AuroraMiddleware/SPIRV-Cross
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Merge pull request #624 from KhronosGroup/fix-619

Browse Source 
Support branch/loop hints in HLSL.
This commit is contained in:
Hans-Kristian Arntzen 2018-06-25 10:53:52 +02:00 committed by GitHub
commit 994f789465
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
9 changed files with 279 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

32
reference/opt/shaders-hlsl/asm/comp/control-flow-hints.asm.comp Normal file
View File

@ -0,0 +1,32 @@
RWByteAddressBuffer bar : register(u0);
RWByteAddressBuffer foo : register(u1);
void comp_main()
{
[unroll]
for (int _135 = 0; _135 < 16; )
{
bar.Store4(_135 * 16 + 0, asuint(asfloat(foo.Load4(_135 * 16 + 0))));
_135++;
continue;
}
[loop]
for (int _136 = 0; _136 < 16; )
{
bar.Store4((15 - _136) * 16 + 0, asuint(asfloat(foo.Load4(_136 * 16 + 0))));
_136++;
continue;
}
[branch]
if (asfloat(bar.Load(160)) > 10.0f)
{
foo.Store4(320, asuint(5.0f.xxxx));
}
foo.Store4(320, asuint(20.0f.xxxx));
}
[numthreads(1, 1, 1)]
void main()
{
comp_main();
}

41
reference/shaders-hlsl/asm/comp/control-flow-hints.asm.comp Normal file
View File

@ -0,0 +1,41 @@
RWByteAddressBuffer bar : register(u0);
RWByteAddressBuffer foo : register(u1);
void _main()
{
[unroll]
for (int i = 0; i < 16; i++)
{
bar.Store4(i * 16 + 0, asuint(asfloat(foo.Load4(i * 16 + 0))));
}
[loop]
for (int i_1 = 0; i_1 < 16; i_1++)
{
bar.Store4((15 - i_1) * 16 + 0, asuint(asfloat(foo.Load4(i_1 * 16 + 0))));
}
float v = asfloat(bar.Load(160));
float w = asfloat(foo.Load(160));
[branch]
if (v > 10.0f)
{
foo.Store4(320, asuint(5.0f.xxxx));
}
float value = 20.0f;
[flatten]
if (w > 40.0f)
{
value = 20.0f;
}
foo.Store4(320, asuint(value.xxxx));
}
void comp_main()
{
_main();
}
[numthreads(1, 1, 1)]
void main()
{
comp_main();
}

146
shaders-hlsl/asm/comp/control-flow-hints.asm.comp Normal file
View File

@ -0,0 +1,146 @@
; SPIR-V
; Version: 1.0
; Generator: Khronos Glslang Reference Front End; 6
; Bound: 85
; Schema: 0
OpCapability Shader
%1 = OpExtInstImport "GLSL.std.450"
OpMemoryModel Logical GLSL450
OpEntryPoint GLCompute %main "main"
OpExecutionMode %main LocalSize 1 1 1
OpSource HLSL 500
OpName %main "main"
OpName %_main_ "@main("
OpName %i "i"
OpName %bar "bar"
OpMemberName %bar 0 "@data"
OpName %bar_0 "bar"
OpName %foo "foo"
OpName %i_0 "i"
OpName %v "v"
OpName %w "w"
OpName %value "value"
OpDecorate %_runtimearr_v4float ArrayStride 16
OpMemberDecorate %bar 0 Offset 0
OpDecorate %bar BufferBlock
OpDecorate %bar_0 DescriptorSet 0
OpDecorate %bar_0 Binding 0
OpDecorate %foo DescriptorSet 0
OpDecorate %foo Binding 1
%void = OpTypeVoid
%3 = OpTypeFunction %void
%int = OpTypeInt 32 1
%_ptr_Function_int = OpTypePointer Function %int
%int_0 = OpConstant %int 0
%int_16 = OpConstant %int 16
%bool = OpTypeBool
%float = OpTypeFloat 32
%v4float = OpTypeVector %float 4
%_runtimearr_v4float = OpTypeRuntimeArray %v4float
%bar = OpTypeStruct %_runtimearr_v4float
%_ptr_Uniform_bar = OpTypePointer Uniform %bar
%bar_0 = OpVariable %_ptr_Uniform_bar Uniform
%foo = OpVariable %_ptr_Uniform_bar Uniform
%_ptr_Uniform_v4float = OpTypePointer Uniform %v4float
%int_1 = OpConstant %int 1
%int_15 = OpConstant %int 15
%_ptr_Function_float = OpTypePointer Function %float
%int_10 = OpConstant %int 10
%uint = OpTypeInt 32 0
%uint_0 = OpConstant %uint 0
%_ptr_Uniform_float = OpTypePointer Uniform %float
%float_10 = OpConstant %float 10
%int_20 = OpConstant %int 20
%float_5 = OpConstant %float 5
%72 = OpConstantComposite %v4float %float_5 %float_5 %float_5 %float_5
%float_20 = OpConstant %float 20
%float_40 = OpConstant %float 40
%main = OpFunction %void None %3
%5 = OpLabel
%84 = OpFunctionCall %void %_main_
OpReturn
OpFunctionEnd
%_main_ = OpFunction %void None %3
%7 = OpLabel
%i = OpVariable %_ptr_Function_int Function
%i_0 = OpVariable %_ptr_Function_int Function
%v = OpVariable %_ptr_Function_float Function
%w = OpVariable %_ptr_Function_float Function
%value = OpVariable %_ptr_Function_float Function
OpStore %i %int_0
OpBranch %12
%12 = OpLabel
OpLoopMerge %14 %15 Unroll
OpBranch %16
%16 = OpLabel
%17 = OpLoad %int %i
%20 = OpSLessThan %bool %17 %int_16
OpBranchConditional %20 %13 %14
%13 = OpLabel
%27 = OpLoad %int %i
%29 = OpLoad %int %i
%31 = OpAccessChain %_ptr_Uniform_v4float %foo %int_0 %29
%32 = OpLoad %v4float %31
%33 = OpAccessChain %_ptr_Uniform_v4float %bar_0 %int_0 %27
OpStore %33 %32
OpBranch %15
%15 = OpLabel
%34 = OpLoad %int %i
%36 = OpIAdd %int %34 %int_1
OpStore %i %36
OpBranch %12
%14 = OpLabel
OpStore %i_0 %int_0
OpBranch %38
%38 = OpLabel
OpLoopMerge %40 %41 DontUnroll
OpBranch %42
%42 = OpLabel
%43 = OpLoad %int %i_0
%44 = OpSLessThan %bool %43 %int_16
OpBranchConditional %44 %39 %40
%39 = OpLabel
%46 = OpLoad %int %i_0
%47 = OpISub %int %int_15 %46
%48 = OpLoad %int %i_0
%49 = OpAccessChain %_ptr_Uniform_v4float %foo %int_0 %48
%50 = OpLoad %v4float %49
%51 = OpAccessChain %_ptr_Uniform_v4float %bar_0 %int_0 %47
OpStore %51 %50
OpBranch %41
%41 = OpLabel
%52 = OpLoad %int %i_0
%53 = OpIAdd %int %52 %int_1
OpStore %i_0 %53
OpBranch %38
%40 = OpLabel
%60 = OpAccessChain %_ptr_Uniform_float %bar_0 %int_0 %int_10 %uint_0
%61 = OpLoad %float %60
OpStore %v %61
%63 = OpAccessChain %_ptr_Uniform_float %foo %int_0 %int_10 %uint_0
%64 = OpLoad %float %63
OpStore %w %64
%65 = OpLoad %float %v
%67 = OpFOrdGreaterThan %bool %65 %float_10
OpSelectionMerge %69 DontFlatten
OpBranchConditional %67 %68 %69
%68 = OpLabel
%73 = OpAccessChain %_ptr_Uniform_v4float %foo %int_0 %int_20
OpStore %73 %72
OpBranch %69
%69 = OpLabel
OpStore %value %float_20
%76 = OpLoad %float %w
%78 = OpFOrdGreaterThan %bool %76 %float_40
OpSelectionMerge %80 Flatten
OpBranchConditional %78 %79 %80
%79 = OpLabel
OpStore %value %float_20
OpBranch %80
%80 = OpLabel
%81 = OpLoad %float %value
%82 = OpCompositeConstruct %v4float %81 %81 %81 %81
%83 = OpAccessChain %_ptr_Uniform_v4float %foo %int_0 %int_20
OpStore %83 %82
OpReturn
OpFunctionEnd

10
spirv_common.hpp
View File

@ -578,6 +578,15 @@ struct SPIRBlock : IVariant
MergeSelection
};
enum Hints
{
HintNone,
HintUnroll,
HintDontUnroll,
HintFlatten,
HintDontFlatten
};
enum Method
{
MergeToSelectForLoop,
@ -610,6 +619,7 @@ struct SPIRBlock : IVariant
Terminator terminator = Unknown;
Merge merge = MergeNone;
Hints hint = HintNone;
uint32_t next_block = 0;
uint32_t merge_block = 0;
uint32_t continue_block = 0;

16
spirv_cross.cpp
View File

@ -2221,6 +2221,14 @@ void Compiler::parse(const Instruction &instruction)
current_block->next_block = ops[0];
current_block->merge = SPIRBlock::MergeSelection;
selection_merge_targets.insert(current_block->next_block);
if (length >= 2)
{
if (ops[1] & SelectionControlFlattenMask)
current_block->hint = SPIRBlock::HintFlatten;
else if (ops[1] & SelectionControlDontFlattenMask)
current_block->hint = SPIRBlock::HintDontFlatten;
}
break;
}
@ -2243,6 +2251,14 @@ void Compiler::parse(const Instruction &instruction)
// they are treated as continues.
if (current_block->continue_block != current_block->self)
continue_blocks.insert(current_block->continue_block);
if (length >= 3)
{
if (ops[2] & LoopControlUnrollMask)
current_block->hint = SPIRBlock::HintUnroll;
else if (ops[2] & LoopControlDontUnrollMask)
current_block->hint = SPIRBlock::HintDontUnroll;
}
break;
}

11
spirv_glsl.cpp
View File

@ -9101,6 +9101,7 @@ void CompilerGLSL::branch(uint32_t from, uint32_t cond, uint32_t true_block, uin
if (true_sub)
{
emit_block_hints(get<SPIRBlock>(from));
statement("if (", to_expression(cond), ")");
begin_scope();
branch(from, true_block);
@ -9124,6 +9125,7 @@ void CompilerGLSL::branch(uint32_t from, uint32_t cond, uint32_t true_block, uin
else if (false_sub && !true_sub)
{
// Only need false path, use negative conditional.
emit_block_hints(get<SPIRBlock>(from));
statement("if (!", to_enclosed_expression(cond), ")");
begin_scope();
branch(from, false_block);
@ -9352,6 +9354,7 @@ bool CompilerGLSL::attempt_emit_loop_header(SPIRBlock &block, SPIRBlock::Method
{
// This block may be a dominating block, so make sure we flush undeclared variables before building the for loop header.
flush_undeclared_variables(block);
emit_block_hints(block);
// Important that we do this in this order because
// emitting the continue block can invalidate the condition expression.
@ -9370,6 +9373,7 @@ bool CompilerGLSL::attempt_emit_loop_header(SPIRBlock &block, SPIRBlock::Method
case SPIRBlock::WhileLoop:
// This block may be a dominating block, so make sure we flush undeclared variables before building the while loop header.
flush_undeclared_variables(block);
emit_block_hints(block);
statement("while (", to_expression(block.condition), ")");
break;
@ -9417,11 +9421,13 @@ bool CompilerGLSL::attempt_emit_loop_header(SPIRBlock &block, SPIRBlock::Method
auto initializer = emit_for_loop_initializers(block);
auto condition = to_expression(child.condition);
auto continue_block = emit_continue_block(block.continue_block);
emit_block_hints(block);
statement("for (", initializer, "; ", condition, "; ", continue_block, ")");
break;
}
case SPIRBlock::WhileLoop:
emit_block_hints(block);
statement("while (", to_expression(child.condition), ")");
break;
@ -9632,6 +9638,7 @@ void CompilerGLSL::emit_block_chain(SPIRBlock &block)
auto &type = expression_type(block.condition);
bool uint32_t_case = type.basetype == SPIRType::UInt;
emit_block_hints(block);
statement("switch (", to_expression(block.condition), ")");
begin_scope();
@ -9944,3 +9951,7 @@ void CompilerGLSL::bitcast_to_builtin_store(uint32_t target_id, std::string &exp
expr = bitcast_expression(type, expr_type.basetype, expr);
}
}
void CompilerGLSL::emit_block_hints(const SPIRBlock &)
{
}

1
spirv_glsl.hpp
View File

@ -462,6 +462,7 @@ protected:
std::string to_combined_image_sampler(uint32_t image_id, uint32_t samp_id);
virtual bool skip_argument(uint32_t id) const;
virtual void emit_array_copy(const std::string &lhs, uint32_t rhs_id);
virtual void emit_block_hints(const SPIRBlock &block);
bool buffer_is_packing_standard(const SPIRType &type, BufferPackingStandard packing, uint32_t start_offset = 0,
uint32_t end_offset = std::numeric_limits<uint32_t>::max());

21
spirv_hlsl.cpp
View File

@ -4610,3 +4610,24 @@ string CompilerHLSL::compile()
return buffer->str();
}
void CompilerHLSL::emit_block_hints(const SPIRBlock &block)
{
switch (block.hint)
{
case SPIRBlock::HintFlatten:
statement("[flatten]");
break;
case SPIRBlock::HintDontFlatten:
statement("[branch]");
break;
case SPIRBlock::HintUnroll:
statement("[unroll]");
break;
case SPIRBlock::HintDontUnroll:
statement("[loop]");
break;
default:
break;
}
}

1
spirv_hlsl.hpp
View File

@ -158,6 +158,7 @@ private:
void emit_store(const Instruction &instruction);
void emit_atomic(const uint32_t *ops, uint32_t length, spv::Op op);
void emit_subgroup_op(const Instruction &i) override;
void emit_block_hints(const SPIRBlock &block) override;
void emit_struct_member(const SPIRType &type, uint32_t member_type_id, uint32_t index, const std::string &qualifier,
uint32_t base_offset = 0) override;