Skip to content

Commit

Permalink
Handle missing xfb execution mode in addCallInstForXfbOutput
Browse files Browse the repository at this point in the history 
There is a case that the vertex outputs has xfb decoration but without
declared with xfb execution mode. As vulkan spec, these outputs will not
be captured by xfb buffer. This change will add the missing handling for
this case in `addCallInstForXfbOutput`. The check for existing xfb
outputis based on lgc.xfb.state metadata.
  • Loading branch information
@xuechen417
xuechen417 committed Oct 27, 2023
1 parent a2ed744 commit b26fa0f
Show file tree
Hide file tree
Showing 2 changed files with 191 additions and 2 deletions.
5 changes: 3 additions & 2 deletions llpc/lower/llpcSpirvLowerGlobal.cpp
View file
Original file line number Diff line number Diff line change
Expand Up @@ -2421,9 +2421,10 @@ void SpirvLowerGlobal::addCallInstForXfbOutput(const ShaderInOutMetadata &output
unsigned xfbBufferAdjust, unsigned xfbOffsetAdjust, unsigned locOffset,
lgc::InOutInfo outputInfo) {
assert(m_shaderStage == m_lastVertexProcessingStage);
auto pipelineBuildInfo = static_cast<const Vkgc::GraphicsPipelineBuildInfo *>(m_context->getPipelineBuildInfo());
DenseMap<unsigned, Vkgc::XfbOutInfo> *locXfbMapPtr = outputMeta.IsBuiltIn ? &m_builtInXfbMap : &m_genericXfbMap;
if (pipelineBuildInfo->apiXfbOutData.forceDisableStreamOut || (locXfbMapPtr->empty() && !outputMeta.IsXfb))
bool hasXfbMetadata = m_entryPoint->getMetadata(lgc::XfbStateMetadataName);
bool hasXfbOut = hasXfbMetadata && (!locXfbMapPtr->empty() || outputMeta.IsXfb);
if (!hasXfbOut)
return;

// If the XFB info is specified from API interface so we try to retrieve the info from m_locXfbMap. Otherwise, the XFB
Expand Down
188 changes: 188 additions & 0 deletions llpc/test/shaderdb/extensions/ExtXfb_TestNoXfbExecutionMode.spvasm
View file
Original file line number Diff line number Diff line change
@@ -0,0 +1,188 @@
; NOTE: Assertions have been autogenerated by tool/update_llpc_test_checks.py
; RUN: amdllpc -o - -gfxip 10.3 -emit-lgc %s | FileCheck -check-prefixes=CHECK %s
; SPIR-V
; Version: 1.3
; Generator: Khronos; 4660
; Bound: 122
; Schema: 0
OpCapability Shader
OpCapability SampledCubeArray
OpCapability ImageBuffer
OpCapability ImageGatherExtended
OpCapability ImageQuery
OpCapability TransformFeedback
OpCapability Tessellation
OpCapability SignedZeroInfNanPreserve
OpExtension "SPV_KHR_float_controls"
%87 = OpExtInstImport "GLSL.std.450"
OpMemoryModel Logical GLSL450
OpEntryPoint TessellationEvaluation %main "main" %DomainPoint %gl_PrimitiveID %o0xyzw %ipatch0y %gl_PointSize
OpExecutionMode %main Quads
OpExecutionMode %main SignedZeroInfNanPreserve 32
%2 = OpString "VMware VMGI Translator (shader 21)"
%3 = OpString "DOMA"
%71 = OpString "PROPERTY REFACTORING_ALLOWED"
%72 = OpString "DCL INPUT_PATCH_CONSTANT[0].y"
%73 = OpString "DCL INPUT_DOMAIN_POINT[0].xy"
%75 = OpString "PROPERTY INPUT_CONTROL_POINT_COUNT 4"
%76 = OpString "PROPERTY TESS_DOMAIN QUAD"
%77 = OpString "DCL OUT[0], POSITION"
%78 = OpString "DCL OUT[1].xy, GENERIC[1]"
%79 = OpString "IMM INT32 {0, 1065353216, 0, 0}"
%83 = OpString "DCL PRIMID[0]"
%85 = OpString "soEntry[0] = OUT[0].xyzw, buffer 0, stream 0"
%86 = OpString "RasterizedStream=255, DefaultStream=0"
%91 = OpString " 0: MOV OUT[0].xy, INPUT_DOMAIN_POINT[0].xyxx"
%102 = OpString " 1: MOV OUT[0].zw, IMM[0].xxxy"
%111 = OpString " 2: MOV OUT[1].x, INPUT_PATCH_CONSTANT[0].y"
%115 = OpString " 3: MOV OUT[1].y, PRIMID[0]"
%120 = OpString " 4: RET"
%121 = OpString " 5: END"
OpName %main "main"
OpName %void_Func "void_Func"
OpName %struct_2_ivec4 "struct_2_ivec4"
OpName %struct_2_uvec4 "struct_2_uvec4"
OpName %main_begin "main_begin"
OpName %DomainPoint "DomainPoint"
OpName %o0xyzw "o0xyzw"
OpName %ipatch0y "ipatch0y"
OpDecorate %DomainPoint BuiltIn TessCoord
OpDecorate %gl_PrimitiveID BuiltIn PrimitiveId
OpDecorate %o0xyzw Location 0
OpDecorate %o0xyzw XfbBuffer 0
OpDecorate %o0xyzw XfbStride 16
OpDecorate %o0xyzw Offset 0
OpDecorate %ipatch0y Patch
OpDecorate %ipatch0y Location 0
OpDecorate %ipatch0y Component 1
OpDecorate %gl_PointSize BuiltIn PointSize
%void = OpTypeVoid
%bool = OpTypeBool
%float = OpTypeFloat 32
%int = OpTypeInt 32 1
%uint = OpTypeInt 32 0
%v2float = OpTypeVector %float 2
%v3float = OpTypeVector %float 3
%v4float = OpTypeVector %float 4
%v2int = OpTypeVector %int 2
%v3int = OpTypeVector %int 3
%v4int = OpTypeVector %int 4
%v2uint = OpTypeVector %uint 2
%v3uint = OpTypeVector %uint 3
%v4uint = OpTypeVector %uint 4
%v2bool = OpTypeVector %bool 2
%v3bool = OpTypeVector %bool 3
%v4bool = OpTypeVector %bool 4
%void_Func = OpTypeFunction %void
%42 = OpTypeFunction %v4float %v4float
%struct_2_ivec4 = OpTypeStruct %v4int %v4int
%struct_2_uvec4 = OpTypeStruct %v4uint %v4uint
%int_0 = OpConstant %int 0
%int_1065353216 = OpConstant %int 1065353216
%82 = OpConstantComposite %v4int %int_0 %int_1065353216 %int_0 %int_0
%float_0 = OpConstant %float 0
%float_1 = OpConstant %float 1
%int_1 = OpConstant %int 1
%int_2 = OpConstant %int 2
%int_3 = OpConstant %int 3
%_ptr_Private_float = OpTypePointer Private %float
%_ptr_Private_v2float = OpTypePointer Private %v2float
%_ptr_Private_v3float = OpTypePointer Private %v3float
%_ptr_Private_v4float = OpTypePointer Private %v4float
%_ptr_Private_int = OpTypePointer Private %int
%_ptr_Private_v2int = OpTypePointer Private %v2int
%_ptr_Private_v3int = OpTypePointer Private %v3int
%_ptr_Private_v4int = OpTypePointer Private %v4int
%_ptr_Private_uint = OpTypePointer Private %uint
%_ptr_Private_v2uint = OpTypePointer Private %v2uint
%_ptr_Private_v3uint = OpTypePointer Private %v3uint
%_ptr_Private_v4uint = OpTypePointer Private %v4uint
%_ptr_Private_bool = OpTypePointer Private %bool
%_ptr_Private_v2bool = OpTypePointer Private %v2bool
%_ptr_Private_v3bool = OpTypePointer Private %v3bool
%_ptr_Private_v4bool = OpTypePointer Private %v4bool
%_ptr_Uniform_v4float = OpTypePointer Uniform %v4float
%_ptr_Uniform_v4int = OpTypePointer Uniform %v4int
%_ptr_PushConstant_v4float = OpTypePointer PushConstant %v4float
%_ptr_PushConstant_v4int = OpTypePointer PushConstant %v4int
%_ptr_Input_float = OpTypePointer Input %float
%_ptr_Input_v2float = OpTypePointer Input %v2float
%_ptr_Input_v3float = OpTypePointer Input %v3float
%_ptr_Input_v4float = OpTypePointer Input %v4float
%_ptr_Input_int = OpTypePointer Input %int
%_ptr_Input_v2int = OpTypePointer Input %v2int
%_ptr_Input_v3int = OpTypePointer Input %v3int
%_ptr_Input_v4int = OpTypePointer Input %v4int
%_ptr_Input_uint = OpTypePointer Input %uint
%_ptr_Input_v2uint = OpTypePointer Input %v2uint
%_ptr_Input_v3uint = OpTypePointer Input %v3uint
%_ptr_Input_v4uint = OpTypePointer Input %v4uint
%_ptr_Input_bool = OpTypePointer Input %bool
%_ptr_Output_float = OpTypePointer Output %float
%_ptr_Output_v2float = OpTypePointer Output %v2float
%_ptr_Output_v3float = OpTypePointer Output %v3float
%_ptr_Output_v4float = OpTypePointer Output %v4float
%_ptr_Output_int = OpTypePointer Output %int
%_ptr_Output_v2int = OpTypePointer Output %v2int
%_ptr_Output_v3int = OpTypePointer Output %v3int
%_ptr_Output_v4int = OpTypePointer Output %v4int
%_ptr_Output_uint = OpTypePointer Output %uint
%_ptr_Output_v2uint = OpTypePointer Output %v2uint
%_ptr_Output_v3uint = OpTypePointer Output %v3uint
%_ptr_Output_v4uint = OpTypePointer Output %v4uint
%DomainPoint = OpVariable %_ptr_Input_v3float Input
%gl_PrimitiveID = OpVariable %_ptr_Input_int Input
%o0xyzw = OpVariable %_ptr_Output_v4float Output
%ipatch0y = OpVariable %_ptr_Input_float Input
%gl_PointSize = OpVariable %_ptr_Output_float Output
%main = OpFunction %void None %void_Func
%main_begin = OpLabel
OpLine %71 0 0
OpLine %72 0 0
OpLine %73 0 0
OpLine %75 0 0
OpLine %76 0 0
OpLine %77 0 0
OpLine %78 0 0
OpLine %79 0 0
OpLine %83 0 0
OpLine %91 0 0
%92 = OpLoad %v3float %DomainPoint
%95 = OpCompositeConstruct %v4float %92 %float_1
%96 = OpVectorShuffle %v4float %95 %95 0 1 0 0
%97 = OpCompositeExtract %float %96 0
%98 = OpAccessChain %_ptr_Output_float %o0xyzw %int_0
OpStore %98 %97
%99 = OpCompositeExtract %float %96 1
%101 = OpAccessChain %_ptr_Output_float %o0xyzw %int_1
OpStore %101 %99
OpLine %102 1 0
%103 = OpBitcast %v4float %82
%104 = OpVectorShuffle %v4float %103 %103 0 0 0 1
%105 = OpCompositeExtract %float %104 2
%107 = OpAccessChain %_ptr_Output_float %o0xyzw %int_2
OpStore %107 %105
%108 = OpCompositeExtract %float %104 3
%110 = OpAccessChain %_ptr_Output_float %o0xyzw %int_3
OpStore %110 %108
OpLine %111 2 0
%112 = OpLoad %float %ipatch0y
%113 = OpCompositeConstruct %v4float %112 %112 %112 %112
%114 = OpCompositeExtract %float %113 0
OpLine %115 3 0
%116 = OpLoad %int %gl_PrimitiveID
%117 = OpBitcast %float %116
%118 = OpCompositeConstruct %v4float %117 %117 %117 %117
%119 = OpCompositeExtract %float %118 1
OpLine %120 4 0
OpReturn
OpFunctionEnd
OpLine %121 5 0
; CHECK-LABEL: @lgc.shader.TES.main(
; CHECK-NEXT: main_begin:
; CHECK-NEXT: [[TMP0:%.*]] = call <3 x float> (...) @lgc.create.read.builtin.input.v3f32(i32 13, i32 0, i32 poison, i32 poison)
; CHECK-NEXT: [[TMP1:%.*]] = shufflevector <3 x float> [[TMP0]], <3 x float> poison, <4 x i32> <i32 0, i32 1, i32 poison, i32 poison>
; CHECK-NEXT: [[__LLPC_OUTPUT_PROXY_O0XYZW_12_VEC_INSERT:%.*]] = shufflevector <4 x float> [[TMP1]], <4 x float> <float poison, float poison, float 0.000000e+00, float 1.000000e+00>, <4 x i32> <i32 0, i32 1, i32 6, i32 7>
; CHECK-NEXT: call void (...) @lgc.create.write.generic.output(<4 x float> [[__LLPC_OUTPUT_PROXY_O0XYZW_12_VEC_INSERT]], i32 0, i32 0, i32 0, i32 0, i32 0, i32 poison)
; CHECK-NEXT: ret void
;

0 comments on commit b26fa0f

Please sign in to comment.