summaryrefslogtreecommitdiffstats
path: root/source/slang/slang-emit-cuda.cpp
diff options
context:
space:
mode:
authorjsmall-nvidia <jsmall@nvidia.com>2023-08-01 15:39:28 -0400
committerGitHub <noreply@github.com>2023-08-01 15:39:28 -0400
commit1653731718e75c297730dfb878e9f23895d1051d (patch)
tree76f05056594f9910e5baf464b05a41e48398fe18 /source/slang/slang-emit-cuda.cpp
parentedcc50cdcaf3743d4140b439375d0d40e3a941f7 (diff)
Fix literals needing cast (#3039)
* Cast integer literals. * Fix expected output. * For CUDA, search global instructions to see what types are used. Improve lookup for fp16 header in CUDA. * Fix issue with f16tof32 * Small improvement around finding used base types.
Diffstat (limited to 'source/slang/slang-emit-cuda.cpp')
-rw-r--r--source/slang/slang-emit-cuda.cpp47
1 files changed, 35 insertions, 12 deletions
diff --git a/source/slang/slang-emit-cuda.cpp b/source/slang/slang-emit-cuda.cpp
index a6501b5be..345aa3168 100644
--- a/source/slang/slang-emit-cuda.cpp
+++ b/source/slang/slang-emit-cuda.cpp
@@ -10,7 +10,29 @@
namespace Slang {
+static CUDAExtensionTracker::BaseTypeFlags _findBaseTypesUsed(IRModule* module)
+{
+ typedef CUDAExtensionTracker::BaseTypeFlags Flags;
+
+ // All basic types are hoistable so must be in global scope.
+ Flags baseTypesUsed = 0;
+
+ auto moduleInst = module->getModuleInst();
+
+ // Search all the insts in global scope, for BasicTypes
+ for (auto inst : moduleInst->getChildren())
+ {
+ if (auto basicType = as<IRBasicType>(inst))
+ {
+ // Get the base type, and set the bit
+ const auto baseTypeEnum = basicType->getBaseType();
+ baseTypesUsed |= Flags(1) << int(baseTypeEnum);
+ }
+ }
+
+ return baseTypesUsed;
+}
void CUDAExtensionTracker::finalize()
{
@@ -48,12 +70,8 @@ UnownedStringSlice CUDASourceEmitter::getBuiltinTypeName(IROp op)
case kIROp_IntPtrType: return UnownedStringSlice("int");
case kIROp_UIntPtrType: return UnownedStringSlice("uint");
#endif
- case kIROp_HalfType:
- {
- m_extensionTracker->requireBaseType(BaseType::Half);
- return UnownedStringSlice("__half");
- }
-
+ case kIROp_HalfType: return UnownedStringSlice("__half");
+
case kIROp_FloatType: return UnownedStringSlice("float");
case kIROp_DoubleType: return UnownedStringSlice("double");
default: return UnownedStringSlice();
@@ -77,11 +95,7 @@ UnownedStringSlice CUDASourceEmitter::getVectorPrefix(IROp op)
case kIROp_UIntType: return UnownedStringSlice("uint");
case kIROp_UInt64Type: return UnownedStringSlice("ulonglong");
- case kIROp_HalfType:
- {
- m_extensionTracker->requireBaseType(BaseType::Half);
- return UnownedStringSlice("__half");
- }
+ case kIROp_HalfType: return UnownedStringSlice("__half");
case kIROp_FloatType: return UnownedStringSlice("float");
case kIROp_DoubleType: return UnownedStringSlice("double");
@@ -424,8 +438,14 @@ void CUDASourceEmitter::_emitInitializerList(IRType* elementType, IRUse* operand
void CUDASourceEmitter::emitIntrinsicCallExprImpl(IRCall* inst, IRTargetIntrinsicDecoration* targetIntrinsic, EmitOpInfo const& inOuterPrec)
{
- if (targetIntrinsic->getDefinition().startsWith("__half"))
+ // This works around the problem, where some intrinsics that require the "half" type enabled don't use the half/float16_t type.
+ // For example `f16tof32` can operate on float16_t *and* uint. If the input is uint, although we are
+ // using the half feature (as far as CUDA is concerned), the half/float16_t type is not visible/directly used.
+ if (targetIntrinsic->getDefinition().startsWith(toSlice("__half")))
+ {
m_extensionTracker->requireBaseType(BaseType::Half);
+ }
+
Super::emitIntrinsicCallExprImpl(inst, targetIntrinsic, inOuterPrec);
}
@@ -795,6 +815,9 @@ bool CUDASourceEmitter::tryEmitGlobalParamImpl(IRGlobalParam* varDecl, IRType* v
void CUDASourceEmitter::emitModuleImpl(IRModule* module, DiagnosticSink* sink)
{
+ // Set up with all of the base types used in the module
+ m_extensionTracker->requireBaseTypes(_findBaseTypesUsed(module));
+
CLikeSourceEmitter::emitModuleImpl(module, sink);
// Emit all witness table definitions.