@@ -779,8 +779,7 @@ func.func @warpgroup_mma_store(
779
779
%result2 : !nvgpu.warpgroup.accumulator < fragmented = vector <64 x128 xf32 >>,
780
780
%matrixD: memref <128 x128 xf32 ,3 >) {
781
781
// CHECK: %[[S0:.+]] = builtin.unrealized_conversion_cast %[[arg0]] : !nvgpu.warpgroup.accumulator<fragmented = vector<64x128xf32>> to !llvm.struct<(f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32)>
782
- // CHECK: %[[DB:.+]] = builtin.unrealized_conversion_cast %[[arg1]] : !nvgpu.warpgroup.accumulator<fragmented = vector<64x128xf32>> to !llvm.struct<(f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32)>
783
- // CHECK: %[[DST:.+]] = builtin.unrealized_conversion_cast %[[arg2]] :
782
+ // CHECK: %[[S1:.+]] = builtin.unrealized_conversion_cast %[[arg1]] : !nvgpu.warpgroup.accumulator<fragmented = vector<64x128xf32>> to !llvm.struct<(f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32)>
784
783
// CHECK: %[[S2:.+]] = llvm.mlir.constant(4 : i32) : i32
785
784
// CHECK: %[[S3:.+]] = llvm.mlir.constant(32 : i32) : i32
786
785
// CHECK: %[[S4:.+]] = llvm.mlir.constant(8 : i32) : i32
0 commit comments