[Mlir-commits] [mlir] [ROCDL] Fix cluster.load.async.to.lds mask parameter to be SSA value (PR #174411)

[Alexander Weinrauch]

https://github.com/AlexAUT created https://github.com/llvm/llvm-project/pull/174411

The (broadcast) `mask` parameter in `rocdl.cluster.load.async.to.lds.*` operations was incorrectly defined as an `I32Attr` when it should be an `I32` as defined by the intrinsic.

>From 9b53344a9e8f3cb71e9181a2ec5cc9fe64b06e78 Mon Sep 17 00:00:00 2001
From: Alexander Weinrauch <alexander.weinrauch at amd.com>
Date: Mon, 5 Jan 2026 14:08:56 +0000
Subject: [PATCH] Mask should not be an int attribute

---
 mlir/include/mlir/Dialect/LLVMIR/ROCDLOps.td |  4 ++--
 mlir/test/Dialect/LLVMIR/rocdl.mlir          | 18 +++++++++---------
 mlir/test/Target/LLVMIR/rocdl.mlir           | 10 +++++-----
 3 files changed, 16 insertions(+), 16 deletions(-)

diff --git a/mlir/include/mlir/Dialect/LLVMIR/ROCDLOps.td b/mlir/include/mlir/Dialect/LLVMIR/ROCDLOps.td
index f0a9d97b6daaf..7968d14be6592 100644
--- a/mlir/include/mlir/Dialect/LLVMIR/ROCDLOps.td
+++ b/mlir/include/mlir/Dialect/LLVMIR/ROCDLOps.td
@@ -1063,12 +1063,12 @@ foreach bitsVal = [8, 32, 64, 128] in {
 foreach bitsVal = [8, 32, 64, 128] in {
   defvar bitsStr = "b" # !cast<string>(bitsVal);
   def ROCDL_ClusterLoadAsyncToLDS # !toupper(bitsStr) # Op :
-    ROCDL_IntrOp<"cluster.load.async.to.lds." # bitsStr, [], [], [], 0, 0, 1, 0, [2, 3, 4], ["offset", "cpol", "mask"]> {
+    ROCDL_IntrOp<"cluster.load.async.to.lds." # bitsStr, [], [], [], 0, 0, 1, 0, [2, 3], ["offset", "cpol"]> {
     dag args = (ins Arg<ROCDLGlobalBuffer, "", [MemRead]>:$globalPtr,
                    Arg<ROCDLBufferLDS, "", [MemWrite]>:$ldsPtr,
                    I32Attr:$offset,
                    I32Attr:$cpol,
-                   I32Attr:$mask);
+                   I32:$mask);
     let arguments = !con(args, baseArgs);
     let assemblyFormat = [{
       $globalPtr `,`  $ldsPtr `,` $offset `,` $cpol `,` $mask
diff --git a/mlir/test/Dialect/LLVMIR/rocdl.mlir b/mlir/test/Dialect/LLVMIR/rocdl.mlir
index b630fb86a5ab2..cf2b144219f36 100644
--- a/mlir/test/Dialect/LLVMIR/rocdl.mlir
+++ b/mlir/test/Dialect/LLVMIR/rocdl.mlir
@@ -843,16 +843,16 @@ llvm.func @rocdl.global.load.async.to.lds(%src : !llvm.ptr<1>, %dst: !llvm.ptr<3
   llvm.return
 }
 
-llvm.func @rocdl.cluster.load.async.to.lds(%src : !llvm.ptr<1>, %dst: !llvm.ptr<3>) {
+llvm.func @rocdl.cluster.load.async.to.lds(%src : !llvm.ptr<1>, %dst: !llvm.ptr<3>, %mask: i32) {
   // CHECK-LABEL @rocdl.cluster.load.async.to.lds
-  // CHECK: rocdl.cluster.load.async.to.lds.b8 %{{.*}}, %{{.*}}, 0, 0, 0
-  // CHECK: rocdl.cluster.load.async.to.lds.b32 %{{.*}}, %{{.*}}, 0, 0, 0
-  // CHECK: rocdl.cluster.load.async.to.lds.b64 %{{.*}}, %{{.*}}, 0, 0, 0
-  // CHECK: rocdl.cluster.load.async.to.lds.b128 %{{.*}}, %{{.*}}, 0, 0, 0
-  rocdl.cluster.load.async.to.lds.b8 %src, %dst, 0, 0, 0 : !llvm.ptr<1>, !llvm.ptr<3>
-  rocdl.cluster.load.async.to.lds.b32 %src, %dst, 0, 0, 0 : !llvm.ptr<1>, !llvm.ptr<3>
-  rocdl.cluster.load.async.to.lds.b64 %src, %dst, 0, 0, 0 : !llvm.ptr<1>, !llvm.ptr<3>
-  rocdl.cluster.load.async.to.lds.b128 %src, %dst, 0, 0, 0 : !llvm.ptr<1>, !llvm.ptr<3>
+  // CHECK: rocdl.cluster.load.async.to.lds.b8 %{{.*}}, %{{.*}}, 0, 0, %{{.*}}
+  // CHECK: rocdl.cluster.load.async.to.lds.b32 %{{.*}}, %{{.*}}, 0, 0, %{{.*}}
+  // CHECK: rocdl.cluster.load.async.to.lds.b64 %{{.*}}, %{{.*}}, 0, 0, %{{.*}}
+  // CHECK: rocdl.cluster.load.async.to.lds.b128 %{{.*}}, %{{.*}}, 0, 0, %{{.*}}
+  rocdl.cluster.load.async.to.lds.b8 %src, %dst, 0, 0, %mask : !llvm.ptr<1>, !llvm.ptr<3>
+  rocdl.cluster.load.async.to.lds.b32 %src, %dst, 0, 0, %mask : !llvm.ptr<1>, !llvm.ptr<3>
+  rocdl.cluster.load.async.to.lds.b64 %src, %dst, 0, 0, %mask : !llvm.ptr<1>, !llvm.ptr<3>
+  rocdl.cluster.load.async.to.lds.b128 %src, %dst, 0, 0, %mask : !llvm.ptr<1>, !llvm.ptr<3>
   llvm.return
 }
 
diff --git a/mlir/test/Target/LLVMIR/rocdl.mlir b/mlir/test/Target/LLVMIR/rocdl.mlir
index 9022beb71ee31..cc3df8cd05087 100644
--- a/mlir/test/Target/LLVMIR/rocdl.mlir
+++ b/mlir/test/Target/LLVMIR/rocdl.mlir
@@ -1282,15 +1282,15 @@ llvm.func @rocdl.global.load.async.to.lds(%src : !llvm.ptr<1>, %dst: !llvm.ptr<3
 }
 
 // CHECK-LABEL: rocdl.cluster.load.async.to.lds
-llvm.func @rocdl.cluster.load.async.to.lds(%src : !llvm.ptr<1>, %dst: !llvm.ptr<3>) {
+llvm.func @rocdl.cluster.load.async.to.lds(%src : !llvm.ptr<1>, %dst: !llvm.ptr<3>, %mask: i32) {
   // CHECK: call void @llvm.amdgcn.cluster.load.async.to.lds.b8
-  rocdl.cluster.load.async.to.lds.b8 %src, %dst, 0, 0, 0 : !llvm.ptr<1>, !llvm.ptr<3>
+  rocdl.cluster.load.async.to.lds.b8 %src, %dst, 0, 0, %mask : !llvm.ptr<1>, !llvm.ptr<3>
   // CHECK: call void @llvm.amdgcn.cluster.load.async.to.lds.b32
-  rocdl.cluster.load.async.to.lds.b32 %src, %dst, 0, 0, 0 : !llvm.ptr<1>, !llvm.ptr<3>
+  rocdl.cluster.load.async.to.lds.b32 %src, %dst, 0, 0, %mask : !llvm.ptr<1>, !llvm.ptr<3>
   // CHECK: call void @llvm.amdgcn.cluster.load.async.to.lds.b64
-  rocdl.cluster.load.async.to.lds.b64 %src, %dst, 0, 0, 0 : !llvm.ptr<1>, !llvm.ptr<3>
+  rocdl.cluster.load.async.to.lds.b64 %src, %dst, 0, 0, %mask : !llvm.ptr<1>, !llvm.ptr<3>
   // CHECK: call void @llvm.amdgcn.cluster.load.async.to.lds.b128
-  rocdl.cluster.load.async.to.lds.b128 %src, %dst, 0, 0, 0 : !llvm.ptr<1>, !llvm.ptr<3>
+  rocdl.cluster.load.async.to.lds.b128 %src, %dst, 0, 0, %mask : !llvm.ptr<1>, !llvm.ptr<3>
   llvm.return
 }