diff options
Diffstat (limited to 'cuda-kernels/v8p_genericMatrixMultiply.cu')
| -rw-r--r-- | cuda-kernels/v8p_genericMatrixMultiply.cu | 2 |
1 files changed, 1 insertions, 1 deletions
diff --git a/cuda-kernels/v8p_genericMatrixMultiply.cu b/cuda-kernels/v8p_genericMatrixMultiply.cu index 2e487e8..208c369 100644 --- a/cuda-kernels/v8p_genericMatrixMultiply.cu +++ b/cuda-kernels/v8p_genericMatrixMultiply.cu @@ -92,7 +92,7 @@ __global__ void vp_example(int *a, int *b, int *c, int M, int N, int K ) { //vp::mma_sync(acc_frag, a_frag, b_frag, acc_frag); asm("/*"); asm("CPTX_BEGIN"); - asm("vp.mma.sync.row.row.m16n16k16.s32 {%0, %1, %2, %3, %4, %5, %6, %7}, {%8, %9, %10, %11, %12, %13, %14, %15}, {%16, %17}, {%18, %19, %20, %21, %22, %23, %24, %25};" : + asm("vp.mma8.sync.row.row.m16n16k16.s32 {%0, %1, %2, %3, %4, %5, %6, %7}, {%8, %9, %10, %11, %12, %13, %14, %15}, {%16, %17}, {%18, %19, %20, %21, %22, %23, %24, %25};" : "=r"(acc_frag[0]), "=r"(acc_frag[1]),"=r"(acc_frag[2]),"=r"(acc_frag[3]), "=r"(acc_frag[4]),"=r"(acc_frag[5]),"=r"(acc_frag[6]),"=r"(acc_frag[7]): "r"(a_frag[0]),"r"(a_frag[1]),"r"(a_frag[2]),"r"(a_frag[3]), |