File tree Expand file tree Collapse file tree 1 file changed +5
-4
lines changed Expand file tree Collapse file tree 1 file changed +5
-4
lines changed Original file line number Diff line number Diff line change @@ -43,19 +43,20 @@ struct TrtllmGenGemmRunnerOptions {
43
43
int64_t select_kernel_fp8 (int32_t M, int32_t N, int32_t K,
44
44
const gemm::gemm::GemmInterface& interface) {
45
45
static constexpr const char * KERNEL_NAME_HIGH_N_K_RATIO =
46
- " gemm_Bfloat16_E4m3E4m3_Fp32_t128x8x128u2_s6_et64x8_m64x8x32_cga1x1x1_16dp256b_TN_transOut_"
46
+ " gemm_Bfloat16_E4m3E4m3_Fp32_t128x8x128u2_s6_et64x8_m64x8x32_cga1x1x1_16dp256b_rM_TN_"
47
+ " transOut_"
47
48
" noShflA_dsFp8_schedP2x2x1x3_sm100f" ;
48
49
49
50
static constexpr const char * KERNEL_NAME_LOW_N_K_RATIO =
50
- " gemm_Bfloat16_E4m3E4m3_Fp32_t128x32x128u2_s6_et64x32_m64x32x32_cga1x1x1_16dp256b_TN_ "
51
+ " gemm_Bfloat16_E4m3E4m3_Fp32_t128x32x128u2_s6_et64x32_m64x32x32_cga1x1x1_16dp256b_rM_TN_ "
51
52
" transOut_noShflA_dsFp8_schedS_sm100f" ;
52
53
53
54
static constexpr const char * KERNEL_NAME_LARGE_N =
54
- " gemm_Bfloat16_E4m3E4m3_Fp32_t128x32x128u2_s6_et64x32_m64x32x32_cga1x1x1_16dp256b_TN_ "
55
+ " gemm_Bfloat16_E4m3E4m3_Fp32_t128x32x128u2_s6_et64x32_m64x32x32_cga1x1x1_16dp256b_rM_TN_ "
55
56
" transOut_noShflA_dsFp8_schedP2x2x1x3_sm100f" ;
56
57
57
58
static constexpr const char * KERNEL_NAME_DEFAULT =
58
- " gemm_Bfloat16_E4m3E4m3_Fp32_t128x16x128u2_s6_et64x16_m64x16x32_cga1x1x1_16dp256b_TN_ "
59
+ " gemm_Bfloat16_E4m3E4m3_Fp32_t128x16x128u2_s6_et64x16_m64x16x32_cga1x1x1_16dp256b_rM_TN_ "
59
60
" transOut_noShflA_dsFp8_schedS_sm100f" ;
60
61
61
62
double const n_k_ratio = static_cast <double >(N) / static_cast <double >(K);
You can’t perform that action at this time.
0 commit comments