Merge pull request #639 from KhronosGroup/fix-638

Properly track read dependencies for UAV access chain.
This commit is contained in:
Hans-Kristian Arntzen 2018-07-09 14:30:16 +02:00 committed by GitHub
commit be7425ef70
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
5 changed files with 102 additions and 1 deletions

View File

@ -0,0 +1,19 @@
RWByteAddressBuffer _4 : register(u0);
void comp_main()
{
uint _21 = _4.Load(_4.Load(0) * 4 + 4);
for (uint _23 = 0u; _23 < 64u; )
{
_4.Store(_23 * 4 + 4, 0u);
_23++;
continue;
}
_4.Store(_4.Load(0) * 4 + 4, _21);
}
[numthreads(1, 1, 1)]
void main()
{
comp_main();
}

View File

@ -0,0 +1,19 @@
RWByteAddressBuffer _4 : register(u0);
void comp_main()
{
uint _21 = _4.Load(_4.Load(0) * 4 + 4);
for (uint _23 = 0u; _23 < 64u; )
{
_4.Store(_23 * 4 + 4, 0u);
_23++;
continue;
}
_4.Store(_4.Load(0) * 4 + 4, _21);
}
[numthreads(1, 1, 1)]
void main()
{
comp_main();
}

View File

@ -0,0 +1,61 @@
; SPIR-V
; Version: 1.0
; Generator: Google Shaderc over Glslang; 7
; Bound: 41
; Schema: 0
OpCapability Shader
%1 = OpExtInstImport "GLSL.std.450"
OpMemoryModel Logical GLSL450
OpEntryPoint GLCompute %main "main"
OpExecutionMode %main LocalSize 1 1 1
OpSource GLSL 450
OpSourceExtension "GL_GOOGLE_cpp_style_line_directive"
OpSourceExtension "GL_GOOGLE_include_directive"
OpName %main "main"
OpName %SSBO "SSBO"
OpMemberName %SSBO 0 "index"
OpMemberName %SSBO 1 "array"
OpName %_ ""
OpDecorate %_arr_uint_uint_64 ArrayStride 4
OpMemberDecorate %SSBO 0 Offset 0
OpMemberDecorate %SSBO 1 Offset 4
OpDecorate %SSBO BufferBlock
OpDecorate %_ DescriptorSet 0
OpDecorate %_ Binding 0
%void = OpTypeVoid
%3 = OpTypeFunction %void
%uint = OpTypeInt 32 0
%uint_64 = OpConstant %uint 64
%_arr_uint_uint_64 = OpTypeArray %uint %uint_64
%SSBO = OpTypeStruct %uint %_arr_uint_uint_64
%_ptr_Uniform_SSBO = OpTypePointer Uniform %SSBO
%_ = OpVariable %_ptr_Uniform_SSBO Uniform
%int = OpTypeInt 32 1
%int_1 = OpConstant %int 1
%int_0 = OpConstant %int 0
%_ptr_Uniform_uint = OpTypePointer Uniform %uint
%uint_0 = OpConstant %uint 0
%bool = OpTypeBool
%main = OpFunction %void None %3
%5 = OpLabel
%18 = OpAccessChain %_ptr_Uniform_uint %_ %int_0
%19 = OpLoad %uint %18
%20 = OpAccessChain %_ptr_Uniform_uint %_ %int_1 %19
%21 = OpLoad %uint %20
OpBranch %24
%24 = OpLabel
%40 = OpPhi %uint %uint_0 %5 %35 %25
%31 = OpULessThan %bool %40 %uint_64
OpLoopMerge %26 %25 None
OpBranchConditional %31 %25 %26
%25 = OpLabel
%33 = OpAccessChain %_ptr_Uniform_uint %_ %int_1 %40
OpStore %33 %uint_0
%35 = OpIAdd %uint %40 %int_1
OpBranch %24
%26 = OpLabel
%37 = OpLoad %uint %18
%39 = OpAccessChain %_ptr_Uniform_uint %_ %int_1 %37
OpStore %39 %21
OpReturn
OpFunctionEnd

View File

@ -6147,7 +6147,7 @@ void CompilerGLSL::emit_instruction(const Instruction &instruction)
if (ptr_expression) if (ptr_expression)
ptr_expression->need_transpose = old_need_transpose; ptr_expression->need_transpose = old_need_transpose;
// By default, suppress usage tracking since using same expressio multiple times does not imply any extra work. // By default, suppress usage tracking since using same expression multiple times does not imply any extra work.
// However, if we try to load a complex, composite object from a flattened buffer, // However, if we try to load a complex, composite object from a flattened buffer,
// we should avoid emitting the same code over and over and lower the result to a temporary. // we should avoid emitting the same code over and over and lower the result to a temporary.
auto &type = get<SPIRType>(result_type); auto &type = get<SPIRType>(result_type);

View File

@ -3519,6 +3519,7 @@ void CompilerHLSL::emit_access_chain(const Instruction &instruction)
if (need_byte_access_chain) if (need_byte_access_chain)
{ {
uint32_t to_plain_buffer_length = static_cast<uint32_t>(type.array.size()); uint32_t to_plain_buffer_length = static_cast<uint32_t>(type.array.size());
auto *backing_variable = maybe_get_backing_variable(ops[2]);
string base; string base;
if (to_plain_buffer_length != 0) if (to_plain_buffer_length != 0)
@ -3557,6 +3558,7 @@ void CompilerHLSL::emit_access_chain(const Instruction &instruction)
e.row_major_matrix = row_major_matrix; e.row_major_matrix = row_major_matrix;
e.matrix_stride = matrix_stride; e.matrix_stride = matrix_stride;
e.immutable = should_forward(ops[2]); e.immutable = should_forward(ops[2]);
e.loaded_from = backing_variable ? backing_variable->self : 0;
if (chain) if (chain)
{ {