[flang][cuda] Fix pred type for vote functions

[clementval]

No description provided.

[llvmbot]

@llvm/pr-subscribers-flang-fir-hlfir

Author: Valentin Clement (バレンタイン クレメン) (clementval)

Changes

---

Full diff: https://github.com/llvm/llvm-project/pull/134166.diff

2 Files Affected:

• (modified) flang/lib/Optimizer/Builder/IntrinsicCall.cpp (+9-5)
• (modified) flang/test/Lower/CUDA/cuda-device-proc.cuf (+5-4)

diff --git a/flang/lib/Optimizer/Builder/IntrinsicCall.cpp b/flang/lib/Optimizer/Builder/IntrinsicCall.cpp
index 8aed288d128b6..4988b6bfb3d3f 100644
--- a/flang/lib/Optimizer/Builder/IntrinsicCall.cpp
+++ b/flang/lib/Optimizer/Builder/IntrinsicCall.cpp
@@ -6508,12 +6508,13 @@ IntrinsicLibrary::genMatchAllSync(mlir::Type resultType,
 }
 
 static mlir::Value genVoteSync(fir::FirOpBuilder &builder, mlir::Location loc,
-                               llvm::StringRef funcName,
+                               llvm::StringRef funcName, mlir::Type resTy,
                                llvm::ArrayRef<mlir::Value> args) {
   mlir::MLIRContext *context = builder.getContext();
   mlir::Type i32Ty = builder.getI32Type();
+  mlir::Type i1Ty = builder.getI1Type();
   mlir::FunctionType ftype =
-      mlir::FunctionType::get(context, {i32Ty, i32Ty}, {i32Ty});
+      mlir::FunctionType::get(context, {i32Ty, i1Ty}, {resTy});
   auto funcOp = builder.createFunction(loc, funcName, ftype);
   llvm::SmallVector<mlir::Value> filteredArgs;
   return builder.create<fir::CallOp>(loc, funcOp, args).getResult(0);
@@ -6523,14 +6524,16 @@ static mlir::Value genVoteSync(fir::FirOpBuilder &builder, mlir::Location loc,
 mlir::Value IntrinsicLibrary::genVoteAllSync(mlir::Type resultType,
                                              llvm::ArrayRef<mlir::Value> args) {
   assert(args.size() == 2);
-  return genVoteSync(builder, loc, "llvm.nvvm.vote.all.sync", args);
+  return genVoteSync(builder, loc, "llvm.nvvm.vote.all.sync",
+                     builder.getI1Type(), args);
 }
 
 // ANY_SYNC
 mlir::Value IntrinsicLibrary::genVoteAnySync(mlir::Type resultType,
                                              llvm::ArrayRef<mlir::Value> args) {
   assert(args.size() == 2);
-  return genVoteSync(builder, loc, "llvm.nvvm.vote.any.sync", args);
+  return genVoteSync(builder, loc, "llvm.nvvm.vote.any.sync",
+                     builder.getI1Type(), args);
 }
 
 // BALLOT_SYNC
@@ -6538,7 +6541,8 @@ mlir::Value
 IntrinsicLibrary::genVoteBallotSync(mlir::Type resultType,
                                     llvm::ArrayRef<mlir::Value> args) {
   assert(args.size() == 2);
-  return genVoteSync(builder, loc, "llvm.nvvm.vote.ballot.sync", args);
+  return genVoteSync(builder, loc, "llvm.nvvm.vote.ballot.sync",
+                     builder.getI32Type(), args);
 }
 
 // MATCH_ANY_SYNC
diff --git a/flang/test/Lower/CUDA/cuda-device-proc.cuf b/flang/test/Lower/CUDA/cuda-device-proc.cuf
index 6a7fee73f338a..a4a4750dd61e6 100644
--- a/flang/test/Lower/CUDA/cuda-device-proc.cuf
+++ b/flang/test/Lower/CUDA/cuda-device-proc.cuf
@@ -297,10 +297,11 @@ end
 ! CHECK: fir.call @__ldcv_r8x2_(%{{.*}}, %{{.*}}) fastmath<contract> : (!fir.ref<!fir.array<2xf64>>, !fir.ref<!fir.array<?xf64>>) -> ()
 
 attributes(device) subroutine testVote()
-  integer :: a, ipred, mask, v32
-  a = all_sync(mask, v32)
-  a = any_sync(mask, v32)
-  a = ballot_sync(mask, v32)
+  integer :: a, ipred, mask
+  logical(4) :: pred
+  a = all_sync(mask, pred)
+  a = any_sync(mask, pred)
+  a = ballot_sync(mask, pred)
 end subroutine
 
 ! CHECK-LABEL: func.func @_QPtestvote()