
2017-12-18 17:13:10 +00:00
const std::vector<std::string> CommonTypes = {
// clang-format off
// scalar types
"%bool = OpTypeBool",
"%uint = OpTypeInt 32 0",
"%int = OpTypeInt 32 1",
"%float = OpTypeFloat 32",
"%double = OpTypeFloat 64",
// vector types
"%v2bool = OpTypeVector %bool 2",
"%v2uint = OpTypeVector %uint 2",
"%v2int = OpTypeVector %int 2",
"%v3int = OpTypeVector %int 3",
"%v4int = OpTypeVector %int 4",
"%v2float = OpTypeVector %float 2",
"%v3float = OpTypeVector %float 3",
"%v2double = OpTypeVector %double 2",
// variable pointer types
"%_pf_bool = OpTypePointer Output %bool",
"%_pf_uint = OpTypePointer Output %uint",
"%_pf_int = OpTypePointer Output %int",
"%_pf_float = OpTypePointer Output %float",
"%_pf_double = OpTypePointer Output %double",
"%_pf_v2int = OpTypePointer Output %v2int",
"%_pf_v3int = OpTypePointer Output %v3int",
"%_pf_v2float = OpTypePointer Output %v2float",
"%_pf_v3float = OpTypePointer Output %v3float",
"%_pf_v2double = OpTypePointer Output %v2double",
// struct types
"%inner_struct = OpTypeStruct %bool %int %float %double",
"%outer_struct = OpTypeStruct %inner_struct %int %double",
"%flat_struct = OpTypeStruct %bool %int %float %double",
// clang-format on
};
Reduce number of test targets (#2024) This CL takes the various opt unit tests and makes a single executable instead of one per test. This reduces the number of build targets by ~125 when building with ninja.
2018-11-01 14:19:37 +00:00
using AggressiveEliminateDeadConstantTest =
PassTest<::testing::TestWithParam<AggressiveEliminateDeadConstantTestCase>>;
Adding additional functionality to ADCE. Modified ADCE to remove dead globals. * Entry point and execution mode instructions are marked as alive * Reachable functions and their parameters are marked as alive * Instruction deletion now deferred until the end of the pass * Eliminated dead insts set, added IsDead to calculate that value instead * Ported applicable dead variable elimination tests * Ported dead constant elim tests Added dead function elimination to ADCE * ported dead function elim tests Added handling of decoration groups in ADCE * Uses a custom sorter to traverse decorations in a specific order * Simplifies necessary checks Updated -O and -Os pass lists.
2017-12-18 17:13:10 +00:00
Reduce number of test targets (#2024) This CL takes the various opt unit tests and makes a single executable instead of one per test. This reduces the number of build targets by ~125 when building with ninja.
2018-11-01 14:19:37 +00:00
TEST_P(AggressiveEliminateDeadConstantTest, Custom) {
Adding additional functionality to ADCE. Modified ADCE to remove dead globals. * Entry point and execution mode instructions are marked as alive * Reachable functions and their parameters are marked as alive * Instruction deletion now deferred until the end of the pass * Eliminated dead insts set, added IsDead to calculate that value instead * Ported applicable dead variable elimination tests * Ported dead constant elim tests Added dead function elimination to ADCE * ported dead function elim tests Added handling of decoration groups in ADCE * Uses a custom sorter to traverse decorations in a specific order * Simplifies necessary checks Updated -O and -Os pass lists.
2017-12-18 17:13:10 +00:00
auto& tc = GetParam();
AssemblyBuilder builder;
builder.AppendTypesConstantsGlobals(CommonTypes)
.AppendTypesConstantsGlobals(tc.used_consts)
.AppendInMain(tc.main_insts);
const std::string expected = builder.GetCode();
builder.AppendTypesConstantsGlobals(tc.dead_consts);
builder.PrependPreamble(tc.checks);
const std::string assembly_with_dead_const = builder.GetCode();
// Do not enable validation. As the input code is invalid from the base
// tests (ported from other passes).
Move tests into namespaces (#1689) This CL moves the test into namespaces based on their directories.
2018-07-11 13:24:49 +00:00
SinglePassRunAndMatch<AggressiveDCEPass>(assembly_with_dead_const, false);
Adding additional functionality to ADCE. Modified ADCE to remove dead globals. * Entry point and execution mode instructions are marked as alive * Reachable functions and their parameters are marked as alive * Instruction deletion now deferred until the end of the pass * Eliminated dead insts set, added IsDead to calculate that value instead * Ported applicable dead variable elimination tests * Ported dead constant elim tests Added dead function elimination to ADCE * ported dead function elim tests Added handling of decoration groups in ADCE * Uses a custom sorter to traverse decorations in a specific order * Simplifies necessary checks Updated -O and -Os pass lists.
2017-12-18 17:13:10 +00:00
}
Remove use of deprecated googletest macro (#2286) * Remove use of deprecated googletest macro INSTANTIATE_TEST_CASE_P has been deprecated. We need to use INSTANTIATE_TEST_SUITE_P instead. * Remove extra commas from test suites.
2019-01-29 23:56:52 +00:00
INSTANTIATE_TEST_SUITE_P(
Reduce number of test targets (#2024) This CL takes the various opt unit tests and makes a single executable instead of one per test. This reduces the number of build targets by ~125 when building with ninja.
2018-11-01 14:19:37 +00:00
ScalarTypeConstants, AggressiveEliminateDeadConstantTest,
::testing::ValuesIn(std::vector<AggressiveEliminateDeadConstantTestCase>({
Adding additional functionality to ADCE. Modified ADCE to remove dead globals. * Entry point and execution mode instructions are marked as alive * Reachable functions and their parameters are marked as alive * Instruction deletion now deferred until the end of the pass * Eliminated dead insts set, added IsDead to calculate that value instead * Ported applicable dead variable elimination tests * Ported dead constant elim tests Added dead function elimination to ADCE * ported dead function elim tests Added handling of decoration groups in ADCE * Uses a custom sorter to traverse decorations in a specific order * Simplifies necessary checks Updated -O and -Os pass lists.
2017-12-18 17:13:10 +00:00
// clang-format off
// Scalar type constants, one dead constant and one used constant.
{
/* .used_consts = */
{
"%used_const_int = OpConstant %int 1",
},
/* .main_insts = */
{
"%int_var = OpVariable %_pf_int Output",
"OpStore %int_var %used_const_int",
},
/* .dead_consts = */
{
"%dead_const_int = OpConstant %int 1",
},
/* .checks = */
{
"; CHECK: [[const:%\\w+]] = OpConstant %int 1",
"; CHECK-NOT: OpConstant",
"; CHECK: OpStore {{%\\w+}} [[const]]",
},
},
{
/* .used_consts = */
{
"%used_const_uint = OpConstant %uint 1",
},
/* .main_insts = */
{
"%uint_var = OpVariable %_pf_uint Output",
"OpStore %uint_var %used_const_uint",
},
/* .dead_consts = */
{
"%dead_const_uint = OpConstant %uint 1",
},
/* .checks = */
{
"; CHECK: [[const:%\\w+]] = OpConstant %uint 1",
"; CHECK-NOT: OpConstant",
"; CHECK: OpStore {{%\\w+}} [[const]]",
},
},
{
/* .used_consts = */
{
hex_float: Use max_digits10 for the float precision CPPreference.com has this description of digits10: “The value of std::numeric_limits<T>::digits10 is the number of base-10 digits that can be represented by the type T without change, that is, any number with this many significant decimal digits can be converted to a value of type T and back to decimal form, without change due to rounding or overflow.” This means that any number with this many digits can be represented accurately in the corresponding type. A change in any digit in a number after that may or may not cause it a different bitwise representation. Therefore this isn’t necessarily enough precision to accurately represent the value in text. Instead we need max_digits10 which has the following description: “The value of std::numeric_limits<T>::max_digits10 is the number of base-10 digits that are necessary to uniquely represent all distinct values of the type T, such as necessary for serialization/deserialization to text.” The patch includes a test case in hex_float_test which tries to do a round-robin conversion of a number that requires more than 6 decimal places to be accurately represented. This would fail without the patch. Sadly this also breaks a bunch of other tests. Some of the tests in hex_float_test use ldexp and then compare it with a value which is not the same as the one returned by ldexp but instead is the value rounded to 6 decimals. Others use values that are not evenly representable as a binary floating fraction but then happened to generate the same value when rounded to 6 decimals. Where the actual value didn’t seem to matter these have been changed with different values that can be represented as a binary fraction.
2018-03-30 23:35:45 +00:00
"%used_const_float = OpConstant %float 3.1415",
Adding additional functionality to ADCE. Modified ADCE to remove dead globals. * Entry point and execution mode instructions are marked as alive * Reachable functions and their parameters are marked as alive * Instruction deletion now deferred until the end of the pass * Eliminated dead insts set, added IsDead to calculate that value instead * Ported applicable dead variable elimination tests * Ported dead constant elim tests Added dead function elimination to ADCE * ported dead function elim tests Added handling of decoration groups in ADCE * Uses a custom sorter to traverse decorations in a specific order * Simplifies necessary checks Updated -O and -Os pass lists.
2017-12-18 17:13:10 +00:00
},
/* .main_insts = */
{
"%float_var = OpVariable %_pf_float Output",
"OpStore %float_var %used_const_float",
},
/* .dead_consts = */
{
hex_float: Use max_digits10 for the float precision CPPreference.com has this description of digits10: “The value of std::numeric_limits<T>::digits10 is the number of base-10 digits that can be represented by the type T without change, that is, any number with this many significant decimal digits can be converted to a value of type T and back to decimal form, without change due to rounding or overflow.” This means that any number with this many digits can be represented accurately in the corresponding type. A change in any digit in a number after that may or may not cause it a different bitwise representation. Therefore this isn’t necessarily enough precision to accurately represent the value in text. Instead we need max_digits10 which has the following description: “The value of std::numeric_limits<T>::max_digits10 is the number of base-10 digits that are necessary to uniquely represent all distinct values of the type T, such as necessary for serialization/deserialization to text.” The patch includes a test case in hex_float_test which tries to do a round-robin conversion of a number that requires more than 6 decimal places to be accurately represented. This would fail without the patch. Sadly this also breaks a bunch of other tests. Some of the tests in hex_float_test use ldexp and then compare it with a value which is not the same as the one returned by ldexp but instead is the value rounded to 6 decimals. Others use values that are not evenly representable as a binary floating fraction but then happened to generate the same value when rounded to 6 decimals. Where the actual value didn’t seem to matter these have been changed with different values that can be represented as a binary fraction.
2018-03-30 23:35:45 +00:00
"%dead_const_float = OpConstant %float 3.1415",
Adding additional functionality to ADCE. Modified ADCE to remove dead globals. * Entry point and execution mode instructions are marked as alive * Reachable functions and their parameters are marked as alive * Instruction deletion now deferred until the end of the pass * Eliminated dead insts set, added IsDead to calculate that value instead * Ported applicable dead variable elimination tests * Ported dead constant elim tests Added dead function elimination to ADCE * ported dead function elim tests Added handling of decoration groups in ADCE * Uses a custom sorter to traverse decorations in a specific order * Simplifies necessary checks Updated -O and -Os pass lists.
2017-12-18 17:13:10 +00:00
},
/* .checks = */
{
hex_float: Use max_digits10 for the float precision CPPreference.com has this description of digits10: “The value of std::numeric_limits<T>::digits10 is the number of base-10 digits that can be represented by the type T without change, that is, any number with this many significant decimal digits can be converted to a value of type T and back to decimal form, without change due to rounding or overflow.” This means that any number with this many digits can be represented accurately in the corresponding type. A change in any digit in a number after that may or may not cause it a different bitwise representation. Therefore this isn’t necessarily enough precision to accurately represent the value in text. Instead we need max_digits10 which has the following description: “The value of std::numeric_limits<T>::max_digits10 is the number of base-10 digits that are necessary to uniquely represent all distinct values of the type T, such as necessary for serialization/deserialization to text.” The patch includes a test case in hex_float_test which tries to do a round-robin conversion of a number that requires more than 6 decimal places to be accurately represented. This would fail without the patch. Sadly this also breaks a bunch of other tests. Some of the tests in hex_float_test use ldexp and then compare it with a value which is not the same as the one returned by ldexp but instead is the value rounded to 6 decimals. Others use values that are not evenly representable as a binary floating fraction but then happened to generate the same value when rounded to 6 decimals. Where the actual value didn’t seem to matter these have been changed with different values that can be represented as a binary fraction.
2018-03-30 23:35:45 +00:00
"; CHECK: [[const:%\\w+]] = OpConstant %float 3.1415",
Adding additional functionality to ADCE. Modified ADCE to remove dead globals. * Entry point and execution mode instructions are marked as alive * Reachable functions and their parameters are marked as alive * Instruction deletion now deferred until the end of the pass * Eliminated dead insts set, added IsDead to calculate that value instead * Ported applicable dead variable elimination tests * Ported dead constant elim tests Added dead function elimination to ADCE * ported dead function elim tests Added handling of decoration groups in ADCE * Uses a custom sorter to traverse decorations in a specific order * Simplifies necessary checks Updated -O and -Os pass lists.
2017-12-18 17:13:10 +00:00
"; CHECK-NOT: OpConstant",
"; CHECK: OpStore {{%\\w+}} [[const]]",
},
},
{
/* .used_consts = */
{
"%used_const_double = OpConstant %double 3.14",
},
/* .main_insts = */
{
"%double_var = OpVariable %_pf_double Output",
"OpStore %double_var %used_const_double",
},
/* .dead_consts = */
{
"%dead_const_double = OpConstant %double 3.14",
},
/* .checks = */
{
"; CHECK: [[const:%\\w+]] = OpConstant %double 3.14",
"; CHECK-NOT: OpConstant",
"; CHECK: OpStore {{%\\w+}} [[const]]",
},
},
// clang-format on
})));
Remove use of deprecated googletest macro (#2286) * Remove use of deprecated googletest macro INSTANTIATE_TEST_CASE_P has been deprecated. We need to use INSTANTIATE_TEST_SUITE_P instead. * Remove extra commas from test suites.
2019-01-29 23:56:52 +00:00
INSTANTIATE_TEST_SUITE_P(
Reduce number of test targets (#2024) This CL takes the various opt unit tests and makes a single executable instead of one per test. This reduces the number of build targets by ~125 when building with ninja.
2018-11-01 14:19:37 +00:00
VectorTypeConstants, AggressiveEliminateDeadConstantTest,
::testing::ValuesIn(std::vector<AggressiveEliminateDeadConstantTestCase>({
Adding additional functionality to ADCE. Modified ADCE to remove dead globals. * Entry point and execution mode instructions are marked as alive * Reachable functions and their parameters are marked as alive * Instruction deletion now deferred until the end of the pass * Eliminated dead insts set, added IsDead to calculate that value instead * Ported applicable dead variable elimination tests * Ported dead constant elim tests Added dead function elimination to ADCE * ported dead function elim tests Added handling of decoration groups in ADCE * Uses a custom sorter to traverse decorations in a specific order * Simplifies necessary checks Updated -O and -Os pass lists.
2017-12-18 17:13:10 +00:00
// clang-format off
// Tests eliminating dead constant type ivec2. One dead constant vector
// and one used constant vector, each built from its own group of
// scalar constants.
{
/* .used_consts = */
{
"%used_int_x = OpConstant %int 1",
"%used_int_y = OpConstant %int 2",
"%used_v2int = OpConstantComposite %v2int %used_int_x %used_int_y",
},
/* .main_insts = */
{
"%v2int_var = OpVariable %_pf_v2int Output",
"OpStore %v2int_var %used_v2int",
},
/* .dead_consts = */
{
"%dead_int_x = OpConstant %int 1",
"%dead_int_y = OpConstant %int 2",
"%dead_v2int = OpConstantComposite %v2int %dead_int_x %dead_int_y",
},
/* .checks = */
{
"; CHECK: [[constx:%\\w+]] = OpConstant %int 1",
"; CHECK: [[consty:%\\w+]] = OpConstant %int 2",
"; CHECK: [[const:%\\w+]] = OpConstantComposite %v2int [[constx]] [[consty]]",
"; CHECK-NOT: OpConstant",
"; CHECK: OpStore {{%\\w+}} [[const]]",
},
},
// Tests eliminating dead constant ivec3. One dead constant vector and
// one used constant vector. But both built from a same group of
// scalar constants.
{
/* .used_consts = */
{
"%used_int_x = OpConstant %int 1",
"%used_int_y = OpConstant %int 2",
"%used_int_z = OpConstant %int 3",
"%used_v3int = OpConstantComposite %v3int %used_int_x %used_int_y %used_int_z",
},
/* .main_insts = */
{
"%v3int_var = OpVariable %_pf_v3int Output",
"OpStore %v3int_var %used_v3int",
},
/* .dead_consts = */
{
"%dead_v3int = OpConstantComposite %v3int %used_int_x %used_int_y %used_int_z",
},
/* .checks = */
{
"; CHECK: [[constx:%\\w+]] = OpConstant %int 1",
"; CHECK: [[consty:%\\w+]] = OpConstant %int 2",
"; CHECK: [[constz:%\\w+]] = OpConstant %int 3",
"; CHECK: [[const:%\\w+]] = OpConstantComposite %v3int [[constx]] [[consty]] [[constz]]",
"; CHECK-NOT: OpConstant",
"; CHECK: OpStore {{%\\w+}} [[const]]",
},
},
// Tests eliminating dead constant vec2. One dead constant vector and
// one used constant vector. Each built from its own group of scalar
// constants.
{
/* .used_consts = */
{
hex_float: Use max_digits10 for the float precision CPPreference.com has this description of digits10: “The value of std::numeric_limits<T>::digits10 is the number of base-10 digits that can be represented by the type T without change, that is, any number with this many significant decimal digits can be converted to a value of type T and back to decimal form, without change due to rounding or overflow.” This means that any number with this many digits can be represented accurately in the corresponding type. A change in any digit in a number after that may or may not cause it a different bitwise representation. Therefore this isn’t necessarily enough precision to accurately represent the value in text. Instead we need max_digits10 which has the following description: “The value of std::numeric_limits<T>::max_digits10 is the number of base-10 digits that are necessary to uniquely represent all distinct values of the type T, such as necessary for serialization/deserialization to text.” The patch includes a test case in hex_float_test which tries to do a round-robin conversion of a number that requires more than 6 decimal places to be accurately represented. This would fail without the patch. Sadly this also breaks a bunch of other tests. Some of the tests in hex_float_test use ldexp and then compare it with a value which is not the same as the one returned by ldexp but instead is the value rounded to 6 decimals. Others use values that are not evenly representable as a binary floating fraction but then happened to generate the same value when rounded to 6 decimals. Where the actual value didn’t seem to matter these have been changed with different values that can be represented as a binary fraction.
2018-03-30 23:35:45 +00:00
"%used_float_x = OpConstant %float 3.1415",
Adding additional functionality to ADCE. Modified ADCE to remove dead globals. * Entry point and execution mode instructions are marked as alive * Reachable functions and their parameters are marked as alive * Instruction deletion now deferred until the end of the pass * Eliminated dead insts set, added IsDead to calculate that value instead * Ported applicable dead variable elimination tests * Ported dead constant elim tests Added dead function elimination to ADCE * ported dead function elim tests Added handling of decoration groups in ADCE * Uses a custom sorter to traverse decorations in a specific order * Simplifies necessary checks Updated -O and -Os pass lists.
2017-12-18 17:13:10 +00:00
"%used_float_y = OpConstant %float 4.13",
"%used_v2float = OpConstantComposite %v2float %used_float_x %used_float_y",
},
/* .main_insts = */
{
"%v2float_var = OpVariable %_pf_v2float Output",
"OpStore %v2float_var %used_v2float",
},
/* .dead_consts = */
{
hex_float: Use max_digits10 for the float precision CPPreference.com has this description of digits10: “The value of std::numeric_limits<T>::digits10 is the number of base-10 digits that can be represented by the type T without change, that is, any number with this many significant decimal digits can be converted to a value of type T and back to decimal form, without change due to rounding or overflow.” This means that any number with this many digits can be represented accurately in the corresponding type. A change in any digit in a number after that may or may not cause it a different bitwise representation. Therefore this isn’t necessarily enough precision to accurately represent the value in text. Instead we need max_digits10 which has the following description: “The value of std::numeric_limits<T>::max_digits10 is the number of base-10 digits that are necessary to uniquely represent all distinct values of the type T, such as necessary for serialization/deserialization to text.” The patch includes a test case in hex_float_test which tries to do a round-robin conversion of a number that requires more than 6 decimal places to be accurately represented. This would fail without the patch. Sadly this also breaks a bunch of other tests. Some of the tests in hex_float_test use ldexp and then compare it with a value which is not the same as the one returned by ldexp but instead is the value rounded to 6 decimals. Others use values that are not evenly representable as a binary floating fraction but then happened to generate the same value when rounded to 6 decimals. Where the actual value didn’t seem to matter these have been changed with different values that can be represented as a binary fraction.
2018-03-30 23:35:45 +00:00
"%dead_float_x = OpConstant %float 3.1415",
Adding additional functionality to ADCE. Modified ADCE to remove dead globals. * Entry point and execution mode instructions are marked as alive * Reachable functions and their parameters are marked as alive * Instruction deletion now deferred until the end of the pass * Eliminated dead insts set, added IsDead to calculate that value instead * Ported applicable dead variable elimination tests * Ported dead constant elim tests Added dead function elimination to ADCE * ported dead function elim tests Added handling of decoration groups in ADCE * Uses a custom sorter to traverse decorations in a specific order * Simplifies necessary checks Updated -O and -Os pass lists.
2017-12-18 17:13:10 +00:00
"%dead_float_y = OpConstant %float 4.13",
"%dead_v2float = OpConstantComposite %v2float %dead_float_x %dead_float_y",
},
/* .checks = */
{
hex_float: Use max_digits10 for the float precision CPPreference.com has this description of digits10: “The value of std::numeric_limits<T>::digits10 is the number of base-10 digits that can be represented by the type T without change, that is, any number with this many significant decimal digits can be converted to a value of type T and back to decimal form, without change due to rounding or overflow.” This means that any number with this many digits can be represented accurately in the corresponding type. A change in any digit in a number after that may or may not cause it a different bitwise representation. Therefore this isn’t necessarily enough precision to accurately represent the value in text. Instead we need max_digits10 which has the following description: “The value of std::numeric_limits<T>::max_digits10 is the number of base-10 digits that are necessary to uniquely represent all distinct values of the type T, such as necessary for serialization/deserialization to text.” The patch includes a test case in hex_float_test which tries to do a round-robin conversion of a number that requires more than 6 decimal places to be accurately represented. This would fail without the patch. Sadly this also breaks a bunch of other tests. Some of the tests in hex_float_test use ldexp and then compare it with a value which is not the same as the one returned by ldexp but instead is the value rounded to 6 decimals. Others use values that are not evenly representable as a binary floating fraction but then happened to generate the same value when rounded to 6 decimals. Where the actual value didn’t seem to matter these have been changed with different values that can be represented as a binary fraction.
2018-03-30 23:35:45 +00:00
"; CHECK: [[constx:%\\w+]] = OpConstant %float 3.1415",
Adding additional functionality to ADCE. Modified ADCE to remove dead globals. * Entry point and execution mode instructions are marked as alive * Reachable functions and their parameters are marked as alive * Instruction deletion now deferred until the end of the pass * Eliminated dead insts set, added IsDead to calculate that value instead * Ported applicable dead variable elimination tests * Ported dead constant elim tests Added dead function elimination to ADCE * ported dead function elim tests Added handling of decoration groups in ADCE * Uses a custom sorter to traverse decorations in a specific order * Simplifies necessary checks Updated -O and -Os pass lists.
2017-12-18 17:13:10 +00:00
"; CHECK: [[consty:%\\w+]] = OpConstant %float 4.13",
"; CHECK: [[const:%\\w+]] = OpConstantComposite %v2float [[constx]] [[consty]]",
"; CHECK-NOT: OpConstant",
"; CHECK: OpStore {{%\\w+}} [[const]]",
},
},
// Tests eliminating dead constant vec3. One dead constant vector and
// one used constant vector. Both built from a same group of scalar
// constants.
{
/* .used_consts = */
{
hex_float: Use max_digits10 for the float precision CPPreference.com has this description of digits10: “The value of std::numeric_limits<T>::digits10 is the number of base-10 digits that can be represented by the type T without change, that is, any number with this many significant decimal digits can be converted to a value of type T and back to decimal form, without change due to rounding or overflow.” This means that any number with this many digits can be represented accurately in the corresponding type. A change in any digit in a number after that may or may not cause it a different bitwise representation. Therefore this isn’t necessarily enough precision to accurately represent the value in text. Instead we need max_digits10 which has the following description: “The value of std::numeric_limits<T>::max_digits10 is the number of base-10 digits that are necessary to uniquely represent all distinct values of the type T, such as necessary for serialization/deserialization to text.” The patch includes a test case in hex_float_test which tries to do a round-robin conversion of a number that requires more than 6 decimal places to be accurately represented. This would fail without the patch. Sadly this also breaks a bunch of other tests. Some of the tests in hex_float_test use ldexp and then compare it with a value which is not the same as the one returned by ldexp but instead is the value rounded to 6 decimals. Others use values that are not evenly representable as a binary floating fraction but then happened to generate the same value when rounded to 6 decimals. Where the actual value didn’t seem to matter these have been changed with different values that can be represented as a binary fraction.
2018-03-30 23:35:45 +00:00
"%used_float_x = OpConstant %float 3.1415",
"%used_float_y = OpConstant %float 4.25",
"%used_float_z = OpConstant %float 4.75",
Adding additional functionality to ADCE. Modified ADCE to remove dead globals. * Entry point and execution mode instructions are marked as alive * Reachable functions and their parameters are marked as alive * Instruction deletion now deferred until the end of the pass * Eliminated dead insts set, added IsDead to calculate that value instead * Ported applicable dead variable elimination tests * Ported dead constant elim tests Added dead function elimination to ADCE * ported dead function elim tests Added handling of decoration groups in ADCE * Uses a custom sorter to traverse decorations in a specific order * Simplifies necessary checks Updated -O and -Os pass lists.
2017-12-18 17:13:10 +00:00
"%used_v3float = OpConstantComposite %v3float %used_float_x %used_float_y %used_float_z",
},
/* .main_insts = */
{
"%v3float_var = OpVariable %_pf_v3float Output",
"OpStore %v3float_var %used_v3float",
},
/* .dead_consts = */
{
"%dead_v3float = OpConstantComposite %v3float %used_float_x %used_float_y %used_float_z",
},
/* .checks = */
{
hex_float: Use max_digits10 for the float precision CPPreference.com has this description of digits10: “The value of std::numeric_limits<T>::digits10 is the number of base-10 digits that can be represented by the type T without change, that is, any number with this many significant decimal digits can be converted to a value of type T and back to decimal form, without change due to rounding or overflow.” This means that any number with this many digits can be represented accurately in the corresponding type. A change in any digit in a number after that may or may not cause it a different bitwise representation. Therefore this isn’t necessarily enough precision to accurately represent the value in text. Instead we need max_digits10 which has the following description: “The value of std::numeric_limits<T>::max_digits10 is the number of base-10 digits that are necessary to uniquely represent all distinct values of the type T, such as necessary for serialization/deserialization to text.” The patch includes a test case in hex_float_test which tries to do a round-robin conversion of a number that requires more than 6 decimal places to be accurately represented. This would fail without the patch. Sadly this also breaks a bunch of other tests. Some of the tests in hex_float_test use ldexp and then compare it with a value which is not the same as the one returned by ldexp but instead is the value rounded to 6 decimals. Others use values that are not evenly representable as a binary floating fraction but then happened to generate the same value when rounded to 6 decimals. Where the actual value didn’t seem to matter these have been changed with different values that can be represented as a binary fraction.
2018-03-30 23:35:45 +00:00
"; CHECK: [[constx:%\\w+]] = OpConstant %float 3.1415",
"; CHECK: [[consty:%\\w+]] = OpConstant %float 4.25",
"; CHECK: [[constz:%\\w+]] = OpConstant %float 4.75",
Adding additional functionality to ADCE. Modified ADCE to remove dead globals. * Entry point and execution mode instructions are marked as alive * Reachable functions and their parameters are marked as alive * Instruction deletion now deferred until the end of the pass * Eliminated dead insts set, added IsDead to calculate that value instead * Ported applicable dead variable elimination tests * Ported dead constant elim tests Added dead function elimination to ADCE * ported dead function elim tests Added handling of decoration groups in ADCE * Uses a custom sorter to traverse decorations in a specific order * Simplifies necessary checks Updated -O and -Os pass lists.
2017-12-18 17:13:10 +00:00
"; CHECK: [[const:%\\w+]] = OpConstantComposite %v3float [[constx]] [[consty]]",
"; CHECK-NOT: OpConstant",
"; CHECK: OpStore {{%\\w+}} [[const]]",
},
},
// clang-format on
})));
Remove use of deprecated googletest macro (#2286) * Remove use of deprecated googletest macro INSTANTIATE_TEST_CASE_P has been deprecated. We need to use INSTANTIATE_TEST_SUITE_P instead. * Remove extra commas from test suites.
2019-01-29 23:56:52 +00:00
INSTANTIATE_TEST_SUITE_P(
Reduce number of test targets (#2024) This CL takes the various opt unit tests and makes a single executable instead of one per test. This reduces the number of build targets by ~125 when building with ninja.
2018-11-01 14:19:37 +00:00
StructTypeConstants, AggressiveEliminateDeadConstantTest,
::testing::ValuesIn(std::vector<AggressiveEliminateDeadConstantTestCase>({
Adding additional functionality to ADCE. Modified ADCE to remove dead globals. * Entry point and execution mode instructions are marked as alive * Reachable functions and their parameters are marked as alive * Instruction deletion now deferred until the end of the pass * Eliminated dead insts set, added IsDead to calculate that value instead * Ported applicable dead variable elimination tests * Ported dead constant elim tests Added dead function elimination to ADCE * ported dead function elim tests Added handling of decoration groups in ADCE * Uses a custom sorter to traverse decorations in a specific order * Simplifies necessary checks Updated -O and -Os pass lists.
2017-12-18 17:13:10 +00:00
// clang-format off
// A plain struct type dead constants. All of its components are dead
// constants too.
{
/* .used_consts = */ {},
/* .main_insts = */ {},
/* .dead_consts = */
{
"%dead_bool = OpConstantTrue %bool",
"%dead_int = OpConstant %int 1",
"%dead_float = OpConstant %float 2.5",
"%dead_double = OpConstant %double 3.14159265358979",
"%dead_struct = OpConstantComposite %flat_struct %dead_bool %dead_int %dead_float %dead_double",
},
/* .checks = */
{
"; CHECK-NOT: OpConstant",
},
},
// A plain struct type dead constants. Some of its components are dead
// constants while others are not.
{
/* .used_consts = */
{
"%used_int = OpConstant %int 1",
"%used_double = OpConstant %double 3.14159265358979",
},
/* .main_insts = */
{
"%int_var = OpVariable %_pf_int Output",
"OpStore %int_var %used_int",
"%double_var = OpVariable %_pf_double Output",
"OpStore %double_var %used_double",
},
/* .dead_consts = */
{
"%dead_bool = OpConstantTrue %bool",
"%dead_float = OpConstant %float 2.5",
"%dead_struct = OpConstantComposite %flat_struct %dead_bool %used_int %dead_float %used_double",
},
/* .checks = */
{
"; CHECK: [[int:%\\w+]] = OpConstant %int 1",
"; CHECK: [[double:%\\w+]] = OpConstant %double 3.14159265358979",
"; CHECK-NOT: OpConstant",
"; CHECK: OpStore {{%\\w+}} [[int]]",
"; CHECK: OpStore {{%\\w+}} [[double]]",
},
},
// A nesting struct type dead constants. All components of both outer
// and inner structs are dead and should be removed after dead constant
// elimination.
{
/* .used_consts = */ {},
/* .main_insts = */ {},
/* .dead_consts = */
{
"%dead_bool = OpConstantTrue %bool",
"%dead_int = OpConstant %int 1",
"%dead_float = OpConstant %float 2.5",
"%dead_double = OpConstant %double 3.1415926535",
"%dead_inner_struct = OpConstantComposite %inner_struct %dead_bool %dead_int %dead_float %dead_double",
"%dead_int2 = OpConstant %int 2",
"%dead_double2 = OpConstant %double 1.428571428514",
"%dead_outer_struct = OpConstantComposite %outer_struct %dead_inner_struct %dead_int2 %dead_double2",
},
/* .checks = */
{
"; CHECK-NOT: OpConstant",
},
},
// A nesting struct type dead constants. Some of its components are
// dead constants while others are not.
{
/* .used_consts = */
{
"%used_int = OpConstant %int 1",
"%used_double = OpConstant %double 3.14159265358979",
},
/* .main_insts = */
{
"%int_var = OpVariable %_pf_int Output",
"OpStore %int_var %used_int",
"%double_var = OpVariable %_pf_double Output",
"OpStore %double_var %used_double",
},
/* .dead_consts = */
{
"%dead_bool = OpConstantTrue %bool",
"%dead_float = OpConstant %float 2.5",
"%dead_inner_struct = OpConstantComposite %inner_struct %dead_bool %used_int %dead_float %used_double",
"%dead_int = OpConstant %int 2",
"%dead_outer_struct = OpConstantComposite %outer_struct %dead_inner_struct %dead_int %used_double",
},
/* .checks = */
{
"; CHECK: [[int:%\\w+]] = OpConstant %int 1",
"; CHECK: [[double:%\\w+]] = OpConstant %double 3.14159265358979",
"; CHECK-NOT: OpConstant",
"; CHECK: OpStore {{%\\w+}} [[int]]",
"; CHECK: OpStore {{%\\w+}} [[double]]",
},
},
// A nesting struct case. The inner struct is used while the outer struct is not
{
/* .used_const = */
{
"%used_bool = OpConstantTrue %bool",
"%used_int = OpConstant %int 1",
"%used_float = OpConstant %float 1.23",
"%used_double = OpConstant %double 1.2345678901234",
"%used_inner_struct = OpConstantComposite %inner_struct %used_bool %used_int %used_float %used_double",
},
/* .main_insts = */
{
"%bool_var = OpVariable %_pf_bool Output",
"%bool_from_inner_struct = OpCompositeExtract %bool %used_inner_struct 0",
"OpStore %bool_var %bool_from_inner_struct",
},
/* .dead_consts = */
{
"%dead_int = OpConstant %int 2",
"%dead_outer_struct = OpConstantComposite %outer_struct %used_inner_struct %dead_int %used_double"
},
/* .checks = */
{
"; CHECK: [[bool:%\\w+]] = OpConstantTrue",
"; CHECK: [[int:%\\w+]] = OpConstant %int 1",
"; CHECK: [[float:%\\w+]] = OpConstant %float 1.23",
"; CHECK: [[double:%\\w+]] = OpConstant %double 1.2345678901234",
"; CHECK: [[struct:%\\w+]] = OpConstantComposite %inner_struct [[bool]] [[int]] [[float]] [[double]]",
"; CHECK-NOT: OpConstant",
"; CHECK: OpCompositeExtract %bool [[struct]]",
}
},
// A nesting struct case. The outer struct is used, so the inner struct should not
// be removed even though it is not used anywhere.
{
/* .used_const = */
{
"%used_bool = OpConstantTrue %bool",
"%used_int = OpConstant %int 1",
"%used_float = OpConstant %float 1.23",
"%used_double = OpConstant %double 1.2345678901234",
"%used_inner_struct = OpConstantComposite %inner_struct %used_bool %used_int %used_float %used_double",
"%used_outer_struct = OpConstantComposite %outer_struct %used_inner_struct %used_int %used_double"
},
/* .main_insts = */
{
"%int_var = OpVariable %_pf_int Output",
"%int_from_outer_struct = OpCompositeExtract %int %used_outer_struct 1",
"OpStore %int_var %int_from_outer_struct",
},
/* .dead_consts = */ {},
/* .checks = */
{
"; CHECK: [[bool:%\\w+]] = OpConstantTrue %bool",
"; CHECK: [[int:%\\w+]] = OpConstant %int 1",
"; CHECK: [[float:%\\w+]] = OpConstant %float 1.23",
"; CHECK: [[double:%\\w+]] = OpConstant %double 1.2345678901234",
"; CHECK: [[inner_struct:%\\w+]] = OpConstantComposite %inner_struct %used_bool %used_int %used_float %used_double",
"; CHECK: [[outer_struct:%\\w+]] = OpConstantComposite %outer_struct %used_inner_struct %used_int %used_double",
"; CHECK: OpCompositeExtract %int [[outer_struct]]",
},
},
// clang-format on
})));
Remove use of deprecated googletest macro (#2286) * Remove use of deprecated googletest macro INSTANTIATE_TEST_CASE_P has been deprecated. We need to use INSTANTIATE_TEST_SUITE_P instead. * Remove extra commas from test suites.
2019-01-29 23:56:52 +00:00
INSTANTIATE_TEST_SUITE_P(
Reduce number of test targets (#2024) This CL takes the various opt unit tests and makes a single executable instead of one per test. This reduces the number of build targets by ~125 when building with ninja.
2018-11-01 14:19:37 +00:00
ScalarTypeSpecConstants, AggressiveEliminateDeadConstantTest,
::testing::ValuesIn(std::vector<AggressiveEliminateDeadConstantTestCase>({
Adding additional functionality to ADCE. Modified ADCE to remove dead globals. * Entry point and execution mode instructions are marked as alive * Reachable functions and their parameters are marked as alive * Instruction deletion now deferred until the end of the pass * Eliminated dead insts set, added IsDead to calculate that value instead * Ported applicable dead variable elimination tests * Ported dead constant elim tests Added dead function elimination to ADCE * ported dead function elim tests Added handling of decoration groups in ADCE * Uses a custom sorter to traverse decorations in a specific order * Simplifies necessary checks Updated -O and -Os pass lists.
2017-12-18 17:13:10 +00:00
// clang-format off
// All scalar type spec constants.
{
/* .used_consts = */
{
"%used_bool = OpSpecConstantTrue %bool",
"%used_uint = OpSpecConstant %uint 2",
"%used_int = OpSpecConstant %int 2",
"%used_float = OpSpecConstant %float 2.5",
"%used_double = OpSpecConstant %double 1.428571428514",
},
/* .main_insts = */
{
"%bool_var = OpVariable %_pf_bool Output",
"%uint_var = OpVariable %_pf_uint Output",
"%int_var = OpVariable %_pf_int Output",
"%float_var = OpVariable %_pf_float Output",
"%double_var = OpVariable %_pf_double Output",
"OpStore %bool_var %used_bool",
"OpStore %uint_var %used_uint",
"OpStore %int_var %used_int",
"OpStore %float_var %used_float",
"OpStore %double_var %used_double",
},
/* .dead_consts = */
{
"%dead_bool = OpSpecConstantTrue %bool",
"%dead_uint = OpSpecConstant %uint 2",
"%dead_int = OpSpecConstant %int 2",
"%dead_float = OpSpecConstant %float 2.5",
"%dead_double = OpSpecConstant %double 1.428571428514",
},
/* .checks = */
{
"; CHECK: [[bool:%\\w+]] = OpSpecConstantTrue %bool",
"; CHECK: [[uint:%\\w+]] = OpSpecConstant %uint 2",
"; CHECK: [[int:%\\w+]] = OpSpecConstant %int 2",
"; CHECK: [[float:%\\w+]] = OpSpecConstant %float 2.5",
"; CHECK: [[double:%\\w+]] = OpSpecConstant %double 1.428571428514",
"; CHECK-NOT: OpSpecConstant",
"; CHECK: OpStore {{%\\w+}} [[bool]]",
"; CHECK: OpStore {{%\\w+}} [[uint]]",
"; CHECK: OpStore {{%\\w+}} [[int]]",
"; CHECK: OpStore {{%\\w+}} [[float]]",
"; CHECK: OpStore {{%\\w+}} [[double]]",
},
},
// clang-format on
})));
Remove use of deprecated googletest macro (#2286) * Remove use of deprecated googletest macro INSTANTIATE_TEST_CASE_P has been deprecated. We need to use INSTANTIATE_TEST_SUITE_P instead. * Remove extra commas from test suites.
2019-01-29 23:56:52 +00:00
INSTANTIATE_TEST_SUITE_P(
Reduce number of test targets (#2024) This CL takes the various opt unit tests and makes a single executable instead of one per test. This reduces the number of build targets by ~125 when building with ninja.
2018-11-01 14:19:37 +00:00
VectorTypeSpecConstants, AggressiveEliminateDeadConstantTest,
::testing::ValuesIn(std::vector<AggressiveEliminateDeadConstantTestCase>({
Adding additional functionality to ADCE. Modified ADCE to remove dead globals. * Entry point and execution mode instructions are marked as alive * Reachable functions and their parameters are marked as alive * Instruction deletion now deferred until the end of the pass * Eliminated dead insts set, added IsDead to calculate that value instead * Ported applicable dead variable elimination tests * Ported dead constant elim tests Added dead function elimination to ADCE * ported dead function elim tests Added handling of decoration groups in ADCE * Uses a custom sorter to traverse decorations in a specific order * Simplifies necessary checks Updated -O and -Os pass lists.
2017-12-18 17:13:10 +00:00
// clang-format off
// Bool vector type spec constants. One vector has all component dead,
// another vector has one dead boolean and one used boolean.
{
/* .used_consts = */
{
"%used_bool = OpSpecConstantTrue %bool",
},
/* .main_insts = */
{
"%bool_var = OpVariable %_pf_bool Output",
"OpStore %bool_var %used_bool",
},
/* .dead_consts = */
{
"%dead_bool = OpSpecConstantFalse %bool",
"%dead_bool_vec1 = OpSpecConstantComposite %v2bool %dead_bool %dead_bool",
"%dead_bool_vec2 = OpSpecConstantComposite %v2bool %dead_bool %used_bool",
},
/* .checks = */
{
"; CHECK: [[bool:%\\w+]] = OpSpecConstantTrue %bool",
"; CHECK-NOT: OpSpecConstant",
"; CHECK: OpStore {{%\\w+}} [[bool]]",
},
},
// Uint vector type spec constants. One vector has all component dead,
Fix various source comment (doxygen) typos (#4680) Found via `codespell -q 3 -L fo,lod,parm
2022-01-26 20:13:08 +00:00
// another vector has one dead unsigned integer and one used unsigned
Adding additional functionality to ADCE. Modified ADCE to remove dead globals. * Entry point and execution mode instructions are marked as alive * Reachable functions and their parameters are marked as alive * Instruction deletion now deferred until the end of the pass * Eliminated dead insts set, added IsDead to calculate that value instead * Ported applicable dead variable elimination tests * Ported dead constant elim tests Added dead function elimination to ADCE * ported dead function elim tests Added handling of decoration groups in ADCE * Uses a custom sorter to traverse decorations in a specific order * Simplifies necessary checks Updated -O and -Os pass lists.
2017-12-18 17:13:10 +00:00
// integer.
{
/* .used_consts = */
{
"%used_uint = OpSpecConstant %uint 3",
},
/* .main_insts = */
{
"%uint_var = OpVariable %_pf_uint Output",
"OpStore %uint_var %used_uint",
},
/* .dead_consts = */
{
"%dead_uint = OpSpecConstant %uint 1",
"%dead_uint_vec1 = OpSpecConstantComposite %v2uint %dead_uint %dead_uint",
"%dead_uint_vec2 = OpSpecConstantComposite %v2uint %dead_uint %used_uint",
},
/* .checks = */
{
"; CHECK: [[uint:%\\w+]] = OpSpecConstant %uint 3",
"; CHECK-NOT: OpSpecConstant",
"; CHECK: OpStore {{%\\w+}} [[uint]]",
},
},
// Int vector type spec constants. One vector has all component dead,
// another vector has one dead integer and one used integer.
{
/* .used_consts = */
{
"%used_int = OpSpecConstant %int 3",
},
/* .main_insts = */
{
"%int_var = OpVariable %_pf_int Output",
"OpStore %int_var %used_int",
},
/* .dead_consts = */
{
"%dead_int = OpSpecConstant %int 1",
"%dead_int_vec1 = OpSpecConstantComposite %v2int %dead_int %dead_int",
"%dead_int_vec2 = OpSpecConstantComposite %v2int %dead_int %used_int",
},
/* .checks = */
{
"; CHECK: [[int:%\\w+]] = OpSpecConstant %int 3",
"; CHECK-NOT: OpSpecConstant",
"; CHECK: OpStore {{%\\w+}} [[int]]",
},
},
// Int vector type spec constants built with both spec constants and
// front-end constants.
{
/* .used_consts = */
{
"%used_spec_int = OpSpecConstant %int 3",
"%used_front_end_int = OpConstant %int 3",
},
/* .main_insts = */
{
"%int_var1 = OpVariable %_pf_int Output",
"OpStore %int_var1 %used_spec_int",
"%int_var2 = OpVariable %_pf_int Output",
"OpStore %int_var2 %used_front_end_int",
},
/* .dead_consts = */
{
"%dead_spec_int = OpSpecConstant %int 1",
"%dead_front_end_int = OpConstant %int 1",
// Dead front-end and dead spec constants
"%dead_int_vec1 = OpSpecConstantComposite %v2int %dead_spec_int %dead_front_end_int",
// Used front-end and dead spec constants
"%dead_int_vec2 = OpSpecConstantComposite %v2int %dead_spec_int %used_front_end_int",
// Dead front-end and used spec constants
"%dead_int_vec3 = OpSpecConstantComposite %v2int %dead_front_end_int %used_spec_int",
},
/* .checks = */
{
"; CHECK: [[int1:%\\w+]] = OpSpecConstant %int 3",
"; CHECK: [[int2:%\\w+]] = OpConstant %int 3",
"; CHECK-NOT: OpSpecConstant",
"; CHECK-NOT: OpConstant",
"; CHECK: OpStore {{%\\w+}} [[int1]]",
"; CHECK: OpStore {{%\\w+}} [[int2]]",
},
},
// clang-format on
})));
Remove use of deprecated googletest macro (#2286) * Remove use of deprecated googletest macro INSTANTIATE_TEST_CASE_P has been deprecated. We need to use INSTANTIATE_TEST_SUITE_P instead. * Remove extra commas from test suites.
2019-01-29 23:56:52 +00:00
INSTANTIATE_TEST_SUITE_P(
Reduce number of test targets (#2024) This CL takes the various opt unit tests and makes a single executable instead of one per test. This reduces the number of build targets by ~125 when building with ninja.
2018-11-01 14:19:37 +00:00
SpecConstantOp, AggressiveEliminateDeadConstantTest,
::testing::ValuesIn(std::vector<AggressiveEliminateDeadConstantTestCase>({
Adding additional functionality to ADCE. Modified ADCE to remove dead globals. * Entry point and execution mode instructions are marked as alive * Reachable functions and their parameters are marked as alive * Instruction deletion now deferred until the end of the pass * Eliminated dead insts set, added IsDead to calculate that value instead * Ported applicable dead variable elimination tests * Ported dead constant elim tests Added dead function elimination to ADCE * ported dead function elim tests Added handling of decoration groups in ADCE * Uses a custom sorter to traverse decorations in a specific order * Simplifies necessary checks Updated -O and -Os pass lists.
2017-12-18 17:13:10 +00:00
// clang-format off
// Cast operations: uint <-> int <-> bool
{
/* .used_consts = */ {},
/* .main_insts = */ {},
/* .dead_consts = */
{
// Assistant constants, only used in dead spec constant
// operations.
"%signed_zero = OpConstant %int 0",
"%signed_zero_vec = OpConstantComposite %v2int %signed_zero %signed_zero",
"%unsigned_zero = OpConstant %uint 0",
"%unsigned_zero_vec = OpConstantComposite %v2uint %unsigned_zero %unsigned_zero",
"%signed_one = OpConstant %int 1",
"%signed_one_vec = OpConstantComposite %v2int %signed_one %signed_one",
"%unsigned_one = OpConstant %uint 1",
"%unsigned_one_vec = OpConstantComposite %v2uint %unsigned_one %unsigned_one",
// Spec constants that support casting to each other.
"%dead_bool = OpSpecConstantTrue %bool",
"%dead_uint = OpSpecConstant %uint 1",
"%dead_int = OpSpecConstant %int 2",
"%dead_bool_vec = OpSpecConstantComposite %v2bool %dead_bool %dead_bool",
"%dead_uint_vec = OpSpecConstantComposite %v2uint %dead_uint %dead_uint",
"%dead_int_vec = OpSpecConstantComposite %v2int %dead_int %dead_int",
// Scalar cast to boolean spec constant.
"%int_to_bool = OpSpecConstantOp %bool INotEqual %dead_int %signed_zero",
"%uint_to_bool = OpSpecConstantOp %bool INotEqual %dead_uint %unsigned_zero",
// Vector cast to boolean spec constant.
"%int_to_bool_vec = OpSpecConstantOp %v2bool INotEqual %dead_int_vec %signed_zero_vec",
"%uint_to_bool_vec = OpSpecConstantOp %v2bool INotEqual %dead_uint_vec %unsigned_zero_vec",
// Scalar cast to int spec constant.
"%bool_to_int = OpSpecConstantOp %int Select %dead_bool %signed_one %signed_zero",
"%uint_to_int = OpSpecConstantOp %uint IAdd %dead_uint %unsigned_zero",
// Vector cast to int spec constant.
"%bool_to_int_vec = OpSpecConstantOp %v2int Select %dead_bool_vec %signed_one_vec %signed_zero_vec",
"%uint_to_int_vec = OpSpecConstantOp %v2uint IAdd %dead_uint_vec %unsigned_zero_vec",
// Scalar cast to uint spec constant.
"%bool_to_uint = OpSpecConstantOp %uint Select %dead_bool %unsigned_one %unsigned_zero",
"%int_to_uint_vec = OpSpecConstantOp %uint IAdd %dead_int %signed_zero",
// Vector cast to uint spec constant.
"%bool_to_uint_vec = OpSpecConstantOp %v2uint Select %dead_bool_vec %unsigned_one_vec %unsigned_zero_vec",
"%int_to_uint = OpSpecConstantOp %v2uint IAdd %dead_int_vec %signed_zero_vec",
},
/* .checks = */
{
"; CHECK-NOT: OpConstant",
"; CHECK-NOT: OpSpecConstant",
},
},
// Add, sub, mul, div, rem.
{
/* .used_consts = */ {},
/* .main_insts = */ {},
/* .dead_consts = */
{
"%dead_spec_int_a = OpSpecConstant %int 1",
"%dead_spec_int_a_vec = OpSpecConstantComposite %v2int %dead_spec_int_a %dead_spec_int_a",
"%dead_spec_int_b = OpSpecConstant %int 2",
"%dead_spec_int_b_vec = OpSpecConstantComposite %v2int %dead_spec_int_b %dead_spec_int_b",
"%dead_const_int_c = OpConstant %int 3",
"%dead_const_int_c_vec = OpConstantComposite %v2int %dead_const_int_c %dead_const_int_c",
// Add
"%add_a_b = OpSpecConstantOp %int IAdd %dead_spec_int_a %dead_spec_int_b",
"%add_a_b_vec = OpSpecConstantOp %v2int IAdd %dead_spec_int_a_vec %dead_spec_int_b_vec",
// Sub
"%sub_a_b = OpSpecConstantOp %int ISub %dead_spec_int_a %dead_spec_int_b",
"%sub_a_b_vec = OpSpecConstantOp %v2int ISub %dead_spec_int_a_vec %dead_spec_int_b_vec",
// Mul
"%mul_a_b = OpSpecConstantOp %int IMul %dead_spec_int_a %dead_spec_int_b",
"%mul_a_b_vec = OpSpecConstantOp %v2int IMul %dead_spec_int_a_vec %dead_spec_int_b_vec",
// Div
"%div_a_b = OpSpecConstantOp %int SDiv %dead_spec_int_a %dead_spec_int_b",
"%div_a_b_vec = OpSpecConstantOp %v2int SDiv %dead_spec_int_a_vec %dead_spec_int_b_vec",
// Bitwise Xor
"%xor_a_b = OpSpecConstantOp %int BitwiseXor %dead_spec_int_a %dead_spec_int_b",
"%xor_a_b_vec = OpSpecConstantOp %v2int BitwiseXor %dead_spec_int_a_vec %dead_spec_int_b_vec",
// Scalar Comparison
"%less_a_b = OpSpecConstantOp %bool SLessThan %dead_spec_int_a %dead_spec_int_b",
},
/* .checks = */
{
"; CHECK-NOT: OpConstant",
"; CHECK-NOT: OpSpecConstant",
},
},
// Vectors without used swizzles should be removed.
{
/* .used_consts = */
{
"%used_int = OpConstant %int 3",
},
/* .main_insts = */
{
"%int_var = OpVariable %_pf_int Output",
"OpStore %int_var %used_int",
},
/* .dead_consts = */
{
"%dead_int = OpConstant %int 3",
"%dead_spec_int_a = OpSpecConstant %int 1",
"%vec_a = OpSpecConstantComposite %v4int %dead_spec_int_a %dead_spec_int_a %dead_int %dead_int",
"%dead_spec_int_b = OpSpecConstant %int 2",
"%vec_b = OpSpecConstantComposite %v4int %dead_spec_int_b %dead_spec_int_b %used_int %used_int",
// Extract scalar
"%a_x = OpSpecConstantOp %int CompositeExtract %vec_a 0",
"%b_x = OpSpecConstantOp %int CompositeExtract %vec_b 0",
// Extract vector
"%a_xy = OpSpecConstantOp %v2int VectorShuffle %vec_a %vec_a 0 1",
"%b_xy = OpSpecConstantOp %v2int VectorShuffle %vec_b %vec_b 0 1",
},
/* .checks = */
{
"; CHECK: [[int:%\\w+]] = OpConstant %int 3",
"; CHECK-NOT: OpConstant",
"; CHECK-NOT: OpSpecConstant",
"; CHECK: OpStore {{%\\w+}} [[int]]",
},
},
// Vectors with used swizzles should not be removed.
{
/* .used_consts = */
{
"%used_int = OpConstant %int 3",
"%used_spec_int_a = OpSpecConstant %int 1",
"%used_spec_int_b = OpSpecConstant %int 2",
// Create vectors
"%vec_a = OpSpecConstantComposite %v4int %used_spec_int_a %used_spec_int_a %used_int %used_int",
"%vec_b = OpSpecConstantComposite %v4int %used_spec_int_b %used_spec_int_b %used_int %used_int",
// Extract vector
"%a_xy = OpSpecConstantOp %v2int VectorShuffle %vec_a %vec_a 0 1",
"%b_xy = OpSpecConstantOp %v2int VectorShuffle %vec_b %vec_b 0 1",
},
/* .main_insts = */
{
"%v2int_var_a = OpVariable %_pf_v2int Output",
"%v2int_var_b = OpVariable %_pf_v2int Output",
"OpStore %v2int_var_a %a_xy",
"OpStore %v2int_var_b %b_xy",
},
/* .dead_consts = */ {},
/* .checks = */
{
"; CHECK: [[int:%\\w+]] = OpConstant %int 3",
"; CHECK: [[a:%\\w+]] = OpSpecConstant %int 1",
"; CHECK: [[b:%\\w+]] = OpSpecConstant %int 2",
"; CHECK: [[veca:%\\w+]] = OpSpecConstantComposite %v4int [[a]] [[a]] [[int]] [[int]]",
"; CHECK: [[vecb:%\\w+]] = OpSpecConstantComposite %v4int [[b]] [[b]] [[int]] [[int]]",
"; CHECK: [[exa:%\\w+]] = OpSpecConstantOp %v2int VectorShuffle [[veca]] [[veca]] 0 1",
"; CHECK: [[exb:%\\w+]] = OpSpecConstantOp %v2int VectorShuffle [[vecb]] [[vecb]] 0 1",
"; CHECK-NOT: OpConstant",
"; CHECK-NOT: OpSpecConstant",
"; CHECK: OpStore {{%\\w+}} [[exa]]",
"; CHECK: OpStore {{%\\w+}} [[exb]]",
},
},
// clang-format on
})));
Remove use of deprecated googletest macro (#2286) * Remove use of deprecated googletest macro INSTANTIATE_TEST_CASE_P has been deprecated. We need to use INSTANTIATE_TEST_SUITE_P instead. * Remove extra commas from test suites.
2019-01-29 23:56:52 +00:00
INSTANTIATE_TEST_SUITE_P(
Reduce number of test targets (#2024) This CL takes the various opt unit tests and makes a single executable instead of one per test. This reduces the number of build targets by ~125 when building with ninja.
2018-11-01 14:19:37 +00:00
LongDefUseChain, AggressiveEliminateDeadConstantTest,
::testing::ValuesIn(std::vector<AggressiveEliminateDeadConstantTestCase>({
Adding additional functionality to ADCE. Modified ADCE to remove dead globals. * Entry point and execution mode instructions are marked as alive * Reachable functions and their parameters are marked as alive * Instruction deletion now deferred until the end of the pass * Eliminated dead insts set, added IsDead to calculate that value instead * Ported applicable dead variable elimination tests * Ported dead constant elim tests Added dead function elimination to ADCE * ported dead function elim tests Added handling of decoration groups in ADCE * Uses a custom sorter to traverse decorations in a specific order * Simplifies necessary checks Updated -O and -Os pass lists.
2017-12-18 17:13:10 +00:00
// clang-format off
// Long Def-Use chain with binary operations.
{
/* .used_consts = */
{
"%array_size = OpConstant %int 4",
"%type_arr_int_4 = OpTypeArray %int %array_size",
"%used_int_0 = OpConstant %int 100",
"%used_int_1 = OpConstant %int 1",
"%used_int_2 = OpSpecConstantOp %int IAdd %used_int_0 %used_int_1",
"%used_int_3 = OpSpecConstantOp %int ISub %used_int_0 %used_int_2",
"%used_int_4 = OpSpecConstantOp %int IAdd %used_int_0 %used_int_3",
"%used_int_5 = OpSpecConstantOp %int ISub %used_int_0 %used_int_4",
"%used_int_6 = OpSpecConstantOp %int IAdd %used_int_0 %used_int_5",
"%used_int_7 = OpSpecConstantOp %int ISub %used_int_0 %used_int_6",
"%used_int_8 = OpSpecConstantOp %int IAdd %used_int_0 %used_int_7",
"%used_int_9 = OpSpecConstantOp %int ISub %used_int_0 %used_int_8",
"%used_int_10 = OpSpecConstantOp %int IAdd %used_int_0 %used_int_9",
"%used_int_11 = OpSpecConstantOp %int ISub %used_int_0 %used_int_10",
"%used_int_12 = OpSpecConstantOp %int IAdd %used_int_0 %used_int_11",
"%used_int_13 = OpSpecConstantOp %int ISub %used_int_0 %used_int_12",
"%used_int_14 = OpSpecConstantOp %int IAdd %used_int_0 %used_int_13",
"%used_int_15 = OpSpecConstantOp %int ISub %used_int_0 %used_int_14",
"%used_int_16 = OpSpecConstantOp %int ISub %used_int_0 %used_int_15",
"%used_int_17 = OpSpecConstantOp %int IAdd %used_int_0 %used_int_16",
"%used_int_18 = OpSpecConstantOp %int ISub %used_int_0 %used_int_17",
"%used_int_19 = OpSpecConstantOp %int IAdd %used_int_0 %used_int_18",
"%used_int_20 = OpSpecConstantOp %int ISub %used_int_0 %used_int_19",
"%used_vec_a = OpSpecConstantComposite %v2int %used_int_18 %used_int_19",
"%used_vec_b = OpSpecConstantOp %v2int IMul %used_vec_a %used_vec_a",
"%used_int_21 = OpSpecConstantOp %int CompositeExtract %used_vec_b 0",
"%used_array = OpConstantComposite %type_arr_int_4 %used_int_20 %used_int_20 %used_int_21 %used_int_21",
},
/* .main_insts = */
{
"%int_var = OpVariable %_pf_int Output",
"%used_array_2 = OpCompositeExtract %int %used_array 2",
"OpStore %int_var %used_array_2",
},
/* .dead_consts = */
{
"%dead_int_1 = OpConstant %int 2",
"%dead_int_2 = OpSpecConstantOp %int IAdd %used_int_0 %dead_int_1",
"%dead_int_3 = OpSpecConstantOp %int ISub %used_int_0 %dead_int_2",
"%dead_int_4 = OpSpecConstantOp %int IAdd %used_int_0 %dead_int_3",
"%dead_int_5 = OpSpecConstantOp %int ISub %used_int_0 %dead_int_4",
"%dead_int_6 = OpSpecConstantOp %int IAdd %used_int_0 %dead_int_5",
"%dead_int_7 = OpSpecConstantOp %int ISub %used_int_0 %dead_int_6",
"%dead_int_8 = OpSpecConstantOp %int IAdd %used_int_0 %dead_int_7",
"%dead_int_9 = OpSpecConstantOp %int ISub %used_int_0 %dead_int_8",
"%dead_int_10 = OpSpecConstantOp %int IAdd %used_int_0 %dead_int_9",
"%dead_int_11 = OpSpecConstantOp %int ISub %used_int_0 %dead_int_10",
"%dead_int_12 = OpSpecConstantOp %int IAdd %used_int_0 %dead_int_11",
"%dead_int_13 = OpSpecConstantOp %int ISub %used_int_0 %dead_int_12",
"%dead_int_14 = OpSpecConstantOp %int IAdd %used_int_0 %dead_int_13",
"%dead_int_15 = OpSpecConstantOp %int ISub %used_int_0 %dead_int_14",
"%dead_int_16 = OpSpecConstantOp %int ISub %used_int_0 %dead_int_15",
"%dead_int_17 = OpSpecConstantOp %int IAdd %used_int_0 %dead_int_16",
"%dead_int_18 = OpSpecConstantOp %int ISub %used_int_0 %dead_int_17",
"%dead_int_19 = OpSpecConstantOp %int IAdd %used_int_0 %dead_int_18",
"%dead_int_20 = OpSpecConstantOp %int ISub %used_int_0 %dead_int_19",
"%dead_vec_a = OpSpecConstantComposite %v2int %dead_int_18 %dead_int_19",
"%dead_vec_b = OpSpecConstantOp %v2int IMul %dead_vec_a %dead_vec_a",
"%dead_int_21 = OpSpecConstantOp %int CompositeExtract %dead_vec_b 0",
"%dead_array = OpConstantComposite %type_arr_int_4 %dead_int_20 %used_int_20 %dead_int_19 %used_int_19",
},
/* .checks = */
{
"; CHECK: OpConstant %int 4",
"; CHECK: [[array:%\\w+]] = OpConstantComposite %type_arr_int_4 %used_int_20 %used_int_20 %used_int_21 %used_int_21",
"; CHECK-NOT: OpConstant",
"; CHECK-NOT: OpSpecConstant",
"; CHECK: OpStore {{%\\w+}} [[array]]",
},
},
// Long Def-Use chain with swizzle
// clang-format on
})));
TEST_F(AggressiveDCETest, DeadDecorationGroup) {
// The decoration group should be eliminated because the target of group
// decorate is dead.
const std::string text = R"(
; CHECK-NOT: OpDecorat
; CHECK-NOT: OpGroupDecorate
OpCapability Shader
OpMemoryModel Logical GLSL450
OpEntryPoint Fragment %main "main"
OpExecutionMode %main OriginUpperLeft
OpDecorate %1 Restrict
OpDecorate %1 Aliased
%1 = OpDecorationGroup
OpGroupDecorate %1 %var
%void = OpTypeVoid
%func = OpTypeFunction %void
%uint = OpTypeInt 32 0
%uint_ptr = OpTypePointer Function %uint
%main = OpFunction %void None %func
%2 = OpLabel
%var = OpVariable %uint_ptr Function
OpReturn
OpFunctionEnd
)";
Move tests into namespaces (#1689) This CL moves the test into namespaces based on their directories.
2018-07-11 13:24:49 +00:00
SinglePassRunAndMatch<AggressiveDCEPass>(text, true);
Adding additional functionality to ADCE. Modified ADCE to remove dead globals. * Entry point and execution mode instructions are marked as alive * Reachable functions and their parameters are marked as alive * Instruction deletion now deferred until the end of the pass * Eliminated dead insts set, added IsDead to calculate that value instead * Ported applicable dead variable elimination tests * Ported dead constant elim tests Added dead function elimination to ADCE * ported dead function elim tests Added handling of decoration groups in ADCE * Uses a custom sorter to traverse decorations in a specific order * Simplifies necessary checks Updated -O and -Os pass lists.
2017-12-18 17:13:10 +00:00
}
Invalidate the decoration manager at the start of ADCE. (#2189) * Invalidate the decoration manager at the start of ADCE. If the decoration manager is kept live the the contex will try to keep it up to date. ADCE deals with group decorations by changing the operands in |OpGroupDecorate| instructions directly without informing the decoration manager. This puts it in an invalid state, which will cause an error when the context tries to update it. To Avoid this problem, we will invalidate the decoration manager upfront. At the same time, the decoration manager is now considered when checking the consistency of the decoration manager.
2018-12-10 18:24:33 +00:00
TEST_F(AggressiveDCETest, DeadDecorationGroupAndValidDecorationMgr) {
// The decoration group should be eliminated because the target of group
// decorate is dead.
const std::string text = R"(
OpCapability Shader
OpMemoryModel Logical GLSL450
OpEntryPoint Fragment %main "main"
OpExecutionMode %main OriginUpperLeft
OpDecorate %1 Restrict
OpDecorate %1 Aliased
%1 = OpDecorationGroup
OpGroupDecorate %1 %var
%void = OpTypeVoid
%func = OpTypeFunction %void
%uint = OpTypeInt 32 0
%uint_ptr = OpTypePointer Function %uint
%main = OpFunction %void None %func
%2 = OpLabel
%var = OpVariable %uint_ptr Function
OpReturn
OpFunctionEnd
)";
auto pass = MakeUnique<AggressiveDCEPass>();
auto consumer = [](spv_message_level_t, const char*, const spv_position_t&,
const char* message) {
std::cerr << message << std::endl;
};
auto context = BuildModule(SPV_ENV_UNIVERSAL_1_1, consumer, text);
// Build the decoration manager before the pass.
context->get_decoration_mgr();
const auto status = pass->Run(context.get());
EXPECT_EQ(status, Pass::Status::SuccessWithChange);
}
Adding additional functionality to ADCE. Modified ADCE to remove dead globals. * Entry point and execution mode instructions are marked as alive * Reachable functions and their parameters are marked as alive * Instruction deletion now deferred until the end of the pass * Eliminated dead insts set, added IsDead to calculate that value instead * Ported applicable dead variable elimination tests * Ported dead constant elim tests Added dead function elimination to ADCE * ported dead function elim tests Added handling of decoration groups in ADCE * Uses a custom sorter to traverse decorations in a specific order * Simplifies necessary checks Updated -O and -Os pass lists.
2017-12-18 17:13:10 +00:00
TEST_F(AggressiveDCETest, ParitallyDeadDecorationGroup) {
const std::string text = R"(
; CHECK: OpDecorate [[grp:%\w+]] Restrict
; CHECK: [[grp]] = OpDecorationGroup
; CHECK: OpGroupDecorate [[grp]] [[output:%\w+]]
; CHECK: [[output]] = OpVariable {{%\w+}} Output
; CHECK-NOT: OpVariable {{%\w+}} Function
OpCapability Shader
OpMemoryModel Logical GLSL450
OpEntryPoint Fragment %main "main" %output
OpExecutionMode %main OriginUpperLeft
OpDecorate %1 Restrict
%1 = OpDecorationGroup
OpGroupDecorate %1 %var %output
%void = OpTypeVoid
%func = OpTypeFunction %void
%uint = OpTypeInt 32 0
%uint_ptr_Function = OpTypePointer Function %uint
%uint_ptr_Output = OpTypePointer Output %uint
%uint_0 = OpConstant %uint 0
%output = OpVariable %uint_ptr_Output Output
%main = OpFunction %void None %func
%2 = OpLabel
%var = OpVariable %uint_ptr_Function Function
OpStore %output %uint_0
OpReturn
OpFunctionEnd
)";
Move tests into namespaces (#1689) This CL moves the test into namespaces based on their directories.
2018-07-11 13:24:49 +00:00
SinglePassRunAndMatch<AggressiveDCEPass>(text, true);
Adding additional functionality to ADCE. Modified ADCE to remove dead globals. * Entry point and execution mode instructions are marked as alive * Reachable functions and their parameters are marked as alive * Instruction deletion now deferred until the end of the pass * Eliminated dead insts set, added IsDead to calculate that value instead * Ported applicable dead variable elimination tests * Ported dead constant elim tests Added dead function elimination to ADCE * ported dead function elim tests Added handling of decoration groups in ADCE * Uses a custom sorter to traverse decorations in a specific order * Simplifies necessary checks Updated -O and -Os pass lists.
2017-12-18 17:13:10 +00:00
}
TEST_F(AggressiveDCETest, ParitallyDeadDecorationGroupDifferentGroupDecorate) {
const std::string text = R"(
; CHECK: OpDecorate [[grp:%\w+]] Restrict
; CHECK: [[grp]] = OpDecorationGroup
; CHECK: OpGroupDecorate [[grp]] [[output:%\w+]]
; CHECK-NOT: OpGroupDecorate
; CHECK: [[output]] = OpVariable {{%\w+}} Output
; CHECK-NOT: OpVariable {{%\w+}} Function
OpCapability Shader
OpMemoryModel Logical GLSL450
OpEntryPoint Fragment %main "main" %output
OpExecutionMode %main OriginUpperLeft
OpDecorate %1 Restrict
%1 = OpDecorationGroup
OpGroupDecorate %1 %output
OpGroupDecorate %1 %var
%void = OpTypeVoid
%func = OpTypeFunction %void
%uint = OpTypeInt 32 0
%uint_ptr_Function = OpTypePointer Function %uint
%uint_ptr_Output = OpTypePointer Output %uint
%uint_0 = OpConstant %uint 0
%output = OpVariable %uint_ptr_Output Output
%main = OpFunction %void None %func
%2 = OpLabel
%var = OpVariable %uint_ptr_Function Function
OpStore %output %uint_0
OpReturn
OpFunctionEnd
)";
Move tests into namespaces (#1689) This CL moves the test into namespaces based on their directories.
2018-07-11 13:24:49 +00:00
SinglePassRunAndMatch<AggressiveDCEPass>(text, true);
Adding additional functionality to ADCE. Modified ADCE to remove dead globals. * Entry point and execution mode instructions are marked as alive * Reachable functions and their parameters are marked as alive * Instruction deletion now deferred until the end of the pass * Eliminated dead insts set, added IsDead to calculate that value instead * Ported applicable dead variable elimination tests * Ported dead constant elim tests Added dead function elimination to ADCE * ported dead function elim tests Added handling of decoration groups in ADCE * Uses a custom sorter to traverse decorations in a specific order * Simplifies necessary checks Updated -O and -Os pass lists.
2017-12-18 17:13:10 +00:00
}
TEST_F(AggressiveDCETest, DeadGroupMemberDecorate) {
const std::string text = R"(
; CHECK-NOT: OpDec
; CHECK-NOT: OpGroup
OpCapability Shader
OpMemoryModel Logical GLSL450
OpEntryPoint Fragment %main "main"
OpExecutionMode %main OriginUpperLeft
OpDecorate %1 Offset 0
OpDecorate %1 Uniform
%1 = OpDecorationGroup
OpGroupMemberDecorate %1 %var 0
%void = OpTypeVoid
%func = OpTypeFunction %void
%uint = OpTypeInt 32 0
%struct = OpTypeStruct %uint %uint
%struct_ptr = OpTypePointer Function %struct
%main = OpFunction %void None %func
%2 = OpLabel
%var = OpVariable %struct_ptr Function
OpReturn
OpFunctionEnd
)";
Move tests into namespaces (#1689) This CL moves the test into namespaces based on their directories.
2018-07-11 13:24:49 +00:00
SinglePassRunAndMatch<AggressiveDCEPass>(text, true);
Adding additional functionality to ADCE. Modified ADCE to remove dead globals. * Entry point and execution mode instructions are marked as alive * Reachable functions and their parameters are marked as alive * Instruction deletion now deferred until the end of the pass * Eliminated dead insts set, added IsDead to calculate that value instead * Ported applicable dead variable elimination tests * Ported dead constant elim tests Added dead function elimination to ADCE * ported dead function elim tests Added handling of decoration groups in ADCE * Uses a custom sorter to traverse decorations in a specific order * Simplifies necessary checks Updated -O and -Os pass lists.
2017-12-18 17:13:10 +00:00
}
TEST_F(AggressiveDCETest, PartiallyDeadGroupMemberDecorate) {
const std::string text = R"(
; CHECK: OpDecorate [[grp:%\w+]] Offset 0
Validate Uniform decoration (#2181)
2018-12-07 14:32:57 +00:00
; CHECK: OpDecorate [[grp]] RelaxedPrecision
Adding additional functionality to ADCE. Modified ADCE to remove dead globals. * Entry point and execution mode instructions are marked as alive * Reachable functions and their parameters are marked as alive * Instruction deletion now deferred until the end of the pass * Eliminated dead insts set, added IsDead to calculate that value instead * Ported applicable dead variable elimination tests * Ported dead constant elim tests Added dead function elimination to ADCE * ported dead function elim tests Added handling of decoration groups in ADCE * Uses a custom sorter to traverse decorations in a specific order * Simplifies necessary checks Updated -O and -Os pass lists.
2017-12-18 17:13:10 +00:00
; CHECK: [[grp]] = OpDecorationGroup
; CHECK: OpGroupMemberDecorate [[grp]] [[output:%\w+]] 1
; CHECK: [[output]] = OpTypeStruct
; CHECK-NOT: OpTypeStruct
OpCapability Shader
OpMemoryModel Logical GLSL450
OpEntryPoint Fragment %main "main" %output
OpExecutionMode %main OriginUpperLeft
OpDecorate %1 Offset 0
Validate Uniform decoration (#2181)
2018-12-07 14:32:57 +00:00
OpDecorate %1 RelaxedPrecision
Adding additional functionality to ADCE. Modified ADCE to remove dead globals. * Entry point and execution mode instructions are marked as alive * Reachable functions and their parameters are marked as alive * Instruction deletion now deferred until the end of the pass * Eliminated dead insts set, added IsDead to calculate that value instead * Ported applicable dead variable elimination tests * Ported dead constant elim tests Added dead function elimination to ADCE * ported dead function elim tests Added handling of decoration groups in ADCE * Uses a custom sorter to traverse decorations in a specific order * Simplifies necessary checks Updated -O and -Os pass lists.
2017-12-18 17:13:10 +00:00
%1 = OpDecorationGroup
OpGroupMemberDecorate %1 %var_struct 0 %output_struct 1
%void = OpTypeVoid
%func = OpTypeFunction %void
%uint = OpTypeInt 32 0
%var_struct = OpTypeStruct %uint %uint
%output_struct = OpTypeStruct %uint %uint
%struct_ptr_Function = OpTypePointer Function %var_struct
%struct_ptr_Output = OpTypePointer Output %output_struct
%uint_ptr_Output = OpTypePointer Output %uint
%output = OpVariable %struct_ptr_Output Output
%uint_0 = OpConstant %uint 0
%main = OpFunction %void None %func
%2 = OpLabel
%var = OpVariable %struct_ptr_Function Function
%3 = OpAccessChain %uint_ptr_Output %output %uint_0
OpStore %3 %uint_0
OpReturn
OpFunctionEnd
)";
Move tests into namespaces (#1689) This CL moves the test into namespaces based on their directories.
2018-07-11 13:24:49 +00:00
SinglePassRunAndMatch<AggressiveDCEPass>(text, true);
Adding additional functionality to ADCE. Modified ADCE to remove dead globals. * Entry point and execution mode instructions are marked as alive * Reachable functions and their parameters are marked as alive * Instruction deletion now deferred until the end of the pass * Eliminated dead insts set, added IsDead to calculate that value instead * Ported applicable dead variable elimination tests * Ported dead constant elim tests Added dead function elimination to ADCE * ported dead function elim tests Added handling of decoration groups in ADCE * Uses a custom sorter to traverse decorations in a specific order * Simplifies necessary checks Updated -O and -Os pass lists.
2017-12-18 17:13:10 +00:00
}
TEST_F(AggressiveDCETest,
PartiallyDeadGroupMemberDecorateDifferentGroupDecorate) {
const std::string text = R"(
; CHECK: OpDecorate [[grp:%\w+]] Offset 0
Validate Uniform decoration (#2181)
2018-12-07 14:32:57 +00:00
; CHECK: OpDecorate [[grp]] RelaxedPrecision
Adding additional functionality to ADCE. Modified ADCE to remove dead globals. * Entry point and execution mode instructions are marked as alive * Reachable functions and their parameters are marked as alive * Instruction deletion now deferred until the end of the pass * Eliminated dead insts set, added IsDead to calculate that value instead * Ported applicable dead variable elimination tests * Ported dead constant elim tests Added dead function elimination to ADCE * ported dead function elim tests Added handling of decoration groups in ADCE * Uses a custom sorter to traverse decorations in a specific order * Simplifies necessary checks Updated -O and -Os pass lists.
2017-12-18 17:13:10 +00:00
; CHECK: [[grp]] = OpDecorationGroup
; CHECK: OpGroupMemberDecorate [[grp]] [[output:%\w+]] 1
; CHECK-NOT: OpGroupMemberDecorate
; CHECK: [[output]] = OpTypeStruct
; CHECK-NOT: OpTypeStruct
OpCapability Shader
OpMemoryModel Logical GLSL450
OpEntryPoint Fragment %main "main" %output
OpExecutionMode %main OriginUpperLeft
OpDecorate %1 Offset 0
Validate Uniform decoration (#2181)
2018-12-07 14:32:57 +00:00
OpDecorate %1 RelaxedPrecision
Adding additional functionality to ADCE. Modified ADCE to remove dead globals. * Entry point and execution mode instructions are marked as alive * Reachable functions and their parameters are marked as alive * Instruction deletion now deferred until the end of the pass * Eliminated dead insts set, added IsDead to calculate that value instead * Ported applicable dead variable elimination tests * Ported dead constant elim tests Added dead function elimination to ADCE * ported dead function elim tests Added handling of decoration groups in ADCE * Uses a custom sorter to traverse decorations in a specific order * Simplifies necessary checks Updated -O and -Os pass lists.
2017-12-18 17:13:10 +00:00
%1 = OpDecorationGroup
OpGroupMemberDecorate %1 %var_struct 0
OpGroupMemberDecorate %1 %output_struct 1
%void = OpTypeVoid
%func = OpTypeFunction %void
%uint = OpTypeInt 32 0
%var_struct = OpTypeStruct %uint %uint
%output_struct = OpTypeStruct %uint %uint
%struct_ptr_Function = OpTypePointer Function %var_struct
%struct_ptr_Output = OpTypePointer Output %output_struct
%uint_ptr_Output = OpTypePointer Output %uint
%output = OpVariable %struct_ptr_Output Output
%uint_0 = OpConstant %uint 0
%main = OpFunction %void None %func
%2 = OpLabel
%var = OpVariable %struct_ptr_Function Function
%3 = OpAccessChain %uint_ptr_Output %output %uint_0
OpStore %3 %uint_0
OpReturn
OpFunctionEnd
)";
Move tests into namespaces (#1689) This CL moves the test into namespaces based on their directories.
2018-07-11 13:24:49 +00:00
SinglePassRunAndMatch<AggressiveDCEPass>(text, true);
Adding additional functionality to ADCE. Modified ADCE to remove dead globals. * Entry point and execution mode instructions are marked as alive * Reachable functions and their parameters are marked as alive * Instruction deletion now deferred until the end of the pass * Eliminated dead insts set, added IsDead to calculate that value instead * Ported applicable dead variable elimination tests * Ported dead constant elim tests Added dead function elimination to ADCE * ported dead function elim tests Added handling of decoration groups in ADCE * Uses a custom sorter to traverse decorations in a specific order * Simplifies necessary checks Updated -O and -Os pass lists.
2017-12-18 17:13:10 +00:00
}
Fixes #1404. Don't DCE workgroup size Added test.
2018-03-13 17:48:48 +00:00
// Test for #1404
TEST_F(AggressiveDCETest, DontRemoveWorkgroupSize) {
const std::string text = R"(
; CHECK: OpDecorate [[wgs:%\w+]] BuiltIn WorkgroupSize
; CHECK: [[wgs]] = OpSpecConstantComposite
OpCapability Shader
OpMemoryModel Logical GLSL450
OpEntryPoint GLCompute %func "func"
OpExecutionMode %func LocalSize 1 1 1
OpDecorate %1 BuiltIn WorkgroupSize
%void = OpTypeVoid
%int = OpTypeInt 32 0
%functy = OpTypeFunction %void
%v3int = OpTypeVector %int 3
%2 = OpSpecConstant %int 1
%1 = OpSpecConstantComposite %v3int %2 %2 %2
%func = OpFunction %void None %functy
%3 = OpLabel
OpReturn
OpFunctionEnd
)";
Move tests into namespaces (#1689) This CL moves the test into namespaces based on their directories.
2018-07-11 13:24:49 +00:00
SinglePassRunAndMatch<AggressiveDCEPass>(text, true);
Fixes #1404. Don't DCE workgroup size Added test.
2018-03-13 17:48:48 +00:00
}
Adding testcase for #1210
2018-01-18 16:02:57 +00:00
Fixes infinite loop in ADCE * Addresses how breaks are indentified to prevent infinite loops when back to back loop share a merge and header * Added test to catch the bug
2018-01-18 20:44:00 +00:00
// Test for #1214
TEST_F(AggressiveDCETest, LoopHeaderIsAlsoAnotherLoopMerge) {
const std::string text = R"(OpCapability Shader
OpMemoryModel Logical GLSL450
OpEntryPoint Fragment %1 "func" %2
OpExecutionMode %1 OriginUpperLeft
%void = OpTypeVoid
%bool = OpTypeBool
%true = OpConstantTrue %bool
%uint = OpTypeInt 32 0
%_ptr_Output_uint = OpTypePointer Output %uint
%2 = OpVariable %_ptr_Output_uint Output
%uint_0 = OpConstant %uint 0
%9 = OpTypeFunction %void
%1 = OpFunction %void None %9
%10 = OpLabel
OpBranch %11
%11 = OpLabel
OpLoopMerge %12 %13 None
OpBranchConditional %true %14 %13
%14 = OpLabel
OpStore %2 %uint_0
OpLoopMerge %15 %16 None
OpBranchConditional %true %15 %16
%16 = OpLabel
OpBranch %14
%15 = OpLabel
OpBranchConditional %true %12 %13
%13 = OpLabel
OpBranch %11
%12 = OpLabel
%17 = OpPhi %uint %uint_0 %15 %uint_0 %18
OpStore %2 %17
OpLoopMerge %19 %18 None
OpBranchConditional %true %19 %18
%18 = OpLabel
OpBranch %12
%19 = OpLabel
OpReturn
OpFunctionEnd
)";
Move tests into namespaces (#1689) This CL moves the test into namespaces based on their directories.
2018-07-11 13:24:49 +00:00
SinglePassRunAndCheck<AggressiveDCEPass>(text, text, true, true);
Fixes infinite loop in ADCE * Addresses how breaks are indentified to prevent infinite loops when back to back loop share a merge and header * Added test to catch the bug
2018-01-18 20:44:00 +00:00
}
Adding testcase for #1210
2018-01-18 16:02:57 +00:00
TEST_F(AggressiveDCETest, BreaksDontVisitPhis) {
const std::string text = R"(
OpCapability Shader
OpMemoryModel Logical GLSL450
OpEntryPoint Fragment %func "func" %var
OpExecutionMode %func OriginUpperLeft
%void = OpTypeVoid
%bool = OpTypeBool
%true = OpConstantTrue %bool
%int = OpTypeInt 32 0
%int_ptr_Output = OpTypePointer Output %int
%var = OpVariable %int_ptr_Output Output
%int0 = OpConstant %int 0
%functy = OpTypeFunction %void
%func = OpFunction %void None %functy
%entry = OpLabel
OpBranch %outer_header
%outer_header = OpLabel
OpLoopMerge %outer_merge %outer_continue None
OpBranchConditional %true %inner_header %outer_continue
%inner_header = OpLabel
%phi = OpPhi %int %int0 %outer_header %int0 %inner_continue
OpStore %var %phi
OpLoopMerge %inner_merge %inner_continue None
OpBranchConditional %true %inner_merge %inner_continue
%inner_continue = OpLabel
OpBranch %inner_header
%inner_merge = OpLabel
OpBranch %outer_continue
%outer_continue = OpLabel
%p = OpPhi %int %int0 %outer_header %int0 %inner_merge
OpStore %var %p
OpBranch %outer_header
%outer_merge = OpLabel
OpReturn
OpFunctionEnd
)";
Move tests into namespaces (#1689) This CL moves the test into namespaces based on their directories.
2018-07-11 13:24:49 +00:00
EXPECT_EQ(Pass::Status::SuccessWithoutChange,
std::get<1>(SinglePassRunAndDisassemble<AggressiveDCEPass>(
Adding testcase for #1210
2018-01-18 16:02:57 +00:00
text, false, true)));
}
Simplifying code for adding instructions to worklist * AddToWorklist can now be called unconditionally * It will only add instructions that have not already been marked as live * Fixes a case where a merge was not added to the worklist because the branch was already marked as live * Added two similar tests that fail without the fix
2018-01-18 19:15:54 +00:00
// Test for #1212
TEST_F(AggressiveDCETest, ConstStoreInnerLoop) {
const std::string text = R"(OpCapability Shader
OpMemoryModel Logical GLSL450
OpEntryPoint Vertex %1 "main" %2
%void = OpTypeVoid
%4 = OpTypeFunction %void
%float = OpTypeFloat 32
%bool = OpTypeBool
%true = OpConstantTrue %bool
%_ptr_Output_float = OpTypePointer Output %float
%2 = OpVariable %_ptr_Output_float Output
%float_3 = OpConstant %float 3
%1 = OpFunction %void None %4
%13 = OpLabel
OpBranch %14
%14 = OpLabel
OpLoopMerge %15 %16 None
OpBranchConditional %true %17 %15
%17 = OpLabel
OpStore %2 %float_3
OpLoopMerge %18 %17 None
OpBranchConditional %true %18 %17
%18 = OpLabel
OpBranch %15
%16 = OpLabel
OpBranch %14
%15 = OpLabel
OpBranch %20
%20 = OpLabel
OpReturn
OpFunctionEnd
)";
SetAssembleOptions(SPV_TEXT_TO_BINARY_OPTION_PRESERVE_NUMERIC_IDS);
Move tests into namespaces (#1689) This CL moves the test into namespaces based on their directories.
2018-07-11 13:24:49 +00:00
SinglePassRunAndCheck<AggressiveDCEPass>(text, text, true, true);
Simplifying code for adding instructions to worklist * AddToWorklist can now be called unconditionally * It will only add instructions that have not already been marked as live * Fixes a case where a merge was not added to the worklist because the branch was already marked as live * Added two similar tests that fail without the fix
2018-01-18 19:15:54 +00:00
}
// Test for #1212
TEST_F(AggressiveDCETest, InnerLoopCopy) {
const std::string text = R"(OpCapability Shader
OpMemoryModel Logical GLSL450
OpEntryPoint Vertex %1 "main" %2 %3
%void = OpTypeVoid
%5 = OpTypeFunction %void
%float = OpTypeFloat 32
%bool = OpTypeBool
%true = OpConstantTrue %bool
%_ptr_Output_float = OpTypePointer Output %float
%_ptr_Input_float = OpTypePointer Input %float
%2 = OpVariable %_ptr_Output_float Output
%3 = OpVariable %_ptr_Input_float Input
%1 = OpFunction %void None %5
%14 = OpLabel
OpBranch %15
%15 = OpLabel
OpLoopMerge %16 %17 None
OpBranchConditional %true %18 %16
%18 = OpLabel
%19 = OpLoad %float %3
OpStore %2 %19
OpLoopMerge %20 %18 None
OpBranchConditional %true %20 %18
%20 = OpLabel
OpBranch %16
%17 = OpLabel
OpBranch %15
%16 = OpLabel
OpBranch %22
%22 = OpLabel
OpReturn
OpFunctionEnd
)";
SetAssembleOptions(SPV_TEXT_TO_BINARY_OPTION_PRESERVE_NUMERIC_IDS);
Move tests into namespaces (#1689) This CL moves the test into namespaces based on their directories.
2018-07-11 13:24:49 +00:00
SinglePassRunAndCheck<AggressiveDCEPass>(text, text, true, true);
Simplifying code for adding instructions to worklist * AddToWorklist can now be called unconditionally * It will only add instructions that have not already been marked as live * Fixes a case where a merge was not added to the worklist because the branch was already marked as live * Added two similar tests that fail without the fix
2018-01-18 19:15:54 +00:00
}
Adding additional functionality to ADCE. Modified ADCE to remove dead globals. * Entry point and execution mode instructions are marked as alive * Reachable functions and their parameters are marked as alive * Instruction deletion now deferred until the end of the pass * Eliminated dead insts set, added IsDead to calculate that value instead * Ported applicable dead variable elimination tests * Ported dead constant elim tests Added dead function elimination to ADCE * ported dead function elim tests Added handling of decoration groups in ADCE * Uses a custom sorter to traverse decorations in a specific order * Simplifies necessary checks Updated -O and -Os pass lists.
2017-12-18 17:13:10 +00:00
Teach ADCE about OpImageTexelPointer Currently OpImageTexelPointer operations are treat like a use of the pointer, but it does not look for the memory being referenced to make sure stores are not removed. This change teaches it so identify the memory being accessed, and treats it as if that memory is loaded. Fixes to #1445.
2018-04-02 17:40:53 +00:00
TEST_F(AggressiveDCETest, AtomicAdd) {
const std::string text = R"(OpCapability SampledBuffer
OpCapability StorageImageExtendedFormats
OpCapability ImageBuffer
OpCapability Shader
%1 = OpExtInstImport "GLSL.std.450"
OpMemoryModel Logical GLSL450
Allow ADCE to remove dead inputs (#4629) * https://github.com/KhronosGroup/Vulkan-Docs/issues/666 clearly specified that interfaces do not require an input if there is an associated output * ADCE can now remove unused input variables (though they are kept if the preserve interfaces option is used)
2021-11-16 20:54:17 +00:00
OpEntryPoint GLCompute %2 "min"
Teach ADCE about OpImageTexelPointer Currently OpImageTexelPointer operations are treat like a use of the pointer, but it does not look for the memory being referenced to make sure stores are not removed. This change teaches it so identify the memory being accessed, and treats it as if that memory is loaded. Fixes to #1445.
2018-04-02 17:40:53 +00:00
OpExecutionMode %2 LocalSize 64 1 1
OpSource HLSL 600
OpDecorate %4 DescriptorSet 4
OpDecorate %4 Binding 70
%uint = OpTypeInt 32 0
%6 = OpTypeImage %uint Buffer 0 0 0 2 R32ui
%_ptr_UniformConstant_6 = OpTypePointer UniformConstant %6
%_ptr_Private_6 = OpTypePointer Private %6
%void = OpTypeVoid
%10 = OpTypeFunction %void
%uint_0 = OpConstant %uint 0
%uint_1 = OpConstant %uint 1
%_ptr_Image_uint = OpTypePointer Image %uint
%4 = OpVariable %_ptr_UniformConstant_6 UniformConstant
%16 = OpVariable %_ptr_Private_6 Private
%2 = OpFunction %void None %10
%17 = OpLabel
%18 = OpLoad %6 %4
OpStore %16 %18
%19 = OpImageTexelPointer %_ptr_Image_uint %16 %uint_0 %uint_0
%20 = OpAtomicIAdd %uint %19 %uint_1 %uint_0 %uint_1
OpReturn
OpFunctionEnd
)";
SetAssembleOptions(SPV_TEXT_TO_BINARY_OPTION_PRESERVE_NUMERIC_IDS);
Move tests into namespaces (#1689) This CL moves the test into namespaces based on their directories.
2018-07-11 13:24:49 +00:00
SinglePassRunAndCheck<AggressiveDCEPass>(text, text, true, true);
Teach ADCE about OpImageTexelPointer Currently OpImageTexelPointer operations are treat like a use of the pointer, but it does not look for the memory being referenced to make sure stores are not removed. This change teaches it so identify the memory being accessed, and treats it as if that memory is loaded. Fixes to #1445.
2018-04-02 17:40:53 +00:00
}
ADCE: Remove OpDecorateStringGOOGLE Also fix a few failures to set "modified" status when removing global values. Add OpDecorateStringGOOGLE to decoration ordering Fixes #1492
2018-04-16 15:33:13 +00:00
TEST_F(AggressiveDCETest, SafelyRemoveDecorateString) {
const std::string preamble = R"(OpCapability Shader
OpExtension "SPV_GOOGLE_hlsl_functionality1"
OpMemoryModel Logical GLSL450
OpEntryPoint Fragment %1 "main"
Add validation for execution modes * Check rules from Execution Mode tables, 2.16.2 and the Vulkan environment spec * Allows MeshNV execution model with the following execution modes * LocalSize, LocalSizeId, OutputPoints and OutputVertices * Done to not break their validation
2018-09-28 20:41:56 +00:00
OpExecutionMode %1 OriginUpperLeft
ADCE: Remove OpDecorateStringGOOGLE Also fix a few failures to set "modified" status when removing global values. Add OpDecorateStringGOOGLE to decoration ordering Fixes #1492
2018-04-16 15:33:13 +00:00
)";
const std::string body_before =
R"(OpDecorateStringGOOGLE %2 HlslSemanticGOOGLE "FOOBAR"
%void = OpTypeVoid
%4 = OpTypeFunction %void
%uint = OpTypeInt 32 0
%_ptr_StorageBuffer_uint = OpTypePointer StorageBuffer %uint
%2 = OpVariable %_ptr_StorageBuffer_uint StorageBuffer
%1 = OpFunction %void None %4
%7 = OpLabel
OpReturn
OpFunctionEnd
)";
const std::string body_after = R"(%void = OpTypeVoid
%4 = OpTypeFunction %void
%1 = OpFunction %void None %4
%7 = OpLabel
OpReturn
OpFunctionEnd
)";
SetAssembleOptions(SPV_TEXT_TO_BINARY_OPTION_PRESERVE_NUMERIC_IDS);
Move tests into namespaces (#1689) This CL moves the test into namespaces based on their directories.
2018-07-11 13:24:49 +00:00
SinglePassRunAndCheck<AggressiveDCEPass>(preamble + body_before,
preamble + body_after, true, true);
ADCE: Remove OpDecorateStringGOOGLE Also fix a few failures to set "modified" status when removing global values. Add OpDecorateStringGOOGLE to decoration ordering Fixes #1492
2018-04-16 15:33:13 +00:00
}
Get ADCE to handle OpCopyMemory ADCE does not treat OpCopyMemory as an instruction that references memory. Because of that stores are removed that should not be. This change teaches ADCE that OpCopyMemory and OpCopyMemorySize both loads from and stores to memory. This will keep other stores live when needed, and will allows ADCE to remove OpCopyMemory instructions as well. Fixes https://github.com/KhronosGroup/SPIRV-Tools/issues/1556.
2018-05-16 14:41:40 +00:00
TEST_F(AggressiveDCETest, CopyMemoryToGlobal) {
// |local| is loaded in an OpCopyMemory instruction. So the store must be
// kept alive.
const std::string test =
R"(OpCapability Geometry
%1 = OpExtInstImport "GLSL.std.450"
OpMemoryModel Logical GLSL450
OpEntryPoint Geometry %main "main" %global
OpExecutionMode %main Triangles
OpExecutionMode %main Invocations 1
OpExecutionMode %main OutputTriangleStrip
OpExecutionMode %main OutputVertices 5
OpSource GLSL 440
OpName %main "main"
OpName %local "local"
OpName %global "global"
%void = OpTypeVoid
%7 = OpTypeFunction %void
%float = OpTypeFloat 32
%v4float = OpTypeVector %float 4
%12 = OpConstantNull %v4float
%_ptr_Function_v4float = OpTypePointer Function %v4float
%_ptr_Output_v4float = OpTypePointer Output %v4float
%global = OpVariable %_ptr_Output_v4float Output
%main = OpFunction %void None %7
%19 = OpLabel
%local = OpVariable %_ptr_Function_v4float Function
OpStore %local %12
OpCopyMemory %global %local
OpEndPrimitive
OpReturn
OpFunctionEnd
)";
SetAssembleOptions(SPV_TEXT_TO_BINARY_OPTION_PRESERVE_NUMERIC_IDS);
Move tests into namespaces (#1689) This CL moves the test into namespaces based on their directories.
2018-07-11 13:24:49 +00:00
SinglePassRunAndCheck<AggressiveDCEPass>(test, test, true, true);
Get ADCE to handle OpCopyMemory ADCE does not treat OpCopyMemory as an instruction that references memory. Because of that stores are removed that should not be. This change teaches ADCE that OpCopyMemory and OpCopyMemorySize both loads from and stores to memory. This will keep other stores live when needed, and will allows ADCE to remove OpCopyMemory instructions as well. Fixes https://github.com/KhronosGroup/SPIRV-Tools/issues/1556.
2018-05-16 14:41:40 +00:00
}
TEST_F(AggressiveDCETest, CopyMemoryToLocal) {
// Make sure the store to |local2| using OpCopyMemory is kept and keeps
// |local1| alive.
const std::string test =
R"(OpCapability Geometry
%1 = OpExtInstImport "GLSL.std.450"
OpMemoryModel Logical GLSL450
OpEntryPoint Geometry %main "main" %global
OpExecutionMode %main Triangles
OpExecutionMode %main Invocations 1
OpExecutionMode %main OutputTriangleStrip
OpExecutionMode %main OutputVertices 5
OpSource GLSL 440
OpName %main "main"
OpName %local1 "local1"
OpName %local2 "local2"
OpName %global "global"
%void = OpTypeVoid
%7 = OpTypeFunction %void
%float = OpTypeFloat 32
%v4float = OpTypeVector %float 4
%12 = OpConstantNull %v4float
%_ptr_Function_v4float = OpTypePointer Function %v4float
%_ptr_Output_v4float = OpTypePointer Output %v4float
%global = OpVariable %_ptr_Output_v4float Output
%main = OpFunction %void None %7
%19 = OpLabel
%local1 = OpVariable %_ptr_Function_v4float Function
%local2 = OpVariable %_ptr_Function_v4float Function
OpStore %local1 %12
OpCopyMemory %local2 %local1
OpCopyMemory %global %local2
OpEndPrimitive
OpReturn
OpFunctionEnd
)";
SetAssembleOptions(SPV_TEXT_TO_BINARY_OPTION_PRESERVE_NUMERIC_IDS);
Move tests into namespaces (#1689) This CL moves the test into namespaces based on their directories.
2018-07-11 13:24:49 +00:00
SinglePassRunAndCheck<AggressiveDCEPass>(test, test, true, true);
Get ADCE to handle OpCopyMemory ADCE does not treat OpCopyMemory as an instruction that references memory. Because of that stores are removed that should not be. This change teaches ADCE that OpCopyMemory and OpCopyMemorySize both loads from and stores to memory. This will keep other stores live when needed, and will allows ADCE to remove OpCopyMemory instructions as well. Fixes https://github.com/KhronosGroup/SPIRV-Tools/issues/1556.
2018-05-16 14:41:40 +00:00
}
TEST_F(AggressiveDCETest, RemoveCopyMemoryToLocal) {
// Test that we remove function scope variables that are stored to using
// OpCopyMemory, but are never loaded. We can remove both |local1| and
// |local2|.
const std::string test =
R"(OpCapability Geometry
%1 = OpExtInstImport "GLSL.std.450"
OpMemoryModel Logical GLSL450
OpEntryPoint Geometry %main "main" %global
OpExecutionMode %main Triangles
OpExecutionMode %main Invocations 1
OpExecutionMode %main OutputTriangleStrip
OpExecutionMode %main OutputVertices 5
OpSource GLSL 440
OpName %main "main"
OpName %local1 "local1"
OpName %local2 "local2"
OpName %global "global"
%void = OpTypeVoid
%7 = OpTypeFunction %void
%float = OpTypeFloat 32
%v4float = OpTypeVector %float 4
%12 = OpConstantNull %v4float
%_ptr_Function_v4float = OpTypePointer Function %v4float
%_ptr_Output_v4float = OpTypePointer Output %v4float
%global = OpVariable %_ptr_Output_v4float Output
%main = OpFunction %void None %7
%19 = OpLabel
%local1 = OpVariable %_ptr_Function_v4float Function
%local2 = OpVariable %_ptr_Function_v4float Function
OpStore %local1 %12
OpCopyMemory %local2 %local1
OpEndPrimitive
OpReturn
OpFunctionEnd
)";
const std::string result =
R"(OpCapability Geometry
%1 = OpExtInstImport "GLSL.std.450"
OpMemoryModel Logical GLSL450
OpEntryPoint Geometry %main "main" %global
OpExecutionMode %main Triangles
OpExecutionMode %main Invocations 1
OpExecutionMode %main OutputTriangleStrip
OpExecutionMode %main OutputVertices 5
OpSource GLSL 440
OpName %main "main"
OpName %global "global"
%void = OpTypeVoid
%7 = OpTypeFunction %void
%float = OpTypeFloat 32
%v4float = OpTypeVector %float 4
%_ptr_Output_v4float = OpTypePointer Output %v4float
%global = OpVariable %_ptr_Output_v4float Output
%main = OpFunction %void None %7
%19 = OpLabel
OpEndPrimitive
OpReturn
OpFunctionEnd
)";
SetAssembleOptions(SPV_TEXT_TO_BINARY_OPTION_PRESERVE_NUMERIC_IDS);
Move tests into namespaces (#1689) This CL moves the test into namespaces based on their directories.
2018-07-11 13:24:49 +00:00
SinglePassRunAndCheck<AggressiveDCEPass>(test, result, true, true);
Get ADCE to handle OpCopyMemory ADCE does not treat OpCopyMemory as an instruction that references memory. Because of that stores are removed that should not be. This change teaches ADCE that OpCopyMemory and OpCopyMemorySize both loads from and stores to memory. This will keep other stores live when needed, and will allows ADCE to remove OpCopyMemory instructions as well. Fixes https://github.com/KhronosGroup/SPIRV-Tools/issues/1556.
2018-05-16 14:41:40 +00:00
}
TEST_F(AggressiveDCETest, RemoveCopyMemoryToLocal2) {
// We are able to remove "local2" because it is not loaded, but have to keep
// the stores to "local1".
const std::string test =
R"(OpCapability Geometry
%1 = OpExtInstImport "GLSL.std.450"
OpMemoryModel Logical GLSL450
OpEntryPoint Geometry %main "main" %global
OpExecutionMode %main Triangles
OpExecutionMode %main Invocations 1
OpExecutionMode %main OutputTriangleStrip
OpExecutionMode %main OutputVertices 5
OpSource GLSL 440
OpName %main "main"
OpName %local1 "local1"
OpName %local2 "local2"
OpName %global "global"
%void = OpTypeVoid
%7 = OpTypeFunction %void
%float = OpTypeFloat 32
%v4float = OpTypeVector %float 4
%12 = OpConstantNull %v4float
%_ptr_Function_v4float = OpTypePointer Function %v4float
%_ptr_Output_v4float = OpTypePointer Output %v4float
%global = OpVariable %_ptr_Output_v4float Output
%main = OpFunction %void None %7
%19 = OpLabel
%local1 = OpVariable %_ptr_Function_v4float Function
%local2 = OpVariable %_ptr_Function_v4float Function
OpStore %local1 %12
OpCopyMemory %local2 %local1
OpCopyMemory %global %local1
OpEndPrimitive
OpReturn
OpFunctionEnd
)";
const std::string result =
R"(OpCapability Geometry
%1 = OpExtInstImport "GLSL.std.450"
OpMemoryModel Logical GLSL450
OpEntryPoint Geometry %main "main" %global
OpExecutionMode %main Triangles
OpExecutionMode %main Invocations 1
OpExecutionMode %main OutputTriangleStrip
OpExecutionMode %main OutputVertices 5
OpSource GLSL 440
OpName %main "main"
OpName %local1 "local1"
OpName %global "global"
%void = OpTypeVoid
%7 = OpTypeFunction %void
%float = OpTypeFloat 32
%v4float = OpTypeVector %float 4
%12 = OpConstantNull %v4float
%_ptr_Function_v4float = OpTypePointer Function %v4float
%_ptr_Output_v4float = OpTypePointer Output %v4float
%global = OpVariable %_ptr_Output_v4float Output
%main = OpFunction %void None %7
%19 = OpLabel
%local1 = OpVariable %_ptr_Function_v4float Function
OpStore %local1 %12
OpCopyMemory %global %local1
OpEndPrimitive
OpReturn
OpFunctionEnd
)";
SetAssembleOptions(SPV_TEXT_TO_BINARY_OPTION_PRESERVE_NUMERIC_IDS);
Move tests into namespaces (#1689) This CL moves the test into namespaces based on their directories.
2018-07-11 13:24:49 +00:00
SinglePassRunAndCheck<AggressiveDCEPass>(test, result, true, true);
Get ADCE to handle OpCopyMemory ADCE does not treat OpCopyMemory as an instruction that references memory. Because of that stores are removed that should not be. This change teaches ADCE that OpCopyMemory and OpCopyMemorySize both loads from and stores to memory. This will keep other stores live when needed, and will allows ADCE to remove OpCopyMemory instructions as well. Fixes https://github.com/KhronosGroup/SPIRV-Tools/issues/1556.
2018-05-16 14:41:40 +00:00
}
Handle breaks from structured-ifs in DCE. (#1848) * Handle breaks from structured-ifs in DCE. dead code elimination assumes that are conditional branches except for breaks and continues in loops will have an OpSelectionMerge before them. That is not true when breaking out of a selection construct. The fix is to look for breaks in selection constructs in the same place we look for breaks and continues for loops.
2018-08-21 15:54:44 +00:00
TEST_F(AggressiveDCETest, StructuredIfWithConditionalExit) {
// We are able to remove "local2" because it is not loaded, but have to keep
// the stores to "local1".
const std::string test =
R"(OpCapability Shader
%1 = OpExtInstImport "GLSL.std.450"
OpMemoryModel Logical GLSL450
OpEntryPoint Fragment %main "main"
OpExecutionMode %main OriginUpperLeft
OpSource GLSL 140
OpSourceExtension "GL_GOOGLE_cpp_style_line_directive"
OpSourceExtension "GL_GOOGLE_include_directive"
OpName %main "main"
OpName %a "a"
%void = OpTypeVoid
%5 = OpTypeFunction %void
%int = OpTypeInt 32 1
%_ptr_Uniform_int = OpTypePointer Uniform %int
%int_0 = OpConstant %int 0
%bool = OpTypeBool
%int_100 = OpConstant %int 100
%int_1 = OpConstant %int 1
%a = OpVariable %_ptr_Uniform_int Uniform
%main = OpFunction %void None %5
%12 = OpLabel
%13 = OpLoad %int %a
%14 = OpSGreaterThan %bool %13 %int_0
OpSelectionMerge %15 None
OpBranchConditional %14 %16 %15
%16 = OpLabel
%17 = OpLoad %int %a
%18 = OpSLessThan %bool %17 %int_100
OpBranchConditional %18 %19 %15
%19 = OpLabel
OpStore %a %int_1
OpBranch %15
%15 = OpLabel
OpReturn
OpFunctionEnd
)";
SetAssembleOptions(SPV_TEXT_TO_BINARY_OPTION_PRESERVE_NUMERIC_IDS);
SinglePassRunAndCheck<AggressiveDCEPass>(test, test, true, true);
}
Removing HLSLCounterBuffer decorations when not needed. (#1954) The HlslCounterBufferGOOGLE that was introduced changed the OpDecorateId so that is can now reference an id other than the target. If that other id is used only in the decoration, then the definition of the id will be removed because decoration do not count as real uses. However, if the target of the decoration is still live the decoration will not be removed. This leaves a reference to an id that is not defined. There are two solutions to consider. The first is that is the decoration is kept, then the definition of the id should be kept live. Implementing this change would be involved because the way ADCE handles decorations will have to be reimplemented. The other solution is to remove the decoration the id is otherwise dead. This works for this specific case. Also this is the more desirable behaviour in this case. The id will always be the id of a variable that belongs to a descriptor set. If that variable is not bound and we do not remove it, the driver will complain. I chose to implement the second solution. The first will be left to when a case for it comes up. Fixes https://github.com/KhronosGroup/SPIRV-Tools/issues/1885.
2018-10-05 12:23:09 +00:00
Fix ADCE liveness to include all enclosing control structures. (#1975) Was removing control structures which didn't have data dependency with enclosed live loop and otherwise did not contain live code. An example is a counting loop around a live loop. Fixes #1967.
2018-10-16 12:00:07 +00:00
TEST_F(AggressiveDCETest, CountingLoopNotEliminated) {
// #version 310 es
//
// precision highp float;
// precision highp int;
//
// layout(location = 0) out vec4 _GLF_color;
//
// void main()
// {
// float data[1];
// for (int c = 0; c < 1; c++) {
// if (true) {
// do {
// for (int i = 0; i < 1; i++) {
// data[i] = 1.0;
// }
// } while (false);
// }
// }
// _GLF_color = vec4(data[0], 0.0, 0.0, 1.0);
// }
const std::string test =
R"(OpCapability Shader
%1 = OpExtInstImport "GLSL.std.450"
OpMemoryModel Logical GLSL450
OpEntryPoint Fragment %main "main" %_GLF_color
OpExecutionMode %main OriginUpperLeft
OpSource ESSL 310
OpName %main "main"
OpName %c "c"
OpName %i "i"
OpName %data "data"
OpName %_GLF_color "_GLF_color"
OpDecorate %_GLF_color Location 0
%void = OpTypeVoid
%8 = OpTypeFunction %void
%int = OpTypeInt 32 1
%_ptr_Function_int = OpTypePointer Function %int
%int_0 = OpConstant %int 0
%int_1 = OpConstant %int 1
%bool = OpTypeBool
%float = OpTypeFloat 32
%uint = OpTypeInt 32 0
%uint_1 = OpConstant %uint 1
%_arr_float_uint_1 = OpTypeArray %float %uint_1
%_ptr_Function__arr_float_uint_1 = OpTypePointer Function %_arr_float_uint_1
%float_1 = OpConstant %float 1
%_ptr_Function_float = OpTypePointer Function %float
%false = OpConstantFalse %bool
%v4float = OpTypeVector %float 4
%_ptr_Output_v4float = OpTypePointer Output %v4float
%_GLF_color = OpVariable %_ptr_Output_v4float Output
%float_0 = OpConstant %float 0
%main = OpFunction %void None %8
%26 = OpLabel
%c = OpVariable %_ptr_Function_int Function
%i = OpVariable %_ptr_Function_int Function
%data = OpVariable %_ptr_Function__arr_float_uint_1 Function
OpStore %c %int_0
OpBranch %27
%27 = OpLabel
OpLoopMerge %28 %29 None
OpBranch %30
%30 = OpLabel
%31 = OpLoad %int %c
%32 = OpSLessThan %bool %31 %int_1
OpBranchConditional %32 %33 %28
%33 = OpLabel
OpBranch %34
%34 = OpLabel
OpBranch %35
%35 = OpLabel
OpLoopMerge %36 %37 None
OpBranch %38
%38 = OpLabel
OpStore %i %int_0
OpBranch %39
%39 = OpLabel
OpLoopMerge %40 %41 None
OpBranch %42
%42 = OpLabel
%43 = OpLoad %int %i
%44 = OpSLessThan %bool %43 %int_1
OpBranchConditional %44 %46 %40
%46 = OpLabel
%47 = OpLoad %int %i
%48 = OpAccessChain %_ptr_Function_float %data %47
OpStore %48 %float_1
OpBranch %41
%41 = OpLabel
%49 = OpLoad %int %i
%50 = OpIAdd %int %49 %int_1
OpStore %i %50
OpBranch %39
%40 = OpLabel
OpBranch %37
%37 = OpLabel
OpBranchConditional %false %35 %36
%36 = OpLabel
OpBranch %45
%45 = OpLabel
OpBranch %29
%29 = OpLabel
%51 = OpLoad %int %c
%52 = OpIAdd %int %51 %int_1
OpStore %c %52
OpBranch %27
%28 = OpLabel
%53 = OpAccessChain %_ptr_Function_float %data %int_0
%54 = OpLoad %float %53
%55 = OpCompositeConstruct %v4float %54 %float_0 %float_0 %float_1
OpStore %_GLF_color %55
OpReturn
OpFunctionEnd
)";
SetAssembleOptions(SPV_TEXT_TO_BINARY_OPTION_PRESERVE_NUMERIC_IDS);
SinglePassRunAndCheck<AggressiveDCEPass>(test, test, true, true);
}
Fix ADCE to treat OpUnreachable correctly during liveness analysis (#1984) ADCE liveness algorithm should treat OpUnreachable at least like other branch instructions. It was being treated as always live which was preventing useless structured constructs from being eliminated. OpUnreachable is generated by dead branch elimination which is now being required by merge return, so this fix should accompany that change.
2018-10-19 14:16:35 +00:00
TEST_F(AggressiveDCETest, EliminateLoopWithUnreachable) {
// #version 430
//
// layout(std430) buffer U_t
// {
// float g_F[10];
// float g_S;
// };
//
// layout(location = 0)out float o;
//
// void main(void)
// {
// // Useless loop
// for (int i = 0; i<10; i++) {
// if (g_F[i] == 0.0)
// break;
// else
// break;
// // Unreachable merge block created here.
// // Need to edit SPIR-V to change to OpUnreachable
// }
// o = g_S;
// }
const std::string before =
R"(OpCapability Shader
%1 = OpExtInstImport "GLSL.std.450"
OpMemoryModel Logical GLSL450
OpEntryPoint Fragment %main "main" %o
OpExecutionMode %main OriginUpperLeft
OpSource GLSL 430
OpName %main "main"
OpName %i "i"
OpName %U_t "U_t"
OpMemberName %U_t 0 "g_F"
OpMemberName %U_t 1 "g_S"
OpName %_ ""
OpName %o "o"
OpDecorate %_arr_float_uint_10 ArrayStride 4
OpMemberDecorate %U_t 0 Offset 0
OpMemberDecorate %U_t 1 Offset 40
OpDecorate %U_t BufferBlock
OpDecorate %_ DescriptorSet 0
OpDecorate %o Location 0
%void = OpTypeVoid
%9 = OpTypeFunction %void
%int = OpTypeInt 32 1
%_ptr_Function_int = OpTypePointer Function %int
%int_0 = OpConstant %int 0
%int_10 = OpConstant %int 10
%bool = OpTypeBool
%float = OpTypeFloat 32
%uint = OpTypeInt 32 0
%uint_10 = OpConstant %uint 10
%_arr_float_uint_10 = OpTypeArray %float %uint_10
%U_t = OpTypeStruct %_arr_float_uint_10 %float
%_ptr_Uniform_U_t = OpTypePointer Uniform %U_t
%_ = OpVariable %_ptr_Uniform_U_t Uniform
%_ptr_Uniform_float = OpTypePointer Uniform %float
%float_0 = OpConstant %float 0
%int_1 = OpConstant %int 1
%_ptr_Output_float = OpTypePointer Output %float
%o = OpVariable %_ptr_Output_float Output
%main = OpFunction %void None %9
%23 = OpLabel
%i = OpVariable %_ptr_Function_int Function
OpStore %i %int_0
OpBranch %24
%24 = OpLabel
OpLoopMerge %25 %26 None
OpBranch %27
%27 = OpLabel
%28 = OpLoad %int %i
%29 = OpSLessThan %bool %28 %int_10
OpBranchConditional %29 %30 %25
%30 = OpLabel
%31 = OpLoad %int %i
%32 = OpAccessChain %_ptr_Uniform_float %_ %int_0 %31
%33 = OpLoad %float %32
%34 = OpFOrdEqual %bool %33 %float_0
OpSelectionMerge %35 None
OpBranchConditional %34 %36 %37
%36 = OpLabel
OpBranch %25
%37 = OpLabel
OpBranch %25
%35 = OpLabel
OpUnreachable
%26 = OpLabel
%38 = OpLoad %int %i
%39 = OpIAdd %int %38 %int_1
OpStore %i %39
OpBranch %24
%25 = OpLabel
%40 = OpAccessChain %_ptr_Uniform_float %_ %int_1
%41 = OpLoad %float %40
OpStore %o %41
OpReturn
OpFunctionEnd
)";
const std::string after =
R"(OpCapability Shader
%1 = OpExtInstImport "GLSL.std.450"
OpMemoryModel Logical GLSL450
OpEntryPoint Fragment %main "main" %o
OpExecutionMode %main OriginUpperLeft
OpSource GLSL 430
OpName %main "main"
OpName %U_t "U_t"
OpMemberName %U_t 0 "g_F"
OpMemberName %U_t 1 "g_S"
OpName %_ ""
OpName %o "o"
OpDecorate %_arr_float_uint_10 ArrayStride 4
OpMemberDecorate %U_t 0 Offset 0
OpMemberDecorate %U_t 1 Offset 40
OpDecorate %U_t BufferBlock
OpDecorate %_ DescriptorSet 0
OpDecorate %o Location 0
%void = OpTypeVoid
%9 = OpTypeFunction %void
%int = OpTypeInt 32 1
%float = OpTypeFloat 32
%uint = OpTypeInt 32 0
%uint_10 = OpConstant %uint 10
%_arr_float_uint_10 = OpTypeArray %float %uint_10
%U_t = OpTypeStruct %_arr_float_uint_10 %float
%_ptr_Uniform_U_t = OpTypePointer Uniform %U_t
%_ = OpVariable %_ptr_Uniform_U_t Uniform
%_ptr_Uniform_float = OpTypePointer Uniform %float
%int_1 = OpConstant %int 1
%_ptr_Output_float = OpTypePointer Output %float
%o = OpVariable %_ptr_Output_float Output
%main = OpFunction %void None %9
%23 = OpLabel
OpBranch %24
%24 = OpLabel
OpBranch %25
%25 = OpLabel
%40 = OpAccessChain %_ptr_Uniform_float %_ %int_1
%41 = OpLoad %float %40
OpStore %o %41
OpReturn
OpFunctionEnd
)";
SetAssembleOptions(SPV_TEXT_TO_BINARY_OPTION_PRESERVE_NUMERIC_IDS);
SinglePassRunAndCheck<AggressiveDCEPass>(before, after, true, true);
}
Removing HLSLCounterBuffer decorations when not needed. (#1954) The HlslCounterBufferGOOGLE that was introduced changed the OpDecorateId so that is can now reference an id other than the target. If that other id is used only in the decoration, then the definition of the id will be removed because decoration do not count as real uses. However, if the target of the decoration is still live the decoration will not be removed. This leaves a reference to an id that is not defined. There are two solutions to consider. The first is that is the decoration is kept, then the definition of the id should be kept live. Implementing this change would be involved because the way ADCE handles decorations will have to be reimplemented. The other solution is to remove the decoration the id is otherwise dead. This works for this specific case. Also this is the more desirable behaviour in this case. The id will always be the id of a variable that belongs to a descriptor set. If that variable is not bound and we do not remove it, the driver will complain. I chose to implement the second solution. The first will be left to when a case for it comes up. Fixes https://github.com/KhronosGroup/SPIRV-Tools/issues/1885.
2018-10-05 12:23:09 +00:00
TEST_F(AggressiveDCETest, DeadHlslCounterBufferGOOGLE) {
// We are able to remove "local2" because it is not loaded, but have to keep
// the stores to "local1".
const std::string test =
R"(
; CHECK-NOT: OpDecorateId
; CHECK: [[var:%\w+]] = OpVariable
; CHECK-NOT: OpVariable
; CHECK: [[ac:%\w+]] = OpAccessChain {{%\w+}} [[var]]
; CHECK: OpStore [[ac]]
OpCapability Shader
OpExtension "SPV_GOOGLE_hlsl_functionality1"
OpMemoryModel Logical GLSL450
OpEntryPoint GLCompute %1 "main"
OpExecutionMode %1 LocalSize 32 1 1
OpSource HLSL 600
OpDecorate %_runtimearr_v2float ArrayStride 8
OpMemberDecorate %_struct_3 0 Offset 0
OpDecorate %_struct_3 BufferBlock
OpMemberDecorate %_struct_4 0 Offset 0
OpDecorate %_struct_4 BufferBlock
OpDecorateId %5 HlslCounterBufferGOOGLE %6
OpDecorate %5 DescriptorSet 0
OpDecorate %5 Binding 0
OpDecorate %6 DescriptorSet 0
OpDecorate %6 Binding 1
%float = OpTypeFloat 32
%v2float = OpTypeVector %float 2
%_runtimearr_v2float = OpTypeRuntimeArray %v2float
%_struct_3 = OpTypeStruct %_runtimearr_v2float
%_ptr_Uniform__struct_3 = OpTypePointer Uniform %_struct_3
%int = OpTypeInt 32 1
%_struct_4 = OpTypeStruct %int
%_ptr_Uniform__struct_4 = OpTypePointer Uniform %_struct_4
%void = OpTypeVoid
%13 = OpTypeFunction %void
%19 = OpConstantNull %v2float
%int_0 = OpConstant %int 0
%_ptr_Uniform_v2float = OpTypePointer Uniform %v2float
%5 = OpVariable %_ptr_Uniform__struct_3 Uniform
%6 = OpVariable %_ptr_Uniform__struct_4 Uniform
%1 = OpFunction %void None %13
%22 = OpLabel
%23 = OpAccessChain %_ptr_Uniform_v2float %5 %int_0 %int_0
OpStore %23 %19
OpReturn
OpFunctionEnd
)";
SetAssembleOptions(SPV_TEXT_TO_BINARY_OPTION_PRESERVE_NUMERIC_IDS);
SinglePassRunAndMatch<AggressiveDCEPass>(test, true);
}
Handle OpMemberDecorateStringGOOGLE in ACDE (#2029) Add missing case to the switch statement for the annotation instructions. See https://github.com/KhronosGroup/glslang/issues/1561.
2018-11-02 17:42:45 +00:00
TEST_F(AggressiveDCETest, Dead) {
// We are able to remove "local2" because it is not loaded, but have to keep
// the stores to "local1".
const std::string test =
R"(
; CHECK: OpCapability
; CHECK-NOT: OpMemberDecorateStringGOOGLE
; CHECK: OpFunctionEnd
OpCapability Shader
OpExtension "SPV_GOOGLE_hlsl_functionality1"
%1 = OpExtInstImport "GLSL.std.450"
OpMemoryModel Logical GLSL450
OpEntryPoint Vertex %VSMain "VSMain"
OpSource HLSL 500
OpName %VSMain "VSMain"
OpName %PSInput "PSInput"
OpMemberName %PSInput 0 "Pos"
OpMemberName %PSInput 1 "uv"
OpMemberDecorateStringGOOGLE %PSInput 0 HlslSemanticGOOGLE "SV_POSITION"
OpMemberDecorateStringGOOGLE %PSInput 1 HlslSemanticGOOGLE "TEX_COORD"
%void = OpTypeVoid
%5 = OpTypeFunction %void
%float = OpTypeFloat 32
%v2float = OpTypeVector %float 2
%v4float = OpTypeVector %float 4
%PSInput = OpTypeStruct %v4float %v2float
%VSMain = OpFunction %void None %5
%9 = OpLabel
OpReturn
OpFunctionEnd
)";
SetAssembleOptions(SPV_TEXT_TO_BINARY_OPTION_PRESERVE_NUMERIC_IDS);
SinglePassRunAndMatch<AggressiveDCEPass>(test, true);
}
Handle dead infinite loops in DCE (#2471) Fixes #2456 * When eliminating a structured construct that has an unreachable merge, replace that unreachable terminator with an appropriate return * New tests
2019-04-03 14:30:12 +00:00
TEST_F(AggressiveDCETest, DeadInfiniteLoop) {
const std::string test = R"(
; CHECK: OpSwitch {{%\w+}} {{%\w+}} {{\w+}} {{%\w+}} {{\w+}} [[block:%\w+]]
; CHECK: [[block]] = OpLabel
; CHECK-NEXT: OpBranch [[block:%\w+]]
; CHECK: [[block]] = OpLabel
; CHECK-NEXT: OpBranch [[block:%\w+]]
; CHECK: [[block]] = OpLabel
; CHECK-NEXT: OpReturn
OpCapability Shader
OpMemoryModel Logical GLSL450
OpEntryPoint Fragment %2 "main"
OpExecutionMode %2 OriginUpperLeft
%6 = OpTypeVoid
%7 = OpTypeFunction %6
%8 = OpTypeFloat 32
%9 = OpTypeVector %8 3
%10 = OpTypeFunction %9
%11 = OpConstant %8 1
%12 = OpConstantComposite %9 %11 %11 %11
%13 = OpTypeInt 32 1
%32 = OpUndef %13
%2 = OpFunction %6 None %7
%33 = OpLabel
OpBranch %34
%34 = OpLabel
OpLoopMerge %35 %36 None
OpBranch %37
%37 = OpLabel
%38 = OpFunctionCall %9 %39
OpSelectionMerge %40 None
OpSwitch %32 %40 14 %41 58 %42
%42 = OpLabel
OpBranch %43
%43 = OpLabel
OpLoopMerge %44 %45 None
OpBranch %45
%45 = OpLabel
OpBranch %43
%44 = OpLabel
OpUnreachable
%41 = OpLabel
OpBranch %36
%40 = OpLabel
OpBranch %36
%36 = OpLabel
OpBranch %34
%35 = OpLabel
OpReturn
OpFunctionEnd
%39 = OpFunction %9 None %10
%46 = OpLabel
OpReturnValue %12
OpFunctionEnd
)";
SetAssembleOptions(SPV_TEXT_TO_BINARY_OPTION_PRESERVE_NUMERIC_IDS);
SinglePassRunAndMatch<AggressiveDCEPass>(test, true);
}
TEST_F(AggressiveDCETest, DeadInfiniteLoopReturnValue) {
const std::string test = R"(
; CHECK: [[vec3:%\w+]] = OpTypeVector
; CHECK: [[undef:%\w+]] = OpUndef [[vec3]]
; CHECK: OpSwitch {{%\w+}} {{%\w+}} {{\w+}} {{%\w+}} {{\w+}} [[block:%\w+]]
; CHECK: [[block]] = OpLabel
; CHECK-NEXT: OpBranch [[block:%\w+]]
; CHECK: [[block]] = OpLabel
; CHECK-NEXT: OpBranch [[block:%\w+]]
; CHECK: [[block]] = OpLabel
; CHECK-NEXT: OpReturnValue [[undef]]
OpCapability Shader
OpMemoryModel Logical GLSL450
OpEntryPoint Fragment %2 "main"
OpExecutionMode %2 OriginUpperLeft
%6 = OpTypeVoid
%7 = OpTypeFunction %6
%8 = OpTypeFloat 32
%9 = OpTypeVector %8 3
%10 = OpTypeFunction %9
%11 = OpConstant %8 1
%12 = OpConstantComposite %9 %11 %11 %11
%13 = OpTypeInt 32 1
%32 = OpUndef %13
%2 = OpFunction %6 None %7
%entry = OpLabel
%call = OpFunctionCall %9 %func
OpReturn
OpFunctionEnd
%func = OpFunction %9 None %10
%33 = OpLabel
OpBranch %34
%34 = OpLabel
OpLoopMerge %35 %36 None
OpBranch %37
%37 = OpLabel
%38 = OpFunctionCall %9 %39
OpSelectionMerge %40 None
OpSwitch %32 %40 14 %41 58 %42
%42 = OpLabel
OpBranch %43
%43 = OpLabel
OpLoopMerge %44 %45 None
OpBranch %45
%45 = OpLabel
OpBranch %43
%44 = OpLabel
OpUnreachable
%41 = OpLabel
OpBranch %36
%40 = OpLabel
OpBranch %36
%36 = OpLabel
OpBranch %34
%35 = OpLabel
OpReturnValue %12
OpFunctionEnd
%39 = OpFunction %9 None %10
%46 = OpLabel
OpReturnValue %12
OpFunctionEnd
)";
SetAssembleOptions(SPV_TEXT_TO_BINARY_OPTION_PRESERVE_NUMERIC_IDS);
SinglePassRunAndMatch<AggressiveDCEPass>(test, true);
}
Handle variable pointer in some optimizations (#2490) * Check var pointer capability in ADCE. * Check var ptr capability for common uniform. * Check var ptr capability in access chain convert. Since we want this pass to run even if there are variable pointer on storage buffers, we had to remove asserts that assumed there were no variable pointers. The functions with the asserts will now work, it becomes the responsibility of the callers to deal with the output as appropriate. * Single block elimination and variable pointers. It seems like the code in local single block elimination is able to handle cases with variable pointers already. This is because the function `HasOnlySupportedRefs` ensures that variables that feed a variable pointer are not candidates. * Single store elimination and variable pointers. It seems like the code in local single stroe elimination is able to handle cases with variable pointers already. This is because the function `FindSingleStoreAndCheckUses` ensures that variables that feed a variable pointer are not candidates. * SSA rewriter and variable pointers. It seems like the code in the two passes that call the SSA rewriter are able to handle cases with variable pointers already. This is because the function `HasOnlySupportedRefs` ensures that variables that feed a variable pointer are not candidates. Fixes #2458.
2019-04-03 16:47:51 +00:00
TEST_F(AggressiveDCETest, TestVariablePointer) {
const std::string before =
R"(OpCapability Shader
OpCapability VariablePointers
%1 = OpExtInstImport "GLSL.std.450"
OpMemoryModel Logical GLSL450
OpEntryPoint GLCompute %2 "main"
OpExecutionMode %2 LocalSize 1 1 1
OpSource GLSL 450
OpMemberDecorate %_struct_3 0 Offset 0
OpDecorate %_struct_3 Block
OpDecorate %4 DescriptorSet 0
OpDecorate %4 Binding 0
OpDecorate %_ptr_StorageBuffer_int ArrayStride 4
OpDecorate %_arr_int_int_128 ArrayStride 4
%void = OpTypeVoid
%8 = OpTypeFunction %void
%int = OpTypeInt 32 1
%int_128 = OpConstant %int 128
%_arr_int_int_128 = OpTypeArray %int %int_128
%_struct_3 = OpTypeStruct %_arr_int_int_128
%_ptr_StorageBuffer__struct_3 = OpTypePointer StorageBuffer %_struct_3
%4 = OpVariable %_ptr_StorageBuffer__struct_3 StorageBuffer
%bool = OpTypeBool
%true = OpConstantTrue %bool
%int_0 = OpConstant %int 0
%int_1 = OpConstant %int 1
%_ptr_StorageBuffer_int = OpTypePointer StorageBuffer %int
%2 = OpFunction %void None %8
%16 = OpLabel
%17 = OpAccessChain %_ptr_StorageBuffer_int %4 %int_0 %int_0
OpBranch %18
%18 = OpLabel
%19 = OpPhi %_ptr_StorageBuffer_int %17 %16 %20 %21
OpLoopMerge %22 %21 None
OpBranchConditional %true %23 %22
%23 = OpLabel
OpStore %19 %int_0
OpBranch %21
%21 = OpLabel
%20 = OpPtrAccessChain %_ptr_StorageBuffer_int %19 %int_1
OpBranch %18
%22 = OpLabel
OpReturn
OpFunctionEnd
)";
SetAssembleOptions(SPV_TEXT_TO_BINARY_OPTION_PRESERVE_NUMERIC_IDS);
SinglePassRunAndCheck<AggressiveDCEPass>(before, before, true, true);
}
ADCE support for SPIR-V 1.4 entry points (#2561) Fixes #2551 * Add support for 1.4 entry point interface lists * only input and output variables are automatically live * can clean up interfaces after DCE * added tests * allow opt tests to specify a target environment
2019-05-07 18:52:22 +00:00
TEST_F(AggressiveDCETest, DeadInputInterfaceV13) {
const std::string spirv = R"(
Allow ADCE to remove dead inputs (#4629) * https://github.com/KhronosGroup/Vulkan-Docs/issues/666 clearly specified that interfaces do not require an input if there is an associated output * ADCE can now remove unused input variables (though they are kept if the preserve interfaces option is used)
2021-11-16 20:54:17 +00:00
; CHECK: OpEntryPoint GLCompute %main "main"
; CHECK-NOT: OpVariable
ADCE support for SPIR-V 1.4 entry points (#2561) Fixes #2551 * Add support for 1.4 entry point interface lists * only input and output variables are automatically live * can clean up interfaces after DCE * added tests * allow opt tests to specify a target environment
2019-05-07 18:52:22 +00:00
OpCapability Shader
OpMemoryModel Logical GLSL450
OpEntryPoint GLCompute %main "main" %dead
OpExecutionMode %main LocalSize 1 1 1
OpName %main "main"
%void = OpTypeVoid
%int = OpTypeInt 32 0
%ptr_input_int = OpTypePointer Input %int
%dead = OpVariable %ptr_input_int Input
%void_fn = OpTypeFunction %void
%main = OpFunction %void None %void_fn
%entry = OpLabel
OpReturn
OpFunctionEnd
)";
SetTargetEnv(SPV_ENV_UNIVERSAL_1_3);
SinglePassRunAndMatch<AggressiveDCEPass>(spirv, true);
}
TEST_F(AggressiveDCETest, DeadInputInterfaceV14) {
const std::string spirv = R"(
Allow ADCE to remove dead inputs (#4629) * https://github.com/KhronosGroup/Vulkan-Docs/issues/666 clearly specified that interfaces do not require an input if there is an associated output * ADCE can now remove unused input variables (though they are kept if the preserve interfaces option is used)
2021-11-16 20:54:17 +00:00
; CHECK: OpEntryPoint GLCompute %main "main"
; CHECK-NOT: OpVariable
ADCE support for SPIR-V 1.4 entry points (#2561) Fixes #2551 * Add support for 1.4 entry point interface lists * only input and output variables are automatically live * can clean up interfaces after DCE * added tests * allow opt tests to specify a target environment
2019-05-07 18:52:22 +00:00
OpCapability Shader
OpMemoryModel Logical GLSL450
OpEntryPoint GLCompute %main "main" %dead
OpExecutionMode %main LocalSize 1 1 1
OpName %main "main"
%void = OpTypeVoid
%int = OpTypeInt 32 0
%ptr_input_int = OpTypePointer Input %int
%dead = OpVariable %ptr_input_int Input
%void_fn = OpTypeFunction %void
%main = OpFunction %void None %void_fn
%entry = OpLabel
OpReturn
OpFunctionEnd
)";
SetTargetEnv(SPV_ENV_UNIVERSAL_1_4);
SinglePassRunAndMatch<AggressiveDCEPass>(spirv, true);
}
TEST_F(AggressiveDCETest, DeadInterfaceV14) {
const std::string spirv = R"(
; CHECK-NOT: OpEntryPoint GLCompute %main "main" %
; CHECK: OpEntryPoint GLCompute %main "main"
; CHECK-NOT: OpVariable
OpCapability Shader
OpMemoryModel Logical GLSL450
OpEntryPoint GLCompute %main "main" %dead
OpExecutionMode %main LocalSize 1 1 1
OpName %main "main"
%void = OpTypeVoid
%int = OpTypeInt 32 0
%ptr_private_int = OpTypePointer Private %int
%dead = OpVariable %ptr_private_int Private
%void_fn = OpTypeFunction %void
%main = OpFunction %void None %void_fn
%entry = OpLabel
OpReturn
OpFunctionEnd
)";
SetTargetEnv(SPV_ENV_UNIVERSAL_1_4);
SinglePassRunAndMatch<AggressiveDCEPass>(spirv, true);
}
TEST_F(AggressiveDCETest, DeadInterfacesV14) {
const std::string spirv = R"(
; CHECK: OpEntryPoint GLCompute %main "main" %live1 %live2
; CHECK-NOT: %dead
OpCapability Shader
OpMemoryModel Logical GLSL450
OpEntryPoint GLCompute %main "main" %live1 %dead1 %dead2 %live2
OpExecutionMode %main LocalSize 1 1 1
OpName %main "main"
OpName %live1 "live1"
OpName %live2 "live2"
OpName %dead1 "dead1"
OpName %dead2 "dead2"
%void = OpTypeVoid
%int = OpTypeInt 32 0
%int0 = OpConstant %int 0
%ptr_ssbo_int = OpTypePointer StorageBuffer %int
%live1 = OpVariable %ptr_ssbo_int StorageBuffer
%live2 = OpVariable %ptr_ssbo_int StorageBuffer
%dead1 = OpVariable %ptr_ssbo_int StorageBuffer
%dead2 = OpVariable %ptr_ssbo_int StorageBuffer
%void_fn = OpTypeFunction %void
%main = OpFunction %void None %void_fn
%entry = OpLabel
OpStore %live1 %int0
OpStore %live2 %int0
OpReturn
OpFunctionEnd
)";
SetTargetEnv(SPV_ENV_UNIVERSAL_1_4);
SinglePassRunAndMatch<AggressiveDCEPass>(spirv, true);
}
Add —preserve-bindings and —preserve-spec-constants (#2693) Add optimizer options to for preservation of spec constants and variable with binding decorations. They are to be preserved even if they are unused.
2019-07-10 18:12:19 +00:00
TEST_F(AggressiveDCETest, PreserveBindings) {
const std::string spirv = R"(
; CHECK: OpDecorate %unusedSampler DescriptorSet 0
; CHECK: OpDecorate %unusedSampler Binding 0
OpCapability Shader
%1 = OpExtInstImport "GLSL.std.450"
OpMemoryModel Logical GLSL450
OpEntryPoint Fragment %main "main"
OpExecutionMode %main OriginUpperLeft
OpSource GLSL 430
OpName %main "main"
OpName %unusedSampler "unusedSampler"
OpDecorate %unusedSampler DescriptorSet 0
OpDecorate %unusedSampler Binding 0
%void = OpTypeVoid
%5 = OpTypeFunction %void
%float = OpTypeFloat 32
%7 = OpTypeImage %float 2D 0 0 0 1 Unknown
%8 = OpTypeSampledImage %7
%_ptr_UniformConstant_8 = OpTypePointer UniformConstant %8
%unusedSampler = OpVariable %_ptr_UniformConstant_8 UniformConstant
%main = OpFunction %void None %5
%10 = OpLabel
OpReturn
OpFunctionEnd
)";
SetTargetEnv(SPV_ENV_UNIVERSAL_1_4);
OptimizerOptions()->preserve_bindings_ = true;
SinglePassRunAndMatch<AggressiveDCEPass>(spirv, true);
}
TEST_F(AggressiveDCETest, PreserveSpecConstants) {
const std::string spirv = R"(
; CHECK: OpName %specConstant "specConstant"
; CHECK: %specConstant = OpSpecConstant %int 0
OpCapability Shader
%1 = OpExtInstImport "GLSL.std.450"
OpMemoryModel Logical GLSL450
OpEntryPoint Fragment %main "main"
OpExecutionMode %main OriginUpperLeft
OpSource GLSL 430
OpName %main "main"
OpName %specConstant "specConstant"
OpDecorate %specConstant SpecId 0
%void = OpTypeVoid
%3 = OpTypeFunction %void
%int = OpTypeInt 32 1
%specConstant = OpSpecConstant %int 0
%main = OpFunction %void None %3
%5 = OpLabel
OpReturn
OpFunctionEnd
)";
SetTargetEnv(SPV_ENV_UNIVERSAL_1_4);
OptimizerOptions()->preserve_spec_constants_ = true;
SinglePassRunAndMatch<AggressiveDCEPass>(spirv, true);
}
Process OpDecorateId in ADCE (#2761) * Process OpDecorateId in ADCE When there is an OpDecorateId instruction that is live, the ids that is references must be kept live. This change adds them to the worklist. I've also updated a validator check to allow OpDecorateId to be able to apply to decoration groups. Fixes #1759. * Remove dead code.
2019-07-24 18:43:49 +00:00
TEST_F(AggressiveDCETest, LiveDecorateId) {
const std::string spirv = R"(OpCapability Shader
OpMemoryModel Logical GLSL450
OpEntryPoint GLCompute %1 "main" %2
OpExecutionMode %1 LocalSize 8 1 1
OpDecorate %2 DescriptorSet 0
OpDecorate %2 Binding 0
OpDecorateId %3 UniformId %uint_2
%void = OpTypeVoid
%uint = OpTypeInt 32 0
%uint_2 = OpConstant %uint 2
%_ptr_StorageBuffer_uint = OpTypePointer StorageBuffer %uint
%2 = OpVariable %_ptr_StorageBuffer_uint StorageBuffer
%8 = OpTypeFunction %void
%1 = OpFunction %void None %8
%9 = OpLabel
%3 = OpLoad %uint %2
OpStore %2 %3
OpReturn
OpFunctionEnd
)";
SetTargetEnv(SPV_ENV_UNIVERSAL_1_4);
OptimizerOptions()->preserve_spec_constants_ = true;
SinglePassRunAndCheck<AggressiveDCEPass>(spirv, spirv, true);
}
TEST_F(AggressiveDCETest, LiveDecorateIdOnGroup) {
const std::string spirv = R"(OpCapability Shader
OpMemoryModel Logical GLSL450
OpEntryPoint GLCompute %1 "main" %2
OpExecutionMode %1 LocalSize 8 1 1
OpDecorate %2 DescriptorSet 0
OpDecorate %2 Binding 0
OpDecorateId %3 UniformId %uint_2
%3 = OpDecorationGroup
OpGroupDecorate %3 %5
%void = OpTypeVoid
%uint = OpTypeInt 32 0
%uint_2 = OpConstant %uint 2
%_ptr_StorageBuffer_uint = OpTypePointer StorageBuffer %uint
%2 = OpVariable %_ptr_StorageBuffer_uint StorageBuffer
%9 = OpTypeFunction %void
%1 = OpFunction %void None %9
%10 = OpLabel
%5 = OpLoad %uint %2
OpStore %2 %5
OpReturn
OpFunctionEnd
)";
SetTargetEnv(SPV_ENV_UNIVERSAL_1_4);
OptimizerOptions()->preserve_spec_constants_ = true;
SinglePassRunAndCheck<AggressiveDCEPass>(spirv, spirv, true);
}
Add SPV_EXT_physical_storage_buffer to opt whitelists (#2779) This also fixes ADCE to not remove possibly needed OpTypeForwardPointer. The bug, its fix and the corresponding test have a circular dependency with the extension, so they are packaged together.
2019-08-08 13:45:59 +00:00
TEST_F(AggressiveDCETest, NoEliminateForwardPointer) {
// clang-format off
//
// #version 450
// #extension GL_EXT_buffer_reference : enable
//
// // forward reference
// layout(buffer_reference) buffer blockType;
//
// layout(buffer_reference, std430, buffer_reference_align = 16) buffer blockType {
// int x;
// blockType next;
// };
//
// layout(std430) buffer rootBlock {
// blockType root;
// } r;
//
// void main()
// {
// blockType b = r.root;
// b = b.next;
// b.x = 531;
// }
//
// clang-format on
const std::string predefs1 =
R"(OpCapability Shader
SPIRV-Tools support for SPIR-V 1.5 (#2865) * Ensure same enum values have consistent extension lists * val: fix checking of capabilities The operand for an OpCapability should only be checked for the extension or core version. The InstructionPass registers a capability, and all its implied sub-capabilities before actually checking the operand to an OpCapability. * Add basic support for SPIR-V 1.5 - Adds SPV_ENV_UNIVERSAL_1_5 - Command line tools default to spv1.5 environment - SPIR-V 1.5 incorporates several extensions. Now the disassembler prefers outputing the non-EXT or non-KHR names. This requires updates to many tests, to make strings match again. - Command line tests: Expect SPIR-V 1.5 by default * Test validation of SPIR-V 1.5 incorporated extensions Starting with 1.5, incorporated features no longer require the associated OpExtension instruction.
2019-09-13 18:59:02 +00:00
OpCapability PhysicalStorageBufferAddresses
Add SPV_EXT_physical_storage_buffer to opt whitelists (#2779) This also fixes ADCE to not remove possibly needed OpTypeForwardPointer. The bug, its fix and the corresponding test have a circular dependency with the extension, so they are packaged together.
2019-08-08 13:45:59 +00:00
OpExtension "SPV_EXT_physical_storage_buffer"
OpExtension "SPV_KHR_storage_buffer_storage_class"
%1 = OpExtInstImport "GLSL.std.450"
SPIRV-Tools support for SPIR-V 1.5 (#2865) * Ensure same enum values have consistent extension lists * val: fix checking of capabilities The operand for an OpCapability should only be checked for the extension or core version. The InstructionPass registers a capability, and all its implied sub-capabilities before actually checking the operand to an OpCapability. * Add basic support for SPIR-V 1.5 - Adds SPV_ENV_UNIVERSAL_1_5 - Command line tools default to spv1.5 environment - SPIR-V 1.5 incorporates several extensions. Now the disassembler prefers outputing the non-EXT or non-KHR names. This requires updates to many tests, to make strings match again. - Command line tests: Expect SPIR-V 1.5 by default * Test validation of SPIR-V 1.5 incorporated extensions Starting with 1.5, incorporated features no longer require the associated OpExtension instruction.
2019-09-13 18:59:02 +00:00
OpMemoryModel PhysicalStorageBuffer64 GLSL450
Add SPV_EXT_physical_storage_buffer to opt whitelists (#2779) This also fixes ADCE to not remove possibly needed OpTypeForwardPointer. The bug, its fix and the corresponding test have a circular dependency with the extension, so they are packaged together.
2019-08-08 13:45:59 +00:00
OpEntryPoint GLCompute %main "main"
OpExecutionMode %main LocalSize 1 1 1
OpSource GLSL 450
OpSourceExtension "GL_EXT_buffer_reference"
)";
const std::string names_before =
R"(OpName %main "main"
OpName %blockType "blockType"
OpMemberName %blockType 0 "x"
OpMemberName %blockType 1 "next"
OpName %b "b"
OpName %rootBlock "rootBlock"
OpMemberName %rootBlock 0 "root"
OpName %r "r"
OpMemberDecorate %blockType 0 Offset 0
OpMemberDecorate %blockType 1 Offset 8
OpDecorate %blockType Block
SPIRV-Tools support for SPIR-V 1.5 (#2865) * Ensure same enum values have consistent extension lists * val: fix checking of capabilities The operand for an OpCapability should only be checked for the extension or core version. The InstructionPass registers a capability, and all its implied sub-capabilities before actually checking the operand to an OpCapability. * Add basic support for SPIR-V 1.5 - Adds SPV_ENV_UNIVERSAL_1_5 - Command line tools default to spv1.5 environment - SPIR-V 1.5 incorporates several extensions. Now the disassembler prefers outputing the non-EXT or non-KHR names. This requires updates to many tests, to make strings match again. - Command line tests: Expect SPIR-V 1.5 by default * Test validation of SPIR-V 1.5 incorporated extensions Starting with 1.5, incorporated features no longer require the associated OpExtension instruction.
2019-09-13 18:59:02 +00:00
OpDecorate %b AliasedPointer
Add SPV_EXT_physical_storage_buffer to opt whitelists (#2779) This also fixes ADCE to not remove possibly needed OpTypeForwardPointer. The bug, its fix and the corresponding test have a circular dependency with the extension, so they are packaged together.
2019-08-08 13:45:59 +00:00
OpMemberDecorate %rootBlock 0 Offset 0
OpDecorate %rootBlock Block
OpDecorate %r DescriptorSet 0
OpDecorate %r Binding 0
)";
const std::string names_after =
R"(OpName %main "main"
OpName %blockType "blockType"
OpMemberName %blockType 0 "x"
OpMemberName %blockType 1 "next"
OpName %rootBlock "rootBlock"
OpMemberName %rootBlock 0 "root"
OpName %r "r"
OpMemberDecorate %blockType 0 Offset 0
OpMemberDecorate %blockType 1 Offset 8
OpDecorate %blockType Block
OpMemberDecorate %rootBlock 0 Offset 0
OpDecorate %rootBlock Block
OpDecorate %r DescriptorSet 0
OpDecorate %r Binding 0
)";
const std::string predefs2_before =
R"(%void = OpTypeVoid
%3 = OpTypeFunction %void
SPIRV-Tools support for SPIR-V 1.5 (#2865) * Ensure same enum values have consistent extension lists * val: fix checking of capabilities The operand for an OpCapability should only be checked for the extension or core version. The InstructionPass registers a capability, and all its implied sub-capabilities before actually checking the operand to an OpCapability. * Add basic support for SPIR-V 1.5 - Adds SPV_ENV_UNIVERSAL_1_5 - Command line tools default to spv1.5 environment - SPIR-V 1.5 incorporates several extensions. Now the disassembler prefers outputing the non-EXT or non-KHR names. This requires updates to many tests, to make strings match again. - Command line tests: Expect SPIR-V 1.5 by default * Test validation of SPIR-V 1.5 incorporated extensions Starting with 1.5, incorporated features no longer require the associated OpExtension instruction.
2019-09-13 18:59:02 +00:00
OpTypeForwardPointer %_ptr_PhysicalStorageBuffer_blockType PhysicalStorageBuffer
Add SPV_EXT_physical_storage_buffer to opt whitelists (#2779) This also fixes ADCE to not remove possibly needed OpTypeForwardPointer. The bug, its fix and the corresponding test have a circular dependency with the extension, so they are packaged together.
2019-08-08 13:45:59 +00:00
%int = OpTypeInt 32 1
SPIRV-Tools support for SPIR-V 1.5 (#2865) * Ensure same enum values have consistent extension lists * val: fix checking of capabilities The operand for an OpCapability should only be checked for the extension or core version. The InstructionPass registers a capability, and all its implied sub-capabilities before actually checking the operand to an OpCapability. * Add basic support for SPIR-V 1.5 - Adds SPV_ENV_UNIVERSAL_1_5 - Command line tools default to spv1.5 environment - SPIR-V 1.5 incorporates several extensions. Now the disassembler prefers outputing the non-EXT or non-KHR names. This requires updates to many tests, to make strings match again. - Command line tests: Expect SPIR-V 1.5 by default * Test validation of SPIR-V 1.5 incorporated extensions Starting with 1.5, incorporated features no longer require the associated OpExtension instruction.
2019-09-13 18:59:02 +00:00
%blockType = OpTypeStruct %int %_ptr_PhysicalStorageBuffer_blockType
%_ptr_PhysicalStorageBuffer_blockType = OpTypePointer PhysicalStorageBuffer %blockType
%_ptr_Function__ptr_PhysicalStorageBuffer_blockType = OpTypePointer Function %_ptr_PhysicalStorageBuffer_blockType
%rootBlock = OpTypeStruct %_ptr_PhysicalStorageBuffer_blockType
Add SPV_EXT_physical_storage_buffer to opt whitelists (#2779) This also fixes ADCE to not remove possibly needed OpTypeForwardPointer. The bug, its fix and the corresponding test have a circular dependency with the extension, so they are packaged together.
2019-08-08 13:45:59 +00:00
%_ptr_StorageBuffer_rootBlock = OpTypePointer StorageBuffer %rootBlock
%r = OpVariable %_ptr_StorageBuffer_rootBlock StorageBuffer
%int_0 = OpConstant %int 0
SPIRV-Tools support for SPIR-V 1.5 (#2865) * Ensure same enum values have consistent extension lists * val: fix checking of capabilities The operand for an OpCapability should only be checked for the extension or core version. The InstructionPass registers a capability, and all its implied sub-capabilities before actually checking the operand to an OpCapability. * Add basic support for SPIR-V 1.5 - Adds SPV_ENV_UNIVERSAL_1_5 - Command line tools default to spv1.5 environment - SPIR-V 1.5 incorporates several extensions. Now the disassembler prefers outputing the non-EXT or non-KHR names. This requires updates to many tests, to make strings match again. - Command line tests: Expect SPIR-V 1.5 by default * Test validation of SPIR-V 1.5 incorporated extensions Starting with 1.5, incorporated features no longer require the associated OpExtension instruction.
2019-09-13 18:59:02 +00:00
%_ptr_StorageBuffer__ptr_PhysicalStorageBuffer_blockType = OpTypePointer StorageBuffer %_ptr_PhysicalStorageBuffer_blockType
Add SPV_EXT_physical_storage_buffer to opt whitelists (#2779) This also fixes ADCE to not remove possibly needed OpTypeForwardPointer. The bug, its fix and the corresponding test have a circular dependency with the extension, so they are packaged together.
2019-08-08 13:45:59 +00:00
%int_1 = OpConstant %int 1
SPIRV-Tools support for SPIR-V 1.5 (#2865) * Ensure same enum values have consistent extension lists * val: fix checking of capabilities The operand for an OpCapability should only be checked for the extension or core version. The InstructionPass registers a capability, and all its implied sub-capabilities before actually checking the operand to an OpCapability. * Add basic support for SPIR-V 1.5 - Adds SPV_ENV_UNIVERSAL_1_5 - Command line tools default to spv1.5 environment - SPIR-V 1.5 incorporates several extensions. Now the disassembler prefers outputing the non-EXT or non-KHR names. This requires updates to many tests, to make strings match again. - Command line tests: Expect SPIR-V 1.5 by default * Test validation of SPIR-V 1.5 incorporated extensions Starting with 1.5, incorporated features no longer require the associated OpExtension instruction.
2019-09-13 18:59:02 +00:00
%_ptr_PhysicalStorageBuffer__ptr_PhysicalStorageBuffer_blockType = OpTypePointer PhysicalStorageBuffer %_ptr_PhysicalStorageBuffer_blockType
Add SPV_EXT_physical_storage_buffer to opt whitelists (#2779) This also fixes ADCE to not remove possibly needed OpTypeForwardPointer. The bug, its fix and the corresponding test have a circular dependency with the extension, so they are packaged together.
2019-08-08 13:45:59 +00:00
%int_531 = OpConstant %int 531
SPIRV-Tools support for SPIR-V 1.5 (#2865) * Ensure same enum values have consistent extension lists * val: fix checking of capabilities The operand for an OpCapability should only be checked for the extension or core version. The InstructionPass registers a capability, and all its implied sub-capabilities before actually checking the operand to an OpCapability. * Add basic support for SPIR-V 1.5 - Adds SPV_ENV_UNIVERSAL_1_5 - Command line tools default to spv1.5 environment - SPIR-V 1.5 incorporates several extensions. Now the disassembler prefers outputing the non-EXT or non-KHR names. This requires updates to many tests, to make strings match again. - Command line tests: Expect SPIR-V 1.5 by default * Test validation of SPIR-V 1.5 incorporated extensions Starting with 1.5, incorporated features no longer require the associated OpExtension instruction.
2019-09-13 18:59:02 +00:00
%_ptr_PhysicalStorageBuffer_int = OpTypePointer PhysicalStorageBuffer %int
Add SPV_EXT_physical_storage_buffer to opt whitelists (#2779) This also fixes ADCE to not remove possibly needed OpTypeForwardPointer. The bug, its fix and the corresponding test have a circular dependency with the extension, so they are packaged together.
2019-08-08 13:45:59 +00:00
)";
const std::string predefs2_after =
R"(%void = OpTypeVoid
%8 = OpTypeFunction %void
SPIRV-Tools support for SPIR-V 1.5 (#2865) * Ensure same enum values have consistent extension lists * val: fix checking of capabilities The operand for an OpCapability should only be checked for the extension or core version. The InstructionPass registers a capability, and all its implied sub-capabilities before actually checking the operand to an OpCapability. * Add basic support for SPIR-V 1.5 - Adds SPV_ENV_UNIVERSAL_1_5 - Command line tools default to spv1.5 environment - SPIR-V 1.5 incorporates several extensions. Now the disassembler prefers outputing the non-EXT or non-KHR names. This requires updates to many tests, to make strings match again. - Command line tests: Expect SPIR-V 1.5 by default * Test validation of SPIR-V 1.5 incorporated extensions Starting with 1.5, incorporated features no longer require the associated OpExtension instruction.
2019-09-13 18:59:02 +00:00
OpTypeForwardPointer %_ptr_PhysicalStorageBuffer_blockType PhysicalStorageBuffer
Add SPV_EXT_physical_storage_buffer to opt whitelists (#2779) This also fixes ADCE to not remove possibly needed OpTypeForwardPointer. The bug, its fix and the corresponding test have a circular dependency with the extension, so they are packaged together.
2019-08-08 13:45:59 +00:00
%int = OpTypeInt 32 1
SPIRV-Tools support for SPIR-V 1.5 (#2865) * Ensure same enum values have consistent extension lists * val: fix checking of capabilities The operand for an OpCapability should only be checked for the extension or core version. The InstructionPass registers a capability, and all its implied sub-capabilities before actually checking the operand to an OpCapability. * Add basic support for SPIR-V 1.5 - Adds SPV_ENV_UNIVERSAL_1_5 - Command line tools default to spv1.5 environment - SPIR-V 1.5 incorporates several extensions. Now the disassembler prefers outputing the non-EXT or non-KHR names. This requires updates to many tests, to make strings match again. - Command line tests: Expect SPIR-V 1.5 by default * Test validation of SPIR-V 1.5 incorporated extensions Starting with 1.5, incorporated features no longer require the associated OpExtension instruction.
2019-09-13 18:59:02 +00:00
%blockType = OpTypeStruct %int %_ptr_PhysicalStorageBuffer_blockType
%_ptr_PhysicalStorageBuffer_blockType = OpTypePointer PhysicalStorageBuffer %blockType
%rootBlock = OpTypeStruct %_ptr_PhysicalStorageBuffer_blockType
Add SPV_EXT_physical_storage_buffer to opt whitelists (#2779) This also fixes ADCE to not remove possibly needed OpTypeForwardPointer. The bug, its fix and the corresponding test have a circular dependency with the extension, so they are packaged together.
2019-08-08 13:45:59 +00:00
%_ptr_StorageBuffer_rootBlock = OpTypePointer StorageBuffer %rootBlock
%r = OpVariable %_ptr_StorageBuffer_rootBlock StorageBuffer
%int_0 = OpConstant %int 0
SPIRV-Tools support for SPIR-V 1.5 (#2865) * Ensure same enum values have consistent extension lists * val: fix checking of capabilities The operand for an OpCapability should only be checked for the extension or core version. The InstructionPass registers a capability, and all its implied sub-capabilities before actually checking the operand to an OpCapability. * Add basic support for SPIR-V 1.5 - Adds SPV_ENV_UNIVERSAL_1_5 - Command line tools default to spv1.5 environment - SPIR-V 1.5 incorporates several extensions. Now the disassembler prefers outputing the non-EXT or non-KHR names. This requires updates to many tests, to make strings match again. - Command line tests: Expect SPIR-V 1.5 by default * Test validation of SPIR-V 1.5 incorporated extensions Starting with 1.5, incorporated features no longer require the associated OpExtension instruction.
2019-09-13 18:59:02 +00:00
%_ptr_StorageBuffer__ptr_PhysicalStorageBuffer_blockType = OpTypePointer StorageBuffer %_ptr_PhysicalStorageBuffer_blockType
Add SPV_EXT_physical_storage_buffer to opt whitelists (#2779) This also fixes ADCE to not remove possibly needed OpTypeForwardPointer. The bug, its fix and the corresponding test have a circular dependency with the extension, so they are packaged together.
2019-08-08 13:45:59 +00:00
%int_1 = OpConstant %int 1
SPIRV-Tools support for SPIR-V 1.5 (#2865) * Ensure same enum values have consistent extension lists * val: fix checking of capabilities The operand for an OpCapability should only be checked for the extension or core version. The InstructionPass registers a capability, and all its implied sub-capabilities before actually checking the operand to an OpCapability. * Add basic support for SPIR-V 1.5 - Adds SPV_ENV_UNIVERSAL_1_5 - Command line tools default to spv1.5 environment - SPIR-V 1.5 incorporates several extensions. Now the disassembler prefers outputing the non-EXT or non-KHR names. This requires updates to many tests, to make strings match again. - Command line tests: Expect SPIR-V 1.5 by default * Test validation of SPIR-V 1.5 incorporated extensions Starting with 1.5, incorporated features no longer require the associated OpExtension instruction.
2019-09-13 18:59:02 +00:00
%_ptr_PhysicalStorageBuffer__ptr_PhysicalStorageBuffer_blockType = OpTypePointer PhysicalStorageBuffer %_ptr_PhysicalStorageBuffer_blockType
Add SPV_EXT_physical_storage_buffer to opt whitelists (#2779) This also fixes ADCE to not remove possibly needed OpTypeForwardPointer. The bug, its fix and the corresponding test have a circular dependency with the extension, so they are packaged together.
2019-08-08 13:45:59 +00:00
%int_531 = OpConstant %int 531
SPIRV-Tools support for SPIR-V 1.5 (#2865) * Ensure same enum values have consistent extension lists * val: fix checking of capabilities The operand for an OpCapability should only be checked for the extension or core version. The InstructionPass registers a capability, and all its implied sub-capabilities before actually checking the operand to an OpCapability. * Add basic support for SPIR-V 1.5 - Adds SPV_ENV_UNIVERSAL_1_5 - Command line tools default to spv1.5 environment - SPIR-V 1.5 incorporates several extensions. Now the disassembler prefers outputing the non-EXT or non-KHR names. This requires updates to many tests, to make strings match again. - Command line tests: Expect SPIR-V 1.5 by default * Test validation of SPIR-V 1.5 incorporated extensions Starting with 1.5, incorporated features no longer require the associated OpExtension instruction.
2019-09-13 18:59:02 +00:00
%_ptr_PhysicalStorageBuffer_int = OpTypePointer PhysicalStorageBuffer %int
Add SPV_EXT_physical_storage_buffer to opt whitelists (#2779) This also fixes ADCE to not remove possibly needed OpTypeForwardPointer. The bug, its fix and the corresponding test have a circular dependency with the extension, so they are packaged together.
2019-08-08 13:45:59 +00:00
)";
const std::string func_before =
R"(%main = OpFunction %void None %3
%5 = OpLabel
SPIRV-Tools support for SPIR-V 1.5 (#2865) * Ensure same enum values have consistent extension lists * val: fix checking of capabilities The operand for an OpCapability should only be checked for the extension or core version. The InstructionPass registers a capability, and all its implied sub-capabilities before actually checking the operand to an OpCapability. * Add basic support for SPIR-V 1.5 - Adds SPV_ENV_UNIVERSAL_1_5 - Command line tools default to spv1.5 environment - SPIR-V 1.5 incorporates several extensions. Now the disassembler prefers outputing the non-EXT or non-KHR names. This requires updates to many tests, to make strings match again. - Command line tests: Expect SPIR-V 1.5 by default * Test validation of SPIR-V 1.5 incorporated extensions Starting with 1.5, incorporated features no longer require the associated OpExtension instruction.
2019-09-13 18:59:02 +00:00
%b = OpVariable %_ptr_Function__ptr_PhysicalStorageBuffer_blockType Function
%16 = OpAccessChain %_ptr_StorageBuffer__ptr_PhysicalStorageBuffer_blockType %r %int_0
%17 = OpLoad %_ptr_PhysicalStorageBuffer_blockType %16
%21 = OpAccessChain %_ptr_PhysicalStorageBuffer__ptr_PhysicalStorageBuffer_blockType %17 %int_1
%22 = OpLoad %_ptr_PhysicalStorageBuffer_blockType %21 Aligned 8
Add SPV_EXT_physical_storage_buffer to opt whitelists (#2779) This also fixes ADCE to not remove possibly needed OpTypeForwardPointer. The bug, its fix and the corresponding test have a circular dependency with the extension, so they are packaged together.
2019-08-08 13:45:59 +00:00
OpStore %b %22
SPIRV-Tools support for SPIR-V 1.5 (#2865) * Ensure same enum values have consistent extension lists * val: fix checking of capabilities The operand for an OpCapability should only be checked for the extension or core version. The InstructionPass registers a capability, and all its implied sub-capabilities before actually checking the operand to an OpCapability. * Add basic support for SPIR-V 1.5 - Adds SPV_ENV_UNIVERSAL_1_5 - Command line tools default to spv1.5 environment - SPIR-V 1.5 incorporates several extensions. Now the disassembler prefers outputing the non-EXT or non-KHR names. This requires updates to many tests, to make strings match again. - Command line tests: Expect SPIR-V 1.5 by default * Test validation of SPIR-V 1.5 incorporated extensions Starting with 1.5, incorporated features no longer require the associated OpExtension instruction.
2019-09-13 18:59:02 +00:00
%26 = OpAccessChain %_ptr_PhysicalStorageBuffer_int %22 %int_0
Add SPV_EXT_physical_storage_buffer to opt whitelists (#2779) This also fixes ADCE to not remove possibly needed OpTypeForwardPointer. The bug, its fix and the corresponding test have a circular dependency with the extension, so they are packaged together.
2019-08-08 13:45:59 +00:00
OpStore %26 %int_531 Aligned 16
OpReturn
OpFunctionEnd
)";
const std::string func_after =
R"(%main = OpFunction %void None %8
%19 = OpLabel
SPIRV-Tools support for SPIR-V 1.5 (#2865) * Ensure same enum values have consistent extension lists * val: fix checking of capabilities The operand for an OpCapability should only be checked for the extension or core version. The InstructionPass registers a capability, and all its implied sub-capabilities before actually checking the operand to an OpCapability. * Add basic support for SPIR-V 1.5 - Adds SPV_ENV_UNIVERSAL_1_5 - Command line tools default to spv1.5 environment - SPIR-V 1.5 incorporates several extensions. Now the disassembler prefers outputing the non-EXT or non-KHR names. This requires updates to many tests, to make strings match again. - Command line tests: Expect SPIR-V 1.5 by default * Test validation of SPIR-V 1.5 incorporated extensions Starting with 1.5, incorporated features no longer require the associated OpExtension instruction.
2019-09-13 18:59:02 +00:00
%20 = OpAccessChain %_ptr_StorageBuffer__ptr_PhysicalStorageBuffer_blockType %r %int_0
%21 = OpLoad %_ptr_PhysicalStorageBuffer_blockType %20
%22 = OpAccessChain %_ptr_PhysicalStorageBuffer__ptr_PhysicalStorageBuffer_blockType %21 %int_1
%23 = OpLoad %_ptr_PhysicalStorageBuffer_blockType %22 Aligned 8
%24 = OpAccessChain %_ptr_PhysicalStorageBuffer_int %23 %int_0
Add SPV_EXT_physical_storage_buffer to opt whitelists (#2779) This also fixes ADCE to not remove possibly needed OpTypeForwardPointer. The bug, its fix and the corresponding test have a circular dependency with the extension, so they are packaged together.
2019-08-08 13:45:59 +00:00
OpStore %24 %int_531 Aligned 16
OpReturn
OpFunctionEnd
)";
SinglePassRunAndCheck<AggressiveDCEPass>(
predefs1 + names_before + predefs2_before + func_before,
predefs1 + names_after + predefs2_after + func_after, true, true);
}
Fix ADCE pass bug for mulitple entries (#3470) When there are multiple entries and the shader has a variable with WorkGroup storage class, those multiple entry functions store values to the variable. Since ADCE pass uses def-use chains to propagate the work list, some of instructions in the work list are not actually a part of the currently processed function. As a result, it adds instructions in other functions and put them in |live_insts_|. However, it does not have the control flow information for those instructions in other functions i.e., |block2headerBranch_| and |header2nextHeaderBranch_|. When it processes those instructions (they are added when it processes a different function), it skips handling them because they are already in |live_insts_| and does not check |block2headerBranch_| and |header2nextHeaderBranch_|, which results in skipping some branches. Even though those branches are live branches, it considers they are dead branches.
2020-06-29 17:08:48 +00:00
TEST_F(AggressiveDCETest, MultipleFunctionProcessIndependently) {
const std::string spirv = R"(
OpCapability Shader
OpMemoryModel Logical GLSL450
OpEntryPoint GLCompute %entryHistogram "entryHistogram" %gl_GlobalInvocationID %gl_LocalInvocationIndex
OpEntryPoint GLCompute %entryAverage "entryAverage" %gl_GlobalInvocationID %gl_LocalInvocationIndex
OpExecutionMode %entryHistogram LocalSize 16 16 1
OpExecutionMode %entryAverage LocalSize 256 1 1
OpSource HLSL 640
OpName %type_RWStructuredBuffer_uint "type.RWStructuredBuffer.uint"
OpName %uHistogram "uHistogram"
OpName %type_ACSBuffer_counter "type.ACSBuffer.counter"
OpMemberName %type_ACSBuffer_counter 0 "counter"
OpName %counter_var_uHistogram "counter.var.uHistogram"
OpName %sharedHistogram "sharedHistogram"
OpName %entryHistogram "entryHistogram"
OpName %param_var_id "param.var.id"
OpName %param_var_idx "param.var.idx"
OpName %entryAverage "entryAverage"
OpName %param_var_id_0 "param.var.id"
OpName %param_var_idx_0 "param.var.idx"
OpDecorate %gl_GlobalInvocationID BuiltIn GlobalInvocationId
OpDecorate %gl_LocalInvocationIndex BuiltIn LocalInvocationIndex
OpDecorate %uHistogram DescriptorSet 0
OpDecorate %uHistogram Binding 0
OpDecorate %counter_var_uHistogram DescriptorSet 0
OpDecorate %counter_var_uHistogram Binding 1
OpDecorate %_runtimearr_uint ArrayStride 4
OpMemberDecorate %type_RWStructuredBuffer_uint 0 Offset 0
OpDecorate %type_RWStructuredBuffer_uint BufferBlock
OpMemberDecorate %type_ACSBuffer_counter 0 Offset 0
OpDecorate %type_ACSBuffer_counter BufferBlock
%uint = OpTypeInt 32 0
%uint_0 = OpConstant %uint 0
%uint_1 = OpConstant %uint 1
%uint_2 = OpConstant %uint 2
%uint_4 = OpConstant %uint 4
%uint_8 = OpConstant %uint 8
%uint_16 = OpConstant %uint 16
%uint_32 = OpConstant %uint 32
%uint_64 = OpConstant %uint 64
%uint_128 = OpConstant %uint 128
%uint_256 = OpConstant %uint 256
%uint_512 = OpConstant %uint 512
%uint_254 = OpConstant %uint 254
%uint_255 = OpConstant %uint 255
%int = OpTypeInt 32 1
%int_0 = OpConstant %int 0
%_runtimearr_uint = OpTypeRuntimeArray %uint
%type_RWStructuredBuffer_uint = OpTypeStruct %_runtimearr_uint
%_ptr_Uniform_type_RWStructuredBuffer_uint = OpTypePointer Uniform %type_RWStructuredBuffer_uint
%type_ACSBuffer_counter = OpTypeStruct %int
%_ptr_Uniform_type_ACSBuffer_counter = OpTypePointer Uniform %type_ACSBuffer_counter
%_arr_uint_uint_256 = OpTypeArray %uint %uint_256
%_ptr_Workgroup__arr_uint_uint_256 = OpTypePointer Workgroup %_arr_uint_uint_256
%v3uint = OpTypeVector %uint 3
%_ptr_Input_v3uint = OpTypePointer Input %v3uint
%_ptr_Input_uint = OpTypePointer Input %uint
%void = OpTypeVoid
%49 = OpTypeFunction %void
%_ptr_Function_v3uint = OpTypePointer Function %v3uint
%_ptr_Function_uint = OpTypePointer Function %uint
%52 = OpTypeFunction %void %_ptr_Function_v3uint %_ptr_Function_uint
%_ptr_Workgroup_uint = OpTypePointer Workgroup %uint
%uint_264 = OpConstant %uint 264
%bool = OpTypeBool
%_ptr_Uniform_uint = OpTypePointer Uniform %uint
%uHistogram = OpVariable %_ptr_Uniform_type_RWStructuredBuffer_uint Uniform
%counter_var_uHistogram = OpVariable %_ptr_Uniform_type_ACSBuffer_counter Uniform
%sharedHistogram = OpVariable %_ptr_Workgroup__arr_uint_uint_256 Workgroup
%gl_GlobalInvocationID = OpVariable %_ptr_Input_v3uint Input
%gl_LocalInvocationIndex = OpVariable %_ptr_Input_uint Input
%entryHistogram = OpFunction %void None %49
%57 = OpLabel
%param_var_id = OpVariable %_ptr_Function_v3uint Function
%param_var_idx = OpVariable %_ptr_Function_uint Function
%58 = OpLoad %v3uint %gl_GlobalInvocationID
%59 = OpLoad %uint %gl_LocalInvocationIndex
%79 = OpAccessChain %_ptr_Workgroup_uint %sharedHistogram %int_0
%80 = OpAtomicIAdd %uint %79 %uint_1 %uint_0 %uint_1
OpReturn
OpFunctionEnd
%entryAverage = OpFunction %void None %49
%63 = OpLabel
%param_var_id_0 = OpVariable %_ptr_Function_v3uint Function
%param_var_idx_0 = OpVariable %_ptr_Function_uint Function
%64 = OpLoad %v3uint %gl_GlobalInvocationID
%65 = OpLoad %uint %gl_LocalInvocationIndex
OpStore %param_var_idx_0 %65
%83 = OpAccessChain %_ptr_Workgroup_uint %sharedHistogram %65
OpStore %83 %uint_0
; CHECK: [[ieq:%\w+]] = OpIEqual
; CHECK-NEXT: OpSelectionMerge [[merge:%\w+]]
; CHECK-NEXT: OpBranchConditional [[ieq]] [[not_elim:%\w+]] [[merge]]
; CHECK-NEXT: [[not_elim]] = OpLabel
; CHECK: [[merge]] = OpLabel
OpControlBarrier %uint_2 %uint_2 %uint_264
%85 = OpIEqual %bool %65 %uint_0
OpSelectionMerge %89 None
OpBranchConditional %85 %86 %89
%86 = OpLabel
%88 = OpAccessChain %_ptr_Workgroup_uint %sharedHistogram %65
OpStore %88 %uint_1
OpBranch %89
%89 = OpLabel
OpControlBarrier %uint_2 %uint_2 %uint_264
%91 = OpAccessChain %_ptr_Workgroup_uint %sharedHistogram %65
%92 = OpLoad %uint %91
%94 = OpAccessChain %_ptr_Uniform_uint %uHistogram %int_0 %65
OpStore %94 %92
OpReturn
OpFunctionEnd
)";
SetTargetEnv(SPV_ENV_UNIVERSAL_1_3);
SinglePassRunAndMatch<AggressiveDCEPass>(spirv, true);
}
Preserve OpenCL.DebugInfo.100 through elim-dead-code-aggressive (#3542) Essentially, it marks all DebugInfo instructions in functions (and their operands) as live. It treats DebugDeclare and DebugValue with Deref as loads and so marks Stores of their variables as live. It marks each DebugGlobalVariables as live except for its variable. After closure, it rechecks if the variable is live. If not, the DebugGlobalVariable instruction's variable operand is set to DebugInfoNone, per the DebugInfo spec.
2020-07-21 20:10:09 +00:00
TEST_F(AggressiveDCETest, DebugInfoKeepInFunctionElimStoreVar) {
// Verify that dead local variable tc and store eliminated but all
// in-function debuginfo kept.
//
// The SPIR-V has been inlined and local single store eliminated
//
// Texture2D g_tColor;
// SamplerState g_sAniso;
//
// struct PS_INPUT {
// float2 vTextureCoords : TEXCOORD2;
// };
//
// struct PS_OUTPUT {
// float4 vColor : SV_Target0;
// };
//
// PS_OUTPUT MainPs(PS_INPUT i) {
// PS_OUTPUT ps_output;
// float2 tc = i.vTextureCoords.xy;
// ps_output.vColor = g_tColor.Sample(g_sAniso, tc);
// return ps_output;
// }
const std::string text = R"(
OpCapability Shader
%1 = OpExtInstImport "OpenCL.DebugInfo.100"
OpMemoryModel Logical GLSL450
OpEntryPoint Fragment %MainPs "MainPs" %g_tColor %g_sAniso %in_var_TEXCOORD2 %out_var_SV_Target0
OpExecutionMode %MainPs OriginUpperLeft
%7 = OpString "foo.frag"
%8 = OpString "PS_OUTPUT"
%9 = OpString "float"
%10 = OpString "vColor"
%11 = OpString "PS_INPUT"
%12 = OpString "vTextureCoords"
%13 = OpString "@type.2d.image"
%14 = OpString "type.2d.image"
%15 = OpString "Texture2D.TemplateParam"
%16 = OpString "src.MainPs"
%17 = OpString "tc"
%18 = OpString "ps_output"
%19 = OpString "i"
%20 = OpString "@type.sampler"
%21 = OpString "type.sampler"
%22 = OpString "g_sAniso"
%23 = OpString "g_tColor"
OpName %type_2d_image "type.2d.image"
OpName %g_tColor "g_tColor"
OpName %type_sampler "type.sampler"
OpName %g_sAniso "g_sAniso"
OpName %in_var_TEXCOORD2 "in.var.TEXCOORD2"
OpName %out_var_SV_Target0 "out.var.SV_Target0"
OpName %MainPs "MainPs"
OpName %PS_INPUT "PS_INPUT"
OpMemberName %PS_INPUT 0 "vTextureCoords"
OpName %param_var_i "param.var.i"
OpName %PS_OUTPUT "PS_OUTPUT"
OpMemberName %PS_OUTPUT 0 "vColor"
OpName %type_sampled_image "type.sampled.image"
OpDecorate %in_var_TEXCOORD2 Location 0
OpDecorate %out_var_SV_Target0 Location 0
OpDecorate %g_tColor DescriptorSet 0
OpDecorate %g_tColor Binding 0
OpDecorate %g_sAniso DescriptorSet 0
OpDecorate %g_sAniso Binding 1
%int = OpTypeInt 32 1
%int_0 = OpConstant %int 0
%uint = OpTypeInt 32 0
%uint_32 = OpConstant %uint 32
%float = OpTypeFloat 32
%type_2d_image = OpTypeImage %float 2D 2 0 0 1 Unknown
%_ptr_UniformConstant_type_2d_image = OpTypePointer UniformConstant %type_2d_image
%type_sampler = OpTypeSampler
%_ptr_UniformConstant_type_sampler = OpTypePointer UniformConstant %type_sampler
%v2float = OpTypeVector %float 2
%_ptr_Input_v2float = OpTypePointer Input %v2float
%v4float = OpTypeVector %float 4
%_ptr_Output_v4float = OpTypePointer Output %v4float
%void = OpTypeVoid
%uint_128 = OpConstant %uint 128
%uint_0 = OpConstant %uint 0
%uint_64 = OpConstant %uint 64
%45 = OpTypeFunction %void
%PS_INPUT = OpTypeStruct %v2float
%_ptr_Function_PS_INPUT = OpTypePointer Function %PS_INPUT
%PS_OUTPUT = OpTypeStruct %v4float
%47 = OpTypeFunction %PS_OUTPUT %_ptr_Function_PS_INPUT
%_ptr_Function_PS_OUTPUT = OpTypePointer Function %PS_OUTPUT
%_ptr_Function_v2float = OpTypePointer Function %v2float
%type_sampled_image = OpTypeSampledImage %type_2d_image
%_ptr_Function_v4float = OpTypePointer Function %v4float
%g_tColor = OpVariable %_ptr_UniformConstant_type_2d_image UniformConstant
%g_sAniso = OpVariable %_ptr_UniformConstant_type_sampler UniformConstant
%in_var_TEXCOORD2 = OpVariable %_ptr_Input_v2float Input
%out_var_SV_Target0 = OpVariable %_ptr_Output_v4float Output
%51 = OpExtInst %void %1 DebugInfoNone
%52 = OpExtInst %void %1 DebugExpression
%53 = OpExtInst %void %1 DebugOperation Deref
%54 = OpExtInst %void %1 DebugExpression %53
%55 = OpExtInst %void %1 DebugSource %7
%56 = OpExtInst %void %1 DebugCompilationUnit 1 4 %55 HLSL
%57 = OpExtInst %void %1 DebugTypeComposite %8 Structure %55 10 1 %56 %8 %uint_128 FlagIsProtected|FlagIsPrivate %58
%59 = OpExtInst %void %1 DebugTypeBasic %9 %uint_32 Float
%60 = OpExtInst %void %1 DebugTypeVector %59 4
%58 = OpExtInst %void %1 DebugTypeMember %10 %60 %55 12 5 %57 %uint_0 %uint_128 FlagIsProtected|FlagIsPrivate
%61 = OpExtInst %void %1 DebugTypeComposite %11 Structure %55 5 1 %56 %11 %uint_64 FlagIsProtected|FlagIsPrivate %62
%63 = OpExtInst %void %1 DebugTypeVector %59 2
%62 = OpExtInst %void %1 DebugTypeMember %12 %63 %55 7 5 %61 %uint_0 %uint_64 FlagIsProtected|FlagIsPrivate
%64 = OpExtInst %void %1 DebugTypeComposite %13 Class %55 0 0 %56 %14 %51 FlagIsProtected|FlagIsPrivate
%65 = OpExtInst %void %1 DebugTypeTemplateParameter %15 %59 %51 %55 0 0
%66 = OpExtInst %void %1 DebugTypeTemplate %64 %65
%67 = OpExtInst %void %1 DebugTypeFunction FlagIsProtected|FlagIsPrivate %57 %61
%68 = OpExtInst %void %1 DebugFunction %16 %67 %55 15 1 %56 %16 FlagIsProtected|FlagIsPrivate 16 %51
%69 = OpExtInst %void %1 DebugLexicalBlock %55 16 1 %68
%70 = OpExtInst %void %1 DebugLocalVariable %17 %63 %55 19 12 %69 FlagIsLocal
%71 = OpExtInst %void %1 DebugLocalVariable %18 %57 %55 17 15 %69 FlagIsLocal
%72 = OpExtInst %void %1 DebugLocalVariable %19 %61 %55 15 29 %68 FlagIsLocal 1
%73 = OpExtInst %void %1 DebugTypeComposite %20 Structure %55 0 0 %56 %21 %51 FlagIsProtected|FlagIsPrivate
%74 = OpExtInst %void %1 DebugGlobalVariable %22 %73 %55 3 14 %56 %22 %g_sAniso FlagIsDefinition
%75 = OpExtInst %void %1 DebugGlobalVariable %23 %64 %55 1 11 %56 %23 %g_tColor FlagIsDefinition
%MainPs = OpFunction %void None %45
%76 = OpLabel
%107 = OpExtInst %void %1 DebugScope %69
;CHECK: {{%\w+}} = OpExtInst %void %1 DebugScope %69
%78 = OpVariable %_ptr_Function_PS_OUTPUT Function
%79 = OpVariable %_ptr_Function_v2float Function
%108 = OpExtInst %void %1 DebugNoScope
;CHECK: {{%\w+}} = OpExtInst %void %1 DebugNoScope
%81 = OpVariable %_ptr_Function_PS_OUTPUT Function
%param_var_i = OpVariable %_ptr_Function_PS_INPUT Function
%82 = OpLoad %v2float %in_var_TEXCOORD2
%83 = OpCompositeConstruct %PS_INPUT %82
OpStore %param_var_i %83
%109 = OpExtInst %void %1 DebugScope %68
%85 = OpExtInst %void %1 DebugDeclare %72 %param_var_i %52
%110 = OpExtInst %void %1 DebugScope %69
%87 = OpExtInst %void %1 DebugDeclare %71 %78 %52
;CHECK: {{%\w+}} = OpExtInst %void %1 DebugScope %68
;CHECK: {{%\w+}} = OpExtInst %void %1 DebugDeclare %72 %param_var_i %52
;CHECK: {{%\w+}} = OpExtInst %void %1 DebugScope %69
;CHECK: {{%\w+}} = OpExtInst %void %1 DebugDeclare %71 %78 %52
OpLine %7 19 17
%88 = OpAccessChain %_ptr_Function_v2float %param_var_i %int_0
%89 = OpLoad %v2float %88
OpLine %7 19 12
OpStore %79 %89
;CHECK-NOT: OpStore %79 %89
OpLine %7 19 12
%106 = OpExtInst %void %1 DebugValue %70 %89 %52
;CHECK: {{%\w+}} = OpExtInst %void %1 DebugValue %70 %89 %52
OpLine %7 20 26
%91 = OpLoad %type_2d_image %g_tColor
OpLine %7 20 46
%92 = OpLoad %type_sampler %g_sAniso
OpLine %7 20 26
%94 = OpSampledImage %type_sampled_image %91 %92
%95 = OpImageSampleImplicitLod %v4float %94 %89 None
OpLine %7 20 5
%96 = OpAccessChain %_ptr_Function_v4float %78 %int_0
OpStore %96 %95
OpLine %7 21 12
%97 = OpLoad %PS_OUTPUT %78
OpLine %7 21 5
OpStore %81 %97
%111 = OpExtInst %void %1 DebugNoScope
;CHECK: {{%\w+}} = OpExtInst %void %1 DebugNoScope
%100 = OpCompositeExtract %v4float %97 0
OpStore %out_var_SV_Target0 %100
OpReturn
OpFunctionEnd
)";
SetTargetEnv(SPV_ENV_VULKAN_1_2);
SetAssembleOptions(SPV_TEXT_TO_BINARY_OPTION_PRESERVE_NUMERIC_IDS);
SinglePassRunAndMatch<AggressiveDCEPass>(text, true);
spirv-opt: Switch from Vulkan.DebugInfo to Shader.DebugInfo (#4493) Includes: - Shift to use of spirv-header extinst.nonsemantic.shader grammar.json - Remove extinst.nonsemantic.vulkan.debuginfo.100.grammar.json - Enable all optimizations for Shader.DebugInfo Also fixes scalar replacement to only insert DebugValue after all OpVariables. This is not necessary for OpenCL.DebugInfo, but it is for Shader.DebugInfo. Likewise, fixes Private-to-Local to insert DebugDeclare after all OpVariables. Also fixes inlining to handle FunctionDefinition which can show up after first block if early return processing happens. Co-authored-by: baldurk <baldurk@baldurk.org>
2021-09-15 18:38:53 +00:00
}
TEST_F(AggressiveDCETest, ShaderDebugInfoKeepInFunctionElimStoreVar) {
// Verify that dead local variable tc and store eliminated but all
// in-function NonSemantic Shader debuginfo kept.
const std::string text = R"(
OpCapability Shader
OpExtension "SPV_KHR_non_semantic_info"
%1 = OpExtInstImport "NonSemantic.Shader.DebugInfo.100"
OpMemoryModel Logical GLSL450
OpEntryPoint Fragment %MainPs "MainPs" %g_tColor %g_sAniso %in_var_TEXCOORD2 %out_var_SV_Target0
OpExecutionMode %MainPs OriginUpperLeft
%7 = OpString "foo.frag"
%8 = OpString "PS_OUTPUT"
%9 = OpString "float"
%10 = OpString "vColor"
%11 = OpString "PS_INPUT"
%12 = OpString "vTextureCoords"
%13 = OpString "@type.2d.image"
%14 = OpString "type.2d.image"
%15 = OpString "Texture2D.TemplateParam"
%16 = OpString "src.MainPs"
%17 = OpString "tc"
%18 = OpString "ps_output"
%19 = OpString "i"
%20 = OpString "@type.sampler"
%21 = OpString "type.sampler"
%22 = OpString "g_sAniso"
%23 = OpString "g_tColor"
OpName %type_2d_image "type.2d.image"
OpName %g_tColor "g_tColor"
OpName %type_sampler "type.sampler"
OpName %g_sAniso "g_sAniso"
OpName %in_var_TEXCOORD2 "in.var.TEXCOORD2"
OpName %out_var_SV_Target0 "out.var.SV_Target0"
OpName %MainPs "MainPs"
OpName %PS_INPUT "PS_INPUT"
OpMemberName %PS_INPUT 0 "vTextureCoords"
OpName %param_var_i "param.var.i"
OpName %PS_OUTPUT "PS_OUTPUT"
OpMemberName %PS_OUTPUT 0 "vColor"
OpName %type_sampled_image "type.sampled.image"
OpDecorate %in_var_TEXCOORD2 Location 0
OpDecorate %out_var_SV_Target0 Location 0
OpDecorate %g_tColor DescriptorSet 0
OpDecorate %g_tColor Binding 0
OpDecorate %g_sAniso DescriptorSet 0
OpDecorate %g_sAniso Binding 1
%int = OpTypeInt 32 1
%int_0 = OpConstant %int 0
%uint = OpTypeInt 32 0
%uint_32 = OpConstant %uint 32
%float = OpTypeFloat 32
%type_2d_image = OpTypeImage %float 2D 2 0 0 1 Unknown
%_ptr_UniformConstant_type_2d_image = OpTypePointer UniformConstant %type_2d_image
%type_sampler = OpTypeSampler
%_ptr_UniformConstant_type_sampler = OpTypePointer UniformConstant %type_sampler
%v2float = OpTypeVector %float 2
%_ptr_Input_v2float = OpTypePointer Input %v2float
%v4float = OpTypeVector %float 4
%_ptr_Output_v4float = OpTypePointer Output %v4float
%void = OpTypeVoid
%uint_128 = OpConstant %uint 128
%uint_0 = OpConstant %uint 0
%uint_1 = OpConstant %uint 1
%uint_2 = OpConstant %uint 2
%uint_3 = OpConstant %uint 3
%uint_4 = OpConstant %uint 4
%uint_5 = OpConstant %uint 5
%uint_7 = OpConstant %uint 7
%uint_8 = OpConstant %uint 8
%uint_10 = OpConstant %uint 10
%uint_11 = OpConstant %uint 11
%uint_12 = OpConstant %uint 12
%uint_14 = OpConstant %uint 14
%uint_15 = OpConstant %uint 15
%uint_16 = OpConstant %uint 16
%uint_17 = OpConstant %uint 17
%uint_19 = OpConstant %uint 19
Handle NonSemantic.Shader Debug[No]Line (#4530) Debug[No]Line are tracked and optimized using the same mechanism that tracks and optimizes Op[No]Line. Also: - Fix missing DebugScope at top of block. - Allow scalar replacement of access chain in DebugDeclare
2021-09-24 14:56:08 +00:00
%uint_20 = OpConstant %uint 20
%uint_21 = OpConstant %uint 21
%uint_25 = OpConstant %uint 25
spirv-opt: Switch from Vulkan.DebugInfo to Shader.DebugInfo (#4493) Includes: - Shift to use of spirv-header extinst.nonsemantic.shader grammar.json - Remove extinst.nonsemantic.vulkan.debuginfo.100.grammar.json - Enable all optimizations for Shader.DebugInfo Also fixes scalar replacement to only insert DebugValue after all OpVariables. This is not necessary for OpenCL.DebugInfo, but it is for Shader.DebugInfo. Likewise, fixes Private-to-Local to insert DebugDeclare after all OpVariables. Also fixes inlining to handle FunctionDefinition which can show up after first block if early return processing happens. Co-authored-by: baldurk <baldurk@baldurk.org>
2021-09-15 18:38:53 +00:00
%uint_29 = OpConstant %uint 29
Handle NonSemantic.Shader Debug[No]Line (#4530) Debug[No]Line are tracked and optimized using the same mechanism that tracks and optimizes Op[No]Line. Also: - Fix missing DebugScope at top of block. - Allow scalar replacement of access chain in DebugDeclare
2021-09-24 14:56:08 +00:00
%uint_30 = OpConstant %uint 30
%uint_35 = OpConstant %uint 35
%uint_41 = OpConstant %uint 41
%uint_48 = OpConstant %uint 48
%uint_53 = OpConstant %uint 53
spirv-opt: Switch from Vulkan.DebugInfo to Shader.DebugInfo (#4493) Includes: - Shift to use of spirv-header extinst.nonsemantic.shader grammar.json - Remove extinst.nonsemantic.vulkan.debuginfo.100.grammar.json - Enable all optimizations for Shader.DebugInfo Also fixes scalar replacement to only insert DebugValue after all OpVariables. This is not necessary for OpenCL.DebugInfo, but it is for Shader.DebugInfo. Likewise, fixes Private-to-Local to insert DebugDeclare after all OpVariables. Also fixes inlining to handle FunctionDefinition which can show up after first block if early return processing happens. Co-authored-by: baldurk <baldurk@baldurk.org>
2021-09-15 18:38:53 +00:00
%uint_64 = OpConstant %uint 64
%45 = OpTypeFunction %void
%PS_INPUT = OpTypeStruct %v2float
%_ptr_Function_PS_INPUT = OpTypePointer Function %PS_INPUT
%PS_OUTPUT = OpTypeStruct %v4float
%47 = OpTypeFunction %PS_OUTPUT %_ptr_Function_PS_INPUT
%_ptr_Function_PS_OUTPUT = OpTypePointer Function %PS_OUTPUT
%_ptr_Function_v2float = OpTypePointer Function %v2float
%type_sampled_image = OpTypeSampledImage %type_2d_image
%_ptr_Function_v4float = OpTypePointer Function %v4float
%g_tColor = OpVariable %_ptr_UniformConstant_type_2d_image UniformConstant
%g_sAniso = OpVariable %_ptr_UniformConstant_type_sampler UniformConstant
%in_var_TEXCOORD2 = OpVariable %_ptr_Input_v2float Input
%out_var_SV_Target0 = OpVariable %_ptr_Output_v4float Output
%51 = OpExtInst %void %1 DebugInfoNone
%52 = OpExtInst %void %1 DebugExpression
%53 = OpExtInst %void %1 DebugOperation %uint_0
%54 = OpExtInst %void %1 DebugExpression %53
%55 = OpExtInst %void %1 DebugSource %7
%56 = OpExtInst %void %1 DebugCompilationUnit %uint_1 %uint_4 %55 %uint_5
%59 = OpExtInst %void %1 DebugTypeBasic %9 %uint_32 %uint_3 %uint_0
%60 = OpExtInst %void %1 DebugTypeVector %59 %uint_4
%58 = OpExtInst %void %1 DebugTypeMember %10 %60 %55 %uint_12 %uint_5 %uint_0 %uint_128 %uint_3
%57 = OpExtInst %void %1 DebugTypeComposite %8 %uint_1 %55 %uint_10 %uint_1 %56 %8 %uint_128 %uint_3 %58
%63 = OpExtInst %void %1 DebugTypeVector %59 %uint_2
%62 = OpExtInst %void %1 DebugTypeMember %12 %63 %55 %uint_7 %uint_5 %uint_0 %uint_64 %uint_3
%61 = OpExtInst %void %1 DebugTypeComposite %11 %uint_1 %55 %uint_5 %uint_1 %56 %11 %uint_64 %uint_3 %62
%64 = OpExtInst %void %1 DebugTypeComposite %13 %uint_0 %55 %uint_0 %uint_0 %56 %14 %51 %uint_3
%67 = OpExtInst %void %1 DebugTypeFunction %uint_3 %57 %61
%68 = OpExtInst %void %1 DebugFunction %16 %67 %55 %uint_15 %uint_1 %56 %16 %uint_3 %uint_16
%69 = OpExtInst %void %1 DebugLexicalBlock %55 %uint_16 %uint_1 %68
%70 = OpExtInst %void %1 DebugLocalVariable %17 %63 %55 %uint_19 %uint_12 %69 %uint_4
%71 = OpExtInst %void %1 DebugLocalVariable %18 %57 %55 %uint_17 %uint_15 %69 %uint_4
%72 = OpExtInst %void %1 DebugLocalVariable %19 %61 %55 %uint_15 %uint_29 %68 %uint_4 %uint_1
%73 = OpExtInst %void %1 DebugTypeComposite %20 %uint_1 %55 %uint_0 %uint_0 %56 %21 %51 %uint_3
%74 = OpExtInst %void %1 DebugGlobalVariable %22 %73 %55 %uint_3 %uint_14 %56 %22 %g_sAniso %uint_8
%75 = OpExtInst %void %1 DebugGlobalVariable %23 %64 %55 %uint_1 %uint_11 %56 %23 %g_tColor %uint_8
%MainPs = OpFunction %void None %45
%76 = OpLabel
%78 = OpVariable %_ptr_Function_PS_OUTPUT Function
%79 = OpVariable %_ptr_Function_v2float Function
%81 = OpVariable %_ptr_Function_PS_OUTPUT Function
%param_var_i = OpVariable %_ptr_Function_PS_INPUT Function
%82 = OpLoad %v2float %in_var_TEXCOORD2
%83 = OpCompositeConstruct %PS_INPUT %82
OpStore %param_var_i %83
%112 = OpExtInst %void %1 DebugFunctionDefinition %68 %MainPs
%109 = OpExtInst %void %1 DebugScope %68
%85 = OpExtInst %void %1 DebugDeclare %72 %param_var_i %52
%110 = OpExtInst %void %1 DebugScope %69
%87 = OpExtInst %void %1 DebugDeclare %71 %78 %52
;CHECK: {{%\w+}} = OpExtInst %void %1 DebugFunctionDefinition %68 %MainPs
;CHECK: {{%\w+}} = OpExtInst %void %1 DebugScope %68
;CHECK: {{%\w+}} = OpExtInst %void %1 DebugDeclare %72 %param_var_i %52
;CHECK: {{%\w+}} = OpExtInst %void %1 DebugScope %69
;CHECK: {{%\w+}} = OpExtInst %void %1 DebugDeclare %71 %78 %52
Handle NonSemantic.Shader Debug[No]Line (#4530) Debug[No]Line are tracked and optimized using the same mechanism that tracks and optimizes Op[No]Line. Also: - Fix missing DebugScope at top of block. - Allow scalar replacement of access chain in DebugDeclare
2021-09-24 14:56:08 +00:00
%300 = OpExtInst %void %1 DebugLine %55 %uint_19 %uint_19 %uint_17 %uint_30
;CHECK: {{%\w+}} = OpExtInst %void %1 DebugLine %55 %uint_19 %uint_19 %uint_17 %uint_30
spirv-opt: Switch from Vulkan.DebugInfo to Shader.DebugInfo (#4493) Includes: - Shift to use of spirv-header extinst.nonsemantic.shader grammar.json - Remove extinst.nonsemantic.vulkan.debuginfo.100.grammar.json - Enable all optimizations for Shader.DebugInfo Also fixes scalar replacement to only insert DebugValue after all OpVariables. This is not necessary for OpenCL.DebugInfo, but it is for Shader.DebugInfo. Likewise, fixes Private-to-Local to insert DebugDeclare after all OpVariables. Also fixes inlining to handle FunctionDefinition which can show up after first block if early return processing happens. Co-authored-by: baldurk <baldurk@baldurk.org>
2021-09-15 18:38:53 +00:00
%88 = OpAccessChain %_ptr_Function_v2float %param_var_i %int_0
%89 = OpLoad %v2float %88
Handle NonSemantic.Shader Debug[No]Line (#4530) Debug[No]Line are tracked and optimized using the same mechanism that tracks and optimizes Op[No]Line. Also: - Fix missing DebugScope at top of block. - Allow scalar replacement of access chain in DebugDeclare
2021-09-24 14:56:08 +00:00
%301 = OpExtInst %void %1 DebugLine %55 %uint_19 %uint_19 %uint_12 %uint_35
spirv-opt: Switch from Vulkan.DebugInfo to Shader.DebugInfo (#4493) Includes: - Shift to use of spirv-header extinst.nonsemantic.shader grammar.json - Remove extinst.nonsemantic.vulkan.debuginfo.100.grammar.json - Enable all optimizations for Shader.DebugInfo Also fixes scalar replacement to only insert DebugValue after all OpVariables. This is not necessary for OpenCL.DebugInfo, but it is for Shader.DebugInfo. Likewise, fixes Private-to-Local to insert DebugDeclare after all OpVariables. Also fixes inlining to handle FunctionDefinition which can show up after first block if early return processing happens. Co-authored-by: baldurk <baldurk@baldurk.org>
2021-09-15 18:38:53 +00:00
OpStore %79 %89
;CHECK-NOT: OpStore %79 %89
Handle NonSemantic.Shader Debug[No]Line (#4530) Debug[No]Line are tracked and optimized using the same mechanism that tracks and optimizes Op[No]Line. Also: - Fix missing DebugScope at top of block. - Allow scalar replacement of access chain in DebugDeclare
2021-09-24 14:56:08 +00:00
%302 = OpExtInst %void %1 DebugLine %55 %uint_19 %uint_19 %uint_12 %uint_35
;CHECK: {{%\w+}} = OpExtInst %void %1 DebugLine %55 %uint_19 %uint_19 %uint_12 %uint_35
spirv-opt: Switch from Vulkan.DebugInfo to Shader.DebugInfo (#4493) Includes: - Shift to use of spirv-header extinst.nonsemantic.shader grammar.json - Remove extinst.nonsemantic.vulkan.debuginfo.100.grammar.json - Enable all optimizations for Shader.DebugInfo Also fixes scalar replacement to only insert DebugValue after all OpVariables. This is not necessary for OpenCL.DebugInfo, but it is for Shader.DebugInfo. Likewise, fixes Private-to-Local to insert DebugDeclare after all OpVariables. Also fixes inlining to handle FunctionDefinition which can show up after first block if early return processing happens. Co-authored-by: baldurk <baldurk@baldurk.org>
2021-09-15 18:38:53 +00:00
%106 = OpExtInst %void %1 DebugValue %70 %89 %52
;CHECK: {{%\w+}} = OpExtInst %void %1 DebugValue %70 %89 %52
Handle NonSemantic.Shader Debug[No]Line (#4530) Debug[No]Line are tracked and optimized using the same mechanism that tracks and optimizes Op[No]Line. Also: - Fix missing DebugScope at top of block. - Allow scalar replacement of access chain in DebugDeclare
2021-09-24 14:56:08 +00:00
%303 = OpExtInst %void %1 DebugLine %55 %uint_20 %uint_20 %uint_25 %uint_32
spirv-opt: Switch from Vulkan.DebugInfo to Shader.DebugInfo (#4493) Includes: - Shift to use of spirv-header extinst.nonsemantic.shader grammar.json - Remove extinst.nonsemantic.vulkan.debuginfo.100.grammar.json - Enable all optimizations for Shader.DebugInfo Also fixes scalar replacement to only insert DebugValue after all OpVariables. This is not necessary for OpenCL.DebugInfo, but it is for Shader.DebugInfo. Likewise, fixes Private-to-Local to insert DebugDeclare after all OpVariables. Also fixes inlining to handle FunctionDefinition which can show up after first block if early return processing happens. Co-authored-by: baldurk <baldurk@baldurk.org>
2021-09-15 18:38:53 +00:00
%91 = OpLoad %type_2d_image %g_tColor
Handle NonSemantic.Shader Debug[No]Line (#4530) Debug[No]Line are tracked and optimized using the same mechanism that tracks and optimizes Op[No]Line. Also: - Fix missing DebugScope at top of block. - Allow scalar replacement of access chain in DebugDeclare
2021-09-24 14:56:08 +00:00
%304 = OpExtInst %void %1 DebugLine %55 %uint_20 %uint_20 %uint_41 %uint_48
spirv-opt: Switch from Vulkan.DebugInfo to Shader.DebugInfo (#4493) Includes: - Shift to use of spirv-header extinst.nonsemantic.shader grammar.json - Remove extinst.nonsemantic.vulkan.debuginfo.100.grammar.json - Enable all optimizations for Shader.DebugInfo Also fixes scalar replacement to only insert DebugValue after all OpVariables. This is not necessary for OpenCL.DebugInfo, but it is for Shader.DebugInfo. Likewise, fixes Private-to-Local to insert DebugDeclare after all OpVariables. Also fixes inlining to handle FunctionDefinition which can show up after first block if early return processing happens. Co-authored-by: baldurk <baldurk@baldurk.org>
2021-09-15 18:38:53 +00:00
%92 = OpLoad %type_sampler %g_sAniso
Handle NonSemantic.Shader Debug[No]Line (#4530) Debug[No]Line are tracked and optimized using the same mechanism that tracks and optimizes Op[No]Line. Also: - Fix missing DebugScope at top of block. - Allow scalar replacement of access chain in DebugDeclare
2021-09-24 14:56:08 +00:00
%305 = OpExtInst %void %1 DebugLine %55 %uint_20 %uint_20 %uint_25 %uint_53
spirv-opt: Switch from Vulkan.DebugInfo to Shader.DebugInfo (#4493) Includes: - Shift to use of spirv-header extinst.nonsemantic.shader grammar.json - Remove extinst.nonsemantic.vulkan.debuginfo.100.grammar.json - Enable all optimizations for Shader.DebugInfo Also fixes scalar replacement to only insert DebugValue after all OpVariables. This is not necessary for OpenCL.DebugInfo, but it is for Shader.DebugInfo. Likewise, fixes Private-to-Local to insert DebugDeclare after all OpVariables. Also fixes inlining to handle FunctionDefinition which can show up after first block if early return processing happens. Co-authored-by: baldurk <baldurk@baldurk.org>
2021-09-15 18:38:53 +00:00
%94 = OpSampledImage %type_sampled_image %91 %92
%95 = OpImageSampleImplicitLod %v4float %94 %89 None
Handle NonSemantic.Shader Debug[No]Line (#4530) Debug[No]Line are tracked and optimized using the same mechanism that tracks and optimizes Op[No]Line. Also: - Fix missing DebugScope at top of block. - Allow scalar replacement of access chain in DebugDeclare
2021-09-24 14:56:08 +00:00
%306 = OpExtInst %void %1 DebugLine %55 %uint_20 %uint_20 %uint_5 %uint_53
spirv-opt: Switch from Vulkan.DebugInfo to Shader.DebugInfo (#4493) Includes: - Shift to use of spirv-header extinst.nonsemantic.shader grammar.json - Remove extinst.nonsemantic.vulkan.debuginfo.100.grammar.json - Enable all optimizations for Shader.DebugInfo Also fixes scalar replacement to only insert DebugValue after all OpVariables. This is not necessary for OpenCL.DebugInfo, but it is for Shader.DebugInfo. Likewise, fixes Private-to-Local to insert DebugDeclare after all OpVariables. Also fixes inlining to handle FunctionDefinition which can show up after first block if early return processing happens. Co-authored-by: baldurk <baldurk@baldurk.org>
2021-09-15 18:38:53 +00:00
%96 = OpAccessChain %_ptr_Function_v4float %78 %int_0
OpStore %96 %95
Handle NonSemantic.Shader Debug[No]Line (#4530) Debug[No]Line are tracked and optimized using the same mechanism that tracks and optimizes Op[No]Line. Also: - Fix missing DebugScope at top of block. - Allow scalar replacement of access chain in DebugDeclare
2021-09-24 14:56:08 +00:00
%307 = OpExtInst %void %1 DebugLine %55 %uint_21 %uint_21 %uint_12 %uint_20
spirv-opt: Switch from Vulkan.DebugInfo to Shader.DebugInfo (#4493) Includes: - Shift to use of spirv-header extinst.nonsemantic.shader grammar.json - Remove extinst.nonsemantic.vulkan.debuginfo.100.grammar.json - Enable all optimizations for Shader.DebugInfo Also fixes scalar replacement to only insert DebugValue after all OpVariables. This is not necessary for OpenCL.DebugInfo, but it is for Shader.DebugInfo. Likewise, fixes Private-to-Local to insert DebugDeclare after all OpVariables. Also fixes inlining to handle FunctionDefinition which can show up after first block if early return processing happens. Co-authored-by: baldurk <baldurk@baldurk.org>
2021-09-15 18:38:53 +00:00
%97 = OpLoad %PS_OUTPUT %78
Handle NonSemantic.Shader Debug[No]Line (#4530) Debug[No]Line are tracked and optimized using the same mechanism that tracks and optimizes Op[No]Line. Also: - Fix missing DebugScope at top of block. - Allow scalar replacement of access chain in DebugDeclare
2021-09-24 14:56:08 +00:00
%308 = OpExtInst %void %1 DebugLine %55 %uint_21 %uint_21 %uint_5 %uint_20
spirv-opt: Switch from Vulkan.DebugInfo to Shader.DebugInfo (#4493) Includes: - Shift to use of spirv-header extinst.nonsemantic.shader grammar.json - Remove extinst.nonsemantic.vulkan.debuginfo.100.grammar.json - Enable all optimizations for Shader.DebugInfo Also fixes scalar replacement to only insert DebugValue after all OpVariables. This is not necessary for OpenCL.DebugInfo, but it is for Shader.DebugInfo. Likewise, fixes Private-to-Local to insert DebugDeclare after all OpVariables. Also fixes inlining to handle FunctionDefinition which can show up after first block if early return processing happens. Co-authored-by: baldurk <baldurk@baldurk.org>
2021-09-15 18:38:53 +00:00
OpStore %81 %97
Handle NonSemantic.Shader Debug[No]Line (#4530) Debug[No]Line are tracked and optimized using the same mechanism that tracks and optimizes Op[No]Line. Also: - Fix missing DebugScope at top of block. - Allow scalar replacement of access chain in DebugDeclare
2021-09-24 14:56:08 +00:00
%309 = OpExtInst %void %1 DebugNoLine
;CHECK: {{%\w+}} = OpExtInst %void %1 DebugNoLine
spirv-opt: Switch from Vulkan.DebugInfo to Shader.DebugInfo (#4493) Includes: - Shift to use of spirv-header extinst.nonsemantic.shader grammar.json - Remove extinst.nonsemantic.vulkan.debuginfo.100.grammar.json - Enable all optimizations for Shader.DebugInfo Also fixes scalar replacement to only insert DebugValue after all OpVariables. This is not necessary for OpenCL.DebugInfo, but it is for Shader.DebugInfo. Likewise, fixes Private-to-Local to insert DebugDeclare after all OpVariables. Also fixes inlining to handle FunctionDefinition which can show up after first block if early return processing happens. Co-authored-by: baldurk <baldurk@baldurk.org>
2021-09-15 18:38:53 +00:00
%111 = OpExtInst %void %1 DebugNoScope
;CHECK: {{%\w+}} = OpExtInst %void %1 DebugNoScope
%100 = OpCompositeExtract %v4float %97 0
OpStore %out_var_SV_Target0 %100
OpReturn
OpFunctionEnd
)";
SetTargetEnv(SPV_ENV_VULKAN_1_2);
SetAssembleOptions(SPV_TEXT_TO_BINARY_OPTION_PRESERVE_NUMERIC_IDS);
SinglePassRunAndMatch<AggressiveDCEPass>(text, true);
Mark DebugInfoNone as live in ADCE when DebugInfo present (#4568) Otherwise KillInst() tries to generate it when the module is inconsistent.
2021-10-14 17:29:54 +00:00
}
TEST_F(AggressiveDCETest, ShaderDebugInfoGlobalDCE) {
// Verify that DebugGlobalVariable for eliminated private variable has
// variable operand replaced with DebugInfoNone.
const std::string text = R"(OpCapability Shader
OpExtension "SPV_KHR_non_semantic_info"
%1 = OpExtInstImport "NonSemantic.Shader.DebugInfo.100"
OpMemoryModel Logical GLSL450
OpEntryPoint Fragment %MainPs "MainPs" %out_var_SV_Target0 %a
OpExecutionMode %MainPs OriginUpperLeft
%5 = OpString "source2.hlsl"
%24 = OpString "float"
%29 = OpString "vColor"
%33 = OpString "PS_OUTPUT"
%37 = OpString "MainPs"
%38 = OpString ""
%42 = OpString "ps_output"
%46 = OpString "a"
OpName %a "a"
OpName %out_var_SV_Target0 "out.var.SV_Target0"
OpName %MainPs "MainPs"
OpName %PS_OUTPUT "PS_OUTPUT"
OpMemberName %PS_OUTPUT 0 "vColor"
OpDecorate %out_var_SV_Target0 Location 0
%float = OpTypeFloat 32
%v4float = OpTypeVector %float 4
%8 = OpConstantNull %v4float
%float_0 = OpConstant %float 0
%10 = OpConstantComposite %v4float %float_0 %float_0 %float_0 %float_0
%int = OpTypeInt 32 1
%int_0 = OpConstant %int 0
%uint = OpTypeInt 32 0
%uint_32 = OpConstant %uint 32
%_ptr_Private_v4float = OpTypePointer Private %v4float
%_ptr_Output_v4float = OpTypePointer Output %v4float
%void = OpTypeVoid
%uint_1 = OpConstant %uint 1
%uint_4 = OpConstant %uint 4
%uint_5 = OpConstant %uint 5
%uint_3 = OpConstant %uint 3
%uint_0 = OpConstant %uint 0
%uint_128 = OpConstant %uint 128
%uint_12 = OpConstant %uint 12
%uint_8 = OpConstant %uint 8
%uint_9 = OpConstant %uint 9
%uint_10 = OpConstant %uint 10
%uint_15 = OpConstant %uint 15
%48 = OpTypeFunction %void
%PS_OUTPUT = OpTypeStruct %v4float
%54 = OpTypeFunction %PS_OUTPUT
%_ptr_Function_PS_OUTPUT = OpTypePointer Function %PS_OUTPUT
%_ptr_Function_v4float = OpTypePointer Function %v4float
%a = OpVariable %_ptr_Private_v4float Private
;CHECK-NOT: %a = OpVariable %_ptr_Private_v4float Private
%out_var_SV_Target0 = OpVariable %_ptr_Output_v4float Output
;CHECK: [[dbg_none:%\w+]] = OpExtInst %void %1 DebugInfoNone
%18 = OpExtInst %void %1 DebugExpression
%19 = OpExtInst %void %1 DebugSource %5
%20 = OpExtInst %void %1 DebugCompilationUnit %uint_1 %uint_4 %19 %uint_5
%25 = OpExtInst %void %1 DebugTypeBasic %24 %uint_32 %uint_3 %uint_0
%28 = OpExtInst %void %1 DebugTypeVector %25 %uint_4
%31 = OpExtInst %void %1 DebugTypeMember %29 %28 %19 %uint_5 %uint_12 %uint_0 %uint_128 %uint_3
%34 = OpExtInst %void %1 DebugTypeComposite %33 %uint_1 %19 %uint_3 %uint_8 %20 %33 %uint_128 %uint_3 %31
%36 = OpExtInst %void %1 DebugTypeFunction %uint_3 %34
%39 = OpExtInst %void %1 DebugFunction %37 %36 %19 %uint_8 %uint_1 %20 %38 %uint_3 %uint_9
%41 = OpExtInst %void %1 DebugLexicalBlock %19 %uint_9 %uint_1 %39
%43 = OpExtInst %void %1 DebugLocalVariable %42 %34 %19 %uint_10 %uint_15 %41 %uint_4
%47 = OpExtInst %void %1 DebugGlobalVariable %46 %28 %19 %uint_1 %uint_15 %20 %46 %a %uint_8
;CHECK: %47 = OpExtInst %void %1 DebugGlobalVariable %46 %28 %19 %uint_1 %uint_15 %20 %46 [[dbg_none]] %uint_8
%MainPs = OpFunction %void None %48
%49 = OpLabel
%65 = OpVariable %_ptr_Function_PS_OUTPUT Function
%66 = OpVariable %_ptr_Function_PS_OUTPUT Function
OpStore %a %8
%72 = OpExtInst %void %1 DebugScope %41
%69 = OpExtInst %void %1 DebugDeclare %43 %65 %18
OpLine %5 11 5
%70 = OpAccessChain %_ptr_Function_v4float %65 %int_0
OpStore %70 %10
OpLine %5 12 12
%71 = OpLoad %PS_OUTPUT %65
OpLine %5 12 5
OpStore %66 %71
%73 = OpExtInst %void %1 DebugNoLine
%74 = OpExtInst %void %1 DebugNoScope
%51 = OpLoad %PS_OUTPUT %66
%53 = OpCompositeExtract %v4float %51 0
OpStore %out_var_SV_Target0 %53
OpLine %5 13 1
OpReturn
OpFunctionEnd
)";
SetTargetEnv(SPV_ENV_VULKAN_1_2);
SetAssembleOptions(SPV_TEXT_TO_BINARY_OPTION_PRESERVE_NUMERIC_IDS);
SinglePassRunAndMatch<AggressiveDCEPass>(text, true);
Preserve OpenCL.DebugInfo.100 through elim-dead-code-aggressive (#3542) Essentially, it marks all DebugInfo instructions in functions (and their operands) as live. It treats DebugDeclare and DebugValue with Deref as loads and so marks Stores of their variables as live. It marks each DebugGlobalVariables as live except for its variable. After closure, it rechecks if the variable is live. If not, the DebugGlobalVariable instruction's variable operand is set to DebugInfoNone, per the DebugInfo spec.
2020-07-21 20:10:09 +00:00
}
TEST_F(AggressiveDCETest, DebugInfoDeclareKeepsStore) {
// Verify that local variable tc and its store are kept by DebugDeclare.
//
// Same shader source as DebugInfoInFunctionKeepStoreVarElim. The SPIR-V
// has just been inlined.
const std::string text = R"(
OpCapability Shader
%1 = OpExtInstImport "OpenCL.DebugInfo.100"
OpMemoryModel Logical GLSL450
OpEntryPoint Fragment %MainPs "MainPs" %g_tColor %g_sAniso %in_var_TEXCOORD2 %out_var_SV_Target0
OpExecutionMode %MainPs OriginUpperLeft
%20 = OpString "foo.frag"
%24 = OpString "PS_OUTPUT"
%28 = OpString "float"
%31 = OpString "vColor"
%33 = OpString "PS_INPUT"
%38 = OpString "vTextureCoords"
%40 = OpString "@type.2d.image"
%41 = OpString "type.2d.image"
%43 = OpString "Texture2D.TemplateParam"
%47 = OpString "src.MainPs"
%51 = OpString "tc"
%53 = OpString "ps_output"
%56 = OpString "i"
%58 = OpString "@type.sampler"
%59 = OpString "type.sampler"
%61 = OpString "g_sAniso"
%63 = OpString "g_tColor"
OpName %type_2d_image "type.2d.image"
OpName %g_tColor "g_tColor"
OpName %type_sampler "type.sampler"
OpName %g_sAniso "g_sAniso"
OpName %in_var_TEXCOORD2 "in.var.TEXCOORD2"
OpName %out_var_SV_Target0 "out.var.SV_Target0"
OpName %MainPs "MainPs"
OpName %PS_INPUT "PS_INPUT"
OpMemberName %PS_INPUT 0 "vTextureCoords"
OpName %param_var_i "param.var.i"
OpName %PS_OUTPUT "PS_OUTPUT"
OpMemberName %PS_OUTPUT 0 "vColor"
OpName %type_sampled_image "type.sampled.image"
OpDecorate %in_var_TEXCOORD2 Location 0
OpDecorate %out_var_SV_Target0 Location 0
OpDecorate %g_tColor DescriptorSet 0
OpDecorate %g_tColor Binding 0
OpDecorate %g_sAniso DescriptorSet 0
OpDecorate %g_sAniso Binding 1
%int = OpTypeInt 32 1
%int_0 = OpConstant %int 0
%uint = OpTypeInt 32 0
%uint_32 = OpConstant %uint 32
%float = OpTypeFloat 32
%type_2d_image = OpTypeImage %float 2D 2 0 0 1 Unknown
%_ptr_UniformConstant_type_2d_image = OpTypePointer UniformConstant %type_2d_image
%type_sampler = OpTypeSampler
%_ptr_UniformConstant_type_sampler = OpTypePointer UniformConstant %type_sampler
%v2float = OpTypeVector %float 2
%_ptr_Input_v2float = OpTypePointer Input %v2float
%v4float = OpTypeVector %float 4
%_ptr_Output_v4float = OpTypePointer Output %v4float
%void = OpTypeVoid
%uint_128 = OpConstant %uint 128
%uint_0 = OpConstant %uint 0
%uint_64 = OpConstant %uint 64
%65 = OpTypeFunction %void
%PS_INPUT = OpTypeStruct %v2float
%_ptr_Function_PS_INPUT = OpTypePointer Function %PS_INPUT
%PS_OUTPUT = OpTypeStruct %v4float
%75 = OpTypeFunction %PS_OUTPUT %_ptr_Function_PS_INPUT
%_ptr_Function_PS_OUTPUT = OpTypePointer Function %PS_OUTPUT
%_ptr_Function_v2float = OpTypePointer Function %v2float
%type_sampled_image = OpTypeSampledImage %type_2d_image
%_ptr_Function_v4float = OpTypePointer Function %v4float
%g_tColor = OpVariable %_ptr_UniformConstant_type_2d_image UniformConstant
%g_sAniso = OpVariable %_ptr_UniformConstant_type_sampler UniformConstant
%in_var_TEXCOORD2 = OpVariable %_ptr_Input_v2float Input
%out_var_SV_Target0 = OpVariable %_ptr_Output_v4float Output
%39 = OpExtInst %void %1 DebugInfoNone
%55 = OpExtInst %void %1 DebugExpression
%22 = OpExtInst %void %1 DebugSource %20
%23 = OpExtInst %void %1 DebugCompilationUnit 1 4 %22 HLSL
%26 = OpExtInst %void %1 DebugTypeComposite %24 Structure %22 10 1 %23 %24 %uint_128 FlagIsProtected|FlagIsPrivate %27
%29 = OpExtInst %void %1 DebugTypeBasic %28 %uint_32 Float
%30 = OpExtInst %void %1 DebugTypeVector %29 4
%27 = OpExtInst %void %1 DebugTypeMember %31 %30 %22 12 5 %26 %uint_0 %uint_128 FlagIsProtected|FlagIsPrivate
%35 = OpExtInst %void %1 DebugTypeComposite %33 Structure %22 5 1 %23 %33 %uint_64 FlagIsProtected|FlagIsPrivate %36
%37 = OpExtInst %void %1 DebugTypeVector %29 2
%36 = OpExtInst %void %1 DebugTypeMember %38 %37 %22 7 5 %35 %uint_0 %uint_64 FlagIsProtected|FlagIsPrivate
%42 = OpExtInst %void %1 DebugTypeComposite %40 Class %22 0 0 %23 %41 %39 FlagIsProtected|FlagIsPrivate
%44 = OpExtInst %void %1 DebugTypeTemplateParameter %43 %29 %39 %22 0 0
%45 = OpExtInst %void %1 DebugTypeTemplate %42 %44
%46 = OpExtInst %void %1 DebugTypeFunction FlagIsProtected|FlagIsPrivate %26 %35
%48 = OpExtInst %void %1 DebugFunction %47 %46 %22 15 1 %23 %47 FlagIsProtected|FlagIsPrivate 16 %39
%50 = OpExtInst %void %1 DebugLexicalBlock %22 16 1 %48
%52 = OpExtInst %void %1 DebugLocalVariable %51 %37 %22 19 12 %50 FlagIsLocal
%54 = OpExtInst %void %1 DebugLocalVariable %53 %26 %22 17 15 %50 FlagIsLocal
%57 = OpExtInst %void %1 DebugLocalVariable %56 %35 %22 15 29 %48 FlagIsLocal 1
%60 = OpExtInst %void %1 DebugTypeComposite %58 Structure %22 0 0 %23 %59 %39 FlagIsProtected|FlagIsPrivate
%62 = OpExtInst %void %1 DebugGlobalVariable %61 %60 %22 3 14 %23 %61 %g_sAniso FlagIsDefinition
%64 = OpExtInst %void %1 DebugGlobalVariable %63 %42 %22 1 11 %23 %63 %g_tColor FlagIsDefinition
%MainPs = OpFunction %void None %65
%66 = OpLabel
%114 = OpExtInst %void %1 DebugScope %50
%98 = OpVariable %_ptr_Function_PS_OUTPUT Function
%99 = OpVariable %_ptr_Function_v2float Function
%115 = OpExtInst %void %1 DebugNoScope
%100 = OpVariable %_ptr_Function_PS_OUTPUT Function
%param_var_i = OpVariable %_ptr_Function_PS_INPUT Function
%70 = OpLoad %v2float %in_var_TEXCOORD2
%71 = OpCompositeConstruct %PS_INPUT %70
OpStore %param_var_i %71
%116 = OpExtInst %void %1 DebugScope %48
%102 = OpExtInst %void %1 DebugDeclare %57 %param_var_i %55
%117 = OpExtInst %void %1 DebugScope %50
%103 = OpExtInst %void %1 DebugDeclare %54 %98 %55
OpLine %20 19 17
%104 = OpAccessChain %_ptr_Function_v2float %param_var_i %int_0
%105 = OpLoad %v2float %104
OpLine %20 19 12
OpStore %99 %105
;CHECK: OpStore %99 %105
%106 = OpExtInst %void %1 DebugDeclare %52 %99 %55
OpLine %20 20 26
%107 = OpLoad %type_2d_image %g_tColor
OpLine %20 20 46
%108 = OpLoad %type_sampler %g_sAniso
OpLine %20 20 26
%110 = OpSampledImage %type_sampled_image %107 %108
%111 = OpImageSampleImplicitLod %v4float %110 %105 None
OpLine %20 20 5
%112 = OpAccessChain %_ptr_Function_v4float %98 %int_0
OpStore %112 %111
OpLine %20 21 12
%113 = OpLoad %PS_OUTPUT %98
OpLine %20 21 5
OpStore %100 %113
%118 = OpExtInst %void %1 DebugNoScope
%73 = OpLoad %PS_OUTPUT %100
%74 = OpCompositeExtract %v4float %73 0
OpStore %out_var_SV_Target0 %74
OpReturn
OpFunctionEnd
)";
SetTargetEnv(SPV_ENV_VULKAN_1_2);
SetAssembleOptions(SPV_TEXT_TO_BINARY_OPTION_PRESERVE_NUMERIC_IDS);
SinglePassRunAndMatch<AggressiveDCEPass>(text, true);
}
TEST_F(AggressiveDCETest, DebugInfoValueDerefKeepsStore) {
// Verify that local variable tc and its store are kept by DebugValue with
// Deref.
//
// Same shader source as DebugInfoInFunctionKeepStoreVarElim. The SPIR-V
// has just been inlined and edited to replace the DebugDeclare with the
// DebugValue/Deref.
const std::string text = R"(
OpCapability Shader
%1 = OpExtInstImport "OpenCL.DebugInfo.100"
OpMemoryModel Logical GLSL450
OpEntryPoint Fragment %MainPs "MainPs" %g_tColor %g_sAniso %in_var_TEXCOORD2 %out_var_SV_Target0
OpExecutionMode %MainPs OriginUpperLeft
%7 = OpString "foo.frag"
%8 = OpString "PS_OUTPUT"
%9 = OpString "float"
%10 = OpString "vColor"
%11 = OpString "PS_INPUT"
%12 = OpString "vTextureCoords"
%13 = OpString "@type.2d.image"
%14 = OpString "type.2d.image"
%15 = OpString "Texture2D.TemplateParam"
%16 = OpString "src.MainPs"
%17 = OpString "tc"
%18 = OpString "ps_output"
%19 = OpString "i"
%20 = OpString "@type.sampler"
%21 = OpString "type.sampler"
%22 = OpString "g_sAniso"
%23 = OpString "g_tColor"
OpName %type_2d_image "type.2d.image"
OpName %g_tColor "g_tColor"
OpName %type_sampler "type.sampler"
OpName %g_sAniso "g_sAniso"
OpName %in_var_TEXCOORD2 "in.var.TEXCOORD2"
OpName %out_var_SV_Target0 "out.var.SV_Target0"
OpName %MainPs "MainPs"
OpName %PS_INPUT "PS_INPUT"
OpMemberName %PS_INPUT 0 "vTextureCoords"
OpName %param_var_i "param.var.i"
OpName %PS_OUTPUT "PS_OUTPUT"
OpMemberName %PS_OUTPUT 0 "vColor"
OpName %type_sampled_image "type.sampled.image"
OpDecorate %in_var_TEXCOORD2 Location 0
OpDecorate %out_var_SV_Target0 Location 0
OpDecorate %g_tColor DescriptorSet 0
OpDecorate %g_tColor Binding 0
OpDecorate %g_sAniso DescriptorSet 0
OpDecorate %g_sAniso Binding 1
%int = OpTypeInt 32 1
%int_0 = OpConstant %int 0
%uint = OpTypeInt 32 0
%uint_32 = OpConstant %uint 32
%float = OpTypeFloat 32
%type_2d_image = OpTypeImage %float 2D 2 0 0 1 Unknown
%_ptr_UniformConstant_type_2d_image = OpTypePointer UniformConstant %type_2d_image
%type_sampler = OpTypeSampler
%_ptr_UniformConstant_type_sampler = OpTypePointer UniformConstant %type_sampler
%v2float = OpTypeVector %float 2
%_ptr_Input_v2float = OpTypePointer Input %v2float
%v4float = OpTypeVector %float 4
%_ptr_Output_v4float = OpTypePointer Output %v4float
%void = OpTypeVoid
%uint_128 = OpConstant %uint 128
%uint_0 = OpConstant %uint 0
%uint_64 = OpConstant %uint 64
%45 = OpTypeFunction %void
%PS_INPUT = OpTypeStruct %v2float
%_ptr_Function_PS_INPUT = OpTypePointer Function %PS_INPUT
%PS_OUTPUT = OpTypeStruct %v4float
%47 = OpTypeFunction %PS_OUTPUT %_ptr_Function_PS_INPUT
%_ptr_Function_PS_OUTPUT = OpTypePointer Function %PS_OUTPUT
%_ptr_Function_v2float = OpTypePointer Function %v2float
%type_sampled_image = OpTypeSampledImage %type_2d_image
%_ptr_Function_v4float = OpTypePointer Function %v4float
%g_tColor = OpVariable %_ptr_UniformConstant_type_2d_image UniformConstant
%g_sAniso = OpVariable %_ptr_UniformConstant_type_sampler UniformConstant
%in_var_TEXCOORD2 = OpVariable %_ptr_Input_v2float Input
%out_var_SV_Target0 = OpVariable %_ptr_Output_v4float Output
%51 = OpExtInst %void %1 DebugInfoNone
%52 = OpExtInst %void %1 DebugExpression
%53 = OpExtInst %void %1 DebugOperation Deref
%54 = OpExtInst %void %1 DebugExpression %53
%55 = OpExtInst %void %1 DebugSource %7
%56 = OpExtInst %void %1 DebugCompilationUnit 1 4 %55 HLSL
%57 = OpExtInst %void %1 DebugTypeComposite %8 Structure %55 10 1 %56 %8 %uint_128 FlagIsProtected|FlagIsPrivate %58
%59 = OpExtInst %void %1 DebugTypeBasic %9 %uint_32 Float
%60 = OpExtInst %void %1 DebugTypeVector %59 4
%58 = OpExtInst %void %1 DebugTypeMember %10 %60 %55 12 5 %57 %uint_0 %uint_128 FlagIsProtected|FlagIsPrivate
%61 = OpExtInst %void %1 DebugTypeComposite %11 Structure %55 5 1 %56 %11 %uint_64 FlagIsProtected|FlagIsPrivate %62
%63 = OpExtInst %void %1 DebugTypeVector %59 2
%62 = OpExtInst %void %1 DebugTypeMember %12 %63 %55 7 5 %61 %uint_0 %uint_64 FlagIsProtected|FlagIsPrivate
%64 = OpExtInst %void %1 DebugTypeComposite %13 Class %55 0 0 %56 %14 %51 FlagIsProtected|FlagIsPrivate
%65 = OpExtInst %void %1 DebugTypeTemplateParameter %15 %59 %51 %55 0 0
%66 = OpExtInst %void %1 DebugTypeTemplate %64 %65
%67 = OpExtInst %void %1 DebugTypeFunction FlagIsProtected|FlagIsPrivate %57 %61
%68 = OpExtInst %void %1 DebugFunction %16 %67 %55 15 1 %56 %16 FlagIsProtected|FlagIsPrivate 16 %51
%69 = OpExtInst %void %1 DebugLexicalBlock %55 16 1 %68
%70 = OpExtInst %void %1 DebugLocalVariable %17 %63 %55 19 12 %69 FlagIsLocal
%71 = OpExtInst %void %1 DebugLocalVariable %18 %57 %55 17 15 %69 FlagIsLocal
%72 = OpExtInst %void %1 DebugLocalVariable %19 %61 %55 15 29 %68 FlagIsLocal 1
%73 = OpExtInst %void %1 DebugTypeComposite %20 Structure %55 0 0 %56 %21 %51 FlagIsProtected|FlagIsPrivate
%74 = OpExtInst %void %1 DebugGlobalVariable %22 %73 %55 3 14 %56 %22 %g_sAniso FlagIsDefinition
%75 = OpExtInst %void %1 DebugGlobalVariable %23 %64 %55 1 11 %56 %23 %g_tColor FlagIsDefinition
%MainPs = OpFunction %void None %45
%76 = OpLabel
%101 = OpExtInst %void %1 DebugScope %69
%78 = OpVariable %_ptr_Function_PS_OUTPUT Function
%79 = OpVariable %_ptr_Function_v2float Function
%102 = OpExtInst %void %1 DebugNoScope
%81 = OpVariable %_ptr_Function_PS_OUTPUT Function
%param_var_i = OpVariable %_ptr_Function_PS_INPUT Function
%82 = OpLoad %v2float %in_var_TEXCOORD2
%83 = OpCompositeConstruct %PS_INPUT %82
OpStore %param_var_i %83
%103 = OpExtInst %void %1 DebugScope %68
%85 = OpExtInst %void %1 DebugDeclare %72 %param_var_i %52
%104 = OpExtInst %void %1 DebugScope %69
%87 = OpExtInst %void %1 DebugDeclare %71 %78 %52
OpLine %7 19 17
%88 = OpAccessChain %_ptr_Function_v2float %param_var_i %int_0
%89 = OpLoad %v2float %88
OpLine %7 19 12
OpStore %79 %89
;CHECK: OpStore %79 %89
%90 = OpExtInst %void %1 DebugValue %70 %79 %54
OpLine %7 20 26
%91 = OpLoad %type_2d_image %g_tColor
OpLine %7 20 46
%92 = OpLoad %type_sampler %g_sAniso
OpLine %7 20 26
%94 = OpSampledImage %type_sampled_image %91 %92
%95 = OpImageSampleImplicitLod %v4float %94 %89 None
OpLine %7 20 5
%96 = OpAccessChain %_ptr_Function_v4float %78 %int_0
OpStore %96 %95
OpLine %7 21 12
%97 = OpLoad %PS_OUTPUT %78
OpLine %7 21 5
OpStore %81 %97
%105 = OpExtInst %void %1 DebugNoScope
%99 = OpLoad %PS_OUTPUT %81
%100 = OpCompositeExtract %v4float %99 0
OpStore %out_var_SV_Target0 %100
OpReturn
OpFunctionEnd
)";
SetTargetEnv(SPV_ENV_VULKAN_1_2);
SetAssembleOptions(SPV_TEXT_TO_BINARY_OPTION_PRESERVE_NUMERIC_IDS);
SinglePassRunAndMatch<AggressiveDCEPass>(text, true);
}
TEST_F(AggressiveDCETest, DebugInfoElimUnusedTextureKeepGlobalVariable) {
// Verify that unused texture g_tColor2 is eliminated but its
// DebugGlobalVariable is retained but with DebugInfoNone for its Variable.
//
// Same shader source as DebugInfoInFunctionKeepStoreVarElim but with unused
// g_tColor2 added.
const std::string text = R"(
OpCapability Shader
%1 = OpExtInstImport "OpenCL.DebugInfo.100"
OpMemoryModel Logical GLSL450
OpEntryPoint Fragment %MainPs "MainPs" %g_tColor %g_tColor2 %g_sAniso %in_var_TEXCOORD2 %out_var_SV_Target0
OpExecutionMode %MainPs OriginUpperLeft
%21 = OpString "foo6.frag"
%25 = OpString "PS_OUTPUT"
%29 = OpString "float"
%32 = OpString "vColor"
%34 = OpString "PS_INPUT"
%39 = OpString "vTextureCoords"
%41 = OpString "@type.2d.image"
%42 = OpString "type.2d.image"
%44 = OpString "Texture2D.TemplateParam"
%48 = OpString "src.MainPs"
%52 = OpString "tc"
%54 = OpString "ps_output"
%57 = OpString "i"
%59 = OpString "@type.sampler"
%60 = OpString "type.sampler"
%62 = OpString "g_sAniso"
%64 = OpString "g_tColor2"
%66 = OpString "g_tColor"
OpName %type_2d_image "type.2d.image"
OpName %g_tColor "g_tColor"
OpName %g_tColor2 "g_tColor2"
OpName %type_sampler "type.sampler"
OpName %g_sAniso "g_sAniso"
OpName %in_var_TEXCOORD2 "in.var.TEXCOORD2"
OpName %out_var_SV_Target0 "out.var.SV_Target0"
OpName %MainPs "MainPs"
OpName %PS_INPUT "PS_INPUT"
OpMemberName %PS_INPUT 0 "vTextureCoords"
OpName %param_var_i "param.var.i"
OpName %PS_OUTPUT "PS_OUTPUT"
OpMemberName %PS_OUTPUT 0 "vColor"
OpName %type_sampled_image "type.sampled.image"
OpDecorate %in_var_TEXCOORD2 Location 0
OpDecorate %out_var_SV_Target0 Location 0
OpDecorate %g_tColor DescriptorSet 0
OpDecorate %g_tColor Binding 0
OpDecorate %g_tColor2 DescriptorSet 0
OpDecorate %g_tColor2 Binding 1
OpDecorate %g_sAniso DescriptorSet 0
OpDecorate %g_sAniso Binding 2
%int = OpTypeInt 32 1
%int_0 = OpConstant %int 0
%uint = OpTypeInt 32 0
%uint_32 = OpConstant %uint 32
%float = OpTypeFloat 32
%type_2d_image = OpTypeImage %float 2D 2 0 0 1 Unknown
%_ptr_UniformConstant_type_2d_image = OpTypePointer UniformConstant %type_2d_image
%type_sampler = OpTypeSampler
%_ptr_UniformConstant_type_sampler = OpTypePointer UniformConstant %type_sampler
%v2float = OpTypeVector %float 2
%_ptr_Input_v2float = OpTypePointer Input %v2float
%v4float = OpTypeVector %float 4
%_ptr_Output_v4float = OpTypePointer Output %v4float
%void = OpTypeVoid
%uint_128 = OpConstant %uint 128
%uint_0 = OpConstant %uint 0
%uint_64 = OpConstant %uint 64
%68 = OpTypeFunction %void
%PS_INPUT = OpTypeStruct %v2float
%_ptr_Function_PS_INPUT = OpTypePointer Function %PS_INPUT
%PS_OUTPUT = OpTypeStruct %v4float
%78 = OpTypeFunction %PS_OUTPUT %_ptr_Function_PS_INPUT
%_ptr_Function_PS_OUTPUT = OpTypePointer Function %PS_OUTPUT
%_ptr_Function_v2float = OpTypePointer Function %v2float
%type_sampled_image = OpTypeSampledImage %type_2d_image
%_ptr_Function_v4float = OpTypePointer Function %v4float
%g_tColor = OpVariable %_ptr_UniformConstant_type_2d_image UniformConstant
%g_tColor2 = OpVariable %_ptr_UniformConstant_type_2d_image UniformConstant
;CHECK-NOT: %g_tColor2 = OpVariable %_ptr_UniformConstant_type_2d_image UniformConstant
%g_sAniso = OpVariable %_ptr_UniformConstant_type_sampler UniformConstant
%in_var_TEXCOORD2 = OpVariable %_ptr_Input_v2float Input
%out_var_SV_Target0 = OpVariable %_ptr_Output_v4float Output
%40 = OpExtInst %void %1 DebugInfoNone
%56 = OpExtInst %void %1 DebugExpression
%23 = OpExtInst %void %1 DebugSource %21
%24 = OpExtInst %void %1 DebugCompilationUnit 1 4 %23 HLSL
%27 = OpExtInst %void %1 DebugTypeComposite %25 Structure %23 11 1 %24 %25 %uint_128 FlagIsProtected|FlagIsPrivate %28
%30 = OpExtInst %void %1 DebugTypeBasic %29 %uint_32 Float
%31 = OpExtInst %void %1 DebugTypeVector %30 4
%28 = OpExtInst %void %1 DebugTypeMember %32 %31 %23 13 5 %27 %uint_0 %uint_128 FlagIsProtected|FlagIsPrivate
%36 = OpExtInst %void %1 DebugTypeComposite %34 Structure %23 6 1 %24 %34 %uint_64 FlagIsProtected|FlagIsPrivate %37
%38 = OpExtInst %void %1 DebugTypeVector %30 2
%37 = OpExtInst %void %1 DebugTypeMember %39 %38 %23 8 5 %36 %uint_0 %uint_64 FlagIsProtected|FlagIsPrivate
%43 = OpExtInst %void %1 DebugTypeComposite %41 Class %23 0 0 %24 %42 %40 FlagIsProtected|FlagIsPrivate
%45 = OpExtInst %void %1 DebugTypeTemplateParameter %44 %30 %40 %23 0 0
%46 = OpExtInst %void %1 DebugTypeTemplate %43 %45
%47 = OpExtInst %void %1 DebugTypeFunction FlagIsProtected|FlagIsPrivate %27 %36
%49 = OpExtInst %void %1 DebugFunction %48 %47 %23 16 1 %24 %48 FlagIsProtected|FlagIsPrivate 17 %40
%51 = OpExtInst %void %1 DebugLexicalBlock %23 17 1 %49
%53 = OpExtInst %void %1 DebugLocalVariable %52 %38 %23 20 12 %51 FlagIsLocal
%55 = OpExtInst %void %1 DebugLocalVariable %54 %27 %23 18 15 %51 FlagIsLocal
%58 = OpExtInst %void %1 DebugLocalVariable %57 %36 %23 16 29 %49 FlagIsLocal 1
%61 = OpExtInst %void %1 DebugTypeComposite %59 Structure %23 0 0 %24 %60 %40 FlagIsProtected|FlagIsPrivate
%63 = OpExtInst %void %1 DebugGlobalVariable %62 %61 %23 4 14 %24 %62 %g_sAniso FlagIsDefinition
%65 = OpExtInst %void %1 DebugGlobalVariable %64 %43 %23 2 11 %24 %64 %g_tColor2 FlagIsDefinition
;CHECK-NOT: %65 = OpExtInst %void %1 DebugGlobalVariable %64 %43 %23 2 11 %24 %64 %g_tColor2 FlagIsDefinition
;CHECK: %65 = OpExtInst %void %1 DebugGlobalVariable %64 %43 %23 2 11 %24 %64 %40 FlagIsDefinition
%67 = OpExtInst %void %1 DebugGlobalVariable %66 %43 %23 1 11 %24 %66 %g_tColor FlagIsDefinition
%MainPs = OpFunction %void None %68
%69 = OpLabel
%117 = OpExtInst %void %1 DebugScope %51
%101 = OpVariable %_ptr_Function_PS_OUTPUT Function
%102 = OpVariable %_ptr_Function_v2float Function
%118 = OpExtInst %void %1 DebugNoScope
%103 = OpVariable %_ptr_Function_PS_OUTPUT Function
%param_var_i = OpVariable %_ptr_Function_PS_INPUT Function
%73 = OpLoad %v2float %in_var_TEXCOORD2
%74 = OpCompositeConstruct %PS_INPUT %73
OpStore %param_var_i %74
%119 = OpExtInst %void %1 DebugScope %49
%105 = OpExtInst %void %1 DebugDeclare %58 %param_var_i %56
%120 = OpExtInst %void %1 DebugScope %51
%106 = OpExtInst %void %1 DebugDeclare %55 %101 %56
OpLine %21 20 17
%107 = OpAccessChain %_ptr_Function_v2float %param_var_i %int_0
%108 = OpLoad %v2float %107
OpLine %21 20 12
OpStore %102 %108
%109 = OpExtInst %void %1 DebugDeclare %53 %102 %56
OpLine %21 21 26
%110 = OpLoad %type_2d_image %g_tColor
OpLine %21 21 46
%111 = OpLoad %type_sampler %g_sAniso
OpLine %21 21 57
%112 = OpLoad %v2float %102
OpLine %21 21 26
%113 = OpSampledImage %type_sampled_image %110 %111
%114 = OpImageSampleImplicitLod %v4float %113 %112 None
OpLine %21 21 5
%115 = OpAccessChain %_ptr_Function_v4float %101 %int_0
OpStore %115 %114
OpLine %21 22 12
%116 = OpLoad %PS_OUTPUT %101
OpLine %21 22 5
OpStore %103 %116
%121 = OpExtInst %void %1 DebugNoScope
%76 = OpLoad %PS_OUTPUT %103
%77 = OpCompositeExtract %v4float %76 0
OpStore %out_var_SV_Target0 %77
OpReturn
OpFunctionEnd
)";
SetTargetEnv(SPV_ENV_VULKAN_1_2);
SetAssembleOptions(SPV_TEXT_TO_BINARY_OPTION_PRESERVE_NUMERIC_IDS);
SinglePassRunAndMatch<AggressiveDCEPass>(text, true);
}
Let ADCE pass check DebugScope (#3703) In the existing code, ADCE pass does not check DebugScope of an instruction when it checks the users of each instruction, which results in removing OpenCL.Debug.100 instructions that are only used by DebugScope. This commit lets ADCE pass add DebugScope of an instruction to the live instruction set when the instruction is added to the live instruction set.
2020-08-18 13:33:20 +00:00
TEST_F(AggressiveDCETest, KeepDebugScopeParent) {
// Verify that local variable tc and its store are kept by DebugDeclare.
//
// Same shader source as DebugInfoInFunctionKeepStoreVarElim. The SPIR-V
// has just been inlined.
const std::string text = R"(
OpCapability Shader
%1 = OpExtInstImport "OpenCL.DebugInfo.100"
OpMemoryModel Logical GLSL450
OpEntryPoint Fragment %main "main" %out_var_SV_TARGET0
OpExecutionMode %main OriginUpperLeft
%11 = OpString "float"
%16 = OpString "t.hlsl"
%19 = OpString "src.main"
OpName %out_var_SV_TARGET0 "out.var.SV_TARGET0"
OpName %main "main"
OpDecorate %out_var_SV_TARGET0 Location 0
%float = OpTypeFloat 32
%float_0 = OpConstant %float 0
%v4float = OpTypeVector %float 4
%7 = OpConstantComposite %v4float %float_0 %float_0 %float_0 %float_0
%uint = OpTypeInt 32 0
%uint_32 = OpConstant %uint 32
%_ptr_Output_v4float = OpTypePointer Output %v4float
%void = OpTypeVoid
%23 = OpTypeFunction %void
%26 = OpTypeFunction %v4float
%out_var_SV_TARGET0 = OpVariable %_ptr_Output_v4float Output
%_ptr_Function_v4float = OpTypePointer Function %v4float
%33 = OpExtInst %void %1 DebugInfoNone
%13 = OpExtInst %void %1 DebugTypeBasic %11 %uint_32 Float
%14 = OpExtInst %void %1 DebugTypeVector %13 4
%15 = OpExtInst %void %1 DebugTypeFunction FlagIsProtected|FlagIsPrivate %14
%17 = OpExtInst %void %1 DebugSource %16
%18 = OpExtInst %void %1 DebugCompilationUnit 1 4 %17 HLSL
%20 = OpExtInst %void %1 DebugFunction %19 %15 %17 1 1 %18 %19 FlagIsProtected|FlagIsPrivate 2 %33
%22 = OpExtInst %void %1 DebugLexicalBlock %17 2 1 %20
%main = OpFunction %void None %23
%24 = OpLabel
%31 = OpVariable %_ptr_Function_v4float Function
; CHECK: [[block:%\w+]] = OpExtInst %void %1 DebugLexicalBlock
; CHECK: DebugScope [[block]]
%34 = OpExtInst %void %1 DebugScope %22
OpLine %16 3 5
OpStore %31 %7
OpStore %out_var_SV_TARGET0 %7
%35 = OpExtInst %void %1 DebugNoScope
OpReturn
OpFunctionEnd
)";
SetAssembleOptions(SPV_TEXT_TO_BINARY_OPTION_PRESERVE_NUMERIC_IDS);
SinglePassRunAndMatch<AggressiveDCEPass>(text, true);
}
Handle exported functions in ADCE (#4495) ADCE does not handle exported functions. This was an explicit decision because we did not believe that the linkage attribute could be used in shaders, but it can now. This change has been made. While fixing this error, I noticed that the OpName for labels is sometimes removed because the label instructions are not marked explicitly marked as live. This has able been fixed.
2021-08-31 16:39:46 +00:00
TEST_F(AggressiveDCETest, KeepExportFunctions) {
// All functions are reachable. In particular, ExportedFunc and Constant are
// reachable because ExportedFunc is exported. Nothing should be removed.
const std::vector<const char*> text = {
// clang-format off
"OpCapability Shader",
"OpCapability Linkage",
"OpMemoryModel Logical GLSL450",
"OpEntryPoint Fragment %main \"main\"",
"OpName %main \"main\"",
"OpName %ExportedFunc \"ExportedFunc\"",
"OpName %Live \"Live\"",
"OpDecorate %ExportedFunc LinkageAttributes \"ExportedFunc\" Export",
"%void = OpTypeVoid",
"%7 = OpTypeFunction %void",
"%main = OpFunction %void None %7",
"%15 = OpLabel",
"OpReturn",
"OpFunctionEnd",
"%ExportedFunc = OpFunction %void None %7",
"%19 = OpLabel",
"%16 = OpFunctionCall %void %Live",
"OpReturn",
"OpFunctionEnd",
"%Live = OpFunction %void None %7",
"%20 = OpLabel",
"OpReturn",
"OpFunctionEnd"
// clang-format on
};
SetAssembleOptions(SPV_TEXT_TO_BINARY_OPTION_PRESERVE_NUMERIC_IDS);
std::string assembly = JoinAllInsts(text);
auto result = SinglePassRunAndDisassemble<AggressiveDCEPass>(
assembly, /* skip_nop = */ true, /* do_validation = */ false);
EXPECT_EQ(Pass::Status::SuccessWithoutChange, std::get<1>(result));
EXPECT_EQ(assembly, std::get<0>(result));
}
TEST_F(AggressiveDCETest, KeepPrivateVarInExportFunctions) {
// The loads and stores from the private variable should not be removed
// because the functions are exported and could be called.
const std::string text = R"(OpCapability Shader
OpCapability Linkage
OpMemoryModel Logical GLSL450
OpSource HLSL 630
OpName %privateVar "privateVar"
OpName %ReadPrivate "ReadPrivate"
OpName %WritePrivate "WritePrivate"
OpName %value "value"
OpDecorate %ReadPrivate LinkageAttributes "ReadPrivate" Export
OpDecorate %WritePrivate LinkageAttributes "WritePrivate" Export
%int = OpTypeInt 32 1
%_ptr_Private_int = OpTypePointer Private %int
%6 = OpTypeFunction %int
%void = OpTypeVoid
%_ptr_Function_int = OpTypePointer Function %int
%10 = OpTypeFunction %void %_ptr_Function_int
%privateVar = OpVariable %_ptr_Private_int Private
%ReadPrivate = OpFunction %int None %6
%12 = OpLabel
%8 = OpLoad %int %privateVar
OpReturnValue %8
OpFunctionEnd
%WritePrivate = OpFunction %void None %10
%value = OpFunctionParameter %_ptr_Function_int
%13 = OpLabel
%14 = OpLoad %int %value
OpStore %privateVar %14
OpReturn
OpFunctionEnd
)";
SetAssembleOptions(SPV_TEXT_TO_BINARY_OPTION_PRESERVE_NUMERIC_IDS);
auto result = SinglePassRunAndDisassemble<AggressiveDCEPass>(
text, /* skip_nop = */ true, /* do_validation = */ false);
EXPECT_EQ(Pass::Status::SuccessWithoutChange, std::get<1>(result));
EXPECT_EQ(text, std::get<0>(result));
}
TEST_F(AggressiveDCETest, KeepLableNames) {
const std::string text = R"(OpCapability Shader
OpCapability Linkage
OpMemoryModel Logical GLSL450
OpSource HLSL 630
OpName %WritePrivate "WritePrivate"
OpName %entry "entry"
OpName %target "target"
OpDecorate %WritePrivate LinkageAttributes "WritePrivate" Export
%void = OpTypeVoid
%3 = OpTypeFunction %void
%WritePrivate = OpFunction %void None %3
%entry = OpLabel
OpBranch %target
%target = OpLabel
OpReturn
OpFunctionEnd
)";
SetAssembleOptions(SPV_TEXT_TO_BINARY_OPTION_PRESERVE_NUMERIC_IDS);
auto result = SinglePassRunAndDisassemble<AggressiveDCEPass>(
text, /* skip_nop = */ true, /* do_validation = */ false);
EXPECT_EQ(Pass::Status::SuccessWithoutChange, std::get<1>(result));
EXPECT_EQ(text, std::get<0>(result));
}
Add preserve_interface mode to aggressive_dead_code_elim (#4520) This mode is needed by GPU-assisted validation instrumentation which cannot change the shader entry point interface.
2021-09-15 18:38:34 +00:00
TEST_F(AggressiveDCETest, PreserveInterface) {
// Set preserve_interface to true. Verify that unused uniform
// constant in entry point interface is not eliminated.
const std::string text = R"(OpCapability RayTracingKHR
OpExtension "SPV_KHR_ray_tracing"
%1 = OpExtInstImport "GLSL.std.450"
OpMemoryModel Logical GLSL450
OpEntryPoint RayGenerationNV %2 "main" %3 %4
OpDecorate %3 Location 0
OpDecorate %4 DescriptorSet 2
OpDecorate %4 Binding 0
%void = OpTypeVoid
%6 = OpTypeFunction %void
%uint = OpTypeInt 32 0
%uint_0 = OpConstant %uint 0
%float = OpTypeFloat 32
%_ptr_CallableDataNV_float = OpTypePointer CallableDataNV %float
%3 = OpVariable %_ptr_CallableDataNV_float CallableDataNV
%13 = OpTypeAccelerationStructureKHR
%_ptr_UniformConstant_13 = OpTypePointer UniformConstant %13
%4 = OpVariable %_ptr_UniformConstant_13 UniformConstant
%2 = OpFunction %void None %6
%15 = OpLabel
OpExecuteCallableKHR %uint_0 %3
OpReturn
OpFunctionEnd
)";
SetTargetEnv(SPV_ENV_VULKAN_1_2);
SetAssembleOptions(SPV_TEXT_TO_BINARY_OPTION_PRESERVE_NUMERIC_IDS);
auto result = SinglePassRunAndDisassemble<AggressiveDCEPass>(
text, /* skip_nop = */ true, /* do_validation = */ false,
/* preserve_interface */ true);
EXPECT_EQ(Pass::Status::SuccessWithoutChange, std::get<1>(result));
EXPECT_EQ(text, std::get<0>(result));
}
Change branch handling in ADCE to fix errors (#4596) Consider the new test case. The conditional branch in the continue block is never marked as live. However, `IsDead` will say it is not dead, so it does not get deleted. Because it was never marked as live, `%false` was not mark as live either, but it gets deleted. This results in invalid code. To fix this properly, we had to reconsider how branches are handle. We make the following changes: 1) Terminator instructions that are not branch or OpUnreachable must be kept, so they are marked as live when initializing the worklist. 2) Branches and OpUnreachable instructions are marked as live if a) the block does not have a merge instruction and another instruction in the block is marked as live, or b) the merge instruction in the same block is marked as live. 3) Any instruction that is not marked as live is removed. 4) If a terminator is to be removed, an OpUnreachable is added. This happens when the entire block is dead, and the block will be removed. The OpUnreachable is generated to make sure the block still has a terminator, and is valid. Fixes https://github.com/KhronosGroup/SPIRV-Tools/issues/4509.
2021-10-29 14:46:43 +00:00
TEST_F(AggressiveDCETest, EmptyContinueWithConditionalBranch) {
const std::string text = R"(OpCapability Shader
%1 = OpExtInstImport "GLSL.std.450"
OpMemoryModel Logical GLSL450
OpEntryPoint Fragment %2 "main"
OpExecutionMode %2 OriginUpperLeft
%void = OpTypeVoid
%4 = OpTypeFunction %void
%bool = OpTypeBool
%false = OpConstantFalse %bool
%2 = OpFunction %void None %4
%9 = OpLabel
OpBranch %10
%10 = OpLabel
OpLoopMerge %11 %12 None
OpBranch %13
%13 = OpLabel
OpKill
%12 = OpLabel
OpBranchConditional %false %10 %10
%11 = OpLabel
OpUnreachable
OpFunctionEnd
)";
SetTargetEnv(SPV_ENV_VULKAN_1_2);
SetAssembleOptions(SPV_TEXT_TO_BINARY_OPTION_PRESERVE_NUMERIC_IDS);
SinglePassRunAndCheck<AggressiveDCEPass>(text, text, false);
}
Add AggressiveDCEPass Create aggressive dead code elimination pass This pass eliminates unused code from functions. In addition, it detects and eliminates code which may have spurious uses but which do not contribute to the output of the function. The most common cause of such code sequences is summations in loops whose result is no longer used due to dead code elimination. This optimization has additional compile time cost over standard dead code elimination. This pass only processes entry point functions. It also only processes shaders with logical addressing. It currently will not process functions with function calls. It currently only supports the GLSL.std.450 extended instruction set. It currently does not support any extensions. This pass will be made more effective by first running passes that remove dead control flow and inlines function calls. This pass can be especially useful after running Local Access Chain Conversion, which tends to cause cycles of dead code to be left after Store/Load elimination passes are completed. These cycles cannot be eliminated with standard dead code elimination. Additionally: This transform uses a whitelist of instructions that it knows do have side effects, (a.k.a. combinators). It assumes other instructions have side effects: it will not remove them, and assumes they have side effects via their ID operands.
2017-06-08 16:37:21 +00:00
DCE: clean up the cfg for all functions that were processed. (#4840) Which functions are processed is determined by which ones are on the call tree from the entry points before dead code is removed. So it is possible that a function is process because it is called from an entry point, but the CFG is not cleaned up because the call to the function was removed. The fix is to process and cleanup every function in the module. Since all of the dead functions would have already been removed in an earlier step of DCE, it should not make a different in compile time. Fixes #4731
2022-07-05 16:23:32 +00:00
TEST_F(AggressiveDCETest, FunctionBecomesUnreachableAfterDCE) {
const std::string text = R"(
OpCapability Shader
%1 = OpExtInstImport "GLSL.std.450"
OpMemoryModel Logical GLSL450
OpEntryPoint Fragment %2 "main"
OpExecutionMode %2 OriginUpperLeft
OpSource ESSL 320
%void = OpTypeVoid
%4 = OpTypeFunction %void
%int = OpTypeInt 32 1
%_ptr_Function_int = OpTypePointer Function %int
%7 = OpTypeFunction %int
%int_n1 = OpConstant %int -1
%2 = OpFunction %void None %4
%9 = OpLabel
OpKill
%10 = OpLabel
%11 = OpFunctionCall %int %12
OpReturn
OpFunctionEnd
; CHECK: {{%\w+}} = OpFunction %int DontInline|Pure
%12 = OpFunction %int DontInline|Pure %7
; CHECK-NEXT: {{%\w+}} = OpLabel
%13 = OpLabel
%14 = OpVariable %_ptr_Function_int Function
; CHECK-NEXT: OpBranch [[header:%\w+]]
OpBranch %15
; CHECK-NEXT: [[header]] = OpLabel
; CHECK-NEXT: OpBranch [[merge:%\w+]]
%15 = OpLabel
OpLoopMerge %16 %17 None
OpBranch %18
%18 = OpLabel
%19 = OpLoad %int %14
OpBranch %17
%17 = OpLabel
OpBranch %15
; CHECK-NEXT: [[merge]] = OpLabel
%16 = OpLabel
; CHECK-NEXT: OpReturnValue %int_n1
OpReturnValue %int_n1
; CHECK-NEXT: OpFunctionEnd
OpFunctionEnd
)";
SinglePassRunAndMatch<AggressiveDCEPass>(text, true);
}
Fix ADCE to not eliminate top level DebugInfo instructions (#4889) Specifically, DebugSourceContinued, DebugCompilationUnit, and DebugEntryPoint. These instructions are top-level instructions which do not or may not have a user except for the tool and so should not be eliminated.
2022-08-15 21:23:23 +00:00
TEST_F(AggressiveDCETest, KeepTopLevelDebugInfo) {
// Don't eliminate DebugCompilationUnit, DebugSourceContinued, and
// DebugEntryPoint
const std::string text = R"(
OpCapability Shader
OpExtension "SPV_KHR_non_semantic_info"
%1 = OpExtInstImport "NonSemantic.Shader.DebugInfo.100"
OpMemoryModel Logical GLSL450
OpEntryPoint Fragment %MainPs "MainPs" %out_var_SV_Target0
OpExecutionMode %MainPs OriginUpperLeft
%4 = OpString "foo2.frag"
%5 = OpString "
struct PS_OUTPUT
{
float4 vColor : SV_Target0 ;
} ;
"
%6 = OpString "
PS_OUTPUT MainPs ( )
{
PS_OUTPUT ps_output ;
ps_output . vColor = float4( 1.0, 0.0, 0.0, 0.0 );
return ps_output ;
}
"
%7 = OpString "float"
%8 = OpString "vColor"
%9 = OpString "PS_OUTPUT"
%10 = OpString "MainPs"
%11 = OpString ""
%12 = OpString "ps_output"
%13 = OpString "97a939fb"
%14 = OpString " foo2.frag -E MainPs -T ps_6_1 -spirv -fspv-target-env=vulkan1.2 -fspv-debug=vulkan-with-source -fcgl -Fo foo2.frag.nopt.spv -Qembed_debug"
OpName %out_var_SV_Target0 "out.var.SV_Target0"
OpName %MainPs "MainPs"
OpDecorate %out_var_SV_Target0 Location 0
%float = OpTypeFloat 32
%float_1 = OpConstant %float 1
%float_0 = OpConstant %float 0
%v4float = OpTypeVector %float 4
%23 = OpConstantComposite %v4float %float_1 %float_0 %float_0 %float_0
%int = OpTypeInt 32 1
%int_0 = OpConstant %int 0
%uint = OpTypeInt 32 0
%uint_32 = OpConstant %uint 32
%_ptr_Output_v4float = OpTypePointer Output %v4float
%void = OpTypeVoid
%uint_1 = OpConstant %uint 1
%uint_4 = OpConstant %uint 4
%uint_5 = OpConstant %uint 5
%uint_3 = OpConstant %uint 3
%uint_0 = OpConstant %uint 0
%uint_128 = OpConstant %uint 128
%uint_12 = OpConstant %uint 12
%uint_2 = OpConstant %uint 2
%uint_8 = OpConstant %uint 8
%uint_7 = OpConstant %uint 7
%uint_9 = OpConstant %uint 9
%uint_15 = OpConstant %uint 15
%42 = OpTypeFunction %void
%uint_10 = OpConstant %uint 10
%uint_53 = OpConstant %uint 53
%out_var_SV_Target0 = OpVariable %_ptr_Output_v4float Output
%49 = OpExtInst %void %1 DebugExpression
%50 = OpExtInst %void %1 DebugSource %4 %5
%51 = OpExtInst %void %1 DebugSourceContinued %6
; CHECK: %51 = OpExtInst %void %1 DebugSourceContinued %6
%52 = OpExtInst %void %1 DebugCompilationUnit %uint_1 %uint_4 %50 %uint_5
; CHECK: %52 = OpExtInst %void %1 DebugCompilationUnit %uint_1 %uint_4 %50 %uint_5
%53 = OpExtInst %void %1 DebugTypeBasic %7 %uint_32 %uint_3 %uint_0
%54 = OpExtInst %void %1 DebugTypeVector %53 %uint_4
%55 = OpExtInst %void %1 DebugTypeMember %8 %54 %50 %uint_4 %uint_12 %uint_0 %uint_128 %uint_3
%56 = OpExtInst %void %1 DebugTypeComposite %9 %uint_1 %50 %uint_2 %uint_8 %52 %9 %uint_128 %uint_3 %55
%57 = OpExtInst %void %1 DebugTypeFunction %uint_3 %56
%58 = OpExtInst %void %1 DebugFunction %10 %57 %50 %uint_7 %uint_1 %52 %11 %uint_3 %uint_8
%59 = OpExtInst %void %1 DebugLexicalBlock %50 %uint_8 %uint_1 %58
%60 = OpExtInst %void %1 DebugLocalVariable %12 %56 %50 %uint_9 %uint_15 %59 %uint_4
%61 = OpExtInst %void %1 DebugEntryPoint %58 %52 %13 %14
; CHECK: %61 = OpExtInst %void %1 DebugEntryPoint %58 %52 %13 %14
%MainPs = OpFunction %void None %42
%62 = OpLabel
%63 = OpExtInst %void %1 DebugFunctionDefinition %58 %MainPs
%112 = OpExtInst %void %1 DebugScope %59
%111 = OpExtInst %void %1 DebugLine %50 %uint_10 %uint_10 %uint_5 %uint_53
%110 = OpExtInst %void %1 DebugValue %60 %23 %49 %int_0
%113 = OpExtInst %void %1 DebugNoLine
%114 = OpExtInst %void %1 DebugNoScope
OpStore %out_var_SV_Target0 %23
%66 = OpExtInst %void %1 DebugLine %50 %uint_12 %uint_12 %uint_1 %uint_1
OpReturn
OpFunctionEnd
)";
SetTargetEnv(SPV_ENV_VULKAN_1_2);
SetAssembleOptions(SPV_TEXT_TO_BINARY_OPTION_PRESERVE_NUMERIC_IDS);
SinglePassRunAndMatch<AggressiveDCEPass>(text, true);
}
Move tests into namespaces (#1689) This CL moves the test into namespaces based on their directories.
2018-07-11 13:24:49 +00:00
} // namespace
} // namespace opt
} // namespace spvtools
Reference in New Issue Copy Permalink