Skip to content

Commit 4e48e79

Browse files
committed
adjust gemm kernel names for rasterizeAlong parameter
1 parent 4d084a7 commit 4e48e79

File tree

1 file changed

+5
-4
lines changed

1 file changed

+5
-4
lines changed

csrc/trtllm_gemm_runner.cu

Lines changed: 5 additions & 4 deletions
Original file line numberDiff line numberDiff line change
@@ -43,19 +43,20 @@ struct TrtllmGenGemmRunnerOptions {
4343
int64_t select_kernel_fp8(int32_t M, int32_t N, int32_t K,
4444
const gemm::gemm::GemmInterface& interface) {
4545
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_"
4748
"noShflA_dsFp8_schedP2x2x1x3_sm100f";
4849

4950
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_"
5152
"transOut_noShflA_dsFp8_schedS_sm100f";
5253

5354
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_"
5556
"transOut_noShflA_dsFp8_schedP2x2x1x3_sm100f";
5657

5758
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_"
5960
"transOut_noShflA_dsFp8_schedS_sm100f";
6061

6162
double const n_k_ratio = static_cast<double>(N) / static_cast<double>(K);

0 commit comments

Comments
 (0)