[HLSL] Don't use CreateRuntimeFunction for intrinsics

HLSL uses CreateRuntimeFunction for two intrinsics. This is pretty
weird thing to do, and doesn't match what the rest of the file
does.

I suspect this might be because these are convergent calls, but
the intrinsics themselves are already marked convergent, so it's
not necessary for clang to manually add the attribute.

Author: nikic

clang/lib/CodeGen/CGHLSLBuiltins.cpp

@@ -676,35 +676,23 @@ Value *CodeGenFunction::EmitHLSLBuiltinExpr(unsigned BuiltinID,
   case Builtin::BI__builtin_hlsl_wave_active_sum: {
     // Due to the use of variadic arguments, explicitly retreive argument
     Value *OpExpr = EmitScalarExpr(E->getArg(0));
-    llvm::FunctionType *FT = llvm::FunctionType::get(
-        OpExpr->getType(), ArrayRef{OpExpr->getType()}, false);
     Intrinsic::ID IID = getWaveActiveSumIntrinsic(
         getTarget().getTriple().getArch(), CGM.getHLSLRuntime(),
         E->getArg(0)->getType());
 
-    // Get overloaded name
-    std::string Name =
-        Intrinsic::getName(IID, ArrayRef{OpExpr->getType()}, &CGM.getModule());
-    return EmitRuntimeCall(CGM.CreateRuntimeFunction(FT, Name, {},
-                                                     /*Local=*/false,
-                                                     /*AssumeConvergent=*/true),
+    return EmitRuntimeCall(Intrinsic::getOrInsertDeclaration(
+                               &CGM.getModule(), IID, {OpExpr->getType()}),
                            ArrayRef{OpExpr}, "hlsl.wave.active.sum");
   }
   case Builtin::BI__builtin_hlsl_wave_active_max: {
     // Due to the use of variadic arguments, explicitly retreive argument
     Value *OpExpr = EmitScalarExpr(E->getArg(0));
-    llvm::FunctionType *FT = llvm::FunctionType::get(
-        OpExpr->getType(), ArrayRef{OpExpr->getType()}, false);
     Intrinsic::ID IID = getWaveActiveMaxIntrinsic(
         getTarget().getTriple().getArch(), CGM.getHLSLRuntime(),
         E->getArg(0)->getType());
 
-    // Get overloaded name
-    std::string Name =
-        Intrinsic::getName(IID, ArrayRef{OpExpr->getType()}, &CGM.getModule());
-    return EmitRuntimeCall(CGM.CreateRuntimeFunction(FT, Name, {},
-                                                     /*Local=*/false,
-                                                     /*AssumeConvergent=*/true),
+    return EmitRuntimeCall(Intrinsic::getOrInsertDeclaration(
+                               &CGM.getModule(), IID, {OpExpr->getType()}),
                            ArrayRef{OpExpr}, "hlsl.wave.active.max");
   }
   case Builtin::BI__builtin_hlsl_wave_get_lane_index: {

clang/test/CodeGenHLSL/builtins/WaveActiveMax.hlsl

@@ -16,7 +16,7 @@ int test_int(int expr) {
 }
 
 // CHECK-DXIL: declare [[TY]] @llvm.dx.wave.reduce.max.i32([[TY]]) #[[#attr:]]
-// CHECK-SPIRV: declare spir_func [[TY]] @llvm.spv.wave.reduce.max.i32([[TY]]) #[[#attr:]]
+// CHECK-SPIRV: declare [[TY]] @llvm.spv.wave.reduce.max.i32([[TY]]) #[[#attr:]]
 
 // CHECK-LABEL: test_uint64_t
 uint64_t test_uint64_t(uint64_t expr) {
@@ -27,7 +27,7 @@ uint64_t test_uint64_t(uint64_t expr) {
 }
 
 // CHECK-DXIL: declare [[TY]] @llvm.dx.wave.reduce.umax.i64([[TY]]) #[[#attr:]]
-// CHECK-SPIRV: declare spir_func [[TY]] @llvm.spv.wave.reduce.umax.i64([[TY]]) #[[#attr:]]
+// CHECK-SPIRV: declare [[TY]] @llvm.spv.wave.reduce.umax.i64([[TY]]) #[[#attr:]]
 
 // Test basic lowering to runtime function call with array and float value.
 
@@ -40,7 +40,7 @@ float4 test_floatv4(float4 expr) {
 }
 
 // CHECK-DXIL: declare [[TY1]] @llvm.dx.wave.reduce.max.v4f32([[TY1]]) #[[#attr]]
-// CHECK-SPIRV: declare spir_func [[TY1]] @llvm.spv.wave.reduce.max.v4f32([[TY1]]) #[[#attr]]
+// CHECK-SPIRV: declare [[TY1]] @llvm.spv.wave.reduce.max.v4f32([[TY1]]) #[[#attr]]
 
 // CHECK: attributes #[[#attr]] = {{{.*}} convergent {{.*}}}
 

clang/test/CodeGenHLSL/builtins/WaveActiveSum.hlsl

@@ -16,7 +16,7 @@ int test_int(int expr) {
 }
 
 // CHECK-DXIL: declare [[TY]] @llvm.dx.wave.reduce.sum.i32([[TY]]) #[[#attr:]]
-// CHECK-SPIRV: declare spir_func [[TY]] @llvm.spv.wave.reduce.sum.i32([[TY]]) #[[#attr:]]
+// CHECK-SPIRV: declare [[TY]] @llvm.spv.wave.reduce.sum.i32([[TY]]) #[[#attr:]]
 
 // CHECK-LABEL: test_uint64_t
 uint64_t test_uint64_t(uint64_t expr) {
@@ -27,7 +27,7 @@ uint64_t test_uint64_t(uint64_t expr) {
 }
 
 // CHECK-DXIL: declare [[TY]] @llvm.dx.wave.reduce.usum.i64([[TY]]) #[[#attr:]]
-// CHECK-SPIRV: declare spir_func [[TY]] @llvm.spv.wave.reduce.sum.i64([[TY]]) #[[#attr:]]
+// CHECK-SPIRV: declare [[TY]] @llvm.spv.wave.reduce.sum.i64([[TY]]) #[[#attr:]]
 
 // Test basic lowering to runtime function call with array and float value.
 
@@ -40,6 +40,6 @@ float4 test_floatv4(float4 expr) {
 }
 
 // CHECK-DXIL: declare [[TY1]] @llvm.dx.wave.reduce.sum.v4f32([[TY1]]) #[[#attr]]
-// CHECK-SPIRV: declare spir_func [[TY1]] @llvm.spv.wave.reduce.sum.v4f32([[TY1]]) #[[#attr]]
+// CHECK-SPIRV: declare [[TY1]] @llvm.spv.wave.reduce.sum.v4f32([[TY1]]) #[[#attr]]
 
 // CHECK: attributes #[[#attr]] = {{{.*}} convergent {{.*}}}