[Mlir-commits] [mlir] 268a7a7 - [mlir][rocdl] Implement GlobalStoreAsyncFromLDS (gfx1250) (#190877)

[llvmlistbot at llvm.org]

Author: Eric
Date: 2026-04-30T18:05:10-07:00
New Revision: 268a7a73078d294a362df4fb1753a924a6ff4be5

URL: https://github.com/llvm/llvm-project/commit/268a7a73078d294a362df4fb1753a924a6ff4be5
DIFF: https://github.com/llvm/llvm-project/commit/268a7a73078d294a362df4fb1753a924a6ff4be5.diff

LOG: [mlir][rocdl] Implement GlobalStoreAsyncFromLDS (gfx1250) (#190877)

Added: 
    

Modified: 
    mlir/include/mlir/Dialect/LLVMIR/ROCDLOps.td
    mlir/test/Dialect/LLVMIR/rocdl.mlir
    mlir/test/Target/LLVMIR/rocdl.mlir

Removed: 
    


################################################################################
diff  --git a/mlir/include/mlir/Dialect/LLVMIR/ROCDLOps.td b/mlir/include/mlir/Dialect/LLVMIR/ROCDLOps.td
index c887598626d17..a9a5d0f983538 100644
--- a/mlir/include/mlir/Dialect/LLVMIR/ROCDLOps.td
+++ b/mlir/include/mlir/Dialect/LLVMIR/ROCDLOps.td
@@ -1918,6 +1918,44 @@ foreach bitsVal = [8, 32, 64, 128] in {
   }
 }
 
+//===---------------------------------------------------------------------===//
+// Async store from LDS intrinsic (available in GFX1250)
+//===---------------------------------------------------------------------===//
+
+foreach bitsVal = [8, 32, 64, 128] in {
+  defvar bitsStr = "b" # !cast<string>(bitsVal);
+  def ROCDL_GlobalStoreAsyncFromLDS # !toupper(bitsStr) # Op :
+    ROCDL_IntrOp<"global.store.async.from.lds." # bitsStr, [], [], [], 0, 0, 1, 0, [2, 3], ["offset", "aux"]> {
+    dag args = (ins Arg<ROCDLGlobalBuffer, "", [MemWrite]>:$globalPtr,
+                   Arg<ROCDLBufferLDS, "", [MemRead]>:$ldsPtr,
+                   I32Attr:$offset,
+                   I32Attr:$aux);
+    let arguments = !con(args, baseArgs);
+    let assemblyFormat = [{
+      $globalPtr `,`  $ldsPtr `,` $offset `,` $aux
+      attr-dict `:` qualified(type($globalPtr)) `,` qualified(type($ldsPtr))
+    }];
+    let description = [{
+      Asynchronously stores }] # !cast<string>(bitsVal) # [{ bits of data from a Local Data Share (LDS)
+      pointer to a global memory pointer.
+
+      Available on gfx1250+.
+
+      Example:
+      ```mlir
+      // Async }] # !cast<string>(bitsVal) # [{-bit store from LDS to global.
+      rocdl.global.store.async.from.lds.}] # bitsStr # [{ %dst, %src, 0, 0 : !llvm.ptr<1>, !llvm.ptr<3>
+      ```
+    }];
+
+    let extraClassDefinition = [{
+      ::llvm::SmallVector<::mlir::Value> $cppClass::getAccessedOperands() {
+        return {getGlobalPtr(), getLdsPtr()};
+      }
+    }];
+  }
+}
+
 foreach bitsVal = [8, 32, 64, 128] in {
   defvar bitsStr = "b" # !cast<string>(bitsVal);
   def ROCDL_ClusterLoadAsyncToLDS # !toupper(bitsStr) # Op :

diff  --git a/mlir/test/Dialect/LLVMIR/rocdl.mlir b/mlir/test/Dialect/LLVMIR/rocdl.mlir
index 1da7b63efdd7e..e8c2527313c43 100644
--- a/mlir/test/Dialect/LLVMIR/rocdl.mlir
+++ b/mlir/test/Dialect/LLVMIR/rocdl.mlir
@@ -785,6 +785,19 @@ llvm.func @rocdl.global.load.async.to.lds(%src : !llvm.ptr<1>, %dst: !llvm.ptr<3
   llvm.return
 }
 
+// CHECK-LABEL: @rocdl.global.store.async.from.lds
+llvm.func @rocdl.global.store.async.from.lds(%dst : !llvm.ptr<1>, %src: !llvm.ptr<3>) {
+  // CHECK: rocdl.global.store.async.from.lds.b8 %{{.*}}, %{{.*}}, 0, 0
+  // CHECK: rocdl.global.store.async.from.lds.b32 %{{.*}}, %{{.*}}, 0, 0
+  // CHECK: rocdl.global.store.async.from.lds.b64 %{{.*}}, %{{.*}}, 0, 0
+  // CHECK: rocdl.global.store.async.from.lds.b128 %{{.*}}, %{{.*}}, 0, 0
+  rocdl.global.store.async.from.lds.b8 %dst, %src, 0, 0 : !llvm.ptr<1>, !llvm.ptr<3>
+  rocdl.global.store.async.from.lds.b32 %dst, %src, 0, 0 : !llvm.ptr<1>, !llvm.ptr<3>
+  rocdl.global.store.async.from.lds.b64 %dst, %src, 0, 0 : !llvm.ptr<1>, !llvm.ptr<3>
+  rocdl.global.store.async.from.lds.b128 %dst, %src, 0, 0 : !llvm.ptr<1>, !llvm.ptr<3>
+  llvm.return
+}
+
 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, %{{.*}}

diff  --git a/mlir/test/Target/LLVMIR/rocdl.mlir b/mlir/test/Target/LLVMIR/rocdl.mlir
index 169702efd312f..7ea554b897646 100644
--- a/mlir/test/Target/LLVMIR/rocdl.mlir
+++ b/mlir/test/Target/LLVMIR/rocdl.mlir
@@ -1363,6 +1363,19 @@ llvm.func @rocdl.global.load.async.to.lds(%src : !llvm.ptr<1>, %dst: !llvm.ptr<3
   llvm.return
 }
 
+// CHECK-LABEL: rocdl.global.store.async.from.lds
+llvm.func @rocdl.global.store.async.from.lds(%dst : !llvm.ptr<1>, %src: !llvm.ptr<3>) {
+  // CHECK: call void @llvm.amdgcn.global.store.async.from.lds.b8
+  rocdl.global.store.async.from.lds.b8 %dst, %src, 0, 0 : !llvm.ptr<1>, !llvm.ptr<3>
+  // CHECK: call void @llvm.amdgcn.global.store.async.from.lds.b32
+  rocdl.global.store.async.from.lds.b32 %dst, %src, 0, 0 : !llvm.ptr<1>, !llvm.ptr<3>
+  // CHECK: call void @llvm.amdgcn.global.store.async.from.lds.b64
+  rocdl.global.store.async.from.lds.b64 %dst, %src, 0, 0 : !llvm.ptr<1>, !llvm.ptr<3>
+  // CHECK: call void @llvm.amdgcn.global.store.async.from.lds.b128
+  rocdl.global.store.async.from.lds.b128 %dst, %src, 0, 0 : !llvm.ptr<1>, !llvm.ptr<3>
+  llvm.return
+}
+
 // CHECK-LABEL: rocdl.cluster.load.async.to.lds
 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