fix: asm movmatrix.sync.aligned.m8n8.trans.b16 is not supported

a43eabbc · fengzch · 0343f43e · a43eabbc
Commit a43eabbc authored Nov 24, 2025 by fengzch
Hide whitespace changes
Inline Side-by-side

Showing with 4 additions and 3 deletions

src/kernels/zgemm/gemm_utils.cuh src/kernels/zgemm/gemm_utils.cuh +4 -3

No files found.
--- a/src/kernels/zgemm/gemm_utils.cuh
+++ b/src/kernels/zgemm/gemm_utils.cuh
@@ -199,9 +199,10 @@ __device__ __forceinline__ static void ldmatrix(const void *ptr, uint4 &out) {

 template<typename T>
 __device__ __forceinline__ static T movmatrix(T x) {
-    asm volatile("movmatrix.sync.aligned.m8n8.trans.b16 %0, %1;"
-                 : "=r"(*reinterpret_cast<uint32_t *>(&x))
-                 : "r"(*reinterpret_cast<uint32_t *>(&x)));
+    // asm volatile("movmatrix.sync.aligned.m8n8.trans.b16 %0, %1;"
+    //              : "=r"(*reinterpret_cast<uint32_t *>(&x))
+    //              : "r"(*reinterpret_cast<uint32_t *>(&x)));
+    printf("%s: asm movmatrix.sync.aligned.m8n8.trans.b16 is not supported in HIP yet!\n", __func__);
    return x;
 }