BeeEngine/glslang-zig
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

Merge pull request #2353 from vkushwaha-nv/SPV_EXT_shader_atomic_float

Browse source 
Add changes for SPV_EXT_shader_atomic_float_add
This commit is contained in:
John Kessenich 2020-07-22 22:06:01 +07:00 • committed by GitHub
commit 7923b3e3d9
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
13 changed files with 1109 additions and 5 deletions
Download patch file Download diff file Expand all files Collapse all files

43
glslang/MachineIndependent/Initialize.cpp
View file

@ -1241,11 +1241,19 @@ void TBuiltIns::initialize(int version, EProfile profile, const SpvVersion& spvV
" int64_t atomicAdd(coherent volatile inout int64_t, int64_t);"
"uint64_t atomicAdd(coherent volatile inout uint64_t, uint64_t, int, int, int);"
" int64_t atomicAdd(coherent volatile inout int64_t, int64_t, int, int, int);"
" float atomicAdd(coherent volatile inout float, float);"
" float atomicAdd(coherent volatile inout float, float, int, int, int);"
" double atomicAdd(coherent volatile inout double, double);"
" double atomicAdd(coherent volatile inout double, double, int, int, int);"
"uint64_t atomicExchange(coherent volatile inout uint64_t, uint64_t);"
" int64_t atomicExchange(coherent volatile inout int64_t, int64_t);"
"uint64_t atomicExchange(coherent volatile inout uint64_t, uint64_t, int, int, int);"
" int64_t atomicExchange(coherent volatile inout int64_t, int64_t, int, int, int);"
" float atomicExchange(coherent volatile inout float, float);"
" float atomicExchange(coherent volatile inout float, float, int, int, int);"
" double atomicExchange(coherent volatile inout double, double);"
" double atomicExchange(coherent volatile inout double, double, int, int, int);"
"uint64_t atomicCompSwap(coherent volatile inout uint64_t, uint64_t, uint64_t);"
" int64_t atomicCompSwap(coherent volatile inout int64_t, int64_t, int64_t);"
@ -1254,9 +1262,13 @@ void TBuiltIns::initialize(int version, EProfile profile, const SpvVersion& spvV
"uint64_t atomicLoad(coherent volatile in uint64_t, int, int, int);"
" int64_t atomicLoad(coherent volatile in int64_t, int, int, int);"
" float atomicLoad(coherent volatile in float, int, int, int);"
" double atomicLoad(coherent volatile in double, int, int, int);"
"void atomicStore(coherent volatile out uint64_t, uint64_t, int, int, int);"
"void atomicStore(coherent volatile out int64_t, int64_t, int, int, int);"
"void atomicStore(coherent volatile out float, float, int, int, int);"
"void atomicStore(coherent volatile out double, double, int, int, int);"
"\n");
}
#endif
@ -6013,12 +6025,39 @@ void TBuiltIns::addImageFunctions(TSampler sampler, const TString& typeName, int
// not int or uint
// GL_ARB_ES3_1_compatibility
// TODO: spec issue: are there restrictions on the kind of layout() that can be used? what about dropping memory qualifiers?
if ((profile != EEsProfile && version >= 450) ||
(profile == EEsProfile && version >= 310)) {
if (profile == EEsProfile && version >= 310) {
commonBuiltins.append("float imageAtomicExchange(volatile coherent ");
commonBuiltins.append(imageParams);
commonBuiltins.append(", float);\n");
}
if (profile != EEsProfile && version >= 450) {
commonBuiltins.append("float imageAtomicAdd(volatile coherent ");
commonBuiltins.append(imageParams);
commonBuiltins.append(", float);\n");
commonBuiltins.append("float imageAtomicAdd(volatile coherent ");
commonBuiltins.append(imageParams);
commonBuiltins.append(", float");
commonBuiltins.append(", int, int, int);\n");
commonBuiltins.append("float imageAtomicExchange(volatile coherent ");
commonBuiltins.append(imageParams);
commonBuiltins.append(", float);\n");
commonBuiltins.append("float imageAtomicExchange(volatile coherent ");
commonBuiltins.append(imageParams);
commonBuiltins.append(", float");
commonBuiltins.append(", int, int, int);\n");
commonBuiltins.append("float imageAtomicLoad(readonly volatile coherent ");
commonBuiltins.append(imageParams);
commonBuiltins.append(", int, int, int);\n");
commonBuiltins.append("void imageAtomicStore(writeonly volatile coherent ");
commonBuiltins.append(imageParams);
commonBuiltins.append(", float");
commonBuiltins.append(", int, int, int);\n");
}
}
}

17
glslang/MachineIndependent/ParseHelper.cpp
View file

@ -2110,7 +2110,14 @@ void TParseContext::builtInOpCheck(const TSourceLoc& loc, const TFunction& fnCan
if (imageType.getQualifier().getFormat() != ElfR32i && imageType.getQualifier().getFormat() != ElfR32ui)
error(loc, "only supported on image with format r32i or r32ui", fnCandidate.getName().c_str(), "");
} else {
if (fnCandidate.getName().compare(0, 19, "imageAtomicExchange") != 0)
bool isImageAtomicOnFloatAllowed = ((fnCandidate.getName().compare(0, 14, "imageAtomicAdd") == 0) ||
(fnCandidate.getName().compare(0, 15, "imageAtomicLoad") == 0) ||
(fnCandidate.getName().compare(0, 16, "imageAtomicStore") == 0) ||
(fnCandidate.getName().compare(0, 19, "imageAtomicExchange") == 0));
if (imageType.getSampler().type == EbtFloat && isImageAtomicOnFloatAllowed &&
(fnCandidate.getName().compare(0, 19, "imageAtomicExchange") != 0)) // imageAtomicExchange doesn't require GL_EXT_shader_atomic_float
requireExtensions(loc, 1, &E_GL_EXT_shader_atomic_float, fnCandidate.getName().c_str());
if (!isImageAtomicOnFloatAllowed)
error(loc, "only supported on integer images", fnCandidate.getName().c_str(), "");
else if (imageType.getQualifier().getFormat() != ElfR32f && isEsProfile())
error(loc, "only supported on image with format r32f", fnCandidate.getName().c_str(), "");
@ -2139,10 +2146,18 @@ void TParseContext::builtInOpCheck(const TSourceLoc& loc, const TFunction& fnCan
if (argp->size() > 3) {
requireExtensions(loc, 1, &E_GL_KHR_memory_scope_semantics, fnCandidate.getName().c_str());
memorySemanticsCheck(loc, fnCandidate, callNode);
if ((callNode.getOp() == EOpAtomicAdd || callNode.getOp() == EOpAtomicExchange ||
callNode.getOp() == EOpAtomicLoad || callNode.getOp() == EOpAtomicStore) &&
(arg0->getType().isFloatingDomain())) {
requireExtensions(loc, 1, &E_GL_EXT_shader_atomic_float, fnCandidate.getName().c_str());
}
} else if (arg0->getType().getBasicType() == EbtInt64 || arg0->getType().getBasicType() == EbtUint64) {
const char* const extensions[2] = { E_GL_NV_shader_atomic_int64,
E_GL_EXT_shader_atomic_int64 };
requireExtensions(loc, 2, extensions, fnCandidate.getName().c_str());
} else if ((callNode.getOp() == EOpAtomicAdd || callNode.getOp() == EOpAtomicExchange) &&
(arg0->getType().isFloatingDomain())) {
requireExtensions(loc, 1, &E_GL_EXT_shader_atomic_float, fnCandidate.getName().c_str());
}
break;
}

3
glslang/MachineIndependent/Versions.cpp
View file

@ -347,6 +347,7 @@ void TParseVersions::initializeExtensionBehavior()
extensionBehavior[E_GL_EXT_shader_subgroup_extended_types_int16] = EBhDisable;
extensionBehavior[E_GL_EXT_shader_subgroup_extended_types_int64] = EBhDisable;
extensionBehavior[E_GL_EXT_shader_subgroup_extended_types_float16] = EBhDisable;
extensionBehavior[E_GL_EXT_shader_atomic_float] = EBhDisable;
}
#endif // GLSLANG_WEB
@ -520,6 +521,8 @@ void TParseVersions::getPreamble(std::string& preamble)
"#define GL_EXT_shader_subgroup_extended_types_int16 1\n"
"#define GL_EXT_shader_subgroup_extended_types_int64 1\n"
"#define GL_EXT_shader_subgroup_extended_types_float16 1\n"
"#define GL_EXT_shader_atomic_float 1\n"
;
if (version >= 150) {

2
glslang/MachineIndependent/Versions.h
View file

@ -298,6 +298,8 @@ const char* const E_GL_EXT_shader_subgroup_extended_types_int16 = "GL_EXT_shad
const char* const E_GL_EXT_shader_subgroup_extended_types_int64 = "GL_EXT_shader_subgroup_extended_types_int64";
const char* const E_GL_EXT_shader_subgroup_extended_types_float16 = "GL_EXT_shader_subgroup_extended_types_float16";
const char* const E_GL_EXT_shader_atomic_float = "GL_EXT_shader_atomic_float";
// Arrays of extensions for the above AEP duplications
const char* const AEP_geometry_shader[] = { E_GL_EXT_geometry_shader, E_GL_OES_geometry_shader };