AuroraMiddleware/SPIRV-Tools
1
0
Fork 0
mirror of https://github.com/KhronosGroup/SPIRV-Tools synced 2024-10-19 03:20:14 +00:00
Code Issues Packages Projects Releases Wiki Activity
532b327d4d
SPIRV-Tools/source/opt/aggressive_dead_code_elim_pass.cpp

474 lines
17 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 The Khronos Group Inc.
// 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.
#include "aggressive_dead_code_elim_pass.h"
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
#include "cfa.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
#include "iterator.h"
#include "spirv/1.0/GLSL.std.450.h"
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
#include <stack>
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 spvtools {
namespace opt {
namespace {
const uint32_t kTypePointerStorageClassInIdx = 0;
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
const uint32_t kEntryPointFunctionIdInIdx = 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
const uint32_t kSelectionMergeMergeBlockIdInIdx = 0;
const uint32_t kLoopMergeMergeBlockIdInIdx = 0;
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
const uint32_t kLoopMergeContinueBlockIdInIdx = 1;
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
Create a local value numbering pass Creates a pass that removes redundant instructions within the same basic block. This will be implemented using a hash based value numbering algorithm. Added a number of functions that check for the Vulkan descriptor types. These are used to determine if we are variables are read-only or not. Implemented a function to check if loads and variables are read-only. Implemented kernel specific and shader specific versions. A big change is that the Combinator analysis in ADCE is factored out into the IRContext as an analysis. This was done because it is being reused in the value number table.
2017-11-11 01:26:55 +00:00
} // namespace
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
Re-format files in source, source/opt, source/util, source/val and tools. NFC. This just makes sure every file is formatted following the formatting definition in .clang-format. Re-formatted with: $ clang-format -i $(find source tools include -name '*.cpp') $ clang-format -i $(find source tools include -name '*.h')
2017-11-08 17:40:02 +00:00
bool AggressiveDCEPass::IsVarOfStorage(uint32_t varId, uint32_t storageClass) {
Re-factor Phi insertion code out of LocalMultiStoreElimPass Including a re-factor of common behaviour into class Pass: The following functions are now in class Pass: - IsLoopHeader. - ComputeStructuredOrder - ComputeStructuredSuccessors (annoyingly, I could not re-factor all instances of this function, the copy in common_uniform_elim_pass.cpp is slightly different and fails with the common implementation). - GetPointeeTypeId - TakeNextId - FinalizeNextId - MergeBlockIdIfAny This is a NFC (non-functional change)
2017-10-25 17:26:25 +00:00
const ir::Instruction* varInst = get_def_use_mgr()->GetDef(varId);
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
const SpvOp op = varInst->opcode();
Re-format files in source, source/opt, source/util, source/val and tools. NFC. This just makes sure every file is formatted following the formatting definition in .clang-format. Re-formatted with: $ clang-format -i $(find source tools include -name '*.cpp') $ clang-format -i $(find source tools include -name '*.h')
2017-11-08 17:40:02 +00:00
if (op != SpvOpVariable) return 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
const uint32_t varTypeId = varInst->type_id();
Re-factor Phi insertion code out of LocalMultiStoreElimPass Including a re-factor of common behaviour into class Pass: The following functions are now in class Pass: - IsLoopHeader. - ComputeStructuredOrder - ComputeStructuredSuccessors (annoyingly, I could not re-factor all instances of this function, the copy in common_uniform_elim_pass.cpp is slightly different and fails with the common implementation). - GetPointeeTypeId - TakeNextId - FinalizeNextId - MergeBlockIdIfAny This is a NFC (non-functional change)
2017-10-25 17:26:25 +00:00
const ir::Instruction* varTypeInst = get_def_use_mgr()->GetDef(varTypeId);
Re-format files in source, source/opt, source/util, source/val and tools. NFC. This just makes sure every file is formatted following the formatting definition in .clang-format. Re-formatted with: $ clang-format -i $(find source tools include -name '*.cpp') $ clang-format -i $(find source tools include -name '*.h')
2017-11-08 17:40:02 +00:00
if (varTypeInst->opcode() != SpvOpTypePointer) return 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
return varTypeInst->GetSingleWordInOperand(kTypePointerStorageClassInIdx) ==
Re-format files in source, source/opt, source/util, source/val and tools. NFC. This just makes sure every file is formatted following the formatting definition in .clang-format. Re-formatted with: $ clang-format -i $(find source tools include -name '*.cpp') $ clang-format -i $(find source tools include -name '*.h')
2017-11-08 17:40:02 +00:00
storageClass;
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
}
bool AggressiveDCEPass::IsLocalVar(uint32_t varId) {
Re-format files in source, source/opt, source/util, source/val and tools. NFC. This just makes sure every file is formatted following the formatting definition in .clang-format. Re-formatted with: $ clang-format -i $(find source tools include -name '*.cpp') $ clang-format -i $(find source tools include -name '*.h')
2017-11-08 17:40:02 +00:00
return IsVarOfStorage(varId, SpvStorageClassFunction) ||
(IsVarOfStorage(varId, SpvStorageClassPrivate) && private_like_local_);
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 AggressiveDCEPass::AddStores(uint32_t ptrId) {
Adding new def -> use mapping container Replaced representation of uses * Changed uses from unordered_map<uint32_t, UseList> to set<pairInstruction*, Instruction*>> * Replaced GetUses with ForEachUser and ForEachUse functions * updated passes to use new functions * partially updated tests * lots of cleanup still todo Adding an unique id to Instruction generated by IRContext Each instruction is given an unique id that can be used for ordering purposes. The ids are generated via the IRContext. Major changes: * Instructions now contain a uint32_t for unique id and a cached context pointer * Most constructors have been modified to take a context as input * unfortunately I cannot remove the default and copy constructors, but developers should avoid these * Added accessors to parents of basic block and function * Removed the copy constructors for BasicBlock and Function and replaced them with Clone functions * Reworked BuildModule to return an IRContext owning the built module * Since all instructions require a context, the context now becomes the basic unit for IR * Added a constructor to context to create an owned module internally * Replaced uses of Instruction's copy constructor with Clone whereever I found them * Reworked the linker functionality to perform clones into a different context instead of moves * Updated many tests to be consistent with the above changes * Still need to add new tests to cover added functionality * Added comparison operators to Instruction Adding tests for Instruction, IRContext and IR loading Fixed some header comments for BuildModule Fixes to get tests passing again * Reordered two linker steps to avoid use/def problems * Fixed def/use manager uses in merge return pass * Added early return for GetAnnotations * Changed uses of Instruction::ToNop in passes to IRContext::KillInst Simplifying the uses for some contexts in passes
2017-11-14 19:11:50 +00:00
get_def_use_mgr()->ForEachUser(ptrId, [this](ir::Instruction* user) {
switch (user->opcode()) {
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
case SpvOpAccessChain:
case SpvOpInBoundsAccessChain:
Adding new def -> use mapping container Replaced representation of uses * Changed uses from unordered_map<uint32_t, UseList> to set<pairInstruction*, Instruction*>> * Replaced GetUses with ForEachUser and ForEachUse functions * updated passes to use new functions * partially updated tests * lots of cleanup still todo Adding an unique id to Instruction generated by IRContext Each instruction is given an unique id that can be used for ordering purposes. The ids are generated via the IRContext. Major changes: * Instructions now contain a uint32_t for unique id and a cached context pointer * Most constructors have been modified to take a context as input * unfortunately I cannot remove the default and copy constructors, but developers should avoid these * Added accessors to parents of basic block and function * Removed the copy constructors for BasicBlock and Function and replaced them with Clone functions * Reworked BuildModule to return an IRContext owning the built module * Since all instructions require a context, the context now becomes the basic unit for IR * Added a constructor to context to create an owned module internally * Replaced uses of Instruction's copy constructor with Clone whereever I found them * Reworked the linker functionality to perform clones into a different context instead of moves * Updated many tests to be consistent with the above changes * Still need to add new tests to cover added functionality * Added comparison operators to Instruction Adding tests for Instruction, IRContext and IR loading Fixed some header comments for BuildModule Fixes to get tests passing again * Reordered two linker steps to avoid use/def problems * Fixed def/use manager uses in merge return pass * Added early return for GetAnnotations * Changed uses of Instruction::ToNop in passes to IRContext::KillInst Simplifying the uses for some contexts in passes
2017-11-14 19:11:50 +00:00
case SpvOpCopyObject:
this->AddStores(user->result_id());
break;
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
case SpvOpLoad:
break;
Adding new def -> use mapping container Replaced representation of uses * Changed uses from unordered_map<uint32_t, UseList> to set<pairInstruction*, Instruction*>> * Replaced GetUses with ForEachUser and ForEachUse functions * updated passes to use new functions * partially updated tests * lots of cleanup still todo Adding an unique id to Instruction generated by IRContext Each instruction is given an unique id that can be used for ordering purposes. The ids are generated via the IRContext. Major changes: * Instructions now contain a uint32_t for unique id and a cached context pointer * Most constructors have been modified to take a context as input * unfortunately I cannot remove the default and copy constructors, but developers should avoid these * Added accessors to parents of basic block and function * Removed the copy constructors for BasicBlock and Function and replaced them with Clone functions * Reworked BuildModule to return an IRContext owning the built module * Since all instructions require a context, the context now becomes the basic unit for IR * Added a constructor to context to create an owned module internally * Replaced uses of Instruction's copy constructor with Clone whereever I found them * Reworked the linker functionality to perform clones into a different context instead of moves * Updated many tests to be consistent with the above changes * Still need to add new tests to cover added functionality * Added comparison operators to Instruction Adding tests for Instruction, IRContext and IR loading Fixed some header comments for BuildModule Fixes to get tests passing again * Reordered two linker steps to avoid use/def problems * Fixed def/use manager uses in merge return pass * Added early return for GetAnnotations * Changed uses of Instruction::ToNop in passes to IRContext::KillInst Simplifying the uses for some contexts in passes
2017-11-14 19:11:50 +00:00
// If default, assume it stores e.g. frexp, modf, function call
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
case SpvOpStore:
Adding new def -> use mapping container Replaced representation of uses * Changed uses from unordered_map<uint32_t, UseList> to set<pairInstruction*, Instruction*>> * Replaced GetUses with ForEachUser and ForEachUse functions * updated passes to use new functions * partially updated tests * lots of cleanup still todo Adding an unique id to Instruction generated by IRContext Each instruction is given an unique id that can be used for ordering purposes. The ids are generated via the IRContext. Major changes: * Instructions now contain a uint32_t for unique id and a cached context pointer * Most constructors have been modified to take a context as input * unfortunately I cannot remove the default and copy constructors, but developers should avoid these * Added accessors to parents of basic block and function * Removed the copy constructors for BasicBlock and Function and replaced them with Clone functions * Reworked BuildModule to return an IRContext owning the built module * Since all instructions require a context, the context now becomes the basic unit for IR * Added a constructor to context to create an owned module internally * Replaced uses of Instruction's copy constructor with Clone whereever I found them * Reworked the linker functionality to perform clones into a different context instead of moves * Updated many tests to be consistent with the above changes * Still need to add new tests to cover added functionality * Added comparison operators to Instruction Adding tests for Instruction, IRContext and IR loading Fixed some header comments for BuildModule Fixes to get tests passing again * Reordered two linker steps to avoid use/def problems * Fixed def/use manager uses in merge return pass * Added early return for GetAnnotations * Changed uses of Instruction::ToNop in passes to IRContext::KillInst Simplifying the uses for some contexts in passes
2017-11-14 19:11:50 +00:00
default:
if (!IsLive(user)) AddToWorklist(user);
break;
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
}
Adding new def -> use mapping container Replaced representation of uses * Changed uses from unordered_map<uint32_t, UseList> to set<pairInstruction*, Instruction*>> * Replaced GetUses with ForEachUser and ForEachUse functions * updated passes to use new functions * partially updated tests * lots of cleanup still todo Adding an unique id to Instruction generated by IRContext Each instruction is given an unique id that can be used for ordering purposes. The ids are generated via the IRContext. Major changes: * Instructions now contain a uint32_t for unique id and a cached context pointer * Most constructors have been modified to take a context as input * unfortunately I cannot remove the default and copy constructors, but developers should avoid these * Added accessors to parents of basic block and function * Removed the copy constructors for BasicBlock and Function and replaced them with Clone functions * Reworked BuildModule to return an IRContext owning the built module * Since all instructions require a context, the context now becomes the basic unit for IR * Added a constructor to context to create an owned module internally * Replaced uses of Instruction's copy constructor with Clone whereever I found them * Reworked the linker functionality to perform clones into a different context instead of moves * Updated many tests to be consistent with the above changes * Still need to add new tests to cover added functionality * Added comparison operators to Instruction Adding tests for Instruction, IRContext and IR loading Fixed some header comments for BuildModule Fixes to get tests passing again * Reordered two linker steps to avoid use/def problems * Fixed def/use manager uses in merge return pass * Added early return for GetAnnotations * Changed uses of Instruction::ToNop in passes to IRContext::KillInst Simplifying the uses for some contexts in passes
2017-11-14 19:11:50 +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
}
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
bool AggressiveDCEPass::AllExtensionsSupported() const {
// If any extension not in whitelist, return false
Re-factor Phi insertion code out of LocalMultiStoreElimPass Including a re-factor of common behaviour into class Pass: The following functions are now in class Pass: - IsLoopHeader. - ComputeStructuredOrder - ComputeStructuredSuccessors (annoyingly, I could not re-factor all instances of this function, the copy in common_uniform_elim_pass.cpp is slightly different and fails with the common implementation). - GetPointeeTypeId - TakeNextId - FinalizeNextId - MergeBlockIdIfAny This is a NFC (non-functional change)
2017-10-25 17:26:25 +00:00
for (auto& ei : get_module()->extensions()) {
Re-format files in source, source/opt, source/util, source/val and tools. NFC. This just makes sure every file is formatted following the formatting definition in .clang-format. Re-formatted with: $ clang-format -i $(find source tools include -name '*.cpp') $ clang-format -i $(find source tools include -name '*.h')
2017-11-08 17:40:02 +00:00
const char* extName =
reinterpret_cast<const char*>(&ei.GetInOperand(0).words[0]);
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
if (extensions_whitelist_.find(extName) == extensions_whitelist_.end())
return 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
}
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
return 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
}
Change IRContext::KillInst to delete instructions. The current method of removing an instruction is to call ToNop. The problem with this is that it leaves around an instruction that later passes will look at. We should just delete the instruction. In MemPass there is a utility routine called DCEInst. It can delete essentially any instruction, which can invalidate pointers now that they are actually deleted. The interface was changed to add a call back that can be used to update any local data structures that contain ir::Intruction*.
2017-11-21 19:47:46 +00:00
bool AggressiveDCEPass::IsTargetDead(ir::Instruction* inst) {
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 uint32_t tId = inst->GetSingleWordInOperand(0);
Re-factor Phi insertion code out of LocalMultiStoreElimPass Including a re-factor of common behaviour into class Pass: The following functions are now in class Pass: - IsLoopHeader. - ComputeStructuredOrder - ComputeStructuredSuccessors (annoyingly, I could not re-factor all instances of this function, the copy in common_uniform_elim_pass.cpp is slightly different and fails with the common implementation). - GetPointeeTypeId - TakeNextId - FinalizeNextId - MergeBlockIdIfAny This is a NFC (non-functional change)
2017-10-25 17:26:25 +00:00
const ir::Instruction* tInst = get_def_use_mgr()->GetDef(tId);
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
if (dead_insts_.find(tInst) != dead_insts_.end()) {
return true;
}
return false;
}
void AggressiveDCEPass::ProcessLoad(uint32_t varId) {
// Only process locals
Re-format files in source, source/opt, source/util, source/val and tools. NFC. This just makes sure every file is formatted following the formatting definition in .clang-format. Re-formatted with: $ clang-format -i $(find source tools include -name '*.cpp') $ clang-format -i $(find source tools include -name '*.h')
2017-11-08 17:40:02 +00:00
if (!IsLocalVar(varId)) return;
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
// Return if already processed
Re-format files in source, source/opt, source/util, source/val and tools. NFC. This just makes sure every file is formatted following the formatting definition in .clang-format. Re-formatted with: $ clang-format -i $(find source tools include -name '*.cpp') $ clang-format -i $(find source tools include -name '*.h')
2017-11-08 17:40:02 +00:00
if (live_local_vars_.find(varId) != live_local_vars_.end()) return;
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
// Mark all stores to varId as live
AddStores(varId);
// Cache varId as processed
live_local_vars_.insert(varId);
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
}
ADCE: Empty Loop Elimination This entirely eliminates loops which do not contain live code.
2017-10-25 00:58:48 +00:00
bool AggressiveDCEPass::IsStructuredIfOrLoopHeader(ir::BasicBlock* bp,
ir::Instruction** mergeInst,
ir::Instruction** branchInst,
uint32_t* mergeBlockId) {
ir::Instruction* mi = bp->GetMergeInst();
if (mi == nullptr) return false;
ir::Instruction* bri = &*bp->tail();
// Make sure it is not a Switch
if (mi->opcode() == SpvOpSelectionMerge &&
bri->opcode() != SpvOpBranchConditional)
return false;
if (branchInst != nullptr) *branchInst = bri;
if (mergeInst != nullptr) *mergeInst = mi;
if (mergeBlockId != nullptr) *mergeBlockId = mi->GetSingleWordInOperand(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
return true;
}
void AggressiveDCEPass::ComputeBlock2HeaderMaps(
Re-format files in source, source/opt, source/util, source/val and tools. NFC. This just makes sure every file is formatted following the formatting definition in .clang-format. Re-formatted with: $ clang-format -i $(find source tools include -name '*.cpp') $ clang-format -i $(find source tools include -name '*.h')
2017-11-08 17:40:02 +00:00
std::list<ir::BasicBlock*>& structuredOrder) {
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
block2headerMerge_.clear();
block2headerBranch_.clear();
std::stack<ir::Instruction*> currentMergeInst;
std::stack<ir::Instruction*> currentBranchInst;
std::stack<uint32_t> currentMergeBlockId;
currentMergeInst.push(nullptr);
currentBranchInst.push(nullptr);
currentMergeBlockId.push(0);
for (auto bi = structuredOrder.begin(); bi != structuredOrder.end(); ++bi) {
ADCE: Empty Loop Elimination This entirely eliminates loops which do not contain live code.
2017-10-25 00:58:48 +00:00
// If leaving an if or loop, update stacks
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
if ((*bi)->id() == currentMergeBlockId.top()) {
currentMergeBlockId.pop();
currentMergeInst.pop();
currentBranchInst.pop();
}
ir::Instruction* mergeInst;
ir::Instruction* branchInst;
uint32_t mergeBlockId;
ADCE: Empty Loop Elimination This entirely eliminates loops which do not contain live code.
2017-10-25 00:58:48 +00:00
bool is_header =
IsStructuredIfOrLoopHeader(*bi, &mergeInst, &branchInst, &mergeBlockId);
// If there is live code in loop header, the loop is live
if (is_header && mergeInst->opcode() == SpvOpLoopMerge) {
currentMergeBlockId.push(mergeBlockId);
currentMergeInst.push(mergeInst);
currentBranchInst.push(branchInst);
}
block2headerMerge_[*bi] = currentMergeInst.top();
block2headerBranch_[*bi] = currentBranchInst.top();
// If there is live code following if header, the if is live
if (is_header && mergeInst->opcode() == SpvOpSelectionMerge) {
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
currentMergeBlockId.push(mergeBlockId);
currentMergeInst.push(mergeInst);
currentBranchInst.push(branchInst);
}
}
}
void AggressiveDCEPass::ComputeInst2BlockMap(ir::Function* func) {
for (auto& blk : *func) {
Re-format files in source, source/opt, source/util, source/val and tools. NFC. This just makes sure every file is formatted following the formatting definition in .clang-format. Re-formatted with: $ clang-format -i $(find source tools include -name '*.cpp') $ clang-format -i $(find source tools include -name '*.h')
2017-11-08 17:40:02 +00:00
blk.ForEachInst(
[&blk, this](ir::Instruction* ip) { inst2block_[ip] = &blk; });
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 AggressiveDCEPass::AddBranch(uint32_t labelId, ir::BasicBlock* bp) {
Re-format files in source, source/opt, source/util, source/val and tools. NFC. This just makes sure every file is formatted following the formatting definition in .clang-format. Re-formatted with: $ clang-format -i $(find source tools include -name '*.cpp') $ clang-format -i $(find source tools include -name '*.h')
2017-11-08 17:40:02 +00:00
std::unique_ptr<ir::Instruction> newBranch(new ir::Instruction(
Adding an unique id to Instruction generated by IRContext Each instruction is given an unique id that can be used for ordering purposes. The ids are generated via the IRContext. Major changes: * Instructions now contain a uint32_t for unique id and a cached context pointer * Most constructors have been modified to take a context as input * unfortunately I cannot remove the default and copy constructors, but developers should avoid these * Added accessors to parents of basic block and function * Removed the copy constructors for BasicBlock and Function and replaced them with Clone functions * Reworked BuildModule to return an IRContext owning the built module * Since all instructions require a context, the context now becomes the basic unit for IR * Added a constructor to context to create an owned module internally * Replaced uses of Instruction's copy constructor with Clone whereever I found them * Reworked the linker functionality to perform clones into a different context instead of moves * Updated many tests to be consistent with the above changes * Still need to add new tests to cover added functionality * Added comparison operators to Instruction * Added an internal option to LinkerOptions to verify merged ids are unique * Added a test for the linker to verify merged ids are unique * Updated MergeReturnPass to supply a context * Updated DecorationManager to supply a context for cloned decorations * Reworked several portions of the def use tests in anticipation of next set of changes
2017-11-14 19:11:50 +00:00
context(), SpvOpBranch, 0, 0,
Re-format files in source, source/opt, source/util, source/val and tools. NFC. This just makes sure every file is formatted following the formatting definition in .clang-format. Re-formatted with: $ clang-format -i $(find source tools include -name '*.cpp') $ clang-format -i $(find source tools include -name '*.h')
2017-11-08 17:40:02 +00:00
{{spv_operand_type_t::SPV_OPERAND_TYPE_ID, {labelId}}}));
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
get_def_use_mgr()->AnalyzeInstDefUse(&*newBranch);
bp->AddInstruction(std::move(newBranch));
}
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 AggressiveDCEPass::AddBranchesToWorklist(uint32_t labelId) {
get_def_use_mgr()->ForEachUser(labelId, [this](ir::Instruction* user) {
SpvOp op = user->opcode();
if (op == SpvOpBranchConditional || op == SpvOpBranch)
if (!IsLive(user)) AddToWorklist(user);
});
}
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
bool AggressiveDCEPass::AggressiveDCE(ir::Function* func) {
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
// Compute map from instruction to block
ComputeInst2BlockMap(func);
// Compute map from block to controlling conditional branch
std::list<ir::BasicBlock*> structuredOrder;
Add a new class opt::CFG to represent the CFG for the module. This class moves some of the CFG-related functionality into a new class opt::CFG. There is some other code related to the CFG in the inliner and in opt::LocalSingleStoreElimPass that should also be moved, but that require more changes than this pure restructuring. I will move those bits in a follow-up PR. Currently, the CFG is computed every time a pass is instantiated, but this should be later moved to the new IRContext class that @s-perron is working on. Other re-factoring: - Add BasicBlock::ContinueBlockIdIfAny. Re-factored out of MergeBlockIdIfAny - Rewrite IsLoopHeader in terms of GetLoopMergeInst. - Run clang-format on some files.
2017-10-30 21:42:26 +00:00
cfg()->ComputeStructuredOrder(func, &*func->begin(), &structuredOrder);
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
ComputeBlock2HeaderMaps(structuredOrder);
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
bool modified = false;
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
// Add instructions with external side effects to worklist. Also add branches
ADCE: Empty Loop Elimination This entirely eliminates loops which do not contain live code.
2017-10-25 00:58:48 +00:00
// EXCEPT those immediately contained in an "if" selection construct or a loop
// or continue construct.
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
// TODO(greg-lunarg): Handle Frexp, Modf more optimally
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
call_in_func_ = false;
func_is_entry_point_ = false;
private_stores_.clear();
ADCE: Empty Loop Elimination This entirely eliminates loops which do not contain live code.
2017-10-25 00:58:48 +00:00
// Stacks to keep track of when we are inside an if- or loop-construct.
// When not immediately inside an if- or loop-construct, we must assume
// all branches are live as we may be inside of a control construct (ie
// switch) which is not part of the ADCE analysis.
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
std::stack<bool> assume_branches_live;
std::stack<uint32_t> currentMergeBlockId;
// Push sentinel values on stack for when outside of any control flow.
assume_branches_live.push(true);
currentMergeBlockId.push(0);
for (auto bi = structuredOrder.begin(); bi != structuredOrder.end(); ++bi) {
ADCE: Empty Loop Elimination This entirely eliminates loops which do not contain live code.
2017-10-25 00:58:48 +00:00
// If exiting if or loop, update stacks
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
if ((*bi)->id() == currentMergeBlockId.top()) {
assume_branches_live.pop();
currentMergeBlockId.pop();
}
for (auto ii = (*bi)->begin(); ii != (*bi)->end(); ++ii) {
SpvOp op = ii->opcode();
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
switch (op) {
case SpvOpStore: {
uint32_t varId;
Re-format files in source, source/opt, source/util, source/val and tools. NFC. This just makes sure every file is formatted following the formatting definition in .clang-format. Re-formatted with: $ clang-format -i $(find source tools include -name '*.cpp') $ clang-format -i $(find source tools include -name '*.h')
2017-11-08 17:40:02 +00:00
(void)GetPtr(&*ii, &varId);
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
// Mark stores as live if their variable is not function scope
// and is not private scope. Remember private stores for possible
// later inclusion
if (IsVarOfStorage(varId, SpvStorageClassPrivate))
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
private_stores_.push_back(&*ii);
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
else if (!IsVarOfStorage(varId, SpvStorageClassFunction))
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
AddToWorklist(&*ii);
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
} break;
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
case SpvOpLoopMerge: {
ADCE: Empty Loop Elimination This entirely eliminates loops which do not contain live code.
2017-10-25 00:58:48 +00:00
assume_branches_live.push(false);
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
currentMergeBlockId.push(
ii->GetSingleWordInOperand(kLoopMergeMergeBlockIdInIdx));
} break;
case SpvOpSelectionMerge: {
auto brii = ii;
++brii;
bool is_structured_if = brii->opcode() == SpvOpBranchConditional;
assume_branches_live.push(!is_structured_if);
currentMergeBlockId.push(
ii->GetSingleWordInOperand(kSelectionMergeMergeBlockIdInIdx));
Re-format files in source, source/opt, source/util, source/val and tools. NFC. This just makes sure every file is formatted following the formatting definition in .clang-format. Re-formatted with: $ clang-format -i $(find source tools include -name '*.cpp') $ clang-format -i $(find source tools include -name '*.h')
2017-11-08 17:40:02 +00:00
if (!is_structured_if) AddToWorklist(&*ii);
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
} break;
case SpvOpBranch:
case SpvOpBranchConditional: {
Re-format files in source, source/opt, source/util, source/val and tools. NFC. This just makes sure every file is formatted following the formatting definition in .clang-format. Re-formatted with: $ clang-format -i $(find source tools include -name '*.cpp') $ clang-format -i $(find source tools include -name '*.h')
2017-11-08 17:40:02 +00:00
if (assume_branches_live.top()) AddToWorklist(&*ii);
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
} break;
default: {
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
// Function calls, atomics, function params, function returns, etc.
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
// TODO(greg-lunarg): function calls live only if write to non-local
Create a local value numbering pass Creates a pass that removes redundant instructions within the same basic block. This will be implemented using a hash based value numbering algorithm. Added a number of functions that check for the Vulkan descriptor types. These are used to determine if we are variables are read-only or not. Implemented a function to check if loads and variables are read-only. Implemented kernel specific and shader specific versions. A big change is that the Combinator analysis in ADCE is factored out into the IRContext as an analysis. This was done because it is being reused in the value number table.
2017-11-11 01:26:55 +00:00
if (!context()->IsCombinatorInstruction(&*ii)) {
AddToWorklist(&*ii);
}
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
// Remember function calls
Re-format files in source, source/opt, source/util, source/val and tools. NFC. This just makes sure every file is formatted following the formatting definition in .clang-format. Re-formatted with: $ clang-format -i $(find source tools include -name '*.cpp') $ clang-format -i $(find source tools include -name '*.h')
2017-11-08 17:40:02 +00:00
if (op == SpvOpFunctionCall) call_in_func_ = 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
} break;
}
}
}
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
// See if current function is an entry point
Re-factor Phi insertion code out of LocalMultiStoreElimPass Including a re-factor of common behaviour into class Pass: The following functions are now in class Pass: - IsLoopHeader. - ComputeStructuredOrder - ComputeStructuredSuccessors (annoyingly, I could not re-factor all instances of this function, the copy in common_uniform_elim_pass.cpp is slightly different and fails with the common implementation). - GetPointeeTypeId - TakeNextId - FinalizeNextId - MergeBlockIdIfAny This is a NFC (non-functional change)
2017-10-25 17:26:25 +00:00
for (auto& ei : get_module()->entry_points()) {
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
if (ei.GetSingleWordInOperand(kEntryPointFunctionIdInIdx) ==
func->result_id()) {
func_is_entry_point_ = true;
break;
}
}
// If the current function is an entry point and has no function calls,
// we can optimize private variables as locals
private_like_local_ = func_is_entry_point_ && !call_in_func_;
// If privates are not like local, add their stores to worklist
if (!private_like_local_)
Re-format files in source, source/opt, source/util, source/val and tools. NFC. This just makes sure every file is formatted following the formatting definition in .clang-format. Re-formatted with: $ clang-format -i $(find source tools include -name '*.cpp') $ clang-format -i $(find source tools include -name '*.h')
2017-11-08 17:40:02 +00:00
for (auto& ps : private_stores_) AddToWorklist(ps);
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
// Add OpGroupDecorates to worklist because they are a pain to remove
// ids from.
// TODO(greg-lunarg): Handle dead ids in OpGroupDecorate
Re-factor Phi insertion code out of LocalMultiStoreElimPass Including a re-factor of common behaviour into class Pass: The following functions are now in class Pass: - IsLoopHeader. - ComputeStructuredOrder - ComputeStructuredSuccessors (annoyingly, I could not re-factor all instances of this function, the copy in common_uniform_elim_pass.cpp is slightly different and fails with the common implementation). - GetPointeeTypeId - TakeNextId - FinalizeNextId - MergeBlockIdIfAny This is a NFC (non-functional change)
2017-10-25 17:26:25 +00:00
for (auto& ai : get_module()->annotations()) {
Re-format files in source, source/opt, source/util, source/val and tools. NFC. This just makes sure every file is formatted following the formatting definition in .clang-format. Re-formatted with: $ clang-format -i $(find source tools include -name '*.cpp') $ clang-format -i $(find source tools include -name '*.h')
2017-11-08 17:40:02 +00:00
if (ai.opcode() == SpvOpGroupDecorate) AddToWorklist(&ai);
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
}
Re-format files in source, source/opt, source/util, source/val and tools. NFC. This just makes sure every file is formatted following the formatting definition in .clang-format. Re-formatted with: $ clang-format -i $(find source tools include -name '*.cpp') $ clang-format -i $(find source tools include -name '*.h')
2017-11-08 17:40:02 +00:00
// Perform closure on live instruction set.
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
while (!worklist_.empty()) {
ir::Instruction* liveInst = worklist_.front();
// Add all operand instructions if not already live
ADCE: Empty Loop Elimination This entirely eliminates loops which do not contain live code.
2017-10-25 00:58:48 +00:00
liveInst->ForEachInId([&liveInst, this](const uint32_t* iid) {
Re-factor Phi insertion code out of LocalMultiStoreElimPass Including a re-factor of common behaviour into class Pass: The following functions are now in class Pass: - IsLoopHeader. - ComputeStructuredOrder - ComputeStructuredSuccessors (annoyingly, I could not re-factor all instances of this function, the copy in common_uniform_elim_pass.cpp is slightly different and fails with the common implementation). - GetPointeeTypeId - TakeNextId - FinalizeNextId - MergeBlockIdIfAny This is a NFC (non-functional change)
2017-10-25 17:26:25 +00:00
ir::Instruction* inInst = get_def_use_mgr()->GetDef(*iid);
ADCE: Empty Loop Elimination This entirely eliminates loops which do not contain live code.
2017-10-25 00:58:48 +00:00
// Do not add label if an operand of a branch. This is not needed
// as part of live code discovery and can create false live code,
// for example, the branch to a header of a loop.
if (inInst->opcode() == SpvOpLabel && liveInst->IsBranch()) return;
Re-format files in source, source/opt, source/util, source/val and tools. NFC. This just makes sure every file is formatted following the formatting definition in .clang-format. Re-formatted with: $ clang-format -i $(find source tools include -name '*.cpp') $ clang-format -i $(find source tools include -name '*.h')
2017-11-08 17:40:02 +00:00
if (!IsLive(inInst)) AddToWorklist(inInst);
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
});
ADCE: Empty Loop Elimination This entirely eliminates loops which do not contain live code.
2017-10-25 00:58:48 +00:00
// If in a structured if or loop construct, add the controlling
// conditional branch and its merge. Any containing control construct
// is marked live when the merge and branch are processed out of the
// worklist.
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
ir::BasicBlock* blk = inst2block_[liveInst];
ir::Instruction* branchInst = block2headerBranch_[blk];
if (branchInst != nullptr && !IsLive(branchInst)) {
AddToWorklist(branchInst);
ADCE: Empty Loop Elimination This entirely eliminates loops which do not contain live code.
2017-10-25 00:58:48 +00:00
ir::Instruction* mergeInst = block2headerMerge_[blk];
AddToWorklist(mergeInst);
// If in a loop, mark all branches targeting merge block
// and continue block as live.
if (mergeInst->opcode() == SpvOpLoopMerge) {
AddBranchesToWorklist(
mergeInst->GetSingleWordInOperand(kLoopMergeMergeBlockIdInIdx));
AddBranchesToWorklist(
mergeInst->GetSingleWordInOperand(kLoopMergeContinueBlockIdInIdx));
}
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
}
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
// If local load, add all variable's stores if variable not already live
if (liveInst->opcode() == SpvOpLoad) {
uint32_t varId;
Re-format files in source, source/opt, source/util, source/val and tools. NFC. This just makes sure every file is formatted following the formatting definition in .clang-format. Re-formatted with: $ clang-format -i $(find source tools include -name '*.cpp') $ clang-format -i $(find source tools include -name '*.h')
2017-11-08 17:40:02 +00:00
(void)GetPtr(liveInst, &varId);
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
ProcessLoad(varId);
}
// If function call, treat as if it loads from all pointer arguments
else if (liveInst->opcode() == SpvOpFunctionCall) {
liveInst->ForEachInId([this](const uint32_t* iid) {
// Skip non-ptr args
if (!IsPtr(*iid)) return;
uint32_t varId;
Re-format files in source, source/opt, source/util, source/val and tools. NFC. This just makes sure every file is formatted following the formatting definition in .clang-format. Re-formatted with: $ clang-format -i $(find source tools include -name '*.cpp') $ clang-format -i $(find source tools include -name '*.h')
2017-11-08 17:40:02 +00:00
(void)GetPtr(*iid, &varId);
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
ProcessLoad(varId);
});
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
}
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
// If function parameter, treat as if it's result id is loaded from
else if (liveInst->opcode() == SpvOpFunctionParameter) {
ProcessLoad(liveInst->result_id());
}
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
worklist_.pop();
}
ADCE: Empty Loop Elimination This entirely eliminates loops which do not contain live code.
2017-10-25 00:58:48 +00:00
// Mark all non-live instructions dead except non-structured branches, which
// now should be considered live unless their block is deleted.
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
for (auto bi = structuredOrder.begin(); bi != structuredOrder.end(); ++bi) {
for (auto ii = (*bi)->begin(); ii != (*bi)->end(); ++ii) {
Re-format files in source, source/opt, source/util, source/val and tools. NFC. This just makes sure every file is formatted following the formatting definition in .clang-format. Re-formatted with: $ clang-format -i $(find source tools include -name '*.cpp') $ clang-format -i $(find source tools include -name '*.h')
2017-11-08 17:40:02 +00:00
if (IsLive(&*ii)) continue;
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
if (ii->IsBranch() &&
ADCE: Empty Loop Elimination This entirely eliminates loops which do not contain live code.
2017-10-25 00:58:48 +00:00
!IsStructuredIfOrLoopHeader(*bi, nullptr, nullptr, nullptr))
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
continue;
dead_insts_.insert(&*ii);
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
}
}
// Remove debug and annotation statements referencing dead instructions.
// This must be done before killing the instructions, otherwise there are
// dead objects in the def/use database.
Change IRContext::KillInst to delete instructions. The current method of removing an instruction is to call ToNop. The problem with this is that it leaves around an instruction that later passes will look at. We should just delete the instruction. In MemPass there is a utility routine called DCEInst. It can delete essentially any instruction, which can invalidate pointers now that they are actually deleted. The interface was changed to add a call back that can be used to update any local data structures that contain ir::Intruction*.
2017-11-21 19:47:46 +00:00
ir::Instruction* instruction = &*get_module()->debug2_begin();
while (instruction) {
if (instruction->opcode() != SpvOpName) {
instruction = instruction->NextNode();
continue;
}
if (IsTargetDead(instruction)) {
instruction = context()->KillInst(instruction);
modified = true;
} else {
instruction = instruction->NextNode();
}
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
}
Change IRContext::KillInst to delete instructions. The current method of removing an instruction is to call ToNop. The problem with this is that it leaves around an instruction that later passes will look at. We should just delete the instruction. In MemPass there is a utility routine called DCEInst. It can delete essentially any instruction, which can invalidate pointers now that they are actually deleted. The interface was changed to add a call back that can be used to update any local data structures that contain ir::Intruction*.
2017-11-21 19:47:46 +00:00
instruction = &*get_module()->annotation_begin();
while (instruction) {
if (instruction->opcode() != SpvOpDecorate &&
instruction->opcode() != SpvOpDecorateId) {
instruction = instruction->NextNode();
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
continue;
Change IRContext::KillInst to delete instructions. The current method of removing an instruction is to call ToNop. The problem with this is that it leaves around an instruction that later passes will look at. We should just delete the instruction. In MemPass there is a utility routine called DCEInst. It can delete essentially any instruction, which can invalidate pointers now that they are actually deleted. The interface was changed to add a call back that can be used to update any local data structures that contain ir::Intruction*.
2017-11-21 19:47:46 +00:00
}
if (IsTargetDead(instruction)) {
instruction = context()->KillInst(instruction);
modified = true;
} else {
instruction = instruction->NextNode();
}
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
}
Change IRContext::KillInst to delete instructions. The current method of removing an instruction is to call ToNop. The problem with this is that it leaves around an instruction that later passes will look at. We should just delete the instruction. In MemPass there is a utility routine called DCEInst. It can delete essentially any instruction, which can invalidate pointers now that they are actually deleted. The interface was changed to add a call back that can be used to update any local data structures that contain ir::Intruction*.
2017-11-21 19:47:46 +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
// Kill dead instructions and remember dead blocks
for (auto bi = structuredOrder.begin(); bi != structuredOrder.end();) {
uint32_t mergeBlockId = 0;
Change IRContext::KillInst to delete instructions. The current method of removing an instruction is to call ToNop. The problem with this is that it leaves around an instruction that later passes will look at. We should just delete the instruction. In MemPass there is a utility routine called DCEInst. It can delete essentially any instruction, which can invalidate pointers now that they are actually deleted. The interface was changed to add a call back that can be used to update any local data structures that contain ir::Intruction*.
2017-11-21 19:47:46 +00:00
(*bi)->ForEachInst([this, &modified, &mergeBlockId](ir::Instruction* inst) {
if (dead_insts_.find(inst) == dead_insts_.end()) return;
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
// If dead instruction is selection merge, remember merge block
// for new branch at end of block
ADCE: Empty Loop Elimination This entirely eliminates loops which do not contain live code.
2017-10-25 00:58:48 +00:00
if (inst->opcode() == SpvOpSelectionMerge ||
inst->opcode() == SpvOpLoopMerge)
mergeBlockId = inst->GetSingleWordInOperand(0);
Change IRContext::KillInst to delete instructions. The current method of removing an instruction is to call ToNop. The problem with this is that it leaves around an instruction that later passes will look at. We should just delete the instruction. In MemPass there is a utility routine called DCEInst. It can delete essentially any instruction, which can invalidate pointers now that they are actually deleted. The interface was changed to add a call back that can be used to update any local data structures that contain ir::Intruction*.
2017-11-21 19:47:46 +00:00
context()->KillInst(inst);
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
modified = true;
Change IRContext::KillInst to delete instructions. The current method of removing an instruction is to call ToNop. The problem with this is that it leaves around an instruction that later passes will look at. We should just delete the instruction. In MemPass there is a utility routine called DCEInst. It can delete essentially any instruction, which can invalidate pointers now that they are actually deleted. The interface was changed to add a call back that can be used to update any local data structures that contain ir::Intruction*.
2017-11-21 19:47:46 +00:00
});
ADCE: Empty Loop Elimination This entirely eliminates loops which do not contain live code.
2017-10-25 00:58:48 +00:00
// If a structured if or loop was deleted, add a branch to its merge
// block, and traverse to the merge block and continue processing there.
// We know the block still exists because the label is not deleted.
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
if (mergeBlockId != 0) {
AddBranch(mergeBlockId, *bi);
for (++bi; (*bi)->id() != mergeBlockId; ++bi) {
}
Re-format files in source, source/opt, source/util, source/val and tools. NFC. This just makes sure every file is formatted following the formatting definition in .clang-format. Re-formatted with: $ clang-format -i $(find source tools include -name '*.cpp') $ clang-format -i $(find source tools include -name '*.h')
2017-11-08 17:40:02 +00:00
} else {
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
++bi;
}
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
}
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
// Cleanup all CFG including all unreachable blocks
CFGCleanup(func);
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
return modified;
}
Add the IRContext (part 1) This is the first part of adding the IRContext. This class is meant to hold the extra data that is build on top of the module that it owns. The first part will simply create the IRContext class and get it passed to the passes in place of the module. For now it does not have any functionality of its own, but it acts more as a wrapper for the module. The functions that I added to the IRContext are those that either traverse the headers or add to them. I did this because we may decide to have other ways of dealing with these sections (for example adding a type pool, or use the decoration manager). I also added the function that add to the header because the IRContext needs to know when an instruction is added to update other data structures appropriately. Note that there is still lots of work that needs to be done. There are still many places that change the module, and do not inform the context. That will be the next step.
2017-10-30 15:13:24 +00:00
void AggressiveDCEPass::Initialize(ir::IRContext* c) {
InitializeProcessing(c);
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
Minor code review feedback on AggressiveDCE
2017-07-10 15:45:59 +00:00
// Clear collections
worklist_ = std::queue<ir::Instruction*>{};
live_insts_.clear();
live_local_vars_.clear();
dead_insts_.clear();
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
// Initialize extensions whitelist
InitExtensions();
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
}
Pass::Status AggressiveDCEPass::ProcessImpl() {
Re-format files in source, source/opt, source/util, source/val and tools. NFC. This just makes sure every file is formatted following the formatting definition in .clang-format. Re-formatted with: $ clang-format -i $(find source tools include -name '*.cpp') $ clang-format -i $(find source tools include -name '*.h')
2017-11-08 17:40:02 +00:00
// Current functionality assumes shader capability
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
// TODO(greg-lunarg): Handle additional capabilities
Re-factor Phi insertion code out of LocalMultiStoreElimPass Including a re-factor of common behaviour into class Pass: The following functions are now in class Pass: - IsLoopHeader. - ComputeStructuredOrder - ComputeStructuredSuccessors (annoyingly, I could not re-factor all instances of this function, the copy in common_uniform_elim_pass.cpp is slightly different and fails with the common implementation). - GetPointeeTypeId - TakeNextId - FinalizeNextId - MergeBlockIdIfAny This is a NFC (non-functional change)
2017-10-25 17:26:25 +00:00
if (!get_module()->HasCapability(SpvCapabilityShader))
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
return Status::SuccessWithoutChange;
// Current functionality assumes logical addressing only
// TODO(greg-lunarg): Handle non-logical addressing
Re-factor Phi insertion code out of LocalMultiStoreElimPass Including a re-factor of common behaviour into class Pass: The following functions are now in class Pass: - IsLoopHeader. - ComputeStructuredOrder - ComputeStructuredSuccessors (annoyingly, I could not re-factor all instances of this function, the copy in common_uniform_elim_pass.cpp is slightly different and fails with the common implementation). - GetPointeeTypeId - TakeNextId - FinalizeNextId - MergeBlockIdIfAny This is a NFC (non-functional change)
2017-10-25 17:26:25 +00:00
if (get_module()->HasCapability(SpvCapabilityAddresses))
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
return Status::SuccessWithoutChange;
Minor code review feedback on AggressiveDCE
2017-07-10 15:45:59 +00:00
// If any extensions in the module are not explicitly supported,
Re-format files in source, source/opt, source/util, source/val and tools. NFC. This just makes sure every file is formatted following the formatting definition in .clang-format. Re-formatted with: $ clang-format -i $(find source tools include -name '*.cpp') $ clang-format -i $(find source tools include -name '*.h')
2017-11-08 17:40:02 +00:00
// return unmodified.
if (!AllExtensionsSupported()) return Status::SuccessWithoutChange;
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
// Process all entry point functions
Re-format files in source, source/opt, source/util, source/val and tools. NFC. This just makes sure every file is formatted following the formatting definition in .clang-format. Re-formatted with: $ clang-format -i $(find source tools include -name '*.cpp') $ clang-format -i $(find source tools include -name '*.h')
2017-11-08 17:40:02 +00:00
ProcessFunction pfn = [this](ir::Function* fp) { return AggressiveDCE(fp); };
Re-factor Phi insertion code out of LocalMultiStoreElimPass Including a re-factor of common behaviour into class Pass: The following functions are now in class Pass: - IsLoopHeader. - ComputeStructuredOrder - ComputeStructuredSuccessors (annoyingly, I could not re-factor all instances of this function, the copy in common_uniform_elim_pass.cpp is slightly different and fails with the common implementation). - GetPointeeTypeId - TakeNextId - FinalizeNextId - MergeBlockIdIfAny This is a NFC (non-functional change)
2017-10-25 17:26:25 +00:00
bool modified = ProcessEntryPointCallTree(pfn, get_module());
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
return modified ? Status::SuccessWithChange : Status::SuccessWithoutChange;
}
Add MemPass, move all shared functions to it.
2017-07-27 19:30:12 +00:00
AggressiveDCEPass::AggressiveDCEPass() {}
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
Add the IRContext (part 1) This is the first part of adding the IRContext. This class is meant to hold the extra data that is build on top of the module that it owns. The first part will simply create the IRContext class and get it passed to the passes in place of the module. For now it does not have any functionality of its own, but it acts more as a wrapper for the module. The functions that I added to the IRContext are those that either traverse the headers or add to them. I did this because we may decide to have other ways of dealing with these sections (for example adding a type pool, or use the decoration manager). I also added the function that add to the header because the IRContext needs to know when an instruction is added to update other data structures appropriately. Note that there is still lots of work that needs to be done. There are still many places that change the module, and do not inform the context. That will be the next step.
2017-10-30 15:13:24 +00:00
Pass::Status AggressiveDCEPass::Process(ir::IRContext* c) {
Initialize(c);
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
return ProcessImpl();
}
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 AggressiveDCEPass::InitExtensions() {
extensions_whitelist_.clear();
extensions_whitelist_.insert({
Re-format files in source, source/opt, source/util, source/val and tools. NFC. This just makes sure every file is formatted following the formatting definition in .clang-format. Re-formatted with: $ clang-format -i $(find source tools include -name '*.cpp') $ clang-format -i $(find source tools include -name '*.h')
2017-11-08 17:40:02 +00:00
"SPV_AMD_shader_explicit_vertex_parameter",
Create a local value numbering pass Creates a pass that removes redundant instructions within the same basic block. This will be implemented using a hash based value numbering algorithm. Added a number of functions that check for the Vulkan descriptor types. These are used to determine if we are variables are read-only or not. Implemented a function to check if loads and variables are read-only. Implemented kernel specific and shader specific versions. A big change is that the Combinator analysis in ADCE is factored out into the IRContext as an analysis. This was done because it is being reused in the value number table.
2017-11-11 01:26:55 +00:00
"SPV_AMD_shader_trinary_minmax",
"SPV_AMD_gcn_shader",
"SPV_KHR_shader_ballot",
"SPV_AMD_shader_ballot",
"SPV_AMD_gpu_shader_half_float",
"SPV_KHR_shader_draw_parameters",
"SPV_KHR_subgroup_vote",
"SPV_KHR_16bit_storage",
"SPV_KHR_device_group",
"SPV_KHR_multiview",
"SPV_NVX_multiview_per_view_attributes",
"SPV_NV_viewport_array2",
"SPV_NV_stereo_view_rendering",
Re-format files in source, source/opt, source/util, source/val and tools. NFC. This just makes sure every file is formatted following the formatting definition in .clang-format. Re-formatted with: $ clang-format -i $(find source tools include -name '*.cpp') $ clang-format -i $(find source tools include -name '*.h')
2017-11-08 17:40:02 +00:00
"SPV_NV_sample_mask_override_coverage",
Create a local value numbering pass Creates a pass that removes redundant instructions within the same basic block. This will be implemented using a hash based value numbering algorithm. Added a number of functions that check for the Vulkan descriptor types. These are used to determine if we are variables are read-only or not. Implemented a function to check if loads and variables are read-only. Implemented kernel specific and shader specific versions. A big change is that the Combinator analysis in ADCE is factored out into the IRContext as an analysis. This was done because it is being reused in the value number table.
2017-11-11 01:26:55 +00:00
"SPV_NV_geometry_shader_passthrough",
"SPV_AMD_texture_gather_bias_lod",
Re-format files in source, source/opt, source/util, source/val and tools. NFC. This just makes sure every file is formatted following the formatting definition in .clang-format. Re-formatted with: $ clang-format -i $(find source tools include -name '*.cpp') $ clang-format -i $(find source tools include -name '*.h')
2017-11-08 17:40:02 +00:00
"SPV_KHR_storage_buffer_storage_class",
// SPV_KHR_variable_pointers
// Currently do not support extended pointer expressions
Create a local value numbering pass Creates a pass that removes redundant instructions within the same basic block. This will be implemented using a hash based value numbering algorithm. Added a number of functions that check for the Vulkan descriptor types. These are used to determine if we are variables are read-only or not. Implemented a function to check if loads and variables are read-only. Implemented kernel specific and shader specific versions. A big change is that the Combinator analysis in ADCE is factored out into the IRContext as an analysis. This was done because it is being reused in the value number table.
2017-11-11 01:26:55 +00:00
"SPV_AMD_gpu_shader_int16",
"SPV_KHR_post_depth_coverage",
Re-format files in source, source/opt, source/util, source/val and tools. NFC. This just makes sure every file is formatted following the formatting definition in .clang-format. Re-formatted with: $ clang-format -i $(find source tools include -name '*.cpp') $ clang-format -i $(find source tools include -name '*.h')
2017-11-08 17:40:02 +00:00
"SPV_KHR_shader_atomic_counter_ops",
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
});
}
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 opt
} // namespace spvtools
Reference in New Issue Copy Permalink