Commit 3dd1ce5b by John Kessenich

Web: Add SSBOs and a few other missing compute features.

parent 51ed01c8
388096 ../build/install/bin/glslangValidator.exe
399360 ../build/install/bin/glslangValidator.exe
; SPIR-V
; Version: 1.0
; Generator: Khronos Glslang Reference Front End; 7
; Bound: 55
; Bound: 91
; Schema: 0
OpCapability Shader
%1 = OpExtInstImport "GLSL.std.450"
OpMemoryModel Logical GLSL450
OpEntryPoint GLCompute %main "main"
OpEntryPoint GLCompute %main "main" %gl_NumWorkGroups %gl_WorkGroupID %gl_LocalInvocationID %gl_GlobalInvocationID %gl_LocalInvocationIndex
OpExecutionMode %main LocalSize 2 5 7
OpSource ESSL 310
OpName %main "main"
OpName %bName "bName"
OpMemberName %bName 0 "size"
OpMemberName %bName 1 "count"
OpMemberName %bName 2 "data"
OpName %bInst "bInst"
OpName %s "s"
OpName %arrX "arrX"
OpName %arrY "arrY"
OpName %arrZ "arrZ"
OpDecorate %19 SpecId 18
OpDecorate %21 SpecId 19
OpName %gl_NumWorkGroups "gl_NumWorkGroups"
OpName %gl_WorkGroupID "gl_WorkGroupID"
OpName %gl_LocalInvocationID "gl_LocalInvocationID"
OpName %gl_GlobalInvocationID "gl_GlobalInvocationID"
OpName %gl_LocalInvocationIndex "gl_LocalInvocationIndex"
OpDecorate %_runtimearr_v4float ArrayStride 16
OpMemberDecorate %bName 0 Offset 0
OpMemberDecorate %bName 1 Offset 16
OpMemberDecorate %bName 2 Offset 32
OpDecorate %bName BufferBlock
OpDecorate %bInst DescriptorSet 0
OpDecorate %bInst Binding 0
OpDecorate %39 SpecId 18
OpDecorate %41 SpecId 19
OpDecorate %gl_WorkGroupSize BuiltIn WorkgroupSize
OpDecorate %gl_NumWorkGroups BuiltIn NumWorkgroups
OpDecorate %gl_WorkGroupID BuiltIn WorkgroupId
OpDecorate %gl_LocalInvocationID BuiltIn LocalInvocationId
OpDecorate %gl_GlobalInvocationID BuiltIn GlobalInvocationId
OpDecorate %gl_LocalInvocationIndex BuiltIn LocalInvocationIndex
%void = OpTypeVoid
%3 = OpTypeFunction %void
%uint = OpTypeInt 32 0
%uint_2 = OpConstant %uint 2
%uint_264 = OpConstant %uint 264
%int = OpTypeInt 32 1
%v3uint = OpTypeVector %uint 3
%float = OpTypeFloat 32
%v4float = OpTypeVector %float 4
%uint = OpTypeInt 32 0
%_runtimearr_v4float = OpTypeRuntimeArray %v4float
%bName = OpTypeStruct %int %v3uint %_runtimearr_v4float
%_ptr_Uniform_bName = OpTypePointer Uniform %bName
%bInst = OpVariable %_ptr_Uniform_bName Uniform
%int_2 = OpConstant %int 2
%int_0 = OpConstant %int 0
%_ptr_Uniform_int = OpTypePointer Uniform %int
%float_7 = OpConstant %float 7
%24 = OpConstantComposite %v4float %float_7 %float_7 %float_7 %float_7
%_ptr_Uniform_v4float = OpTypePointer Uniform %v4float
%uint_1 = OpConstant %uint 1
%uint_3400 = OpConstant %uint 3400
%uint_72 = OpConstant %uint 72
%uint_197645 = OpConstant %uint 197645
%_arr_v4float_uint_197645 = OpTypeArray %v4float %uint_197645
%_ptr_Function__arr_v4float_uint_197645 = OpTypePointer Function %_arr_v4float_uint_197645
%int = OpTypeInt 32 1
%_ptr_Workgroup__arr_v4float_uint_197645 = OpTypePointer Workgroup %_arr_v4float_uint_197645
%s = OpVariable %_ptr_Workgroup__arr_v4float_uint_197645 Workgroup
%int_3 = OpConstant %int 3
%float_1 = OpConstant %float 1
%16 = OpConstantComposite %v4float %float_1 %float_1 %float_1 %float_1
%_ptr_Function_v4float = OpTypePointer Function %v4float
%19 = OpSpecConstant %uint 2
%float_0 = OpConstant %float 0
%39 = OpSpecConstant %uint 2
%uint_5 = OpConstant %uint 5
%21 = OpSpecConstant %uint 7
%v3uint = OpTypeVector %uint 3
%23 = OpSpecConstantComposite %v3uint %19 %uint_5 %21
%41 = OpSpecConstant %uint 7
%gl_WorkGroupSize = OpSpecConstantComposite %v3uint %39 %uint_5 %41
%uint_0 = OpConstant %uint 0
%_arr_int_25 = OpTypeArray %int %25
%_ptr_Private__arr_int_25 = OpTypePointer Private %_arr_int_25
%arrX = OpVariable %_ptr_Private__arr_int_25 Private
%int_0 = OpConstant %int 0
%_arr_int_44 = OpTypeArray %int %44
%_ptr_Private__arr_int_44 = OpTypePointer Private %_arr_int_44
%arrX = OpVariable %_ptr_Private__arr_int_44 Private
%_ptr_Private_int = OpTypePointer Private %int
%uint_1 = OpConstant %uint 1
%_arr_int_34 = OpTypeArray %int %34
%_ptr_Private__arr_int_34 = OpTypePointer Private %_arr_int_34
%arrY = OpVariable %_ptr_Private__arr_int_34 Private
%uint_2 = OpConstant %uint 2
%_arr_int_42 = OpTypeArray %int %42
%_ptr_Private__arr_int_42 = OpTypePointer Private %_arr_int_42
%arrZ = OpVariable %_ptr_Private__arr_int_42 Private
%_arr_int_52 = OpTypeArray %int %52
%_ptr_Private__arr_int_52 = OpTypePointer Private %_arr_int_52
%arrY = OpVariable %_ptr_Private__arr_int_52 Private
%_arr_int_59 = OpTypeArray %int %59
%_ptr_Private__arr_int_59 = OpTypePointer Private %_arr_int_59
%arrZ = OpVariable %_ptr_Private__arr_int_59 Private
%_ptr_Workgroup_v4float = OpTypePointer Workgroup %v4float
%int_1 = OpConstant %int 1
%_ptr_Input_v3uint = OpTypePointer Input %v3uint
%gl_NumWorkGroups = OpVariable %_ptr_Input_v3uint Input
%gl_WorkGroupID = OpVariable %_ptr_Input_v3uint Input
%gl_LocalInvocationID = OpVariable %_ptr_Input_v3uint Input
%gl_GlobalInvocationID = OpVariable %_ptr_Input_v3uint Input
%_ptr_Input_uint = OpTypePointer Input %uint
%gl_LocalInvocationIndex = OpVariable %_ptr_Input_uint Input
%_ptr_Uniform_v3uint = OpTypePointer Uniform %v3uint
%int_197645 = OpConstant %int 197645
%main = OpFunction %void None %3
%5 = OpLabel
%s = OpVariable %_ptr_Function__arr_v4float_uint_197645 Function
%18 = OpAccessChain %_ptr_Function_v4float %s %int_3
OpStore %18 %16
%25 = OpCompositeExtract %uint %23 0
%31 = OpAccessChain %_ptr_Private_int %arrX %int_0
%32 = OpLoad %int %31
%34 = OpCompositeExtract %uint %23 1
%38 = OpAccessChain %_ptr_Private_int %arrY %int_0
%39 = OpLoad %int %38
%40 = OpIMul %int %32 %39
%42 = OpCompositeExtract %uint %23 2
%46 = OpAccessChain %_ptr_Private_int %arrZ %int_0
%47 = OpLoad %int %46
%48 = OpIMul %int %40 %47
%49 = OpConvertSToF %float %48
%50 = OpAccessChain %_ptr_Function_v4float %s %int_3
%51 = OpLoad %v4float %50
%52 = OpVectorTimesScalar %v4float %51 %49
%53 = OpAccessChain %_ptr_Function_v4float %s %int_3
OpStore %53 %52
OpControlBarrier %uint_2 %uint_2 %uint_264
%20 = OpAccessChain %_ptr_Uniform_int %bInst %int_0
%21 = OpLoad %int %20
%22 = OpSDiv %int %21 %int_2
%26 = OpAccessChain %_ptr_Uniform_v4float %bInst %int_2 %22
%27 = OpLoad %v4float %26
%28 = OpFMul %v4float %27 %24
%29 = OpAccessChain %_ptr_Uniform_v4float %bInst %int_2 %22
OpStore %29 %28
OpMemoryBarrier %uint_1 %uint_3400
OpMemoryBarrier %uint_2 %uint_3400
OpMemoryBarrier %uint_1 %uint_264
OpMemoryBarrier %uint_1 %uint_72
%44 = OpCompositeExtract %uint %gl_WorkGroupSize 0
%49 = OpAccessChain %_ptr_Private_int %arrX %int_0
%50 = OpLoad %int %49
%51 = OpConvertSToF %float %50
%52 = OpCompositeExtract %uint %gl_WorkGroupSize 1
%56 = OpAccessChain %_ptr_Private_int %arrY %int_0
%57 = OpLoad %int %56
%58 = OpConvertSToF %float %57
%59 = OpCompositeExtract %uint %gl_WorkGroupSize 2
%63 = OpAccessChain %_ptr_Private_int %arrZ %int_0
%64 = OpLoad %int %63
%65 = OpConvertSToF %float %64
%66 = OpCompositeConstruct %v4float %float_0 %51 %58 %65
%68 = OpAccessChain %_ptr_Workgroup_v4float %s %int_3
OpStore %68 %66
%72 = OpLoad %v3uint %gl_NumWorkGroups
%73 = OpIAdd %v3uint %72 %gl_WorkGroupSize
%75 = OpLoad %v3uint %gl_WorkGroupID
%76 = OpIAdd %v3uint %73 %75
%78 = OpLoad %v3uint %gl_LocalInvocationID
%79 = OpIAdd %v3uint %76 %78
%81 = OpLoad %v3uint %gl_GlobalInvocationID
%84 = OpLoad %uint %gl_LocalInvocationIndex
%85 = OpCompositeConstruct %v3uint %84 %84 %84
%86 = OpIMul %v3uint %81 %85
%87 = OpIAdd %v3uint %79 %86
%89 = OpAccessChain %_ptr_Uniform_v3uint %bInst %int_1
OpStore %89 %87
OpReturn
OpFunctionEnd
......@@ -5,6 +5,7 @@ layout(local_size_x_id = 18, local_size_z_id = 19) in;
layout(local_size_x = 2) in;
layout(local_size_y = 5) in;
layout(local_size_z = 7) in;
const int total = gl_MaxComputeWorkGroupCount.x
+ gl_MaxComputeWorkGroupCount.y
+ gl_MaxComputeWorkGroupCount.z
......@@ -17,12 +18,24 @@ int arrX[gl_WorkGroupSize.x];
int arrY[gl_WorkGroupSize.y];
int arrZ[gl_WorkGroupSize.z];
layout(binding = 0, set = 0) buffer bName {
int size;
uvec3 count;
vec4 data[];
} bInst;
void main()
{
barrier();
bInst.data[bInst.size / 2] *= vec4(7.0);
memoryBarrier();
memoryBarrierShared();
groupMemoryBarrier();
s[3] = vec4(1.0);
s[3] *= arrX[0] * arrY[0] * arrZ[0];
memoryBarrierShared();
memoryBarrierBuffer();
s[3] = vec4(0, arrX[0], arrY[0], arrZ[0]);
bInst.count = gl_NumWorkGroups + gl_WorkGroupSize + gl_WorkGroupID + gl_LocalInvocationID +
gl_GlobalInvocationID * gl_LocalInvocationIndex;
}
......@@ -3819,13 +3819,13 @@ void TBuiltIns::initialize(int version, EProfile profile, const SpvVersion& spvV
stageBuiltins[EShLangCompute].append(
"void memoryBarrierShared();"
"void groupMemoryBarrier();"
"void memoryBarrierBuffer();"
);
}
#ifndef GLSLANG_WEB
if ((profile != EEsProfile && version >= 420) || esBarrier) {
commonBuiltins.append(
"void memoryBarrierAtomicCounter();"
"void memoryBarrierBuffer();"
"void memoryBarrierImage();"
);
}
......@@ -7692,8 +7692,10 @@ void TBuiltIns::identifyBuiltIns(int version, EProfile profile, const SpvVersion
symbolTable.setFunctionExtensions("dFdyCoarse", 1, &E_GL_NV_compute_shader_derivatives);
symbolTable.setFunctionExtensions("fwidthCoarse", 1, &E_GL_NV_compute_shader_derivatives);
}
#endif
break;
#ifndef GLSLANG_WEB
case EShLangRayGenNV:
case EShLangIntersectNV:
case EShLangAnyHitNV:
......@@ -8108,6 +8110,7 @@ void TBuiltIns::identifyBuiltIns(int version, EProfile profile, const SpvVersion
symbolTable.relateToOperator("barrier", EOpBarrier);
symbolTable.relateToOperator("memoryBarrier", EOpMemoryBarrier);
symbolTable.relateToOperator("memoryBarrierBuffer", EOpMemoryBarrierBuffer);
#ifndef GLSLANG_WEB
symbolTable.relateToOperator("packInt2x16", EOpPackInt2x16);
......@@ -8132,7 +8135,6 @@ void TBuiltIns::identifyBuiltIns(int version, EProfile profile, const SpvVersion
symbolTable.relateToOperator("controlBarrier", EOpBarrier);
symbolTable.relateToOperator("memoryBarrierAtomicCounter", EOpMemoryBarrierAtomicCounter);
symbolTable.relateToOperator("memoryBarrierBuffer", EOpMemoryBarrierBuffer);
symbolTable.relateToOperator("memoryBarrierImage", EOpMemoryBarrierImage);
symbolTable.relateToOperator("atomicLoad", EOpAtomicLoad);
......
......@@ -3813,10 +3813,6 @@ void TParseContext::arraySizesCheck(const TSourceLoc& loc, const TQualifier& qua
// for ES, if size isn't coming from an initializer, it has to be explicitly declared now,
// with very few exceptions
// last member of ssbo block exception:
if (qualifier.storage == EvqBuffer && lastMember)
return;
// implicitly-sized io exceptions:
switch (language) {
case EShLangGeometry:
......@@ -3851,6 +3847,10 @@ void TParseContext::arraySizesCheck(const TSourceLoc& loc, const TQualifier& qua
#endif
// last member of ssbo block exception:
if (qualifier.storage == EvqBuffer && lastMember)
return;
arraySizeRequiredCheck(loc, *arraySizes);
}
......@@ -5384,6 +5384,7 @@ void TParseContext::setLayoutQualifier(const TSourceLoc& loc, TPublicType& publi
#endif
case EShLangCompute:
if (id.compare(0, 11, "local_size_") == 0) {
#ifndef GLSLANG_WEB
if (language == EShLangMeshNV || language == EShLangTaskNV) {
requireExtensions(loc, 1, &E_GL_NV_mesh_shader, "gl_WorkGroupSize");
}
......@@ -5392,6 +5393,7 @@ void TParseContext::setLayoutQualifier(const TSourceLoc& loc, TPublicType& publi
profileRequires(loc, EEsProfile, 310, 0, "gl_WorkGroupSize");
profileRequires(loc, ~EEsProfile, 430, E_GL_ARB_compute_shader, "gl_WorkGroupSize");
}
#endif
if (nonLiteral)
error(loc, "needs a literal integer", "local_size", "");
if (id.size() == 12 && value == 0) {
......@@ -8017,7 +8019,9 @@ void TParseContext::updateStandaloneQualifierDefaults(const TSourceLoc& loc, con
}
if (intermediate.getLocalSize(i) > (unsigned int)max)
error(loc, "too large; see gl_MaxComputeWorkGroupSize", "local_size", "");
} else if (language == EShLangMeshNV) {
}
#ifndef GLSLANG_WEB
else if (language == EShLangMeshNV) {
switch (i) {
case 0: max = resources.maxMeshWorkGroupSizeX_NV; break;
case 1: max = resources.maxMeshWorkGroupSizeY_NV; break;
......@@ -8035,7 +8039,9 @@ void TParseContext::updateStandaloneQualifierDefaults(const TSourceLoc& loc, con
}
if (intermediate.getLocalSize(i) > (unsigned int)max)
error(loc, "too large; see gl_MaxTaskWorkGroupSizeNV", "local_size", "");
} else {
}
#endif
else {
assert(0);
}
......
......@@ -343,6 +343,7 @@ void TScanContext::fillInKeywordMap()
(*KeywordMap)["const"] = CONST;
(*KeywordMap)["uniform"] = UNIFORM;
(*KeywordMap)["buffer"] = BUFFER;
(*KeywordMap)["in"] = IN;
(*KeywordMap)["out"] = OUT;
(*KeywordMap)["smooth"] = SMOOTH;
......@@ -410,7 +411,6 @@ void TScanContext::fillInKeywordMap()
(*KeywordMap)["attribute"] = ATTRIBUTE;
(*KeywordMap)["varying"] = VARYING;
(*KeywordMap)["noperspective"] = NOPERSPECTIVE;
(*KeywordMap)["buffer"] = BUFFER;
(*KeywordMap)["coherent"] = COHERENT;
(*KeywordMap)["devicecoherent"] = DEVICECOHERENT;
(*KeywordMap)["queuefamilycoherent"] = QUEUEFAMILYCOHERENT;
......@@ -905,6 +905,13 @@ int TScanContext::tokenizeIdentifier()
case CASE:
return keyword;
case BUFFER:
afterBuffer = true;
if ((parseContext.isEsProfile() && parseContext.version < 310) ||
(!parseContext.isEsProfile() && parseContext.version < 430))
return identifierOrType();
return keyword;
case STRUCT:
afterStruct = true;
return keyword;
......@@ -998,12 +1005,6 @@ int TScanContext::tokenizeIdentifier()
if (parseContext.isEsProfile() && parseContext.version >= 300)
reservedWord();
return keyword;
case BUFFER:
afterBuffer = true;
if ((parseContext.isEsProfile() && parseContext.version < 310) ||
(!parseContext.isEsProfile() && parseContext.version < 430))
return identifierOrType();
return keyword;
case PAYLOADNV:
case PAYLOADINNV:
case HITATTRNV:
......
......@@ -38,7 +38,7 @@
//
// Do not edit the .y file, only edit the .m4 file.
// The .y bison file is not a source file, it is a derivitive of the .m4 file.
// The .y bison file is not a source file, it is a derivative of the .m4 file.
// The m4 file needs to be processed by m4 to generate the .y bison file.
//
// Code sandwiched between a pair:
......@@ -49,7 +49,7 @@
// ...
// GLSLANG_WEB_EXCLUDE_OFF
//
// Will be exluded from the grammar when m4 is executed as:
// Will be excluded from the grammar when m4 is executed as:
//
// m4 -P -DGLSLANG_WEB
//
......@@ -277,7 +277,7 @@ GLSLANG_WEB_EXCLUDE_OFF
%token <lex> CENTROID IN OUT INOUT
%token <lex> STRUCT VOID WHILE
%token <lex> BREAK CONTINUE DO ELSE FOR IF DISCARD RETURN SWITCH CASE DEFAULT
%token <lex> UNIFORM SHARED
%token <lex> UNIFORM SHARED BUFFER
%token <lex> FLAT SMOOTH LAYOUT
GLSLANG_WEB_EXCLUDE_ON
......@@ -285,7 +285,7 @@ GLSLANG_WEB_EXCLUDE_ON
%token <lex> INT64CONSTANT UINT64CONSTANT
%token <lex> SUBROUTINE DEMOTE
%token <lex> PAYLOADNV PAYLOADINNV HITATTRNV CALLDATANV CALLDATAINNV
%token <lex> PATCH SAMPLE BUFFER NONUNIFORM
%token <lex> PATCH SAMPLE NONUNIFORM
%token <lex> COHERENT VOLATILE RESTRICT READONLY WRITEONLY DEVICECOHERENT QUEUEFAMILYCOHERENT WORKGROUPCOHERENT
%token <lex> SUBGROUPCOHERENT NONPRIVATE
%token <lex> NOPERSPECTIVE EXPLICITINTERPAMD PERVERTEXNV PERPRIMITIVENV PERVIEWNV PERTASKNV
......@@ -1368,12 +1368,12 @@ storage_qualifier
$$.init($1.loc);
$$.qualifier.storage = EvqShared;
}
GLSLANG_WEB_EXCLUDE_ON
| BUFFER {
parseContext.globalCheck($1.loc, "buffer");
$$.init($1.loc);
$$.qualifier.storage = EvqBuffer;
}
GLSLANG_WEB_EXCLUDE_ON
| ATTRIBUTE {
parseContext.requireStage($1.loc, EShLangVertex, "attribute");
parseContext.checkDeprecated($1.loc, ECoreProfile, 130, "attribute");
......
......@@ -38,7 +38,7 @@
//
// Do not edit the .y file, only edit the .m4 file.
// The .y bison file is not a source file, it is a derivitive of the .m4 file.
// The .y bison file is not a source file, it is a derivative of the .m4 file.
// The m4 file needs to be processed by m4 to generate the .y bison file.
//
// Code sandwiched between a pair:
......@@ -49,7 +49,7 @@
// ...
// GLSLANG_WEB_EXCLUDE_OFF
//
// Will be exluded from the grammar when m4 is executed as:
// Will be excluded from the grammar when m4 is executed as:
//
// m4 -P -DGLSLANG_WEB
//
......@@ -194,7 +194,7 @@ extern int yylex(YYSTYPE*, TParseContext&);
%token <lex> CENTROID IN OUT INOUT
%token <lex> STRUCT VOID WHILE
%token <lex> BREAK CONTINUE DO ELSE FOR IF DISCARD RETURN SWITCH CASE DEFAULT
%token <lex> UNIFORM SHARED
%token <lex> UNIFORM SHARED BUFFER
%token <lex> FLAT SMOOTH LAYOUT
......@@ -1164,6 +1164,11 @@ storage_qualifier
$$.init($1.loc);
$$.qualifier.storage = EvqShared;
}
| BUFFER {
parseContext.globalCheck($1.loc, "buffer");
$$.init($1.loc);
$$.qualifier.storage = EvqBuffer;
}
;
......
This source diff could not be displayed because it is too large. You can view the blob instead.
......@@ -182,9 +182,10 @@ extern int yydebug;
DEFAULT = 392,
UNIFORM = 393,
SHARED = 394,
FLAT = 395,
SMOOTH = 396,
LAYOUT = 397
BUFFER = 395,
FLAT = 396,
SMOOTH = 397,
LAYOUT = 398
};
#endif
......@@ -229,7 +230,7 @@ union YYSTYPE
glslang::TArraySizes* typeParameters;
} interm;
#line 233 "MachineIndependent/glslang_tab.cpp.h" /* yacc.c:1909 */
#line 234 "MachineIndependent/glslang_tab.cpp.h" /* yacc.c:1909 */
};
typedef union YYSTYPE YYSTYPE;
......
Markdown is supported
0% or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment