[clang] 9142c0b - [clang][codegen] Hoist parameter attribute setting in function prolog.

Michael Liao via cfe-commits cfe-commits at lists.llvm.org
Tue May 5 12:32:00 PDT 2020


Author: Michael Liao
Date: 2020-05-05T15:31:51-04:00
New Revision: 9142c0b46bfea13d9348ab3d1d706a10ad9e5c8e

URL: https://github.com/llvm/llvm-project/commit/9142c0b46bfea13d9348ab3d1d706a10ad9e5c8e
DIFF: https://github.com/llvm/llvm-project/commit/9142c0b46bfea13d9348ab3d1d706a10ad9e5c8e.diff

LOG: [clang][codegen] Hoist parameter attribute setting in function prolog.

Summary:
- If the coerced type is still a pointer, it should be set with proper
  parameter attributes, such as `noalias`, `nonnull`, and etc. Hoist
  that (pointer) parameter attribute setting so that the coerced pointer
  parameter could be marked properly.

Depends on D79394

Reviewers: rjmccall, kerbowa, yaxunl

Subscribers: jvesely, nhaehnle, cfe-commits

Tags: #clang

Differential Revision: https://reviews.llvm.org/D79395

Added: 
    

Modified: 
    clang/lib/CodeGen/CGCall.cpp
    clang/test/CodeGenCUDA/amdgpu-kernel-arg-pointer-type.cu

Removed: 
    


################################################################################
diff  --git a/clang/lib/CodeGen/CGCall.cpp b/clang/lib/CodeGen/CGCall.cpp
index 44f298892ecf..e336741d9111 100644
--- a/clang/lib/CodeGen/CGCall.cpp
+++ b/clang/lib/CodeGen/CGCall.cpp
@@ -2425,15 +2425,18 @@ void CodeGenFunction::EmitFunctionProlog(const CGFunctionInfo &FI,
 
     case ABIArgInfo::Extend:
     case ABIArgInfo::Direct: {
-
-      // If we have the trivial case, handle it with no muss and fuss.
-      if (!isa<llvm::StructType>(ArgI.getCoerceToType()) &&
-          ArgI.getCoerceToType() == ConvertType(Ty) &&
-          ArgI.getDirectOffset() == 0) {
+      auto AI = Fn->getArg(FirstIRArg);
+      llvm::Type *LTy = ConvertType(Arg->getType());
+
+      // Prepare parameter attributes. So far, only attributes for pointer
+      // parameters are prepared. See
+      // http://llvm.org/docs/LangRef.html#paramattrs.
+      if (ArgI.getDirectOffset() == 0 && LTy->isPointerTy() &&
+          ArgI.getCoerceToType()->isPointerTy()) {
         assert(NumIRArgs == 1);
-        auto AI = Fn->getArg(FirstIRArg);
 
         if (const ParmVarDecl *PVD = dyn_cast<ParmVarDecl>(Arg)) {
+          // Set `nonnull` attribute if any.
           if (getNonNullAttr(CurCodeDecl, PVD, PVD->getType(),
                              PVD->getFunctionScopeIndex()) &&
               !CGM.getCodeGenOpts().NullPointerIsValid)
@@ -2471,6 +2474,7 @@ void CodeGenFunction::EmitFunctionProlog(const CGFunctionInfo &FI,
               AI->addAttr(llvm::Attribute::NonNull);
           }
 
+          // Set `align` attribute if any.
           const auto *AVAttr = PVD->getAttr<AlignValueAttr>();
           if (!AVAttr)
             if (const auto *TOTy = dyn_cast<TypedefType>(OTy))
@@ -2488,8 +2492,17 @@ void CodeGenFunction::EmitFunctionProlog(const CGFunctionInfo &FI,
           }
         }
 
+        // Set 'noalias' if an argument type has the `restrict` qualifier.
         if (Arg->getType().isRestrictQualified())
           AI->addAttr(llvm::Attribute::NoAlias);
+      }
+
+      // Prepare the argument value. If we have the trivial case, handle it
+      // with no muss and fuss.
+      if (!isa<llvm::StructType>(ArgI.getCoerceToType()) &&
+          ArgI.getCoerceToType() == ConvertType(Ty) &&
+          ArgI.getDirectOffset() == 0) {
+        assert(NumIRArgs == 1);
 
         // LLVM expects swifterror parameters to be used in very restricted
         // ways.  Copy the value into a less-restricted temporary.

diff  --git a/clang/test/CodeGenCUDA/amdgpu-kernel-arg-pointer-type.cu b/clang/test/CodeGenCUDA/amdgpu-kernel-arg-pointer-type.cu
index 6e4de1f0f5c3..8aeb0f759e6c 100644
--- a/clang/test/CodeGenCUDA/amdgpu-kernel-arg-pointer-type.cu
+++ b/clang/test/CodeGenCUDA/amdgpu-kernel-arg-pointer-type.cu
@@ -67,3 +67,10 @@ __global__ void kernel6(struct T t) {
   t.x[0][0] += 1.f;
   t.x[1][0] += 2.f;
 }
+
+// Check that coerced pointers retain the noalias attribute when qualified with __restrict.
+// CHECK: define amdgpu_kernel void @_Z7kernel7Pi(i32 addrspace(1)* noalias %x.coerce)
+// HOST: define void @_Z22__device_stub__kernel7Pi(i32* noalias %x)
+__global__ void kernel7(int *__restrict x) {
+  x[0]++;
+}


        


More information about the cfe-commits mailing list