test: add fence+sync between 2 tmem_stores

2026-05-28 09:53:10 +00:00
parent 7a21fa4bd8
commit b69a538ab1
1 changed files with 5 additions and 0 deletions
--- a/tests/unit/test_tmem_cols.cu
+++ b/tests/unit/test_tmem_cols.cu
@@ -45,8 +45,13 @@ __global__ void test_tmem_loop(float* out) {
        memcpy(&u2, &v2, 4); memcpy(&u3, &v3, 4);
        
        tmem_store(tb + 0, u0, u1, u2, u3);
+    }
+    asm volatile("tcgen05.wait::st.sync.aligned;" ::: "memory");
+    __syncthreads();
+    if (threadIdx.x < 32) {
        tmem_store(tb + 1, u0, u1, u2, u3);
    }
+    }
    asm volatile("tcgen05.wait::st.sync.aligned;" ::: "memory");
    __syncthreads();