Skip to content

GitLab

Commit b619815a authored by Muhammed Emin Ozturk's avatar Muhammed Emin Ozturk
Browse files

gemm universal sk fix

parent 2a971875
Hide whitespace changes
Inline Side-by-side
Showing with 38 additions and 38 deletions
library/include/ck/library/tensor_operation_instance/gpu/gemm_universal_streamk.hpp
View file @ b619815a
...@@ -238,188 +238,188 @@ void add_device_gemm_xdl_universal_streamk_f16_f16_f16_mk_nk_mn_mem_v2_mnkpaddin ...@@ -238,188 +238,188 @@ void add_device_gemm_xdl_universal_streamk_f16_f16_f16_mk_nk_mn_mem_v2_mnkpaddin
PassThrough>>>& instances); PassThrough>>>& instances);
#endif #endif
// Emin @Added
#ifdef CK_ENABLE_BF16 #ifdef CK_ENABLE_BF16
void add_device_gemm_xdl_universal_streamk_bf16_bf16_bf16_mk_kn_mn_comp_default_instances( void add_device_gemm_xdl_universal_streamk_bf16_bf16_bf16_mk_kn_mn_comp_default_instances(
std::vector<std::unique_ptr< std::vector<std::unique_ptr<
DeviceGemmV2<Row, Row, Row, BF16, BF16, BF16, PassThrough, PassThrough, PassThrough>>>& DeviceGemm_Streamk_V2<Row, Row, Row, BF16, BF16, BF16, PassThrough, PassThrough, PassThrough>>>&
instances); instances);
void add_device_gemm_xdl_universal_streamk_bf16_bf16_bf16_mk_kn_mn_comp_kpadding_instances( void add_device_gemm_xdl_universal_streamk_bf16_bf16_bf16_mk_kn_mn_comp_kpadding_instances(
std::vector<std::unique_ptr< std::vector<std::unique_ptr<
DeviceGemmV2<Row, Row, Row, BF16, BF16, BF16, PassThrough, PassThrough, PassThrough>>>& DeviceGemm_Streamk_V2<Row, Row, Row, BF16, BF16, BF16, PassThrough, PassThrough, PassThrough>>>&
instances); instances);
void add_device_gemm_xdl_universal_streamk_bf16_bf16_bf16_mk_kn_mn_comp_mnpadding_instances( void add_device_gemm_xdl_universal_streamk_bf16_bf16_bf16_mk_kn_mn_comp_mnpadding_instances(
std::vector<std::unique_ptr< std::vector<std::unique_ptr<
DeviceGemmV2<Row, Row, Row, BF16, BF16, BF16, PassThrough, PassThrough, PassThrough>>>& DeviceGemm_Streamk_V2<Row, Row, Row, BF16, BF16, BF16, PassThrough, PassThrough, PassThrough>>>&
instances); instances);
void add_device_gemm_xdl_universal_streamk_bf16_bf16_bf16_mk_kn_mn_comp_mnkpadding_instances( void add_device_gemm_xdl_universal_streamk_bf16_bf16_bf16_mk_kn_mn_comp_mnkpadding_instances(
std::vector<std::unique_ptr< std::vector<std::unique_ptr<
DeviceGemmV2<Row, Row, Row, BF16, BF16, BF16, PassThrough, PassThrough, PassThrough>>>& DeviceGemm_Streamk_V2<Row, Row, Row, BF16, BF16, BF16, PassThrough, PassThrough, PassThrough>>>&
instances); instances);
void add_device_gemm_xdl_universal_streamk_bf16_bf16_bf16_mk_kn_mn_mem_v1_default_instances( void add_device_gemm_xdl_universal_streamk_bf16_bf16_bf16_mk_kn_mn_mem_v1_default_instances(
std::vector<std::unique_ptr< std::vector<std::unique_ptr<
DeviceGemmV2<Row, Row, Row, BF16, BF16, BF16, PassThrough, PassThrough, PassThrough>>>& DeviceGemm_Streamk_V2<Row, Row, Row, BF16, BF16, BF16, PassThrough, PassThrough, PassThrough>>>&
instances); instances);
void add_device_gemm_xdl_universal_streamk_bf16_bf16_bf16_mk_kn_mn_mem_v1_kpadding_instances( void add_device_gemm_xdl_universal_streamk_bf16_bf16_bf16_mk_kn_mn_mem_v1_kpadding_instances(
std::vector<std::unique_ptr< std::vector<std::unique_ptr<
DeviceGemmV2<Row, Row, Row, BF16, BF16, BF16, PassThrough, PassThrough, PassThrough>>>& DeviceGemm_Streamk_V2<Row, Row, Row, BF16, BF16, BF16, PassThrough, PassThrough, PassThrough>>>&
instances); instances);
void add_device_gemm_xdl_universal_streamk_bf16_bf16_bf16_mk_kn_mn_mem_v1_mnkpadding_instances( void add_device_gemm_xdl_universal_streamk_bf16_bf16_bf16_mk_kn_mn_mem_v1_mnkpadding_instances(
std::vector<std::unique_ptr< std::vector<std::unique_ptr<
DeviceGemmV2<Row, Row, Row, BF16, BF16, BF16, PassThrough, PassThrough, PassThrough>>>& DeviceGemm_Streamk_V2<Row, Row, Row, BF16, BF16, BF16, PassThrough, PassThrough, PassThrough>>>&
instances); instances);
void add_device_gemm_xdl_universal_streamk_bf16_bf16_bf16_mk_kn_mn_mem_v2_default_instances( void add_device_gemm_xdl_universal_streamk_bf16_bf16_bf16_mk_kn_mn_mem_v2_default_instances(
std::vector<std::unique_ptr< std::vector<std::unique_ptr<
DeviceGemmV2<Row, Row, Row, BF16, BF16, BF16, PassThrough, PassThrough, PassThrough>>>& DeviceGemm_Streamk_V2<Row, Row, Row, BF16, BF16, BF16, PassThrough, PassThrough, PassThrough>>>&
instances); instances);
void add_device_gemm_xdl_universal_streamk_bf16_bf16_bf16_mk_kn_mn_mem_v2_kpadding_instances( void add_device_gemm_xdl_universal_streamk_bf16_bf16_bf16_mk_kn_mn_mem_v2_kpadding_instances(
std::vector<std::unique_ptr< std::vector<std::unique_ptr<
DeviceGemmV2<Row, Row, Row, BF16, BF16, BF16, PassThrough, PassThrough, PassThrough>>>& DeviceGemm_Streamk_V2<Row, Row, Row, BF16, BF16, BF16, PassThrough, PassThrough, PassThrough>>>&
instances); instances);
void add_device_gemm_xdl_universal_streamk_bf16_bf16_bf16_mk_kn_mn_mem_v2_mnkpadding_instances( void add_device_gemm_xdl_universal_streamk_bf16_bf16_bf16_mk_kn_mn_mem_v2_mnkpadding_instances(
std::vector<std::unique_ptr< std::vector<std::unique_ptr<
DeviceGemmV2<Row, Row, Row, BF16, BF16, BF16, PassThrough, PassThrough, PassThrough>>>& DeviceGemm_Streamk_V2<Row, Row, Row, BF16, BF16, BF16, PassThrough, PassThrough, PassThrough>>>&
instances); instances);
void add_device_gemm_xdl_universal_streamk_bf16_bf16_bf16_mk_nk_mn_comp_default_instances( void add_device_gemm_xdl_universal_streamk_bf16_bf16_bf16_mk_nk_mn_comp_default_instances(
std::vector<std::unique_ptr< std::vector<std::unique_ptr<
DeviceGemmV2<Row, Col, Row, BF16, BF16, BF16, PassThrough, PassThrough, PassThrough>>>& DeviceGemm_Streamk_V2<Row, Col, Row, BF16, BF16, BF16, PassThrough, PassThrough, PassThrough>>>&
instances); instances);
void add_device_gemm_xdl_universal_streamk_bf16_bf16_bf16_mk_nk_mn_comp_kpadding_instances( void add_device_gemm_xdl_universal_streamk_bf16_bf16_bf16_mk_nk_mn_comp_kpadding_instances(
std::vector<std::unique_ptr< std::vector<std::unique_ptr<
DeviceGemmV2<Row, Col, Row, BF16, BF16, BF16, PassThrough, PassThrough, PassThrough>>>& DeviceGemm_Streamk_V2<Row, Col, Row, BF16, BF16, BF16, PassThrough, PassThrough, PassThrough>>>&
instances); instances);
void add_device_gemm_xdl_universal_streamk_bf16_bf16_bf16_mk_nk_mn_mem_v1_default_instances( void add_device_gemm_xdl_universal_streamk_bf16_bf16_bf16_mk_nk_mn_mem_v1_default_instances(
std::vector<std::unique_ptr< std::vector<std::unique_ptr<
DeviceGemmV2<Row, Col, Row, BF16, BF16, BF16, PassThrough, PassThrough, PassThrough>>>& DeviceGemm_Streamk_V2<Row, Col, Row, BF16, BF16, BF16, PassThrough, PassThrough, PassThrough>>>&
instances); instances);
void add_device_gemm_xdl_universal_streamk_bf16_bf16_bf16_mk_nk_mn_mem_v1_kpadding_instances( void add_device_gemm_xdl_universal_streamk_bf16_bf16_bf16_mk_nk_mn_mem_v1_kpadding_instances(
std::vector<std::unique_ptr< std::vector<std::unique_ptr<
DeviceGemmV2<Row, Col, Row, BF16, BF16, BF16, PassThrough, PassThrough, PassThrough>>>& DeviceGemm_Streamk_V2<Row, Col, Row, BF16, BF16, BF16, PassThrough, PassThrough, PassThrough>>>&
instances); instances);
void add_device_gemm_xdl_universal_streamk_bf16_bf16_bf16_mk_nk_mn_mem_v2_default_instances( void add_device_gemm_xdl_universal_streamk_bf16_bf16_bf16_mk_nk_mn_mem_v2_default_instances(
std::vector<std::unique_ptr< std::vector<std::unique_ptr<
DeviceGemmV2<Row, Col, Row, BF16, BF16, BF16, PassThrough, PassThrough, PassThrough>>>& DeviceGemm_Streamk_V2<Row, Col, Row, BF16, BF16, BF16, PassThrough, PassThrough, PassThrough>>>&
instances); instances);
void add_device_gemm_xdl_universal_streamk_bf16_bf16_bf16_mk_nk_mn_mem_v2_kpadding_instances( void add_device_gemm_xdl_universal_streamk_bf16_bf16_bf16_mk_nk_mn_mem_v2_kpadding_instances(
std::vector<std::unique_ptr< std::vector<std::unique_ptr<
DeviceGemmV2<Row, Col, Row, BF16, BF16, BF16, PassThrough, PassThrough, PassThrough>>>& DeviceGemm_Streamk_V2<Row, Col, Row, BF16, BF16, BF16, PassThrough, PassThrough, PassThrough>>>&
instances); instances);
void add_device_gemm_xdl_universal_streamk_bf16_bf16_bf16_km_kn_mn_comp_default_instances( void add_device_gemm_xdl_universal_streamk_bf16_bf16_bf16_km_kn_mn_comp_default_instances(
std::vector<std::unique_ptr< std::vector<std::unique_ptr<
DeviceGemmV2<Col, Row, Row, BF16, BF16, BF16, PassThrough, PassThrough, PassThrough>>>& DeviceGemm_Streamk_V2<Col, Row, Row, BF16, BF16, BF16, PassThrough, PassThrough, PassThrough>>>&
instances); instances);
void add_device_gemm_xdl_universal_streamk_bf16_bf16_bf16_km_kn_mn_comp_kpadding_instances( void add_device_gemm_xdl_universal_streamk_bf16_bf16_bf16_km_kn_mn_comp_kpadding_instances(
std::vector<std::unique_ptr< std::vector<std::unique_ptr<
DeviceGemmV2<Col, Row, Row, BF16, BF16, BF16, PassThrough, PassThrough, PassThrough>>>& DeviceGemm_Streamk_V2<Col, Row, Row, BF16, BF16, BF16, PassThrough, PassThrough, PassThrough>>>&
instances); instances);
void add_device_gemm_xdl_universal_streamk_bf16_bf16_bf16_km_kn_mn_comp_mnpadding_instances( void add_device_gemm_xdl_universal_streamk_bf16_bf16_bf16_km_kn_mn_comp_mnpadding_instances(
std::vector<std::unique_ptr< std::vector<std::unique_ptr<
DeviceGemmV2<Col, Row, Row, BF16, BF16, BF16, PassThrough, PassThrough, PassThrough>>>& DeviceGemm_Streamk_V2<Col, Row, Row, BF16, BF16, BF16, PassThrough, PassThrough, PassThrough>>>&
instances); instances);
void add_device_gemm_xdl_universal_streamk_bf16_bf16_bf16_km_kn_mn_comp_mnkpadding_instances( void add_device_gemm_xdl_universal_streamk_bf16_bf16_bf16_km_kn_mn_comp_mnkpadding_instances(
std::vector<std::unique_ptr< std::vector<std::unique_ptr<
DeviceGemmV2<Col, Row, Row, BF16, BF16, BF16, PassThrough, PassThrough, PassThrough>>>& DeviceGemm_Streamk_V2<Col, Row, Row, BF16, BF16, BF16, PassThrough, PassThrough, PassThrough>>>&
instances); instances);
void add_device_gemm_xdl_universal_streamk_bf16_bf16_bf16_km_kn_mn_mem_v1_default_instances( void add_device_gemm_xdl_universal_streamk_bf16_bf16_bf16_km_kn_mn_mem_v1_default_instances(
std::vector<std::unique_ptr< std::vector<std::unique_ptr<
DeviceGemmV2<Col, Row, Row, BF16, BF16, BF16, PassThrough, PassThrough, PassThrough>>>& DeviceGemm_Streamk_V2<Col, Row, Row, BF16, BF16, BF16, PassThrough, PassThrough, PassThrough>>>&
instances); instances);
void add_device_gemm_xdl_universal_streamk_bf16_bf16_bf16_km_kn_mn_mem_v1_kpadding_instances( void add_device_gemm_xdl_universal_streamk_bf16_bf16_bf16_km_kn_mn_mem_v1_kpadding_instances(
std::vector<std::unique_ptr< std::vector<std::unique_ptr<
DeviceGemmV2<Col, Row, Row, BF16, BF16, BF16, PassThrough, PassThrough, PassThrough>>>& DeviceGemm_Streamk_V2<Col, Row, Row, BF16, BF16, BF16, PassThrough, PassThrough, PassThrough>>>&
instances); instances);
void add_device_gemm_xdl_universal_streamk_bf16_bf16_bf16_km_kn_mn_mem_v1_mnkpadding_instances( void add_device_gemm_xdl_universal_streamk_bf16_bf16_bf16_km_kn_mn_mem_v1_mnkpadding_instances(
std::vector<std::unique_ptr< std::vector<std::unique_ptr<
DeviceGemmV2<Col, Row, Row, BF16, BF16, BF16, PassThrough, PassThrough, PassThrough>>>& DeviceGemm_Streamk_V2<Col, Row, Row, BF16, BF16, BF16, PassThrough, PassThrough, PassThrough>>>&
instances); instances);
void add_device_gemm_xdl_universal_streamk_bf16_bf16_bf16_km_kn_mn_mem_v2_default_instances( void add_device_gemm_xdl_universal_streamk_bf16_bf16_bf16_km_kn_mn_mem_v2_default_instances(
std::vector<std::unique_ptr< std::vector<std::unique_ptr<
DeviceGemmV2<Col, Row, Row, BF16, BF16, BF16, PassThrough, PassThrough, PassThrough>>>& DeviceGemm_Streamk_V2<Col, Row, Row, BF16, BF16, BF16, PassThrough, PassThrough, PassThrough>>>&
instances); instances);
void add_device_gemm_xdl_universal_streamk_bf16_bf16_bf16_km_kn_mn_mem_v2_kpadding_instances( void add_device_gemm_xdl_universal_streamk_bf16_bf16_bf16_km_kn_mn_mem_v2_kpadding_instances(
std::vector<std::unique_ptr< std::vector<std::unique_ptr<
DeviceGemmV2<Col, Row, Row, BF16, BF16, BF16, PassThrough, PassThrough, PassThrough>>>& DeviceGemm_Streamk_V2<Col, Row, Row, BF16, BF16, BF16, PassThrough, PassThrough, PassThrough>>>&
instances); instances);
void add_device_gemm_xdl_universal_streamk_bf16_bf16_bf16_km_kn_mn_mem_v2_mnkpadding_instances( void add_device_gemm_xdl_universal_streamk_bf16_bf16_bf16_km_kn_mn_mem_v2_mnkpadding_instances(
std::vector<std::unique_ptr< std::vector<std::unique_ptr<
DeviceGemmV2<Col, Row, Row, BF16, BF16, BF16, PassThrough, PassThrough, PassThrough>>>& DeviceGemm_Streamk_V2<Col, Row, Row, BF16, BF16, BF16, PassThrough, PassThrough, PassThrough>>>&
instances); instances);
void add_device_gemm_xdl_universal_streamk_bf16_bf16_bf16_km_nk_mn_comp_default_instances( void add_device_gemm_xdl_universal_streamk_bf16_bf16_bf16_km_nk_mn_comp_default_instances(
std::vector<std::unique_ptr< std::vector<std::unique_ptr<
DeviceGemmV2<Col, Col, Row, BF16, BF16, BF16, PassThrough, PassThrough, PassThrough>>>& DeviceGemm_Streamk_V2<Col, Col, Row, BF16, BF16, BF16, PassThrough, PassThrough, PassThrough>>>&
instances); instances);
void add_device_gemm_xdl_universal_streamk_bf16_bf16_bf16_km_nk_mn_comp_kpadding_instances( void add_device_gemm_xdl_universal_streamk_bf16_bf16_bf16_km_nk_mn_comp_kpadding_instances(
std::vector<std::unique_ptr< std::vector<std::unique_ptr<
DeviceGemmV2<Col, Col, Row, BF16, BF16, BF16, PassThrough, PassThrough, PassThrough>>>& DeviceGemm_Streamk_V2<Col, Col, Row, BF16, BF16, BF16, PassThrough, PassThrough, PassThrough>>>&
instances); instances);
void add_device_gemm_xdl_universal_streamk_bf16_bf16_bf16_km_nk_mn_comp_mpadding_instances( void add_device_gemm_xdl_universal_streamk_bf16_bf16_bf16_km_nk_mn_comp_mpadding_instances(
std::vector<std::unique_ptr< std::vector<std::unique_ptr<
DeviceGemmV2<Col, Col, Row, BF16, BF16, BF16, PassThrough, PassThrough, PassThrough>>>& DeviceGemm_Streamk_V2<Col, Col, Row, BF16, BF16, BF16, PassThrough, PassThrough, PassThrough>>>&
instances); instances);
void add_device_gemm_xdl_universal_streamk_bf16_bf16_bf16_km_nk_mn_comp_mkpadding_instances( void add_device_gemm_xdl_universal_streamk_bf16_bf16_bf16_km_nk_mn_comp_mkpadding_instances(
std::vector<std::unique_ptr< std::vector<std::unique_ptr<
DeviceGemmV2<Col, Col, Row, BF16, BF16, BF16, PassThrough, PassThrough, PassThrough>>>& DeviceGemm_Streamk_V2<Col, Col, Row, BF16, BF16, BF16, PassThrough, PassThrough, PassThrough>>>&
instances); instances);
void add_device_gemm_xdl_universal_streamk_bf16_bf16_bf16_km_nk_mn_mem_v1_default_instances( void add_device_gemm_xdl_universal_streamk_bf16_bf16_bf16_km_nk_mn_mem_v1_default_instances(
std::vector<std::unique_ptr< std::vector<std::unique_ptr<
DeviceGemmV2<Col, Col, Row, BF16, BF16, BF16, PassThrough, PassThrough, PassThrough>>>& DeviceGemm_Streamk_V2<Col, Col, Row, BF16, BF16, BF16, PassThrough, PassThrough, PassThrough>>>&
instances); instances);
void add_device_gemm_xdl_universal_streamk_bf16_bf16_bf16_km_nk_mn_mem_v1_kpadding_instances( void add_device_gemm_xdl_universal_streamk_bf16_bf16_bf16_km_nk_mn_mem_v1_kpadding_instances(
std::vector<std::unique_ptr< std::vector<std::unique_ptr<
DeviceGemmV2<Col, Col, Row, BF16, BF16, BF16, PassThrough, PassThrough, PassThrough>>>& DeviceGemm_Streamk_V2<Col, Col, Row, BF16, BF16, BF16, PassThrough, PassThrough, PassThrough>>>&
instances); instances);
void add_device_gemm_xdl_universal_streamk_bf16_bf16_bf16_km_nk_mn_mem_v1_mkpadding_instances( void add_device_gemm_xdl_universal_streamk_bf16_bf16_bf16_km_nk_mn_mem_v1_mkpadding_instances(
std::vector<std::unique_ptr< std::vector<std::unique_ptr<
DeviceGemmV2<Col, Col, Row, BF16, BF16, BF16, PassThrough, PassThrough, PassThrough>>>& DeviceGemm_Streamk_V2<Col, Col, Row, BF16, BF16, BF16, PassThrough, PassThrough, PassThrough>>>&
instances); instances);
void add_device_gemm_xdl_universal_streamk_bf16_bf16_bf16_km_nk_mn_mem_v2_default_instances( void add_device_gemm_xdl_universal_streamk_bf16_bf16_bf16_km_nk_mn_mem_v2_default_instances(
std::vector<std::unique_ptr< std::vector<std::unique_ptr<
DeviceGemmV2<Col, Col, Row, BF16, BF16, BF16, PassThrough, PassThrough, PassThrough>>>& DeviceGemm_Streamk_V2<Col, Col, Row, BF16, BF16, BF16, PassThrough, PassThrough, PassThrough>>>&
instances); instances);
void add_device_gemm_xdl_universal_streamk_bf16_bf16_bf16_km_nk_mn_mem_v2_kpadding_instances( void add_device_gemm_xdl_universal_streamk_bf16_bf16_bf16_km_nk_mn_mem_v2_kpadding_instances(
std::vector<std::unique_ptr< std::vector<std::unique_ptr<
DeviceGemmV2<Col, Col, Row, BF16, BF16, BF16, PassThrough, PassThrough, PassThrough>>>& DeviceGemm_Streamk_V2<Col, Col, Row, BF16, BF16, BF16, PassThrough, PassThrough, PassThrough>>>&
instances); instances);
void add_device_gemm_xdl_universal_streamk_bf16_bf16_bf16_km_nk_mn_mem_v2_mkpadding_instances( void add_device_gemm_xdl_universal_streamk_bf16_bf16_bf16_km_nk_mn_mem_v2_mkpadding_instances(
std::vector<std::unique_ptr< std::vector<std::unique_ptr<
DeviceGemmV2<Col, Col, Row, BF16, BF16, BF16, PassThrough, PassThrough, PassThrough>>>& DeviceGemm_Streamk_V2<Col, Col, Row, BF16, BF16, BF16, PassThrough, PassThrough, PassThrough>>>&
instances); instances);
#endif #endif
// Emin @Added
#if(defined(CK_ENABLE_FP8)) #if(defined(CK_ENABLE_FP8))
void add_device_gemm_xdl_universal_streamk_f16_f8_f16_mk_kn_mn_comp_default_instances( void add_device_gemm_xdl_universal_streamk_f16_f8_f16_mk_kn_mn_comp_default_instances(
......
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment