[OpaquePointers] Demangle _Float* correctly for determining argument types. (#1849)

Itanium name mangling gives different type names for the _Float16, _Float32,
etc. types than the more common half/float/double names. This adds support for
people who use these types in SPIR-V kernels.

Author: jcranmer-intel

lib/SPIRV/SPIRVUtil.cpp

@@ -653,6 +653,37 @@ static StringRef stringify(const itanium_demangle::NameType *Node) {
   return StringRef(Str.begin(), Str.size());
 }
 
+/// Convert a mangled name that represents a basic integer, floating-point,
+/// etc. type into the corresponding LLVM type.
+static Type *getPrimitiveType(LLVMContext &Ctx,
+                              const llvm::itanium_demangle::Node *N) {
+  using namespace llvm::itanium_demangle;
+  if (auto *Name = dyn_cast<NameType>(N)) {
+    return parsePrimitiveType(Ctx, stringify(Name));
+  }
+  if (auto *BitInt = dyn_cast<BitIntType>(N)) {
+    unsigned BitWidth = 0;
+    BitInt->match([&](const Node *NodeSize, bool) {
+      const StringRef SizeStr(stringify(cast<NameType>(NodeSize)));
+      SizeStr.getAsInteger(10, BitWidth);
+    });
+    return Type::getIntNTy(Ctx, BitWidth);
+  }
+  if (auto *FP = dyn_cast<BinaryFPType>(N)) {
+    StringRef SizeStr;
+    FP->match([&](const Node *NodeDimension) {
+      SizeStr = stringify(cast<NameType>(NodeDimension));
+    });
+    return StringSwitch<Type *>(SizeStr)
+        .Case("16", Type::getHalfTy(Ctx))
+        .Case("32", Type::getFloatTy(Ctx))
+        .Case("64", Type::getDoubleTy(Ctx))
+        .Case("128", Type::getFP128Ty(Ctx))
+        .Default(nullptr);
+  }
+  return nullptr;
+}
+
 template <typename FnType>
 static TypedPointerType *
 parseNode(Module *M, const llvm::itanium_demangle::Node *ParamType,
@@ -724,21 +755,15 @@ parseNode(Module *M, const llvm::itanium_demangle::Node *ParamType,
       } else {
         PointeeTy = parsePrimitiveType(M->getContext(), MangledStructName);
       }
-    } else if (auto *BitInt = dyn_cast<BitIntType>(Pointee)) {
-      unsigned BitWidth = 0;
-      BitInt->match([&](const Node *NodeSize, bool) {
-        const StringRef SizeStr(stringify(cast<NameType>(NodeSize)));
-        SizeStr.getAsInteger(10, BitWidth);
-      });
-      PointeeTy = Type::getIntNTy(M->getContext(), BitWidth);
+    } else if (auto *Ty = getPrimitiveType(M->getContext(), Pointee)) {
+      PointeeTy = Ty;
     } else if (auto *Vec = dyn_cast<itanium_demangle::VectorType>(Pointee)) {
       unsigned ElemCount = 0;
       const StringRef ElemCountStr(
           stringify(cast<NameType>(Vec->getDimension())));
       ElemCountStr.getAsInteger(10, ElemCount);
-      if (auto *Name = dyn_cast<NameType>(Vec->getBaseType())) {
-        PointeeTy = parsePrimitiveType(M->getContext(), stringify(Name));
-        PointeeTy = llvm::VectorType::get(PointeeTy, ElemCount, false);
+      if (auto *Ty = getPrimitiveType(M->getContext(), Vec->getBaseType())) {
+        PointeeTy = llvm::VectorType::get(Ty, ElemCount, false);
       }
     } else if (llvm::isa<itanium_demangle::PointerType>(Pointee)) {
       PointeeTy = parseNode(M, Pointee, GetStructType);

test/transcoding/float16.ll

@@ -0,0 +1,38 @@
+; RUN: llvm-as %s -o %t.bc
+; RUN: llvm-spirv %t.bc -o %t.spv
+; RUN: spirv-val %t.spv
+; RUN: llvm-spirv %t.spv -to-text -o - | FileCheck %s --check-prefix=CHECK-SPIRV
+; RUN: llvm-spirv -r %t.spv -o %t.rev.bc -emit-opaque-pointers
+; RUN: llvm-dis %t.rev.bc -o - | FileCheck %s --check-prefix=CHECK-LLVM
+
+source_filename = "math_builtin_float_half.cpp"
+target datalayout = "e-i64:64-v16:16-v24:32-v32:32-v48:64-v96:128-v192:256-v256:256-v512:512-v1024:1024-n8:16:32:64"
+target triple = "spirv64-unknown-unknown"
+
+; CHECK-SPIRV: 3 TypeFloat [[HALF:[0-9]+]] 16
+; CHECK-SPIRV: 4 TypePointer [[HALFPTR:[0-9]+]] 7 [[HALF]]
+; CHECK-SPIRV: 4 TypeVector [[HALFV2:[0-9]+]] [[HALF]] 2
+; CHECK-SPIRV: 4 TypePointer [[HALFV2PTR:[0-9]+]] 7 [[HALFV2]]
+; CHECK-SPIRV: 4 Constant [[HALF]] [[CONST:[0-9]+]] 14788
+; CHECK-SPIRV: 4 Variable [[HALFPTR]] [[ADDR:[0-9]+]] 7
+; CHECK-SPIRV: 4 Variable [[HALFV2PTR]] [[ADDR2:[0-9]+]] 7
+; CHECK-SPIRV: 7 ExtInst [[HALF]] [[#]] 1 fract [[CONST]] [[ADDR]]
+; CHECK-SPIRV: 7 ExtInst [[HALFV2]] [[#]] 1 fract [[#]] [[ADDR2]]
+
+; CHECK-LLVM: %addr = alloca half
+; CHECK-LLVM: %addr2 = alloca <2 x half>
+; CHECK-LLVM: %res = call spir_func half @_Z5fractDhPDh(half 0xH39C4, ptr %addr)
+; CHECK-LLVM: %res2 = call spir_func <2 x half> @_Z5fractDv2_DhPS_(<2 x half> <half 0xH39C4, half 0xH0000>, ptr %addr2)
+
+define spir_kernel void @test() {
+entry:
+  %addr = alloca half
+  %addr2 = alloca <2 x half>
+  %res = call spir_func noundef half @_Z17__spirv_ocl_fractDF16_PU3AS0DF16_(half noundef 0xH39C4, ptr noundef %addr)
+  %res2 = call spir_func noundef <2 x half> @_Z17__spirv_ocl_fractDv2_DF16_PU3AS0S_(<2 x half> noundef <half 0xH39C4, half 0xH0000>, ptr noundef %addr2)
+  ret void
+}
+
+declare spir_func noundef half @_Z17__spirv_ocl_fractDF16_PU3AS0DF16_(half noundef, ptr noundef) local_unnamed_addr
+
+declare spir_func noundef <2 x half> @_Z17__spirv_ocl_fractDv2_DF16_PU3AS0S_(<2 x half> noundef, ptr noundef) local_unnamed_addr