[llvm-branch-commits] [DXIL][Analysis] Replace #include with forward declaration. NFC (PR #100622)
https://github.com/bogner created https://github.com/llvm/llvm-project/pull/100622 None ___ llvm-branch-commits mailing list llvm-branch-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-branch-commits
[llvm-branch-commits] [DXIL][Analysis] Make the DXILResource binding optional. NFC (PR #100623)
https://github.com/bogner created https://github.com/llvm/llvm-project/pull/100623 This makes the binding structure in a DXILResource default to empty and need a separate call to set up, and also moves the unique ID into it since bindings are the only place where those are actually used. This will put us in a better position when dealing with resource handles in libraries. ___ llvm-branch-commits mailing list llvm-branch-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-branch-commits
[llvm-branch-commits] [DXIL][Analysis] Replace #include with forward declaration. NFC (PR #100622)
https://github.com/bogner updated https://github.com/llvm/llvm-project/pull/100622 ___ llvm-branch-commits mailing list llvm-branch-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-branch-commits
[llvm-branch-commits] [DXIL][Analysis] Replace #include with forward declaration. NFC (PR #100622)
https://github.com/bogner updated https://github.com/llvm/llvm-project/pull/100622 ___ llvm-branch-commits mailing list llvm-branch-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-branch-commits
[llvm-branch-commits] [DXIL][Analysis] Make the DXILResource binding optional. NFC (PR #100623)
https://github.com/bogner updated https://github.com/llvm/llvm-project/pull/100623 ___ llvm-branch-commits mailing list llvm-branch-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-branch-commits
[llvm-branch-commits] [DXIL][Analysis] Make the DXILResource binding optional. NFC (PR #100623)
https://github.com/bogner updated https://github.com/llvm/llvm-project/pull/100623 ___ llvm-branch-commits mailing list llvm-branch-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-branch-commits
[llvm-branch-commits] [DXIL][Analysis] Make alignment on StructuredBuffer optional (PR #100697)
https://github.com/bogner created https://github.com/llvm/llvm-project/pull/100697 HLSL allows StructuredBuffer<> to be defined with scalar or up-to-4-element vectors as well as with structs, but when doing so `dxc` doesn't set the alignment. Emulate this behaviour. ___ llvm-branch-commits mailing list llvm-branch-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-branch-commits
[llvm-branch-commits] [DirectX] Remove new-pm versions of DXILResource passes. NFC (PR #100698)
https://github.com/bogner created https://github.com/llvm/llvm-project/pull/100698 This is a bit of a hack, but these aren't actually used at the moment and they'll cause naming conflicts with DXIL resource passes that I'm adding to llvm/Analysis for replacing all of this with the target extension type based solution. I'd rather leave the existing stuff in place until we're ready to replace it completely, and this is the simplest way to get away with that. ___ llvm-branch-commits mailing list llvm-branch-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-branch-commits
[llvm-branch-commits] [DXIL][Analysis] Implement enough of DXILResourceAnalysis for buffers (PR #100699)
https://github.com/bogner created https://github.com/llvm/llvm-project/pull/100699 This implements the DXILResourceAnalysis pass for `dx.TypedBuffer` and `dx.RawBuffer` types. This should be sufficient to lower `dx.handle.fromBinding` for this set of types, but it leaves a number of TODOs around for other resource types. This also includes a straightforward `print` method in `ResourceInfo` to make the analysis testable. This is deliberately different than the printer in `lib/Target/DirectX/DXILResource.cpp`, which attempts to print bindings in a format compatible with the comments `dxc` prints. We will eventually want to make that functionality driven by this analysis pass, but it isn't sufficient for testing so we need both. ___ llvm-branch-commits mailing list llvm-branch-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-branch-commits
[llvm-branch-commits] [DXIL][Analysis] Boilerplate for DXILResourceAnalysis pass (PR #100700)
https://github.com/bogner created https://github.com/llvm/llvm-project/pull/100700 Broke this out into its own commit to make the next one easier to review. ___ llvm-branch-commits mailing list llvm-branch-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-branch-commits
[llvm-branch-commits] [DXIL][Analysis] Make alignment on StructuredBuffer optional (PR #100697)
https://github.com/bogner updated https://github.com/llvm/llvm-project/pull/100697 ___ llvm-branch-commits mailing list llvm-branch-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-branch-commits
[llvm-branch-commits] [DXIL][Analysis] Make alignment on StructuredBuffer optional (PR #100697)
https://github.com/bogner updated https://github.com/llvm/llvm-project/pull/100697 ___ llvm-branch-commits mailing list llvm-branch-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-branch-commits
[llvm-branch-commits] [DXIL][Analysis] Make alignment on StructuredBuffer optional (PR #100697)
https://github.com/bogner edited https://github.com/llvm/llvm-project/pull/100697 ___ llvm-branch-commits mailing list llvm-branch-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-branch-commits
[llvm-branch-commits] [DXIL][Analysis] Make alignment on StructuredBuffer optional (PR #100697)
https://github.com/bogner updated https://github.com/llvm/llvm-project/pull/100697 ___ llvm-branch-commits mailing list llvm-branch-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-branch-commits
[llvm-branch-commits] [DXIL][Analysis] Make alignment on StructuredBuffer optional (PR #100697)
https://github.com/bogner updated https://github.com/llvm/llvm-project/pull/100697 ___ llvm-branch-commits mailing list llvm-branch-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-branch-commits
[llvm-branch-commits] [DXIL][Analysis] Implement enough of DXILResourceAnalysis for buffers (PR #100699)
@@ -331,6 +336,249 @@ std::pair
ResourceInfo::getAnnotateProps() const {
return {Word0, Word1};
}
+void ResourceInfo::print(raw_ostream &OS) const {
+ OS << " Symbol: ";
+ Symbol->printAsOperand(OS);
+ OS << "\n";
+
+ OS << " Name: \"" << Name << "\"\n"
+ << " Binding:\n"
+ << "Unique ID: " << Binding.UniqueID << "\n"
+ << "Space: " << Binding.Space << "\n"
+ << "Lower Bound: " << Binding.LowerBound << "\n"
+ << "Size: " << Binding.Size << "\n"
+ << " Class: " << static_cast(RC) << "\n"
+ << " Kind: " << static_cast(Kind) << "\n";
+
+ if (isCBuffer()) {
+OS << " CBuffer size: " << CBufferSize << "\n";
+ } else if (isSampler()) {
+OS << " Sampler Type: " << static_cast(SamplerTy) << "\n";
+ } else {
+if (isUAV()) {
+ OS << " Globally Coherent: " << UAVFlags.GloballyCoherent << "\n"
+ << " HasCounter: " << UAVFlags.HasCounter << "\n"
+ << " IsROV: " << UAVFlags.IsROV << "\n";
+}
+if (isMultiSample())
+ OS << " Sample Count: " << MultiSample.Count << "\n";
+
+if (isStruct()) {
+ OS << " Buffer Stride: " << Struct.Stride << "\n";
+ uint32_t AlignLog2 = Struct.Alignment ? Log2(*Struct.Alignment) : 0;
+ OS << " Alignment: " << AlignLog2 << "\n";
+} else if (isTyped()) {
+ OS << " Element Type: " << static_cast(Typed.ElementTy) <<
"\n"
+ << " Element Count: " << static_cast(Typed.ElementCount)
+ << "\n";
+} else if (isFeedback())
+ OS << " Feedback Type: " << static_cast(Feedback.Type) <<
"\n";
+ }
+}
+
+//===--===//
+// ResourceMapper
+
+static dxil::ElementType toDXILElementType(Type *Ty, bool IsSigned) {
+ // TODO: Handle unorm, snorm, and packed.
+ Ty = Ty->getScalarType();
bogner wrote:
No, `Type::getScalarType` cannot return nullptr.
https://github.com/llvm/llvm-project/pull/100699
___
llvm-branch-commits mailing list
llvm-branch-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-branch-commits
[llvm-branch-commits] [DXIL][Analysis] Implement enough of DXILResourceAnalysis for buffers (PR #100699)
@@ -331,6 +336,249 @@ std::pair
ResourceInfo::getAnnotateProps() const {
return {Word0, Word1};
}
+void ResourceInfo::print(raw_ostream &OS) const {
+ OS << " Symbol: ";
+ Symbol->printAsOperand(OS);
+ OS << "\n";
+
+ OS << " Name: \"" << Name << "\"\n"
+ << " Binding:\n"
+ << "Unique ID: " << Binding.UniqueID << "\n"
+ << "Space: " << Binding.Space << "\n"
+ << "Lower Bound: " << Binding.LowerBound << "\n"
+ << "Size: " << Binding.Size << "\n"
+ << " Class: " << static_cast(RC) << "\n"
+ << " Kind: " << static_cast(Kind) << "\n";
+
+ if (isCBuffer()) {
+OS << " CBuffer size: " << CBufferSize << "\n";
+ } else if (isSampler()) {
+OS << " Sampler Type: " << static_cast(SamplerTy) << "\n";
+ } else {
+if (isUAV()) {
+ OS << " Globally Coherent: " << UAVFlags.GloballyCoherent << "\n"
+ << " HasCounter: " << UAVFlags.HasCounter << "\n"
+ << " IsROV: " << UAVFlags.IsROV << "\n";
+}
+if (isMultiSample())
+ OS << " Sample Count: " << MultiSample.Count << "\n";
+
+if (isStruct()) {
+ OS << " Buffer Stride: " << Struct.Stride << "\n";
+ uint32_t AlignLog2 = Struct.Alignment ? Log2(*Struct.Alignment) : 0;
+ OS << " Alignment: " << AlignLog2 << "\n";
+} else if (isTyped()) {
+ OS << " Element Type: " << static_cast(Typed.ElementTy) <<
"\n"
+ << " Element Count: " << static_cast(Typed.ElementCount)
+ << "\n";
+} else if (isFeedback())
+ OS << " Feedback Type: " << static_cast(Feedback.Type) <<
"\n";
+ }
+}
+
+//===--===//
+// ResourceMapper
+
+static dxil::ElementType toDXILElementType(Type *Ty, bool IsSigned) {
+ // TODO: Handle unorm, snorm, and packed.
+ Ty = Ty->getScalarType();
+
+ if (Ty->isIntegerTy()) {
+switch (Ty->getIntegerBitWidth()) {
+case 16:
+ return IsSigned ? ElementType::I16 : ElementType::U16;
+case 32:
+ return IsSigned ? ElementType::I32 : ElementType::U32;
+case 64:
+ return IsSigned ? ElementType::I64 : ElementType::U64;
+case 1:
+default:
+ return ElementType::Invalid;
+}
+ } else if (Ty->isFloatTy()) {
+return ElementType::F32;
+ } else if (Ty->isDoubleTy()) {
+return ElementType::F64;
+ } else if (Ty->isHalfTy()) {
+return ElementType::F16;
+ }
+
+ return ElementType::Invalid;
+}
+
+namespace {
+
+class ResourceMapper {
+ Module &M;
+ LLVMContext &Context;
+ DXILResourceMap &Resources;
+
+ // Unique ID is per resource type to match DXC.
+ uint32_t NextUAV = 0;
+ uint32_t NextSRV = 0;
+ uint32_t NextCBuf = 0;
+ uint32_t NextSmp = 0;
bogner wrote:
This is the "Unique resource record ID" in the [DXIL resource
metadata](https://github.com/microsoft/DirectXShaderCompiler/blob/main/docs/DXIL.rst#metadata-resource-records).
I'll go ahead and rename it to "RecordID" to make this a bit clearer.
https://github.com/llvm/llvm-project/pull/100699
___
llvm-branch-commits mailing list
llvm-branch-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-branch-commits
[llvm-branch-commits] [DXIL][Analysis] Implement enough of DXILResourceAnalysis for buffers (PR #100699)
@@ -331,6 +336,249 @@ std::pair
ResourceInfo::getAnnotateProps() const {
return {Word0, Word1};
}
+void ResourceInfo::print(raw_ostream &OS) const {
+ OS << " Symbol: ";
+ Symbol->printAsOperand(OS);
+ OS << "\n";
+
+ OS << " Name: \"" << Name << "\"\n"
+ << " Binding:\n"
+ << "Unique ID: " << Binding.UniqueID << "\n"
+ << "Space: " << Binding.Space << "\n"
+ << "Lower Bound: " << Binding.LowerBound << "\n"
+ << "Size: " << Binding.Size << "\n"
+ << " Class: " << static_cast(RC) << "\n"
+ << " Kind: " << static_cast(Kind) << "\n";
+
+ if (isCBuffer()) {
+OS << " CBuffer size: " << CBufferSize << "\n";
+ } else if (isSampler()) {
+OS << " Sampler Type: " << static_cast(SamplerTy) << "\n";
+ } else {
+if (isUAV()) {
+ OS << " Globally Coherent: " << UAVFlags.GloballyCoherent << "\n"
+ << " HasCounter: " << UAVFlags.HasCounter << "\n"
+ << " IsROV: " << UAVFlags.IsROV << "\n";
+}
+if (isMultiSample())
+ OS << " Sample Count: " << MultiSample.Count << "\n";
+
+if (isStruct()) {
+ OS << " Buffer Stride: " << Struct.Stride << "\n";
+ uint32_t AlignLog2 = Struct.Alignment ? Log2(*Struct.Alignment) : 0;
+ OS << " Alignment: " << AlignLog2 << "\n";
+} else if (isTyped()) {
+ OS << " Element Type: " << static_cast(Typed.ElementTy) <<
"\n"
+ << " Element Count: " << static_cast(Typed.ElementCount)
+ << "\n";
+} else if (isFeedback())
+ OS << " Feedback Type: " << static_cast(Feedback.Type) <<
"\n";
+ }
+}
+
+//===--===//
+// ResourceMapper
+
+static dxil::ElementType toDXILElementType(Type *Ty, bool IsSigned) {
+ // TODO: Handle unorm, snorm, and packed.
+ Ty = Ty->getScalarType();
+
+ if (Ty->isIntegerTy()) {
+switch (Ty->getIntegerBitWidth()) {
+case 16:
+ return IsSigned ? ElementType::I16 : ElementType::U16;
+case 32:
+ return IsSigned ? ElementType::I32 : ElementType::U32;
+case 64:
+ return IsSigned ? ElementType::I64 : ElementType::U64;
+case 1:
+default:
+ return ElementType::Invalid;
+}
+ } else if (Ty->isFloatTy()) {
+return ElementType::F32;
+ } else if (Ty->isDoubleTy()) {
+return ElementType::F64;
+ } else if (Ty->isHalfTy()) {
+return ElementType::F16;
+ }
+
+ return ElementType::Invalid;
+}
+
+namespace {
+
+class ResourceMapper {
+ Module &M;
+ LLVMContext &Context;
+ DXILResourceMap &Resources;
+
+ // Unique ID is per resource type to match DXC.
+ uint32_t NextUAV = 0;
+ uint32_t NextSRV = 0;
+ uint32_t NextCBuf = 0;
+ uint32_t NextSmp = 0;
+
+public:
+ ResourceMapper(Module &M,
+ MapVector &Resources)
+ : M(M), Context(M.getContext()), Resources(Resources) {}
+
+ void diagnoseHandle(CallInst *CI, const Twine &Msg,
+ DiagnosticSeverity Severity = DS_Error) {
+std::string S;
+raw_string_ostream SS(S);
+CI->printAsOperand(SS);
+DiagnosticInfoUnsupported Diag(*CI->getFunction(), Msg + ": " + SS.str(),
+ CI->getDebugLoc(), Severity);
+Context.diagnose(Diag);
+ }
+
+ ResourceInfo *mapBufferType(CallInst *CI, TargetExtType *HandleTy,
+ bool IsTyped) {
+if (HandleTy->getNumTypeParameters() != 1 ||
+HandleTy->getNumIntParameters() != (IsTyped ? 3 : 2)) {
+ diagnoseHandle(CI, Twine("Invalid buffer target type"));
+ return nullptr;
+}
+
+Type *ElTy = HandleTy->getTypeParameter(0);
+unsigned IsWriteable = HandleTy->getIntParameter(0);
+unsigned IsROV = HandleTy->getIntParameter(1);
+bool IsSigned = IsTyped && HandleTy->getIntParameter(2);
+
+ResourceClass RC = IsWriteable ? ResourceClass::UAV : ResourceClass::SRV;
+ResourceKind Kind;
+if (IsTyped)
+ Kind = ResourceKind::TypedBuffer;
+else if (ElTy->isIntegerTy(8))
+ Kind = ResourceKind::RawBuffer;
+else
+ Kind = ResourceKind::StructuredBuffer;
+
+// TODO: We need to lower to a typed pointer, can we smuggle the type
+// through?
+Value *Symbol = UndefValue::get(PointerType::getUnqual(Context));
+// TODO: We don't actually keep track of the name right now...
+StringRef Name = "";
+
+auto [It, Success] = Resources.try_emplace(CI, RC, Kind, Symbol, Name);
+assert(Success && "Mapping the same CallInst again?");
+(void)Success;
+// We grab a pointer into the map's storage, which isn't generally safe.
+// Since we're just using this to fill in the info the map won't mutate and
+// the pointer stays valid for as long as we need it to.
+ResourceInfo *RI = &(It->second);
+
+if (RI->isUAV())
+ // TODO: We need analysis for GloballyCoherent and HasCounter
+ RI->setUAV(false, false, IsROV);
+
+if (RI->isTyped()) {
+ dxil::ElementType ET = toDXILElementType(ElTy, IsSigned
[llvm-branch-commits] [DirectX] Simplify tablegen'd OpCode and OpClass enums (PR #101249)
https://github.com/bogner created https://github.com/llvm/llvm-project/pull/101249 None ___ llvm-branch-commits mailing list llvm-branch-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-branch-commits
[llvm-branch-commits] [DirectX] Make DXILOpBuilder's API more useable (PR #101250)
https://github.com/bogner created https://github.com/llvm/llvm-project/pull/101250 This adjusts the DXILOpBuilder API in a couple of ways: 1. Remove the need to call `getOverloadTy` before creating Ops 2. Introduce `tryCreateOp` to parallel `createOp` but propagate errors 3. Introduce specialized createOp methods for each DXIL Op This will simplify usage of the builder in upcoming changes, and also allows us to propagate errors via DiagnosticInfo rather than using fatal errors. ___ llvm-branch-commits mailing list llvm-branch-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-branch-commits
[llvm-branch-commits] [llvm] [DirectX] Make DXILOpBuilder's API more useable (PR #101250)
@@ -151,7 +151,11 @@ DXILOperationDesc::DXILOperationDesc(const Record *R) {
assert(knownType && "Specification of multiple differing overload "
"parameter types not yet supported");
} else {
-OverloadParamIndices.push_back(i);
+// Skip the return value - nothing is overloaded on only return, and it
bogner wrote:
That's a good point. This should still happen to work since we treat "not
overloaded" and "overloaded on return type" the same for the purposes of
overloadKind in tryCreateOp, and I've updated the comment to say so.
The right thing to do, which I think should be done in a follow up, is to
generate a table of the function signatures per opcode, which can then be
queried by a simple switch over those records. I think we'll need that soon any
way so I'll do that shortly.
https://github.com/llvm/llvm-project/pull/101250
___
llvm-branch-commits mailing list
llvm-branch-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-branch-commits
[llvm-branch-commits] [llvm] [DirectX] Make DXILOpBuilder's API more useable (PR #101250)
@@ -151,7 +151,11 @@ DXILOperationDesc::DXILOperationDesc(const Record *R) {
assert(knownType && "Specification of multiple differing overload "
"parameter types not yet supported");
} else {
-OverloadParamIndices.push_back(i);
+// Skip the return value - nothing is overloaded on only return, and it
bogner wrote:
Thinking about this some more I don't think we can get away with this - if an
op is overloaded on return type we simply need to provide the one we want. I
think I'll add a `Type *` argument that's defaulted to null for "infer it" and
return an error if we can't do so.
https://github.com/llvm/llvm-project/pull/101250
___
llvm-branch-commits mailing list
llvm-branch-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-branch-commits
[llvm-branch-commits] [llvm] [DirectX] Make DXILOpBuilder's API more useable (PR #101250)
@@ -151,7 +151,11 @@ DXILOperationDesc::DXILOperationDesc(const Record *R) {
assert(knownType && "Specification of multiple differing overload "
"parameter types not yet supported");
} else {
-OverloadParamIndices.push_back(i);
+// Skip the return value - nothing is overloaded on only return, and it
bogner wrote:
Fixed in latest.
https://github.com/llvm/llvm-project/pull/101250
___
llvm-branch-commits mailing list
llvm-branch-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-branch-commits
[llvm-branch-commits] [llvm] [DirectX] Make DXILOpBuilder's API more useable (PR #101250)
@@ -289,24 +289,25 @@ static ShaderKind
getShaderKindEnum(Triple::EnvironmentType EnvType) {
/// its specification in DXIL.td.
/// \param OverloadTy Return type to be used to construct DXIL function type.
bogner wrote:
Removed the parameters from the comment
https://github.com/llvm/llvm-project/pull/101250
___
llvm-branch-commits mailing list
llvm-branch-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-branch-commits
[llvm-branch-commits] [DirectX] Simplify tablegen'd OpCode and OpClass enums (PR #101249)
https://github.com/bogner updated https://github.com/llvm/llvm-project/pull/101249 ___ llvm-branch-commits mailing list llvm-branch-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-branch-commits
[llvm-branch-commits] [DirectX] Simplify tablegen'd OpCode and OpClass enums (PR #101249)
https://github.com/bogner updated https://github.com/llvm/llvm-project/pull/101249 ___ llvm-branch-commits mailing list llvm-branch-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-branch-commits
[llvm-branch-commits] [llvm] [DirectX] Make DXILOpBuilder's API more useable (PR #101250)
bogner wrote: > I do not see the invocation of any of the newly created functions in `class > DXILOpbuilder` specific to DXIL Ops `create##Name##Op(...)` or > `tryCreate##Name##Op(...)` (for example `createSinSin13()` or > `tryCreateSin13()` when lowering `llvm.sin.*`). Are these in preparation for > upcoming changes? The intention is to make it a little bit cleaner when we do custom lowering, so that we can call `Builder->createCreateHandleFromBindingOp(...)` directly rather than having to do something like `Builder->createOp(DXILOp::CreateHandleFromBinding, ...)`. > It appears that macro defs `DXIL_OPCODE` are generated in `DXILEmitter` with > its definition is in `DXILOpBuilder.h`. Would you consider generating the > `tryCreate(...)` and `create(...)` for each of DXIL > Ops in `DXILEmitter`, instead, to keep the entire function generation > functionality in a single place i.e., `DXILEmitter`? I don't understand the suggestion. If you're saying we should emit the full text of the declarations from DXILEmitter rather than the table of simple macros then I disagree with that approach - generating full fledged C++ snippets from tablegen should be avoided when possible, as it's quite a bit harder to see what's happening when it's done that way. https://github.com/llvm/llvm-project/pull/101250 ___ llvm-branch-commits mailing list llvm-branch-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-branch-commits
[llvm-branch-commits] [DXIL][Analysis] Implement enough of DXILResourceAnalysis for buffers (PR #100699)
https://github.com/bogner updated https://github.com/llvm/llvm-project/pull/100699 ___ llvm-branch-commits mailing list llvm-branch-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-branch-commits
[llvm-branch-commits] [DXIL][Analysis] Implement enough of DXILResourceAnalysis for buffers (PR #100699)
https://github.com/bogner updated https://github.com/llvm/llvm-project/pull/100699 ___ llvm-branch-commits mailing list llvm-branch-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-branch-commits
[llvm-branch-commits] [DXIL][Analysis] Implement enough of DXILResourceAnalysis for buffers (PR #100699)
@@ -331,6 +336,249 @@ std::pair
ResourceInfo::getAnnotateProps() const {
return {Word0, Word1};
}
+void ResourceInfo::print(raw_ostream &OS) const {
+ OS << " Symbol: ";
+ Symbol->printAsOperand(OS);
+ OS << "\n";
+
+ OS << " Name: \"" << Name << "\"\n"
+ << " Binding:\n"
+ << "Unique ID: " << Binding.UniqueID << "\n"
+ << "Space: " << Binding.Space << "\n"
+ << "Lower Bound: " << Binding.LowerBound << "\n"
+ << "Size: " << Binding.Size << "\n"
+ << " Class: " << static_cast(RC) << "\n"
+ << " Kind: " << static_cast(Kind) << "\n";
bogner wrote:
I've done this in the latest. Note that I put the functions to translate to
names as statics in this file rather than somewhere that would be accessible to
others. There's an argument that we might want to move some or all of that into
DXILABI.h or something, but given that this (and the old Resource analysis its
replacing) are the only places that currently use the strings this seemed like
the best choice for now.
https://github.com/llvm/llvm-project/pull/100699
___
llvm-branch-commits mailing list
llvm-branch-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-branch-commits
[llvm-branch-commits] [DXIL][Analysis] Boilerplate for DXILResourceAnalysis pass (PR #100700)
https://github.com/bogner updated https://github.com/llvm/llvm-project/pull/100700 ___ llvm-branch-commits mailing list llvm-branch-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-branch-commits
[llvm-branch-commits] [DXIL][Analysis] Boilerplate for DXILResourceAnalysis pass (PR #100700)
https://github.com/bogner updated https://github.com/llvm/llvm-project/pull/100700 ___ llvm-branch-commits mailing list llvm-branch-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-branch-commits
[llvm-branch-commits] [DXIL][Analysis] Implement enough of DXILResourceAnalysis for buffers (PR #100699)
https://github.com/bogner updated https://github.com/llvm/llvm-project/pull/100699 ___ llvm-branch-commits mailing list llvm-branch-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-branch-commits
[llvm-branch-commits] [DXIL][Analysis] Implement enough of DXILResourceAnalysis for buffers (PR #100699)
https://github.com/bogner updated https://github.com/llvm/llvm-project/pull/100699 ___ llvm-branch-commits mailing list llvm-branch-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-branch-commits
[llvm-branch-commits] [DirectX] Remove new-pm versions of DXILResource passes. NFC (PR #100698)
https://github.com/bogner closed https://github.com/llvm/llvm-project/pull/100698 ___ llvm-branch-commits mailing list llvm-branch-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-branch-commits
[llvm-branch-commits] [DirectX] Remove new-pm versions of DXILResource passes. NFC (PR #100698)
bogner wrote: Turns out I was wrong and this *does not* effectively avoid the naming collisions. Tried again here: #101393 https://github.com/llvm/llvm-project/pull/100698 ___ llvm-branch-commits mailing list llvm-branch-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-branch-commits
[llvm-branch-commits] [DXIL][Analysis] Implement enough of DXILResourceAnalysis for buffers (PR #100699)
@@ -331,6 +336,249 @@ std::pair
ResourceInfo::getAnnotateProps() const {
return {Word0, Word1};
}
+void ResourceInfo::print(raw_ostream &OS) const {
+ OS << " Symbol: ";
+ Symbol->printAsOperand(OS);
+ OS << "\n";
+
+ OS << " Name: \"" << Name << "\"\n"
+ << " Binding:\n"
+ << "Unique ID: " << Binding.UniqueID << "\n"
+ << "Space: " << Binding.Space << "\n"
+ << "Lower Bound: " << Binding.LowerBound << "\n"
+ << "Size: " << Binding.Size << "\n"
+ << " Class: " << static_cast(RC) << "\n"
+ << " Kind: " << static_cast(Kind) << "\n";
+
+ if (isCBuffer()) {
+OS << " CBuffer size: " << CBufferSize << "\n";
+ } else if (isSampler()) {
+OS << " Sampler Type: " << static_cast(SamplerTy) << "\n";
+ } else {
+if (isUAV()) {
+ OS << " Globally Coherent: " << UAVFlags.GloballyCoherent << "\n"
+ << " HasCounter: " << UAVFlags.HasCounter << "\n"
+ << " IsROV: " << UAVFlags.IsROV << "\n";
+}
+if (isMultiSample())
+ OS << " Sample Count: " << MultiSample.Count << "\n";
+
+if (isStruct()) {
+ OS << " Buffer Stride: " << Struct.Stride << "\n";
+ uint32_t AlignLog2 = Struct.Alignment ? Log2(*Struct.Alignment) : 0;
+ OS << " Alignment: " << AlignLog2 << "\n";
+} else if (isTyped()) {
+ OS << " Element Type: " << static_cast(Typed.ElementTy) <<
"\n"
+ << " Element Count: " << static_cast(Typed.ElementCount)
+ << "\n";
+} else if (isFeedback())
+ OS << " Feedback Type: " << static_cast(Feedback.Type) <<
"\n";
+ }
+}
+
+//===--===//
+// ResourceMapper
+
+static dxil::ElementType toDXILElementType(Type *Ty, bool IsSigned) {
+ // TODO: Handle unorm, snorm, and packed.
+ Ty = Ty->getScalarType();
+
+ if (Ty->isIntegerTy()) {
+switch (Ty->getIntegerBitWidth()) {
+case 16:
+ return IsSigned ? ElementType::I16 : ElementType::U16;
+case 32:
+ return IsSigned ? ElementType::I32 : ElementType::U32;
+case 64:
+ return IsSigned ? ElementType::I64 : ElementType::U64;
+case 1:
+default:
+ return ElementType::Invalid;
+}
+ } else if (Ty->isFloatTy()) {
+return ElementType::F32;
+ } else if (Ty->isDoubleTy()) {
+return ElementType::F64;
+ } else if (Ty->isHalfTy()) {
+return ElementType::F16;
+ }
+
+ return ElementType::Invalid;
+}
+
+namespace {
+
+class ResourceMapper {
+ Module &M;
+ LLVMContext &Context;
+ DXILResourceMap &Resources;
+
+ // Unique ID is per resource type to match DXC.
+ uint32_t NextUAV = 0;
+ uint32_t NextSRV = 0;
+ uint32_t NextCBuf = 0;
+ uint32_t NextSmp = 0;
+
+public:
+ ResourceMapper(Module &M,
+ MapVector &Resources)
+ : M(M), Context(M.getContext()), Resources(Resources) {}
+
+ void diagnoseHandle(CallInst *CI, const Twine &Msg,
+ DiagnosticSeverity Severity = DS_Error) {
+std::string S;
+raw_string_ostream SS(S);
+CI->printAsOperand(SS);
+DiagnosticInfoUnsupported Diag(*CI->getFunction(), Msg + ": " + SS.str(),
+ CI->getDebugLoc(), Severity);
+Context.diagnose(Diag);
+ }
+
+ ResourceInfo *mapBufferType(CallInst *CI, TargetExtType *HandleTy,
+ bool IsTyped) {
+if (HandleTy->getNumTypeParameters() != 1 ||
+HandleTy->getNumIntParameters() != (IsTyped ? 3 : 2)) {
+ diagnoseHandle(CI, Twine("Invalid buffer target type"));
+ return nullptr;
+}
+
+Type *ElTy = HandleTy->getTypeParameter(0);
+unsigned IsWriteable = HandleTy->getIntParameter(0);
+unsigned IsROV = HandleTy->getIntParameter(1);
+bool IsSigned = IsTyped && HandleTy->getIntParameter(2);
+
+ResourceClass RC = IsWriteable ? ResourceClass::UAV : ResourceClass::SRV;
+ResourceKind Kind;
+if (IsTyped)
+ Kind = ResourceKind::TypedBuffer;
+else if (ElTy->isIntegerTy(8))
+ Kind = ResourceKind::RawBuffer;
+else
+ Kind = ResourceKind::StructuredBuffer;
+
+// TODO: We need to lower to a typed pointer, can we smuggle the type
+// through?
+Value *Symbol = UndefValue::get(PointerType::getUnqual(Context));
+// TODO: We don't actually keep track of the name right now...
+StringRef Name = "";
+
+auto [It, Success] = Resources.try_emplace(CI, RC, Kind, Symbol, Name);
+assert(Success && "Mapping the same CallInst again?");
+(void)Success;
+// We grab a pointer into the map's storage, which isn't generally safe.
+// Since we're just using this to fill in the info the map won't mutate and
+// the pointer stays valid for as long as we need it to.
+ResourceInfo *RI = &(It->second);
+
+if (RI->isUAV())
+ // TODO: We need analysis for GloballyCoherent and HasCounter
+ RI->setUAV(false, false, IsROV);
+
+if (RI->isTyped()) {
+ dxil::ElementType ET = toDXILElementType(ElTy, IsSigned
[llvm-branch-commits] [DXIL][Analysis] Implement enough of DXILResourceAnalysis for buffers (PR #100699)
https://github.com/bogner edited https://github.com/llvm/llvm-project/pull/100699 ___ llvm-branch-commits mailing list llvm-branch-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-branch-commits
[llvm-branch-commits] [DXIL][Analysis] Implement enough of DXILResourceAnalysis for buffers (PR #100699)
@@ -331,6 +336,249 @@ std::pair
ResourceInfo::getAnnotateProps() const {
return {Word0, Word1};
}
+void ResourceInfo::print(raw_ostream &OS) const {
+ OS << " Symbol: ";
+ Symbol->printAsOperand(OS);
+ OS << "\n";
+
+ OS << " Name: \"" << Name << "\"\n"
+ << " Binding:\n"
+ << "Unique ID: " << Binding.UniqueID << "\n"
+ << "Space: " << Binding.Space << "\n"
+ << "Lower Bound: " << Binding.LowerBound << "\n"
+ << "Size: " << Binding.Size << "\n"
+ << " Class: " << static_cast(RC) << "\n"
+ << " Kind: " << static_cast(Kind) << "\n";
+
+ if (isCBuffer()) {
+OS << " CBuffer size: " << CBufferSize << "\n";
+ } else if (isSampler()) {
+OS << " Sampler Type: " << static_cast(SamplerTy) << "\n";
+ } else {
+if (isUAV()) {
+ OS << " Globally Coherent: " << UAVFlags.GloballyCoherent << "\n"
+ << " HasCounter: " << UAVFlags.HasCounter << "\n"
+ << " IsROV: " << UAVFlags.IsROV << "\n";
+}
+if (isMultiSample())
+ OS << " Sample Count: " << MultiSample.Count << "\n";
+
+if (isStruct()) {
+ OS << " Buffer Stride: " << Struct.Stride << "\n";
+ uint32_t AlignLog2 = Struct.Alignment ? Log2(*Struct.Alignment) : 0;
+ OS << " Alignment: " << AlignLog2 << "\n";
+} else if (isTyped()) {
+ OS << " Element Type: " << static_cast(Typed.ElementTy) <<
"\n"
+ << " Element Count: " << static_cast(Typed.ElementCount)
+ << "\n";
+} else if (isFeedback())
+ OS << " Feedback Type: " << static_cast(Feedback.Type) <<
"\n";
+ }
+}
+
+//===--===//
+// ResourceMapper
+
+static dxil::ElementType toDXILElementType(Type *Ty, bool IsSigned) {
+ // TODO: Handle unorm, snorm, and packed.
+ Ty = Ty->getScalarType();
+
+ if (Ty->isIntegerTy()) {
+switch (Ty->getIntegerBitWidth()) {
+case 16:
+ return IsSigned ? ElementType::I16 : ElementType::U16;
+case 32:
+ return IsSigned ? ElementType::I32 : ElementType::U32;
+case 64:
+ return IsSigned ? ElementType::I64 : ElementType::U64;
+case 1:
+default:
+ return ElementType::Invalid;
+}
+ } else if (Ty->isFloatTy()) {
+return ElementType::F32;
+ } else if (Ty->isDoubleTy()) {
+return ElementType::F64;
+ } else if (Ty->isHalfTy()) {
+return ElementType::F16;
+ }
+
+ return ElementType::Invalid;
+}
+
+namespace {
+
+class ResourceMapper {
+ Module &M;
+ LLVMContext &Context;
+ DXILResourceMap &Resources;
+
+ // Unique ID is per resource type to match DXC.
+ uint32_t NextUAV = 0;
+ uint32_t NextSRV = 0;
+ uint32_t NextCBuf = 0;
+ uint32_t NextSmp = 0;
+
+public:
+ ResourceMapper(Module &M,
+ MapVector &Resources)
+ : M(M), Context(M.getContext()), Resources(Resources) {}
+
+ void diagnoseHandle(CallInst *CI, const Twine &Msg,
+ DiagnosticSeverity Severity = DS_Error) {
+std::string S;
+raw_string_ostream SS(S);
+CI->printAsOperand(SS);
+DiagnosticInfoUnsupported Diag(*CI->getFunction(), Msg + ": " + SS.str(),
+ CI->getDebugLoc(), Severity);
+Context.diagnose(Diag);
+ }
+
+ ResourceInfo *mapBufferType(CallInst *CI, TargetExtType *HandleTy,
+ bool IsTyped) {
+if (HandleTy->getNumTypeParameters() != 1 ||
+HandleTy->getNumIntParameters() != (IsTyped ? 3 : 2)) {
+ diagnoseHandle(CI, Twine("Invalid buffer target type"));
+ return nullptr;
+}
+
+Type *ElTy = HandleTy->getTypeParameter(0);
+unsigned IsWriteable = HandleTy->getIntParameter(0);
+unsigned IsROV = HandleTy->getIntParameter(1);
+bool IsSigned = IsTyped && HandleTy->getIntParameter(2);
+
+ResourceClass RC = IsWriteable ? ResourceClass::UAV : ResourceClass::SRV;
+ResourceKind Kind;
+if (IsTyped)
+ Kind = ResourceKind::TypedBuffer;
+else if (ElTy->isIntegerTy(8))
+ Kind = ResourceKind::RawBuffer;
+else
+ Kind = ResourceKind::StructuredBuffer;
+
+// TODO: We need to lower to a typed pointer, can we smuggle the type
+// through?
+Value *Symbol = UndefValue::get(PointerType::getUnqual(Context));
+// TODO: We don't actually keep track of the name right now...
+StringRef Name = "";
+
+auto [It, Success] = Resources.try_emplace(CI, RC, Kind, Symbol, Name);
+assert(Success && "Mapping the same CallInst again?");
+(void)Success;
+// We grab a pointer into the map's storage, which isn't generally safe.
+// Since we're just using this to fill in the info the map won't mutate and
+// the pointer stays valid for as long as we need it to.
+ResourceInfo *RI = &(It->second);
+
+if (RI->isUAV())
+ // TODO: We need analysis for GloballyCoherent and HasCounter
+ RI->setUAV(false, false, IsROV);
+
+if (RI->isTyped()) {
+ dxil::ElementType ET = toDXILElementType(ElTy, IsSigned
[llvm-branch-commits] [DXIL][Analysis] Implement enough of DXILResourceAnalysis for buffers (PR #100699)
https://github.com/bogner edited https://github.com/llvm/llvm-project/pull/100699 ___ llvm-branch-commits mailing list llvm-branch-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-branch-commits
[llvm-branch-commits] [DXIL][Analysis] Implement enough of DXILResourceAnalysis for buffers (PR #100699)
https://github.com/bogner updated https://github.com/llvm/llvm-project/pull/100699 ___ llvm-branch-commits mailing list llvm-branch-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-branch-commits
[llvm-branch-commits] [DXIL][Analysis] Implement enough of DXILResourceAnalysis for buffers (PR #100699)
https://github.com/bogner updated https://github.com/llvm/llvm-project/pull/100699 ___ llvm-branch-commits mailing list llvm-branch-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-branch-commits
[llvm-branch-commits] [DXIL][Analysis] Implement enough of DXILResourceAnalysis for buffers (PR #100699)
https://github.com/bogner updated https://github.com/llvm/llvm-project/pull/100699 ___ llvm-branch-commits mailing list llvm-branch-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-branch-commits
[llvm-branch-commits] [DXIL][Analysis] Implement enough of DXILResourceAnalysis for buffers (PR #100699)
https://github.com/bogner updated https://github.com/llvm/llvm-project/pull/100699 ___ llvm-branch-commits mailing list llvm-branch-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-branch-commits
[llvm-branch-commits] [DXIL][Analysis] Boilerplate for DXILResourceAnalysis pass (PR #100700)
https://github.com/bogner updated https://github.com/llvm/llvm-project/pull/100700 ___ llvm-branch-commits mailing list llvm-branch-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-branch-commits
[llvm-branch-commits] [DXIL][Analysis] Boilerplate for DXILResourceAnalysis pass (PR #100700)
https://github.com/bogner updated https://github.com/llvm/llvm-project/pull/100700 ___ llvm-branch-commits mailing list llvm-branch-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-branch-commits
[llvm-branch-commits] [DXIL][Analysis] Implement enough of DXILResourceAnalysis for buffers (PR #100699)
https://github.com/bogner updated https://github.com/llvm/llvm-project/pull/100699 ___ llvm-branch-commits mailing list llvm-branch-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-branch-commits
[llvm-branch-commits] [DXIL][Analysis] Implement enough of DXILResourceAnalysis for buffers (PR #100699)
https://github.com/bogner updated https://github.com/llvm/llvm-project/pull/100699 ___ llvm-branch-commits mailing list llvm-branch-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-branch-commits
[llvm-branch-commits] [DirectX] Revert specialized createOp methods part of #101250 (PR #104245)
https://github.com/bogner created https://github.com/llvm/llvm-project/pull/104245 In 8cf85653b6f5 "[DirectX] Make DXILOpBuilder's API more useable" we introduced specialized createOp methods for each DirectX op for convenience, but the API is buggy and untested. It also isn't actually as useful as I imagined it would be since we don't have argument names or const-ness represented in DXIL.td currently. Remove these methods for now and we can reintroduce them if we actually need them later. ___ llvm-branch-commits mailing list llvm-branch-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-branch-commits
[llvm-branch-commits] [DirectX] Disentangle DXIL.td's op types from LLVMType. NFC (PR #104247)
https://github.com/bogner created https://github.com/llvm/llvm-project/pull/104247 LLVMType is both too broad and too narrow for defining DXIL operations, in different ways. It's too broad in the sense that we don't need the full set of MVTs - the set of types DXIL operations work on is much smaller. It's too narrow in the sense that it's difficult to use it for the various fixed structure types in DXIL, like `%dx.types.Handle` or `%dx.Types.ResRet.f32`. Replace the usage of LLVMType in DXIL.td with DXILOpParamType, a simple class that we can define an enum of types from. Further, use this to replace the "ParameterKind" enum in DXILABI.h that has nothing to do with DXIL's ABI. ___ llvm-branch-commits mailing list llvm-branch-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-branch-commits
[llvm-branch-commits] [DirectX] Differentiate between 0/1 overloads in the OpBuilder. NFC (PR #104246)
https://github.com/bogner created https://github.com/llvm/llvm-project/pull/104246 DXIL operations that only have one signature behave one of two ways - either they are always suffixed with a type like `dx.op.ThreadId.i32` and hence have exactly one overload, or they're never suffixed like `dx.op.CreateHandle` and hence have zero overloads. Update DXIL.td for operations that have one overload and remove the hack in the builder that was adjusting names for unoverloaded ops. ___ llvm-branch-commits mailing list llvm-branch-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-branch-commits
[llvm-branch-commits] [DirectX] Encapsulate DXILOpLowering's state into a class. NFC (PR #104248)
https://github.com/bogner created https://github.com/llvm/llvm-project/pull/104248 This introduces an anonymous class "OpLowerer" to help with lowering DXIL ops, and moves the DXILOpBuilder there instead of creating a new one for every operation. DXILOpBuilder is also changed to own its IRBuilder, since that makes it simpler to ensure that it isn't misused. ___ llvm-branch-commits mailing list llvm-branch-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-branch-commits
[llvm-branch-commits] [DirectX] Use a more consistent pass name for DXILTranslateMetadata (PR #104249)
https://github.com/bogner created https://github.com/llvm/llvm-project/pull/104249 This updates the "dxil-metadata-emit" pass flag to be spelled "dxil-translate-metadata" to better match the pass name. ___ llvm-branch-commits mailing list llvm-branch-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-branch-commits
[llvm-branch-commits] [DirectX] Register a few DXIL passes with the new PM (PR #104250)
https://github.com/bogner created https://github.com/llvm/llvm-project/pull/104250 This wires up dxil-op-lower, dxil-intrinsic-expansion, dxil-translate-metadata, and dxil-pretty-printer to the new pass manager, both as a matter of future proofing the backend and so that they can be used more flexibly in tests. A few arbitrary tests are updated in order to test the new PM path, and we drop the "print-dxil-resource-md" pass since it's redundant with the pretty printer. ___ llvm-branch-commits mailing list llvm-branch-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-branch-commits
[llvm-branch-commits] [DirectX] Lower `@llvm.dx.handle.fromBinding` to DXIL ops (PR #104251)
https://github.com/bogner created https://github.com/llvm/llvm-project/pull/104251 The `@llvm.dx.handle.fromBinding` intrinsic is lowered either to the `CreateHandle` op or a pair of `CreateHandleFromBinding` and `AnnotateHandle` ops, depending on the DXIL version. Regardless of the DXIL version we need to emit metadata about the binding, but that's left to a separate change. These DXIL ops all need to return the `%dx.types.Handle` type, but the llvm intrinsic returns a target extension type. To facilitate changing the type of the operation and all of its users, we introduce `%llvm.dx.cast.handle`, which can cast between the two handle representations. ___ llvm-branch-commits mailing list llvm-branch-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-branch-commits
[llvm-branch-commits] [DirectX] Lower `@llvm.dx.typedBufferLoad` to DXIL ops (PR #104252)
https://github.com/bogner created https://github.com/llvm/llvm-project/pull/104252 The `@llvm.dx.typedBufferLoad` intrinsic is lowered to `@dx.op.bufferLoad`. There's some complexity here due to translating from a vector return type to a named struct and trying to avoid excessive IR coming out of that. Note that this change includes a bit of a hack in how it deals with `getOverloadKind` for the `dx.ResRet` types - we need to adjust how we deal with operation overloads to generate a table directly rather than proxy through the OverloadKind enum, but that's left for a later change here. ___ llvm-branch-commits mailing list llvm-branch-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-branch-commits
[llvm-branch-commits] [DirectX] Lower `@llvm.dx.typedBufferStore` to DXIL ops (PR #104253)
https://github.com/bogner created https://github.com/llvm/llvm-project/pull/104253 The `@llvm.dx.typedBufferStore` intrinsic is lowered to `@dx.op.bufferStore`. ___ llvm-branch-commits mailing list llvm-branch-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-branch-commits
[llvm-branch-commits] [DirectX] Register a few DXIL passes with the new PM (PR #104250)
bogner wrote: Depends on #104248 https://github.com/llvm/llvm-project/pull/104250 ___ llvm-branch-commits mailing list llvm-branch-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-branch-commits
[llvm-branch-commits] [DirectX] Lower `@llvm.dx.handle.fromBinding` to DXIL ops (PR #104251)
bogner wrote: Depends on #104246, #104247, and #104248 https://github.com/llvm/llvm-project/pull/104251 ___ llvm-branch-commits mailing list llvm-branch-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-branch-commits
[llvm-branch-commits] [DirectX] Lower `@llvm.dx.typedBufferLoad` to DXIL ops (PR #104252)
bogner wrote: Depends on #104251 https://github.com/llvm/llvm-project/pull/104252 ___ llvm-branch-commits mailing list llvm-branch-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-branch-commits
[llvm-branch-commits] [DirectX] Lower `@llvm.dx.typedBufferStore` to DXIL ops (PR #104253)
bogner wrote: Depends on #104252 https://github.com/llvm/llvm-project/pull/104253 ___ llvm-branch-commits mailing list llvm-branch-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-branch-commits
[llvm-branch-commits] [DirectX] Disentangle DXIL.td's op types from LLVMType. NFC (PR #104247)
https://github.com/bogner edited https://github.com/llvm/llvm-project/pull/104247 ___ llvm-branch-commits mailing list llvm-branch-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-branch-commits
[llvm-branch-commits] [DirectX] Lower `@llvm.dx.handle.fromBinding` to DXIL ops (PR #104251)
https://github.com/bogner edited https://github.com/llvm/llvm-project/pull/104251 ___ llvm-branch-commits mailing list llvm-branch-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-branch-commits
[llvm-branch-commits] [DirectX] Lower `@llvm.dx.typedBufferLoad` to DXIL ops (PR #104252)
https://github.com/bogner edited https://github.com/llvm/llvm-project/pull/104252 ___ llvm-branch-commits mailing list llvm-branch-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-branch-commits
[llvm-branch-commits] [DirectX] Lower `@llvm.dx.typedBufferStore` to DXIL ops (PR #104253)
https://github.com/bogner edited https://github.com/llvm/llvm-project/pull/104253 ___ llvm-branch-commits mailing list llvm-branch-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-branch-commits
[llvm-branch-commits] [DirectX] Register a few DXIL passes with the new PM (PR #104250)
@@ -23,7 +23,10 @@ MODULE_ANALYSIS("dxil-resource-md", DXILResourceMDAnalysis())
#ifndef MODULE_PASS
#define MODULE_PASS(NAME, CREATE_PASS)
#endif
+MODULE_PASS("dxil-intrinsic-expansion", DXILIntrinsicExpansion())
+MODULE_PASS("dxil-op-lower", DXILOpLowering())
+MODULE_PASS("dxil-pretty-printer", DXILPrettyPrinterPass(dbgs()))
+MODULE_PASS("dxil-translate-metadata", DXILTranslateMetadata())
// TODO: rename to print after NPM switch
MODULE_PASS("print-dx-shader-flags", dxil::ShaderFlagsAnalysisPrinter(dbgs()))
-MODULE_PASS("print-dxil-resource-md", DXILResourceMDPrinterPass(dbgs()))
bogner wrote:
Yes we should. I currently have that as part of this change that isn't quite
ready for review yet:
https://github.com/bogner/llvm-project/commit/bed151261c5b26519f6e44c44631cc1f770dfd78
Maybe I should pull the deletion into this PR instead?
https://github.com/llvm/llvm-project/pull/104250
___
llvm-branch-commits mailing list
llvm-branch-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-branch-commits
[llvm-branch-commits] [DirectX] Lower `@llvm.dx.handle.fromBinding` to DXIL ops (PR #104251)
@@ -0,0 +1,63 @@
+; RUN: opt -S -dxil-op-lower %s | FileCheck %s
+
+target triple = "dxil-pc-shadermodel6.6-compute"
+
+define void @test_bindings() {
+ ; RWBuffer Buf : register(u5, space3)
+ %typed0 = call target("dx.TypedBuffer", <4 x float>, 1, 0, 0)
+ @llvm.dx.handle.fromBinding.tdx.TypedBuffer_v4f32_1_0_0(
+ i32 3, i32 5, i32 1, i32 4, i1 false)
+ ; CHECK: [[BUF0:%[0-9]*]] = call %dx.types.Handle
@dx.op.createHandleFromBinding(i32 218, %dx.types.ResBind { i32 5, i32 5, i32
3, i8 1 }, i32 4, i1 false)
+ ; CHECK: call %dx.types.Handle @dx.op.annotateHandle(i32 217,
%dx.types.Handle [[BUF0]], %dx.types.ResourceProperties { i32 4106, i32 1033 })
+
+ ; RWBuffer Buf : register(u7, space2)
+ %typed1 = call target("dx.TypedBuffer", i32, 1, 0, 1)
+ @llvm.dx.handle.fromBinding.tdx.TypedBuffer_i32_1_0_0t(
+ i32 2, i32 7, i32 1, i32 6, i1 false)
+ ; CHECK: [[BUF1:%[0-9]*]] = call %dx.types.Handle
@dx.op.createHandleFromBinding(i32 218, %dx.types.ResBind { i32 7, i32 7, i32
2, i8 1 }, i32 6, i1 false)
+ ; CHECK: call %dx.types.Handle @dx.op.annotateHandle(i32 217,
%dx.types.Handle [[BUF1]], %dx.types.ResourceProperties { i32 4106, i32 260 })
+
+ ; Buffer Buf[24] : register(t3, space5)
+ %typed2 = call target("dx.TypedBuffer", <4 x i32>, 0, 0, 0)
bogner wrote:
You can see it in the `createHandleFromBinding` call below, though it looks
more like math there:
```
%dx.types.ResBind { i32 7, i32 30, i32 2, i8 0 }
```
Here we have a resource binding with a lower bound of 7 and an upper bound of
30 - ie, a size of 24.
https://github.com/llvm/llvm-project/pull/104251
___
llvm-branch-commits mailing list
llvm-branch-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-branch-commits
[llvm-branch-commits] [llvm] [DirectX] Register a few DXIL passes with the new PM (PR #104250)
@@ -23,7 +23,10 @@ MODULE_ANALYSIS("dxil-resource-md", DXILResourceMDAnalysis())
#ifndef MODULE_PASS
#define MODULE_PASS(NAME, CREATE_PASS)
#endif
+MODULE_PASS("dxil-intrinsic-expansion", DXILIntrinsicExpansion())
+MODULE_PASS("dxil-op-lower", DXILOpLowering())
+MODULE_PASS("dxil-pretty-printer", DXILPrettyPrinterPass(dbgs()))
+MODULE_PASS("dxil-translate-metadata", DXILTranslateMetadata())
// TODO: rename to print after NPM switch
MODULE_PASS("print-dx-shader-flags", dxil::ShaderFlagsAnalysisPrinter(dbgs()))
-MODULE_PASS("print-dxil-resource-md", DXILResourceMDPrinterPass(dbgs()))
bogner wrote:
I've pulled the removal of the pass into this change.
https://github.com/llvm/llvm-project/pull/104250
___
llvm-branch-commits mailing list
llvm-branch-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-branch-commits
[llvm-branch-commits] [DirectX] Lower `@llvm.dx.handle.fromBinding` to DXIL ops (PR #104251)
https://github.com/bogner edited https://github.com/llvm/llvm-project/pull/104251 ___ llvm-branch-commits mailing list llvm-branch-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-branch-commits
[llvm-branch-commits] [DirectX] Lower `@llvm.dx.typedBufferStore` to DXIL ops (PR #104253)
@@ -289,6 +289,43 @@ class OpLowerer {
});
}
+ void lowerTypedBufferStore(Function &F) {
+IRBuilder<> &IRB = OpBuilder.getIRB();
+Type *Int8Ty = IRB.getInt8Ty();
+Type *Int32Ty = IRB.getInt32Ty();
+
+replaceFunction(F, [&](CallInst *CI) -> Error {
bogner wrote:
It doesn't *need* to be - it just seemed more ergonomic than coming up with a
name for it and spelling it elsewhere. Do you have a different suggestion?
https://github.com/llvm/llvm-project/pull/104253
___
llvm-branch-commits mailing list
llvm-branch-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-branch-commits
[llvm-branch-commits] [DirectX] Disentangle DXIL.td's op types from LLVMType. NFC (PR #104247)
https://github.com/bogner updated https://github.com/llvm/llvm-project/pull/104247 ___ llvm-branch-commits mailing list llvm-branch-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-branch-commits
[llvm-branch-commits] [DirectX] Disentangle DXIL.td's op types from LLVMType. NFC (PR #104247)
https://github.com/bogner edited https://github.com/llvm/llvm-project/pull/104247 ___ llvm-branch-commits mailing list llvm-branch-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-branch-commits
[llvm-branch-commits] [DirectX] Disentangle DXIL.td's op types from LLVMType. NFC (PR #104247)
https://github.com/bogner updated https://github.com/llvm/llvm-project/pull/104247 ___ llvm-branch-commits mailing list llvm-branch-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-branch-commits
[llvm-branch-commits] [DirectX] Disentangle DXIL.td's op types from LLVMType. NFC (PR #104247)
bogner wrote: > Your description and the bug you are adressing make sense. One point of > confusion I see now is our mapping of intrinsics to DXIL Ops. The intrinsics > still all use `LLVMType`. Is that going to be problem? I don't see why it would. LLVMType is perfectly reasonable for describing LLVM intrinsics. DXIL ops, on the other hand, use a number of specific `dx.types.*` struct types that aren't easy to describe with LLVMType and it wouldn't make sense to extend it to do so. Also, there are far fewer types that come up with DXIL ops, so the full flexibility of LLVMType isn't needed, hence this simpler approach. https://github.com/llvm/llvm-project/pull/104247 ___ llvm-branch-commits mailing list llvm-branch-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-branch-commits
[llvm-branch-commits] [DirectX] Use a more consistent pass name for DXILTranslateMetadata (PR #104249)
https://github.com/bogner updated https://github.com/llvm/llvm-project/pull/104249 ___ llvm-branch-commits mailing list llvm-branch-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-branch-commits
[llvm-branch-commits] [DirectX] Use a more consistent pass name for DXILTranslateMetadata (PR #104249)
https://github.com/bogner updated https://github.com/llvm/llvm-project/pull/104249 ___ llvm-branch-commits mailing list llvm-branch-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-branch-commits
[llvm-branch-commits] [DirectX] Lower `@llvm.dx.handle.fromBinding` to DXIL ops (PR #104251)
https://github.com/bogner updated https://github.com/llvm/llvm-project/pull/104251 ___ llvm-branch-commits mailing list llvm-branch-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-branch-commits
[llvm-branch-commits] [DirectX] Lower `@llvm.dx.handle.fromBinding` to DXIL ops (PR #104251)
https://github.com/bogner updated https://github.com/llvm/llvm-project/pull/104251 ___ llvm-branch-commits mailing list llvm-branch-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-branch-commits
[llvm-branch-commits] [DirectX] Lower `@llvm.dx.handle.fromBinding` to DXIL ops (PR #104251)
https://github.com/bogner edited https://github.com/llvm/llvm-project/pull/104251 ___ llvm-branch-commits mailing list llvm-branch-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-branch-commits
[llvm-branch-commits] [DirectX] Lower `@llvm.dx.handle.fromBinding` to DXIL ops (PR #104251)
@@ -0,0 +1,63 @@
+; RUN: opt -S -dxil-op-lower %s | FileCheck %s
+
+target triple = "dxil-pc-shadermodel6.6-compute"
+
+define void @test_bindings() {
+ ; RWBuffer Buf : register(u5, space3)
+ %typed0 = call target("dx.TypedBuffer", <4 x float>, 1, 0, 0)
+ @llvm.dx.handle.fromBinding.tdx.TypedBuffer_v4f32_1_0_0(
+ i32 3, i32 5, i32 1, i32 4, i1 false)
+ ; CHECK: [[BUF0:%[0-9]*]] = call %dx.types.Handle
@dx.op.createHandleFromBinding(i32 218, %dx.types.ResBind { i32 5, i32 5, i32
3, i8 1 }, i32 4, i1 false)
+ ; CHECK: call %dx.types.Handle @dx.op.annotateHandle(i32 217,
%dx.types.Handle [[BUF0]], %dx.types.ResourceProperties { i32 4106, i32 1033 })
+
+ ; RWBuffer Buf : register(u7, space2)
+ %typed1 = call target("dx.TypedBuffer", i32, 1, 0, 1)
+ @llvm.dx.handle.fromBinding.tdx.TypedBuffer_i32_1_0_0t(
+ i32 2, i32 7, i32 1, i32 6, i1 false)
+ ; CHECK: [[BUF1:%[0-9]*]] = call %dx.types.Handle
@dx.op.createHandleFromBinding(i32 218, %dx.types.ResBind { i32 7, i32 7, i32
2, i8 1 }, i32 6, i1 false)
+ ; CHECK: call %dx.types.Handle @dx.op.annotateHandle(i32 217,
%dx.types.Handle [[BUF1]], %dx.types.ResourceProperties { i32 4106, i32 260 })
+
+ ; Buffer Buf[24] : register(t3, space5)
+ %typed2 = call target("dx.TypedBuffer", <4 x i32>, 0, 0, 0)
bogner wrote:
It was technically `Buf[1]` as written, as the index is from the beginning of
the binding space (matching DXC) and the test case didn't match the comment.
I've updated the docs slightly to make that clearer and adjusted the test so
the comment and test case agree.
https://github.com/llvm/llvm-project/pull/104251
___
llvm-branch-commits mailing list
llvm-branch-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-branch-commits
[llvm-branch-commits] [DirectX] Lower `@llvm.dx.typedBufferStore` to DXIL ops (PR #104253)
https://github.com/bogner updated https://github.com/llvm/llvm-project/pull/104253 ___ llvm-branch-commits mailing list llvm-branch-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-branch-commits
[llvm-branch-commits] [DirectX] Lower `@llvm.dx.typedBufferStore` to DXIL ops (PR #104253)
https://github.com/bogner updated https://github.com/llvm/llvm-project/pull/104253 ___ llvm-branch-commits mailing list llvm-branch-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-branch-commits
[llvm-branch-commits] [DirectX] Lower `@llvm.dx.typedBufferStore` to DXIL ops (PR #104253)
https://github.com/bogner edited https://github.com/llvm/llvm-project/pull/104253 ___ llvm-branch-commits mailing list llvm-branch-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-branch-commits
[llvm-branch-commits] [DirectX] Lower `@llvm.dx.typedBufferStore` to DXIL ops (PR #104253)
@@ -289,6 +289,43 @@ class OpLowerer {
});
}
+ void lowerTypedBufferStore(Function &F) {
+IRBuilder<> &IRB = OpBuilder.getIRB();
+Type *Int8Ty = IRB.getInt8Ty();
+Type *Int32Ty = IRB.getInt32Ty();
+
+replaceFunction(F, [&](CallInst *CI) -> Error {
+ IRB.SetInsertPoint(CI);
+
+ Value *Handle =
+ createTmpHandleCast(CI->getArgOperand(0), OpBuilder.getHandleType());
+ Value *Index0 = CI->getArgOperand(1);
+ Value *Index1 = UndefValue::get(Int32Ty);
+ // For typed stores, the mask must always cover all four elements.
+ Constant *Mask = ConstantInt::get(Int8Ty, 0xF);
+
+ Value *Data = CI->getArgOperand(2);
bogner wrote:
The argument to `typedBufferStore` must be a vector of 4 elements, as we must
store 16 bytes exactly. The only exception to this is a vector of 2 doubles,
which isn't supported yet (I've filed #104423 for that).
That said, it'd probably be nice to get decent errors here if we get bad
textual IR or a frontend other than clang messes up, so I've added better
errors in the latest.
https://github.com/llvm/llvm-project/pull/104253
___
llvm-branch-commits mailing list
llvm-branch-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-branch-commits
[llvm-branch-commits] [DirectX] Implement metadata lowering for resources (PR #104447)
https://github.com/bogner created https://github.com/llvm/llvm-project/pull/104447 Generate metadata from target extension type based resources. Part of #91366 ___ llvm-branch-commits mailing list llvm-branch-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-branch-commits
[llvm-branch-commits] [DirectX] Move resource logic into PrettyPrinter and TranslateMetadata (PR #104446)
https://github.com/bogner created https://github.com/llvm/llvm-project/pull/104446 Move the module level logic for resources into the pretty printer and translate metadata passes rather than embedding them in the DXILResource helper. This will make it easier to migrate towards the target extension type based approach to resources. ___ llvm-branch-commits mailing list llvm-branch-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-branch-commits
[llvm-branch-commits] [DirectX] Add resource handling to the DXIL pretty printer (PR #104448)
https://github.com/bogner created https://github.com/llvm/llvm-project/pull/104448 Handle target extension type resources when printing resources to textual IR. ___ llvm-branch-commits mailing list llvm-branch-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-branch-commits
[llvm-branch-commits] [DirectX] Move resource logic into PrettyPrinter and TranslateMetadata. NFC (PR #104446)
https://github.com/bogner edited https://github.com/llvm/llvm-project/pull/104446 ___ llvm-branch-commits mailing list llvm-branch-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-branch-commits
[llvm-branch-commits] [DirectX] Implement metadata lowering for resources (PR #104447)
bogner wrote: Depends on #104446 https://github.com/llvm/llvm-project/pull/104447 ___ llvm-branch-commits mailing list llvm-branch-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-branch-commits
[llvm-branch-commits] [DirectX] Add resource handling to the DXIL pretty printer (PR #104448)
bogner wrote: Depends on #104446 https://github.com/llvm/llvm-project/pull/104448 ___ llvm-branch-commits mailing list llvm-branch-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-branch-commits
[llvm-branch-commits] [DirectX] Implement metadata lowering for resources (PR #104447)
@@ -13,27 +13,52 @@
#include "DXILShaderFlags.h"
#include "DirectX.h"
#include "llvm/ADT/StringSet.h"
+#include "llvm/Analysis/DXILResource.h"
#include "llvm/IR/Constants.h"
#include "llvm/IR/Metadata.h"
#include "llvm/IR/Module.h"
+#include "llvm/InitializePasses.h"
#include "llvm/Pass.h"
#include "llvm/TargetParser/Triple.h"
using namespace llvm;
using namespace llvm::dxil;
-static void emitResourceMetadata(Module &M,
+static void emitResourceMetadata(Module &M, const DXILResourceMap &DRM,
const dxil::Resources &MDResources) {
- Metadata *SRVMD = nullptr, *UAVMD = nullptr, *CBufMD = nullptr,
- *SmpMD = nullptr;
- bool HasResources = false;
+ LLVMContext &Context = M.getContext();
+
+ SmallVector SRVs, UAVs, CBufs, Smps;
+ for (auto [_, RI] : DRM) {
+switch (RI.getResourceClass()) {
+case dxil::ResourceClass::SRV:
+ SRVs.push_back(RI.getAsMetadata(Context));
+ break;
+case dxil::ResourceClass::UAV:
+ UAVs.push_back(RI.getAsMetadata(Context));
+ break;
+case dxil::ResourceClass::CBuffer:
+ CBufs.push_back(RI.getAsMetadata(Context));
+ break;
+case dxil::ResourceClass::Sampler:
+ Smps.push_back(RI.getAsMetadata(Context));
+ break;
+}
+ }
+ Metadata *SRVMD = SRVs.empty() ? nullptr : MDNode::get(Context, SRVs);
+ Metadata *UAVMD = UAVs.empty() ? nullptr : MDNode::get(Context, UAVs);
+ Metadata *CBufMD = CBufs.empty() ? nullptr : MDNode::get(Context, CBufs);
+ Metadata *SmpMD = Smps.empty() ? nullptr : MDNode::get(Context, Smps);
+ bool HasResources = !DRM.empty();
if (MDResources.hasUAVs()) {
+assert(!UAVMD && "Old and new UAV representations can't coexist");
bogner wrote:
Yes. Once the frontend starts generating target extension types it will no
longer generate the uav metadata node and we can remove all of that code.
Similarly for the cbufs.
https://github.com/llvm/llvm-project/pull/104447
___
llvm-branch-commits mailing list
llvm-branch-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-branch-commits
[llvm-branch-commits] [DirectX] Disentangle DXIL.td's op types from LLVMType. NFC (PR #104247)
https://github.com/bogner updated https://github.com/llvm/llvm-project/pull/104247 ___ llvm-branch-commits mailing list llvm-branch-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-branch-commits
[llvm-branch-commits] [DirectX] Disentangle DXIL.td's op types from LLVMType. NFC (PR #104247)
https://github.com/bogner updated https://github.com/llvm/llvm-project/pull/104247 ___ llvm-branch-commits mailing list llvm-branch-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-branch-commits
[llvm-branch-commits] [DirectX] Encapsulate DXILOpLowering's state into a class. NFC (PR #104248)
https://github.com/bogner updated https://github.com/llvm/llvm-project/pull/104248 ___ llvm-branch-commits mailing list llvm-branch-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-branch-commits
[llvm-branch-commits] [DirectX] Encapsulate DXILOpLowering's state into a class. NFC (PR #104248)
https://github.com/bogner updated https://github.com/llvm/llvm-project/pull/104248 ___ llvm-branch-commits mailing list llvm-branch-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-branch-commits
[llvm-branch-commits] [DirectX] Lower `@llvm.dx.handle.fromBinding` to DXIL ops (PR #104251)
https://github.com/bogner updated https://github.com/llvm/llvm-project/pull/104251 ___ llvm-branch-commits mailing list llvm-branch-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-branch-commits
[llvm-branch-commits] [DirectX] Lower `@llvm.dx.handle.fromBinding` to DXIL ops (PR #104251)
https://github.com/bogner updated https://github.com/llvm/llvm-project/pull/104251 ___ llvm-branch-commits mailing list llvm-branch-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-branch-commits
[llvm-branch-commits] [DirectX] Lower `@llvm.dx.handle.fromBinding` to DXIL ops (PR #104251)
@@ -119,6 +123,119 @@ class OpLowerer {
});
}
+ Value *createTmpHandleCast(Value *V, Type *Ty) {
+Function *CastFn = Intrinsic::getDeclaration(&M, Intrinsic::dx_cast_handle,
+ {Ty, V->getType()});
+CallInst *Cast = OpBuilder.getIRB().CreateCall(CastFn, {V});
+CleanupCasts.push_back(Cast);
+return Cast;
+ }
+
+ void cleanupHandleCasts() {
+SmallVector ToRemove;
+SmallVector CastFns;
+
+for (CallInst *Cast : CleanupCasts) {
+ CastFns.push_back(Cast->getCalledFunction());
+ // All of the ops should be using `dx.types.Handle` at this point, so if
+ // we're not producing that we should be part of a pair. Track this so we
+ // can remove it at the end.
+ if (Cast->getType() != OpBuilder.getHandleType()) {
+ToRemove.push_back(Cast);
+continue;
+ }
+ // Otherwise, we're the second handle in a pair. Forward the arguments
and
+ // remove the (second) cast.
+ CallInst *Def = cast(Cast->getOperand(0));
+ assert(Def->getIntrinsicID() == Intrinsic::dx_cast_handle &&
+ "Unbalanced pair of temporary handle casts");
+ Cast->replaceAllUsesWith(Def->getOperand(0));
+ Cast->eraseFromParent();
+}
+for (CallInst *Cast : ToRemove) {
+ assert(Cast->user_empty() && "Temporary handle cast still has users");
+ Cast->eraseFromParent();
+}
+llvm::sort(CastFns);
+CastFns.erase(llvm::unique(CastFns), CastFns.end());
+for (Function *F : CastFns)
+ F->eraseFromParent();
bogner wrote:
Above we add the cast function to the list every time we see a call to a cast -
this means we could end up with duplicates if the same cast function is called
multiple times. So, we need to deduplicate the list before we start erasing the
functions from the module.
https://github.com/llvm/llvm-project/pull/104251
___
llvm-branch-commits mailing list
llvm-branch-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-branch-commits
[llvm-branch-commits] [llvm] [DirectX] Register a few DXIL passes with the new PM (PR #104250)
@@ -46,14 +46,14 @@ define void @test_typedbuffer() {
; Buffer Buf[24] : register(t3, space5)
%typed2 = call target("dx.TypedBuffer", <4 x i32>, 0, 0, 0)
@llvm.dx.handle.fromBinding.tdx.TypedBuffer_i32_0_0t(
- i32 2, i32 7, i32 24, i32 0, i1 false)
+ i32 5, i32 3, i32 24, i32 0, i1 false)
bogner wrote:
That was supposed to be included in #100699 in response to [a
comment](https://github.com/llvm/llvm-project/pull/100699#discussion_r1715835841),
but I guess I somehow put it in the wrong change. Committing that separately
here: #105409
https://github.com/llvm/llvm-project/pull/104250
___
llvm-branch-commits mailing list
llvm-branch-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-branch-commits
[llvm-branch-commits] [clang] cab6bcd - [𝘀𝗽𝗿] initial version
Author: Justin Bogner
Date: 2023-11-03T18:59:49-07:00
New Revision: cab6bcd73081fcbe9807adbf60b345f8d9e654e4
URL:
https://github.com/llvm/llvm-project/commit/cab6bcd73081fcbe9807adbf60b345f8d9e654e4
DIFF:
https://github.com/llvm/llvm-project/commit/cab6bcd73081fcbe9807adbf60b345f8d9e654e4.diff
LOG: [𝘀𝗽𝗿] initial version
Created using spr 1.3.5
Added:
Modified:
clang/lib/Sema/SemaDecl.cpp
clang/lib/Sema/SemaDeclCXX.cpp
clang/lib/Sema/SemaInit.cpp
clang/lib/Sema/SemaLambda.cpp
clang/lib/Sema/SemaTemplate.cpp
clang/lib/Sema/SemaType.cpp
clang/test/SemaHLSL/BuiltIns/vector-errors.hlsl
Removed:
diff --git a/clang/lib/Sema/SemaDecl.cpp b/clang/lib/Sema/SemaDecl.cpp
index 396566a8f10a9b7..a86c428c9121abf 100644
--- a/clang/lib/Sema/SemaDecl.cpp
+++ b/clang/lib/Sema/SemaDecl.cpp
@@ -5929,7 +5929,7 @@ Sema::GetNameFromUnqualifiedId(const UnqualifiedId &Name)
{
Diag(Name.StartLocation,
diag::err_deduction_guide_name_not_class_template)
<< (int)getTemplateNameKindForDiagnostics(TN) << TN;
- if (Template)
+ if (Template && Template->getLocation().isValid())
Diag(Template->getLocation(), diag::note_template_decl_here);
return DeclarationNameInfo();
}
diff --git a/clang/lib/Sema/SemaDeclCXX.cpp b/clang/lib/Sema/SemaDeclCXX.cpp
index 397b7a00e453126..de5cc07f5d42865 100644
--- a/clang/lib/Sema/SemaDeclCXX.cpp
+++ b/clang/lib/Sema/SemaDeclCXX.cpp
@@ -11462,7 +11462,8 @@ bool Sema::CheckDeductionGuideDeclarator(Declarator &D,
QualType &R,
GuidedTemplateDecl->getDeclContext()->getRedeclContext())) {
Diag(D.getIdentifierLoc(), diag::err_deduction_guide_wrong_scope)
<< GuidedTemplateDecl;
-Diag(GuidedTemplateDecl->getLocation(), diag::note_template_decl_here);
+if (GuidedTemplateDecl->getLocation().isValid())
+ Diag(GuidedTemplateDecl->getLocation(), diag::note_template_decl_here);
}
auto &DS = D.getMutableDeclSpec();
diff --git a/clang/lib/Sema/SemaInit.cpp b/clang/lib/Sema/SemaInit.cpp
index ed02d3580f34f9a..0952b065d56b58e 100644
--- a/clang/lib/Sema/SemaInit.cpp
+++ b/clang/lib/Sema/SemaInit.cpp
@@ -10575,7 +10575,8 @@ QualType
Sema::DeduceTemplateSpecializationFromInitializer(
diag::err_deduced_non_class_template_specialization_type)
<< (int)getTemplateNameKindForDiagnostics(TemplateName) << TemplateName;
if (auto *TD = TemplateName.getAsTemplateDecl())
- Diag(TD->getLocation(), diag::note_template_decl_here);
+ if (TD->getLocation().isValid())
+Diag(TD->getLocation(), diag::note_template_decl_here);
return QualType();
}
diff --git a/clang/lib/Sema/SemaLambda.cpp b/clang/lib/Sema/SemaLambda.cpp
index ca09b0481bcac76..603b7465323e237 100644
--- a/clang/lib/Sema/SemaLambda.cpp
+++ b/clang/lib/Sema/SemaLambda.cpp
@@ -1444,7 +1444,8 @@ void Sema::ActOnStartOfLambdaDefinition(LambdaIntroducer
&Intro,
for (const auto &Capture : Intro.Captures) {
if (Capture.Id == TP->getIdentifier()) {
Diag(Capture.Loc, diag::err_template_param_shadow) << Capture.Id;
- Diag(TP->getLocation(), diag::note_template_param_here);
+ if (TP->getLocation().isValid())
+Diag(TP->getLocation(), diag::note_template_param_here);
}
}
}
diff --git a/clang/lib/Sema/SemaTemplate.cpp b/clang/lib/Sema/SemaTemplate.cpp
index ee354862212803f..1a6eaa196db09be 100644
--- a/clang/lib/Sema/SemaTemplate.cpp
+++ b/clang/lib/Sema/SemaTemplate.cpp
@@ -872,7 +872,7 @@ bool Sema::DiagnoseUninstantiableTemplate(SourceLocation
PointOfInstantiation,
Note = diag::note_explicit_instantiation_here;
}
}
- if (Note) // Diagnostics were emitted.
+ if (Note && Pattern->getLocation().isValid()) // Diagnostics were emitted.
Diag(Pattern->getLocation(), *Note);
// In general, Instantiation isn't marked invalid to get more than one
@@ -899,7 +899,8 @@ void Sema::DiagnoseTemplateParameterShadow(SourceLocation
Loc, Decl *PrevDecl) {
unsigned DiagId = getLangOpts().MSVCCompat ? diag::ext_template_param_shadow
: diag::err_template_param_shadow;
Diag(Loc, DiagId) << cast(PrevDecl)->getDeclName();
- Diag(PrevDecl->getLocation(), diag::note_template_param_here);
+ if (PrevDecl->getLocation().isValid())
+Diag(PrevDecl->getLocation(), diag::note_template_param_here);
}
/// AdjustDeclIfTemplate - If the given decl happens to be a template, reset
@@ -4434,7 +4435,8 @@ static void checkMoreSpecializedThanPrimary(Sema &S,
PartialSpecDecl *Partial) {
<< SFINAEArgString;
}
- S.Diag(Template->getLocation(), diag::note_template_decl_here);
+ if (Template->getLocation().isValid())
+S.Diag(Template->getLocation(), diag::note_template_decl_here);
SmallVector PartialAC, TemplateAC;
Template->getAssociatedConstraints(Template
