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

Avoid generating 8/16-bit constants when 8/16-bit arithmetic extensions aren't enabled

Browse source
This commit is contained in:
Jeff Bolz 2019-07-10 13:06:15 -05:00
parent f970253a5d
commit c1ad396258
10 changed files with 136 additions and 80 deletions
Download patch file Download diff file Expand all files Collapse all files

36
glslang/MachineIndependent/Intermediate.cpp
View file

@ -574,24 +574,6 @@ TIntermTyped* TIntermediate::createConversion(TBasicType convertTo, TIntermTyped
TOperator newOp = EOpNull;
// Certain explicit conversions are allowed conditionally
bool arithemeticInt8Enabled = extensionRequested(E_GL_EXT_shader_explicit_arithmetic_types) ||
extensionRequested(E_GL_EXT_shader_explicit_arithmetic_types_int8);
#ifdef AMD_EXTENSIONS
bool arithemeticInt16Enabled = extensionRequested(E_GL_EXT_shader_explicit_arithmetic_types) ||
extensionRequested(E_GL_EXT_shader_explicit_arithmetic_types_int16) ||
extensionRequested(E_GL_AMD_gpu_shader_int16);
bool arithemeticFloat16Enabled = extensionRequested(E_GL_EXT_shader_explicit_arithmetic_types) ||
extensionRequested(E_GL_EXT_shader_explicit_arithmetic_types_float16) ||
extensionRequested(E_GL_AMD_gpu_shader_half_float);
#else
bool arithemeticInt16Enabled = extensionRequested(E_GL_EXT_shader_explicit_arithmetic_types) ||
extensionRequested(E_GL_EXT_shader_explicit_arithmetic_types_int16);
bool arithemeticFloat16Enabled = extensionRequested(E_GL_EXT_shader_explicit_arithmetic_types) ||
extensionRequested(E_GL_EXT_shader_explicit_arithmetic_types_float16);
#endif
bool convertToIntTypes = (convertTo == EbtInt8 || convertTo == EbtUint8 ||
convertTo == EbtInt16 || convertTo == EbtUint16 ||
convertTo == EbtInt || convertTo == EbtUint ||
@ -608,19 +590,19 @@ TIntermTyped* TIntermediate::createConversion(TBasicType convertTo, TIntermTyped
node->getBasicType() == EbtFloat ||
node->getBasicType() == EbtDouble);
if (! arithemeticInt8Enabled) {
if (! getArithemeticInt8Enabled()) {
if (((convertTo == EbtInt8 || convertTo == EbtUint8) && ! convertFromIntTypes) ||
((node->getBasicType() == EbtInt8 || node->getBasicType() == EbtUint8) && ! convertToIntTypes))
return nullptr;
}
if (! arithemeticInt16Enabled) {
if (! getArithemeticInt16Enabled()) {
if (((convertTo == EbtInt16 || convertTo == EbtUint16) && ! convertFromIntTypes) ||
((node->getBasicType() == EbtInt16 || node->getBasicType() == EbtUint16) && ! convertToIntTypes))
return nullptr;
}
if (! arithemeticFloat16Enabled) {
if (! getArithemeticFloat16Enabled()) {
if ((convertTo == EbtFloat16 && ! convertFromFloatTypes) ||
(node->getBasicType() == EbtFloat16 && ! convertToFloatTypes))
return nullptr;
@ -841,9 +823,15 @@ TIntermTyped* TIntermediate::createConversion(TBasicType convertTo, TIntermTyped
newNode = addUnaryNode(newOp, node, node->getLoc(), newType);
if (node->getAsConstantUnion()) {
TIntermTyped* folded = node->getAsConstantUnion()->fold(newOp, newType);
if (folded)
return folded;
// 8/16-bit storage extensions don't support 8/16-bit constants, so don't fold conversions
// to those types
if ((getArithemeticInt8Enabled() || !(convertTo == EbtInt8 || convertTo == EbtUint8)) &&
(getArithemeticInt16Enabled() || !(convertTo == EbtInt16 || convertTo == EbtUint16)) &&
(getArithemeticFloat16Enabled() || !(convertTo == EbtFloat16))) {
TIntermTyped* folded = node->getAsConstantUnion()->fold(newOp, newType);
if (folded)
return folded;
}
}
// Propagate specialization-constant-ness, if allowed

50
glslang/MachineIndependent/ParseHelper.cpp
View file

@ -6948,6 +6948,16 @@ TIntermTyped* TParseContext::constructBuiltIn(const TType& type, TOperator op, T
case EOpConstructF16Mat4x4:
case EOpConstructFloat16:
basicOp = EOpConstructFloat16;
// 8/16-bit storage extensions don't support constructing composites of 8/16-bit types,
// so construct a 32-bit type and convert
if (!intermediate.getArithemeticFloat16Enabled()) {
TType tempType(EbtFloat, EvqTemporary, type.getVectorSize());
newNode = node;
if (tempType != newNode->getType())
newNode = intermediate.setAggregateOperator(newNode, (TOperator)(EOpConstructVec2 + op - EOpConstructF16Vec2), tempType, node->getLoc());
newNode = intermediate.addConversion(EbtFloat16, newNode);
return newNode;
}
break;
case EOpConstructI8Vec2:
@ -6955,6 +6965,16 @@ TIntermTyped* TParseContext::constructBuiltIn(const TType& type, TOperator op, T
case EOpConstructI8Vec4:
case EOpConstructInt8:
basicOp = EOpConstructInt8;
// 8/16-bit storage extensions don't support constructing composites of 8/16-bit types,
// so construct a 32-bit type and convert
if (!intermediate.getArithemeticInt8Enabled()) {
TType tempType(EbtInt, EvqTemporary, type.getVectorSize());
newNode = node;
if (tempType != newNode->getType())
newNode = intermediate.setAggregateOperator(newNode, (TOperator)(EOpConstructIVec2 + op - EOpConstructI8Vec2), tempType, node->getLoc());
newNode = intermediate.addConversion(EbtInt8, newNode);
return newNode;
}
break;
case EOpConstructU8Vec2:
@ -6962,6 +6982,16 @@ TIntermTyped* TParseContext::constructBuiltIn(const TType& type, TOperator op, T
case EOpConstructU8Vec4:
case EOpConstructUint8:
basicOp = EOpConstructUint8;
// 8/16-bit storage extensions don't support constructing composites of 8/16-bit types,
// so construct a 32-bit type and convert
if (!intermediate.getArithemeticInt8Enabled()) {
TType tempType(EbtUint, EvqTemporary, type.getVectorSize());
newNode = node;
if (tempType != newNode->getType())
newNode = intermediate.setAggregateOperator(newNode, (TOperator)(EOpConstructUVec2 + op - EOpConstructU8Vec2), tempType, node->getLoc());
newNode = intermediate.addConversion(EbtUint8, newNode);
return newNode;
}
break;
case EOpConstructI16Vec2:
@ -6969,6 +6999,16 @@ TIntermTyped* TParseContext::constructBuiltIn(const TType& type, TOperator op, T
case EOpConstructI16Vec4:
case EOpConstructInt16:
basicOp = EOpConstructInt16;
// 8/16-bit storage extensions don't support constructing composites of 8/16-bit types,
// so construct a 32-bit type and convert
if (!intermediate.getArithemeticInt16Enabled()) {
TType tempType(EbtInt, EvqTemporary, type.getVectorSize());
newNode = node;
if (tempType != newNode->getType())
newNode = intermediate.setAggregateOperator(newNode, (TOperator)(EOpConstructIVec2 + op - EOpConstructI16Vec2), tempType, node->getLoc());
newNode = intermediate.addConversion(EbtInt16, newNode);
return newNode;
}
break;
case EOpConstructU16Vec2:
@ -6976,6 +7016,16 @@ TIntermTyped* TParseContext::constructBuiltIn(const TType& type, TOperator op, T
case EOpConstructU16Vec4:
case EOpConstructUint16:
basicOp = EOpConstructUint16;
// 8/16-bit storage extensions don't support constructing composites of 8/16-bit types,
// so construct a 32-bit type and convert
if (!intermediate.getArithemeticInt16Enabled()) {
TType tempType(EbtUint, EvqTemporary, type.getVectorSize());
newNode = node;
if (tempType != newNode->getType())
newNode = intermediate.setAggregateOperator(newNode, (TOperator)(EOpConstructUVec2 + op - EOpConstructU16Vec2), tempType, node->getLoc());
newNode = intermediate.addConversion(EbtUint16, newNode);
return newNode;
}
break;
case EOpConstructIVec2:

21
glslang/MachineIndependent/localintermediate.h
View file

@ -780,6 +780,27 @@ public:
const char* const implicitThisName;
const char* const implicitCounterName;
// Certain explicit conversions are allowed conditionally
bool getArithemeticInt8Enabled() const {
return extensionRequested(E_GL_EXT_shader_explicit_arithmetic_types) ||
extensionRequested(E_GL_EXT_shader_explicit_arithmetic_types_int8);
}
bool getArithemeticInt16Enabled() const {
return extensionRequested(E_GL_EXT_shader_explicit_arithmetic_types) ||
#ifdef AMD_EXTENSIONS
extensionRequested(E_GL_AMD_gpu_shader_int16) ||
#endif
extensionRequested(E_GL_EXT_shader_explicit_arithmetic_types_int16);
}
bool getArithemeticFloat16Enabled() const {
return extensionRequested(E_GL_EXT_shader_explicit_arithmetic_types) ||
#ifdef AMD_EXTENSIONS
extensionRequested(E_GL_AMD_gpu_shader_half_float) ||
#endif
extensionRequested(E_GL_EXT_shader_explicit_arithmetic_types_float16);
}
protected:
TIntermSymbol* addSymbol(int Id, const TString&, const TType&, const TConstUnionArray&, TIntermTyped* subtree, const TSourceLoc&);
void error(TInfoSink& infoSink, const char*);