Implement the extension GL_AMD_texture_gather_bias_lod

This commit is contained in:
Rex Xu 2016-11-17 17:47:59 +08:00
parent 94c18a84cd
commit 225e0fcadd
12 changed files with 770 additions and 11 deletions

View file

@ -3844,6 +3844,7 @@ void TBuiltIns::add2ndGenerationSamplingImaging(int version, EProfile profile, c
else {
addSamplingFunctions(sampler, typeName, version, profile);
addGatherFunctions(sampler, typeName, version, profile);
if (spvVersion.vulkan > 0 && sampler.dim == EsdBuffer && sampler.isCombined()) {
// Vulkan wants a textureBuffer to allow texelFetch() --
// a sampled image with no sampler.
@ -4349,6 +4350,7 @@ void TBuiltIns::addGatherFunctions(TSampler sampler, const TString& typeName, in
default:
break;
}
if (sparse)
s.append("ARB");
s.append("(");
@ -4388,6 +4390,116 @@ void TBuiltIns::addGatherFunctions(TSampler sampler, const TString& typeName, in
}
}
}
#ifdef AMD_EXTENSIONS
if (sampler.dim == EsdRect || sampler.shadow)
return;
if (profile == EEsProfile || version < 450)
return;
for (int bias = 0; bias < 2; ++bias) { // loop over presence of bias argument
for (int lod = 0; lod < 2; ++lod) { // loop over presence of lod argument
if ((lod && bias) || (lod == 0 && bias == 0))
continue;
for (int offset = 0; offset < 3; ++offset) { // loop over three forms of offset in the call name: none, Offset, and Offsets
for (int comp = 0; comp < 2; ++comp) { // loop over presence of comp argument
if (comp == 0 && bias)
continue;
if (offset > 0 && sampler.dim == EsdCube)
continue;
for (int sparse = 0; sparse <= 1; ++sparse) { // loop over "bool" sparse or not
if (sparse && (profile == EEsProfile || version < 450))
continue;
TString s;
// return type
if (sparse)
s.append("int ");
else {
s.append(prefixes[sampler.type]);
s.append("vec4 ");
}
// name
if (sparse)
s.append("sparseTextureGather");
else
s.append("textureGather");
if (lod)
s.append("Lod");
switch (offset) {
case 1:
s.append("Offset");
break;
case 2:
s.append("Offsets");
default:
break;
}
if (lod)
s.append("AMD");
else if (sparse)
s.append("ARB");
s.append("(");
// sampler type argument
s.append(typeName);
// P coordinate argument
s.append(",vec");
int totalDims = dimMap[sampler.dim] + (sampler.arrayed ? 1 : 0);
s.append(postfixes[totalDims]);
// lod argument
if (lod)
s.append(",float");
// offset argument
if (offset > 0) {
s.append(",ivec2");
if (offset == 2)
s.append("[4]");
}
// texel out (for sparse texture)
if (sparse) {
s.append(",out ");
s.append(prefixes[sampler.type]);
s.append("vec4 ");
}
// comp argument
if (comp)
s.append(",int");
// bias argument
if (bias)
s.append(",float");
s.append(");\n");
if (bias)
stageBuiltins[EShLangFragment].append(s);
else
commonBuiltins.append(s);
}
}
}
}
}
#endif
}
//
@ -5366,6 +5478,16 @@ void TBuiltIns::identifyBuiltIns(int version, EProfile profile, const SpvVersion
BuiltInVariable("gl_BaryCoordSmoothSampleAMD", EbvBaryCoordSmoothSample, symbolTable);
BuiltInVariable("gl_BaryCoordPullModelAMD", EbvBaryCoordPullModel, symbolTable);
}
// E_GL_AMD_texture_gather_bias_lod
if (profile != EEsProfile) {
symbolTable.setFunctionExtensions("textureGatherLodAMD", 1, &E_GL_AMD_texture_gather_bias_lod);
symbolTable.setFunctionExtensions("textureGatherLodOffsetAMD", 1, &E_GL_AMD_texture_gather_bias_lod);
symbolTable.setFunctionExtensions("textureGatherLodOffsetsAMD", 1, &E_GL_AMD_texture_gather_bias_lod);
symbolTable.setFunctionExtensions("sparseTextureGatherLodAMD", 1, &E_GL_AMD_texture_gather_bias_lod);
symbolTable.setFunctionExtensions("sparseTextureGatherLodOffsetAMD", 1, &E_GL_AMD_texture_gather_bias_lod);
symbolTable.setFunctionExtensions("sparseTextureGatherLodOffsetsAMD", 1, &E_GL_AMD_texture_gather_bias_lod);
}
#endif
symbolTable.setVariableExtensions("gl_FragDepthEXT", 1, &E_GL_EXT_frag_depth);
@ -5752,6 +5874,13 @@ void TBuiltIns::identifyBuiltIns(int version, EProfile profile, const SpvVersion
symbolTable.relateToOperator("cubeFaceIndexAMD", EOpCubeFaceIndex);
symbolTable.relateToOperator("cubeFaceCoordAMD", EOpCubeFaceCoord);
symbolTable.relateToOperator("timeAMD", EOpTime);
symbolTable.relateToOperator("textureGatherLodAMD", EOpTextureGatherLod);
symbolTable.relateToOperator("textureGatherLodOffsetAMD", EOpTextureGatherLodOffset);
symbolTable.relateToOperator("textureGatherLodOffsetsAMD", EOpTextureGatherLodOffsets);
symbolTable.relateToOperator("sparseTextureGatherLodAMD", EOpSparseTextureGatherLod);
symbolTable.relateToOperator("sparseTextureGatherLodOffsetAMD", EOpSparseTextureGatherLodOffset);
symbolTable.relateToOperator("sparseTextureGatherLodOffsetsAMD", EOpSparseTextureGatherLodOffsets);
#endif
}
}

View file

@ -1442,9 +1442,56 @@ void TParseContext::builtInOpCheck(const TSourceLoc& loc, const TFunction& fnCan
error(loc, "must be a compile-time constant:", feature, "component argument");
}
#ifdef AMD_EXTENSIONS
bool bias = false;
if (callNode.getOp() == EOpTextureGather)
bias = fnCandidate.getParamCount() > 3;
else if (callNode.getOp() == EOpTextureGatherOffset ||
callNode.getOp() == EOpTextureGatherOffsets)
bias = fnCandidate.getParamCount() > 4;
if (bias) {
featureString = fnCandidate.getName() + "with bias argument";
feature = featureString.c_str();
profileRequires(loc, ~EEsProfile, 450, nullptr, feature);
requireExtensions(loc, 1, &E_GL_AMD_texture_gather_bias_lod, feature);
}
#endif
break;
}
#ifdef AMD_EXTENSIONS
case EOpSparseTextureGather:
case EOpSparseTextureGatherOffset:
case EOpSparseTextureGatherOffsets:
{
bool bias = false;
if (callNode.getOp() == EOpSparseTextureGather)
bias = fnCandidate.getParamCount() > 4;
else if (callNode.getOp() == EOpSparseTextureGatherOffset ||
callNode.getOp() == EOpSparseTextureGatherOffsets)
bias = fnCandidate.getParamCount() > 5;
if (bias) {
TString featureString = fnCandidate.getName() + "with bias argument";
const char* feature = featureString.c_str();
profileRequires(loc, ~EEsProfile, 450, nullptr, feature);
requireExtensions(loc, 1, &E_GL_AMD_texture_gather_bias_lod, feature);
}
break;
}
case EOpSparseTextureGatherLod:
case EOpSparseTextureGatherLodOffset:
case EOpSparseTextureGatherLodOffsets:
{
requireExtensions(loc, 1, &E_GL_ARB_sparse_texture2, fnCandidate.getName().c_str());
break;
}
#endif
case EOpTextureOffset:
case EOpTextureFetchOffset:
case EOpTextureProjOffset:

View file

@ -194,6 +194,7 @@ void TParseVersions::initializeExtensionBehavior()
extensionBehavior[E_GL_AMD_shader_explicit_vertex_parameter] = EBhDisable;
extensionBehavior[E_GL_AMD_gcn_shader] = EBhDisable;
extensionBehavior[E_GL_AMD_gpu_shader_half_float] = EBhDisable;
extensionBehavior[E_GL_AMD_texture_gather_bias_lod] = EBhDisable;
#endif
#ifdef NV_EXTENSIONS
@ -316,6 +317,7 @@ void TParseVersions::getPreamble(std::string& preamble)
"#define GL_AMD_shader_explicit_vertex_parameter 1\n"
"#define GL_AMD_gcn_shader 1\n"
"#define GL_AMD_gpu_shader_half_float 1\n"
"#define GL_AMD_texture_gather_bias_lod 1\n"
#endif
#ifdef NV_EXTENSIONS

View file

@ -146,6 +146,7 @@ const char* const E_GL_AMD_shader_trinary_minmax = "GL_AMD_shader
const char* const E_GL_AMD_shader_explicit_vertex_parameter = "GL_AMD_shader_explicit_vertex_parameter";
const char* const E_GL_AMD_gcn_shader = "GL_AMD_gcn_shader";
const char* const E_GL_AMD_gpu_shader_half_float = "GL_AMD_gpu_shader_half_float";
const char* const E_GL_AMD_texture_gather_bias_lod = "GL_AMD_texture_gather_bias_lod";
#endif
#ifdef NV_EXTENSIONS

View file

@ -390,6 +390,8 @@ bool TOutputTraverser::visitUnary(TVisit /* visit */, TIntermUnary* node)
case EOpRcp: out.debug << "rcp"; break;
case EOpSaturate: out.debug << "saturate"; break;
case EOpSparseTexelsResident: out.debug << "sparseTexelsResident"; break;
#ifdef AMD_EXTENSIONS
case EOpMinInvocations: out.debug << "minInvocations"; break;
case EOpMaxInvocations: out.debug << "maxInvocations"; break;
@ -647,6 +649,37 @@ bool TOutputTraverser::visitAggregate(TVisit /* visit */, TIntermAggregate* node
case EOpTextureGather: out.debug << "textureGather"; break;
case EOpTextureGatherOffset: out.debug << "textureGatherOffset"; break;
case EOpTextureGatherOffsets: out.debug << "textureGatherOffsets"; break;
case EOpTextureClamp: out.debug << "textureClamp"; break;
case EOpTextureOffsetClamp: out.debug << "textureOffsetClamp"; break;
case EOpTextureGradClamp: out.debug << "textureGradClamp"; break;
case EOpTextureGradOffsetClamp: out.debug << "textureGradOffsetClamp"; break;
#ifdef AMD_EXTENSIONS
case EOpTextureGatherLod: out.debug << "textureGatherLod"; break;
case EOpTextureGatherLodOffset: out.debug << "textureGatherLodOffset"; break;
case EOpTextureGatherLodOffsets: out.debug << "textureGatherLodOffsets"; break;
#endif
case EOpSparseTexture: out.debug << "sparseTexture"; break;
case EOpSparseTextureOffset: out.debug << "sparseTextureOffset"; break;
case EOpSparseTextureLod: out.debug << "sparseTextureLod"; break;
case EOpSparseTextureLodOffset: out.debug << "sparseTextureLodOffset"; break;
case EOpSparseTextureFetch: out.debug << "sparseTexelFetch"; break;
case EOpSparseTextureFetchOffset: out.debug << "sparseTexelFetchOffset"; break;
case EOpSparseTextureGrad: out.debug << "sparseTextureGrad"; break;
case EOpSparseTextureGradOffset: out.debug << "sparseTextureGradOffset"; break;
case EOpSparseTextureGather: out.debug << "sparseTextureGather"; break;
case EOpSparseTextureGatherOffset: out.debug << "sparseTextureGatherOffset"; break;
case EOpSparseTextureGatherOffsets: out.debug << "sparseTextureGatherOffsets"; break;
case EOpSparseImageLoad: out.debug << "sparseImageLoad"; break;
case EOpSparseTextureClamp: out.debug << "sparseTextureClamp"; break;
case EOpSparseTextureOffsetClamp: out.debug << "sparseTextureOffsetClamp"; break;
case EOpSparseTextureGradClamp: out.debug << "sparseTextureGradClamp"; break;
case EOpSparseTextureGradOffsetClamp: out.debug << "sparseTextureGradOffsetClam"; break;
#ifdef AMD_EXTENSIONS
case EOpSparseTextureGatherLod: out.debug << "sparseTextureGatherLod"; break;
case EOpSparseTextureGatherLodOffset: out.debug << "sparseTextureGatherLodOffset"; break;
case EOpSparseTextureGatherLodOffsets: out.debug << "sparseTextureGatherLodOffsets"; break;
#endif
case EOpAddCarry: out.debug << "addCarry"; break;
case EOpSubBorrow: out.debug << "subBorrow"; break;