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

Merge pull request #676 from cdavis5e/image-write-operand-words

Browse Source 
MSL: Correct number of words to skip in OpImageWrite.
This commit is contained in:
Hans-Kristian Arntzen 2018-09-05 17:44:21 +02:00 committed by GitHub
commit 301e5c8076
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
4 changed files with 153 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

28
reference/opt/shaders-msl/asm/comp/buffer-write-relative-addr.asm.comp Normal file
View File

@ -0,0 +1,28 @@
#pragma clang diagnostic ignored "-Wmissing-prototypes"
#include <metal_stdlib>
#include <simd/simd.h>
using namespace metal;
struct cb5_struct
{
float4 _m0[5];
};
// Returns 2D texture coords corresponding to 1D texel buffer coords
uint2 spvTexelBufferCoord(uint tc)
{
return uint2(tc % 4096, tc / 4096);
}
kernel void main0(constant cb5_struct& cb0_5 [[buffer(1)]], texture2d<uint, access::write> u0 [[texture(0)]], uint3 gl_LocalInvocationID [[thread_position_in_threadgroup]])
{
uint _44 = as_type<uint>(as_type<float>(gl_LocalInvocationID.x << 4)) >> 2u;
uint4 _51 = as_type<uint4>(cb0_5._m0[uint(as_type<int>(as_type<float>(gl_LocalInvocationID.x))) + 1u]);
u0.write(_51.xxxx, spvTexelBufferCoord(_44));
u0.write(_51.yyyy, spvTexelBufferCoord((_44 + 1u)));
u0.write(_51.zzzz, spvTexelBufferCoord((_44 + 2u)));
u0.write(_51.wwww, spvTexelBufferCoord((_44 + 3u)));
}

31
reference/shaders-msl/asm/comp/buffer-write-relative-addr.asm.comp Normal file
View File

@ -0,0 +1,31 @@
#pragma clang diagnostic ignored "-Wmissing-prototypes"
#include <metal_stdlib>
#include <simd/simd.h>
using namespace metal;
struct cb5_struct
{
float4 _m0[5];
};
// Returns 2D texture coords corresponding to 1D texel buffer coords
uint2 spvTexelBufferCoord(uint tc)
{
return uint2(tc % 4096, tc / 4096);
}
kernel void main0(constant cb5_struct& cb0_5 [[buffer(1)]], texture2d<uint, access::write> u0 [[texture(0)]], uint3 gl_LocalInvocationID [[thread_position_in_threadgroup]])
{
float4 r0;
r0.x = as_type<float>(gl_LocalInvocationID.x << 4);
r0.y = as_type<float>(gl_LocalInvocationID.x);
uint _44 = as_type<uint>(r0.x) >> 2u;
uint4 _51 = as_type<uint4>(cb0_5._m0[uint(as_type<int>(r0.y)) + 1u]);
u0.write(_51.xxxx, spvTexelBufferCoord(_44));
u0.write(_51.yyyy, spvTexelBufferCoord((_44 + 1u)));
u0.write(_51.zzzz, spvTexelBufferCoord((_44 + 2u)));
u0.write(_51.wwww, spvTexelBufferCoord((_44 + 3u)));
}

93
shaders-msl/asm/comp/buffer-write-relative-addr.asm.comp Normal file
View File

@ -0,0 +1,93 @@
; SPIR-V
; Version: 1.0
; Generator: Wine VKD3D Shader Compiler; 0
; Bound: 59
; Schema: 0
OpCapability Shader
OpCapability UniformBufferArrayDynamicIndexing
OpCapability SampledBuffer
OpCapability ImageBuffer
OpMemoryModel Logical GLSL450
OpEntryPoint GLCompute %main "main" %vThreadIDInGroup
OpExecutionMode %main LocalSize 4 1 1
OpName %main "main"
OpName %cb5_struct "cb5_struct"
OpName %cb0_5 "cb0_5"
OpName %u0 "u0"
OpName %vThreadIDInGroup "vThreadIDInGroup"
OpName %r0 "r0"
OpDecorate %_arr_v4float_uint_5 ArrayStride 16
OpDecorate %cb5_struct Block
OpMemberDecorate %cb5_struct 0 Offset 0
OpDecorate %cb0_5 DescriptorSet 0
OpDecorate %cb0_5 Binding 1
OpDecorate %u0 DescriptorSet 0
OpDecorate %u0 Binding 0
OpDecorate %u0 NonReadable
OpDecorate %vThreadIDInGroup BuiltIn LocalInvocationId
%void = OpTypeVoid
%2 = OpTypeFunction %void
%float = OpTypeFloat 32
%v4float = OpTypeVector %float 4
%uint = OpTypeInt 32 0
%uint_5 = OpConstant %uint 5
%_arr_v4float_uint_5 = OpTypeArray %v4float %uint_5
%cb5_struct = OpTypeStruct %_arr_v4float_uint_5
%_ptr_Uniform_cb5_struct = OpTypePointer Uniform %cb5_struct
%cb0_5 = OpVariable %_ptr_Uniform_cb5_struct Uniform
%13 = OpTypeImage %uint Buffer 0 0 0 2 R32ui
%_ptr_UniformConstant_13 = OpTypePointer UniformConstant %13
%u0 = OpVariable %_ptr_UniformConstant_13 UniformConstant
%int = OpTypeInt 32 1
%v3int = OpTypeVector %int 3
%_ptr_Input_v3int = OpTypePointer Input %v3int
%vThreadIDInGroup = OpVariable %_ptr_Input_v3int Input
%_ptr_Function_v4float = OpTypePointer Function %v4float
%_ptr_Input_int = OpTypePointer Input %int
%uint_0 = OpConstant %uint 0
%int_4 = OpConstant %int 4
%_ptr_Function_float = OpTypePointer Function %float
%uint_1 = OpConstant %uint 1
%uint_2 = OpConstant %uint 2
%_ptr_Uniform_v4float = OpTypePointer Uniform %v4float
%v4uint = OpTypeVector %uint 4
%uint_3 = OpConstant %uint 3
%main = OpFunction %void None %2
%4 = OpLabel
%r0 = OpVariable %_ptr_Function_v4float Function
%24 = OpInBoundsAccessChain %_ptr_Input_int %vThreadIDInGroup %uint_0
%25 = OpLoad %int %24
%27 = OpShiftLeftLogical %int %25 %int_4
%28 = OpBitcast %float %27
%30 = OpInBoundsAccessChain %_ptr_Function_float %r0 %uint_0
OpStore %30 %28
%31 = OpInBoundsAccessChain %_ptr_Input_int %vThreadIDInGroup %uint_0
%32 = OpLoad %int %31
%33 = OpBitcast %float %32
%35 = OpInBoundsAccessChain %_ptr_Function_float %r0 %uint_1
OpStore %35 %33
%36 = OpLoad %13 %u0
%37 = OpInBoundsAccessChain %_ptr_Function_float %r0 %uint_0
%38 = OpLoad %float %37
%39 = OpBitcast %uint %38
%41 = OpShiftRightLogical %uint %39 %uint_2
%42 = OpInBoundsAccessChain %_ptr_Function_float %r0 %uint_1
%43 = OpLoad %float %42
%44 = OpBitcast %int %43
%45 = OpIAdd %uint %44 %uint_1
%47 = OpAccessChain %_ptr_Uniform_v4float %cb0_5 %uint_0 %45
%48 = OpLoad %v4float %47
%50 = OpBitcast %v4uint %48
%51 = OpVectorShuffle %v4uint %50 %50 0 0 0 0
OpImageWrite %36 %41 %51
%52 = OpVectorShuffle %v4uint %50 %50 1 1 1 1
%53 = OpIAdd %uint %41 %uint_1
OpImageWrite %36 %53 %52
%54 = OpVectorShuffle %v4uint %50 %50 2 2 2 2
%55 = OpIAdd %uint %41 %uint_2
OpImageWrite %36 %55 %54
%56 = OpVectorShuffle %v4uint %50 %50 3 3 3 3
%58 = OpIAdd %uint %41 %uint_3
OpImageWrite %36 %58 %56
OpReturn
OpFunctionEnd

2
spirv_msl.cpp
View File

@ -1833,7 +1833,7 @@ void CompilerMSL::emit_instruction(const Instruction &instruction)
uint32_t coord_id = ops[1];
uint32_t texel_id = ops[2];
const uint32_t *opt = &ops[3];
uint32_t length = instruction.length - 4;
uint32_t length = instruction.length - 3;
// Bypass pointers because we need the real image struct
auto &type = expression_type(img_id);