Update

Author: yudongsi

benchmarks/xetla_kernel/python_main.cpp

@@ -309,13 +309,13 @@ PYBIND11_MODULE(xetla_kernel, m) {
         "bf16_gemm_streamk (XeTLA)");
   // gemm split k
   m.def("gemm_splitk_shape_512_32768_8192",
-        &bf16_split_k_gemm<512, 32768, 8192, kslicing_impl_t::none>,
+        &bf16_split_k_gemm<512, 32768, 8192, kslicing_impl_t::global>,
         "bf16_gemm_splitk (XeTLA)");
   m.def("gemm_splitk_shape_1024_28672_8192",
-        &bf16_split_k_gemm<1024, 28672, 8192, kslicing_impl_t::none>,
+        &bf16_split_k_gemm<1024, 28672, 8192, kslicing_impl_t::global>,
         "bf16_gemm_splitk (XeTLA)");
   m.def("gemm_splitk_shape_3072_4096_3072",
-        &bf16_split_k_gemm<3072, 4096, 3072, kslicing_impl_t::none>,
+        &bf16_split_k_gemm<3072, 4096, 3072, kslicing_impl_t::global>,
         "bf16_gemm_splitk (XeTLA)");
   // flash_attn
   m.def("flash_attn_causal_false", &flash_attn<false, false, false>,

benchmarks/xetla_kernel/xetla-library.conf

@@ -1 +1 @@
-ae46a690bac364a93437e248418636c2a8423134
+b9e489ca6a776694a898044a3f2ae023a98db03d