Skip to content

GitLab

Commit d37ba7ea authored by Adam Osewski's avatar Adam Osewski
Browse files

Limit the # of compiled instances. Add instances with different 

ScalarPerVector.
parent e3e38fe3
Show whitespace changes
Inline Side-by-side
Showing with 816 additions and 519 deletions
example/01_gemm/gemm_xdl_direct_c_write_out_fp16.cpp
View file @ d37ba7ea
...@@ -15,6 +15,7 @@ using CShuffleDataType = float; ...@@ -15,6 +15,7 @@ using CShuffleDataType = float;
using CDataType = ck::half_t; using CDataType = ck::half_t;
using F16 = ck::half_t; using F16 = ck::half_t;
using F32 = float;
using ALayout = Row; using ALayout = Row;
using BLayout = Col; using BLayout = Col;
...@@ -25,17 +26,19 @@ using BElementOp = PassThrough; ...@@ -25,17 +26,19 @@ using BElementOp = PassThrough;
using CElementOp = PassThrough; using CElementOp = PassThrough;
static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default; static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default;
static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
static constexpr auto LoopSchedDefault = ck::LoopScheduler::Default; static constexpr auto LoopSchedDefault = ck::LoopScheduler::Default;
static constexpr auto GemmPipeline = ck::PipelineVersion::v1; static constexpr auto GemmPipeline = ck::PipelineVersion::v1;
using DeviceGemmDirectCWriteOut = ck::tensor_operation::device::DeviceGemm_Xdl_DirectCWriteOut using DeviceGemmDirectCWriteOut = ck::tensor_operation::device::DeviceGemm_Xdl_DirectCWriteOut
// clang-format off // clang-format off
// ######| ALayout| BLayout| CLayout| AData| BData| CData| AccData| A| B| C| GEMM| NumGemmK| Block| MPer| NPer| KPer| AK1| BK1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| LoopScheduler| PipelineVersion| // ######| ALayout| BLayout| CLayout| AData| BData| CData| AccData| A| B| C| GEMM| NumGemmK| Block| MPer| NPer| KPer| AK1| BK1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer| CThreadTransfer| LoopScheduler| PipelineVersion|
// ######| | | | Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Spacialization| Prefetch| Size| Block| Block| Block| | | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| | | // ######| | | | Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Spacialization| Prefetch| Size| Block| Block| Block| | | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| DstAccessOrder| DstVectorDim| DstScalarPerVector| | |
// ######| | | | | | | | Operation| Operation| Operation| | Stage| | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | | // ######| | | | | | | | Operation| Operation| Operation| | Stage| | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | | | | |
// ######| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | // ######| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
< ALayout, BLayout, CLayout, ADataType, BDataType, CDataType, AccDataType, AElementOp, BElementOp, CElementOp, GemmDefault, 1, 256, 256, 128, 32, 8, 8, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, LoopSchedDefault, GemmPipeline>; // < ALayout, BLayout, CLayout, ADataType, BDataType, CDataType, AccDataType, AElementOp, BElementOp, CElementOp, GemmDefault, 1, 256, 256, 128, 32, 8, 8, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<0, 1, 2, 3, 4, 5, 6, 7>, 7, 8, LoopSchedDefault, GemmPipeline>;
// < ALayout, BLayout, CLayout, ADataType, BDataType, CDataType, AccDataType, AElementOp, BElementOp, CElementOp, GemmDefault, 1, 64, 32, 32, 32, 8, 8, 32, 32, 1, 1, S<2, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<2, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, LoopSchedDefault, GemmPipeline>; < Row, Col, Row, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 256, 128, 128, 32, 8, 8, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<0, 1, 2, 3, 4, 5, 6, 7>, 7, 8, LoopSchedDefault, GemmPipeline>;
// < ALayout, BLayout, CLayout, ADataType, BDataType, CDataType, AccDataType, AElementOp, BElementOp, CElementOp, GemmDefault, 1, 64, 32, 32, 32, 8, 8, 32, 32, 1, 1, S<2, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<2, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<0, 1, 2, 3, 4, 5, 6, 7>, 7, 8, LoopSchedDefault, GemmPipeline>;
// clang-format on // clang-format on
// clang-format off // clang-format off
......
example/15_grouped_gemm/grouped_gemm_xdl_splitk_direct_c_write_out_fp16.cpp
View file @ d37ba7ea
...@@ -56,7 +56,7 @@ using DeviceGemmInstance = ck::tensor_operation::device::DeviceGroupedGemmXdlSpl ...@@ -56,7 +56,7 @@ using DeviceGemmInstance = ck::tensor_operation::device::DeviceGroupedGemmXdlSpl
//######| | | | | Type| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Spacialization| Prefetch| Size| Block| Block| Block| | | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| | | //######| | | | | Type| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Spacialization| Prefetch| Size| Block| Block| Block| | | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| | |
//######| | | | | | | | | | Operation| Operation| Operation| | Stage| | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | | //######| | | | | | | | | | Operation| Operation| Operation| | Stage| | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | |
//######| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | //######| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
< ALayout, BLayout, DsLayout, ELayout, ADataType, BDataType, AccDataType, DsDataType, EDataType, AElementOp, BElementOp, CDEElementOp, GemmDefault, 1, 256, 256, 128, 32, 8, 8, 32, 32, 4, 2, S<1, 4, 64, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, S<1, 4, 64, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, LoopSchedDefault, GemmPipeline>; < ALayout, BLayout, DsLayout, ELayout, ADataType, BDataType, AccDataType, DsDataType, EDataType, AElementOp, BElementOp, CDEElementOp, GemmDefault, 1, 256, 128, 256, 32, 8, 8, 32, 32, 2, 4, S<1, 4, 64, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, S<1, 4, 64, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, LoopSchedDefault, GemmPipeline>;
// clang-format on // clang-format on
#include "run_grouped_gemm_example.inc" #include "run_grouped_gemm_example.inc"
...@@ -66,10 +66,11 @@ int main(int argc, char* argv[]) ...@@ -66,10 +66,11 @@ int main(int argc, char* argv[])
ProblemSize problem_size; ProblemSize problem_size;
ExecutionConfig config; ExecutionConfig config;
problem_size.group_count = 16; problem_size.group_count = 1;
// problem_size.group_count = 16;
problem_size.Ms = { problem_size.Ms = {167};
167, 183, 177, 181, 153, 139, 156, 173, 163, 150, 204, 184, 168, 156, 168, 148}; // 167, 183, 177, 181, 153, 139, 156, 173, 163, 150, 204, 184, 168, 156, 168, 148};
for(int i = 0; i < problem_size.group_count; i++) for(int i = 0; i < problem_size.group_count; i++)
{ {
......
library/include/ck/library/tensor_operation_instance/gpu/gemm.hpp
View file @ d37ba7ea
...@@ -17,253 +17,268 @@ namespace tensor_operation { ...@@ -17,253 +17,268 @@ namespace tensor_operation {
namespace device { namespace device {
namespace instance { namespace instance {
void add_device_gemm_dl_f16_f16_f16_km_kn_mn_instances( // void add_device_gemm_dl_f16_f16_f16_km_kn_mn_instances(
std::vector<std::unique_ptr< // std::vector<std::unique_ptr<
DeviceGemm<Col, Row, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>& // DeviceGemm<Col, Row, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
instances); // instances);
void add_device_gemm_dl_f16_f16_f16_km_kn_mn_irregular_instances( // void add_device_gemm_dl_f16_f16_f16_km_kn_mn_irregular_instances(
std::vector<std::unique_ptr< // std::vector<std::unique_ptr<
DeviceGemm<Col, Row, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>& // DeviceGemm<Col, Row, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
instances); // instances);
void add_device_gemm_dl_f16_f16_f16_km_nk_mn_instances( // void add_device_gemm_dl_f16_f16_f16_km_nk_mn_instances(
std::vector<std::unique_ptr< // std::vector<std::unique_ptr<
DeviceGemm<Col, Col, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>& // DeviceGemm<Col, Col, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
instances); // instances);
void add_device_gemm_dl_f16_f16_f16_km_nk_mn_irregular_instances( // void add_device_gemm_dl_f16_f16_f16_km_nk_mn_irregular_instances(
std::vector<std::unique_ptr< // std::vector<std::unique_ptr<
DeviceGemm<Col, Col, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>& // DeviceGemm<Col, Col, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
instances); // instances);
void add_device_gemm_dl_f16_f16_f16_mk_kn_mn_instances( // void add_device_gemm_dl_f16_f16_f16_mk_kn_mn_instances(
std::vector<std::unique_ptr< // std::vector<std::unique_ptr<
DeviceGemm<Row, Row, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>& // DeviceGemm<Row, Row, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
instances); // instances);
void add_device_gemm_dl_f16_f16_f16_mk_kn_mn_irregular_instances( // void add_device_gemm_dl_f16_f16_f16_mk_kn_mn_irregular_instances(
std::vector<std::unique_ptr< // std::vector<std::unique_ptr<
DeviceGemm<Row, Row, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>& // DeviceGemm<Row, Row, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
instances); // instances);
void add_device_gemm_dl_f16_f16_f16_mk_nk_mn_instances( // void add_device_gemm_dl_f16_f16_f16_mk_nk_mn_instances(
std::vector<std::unique_ptr< // std::vector<std::unique_ptr<
DeviceGemm<Row, Col, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>& // DeviceGemm<Row, Col, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
instances); // instances);
// void add_device_gemm_dl_f16_f16_f16_mk_nk_mn_irregular_instances(
// std::vector<std::unique_ptr<
// DeviceGemm<Row, Col, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
// instances);
// void add_device_gemm_dl_f32_f32_f32_km_kn_mn_instances(
// std::vector<std::unique_ptr<
// DeviceGemm<Col, Row, Row, F32, F32, F32, PassThrough, PassThrough, PassThrough>>>&
// instances);
// void add_device_gemm_dl_f32_f32_f32_km_nk_mn_instances(
// std::vector<std::unique_ptr<
// DeviceGemm<Col, Col, Row, F32, F32, F32, PassThrough, PassThrough, PassThrough>>>&
// instances);
// void add_device_gemm_dl_f32_f32_f32_mk_kn_mn_instances(
// std::vector<std::unique_ptr<
// DeviceGemm<Row, Row, Row, F32, F32, F32, PassThrough, PassThrough, PassThrough>>>&
// instances);
// void add_device_gemm_dl_f32_f32_f32_mk_nk_mn_instances(
// std::vector<std::unique_ptr<
// DeviceGemm<Row, Col, Row, F32, F32, F32, PassThrough, PassThrough, PassThrough>>>&
// instances);
// void add_device_gemm_dl_i8_i8_i8_km_kn_mn_instances(
// std::vector<std::unique_ptr<
// DeviceGemm<Col, Row, Row, int8_t, int8_t, int8_t, PassThrough, PassThrough,
// PassThrough>>>& instances);
// void add_device_gemm_dl_i8_i8_i8_km_kn_mn_irregular_instances(
// std::vector<std::unique_ptr<
// DeviceGemm<Col, Row, Row, int8_t, int8_t, int8_t, PassThrough, PassThrough,
// PassThrough>>>& instances);
// void add_device_gemm_dl_i8_i8_i8_km_nk_mn_instances(
// std::vector<std::unique_ptr<
// DeviceGemm<Col, Col, Row, int8_t, int8_t, int8_t, PassThrough, PassThrough,
// PassThrough>>>& instances);
// void add_device_gemm_dl_i8_i8_i8_km_nk_mn_irregular_instances(
// std::vector<std::unique_ptr<
// DeviceGemm<Col, Col, Row, int8_t, int8_t, int8_t, PassThrough, PassThrough,
// PassThrough>>>& instances);
// void add_device_gemm_dl_i8_i8_i8_mk_kn_mn_instances(
// std::vector<std::unique_ptr<
// DeviceGemm<Row, Row, Row, int8_t, int8_t, int8_t, PassThrough, PassThrough,
// PassThrough>>>& instances);
// void add_device_gemm_dl_i8_i8_i8_mk_kn_mn_irregular_instances(
// std::vector<std::unique_ptr<
// DeviceGemm<Row, Row, Row, int8_t, int8_t, int8_t, PassThrough, PassThrough,
// PassThrough>>>& instances);
// void add_device_gemm_dl_i8_i8_i8_mk_nk_mn_instances(
// std::vector<std::unique_ptr<
// DeviceGemm<Row, Col, Row, int8_t, int8_t, int8_t, PassThrough, PassThrough,
// PassThrough>>>& instances);
// void add_device_gemm_dl_i8_i8_i8_mk_nk_mn_irregular_instances(
// std::vector<std::unique_ptr<
// DeviceGemm<Row, Col, Row, int8_t, int8_t, int8_t, PassThrough, PassThrough,
// PassThrough>>>& instances);
// void add_device_gemm_xdl_c_shuffle_2_stage_f16_f16_f16_mk_nk_mn_instances(
// std::vector<std::unique_ptr<
// DeviceGemm<Row, Col, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
// instances);
// void add_device_gemm_xdl_c_shuffle_bf16_bf16_bf16_km_kn_mn_instances(
// std::vector<std::unique_ptr<
// DeviceGemm<Col, Row, Row, BF16, BF16, BF16, PassThrough, PassThrough, PassThrough>>>&
// instances);
// void add_device_gemm_xdl_c_shuffle_bf16_bf16_bf16_km_nk_mn_instances(
// std::vector<std::unique_ptr<
// DeviceGemm<Col, Col, Row, BF16, BF16, BF16, PassThrough, PassThrough, PassThrough>>>&
// instances);
// void add_device_gemm_xdl_c_shuffle_bf16_bf16_bf16_mk_kn_mn_instances(
// std::vector<std::unique_ptr<
// DeviceGemm<Row, Row, Row, BF16, BF16, BF16, PassThrough, PassThrough, PassThrough>>>&
// instances);
// void add_device_gemm_xdl_c_shuffle_bf16_bf16_bf16_mk_nk_mn_instances(
// std::vector<std::unique_ptr<
// DeviceGemm<Row, Col, Row, BF16, BF16, BF16, PassThrough, PassThrough, PassThrough>>>&
// instances);
// void add_device_gemm_xdl_c_shuffle_f16_f16_f16_km_kn_mn_instances(
// std::vector<std::unique_ptr<
// DeviceGemm<Col, Row, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
// instances);
// void add_device_gemm_xdl_c_shuffle_f16_f16_f16_km_nk_mn_instances(
// std::vector<std::unique_ptr<
// DeviceGemm<Col, Col, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
// instances);
// void add_device_gemm_xdl_c_shuffle_f16_f16_f16_mk_kn_mn_instances(
// std::vector<std::unique_ptr<
// DeviceGemm<Row, Row, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
// instances);
void add_device_gemm_dl_f16_f16_f16_mk_nk_mn_irregular_instances( void add_device_gemm_xdl_c_shuffle_f16_f16_f16_mk_nk_mn_instances(
std::vector<std::unique_ptr< std::vector<std::unique_ptr<
DeviceGemm<Row, Col, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>& DeviceGemm<Row, Col, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
instances); instances);
void add_device_gemm_dl_f32_f32_f32_km_kn_mn_instances( // void add_device_gemm_xdl_c_shuffle_f32_f32_f32_km_kn_mn_instances(
std::vector<std::unique_ptr< // std::vector<std::unique_ptr<
DeviceGemm<Col, Row, Row, F32, F32, F32, PassThrough, PassThrough, PassThrough>>>& // DeviceGemm<Col, Row, Row, F32, F32, F32, PassThrough, PassThrough, PassThrough>>>&
// instances);
instances);
// void add_device_gemm_xdl_c_shuffle_f32_f32_f32_km_nk_mn_instances(
void add_device_gemm_dl_f32_f32_f32_km_nk_mn_instances( // std::vector<std::unique_ptr<
std::vector<std::unique_ptr< // DeviceGemm<Col, Col, Row, F32, F32, F32, PassThrough, PassThrough, PassThrough>>>&
DeviceGemm<Col, Col, Row, F32, F32, F32, PassThrough, PassThrough, PassThrough>>>& // instances);
instances);
// void add_device_gemm_xdl_c_shuffle_f32_f32_f32_mk_kn_mn_instances(
void add_device_gemm_dl_f32_f32_f32_mk_kn_mn_instances( // std::vector<std::unique_ptr<
std::vector<std::unique_ptr< // DeviceGemm<Row, Row, Row, F32, F32, F32, PassThrough, PassThrough, PassThrough>>>&
DeviceGemm<Row, Row, Row, F32, F32, F32, PassThrough, PassThrough, PassThrough>>>& // instances);
instances);
// void add_device_gemm_xdl_c_shuffle_f32_f32_f32_mk_nk_mn_instances(
void add_device_gemm_dl_f32_f32_f32_mk_nk_mn_instances( // std::vector<std::unique_ptr<
std::vector<std::unique_ptr< // DeviceGemm<Row, Col, Row, F32, F32, F32, PassThrough, PassThrough, PassThrough>>>&
DeviceGemm<Row, Col, Row, F32, F32, F32, PassThrough, PassThrough, PassThrough>>>& // instances);
instances);
// void add_device_gemm_xdl_c_shuffle_i8_i8_i8_km_kn_mn_instances(
void add_device_gemm_dl_i8_i8_i8_km_kn_mn_instances( // std::vector<std::unique_ptr<
std::vector<std::unique_ptr< // DeviceGemm<Col, Row, Row, int8_t, int8_t, int8_t, PassThrough, PassThrough,
DeviceGemm<Col, Row, Row, int8_t, int8_t, int8_t, PassThrough, PassThrough, PassThrough>>>& // PassThrough>>>& instances);
instances);
// void add_device_gemm_xdl_c_shuffle_i8_i8_i8_km_nk_mn_instances(
void add_device_gemm_dl_i8_i8_i8_km_kn_mn_irregular_instances( // std::vector<std::unique_ptr<
std::vector<std::unique_ptr< // DeviceGemm<Col, Col, Row, int8_t, int8_t, int8_t, PassThrough, PassThrough,
DeviceGemm<Col, Row, Row, int8_t, int8_t, int8_t, PassThrough, PassThrough, PassThrough>>>& // PassThrough>>>& instances);
instances);
// void add_device_gemm_xdl_c_shuffle_i8_i8_i8_mk_kn_mn_instances(
void add_device_gemm_dl_i8_i8_i8_km_nk_mn_instances( // std::vector<std::unique_ptr<
std::vector<std::unique_ptr< // DeviceGemm<Row, Row, Row, int8_t, int8_t, int8_t, PassThrough, PassThrough,
DeviceGemm<Col, Col, Row, int8_t, int8_t, int8_t, PassThrough, PassThrough, PassThrough>>>& // PassThrough>>>& instances);
instances);
// void add_device_gemm_xdl_c_shuffle_i8_i8_i8_mk_nk_mn_instances(
void add_device_gemm_dl_i8_i8_i8_km_nk_mn_irregular_instances( // std::vector<std::unique_ptr<
std::vector<std::unique_ptr< // DeviceGemm<Row, Col, Row, int8_t, int8_t, int8_t, PassThrough, PassThrough,
DeviceGemm<Col, Col, Row, int8_t, int8_t, int8_t, PassThrough, PassThrough, PassThrough>>>& // PassThrough>>>& instances);
instances);
// void add_device_gemm_xdl_f16_f16_f16_km_kn_mn_instances(
void add_device_gemm_dl_i8_i8_i8_mk_kn_mn_instances( // std::vector<std::unique_ptr<
std::vector<std::unique_ptr< // DeviceGemm<Col, Row, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
DeviceGemm<Row, Row, Row, int8_t, int8_t, int8_t, PassThrough, PassThrough, PassThrough>>>& // instances);
instances);
// void add_device_gemm_xdl_f16_f16_f16_km_nk_mn_instances(
void add_device_gemm_dl_i8_i8_i8_mk_kn_mn_irregular_instances( // std::vector<std::unique_ptr<
std::vector<std::unique_ptr< // DeviceGemm<Col, Col, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
DeviceGemm<Row, Row, Row, int8_t, int8_t, int8_t, PassThrough, PassThrough, PassThrough>>>& // instances);
instances);
// void add_device_gemm_xdl_f16_f16_f16_mk_kn_mn_instances(
void add_device_gemm_dl_i8_i8_i8_mk_nk_mn_instances( // std::vector<std::unique_ptr<
std::vector<std::unique_ptr< // DeviceGemm<Row, Row, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
DeviceGemm<Row, Col, Row, int8_t, int8_t, int8_t, PassThrough, PassThrough, PassThrough>>>& // instances);
instances);
// void add_device_gemm_xdl_f16_f16_f16_mk_nk_mn_instances(
void add_device_gemm_dl_i8_i8_i8_mk_nk_mn_irregular_instances( // std::vector<std::unique_ptr<
std::vector<std::unique_ptr< // DeviceGemm<Row, Col, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
DeviceGemm<Row, Col, Row, int8_t, int8_t, int8_t, PassThrough, PassThrough, PassThrough>>>& // instances);
instances);
// void add_device_gemm_xdl_f32_f32_f32_km_kn_mn_instances(
void add_device_gemm_xdl_c_shuffle_2_stage_f16_f16_f16_mk_nk_mn_instances( // std::vector<std::unique_ptr<
// DeviceGemm<Col, Row, Row, F32, F32, F32, PassThrough, PassThrough, PassThrough>>>&
// instances);
// void add_device_gemm_xdl_f32_f32_f32_km_nk_mn_instances(
// std::vector<std::unique_ptr<
// DeviceGemm<Col, Col, Row, F32, F32, F32, PassThrough, PassThrough, PassThrough>>>&
// instances);
// void add_device_gemm_xdl_f32_f32_f32_mk_kn_mn_instances(
// std::vector<std::unique_ptr<
// DeviceGemm<Row, Row, Row, F32, F32, F32, PassThrough, PassThrough, PassThrough>>>&
// instances);
// void add_device_gemm_xdl_f32_f32_f32_mk_nk_mn_instances(
// std::vector<std::unique_ptr<
// DeviceGemm<Row, Col, Row, F32, F32, F32, PassThrough, PassThrough, PassThrough>>>&
// instances);
// void add_device_gemm_xdl_f64_f64_f64_km_kn_mn_instances(
// std::vector<std::unique_ptr<
// DeviceGemm<Col, Row, Row, F64, F64, F64, PassThrough, PassThrough, PassThrough>>>&
// instances);
// void add_device_gemm_xdl_f64_f64_f64_km_nk_mn_instances(
// std::vector<std::unique_ptr<
// DeviceGemm<Col, Col, Row, F64, F64, F64, PassThrough, PassThrough, PassThrough>>>&
// instances);
// void add_device_gemm_xdl_f64_f64_f64_mk_kn_mn_instances(
// std::vector<std::unique_ptr<
// DeviceGemm<Row, Row, Row, F64, F64, F64, PassThrough, PassThrough, PassThrough>>>&
// instances);
// void add_device_gemm_xdl_f64_f64_f64_mk_nk_mn_instances(
// std::vector<std::unique_ptr<
// DeviceGemm<Row, Col, Row, F64, F64, F64, PassThrough, PassThrough, PassThrough>>>&
// instances);
void add_device_gemm_xdl_direct_c_write_out_f16_f16_f16_mk_nk_mn_instances_spv8(
std::vector<std::unique_ptr< std::vector<std::unique_ptr<
DeviceGemm<Row, Col, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>& DeviceGemm<Row, Col, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
instances); instances);
void add_device_gemm_xdl_c_shuffle_bf16_bf16_bf16_km_kn_mn_instances( void add_device_gemm_xdl_direct_c_write_out_f16_f16_f16_mk_nk_mn_instances_spv4(
std::vector<std::unique_ptr<
DeviceGemm<Col, Row, Row, BF16, BF16, BF16, PassThrough, PassThrough, PassThrough>>>&
instances);
void add_device_gemm_xdl_c_shuffle_bf16_bf16_bf16_km_nk_mn_instances(
std::vector<std::unique_ptr<
DeviceGemm<Col, Col, Row, BF16, BF16, BF16, PassThrough, PassThrough, PassThrough>>>&
instances);
void add_device_gemm_xdl_c_shuffle_bf16_bf16_bf16_mk_kn_mn_instances(
std::vector<std::unique_ptr<
DeviceGemm<Row, Row, Row, BF16, BF16, BF16, PassThrough, PassThrough, PassThrough>>>&
instances);
void add_device_gemm_xdl_c_shuffle_bf16_bf16_bf16_mk_nk_mn_instances(
std::vector<std::unique_ptr<
DeviceGemm<Row, Col, Row, BF16, BF16, BF16, PassThrough, PassThrough, PassThrough>>>&
instances);
void add_device_gemm_xdl_c_shuffle_f16_f16_f16_km_kn_mn_instances(
std::vector<std::unique_ptr<
DeviceGemm<Col, Row, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
instances);
void add_device_gemm_xdl_c_shuffle_f16_f16_f16_km_nk_mn_instances(
std::vector<std::unique_ptr<
DeviceGemm<Col, Col, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
instances);
void add_device_gemm_xdl_c_shuffle_f16_f16_f16_mk_kn_mn_instances(
std::vector<std::unique_ptr<
DeviceGemm<Row, Row, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
instances);
void add_device_gemm_xdl_c_shuffle_f16_f16_f16_mk_nk_mn_instances(
std::vector<std::unique_ptr< std::vector<std::unique_ptr<
DeviceGemm<Row, Col, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>& DeviceGemm<Row, Col, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
instances); instances);
void add_device_gemm_xdl_c_shuffle_f32_f32_f32_km_kn_mn_instances( void add_device_gemm_xdl_direct_c_write_out_f16_f16_f16_mk_nk_mn_instances_spv2(
std::vector<std::unique_ptr<
DeviceGemm<Col, Row, Row, F32, F32, F32, PassThrough, PassThrough, PassThrough>>>&
instances);
void add_device_gemm_xdl_c_shuffle_f32_f32_f32_km_nk_mn_instances(
std::vector<std::unique_ptr<
DeviceGemm<Col, Col, Row, F32, F32, F32, PassThrough, PassThrough, PassThrough>>>&
instances);
void add_device_gemm_xdl_c_shuffle_f32_f32_f32_mk_kn_mn_instances(
std::vector<std::unique_ptr<
DeviceGemm<Row, Row, Row, F32, F32, F32, PassThrough, PassThrough, PassThrough>>>&
instances);
void add_device_gemm_xdl_c_shuffle_f32_f32_f32_mk_nk_mn_instances(
std::vector<std::unique_ptr<
DeviceGemm<Row, Col, Row, F32, F32, F32, PassThrough, PassThrough, PassThrough>>>&
instances);
void add_device_gemm_xdl_c_shuffle_i8_i8_i8_km_kn_mn_instances(
std::vector<std::unique_ptr<
DeviceGemm<Col, Row, Row, int8_t, int8_t, int8_t, PassThrough, PassThrough, PassThrough>>>&
instances);
void add_device_gemm_xdl_c_shuffle_i8_i8_i8_km_nk_mn_instances(
std::vector<std::unique_ptr<
DeviceGemm<Col, Col, Row, int8_t, int8_t, int8_t, PassThrough, PassThrough, PassThrough>>>&
instances);
void add_device_gemm_xdl_c_shuffle_i8_i8_i8_mk_kn_mn_instances(
std::vector<std::unique_ptr<
DeviceGemm<Row, Row, Row, int8_t, int8_t, int8_t, PassThrough, PassThrough, PassThrough>>>&
instances);
void add_device_gemm_xdl_c_shuffle_i8_i8_i8_mk_nk_mn_instances(
std::vector<std::unique_ptr<
DeviceGemm<Row, Col, Row, int8_t, int8_t, int8_t, PassThrough, PassThrough, PassThrough>>>&
instances);
void add_device_gemm_xdl_f16_f16_f16_km_kn_mn_instances(
std::vector<std::unique_ptr<
DeviceGemm<Col, Row, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
instances);
void add_device_gemm_xdl_f16_f16_f16_km_nk_mn_instances(
std::vector<std::unique_ptr<
DeviceGemm<Col, Col, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
instances);
void add_device_gemm_xdl_f16_f16_f16_mk_kn_mn_instances(
std::vector<std::unique_ptr<
DeviceGemm<Row, Row, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
instances);
void add_device_gemm_xdl_f16_f16_f16_mk_nk_mn_instances(
std::vector<std::unique_ptr< std::vector<std::unique_ptr<
DeviceGemm<Row, Col, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>& DeviceGemm<Row, Col, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
instances); instances);
void add_device_gemm_xdl_f32_f32_f32_km_kn_mn_instances(
std::vector<std::unique_ptr<
DeviceGemm<Col, Row, Row, F32, F32, F32, PassThrough, PassThrough, PassThrough>>>&
instances);
void add_device_gemm_xdl_f32_f32_f32_km_nk_mn_instances(
std::vector<std::unique_ptr<
DeviceGemm<Col, Col, Row, F32, F32, F32, PassThrough, PassThrough, PassThrough>>>&
instances);
void add_device_gemm_xdl_f32_f32_f32_mk_kn_mn_instances(
std::vector<std::unique_ptr<
DeviceGemm<Row, Row, Row, F32, F32, F32, PassThrough, PassThrough, PassThrough>>>&
instances);
void add_device_gemm_xdl_f32_f32_f32_mk_nk_mn_instances(
std::vector<std::unique_ptr<
DeviceGemm<Row, Col, Row, F32, F32, F32, PassThrough, PassThrough, PassThrough>>>&
instances);
void add_device_gemm_xdl_f64_f64_f64_km_kn_mn_instances(
std::vector<std::unique_ptr<
DeviceGemm<Col, Row, Row, F64, F64, F64, PassThrough, PassThrough, PassThrough>>>&
instances);
void add_device_gemm_xdl_f64_f64_f64_km_nk_mn_instances(
std::vector<std::unique_ptr<
DeviceGemm<Col, Col, Row, F64, F64, F64, PassThrough, PassThrough, PassThrough>>>&
instances);
void add_device_gemm_xdl_f64_f64_f64_mk_kn_mn_instances(
std::vector<std::unique_ptr<
DeviceGemm<Row, Row, Row, F64, F64, F64, PassThrough, PassThrough, PassThrough>>>&
instances);
void add_device_gemm_xdl_f64_f64_f64_mk_nk_mn_instances(
std::vector<std::unique_ptr<
DeviceGemm<Row, Col, Row, F64, F64, F64, PassThrough, PassThrough, PassThrough>>>&
instances);
template <typename ALayout, template <typename ALayout,
typename BLayout, typename BLayout,
typename CLayout, typename CLayout,
...@@ -298,128 +313,132 @@ struct DeviceOperationInstanceFactory< ...@@ -298,128 +313,132 @@ struct DeviceOperationInstanceFactory<
if constexpr(is_same_v<ADataType, float> && is_same_v<BDataType, float> && if constexpr(is_same_v<ADataType, float> && is_same_v<BDataType, float> &&
is_same_v<CDataType, float>) is_same_v<CDataType, float>)
{ {
if constexpr(is_same_v<ALayout, Row> && is_same_v<BLayout, Row> && // if constexpr(is_same_v<ALayout, Row> && is_same_v<BLayout, Row> &&
is_same_v<CLayout, Row>) // is_same_v<CLayout, Row>)
{ // {
add_device_gemm_xdl_f32_f32_f32_mk_kn_mn_instances(op_ptrs); // add_device_gemm_xdl_f32_f32_f32_mk_kn_mn_instances(op_ptrs);
add_device_gemm_dl_f32_f32_f32_mk_kn_mn_instances(op_ptrs); // add_device_gemm_dl_f32_f32_f32_mk_kn_mn_instances(op_ptrs);
add_device_gemm_xdl_c_shuffle_f32_f32_f32_mk_kn_mn_instances(op_ptrs); // add_device_gemm_xdl_c_shuffle_f32_f32_f32_mk_kn_mn_instances(op_ptrs);
} // }
else if constexpr(is_same_v<ALayout, Row> && is_same_v<BLayout, Col> && // else if constexpr(is_same_v<ALayout, Row> && is_same_v<BLayout, Col> &&
is_same_v<CLayout, Row>) // is_same_v<CLayout, Row>)
{ // {
add_device_gemm_xdl_f32_f32_f32_mk_nk_mn_instances(op_ptrs); // add_device_gemm_xdl_f32_f32_f32_mk_nk_mn_instances(op_ptrs);
add_device_gemm_dl_f32_f32_f32_mk_nk_mn_instances(op_ptrs); // add_device_gemm_dl_f32_f32_f32_mk_nk_mn_instances(op_ptrs);
add_device_gemm_xdl_c_shuffle_f32_f32_f32_mk_nk_mn_instances(op_ptrs); // add_device_gemm_xdl_c_shuffle_f32_f32_f32_mk_nk_mn_instances(op_ptrs);
} // }
else if constexpr(is_same_v<ALayout, Col> && is_same_v<BLayout, Row> && // else if constexpr(is_same_v<ALayout, Col> && is_same_v<BLayout, Row> &&
is_same_v<CLayout, Row>) // is_same_v<CLayout, Row>)
{ // {
add_device_gemm_xdl_f32_f32_f32_km_kn_mn_instances(op_ptrs); // add_device_gemm_xdl_f32_f32_f32_km_kn_mn_instances(op_ptrs);
add_device_gemm_dl_f32_f32_f32_km_kn_mn_instances(op_ptrs); // add_device_gemm_dl_f32_f32_f32_km_kn_mn_instances(op_ptrs);
add_device_gemm_xdl_c_shuffle_f32_f32_f32_km_kn_mn_instances(op_ptrs); // add_device_gemm_xdl_c_shuffle_f32_f32_f32_km_kn_mn_instances(op_ptrs);
} // }
else if constexpr(is_same_v<ALayout, Col> && is_same_v<BLayout, Col> && // else if constexpr(is_same_v<ALayout, Col> && is_same_v<BLayout, Col> &&
is_same_v<CLayout, Row>) // is_same_v<CLayout, Row>)
{ // {
add_device_gemm_xdl_f32_f32_f32_km_nk_mn_instances(op_ptrs); // add_device_gemm_xdl_f32_f32_f32_km_nk_mn_instances(op_ptrs);
add_device_gemm_dl_f32_f32_f32_km_nk_mn_instances(op_ptrs); // add_device_gemm_dl_f32_f32_f32_km_nk_mn_instances(op_ptrs);
add_device_gemm_xdl_c_shuffle_f32_f32_f32_km_nk_mn_instances(op_ptrs); // add_device_gemm_xdl_c_shuffle_f32_f32_f32_km_nk_mn_instances(op_ptrs);
} // }
} }
else if constexpr(is_same_v<ADataType, half_t> && is_same_v<BDataType, half_t> && else if constexpr(is_same_v<ADataType, half_t> && is_same_v<BDataType, half_t> &&
is_same_v<CDataType, half_t>) is_same_v<CDataType, half_t>)
{ {
if constexpr(is_same_v<ALayout, Row> && is_same_v<BLayout, Row> && // if constexpr(is_same_v<ALayout, Row> && is_same_v<BLayout, Row> &&
is_same_v<CLayout, Row>) // is_same_v<CLayout, Row>)
{ // {
add_device_gemm_xdl_f16_f16_f16_mk_kn_mn_instances(op_ptrs); // add_device_gemm_xdl_f16_f16_f16_mk_kn_mn_instances(op_ptrs);
add_device_gemm_dl_f16_f16_f16_mk_kn_mn_instances(op_ptrs); // add_device_gemm_dl_f16_f16_f16_mk_kn_mn_instances(op_ptrs);
add_device_gemm_dl_f16_f16_f16_mk_kn_mn_irregular_instances(op_ptrs); // add_device_gemm_dl_f16_f16_f16_mk_kn_mn_irregular_instances(op_ptrs);
add_device_gemm_xdl_c_shuffle_f16_f16_f16_mk_kn_mn_instances(op_ptrs); // add_device_gemm_xdl_c_shuffle_f16_f16_f16_mk_kn_mn_instances(op_ptrs);
} // }
else if constexpr(is_same_v<ALayout, Row> && is_same_v<BLayout, Col> && /*else*/ if constexpr(is_same_v<ALayout, Row> && is_same_v<BLayout, Col> &&
is_same_v<CLayout, Row>) is_same_v<CLayout, Row>)
{ {
add_device_gemm_xdl_f16_f16_f16_mk_nk_mn_instances(op_ptrs); // add_device_gemm_xdl_f16_f16_f16_mk_nk_mn_instances(op_ptrs);
add_device_gemm_dl_f16_f16_f16_mk_nk_mn_instances(op_ptrs); // add_device_gemm_dl_f16_f16_f16_mk_nk_mn_instances(op_ptrs);
add_device_gemm_dl_f16_f16_f16_mk_nk_mn_irregular_instances(op_ptrs); // add_device_gemm_dl_f16_f16_f16_mk_nk_mn_irregular_instances(op_ptrs);
add_device_gemm_xdl_c_shuffle_f16_f16_f16_mk_nk_mn_instances(op_ptrs); add_device_gemm_xdl_c_shuffle_f16_f16_f16_mk_nk_mn_instances(op_ptrs);
add_device_gemm_xdl_c_shuffle_2_stage_f16_f16_f16_mk_nk_mn_instances(op_ptrs); // add_device_gemm_xdl_c_shuffle_2_stage_f16_f16_f16_mk_nk_mn_instances(op_ptrs);
} add_device_gemm_xdl_direct_c_write_out_f16_f16_f16_mk_nk_mn_instances_spv8(op_ptrs);
else if constexpr(is_same_v<ALayout, Col> && is_same_v<BLayout, Row> && add_device_gemm_xdl_direct_c_write_out_f16_f16_f16_mk_nk_mn_instances_spv4(op_ptrs);
is_same_v<CLayout, Row>) add_device_gemm_xdl_direct_c_write_out_f16_f16_f16_mk_nk_mn_instances_spv2(op_ptrs);
{
add_device_gemm_xdl_f16_f16_f16_km_kn_mn_instances(op_ptrs);
add_device_gemm_dl_f16_f16_f16_km_kn_mn_instances(op_ptrs);
add_device_gemm_dl_f16_f16_f16_km_kn_mn_irregular_instances(op_ptrs);
add_device_gemm_xdl_c_shuffle_f16_f16_f16_km_kn_mn_instances(op_ptrs);
}
else if constexpr(is_same_v<ALayout, Col> && is_same_v<BLayout, Col> &&
is_same_v<CLayout, Row>)
{
add_device_gemm_xdl_f16_f16_f16_km_nk_mn_instances(op_ptrs);
add_device_gemm_dl_f16_f16_f16_km_nk_mn_instances(op_ptrs);
add_device_gemm_dl_f16_f16_f16_km_nk_mn_irregular_instances(op_ptrs);
add_device_gemm_xdl_c_shuffle_f16_f16_f16_km_nk_mn_instances(op_ptrs);
}
}
else if constexpr(is_same_v<ADataType, ck::bhalf_t> && is_same_v<BDataType, ck::bhalf_t> &&
is_same_v<CDataType, ck::bhalf_t>)
{
if constexpr(is_same_v<ALayout, Row> && is_same_v<BLayout, Row> &&
is_same_v<CLayout, Row>)
{
add_device_gemm_xdl_c_shuffle_bf16_bf16_bf16_mk_kn_mn_instances(op_ptrs);
}
else if constexpr(is_same_v<ALayout, Row> && is_same_v<BLayout, Col> &&
is_same_v<CLayout, Row>)
{
add_device_gemm_xdl_c_shuffle_bf16_bf16_bf16_mk_nk_mn_instances(op_ptrs);
}
else if constexpr(is_same_v<ALayout, Col> && is_same_v<BLayout, Row> &&
is_same_v<CLayout, Row>)
{
add_device_gemm_xdl_c_shuffle_bf16_bf16_bf16_km_kn_mn_instances(op_ptrs);
}
else if constexpr(is_same_v<ALayout, Col> && is_same_v<BLayout, Col> &&
is_same_v<CLayout, Row>)
{
add_device_gemm_xdl_c_shuffle_bf16_bf16_bf16_km_nk_mn_instances(op_ptrs);
}
}
else if constexpr(is_same_v<ADataType, int8_t> && is_same_v<BDataType, int8_t> &&
is_same_v<CDataType, int8_t>)
{
if constexpr(is_same_v<ALayout, Row> && is_same_v<BLayout, Row> &&
is_same_v<CLayout, Row>)
{
add_device_gemm_xdl_c_shuffle_i8_i8_i8_mk_kn_mn_instances(op_ptrs);
add_device_gemm_dl_i8_i8_i8_mk_kn_mn_instances(op_ptrs);
add_device_gemm_dl_i8_i8_i8_mk_kn_mn_irregular_instances(op_ptrs);
}
else if constexpr(is_same_v<ALayout, Row> && is_same_v<BLayout, Col> &&
is_same_v<CLayout, Row>)
{
add_device_gemm_xdl_c_shuffle_i8_i8_i8_mk_nk_mn_instances(op_ptrs);
add_device_gemm_dl_i8_i8_i8_mk_nk_mn_instances(op_ptrs);
add_device_gemm_dl_i8_i8_i8_mk_nk_mn_irregular_instances(op_ptrs);
}
else if constexpr(is_same_v<ALayout, Col> && is_same_v<BLayout, Row> &&
is_same_v<CLayout, Row>)
{
add_device_gemm_xdl_c_shuffle_i8_i8_i8_km_kn_mn_instances(op_ptrs);
add_device_gemm_dl_i8_i8_i8_km_kn_mn_instances(op_ptrs);
add_device_gemm_dl_i8_i8_i8_km_kn_mn_irregular_instances(op_ptrs);
}
else if constexpr(is_same_v<ALayout, Col> && is_same_v<BLayout, Col> &&
is_same_v<CLayout, Row>)
{
add_device_gemm_xdl_c_shuffle_i8_i8_i8_km_nk_mn_instances(op_ptrs);
add_device_gemm_dl_i8_i8_i8_km_nk_mn_instances(op_ptrs);
add_device_gemm_dl_i8_i8_i8_km_nk_mn_irregular_instances(op_ptrs);
} }
// else if constexpr(is_same_v<ALayout, Col> && is_same_v<BLayout, Row> &&
// is_same_v<CLayout, Row>)
// {
// add_device_gemm_xdl_f16_f16_f16_km_kn_mn_instances(op_ptrs);
// add_device_gemm_dl_f16_f16_f16_km_kn_mn_instances(op_ptrs);
// add_device_gemm_dl_f16_f16_f16_km_kn_mn_irregular_instances(op_ptrs);
// add_device_gemm_xdl_c_shuffle_f16_f16_f16_km_kn_mn_instances(op_ptrs);
// }
// else if constexpr(is_same_v<ALayout, Col> && is_same_v<BLayout, Col> &&
// is_same_v<CLayout, Row>)
// {
// add_device_gemm_xdl_f16_f16_f16_km_nk_mn_instances(op_ptrs);
// add_device_gemm_dl_f16_f16_f16_km_nk_mn_instances(op_ptrs);
// add_device_gemm_dl_f16_f16_f16_km_nk_mn_irregular_instances(op_ptrs);
// add_device_gemm_xdl_c_shuffle_f16_f16_f16_km_nk_mn_instances(op_ptrs);
// }
} }
// else if constexpr(is_same_v<ADataType, ck::bhalf_t> && is_same_v<BDataType, ck::bhalf_t>
// &&
// is_same_v<CDataType, ck::bhalf_t>)
// {
// if constexpr(is_same_v<ALayout, Row> && is_same_v<BLayout, Row> &&
// is_same_v<CLayout, Row>)
// {
// add_device_gemm_xdl_c_shuffle_bf16_bf16_bf16_mk_kn_mn_instances(op_ptrs);
// }
// else if constexpr(is_same_v<ALayout, Row> && is_same_v<BLayout, Col> &&
// is_same_v<CLayout, Row>)
// {
// add_device_gemm_xdl_c_shuffle_bf16_bf16_bf16_mk_nk_mn_instances(op_ptrs);
// }
// else if constexpr(is_same_v<ALayout, Col> && is_same_v<BLayout, Row> &&
// is_same_v<CLayout, Row>)
// {
// add_device_gemm_xdl_c_shuffle_bf16_bf16_bf16_km_kn_mn_instances(op_ptrs);
// }
// else if constexpr(is_same_v<ALayout, Col> && is_same_v<BLayout, Col> &&
// is_same_v<CLayout, Row>)
// {
// add_device_gemm_xdl_c_shuffle_bf16_bf16_bf16_km_nk_mn_instances(op_ptrs);
// }
// }
// else if constexpr(is_same_v<ADataType, int8_t> && is_same_v<BDataType, int8_t> &&
// is_same_v<CDataType, int8_t>)
// {
// if constexpr(is_same_v<ALayout, Row> && is_same_v<BLayout, Row> &&
// is_same_v<CLayout, Row>)
// {
// add_device_gemm_xdl_c_shuffle_i8_i8_i8_mk_kn_mn_instances(op_ptrs);
// add_device_gemm_dl_i8_i8_i8_mk_kn_mn_instances(op_ptrs);
// add_device_gemm_dl_i8_i8_i8_mk_kn_mn_irregular_instances(op_ptrs);
// }
// else if constexpr(is_same_v<ALayout, Row> && is_same_v<BLayout, Col> &&
// is_same_v<CLayout, Row>)
// {
// add_device_gemm_xdl_c_shuffle_i8_i8_i8_mk_nk_mn_instances(op_ptrs);
// add_device_gemm_dl_i8_i8_i8_mk_nk_mn_instances(op_ptrs);
// add_device_gemm_dl_i8_i8_i8_mk_nk_mn_irregular_instances(op_ptrs);
// }
// else if constexpr(is_same_v<ALayout, Col> && is_same_v<BLayout, Row> &&
// is_same_v<CLayout, Row>)
// {
// add_device_gemm_xdl_c_shuffle_i8_i8_i8_km_kn_mn_instances(op_ptrs);
// add_device_gemm_dl_i8_i8_i8_km_kn_mn_instances(op_ptrs);
// add_device_gemm_dl_i8_i8_i8_km_kn_mn_irregular_instances(op_ptrs);
// }
// else if constexpr(is_same_v<ALayout, Col> && is_same_v<BLayout, Col> &&
// is_same_v<CLayout, Row>)
// {
// add_device_gemm_xdl_c_shuffle_i8_i8_i8_km_nk_mn_instances(op_ptrs);
// add_device_gemm_dl_i8_i8_i8_km_nk_mn_instances(op_ptrs);
// add_device_gemm_dl_i8_i8_i8_km_nk_mn_irregular_instances(op_ptrs);
// }
// }
return op_ptrs; return op_ptrs;
} }
......
library/src/tensor_operation_instance/gpu/gemm/CMakeLists.txt
View file @ d37ba7ea
add_instance_library(device_gemm_instance add_instance_library(device_gemm_instance
device_gemm_xdl_f64_f64_f64_mk_kn_mn_instance.cpp # device_gemm_xdl_f64_f64_f64_mk_kn_mn_instance.cpp
device_gemm_xdl_f64_f64_f64_mk_nk_mn_instance.cpp # device_gemm_xdl_f64_f64_f64_mk_nk_mn_instance.cpp
device_gemm_xdl_f64_f64_f64_km_kn_mn_instance.cpp # device_gemm_xdl_f64_f64_f64_km_kn_mn_instance.cpp
device_gemm_xdl_f64_f64_f64_km_nk_mn_instance.cpp # device_gemm_xdl_f64_f64_f64_km_nk_mn_instance.cpp
device_gemm_xdl_f32_f32_f32_mk_kn_mn_instance.cpp # device_gemm_xdl_f32_f32_f32_mk_kn_mn_instance.cpp
device_gemm_xdl_f32_f32_f32_mk_nk_mn_instance.cpp # device_gemm_xdl_f32_f32_f32_mk_nk_mn_instance.cpp
device_gemm_xdl_f32_f32_f32_km_kn_mn_instance.cpp # device_gemm_xdl_f32_f32_f32_km_kn_mn_instance.cpp
device_gemm_xdl_f32_f32_f32_km_nk_mn_instance.cpp # device_gemm_xdl_f32_f32_f32_km_nk_mn_instance.cpp
device_gemm_xdl_f16_f16_f16_mk_kn_mn_instance.cpp # device_gemm_xdl_f16_f16_f16_mk_kn_mn_instance.cpp
device_gemm_xdl_f16_f16_f16_mk_nk_mn_instance.cpp # device_gemm_xdl_f16_f16_f16_mk_nk_mn_instance.cpp
device_gemm_xdl_f16_f16_f16_km_kn_mn_instance.cpp # device_gemm_xdl_f16_f16_f16_km_kn_mn_instance.cpp
device_gemm_xdl_f16_f16_f16_km_nk_mn_instance.cpp # device_gemm_xdl_f16_f16_f16_km_nk_mn_instance.cpp
device_gemm_xdl_c_shuffle_i8_i8_i8_mk_kn_mn_instance.cpp # device_gemm_xdl_c_shuffle_i8_i8_i8_mk_kn_mn_instance.cpp
device_gemm_xdl_c_shuffle_i8_i8_i8_mk_nk_mn_instance.cpp # device_gemm_xdl_c_shuffle_i8_i8_i8_mk_nk_mn_instance.cpp
device_gemm_xdl_c_shuffle_i8_i8_i8_km_kn_mn_instance.cpp # device_gemm_xdl_c_shuffle_i8_i8_i8_km_kn_mn_instance.cpp
device_gemm_xdl_c_shuffle_i8_i8_i8_km_nk_mn_instance.cpp # device_gemm_xdl_c_shuffle_i8_i8_i8_km_nk_mn_instance.cpp
device_gemm_xdl_c_shuffle_bf16_bf16_bf16_mk_kn_mn_instance.cpp # device_gemm_xdl_c_shuffle_bf16_bf16_bf16_mk_kn_mn_instance.cpp
device_gemm_xdl_c_shuffle_bf16_bf16_bf16_mk_nk_mn_instance.cpp # device_gemm_xdl_c_shuffle_bf16_bf16_bf16_mk_nk_mn_instance.cpp
device_gemm_xdl_c_shuffle_bf16_bf16_bf16_km_kn_mn_instance.cpp # device_gemm_xdl_c_shuffle_bf16_bf16_bf16_km_kn_mn_instance.cpp
device_gemm_xdl_c_shuffle_bf16_bf16_bf16_km_nk_mn_instance.cpp # device_gemm_xdl_c_shuffle_bf16_bf16_bf16_km_nk_mn_instance.cpp
device_gemm_xdl_c_shuffle_f16_f16_f16_mk_kn_mn_instance.cpp # device_gemm_xdl_c_shuffle_f16_f16_f16_mk_kn_mn_instance.cpp
device_gemm_xdl_c_shuffle_f16_f16_f16_mk_nk_mn_instance.cpp device_gemm_xdl_c_shuffle_f16_f16_f16_mk_nk_mn_instance.cpp
device_gemm_xdl_c_shuffle_f16_f16_f16_km_kn_mn_instance.cpp # device_gemm_xdl_c_shuffle_f16_f16_f16_km_kn_mn_instance.cpp
device_gemm_xdl_c_shuffle_f16_f16_f16_km_nk_mn_instance.cpp # device_gemm_xdl_c_shuffle_f16_f16_f16_km_nk_mn_instance.cpp
device_gemm_xdl_c_shuffle_f32_f32_f32_mk_kn_mn_instance.cpp # device_gemm_xdl_c_shuffle_f32_f32_f32_mk_kn_mn_instance.cpp
device_gemm_xdl_c_shuffle_f32_f32_f32_mk_nk_mn_instance.cpp # device_gemm_xdl_c_shuffle_f32_f32_f32_mk_nk_mn_instance.cpp
device_gemm_xdl_c_shuffle_f32_f32_f32_km_kn_mn_instance.cpp # device_gemm_xdl_c_shuffle_f32_f32_f32_km_kn_mn_instance.cpp
device_gemm_xdl_c_shuffle_f32_f32_f32_km_nk_mn_instance.cpp # device_gemm_xdl_c_shuffle_f32_f32_f32_km_nk_mn_instance.cpp
device_gemm_xdl_c_shuffle_2_stage_f16_f16_f16_mk_nk_mn_instance.cpp # device_gemm_xdl_c_shuffle_2_stage_f16_f16_f16_mk_nk_mn_instance.cpp
device_gemm_dl_f32_f32_f32_mk_kn_mn_instance.cpp # device_gemm_dl_f32_f32_f32_mk_kn_mn_instance.cpp
device_gemm_dl_f32_f32_f32_mk_nk_mn_instance.cpp # device_gemm_dl_f32_f32_f32_mk_nk_mn_instance.cpp
device_gemm_dl_f32_f32_f32_km_kn_mn_instance.cpp # device_gemm_dl_f32_f32_f32_km_kn_mn_instance.cpp
device_gemm_dl_f32_f32_f32_km_nk_mn_instance.cpp # device_gemm_dl_f32_f32_f32_km_nk_mn_instance.cpp
device_gemm_dl_f16_f16_f16_mk_kn_mn_instance.cpp # device_gemm_dl_f16_f16_f16_mk_kn_mn_instance.cpp
device_gemm_dl_f16_f16_f16_mk_kn_mn_irregular_instance.cpp # device_gemm_dl_f16_f16_f16_mk_kn_mn_irregular_instance.cpp
device_gemm_dl_f16_f16_f16_mk_nk_mn_instance.cpp # device_gemm_dl_f16_f16_f16_mk_nk_mn_instance.cpp
device_gemm_dl_f16_f16_f16_mk_nk_mn_irregular_instance.cpp # device_gemm_dl_f16_f16_f16_mk_nk_mn_irregular_instance.cpp
device_gemm_dl_f16_f16_f16_km_kn_mn_instance.cpp # device_gemm_dl_f16_f16_f16_km_kn_mn_instance.cpp
device_gemm_dl_f16_f16_f16_km_kn_mn_irregular_instance.cpp # device_gemm_dl_f16_f16_f16_km_kn_mn_irregular_instance.cpp
device_gemm_dl_f16_f16_f16_km_nk_mn_instance.cpp # device_gemm_dl_f16_f16_f16_km_nk_mn_instance.cpp
device_gemm_dl_f16_f16_f16_km_nk_mn_irregular_instance.cpp # device_gemm_dl_f16_f16_f16_km_nk_mn_irregular_instance.cpp
device_gemm_dl_i8_i8_i8_mk_kn_mn_instance.cpp # device_gemm_dl_i8_i8_i8_mk_kn_mn_instance.cpp
device_gemm_dl_i8_i8_i8_mk_kn_mn_irregular_instance.cpp # device_gemm_dl_i8_i8_i8_mk_kn_mn_irregular_instance.cpp
device_gemm_dl_i8_i8_i8_mk_nk_mn_instance.cpp # device_gemm_dl_i8_i8_i8_mk_nk_mn_instance.cpp
device_gemm_dl_i8_i8_i8_mk_nk_mn_irregular_instance.cpp # device_gemm_dl_i8_i8_i8_mk_nk_mn_irregular_instance.cpp
device_gemm_dl_i8_i8_i8_km_kn_mn_instance.cpp # device_gemm_dl_i8_i8_i8_km_kn_mn_instance.cpp
device_gemm_dl_i8_i8_i8_km_kn_mn_irregular_instance.cpp # device_gemm_dl_i8_i8_i8_km_kn_mn_irregular_instance.cpp
device_gemm_dl_i8_i8_i8_km_nk_mn_instance.cpp # device_gemm_dl_i8_i8_i8_km_nk_mn_instance.cpp
device_gemm_dl_i8_i8_i8_km_nk_mn_irregular_instance.cpp # device_gemm_dl_i8_i8_i8_km_nk_mn_irregular_instance.cpp
device_gemm_xdl_direct_c_write_out_f16_f16_f16_mk_nk_mn_instance_spv8.cpp
device_gemm_xdl_direct_c_write_out_f16_f16_f16_mk_nk_mn_instance_spv4.cpp
device_gemm_xdl_direct_c_write_out_f16_f16_f16_mk_nk_mn_instance_spv2.cpp
) )
library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_f16_f16_f16_mk_nk_mn_instance.cpp
View file @ d37ba7ea
...@@ -27,7 +27,8 @@ using S = ck::Sequence<Is...>; ...@@ -27,7 +27,8 @@ using S = ck::Sequence<Is...>;
using PassThrough = ck::tensor_operation::element_wise::PassThrough; using PassThrough = ck::tensor_operation::element_wise::PassThrough;
static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default; // static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default;
static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
// Compilation parameters for a[m, k] * b[n, k] = c[m, n] // Compilation parameters for a[m, k] * b[n, k] = c[m, n]
using device_gemm_xdl_c_shuffle_f16_f16_f16_mk_nk_mn_instances = std::tuple< using device_gemm_xdl_c_shuffle_f16_f16_f16_mk_nk_mn_instances = std::tuple<
...@@ -37,53 +38,53 @@ using device_gemm_xdl_c_shuffle_f16_f16_f16_mk_nk_mn_instances = std::tuple< ...@@ -37,53 +38,53 @@ using device_gemm_xdl_c_shuffle_f16_f16_f16_mk_nk_mn_instances = std::tuple<
//#####################| | | | | | | | | Operation| Operation| Operation| | Stage| | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| _NBlock_NWaveNPerXdl| _NWaveNPerXdl| | | //#####################| | | | | | | | | Operation| Operation| Operation| | Stage| | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| _NBlock_NWaveNPerXdl| _NWaveNPerXdl| | |
//#####################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | //#####################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
// pipeline v1, 1 wave // pipeline v1, 1 wave
DeviceGemm_Xdl_CShuffle< Row, Col, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 256, 256, 128, 32, 8, 8, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1>, // DeviceGemm_Xdl_CShuffle< Row, Col, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 256, 256, 128, 32, 8, 8, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1>,
DeviceGemm_Xdl_CShuffle< Row, Col, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 256, 128, 256, 32, 8, 8, 32, 32, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1>, // DeviceGemm_Xdl_CShuffle< Row, Col, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 256, 128, 256, 32, 8, 8, 32, 32, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1>,
DeviceGemm_Xdl_CShuffle< Row, Col, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 128, 128, 128, 32, 8, 8, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1>, // DeviceGemm_Xdl_CShuffle< Row, Col, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 128, 128, 128, 32, 8, 8, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1>,
DeviceGemm_Xdl_CShuffle< Row, Col, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 256, 128, 128, 32, 8, 8, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1>, DeviceGemm_Xdl_CShuffle< Row, Col, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 256, 128, 128, 32, 8, 8, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1>
DeviceGemm_Xdl_CShuffle< Row, Col, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 128, 128, 64, 32, 8, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 4>, 8, LoopScheduler::Default, PipelineVersion::v1>, // DeviceGemm_Xdl_CShuffle< Row, Col, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 128, 128, 64, 32, 8, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 4>, 8, LoopScheduler::Default, PipelineVersion::v1>,
DeviceGemm_Xdl_CShuffle< Row, Col, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 128, 64, 128, 32, 8, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1>, // DeviceGemm_Xdl_CShuffle< Row, Col, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 128, 64, 128, 32, 8, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1>,
DeviceGemm_Xdl_CShuffle< Row, Col, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 64, 64, 64, 32, 8, 8, 32, 32, 2, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 4>, 8, LoopScheduler::Default, PipelineVersion::v1>, // DeviceGemm_Xdl_CShuffle< Row, Col, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 64, 64, 64, 32, 8, 8, 32, 32, 2, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 4>, 8, LoopScheduler::Default, PipelineVersion::v1>,
DeviceGemm_Xdl_CShuffle< Row, Col, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 256, 128, 64, 32, 8, 8, 32, 32, 2, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1>, // DeviceGemm_Xdl_CShuffle< Row, Col, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 256, 128, 64, 32, 8, 8, 32, 32, 2, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1>,
DeviceGemm_Xdl_CShuffle< Row, Col, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 256, 64, 128, 32, 8, 8, 32, 32, 1, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1>, // DeviceGemm_Xdl_CShuffle< Row, Col, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 256, 64, 128, 32, 8, 8, 32, 32, 1, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1>,
DeviceGemm_Xdl_CShuffle< Row, Col, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 128, 128, 32, 32, 8, 8, 32, 32, 2, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 4>, 8, LoopScheduler::Default, PipelineVersion::v1>, // DeviceGemm_Xdl_CShuffle< Row, Col, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 128, 128, 32, 32, 8, 8, 32, 32, 2, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 4>, 8, LoopScheduler::Default, PipelineVersion::v1>,
DeviceGemm_Xdl_CShuffle< Row, Col, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 128, 32, 128, 32, 8, 8, 32, 32, 1, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1>, // DeviceGemm_Xdl_CShuffle< Row, Col, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 128, 32, 128, 32, 8, 8, 32, 32, 1, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1>,
DeviceGemm_Xdl_CShuffle< Row, Col, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 64, 64, 32, 32, 8, 8, 32, 32, 2, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 4>, 8, LoopScheduler::Default, PipelineVersion::v1>, // DeviceGemm_Xdl_CShuffle< Row, Col, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 64, 64, 32, 32, 8, 8, 32, 32, 2, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 4>, 8, LoopScheduler::Default, PipelineVersion::v1>
DeviceGemm_Xdl_CShuffle< Row, Col, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 64, 32, 64, 32, 8, 8, 32, 32, 1, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 4>, 8, LoopScheduler::Default, PipelineVersion::v1> // DeviceGemm_Xdl_CShuffle< Row, Col, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 64, 32, 64, 32, 8, 8, 32, 32, 1, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 4>, 8, LoopScheduler::Default, PipelineVersion::v1>
#if CK_EXPERIMENTAL_INTER_WAVE_INSTANCES // #if CK_EXPERIMENTAL_INTER_WAVE_INSTANCES
// pipeline v1, 2 waves // // pipeline v1, 2 waves
, // ,
DeviceGemm_Xdl_CShuffle< Row, Col, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 256, 256, 128, 32, 8, 8, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Interwave, PipelineVersion::v1>, // DeviceGemm_Xdl_CShuffle< Row, Col, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 256, 256, 128, 32, 8, 8, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Interwave, PipelineVersion::v1>,
DeviceGemm_Xdl_CShuffle< Row, Col, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 256, 128, 256, 32, 8, 8, 32, 32, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Interwave, PipelineVersion::v1>, // DeviceGemm_Xdl_CShuffle< Row, Col, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 256, 128, 256, 32, 8, 8, 32, 32, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Interwave, PipelineVersion::v1>,
DeviceGemm_Xdl_CShuffle< Row, Col, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 128, 128, 128, 32, 8, 8, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 8>, 8, LoopScheduler::Interwave, PipelineVersion::v1>, // DeviceGemm_Xdl_CShuffle< Row, Col, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 128, 128, 128, 32, 8, 8, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 8>, 8, LoopScheduler::Interwave, PipelineVersion::v1>,
DeviceGemm_Xdl_CShuffle< Row, Col, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 256, 128, 128, 32, 8, 8, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Interwave, PipelineVersion::v1>, // DeviceGemm_Xdl_CShuffle< Row, Col, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 256, 128, 128, 32, 8, 8, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Interwave, PipelineVersion::v1>,
DeviceGemm_Xdl_CShuffle< Row, Col, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 128, 128, 64, 32, 8, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 4>, 8, LoopScheduler::Interwave, PipelineVersion::v1>, // DeviceGemm_Xdl_CShuffle< Row, Col, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 128, 128, 64, 32, 8, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 4>, 8, LoopScheduler::Interwave, PipelineVersion::v1>,
DeviceGemm_Xdl_CShuffle< Row, Col, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 128, 64, 128, 32, 8, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 8>, 8, LoopScheduler::Interwave, PipelineVersion::v1>, // DeviceGemm_Xdl_CShuffle< Row, Col, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 128, 64, 128, 32, 8, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 8>, 8, LoopScheduler::Interwave, PipelineVersion::v1>,
DeviceGemm_Xdl_CShuffle< Row, Col, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 64, 64, 64, 32, 8, 8, 32, 32, 2, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 4>, 8, LoopScheduler::Interwave, PipelineVersion::v1>, // DeviceGemm_Xdl_CShuffle< Row, Col, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 64, 64, 64, 32, 8, 8, 32, 32, 2, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 4>, 8, LoopScheduler::Interwave, PipelineVersion::v1>,
DeviceGemm_Xdl_CShuffle< Row, Col, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 256, 128, 64, 32, 8, 8, 32, 32, 2, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Interwave, PipelineVersion::v1>, // DeviceGemm_Xdl_CShuffle< Row, Col, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 256, 128, 64, 32, 8, 8, 32, 32, 2, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Interwave, PipelineVersion::v1>,
DeviceGemm_Xdl_CShuffle< Row, Col, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 256, 64, 128, 32, 8, 8, 32, 32, 1, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Interwave, PipelineVersion::v1>, // DeviceGemm_Xdl_CShuffle< Row, Col, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 256, 64, 128, 32, 8, 8, 32, 32, 1, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Interwave, PipelineVersion::v1>,
DeviceGemm_Xdl_CShuffle< Row, Col, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 128, 128, 32, 32, 8, 8, 32, 32, 2, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 4>, 8, LoopScheduler::Interwave, PipelineVersion::v1>, // DeviceGemm_Xdl_CShuffle< Row, Col, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 128, 128, 32, 32, 8, 8, 32, 32, 2, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 4>, 8, LoopScheduler::Interwave, PipelineVersion::v1>,
DeviceGemm_Xdl_CShuffle< Row, Col, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 128, 32, 128, 32, 8, 8, 32, 32, 1, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 8>, 8, LoopScheduler::Interwave, PipelineVersion::v1>, // DeviceGemm_Xdl_CShuffle< Row, Col, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 128, 32, 128, 32, 8, 8, 32, 32, 1, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 8>, 8, LoopScheduler::Interwave, PipelineVersion::v1>,
DeviceGemm_Xdl_CShuffle< Row, Col, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 64, 64, 32, 32, 8, 8, 32, 32, 2, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 4>, 8, LoopScheduler::Interwave, PipelineVersion::v1>, // DeviceGemm_Xdl_CShuffle< Row, Col, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 64, 64, 32, 32, 8, 8, 32, 32, 2, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 4>, 8, LoopScheduler::Interwave, PipelineVersion::v1>,
DeviceGemm_Xdl_CShuffle< Row, Col, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 64, 32, 64, 32, 8, 8, 32, 32, 1, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 4>, 8, LoopScheduler::Interwave, PipelineVersion::v1> // DeviceGemm_Xdl_CShuffle< Row, Col, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 64, 32, 64, 32, 8, 8, 32, 32, 1, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 4>, 8, LoopScheduler::Interwave, PipelineVersion::v1>
#endif // #endif
#if CK_EXPERIMENTAL_PIPELINE_V2_INSTANCES // #if CK_EXPERIMENTAL_PIPELINE_V2_INSTANCES
// pipeline v2, 1 wave // // pipeline v2, 1 wave
, // ,
DeviceGemm_Xdl_CShuffle< Row, Col, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 256, 256, 128, 32, 8, 8, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v2>, // DeviceGemm_Xdl_CShuffle< Row, Col, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 256, 256, 128, 32, 8, 8, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v2>,
DeviceGemm_Xdl_CShuffle< Row, Col, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 256, 128, 256, 32, 8, 8, 32, 32, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v2>, // DeviceGemm_Xdl_CShuffle< Row, Col, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 256, 128, 256, 32, 8, 8, 32, 32, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v2>,
DeviceGemm_Xdl_CShuffle< Row, Col, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 128, 128, 128, 32, 8, 8, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v2>, // DeviceGemm_Xdl_CShuffle< Row, Col, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 128, 128, 128, 32, 8, 8, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v2>,
DeviceGemm_Xdl_CShuffle< Row, Col, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 256, 128, 128, 32, 8, 8, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v2>, // DeviceGemm_Xdl_CShuffle< Row, Col, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 256, 128, 128, 32, 8, 8, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v2>,
DeviceGemm_Xdl_CShuffle< Row, Col, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 128, 128, 64, 32, 8, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 4>, 8, LoopScheduler::Default, PipelineVersion::v2>, // DeviceGemm_Xdl_CShuffle< Row, Col, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 128, 128, 64, 32, 8, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 4>, 8, LoopScheduler::Default, PipelineVersion::v2>,
DeviceGemm_Xdl_CShuffle< Row, Col, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 128, 64, 128, 32, 8, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v2>, // DeviceGemm_Xdl_CShuffle< Row, Col, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 128, 64, 128, 32, 8, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v2>,
DeviceGemm_Xdl_CShuffle< Row, Col, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 64, 64, 64, 32, 8, 8, 32, 32, 2, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 4>, 8, LoopScheduler::Default, PipelineVersion::v2>, // DeviceGemm_Xdl_CShuffle< Row, Col, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 64, 64, 64, 32, 8, 8, 32, 32, 2, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 4>, 8, LoopScheduler::Default, PipelineVersion::v2>,
DeviceGemm_Xdl_CShuffle< Row, Col, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 256, 128, 64, 32, 8, 8, 32, 32, 2, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v2>, // DeviceGemm_Xdl_CShuffle< Row, Col, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 256, 128, 64, 32, 8, 8, 32, 32, 2, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v2>,
DeviceGemm_Xdl_CShuffle< Row, Col, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 256, 64, 128, 32, 8, 8, 32, 32, 1, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v2>, // DeviceGemm_Xdl_CShuffle< Row, Col, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 256, 64, 128, 32, 8, 8, 32, 32, 1, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v2>,
DeviceGemm_Xdl_CShuffle< Row, Col, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 128, 128, 32, 32, 8, 8, 32, 32, 2, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 4>, 8, LoopScheduler::Default, PipelineVersion::v2>, // DeviceGemm_Xdl_CShuffle< Row, Col, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 128, 128, 32, 32, 8, 8, 32, 32, 2, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 4>, 8, LoopScheduler::Default, PipelineVersion::v2>,
DeviceGemm_Xdl_CShuffle< Row, Col, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 128, 32, 128, 32, 8, 8, 32, 32, 1, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v2>, // DeviceGemm_Xdl_CShuffle< Row, Col, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 128, 32, 128, 32, 8, 8, 32, 32, 1, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v2>,
DeviceGemm_Xdl_CShuffle< Row, Col, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 64, 64, 32, 32, 8, 8, 32, 32, 2, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 4>, 8, LoopScheduler::Default, PipelineVersion::v2>, // DeviceGemm_Xdl_CShuffle< Row, Col, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 64, 64, 32, 32, 8, 8, 32, 32, 2, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 4>, 8, LoopScheduler::Default, PipelineVersion::v2>,
DeviceGemm_Xdl_CShuffle< Row, Col, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 64, 32, 64, 32, 8, 8, 32, 32, 1, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 4>, 8, LoopScheduler::Default, PipelineVersion::v2> // DeviceGemm_Xdl_CShuffle< Row, Col, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 64, 32, 64, 32, 8, 8, 32, 32, 1, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 4>, 8, LoopScheduler::Default, PipelineVersion::v2>
#endif // #endif
// clang-format on // clang-format on
>; >;
......
library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_direct_c_write_out_f16_f16_f16_mk_nk_mn_instance_spv2.cpp 0 → 100644
View file @ d37ba7ea
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
#include <cstdlib>
#include "ck/ck.hpp"
#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
#include "ck/tensor_operation/gpu/device/impl/device_gemm_xdl_direct_c_write_out.hpp"
#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
namespace ck {
namespace tensor_operation {
namespace device {
namespace instance {
using F16 = ck::half_t;
using F32 = float;
using Row = ck::tensor_layout::gemm::RowMajor;
using Col = ck::tensor_layout::gemm::ColumnMajor;
template <ck::index_t... Is>
using S = ck::Sequence<Is...>;
using PassThrough = ck::tensor_operation::element_wise::PassThrough;
static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
static constexpr auto LoopSchedDefault = ck::LoopScheduler::Default;
// Compilation parameters for a[m, k] * b[n, k] = c[m, n]
using device_gemm_xdl_f16_f16_f16_mk_nk_mn_instances = std::tuple<
// clang-format off
//############################| ALayout| BLayout| CLayout| AData| BData| CData| AccData| A| B| C| GEMM| NumGemmK| Block| MPer| NPer| KPer| AK1| BK1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer| CThreadTransfer| LoopScheduler| Pipeline|
//############################| | | | Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Specialization| Prefetch| Size| Block| Block| Block| | | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| DstAccessOrder| DstVectorDim| DstScalarPerVector| | |
//############################| | | | | | | | Operation| Operation| Operation| | Stage| | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | | | | |
//############################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
// pipeline v1, 1 wave
DeviceGemm_Xdl_DirectCWriteOut< Row, Col, Row, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 256, 128, 128, 32, 8, 8, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<0, 1, 2, 3, 4, 5, 6, 7>, 7, 2, LoopSchedDefault, PipelineVersion::v1>,
DeviceGemm_Xdl_DirectCWriteOut< Row, Col, Row, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 256, 192, 64, 32, 8, 8, 32, 32, 3, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<0, 1, 2, 3, 4, 5, 6, 7>, 7, 2, LoopSchedDefault, PipelineVersion::v1>,
DeviceGemm_Xdl_DirectCWriteOut< Row, Col, Row, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 256, 64, 192, 32, 8, 8, 32, 32, 1, 3, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 48, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<0, 1, 2, 3, 4, 5, 6, 7>, 7, 2, LoopSchedDefault, PipelineVersion::v1>,
DeviceGemm_Xdl_DirectCWriteOut< Row, Col, Row, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 128, 128, 128, 32, 8, 8, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<0, 1, 2, 3, 4, 5, 6, 7>, 7, 2, LoopSchedDefault, PipelineVersion::v1>,
DeviceGemm_Xdl_DirectCWriteOut< Row, Col, Row, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 128, 128, 64, 32, 8, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<0, 1, 2, 3, 4, 5, 6, 7>, 7, 2, LoopSchedDefault, PipelineVersion::v1>,
DeviceGemm_Xdl_DirectCWriteOut< Row, Col, Row, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 128, 64, 128, 32, 8, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<0, 1, 2, 3, 4, 5, 6, 7>, 7, 2, LoopSchedDefault, PipelineVersion::v1>,
DeviceGemm_Xdl_DirectCWriteOut< Row, Col, Row, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 64, 64, 64, 32, 8, 8, 32, 32, 2, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<0, 1, 2, 3, 4, 5, 6, 7>, 7, 2, LoopSchedDefault, PipelineVersion::v1>,
DeviceGemm_Xdl_DirectCWriteOut< Row, Col, Row, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 64, 64, 32, 32, 8, 8, 32, 32, 2, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<0, 1, 2, 3, 4, 5, 6, 7>, 7, 2, LoopSchedDefault, PipelineVersion::v1>,
DeviceGemm_Xdl_DirectCWriteOut< Row, Col, Row, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 256, 128, 128, 32, 8, 8, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<0, 1, 2, 3, 5, 4, 6, 7>, 7, 2, LoopSchedDefault, PipelineVersion::v1>,
DeviceGemm_Xdl_DirectCWriteOut< Row, Col, Row, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 256, 192, 64, 32, 8, 8, 32, 32, 3, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<0, 1, 2, 3, 5, 4, 6, 7>, 7, 2, LoopSchedDefault, PipelineVersion::v1>,
DeviceGemm_Xdl_DirectCWriteOut< Row, Col, Row, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 256, 64, 192, 32, 8, 8, 32, 32, 1, 3, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 48, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<0, 1, 2, 3, 5, 4, 6, 7>, 7, 2, LoopSchedDefault, PipelineVersion::v1>,
DeviceGemm_Xdl_DirectCWriteOut< Row, Col, Row, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 128, 128, 128, 32, 8, 8, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<0, 1, 2, 3, 5, 4, 6, 7>, 7, 2, LoopSchedDefault, PipelineVersion::v1>,
DeviceGemm_Xdl_DirectCWriteOut< Row, Col, Row, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 128, 128, 64, 32, 8, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<0, 1, 2, 3, 5, 4, 6, 7>, 7, 2, LoopSchedDefault, PipelineVersion::v1>,
DeviceGemm_Xdl_DirectCWriteOut< Row, Col, Row, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 128, 64, 128, 32, 8, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<0, 1, 2, 3, 5, 4, 6, 7>, 7, 2, LoopSchedDefault, PipelineVersion::v1>,
DeviceGemm_Xdl_DirectCWriteOut< Row, Col, Row, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 64, 64, 64, 32, 8, 8, 32, 32, 2, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<0, 1, 2, 3, 5, 4, 6, 7>, 7, 2, LoopSchedDefault, PipelineVersion::v1>,
DeviceGemm_Xdl_DirectCWriteOut< Row, Col, Row, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 64, 64, 32, 32, 8, 8, 32, 32, 2, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<0, 1, 2, 3, 5, 4, 6, 7>, 7, 2, LoopSchedDefault, PipelineVersion::v1>,
DeviceGemm_Xdl_DirectCWriteOut< Row, Col, Row, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 256, 128, 128, 32, 8, 8, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<1, 0, 2, 3, 5, 4, 6, 7>, 7, 2, LoopSchedDefault, PipelineVersion::v1>,
DeviceGemm_Xdl_DirectCWriteOut< Row, Col, Row, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 256, 192, 64, 32, 8, 8, 32, 32, 3, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<1, 0, 2, 3, 5, 4, 6, 7>, 7, 2, LoopSchedDefault, PipelineVersion::v1>,
DeviceGemm_Xdl_DirectCWriteOut< Row, Col, Row, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 256, 64, 192, 32, 8, 8, 32, 32, 1, 3, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 48, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<1, 0, 2, 3, 5, 4, 6, 7>, 7, 2, LoopSchedDefault, PipelineVersion::v1>,
DeviceGemm_Xdl_DirectCWriteOut< Row, Col, Row, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 128, 128, 128, 32, 8, 8, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<1, 0, 2, 3, 5, 4, 6, 7>, 7, 2, LoopSchedDefault, PipelineVersion::v1>,
DeviceGemm_Xdl_DirectCWriteOut< Row, Col, Row, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 128, 128, 64, 32, 8, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<1, 0, 2, 3, 5, 4, 6, 7>, 7, 2, LoopSchedDefault, PipelineVersion::v1>,
DeviceGemm_Xdl_DirectCWriteOut< Row, Col, Row, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 128, 64, 128, 32, 8, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<1, 0, 2, 3, 5, 4, 6, 7>, 7, 2, LoopSchedDefault, PipelineVersion::v1>,
DeviceGemm_Xdl_DirectCWriteOut< Row, Col, Row, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 64, 64, 64, 32, 8, 8, 32, 32, 2, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<1, 0, 2, 3, 5, 4, 6, 7>, 7, 2, LoopSchedDefault, PipelineVersion::v1>,
DeviceGemm_Xdl_DirectCWriteOut< Row, Col, Row, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 64, 64, 32, 32, 8, 8, 32, 32, 2, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<1, 0, 2, 3, 5, 4, 6, 7>, 7, 2, LoopSchedDefault, PipelineVersion::v1>,
DeviceGemm_Xdl_DirectCWriteOut< Row, Col, Row, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 256, 128, 128, 32, 8, 8, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<1, 0, 2, 3, 4, 5, 6, 7>, 7, 2, LoopSchedDefault, PipelineVersion::v1>,
DeviceGemm_Xdl_DirectCWriteOut< Row, Col, Row, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 256, 192, 64, 32, 8, 8, 32, 32, 3, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<1, 0, 2, 3, 4, 5, 6, 7>, 7, 2, LoopSchedDefault, PipelineVersion::v1>,
DeviceGemm_Xdl_DirectCWriteOut< Row, Col, Row, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 256, 64, 192, 32, 8, 8, 32, 32, 1, 3, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 48, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<1, 0, 2, 3, 4, 5, 6, 7>, 7, 2, LoopSchedDefault, PipelineVersion::v1>,
DeviceGemm_Xdl_DirectCWriteOut< Row, Col, Row, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 128, 128, 128, 32, 8, 8, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<1, 0, 2, 3, 4, 5, 6, 7>, 7, 2, LoopSchedDefault, PipelineVersion::v1>,
DeviceGemm_Xdl_DirectCWriteOut< Row, Col, Row, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 128, 128, 64, 32, 8, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<1, 0, 2, 3, 4, 5, 6, 7>, 7, 2, LoopSchedDefault, PipelineVersion::v1>,
DeviceGemm_Xdl_DirectCWriteOut< Row, Col, Row, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 128, 64, 128, 32, 8, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<1, 0, 2, 3, 4, 5, 6, 7>, 7, 2, LoopSchedDefault, PipelineVersion::v1>,
DeviceGemm_Xdl_DirectCWriteOut< Row, Col, Row, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 64, 64, 64, 32, 8, 8, 32, 32, 2, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<1, 0, 2, 3, 4, 5, 6, 7>, 7, 2, LoopSchedDefault, PipelineVersion::v1>,
DeviceGemm_Xdl_DirectCWriteOut< Row, Col, Row, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 64, 64, 32, 32, 8, 8, 32, 32, 2, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<1, 0, 2, 3, 4, 5, 6, 7>, 7, 2, LoopSchedDefault, PipelineVersion::v1>
// clang-format on
>;
void add_device_gemm_xdl_direct_c_write_out_f16_f16_f16_mk_nk_mn_instances_spv2(
std::vector<std::unique_ptr<
DeviceGemm<Row, Col, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
instances)
{
add_device_operation_instances(instances, device_gemm_xdl_f16_f16_f16_mk_nk_mn_instances{});
}
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck
library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_direct_c_write_out_f16_f16_f16_mk_nk_mn_instance_spv4.cpp 0 → 100644
View file @ d37ba7ea
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
#include <cstdlib>
#include "ck/ck.hpp"
#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
#include "ck/tensor_operation/gpu/device/impl/device_gemm_xdl_direct_c_write_out.hpp"
#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
namespace ck {
namespace tensor_operation {
namespace device {
namespace instance {
using F16 = ck::half_t;
using F32 = float;
using Row = ck::tensor_layout::gemm::RowMajor;
using Col = ck::tensor_layout::gemm::ColumnMajor;
template <ck::index_t... Is>
using S = ck::Sequence<Is...>;
using PassThrough = ck::tensor_operation::element_wise::PassThrough;
static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
static constexpr auto LoopSchedDefault = ck::LoopScheduler::Default;
// Compilation parameters for a[m, k] * b[n, k] = c[m, n]
using device_gemm_xdl_f16_f16_f16_mk_nk_mn_instances = std::tuple<
// clang-format off
//############################| ALayout| BLayout| CLayout| AData| BData| CData| AccData| A| B| C| GEMM| NumGemmK| Block| MPer| NPer| KPer| AK1| BK1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer| CThreadTransfer| LoopScheduler| Pipeline|
//############################| | | | Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Specialization| Prefetch| Size| Block| Block| Block| | | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| DstAccessOrder| DstVectorDim| DstScalarPerVector| | |
//############################| | | | | | | | Operation| Operation| Operation| | Stage| | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | | | | |
//############################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
// pipeline v1, 1 wave
DeviceGemm_Xdl_DirectCWriteOut< Row, Col, Row, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 256, 128, 128, 32, 8, 8, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<0, 1, 2, 3, 4, 5, 6, 7>, 7, 4, LoopSchedDefault, PipelineVersion::v1>,
DeviceGemm_Xdl_DirectCWriteOut< Row, Col, Row, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 256, 192, 64, 32, 8, 8, 32, 32, 3, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<0, 1, 2, 3, 4, 5, 6, 7>, 7, 4, LoopSchedDefault, PipelineVersion::v1>,
DeviceGemm_Xdl_DirectCWriteOut< Row, Col, Row, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 256, 64, 192, 32, 8, 8, 32, 32, 1, 3, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 48, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<0, 1, 2, 3, 4, 5, 6, 7>, 7, 4, LoopSchedDefault, PipelineVersion::v1>,
DeviceGemm_Xdl_DirectCWriteOut< Row, Col, Row, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 128, 128, 128, 32, 8, 8, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<0, 1, 2, 3, 4, 5, 6, 7>, 7, 4, LoopSchedDefault, PipelineVersion::v1>,
DeviceGemm_Xdl_DirectCWriteOut< Row, Col, Row, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 128, 128, 64, 32, 8, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<0, 1, 2, 3, 4, 5, 6, 7>, 7, 4, LoopSchedDefault, PipelineVersion::v1>,
DeviceGemm_Xdl_DirectCWriteOut< Row, Col, Row, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 128, 64, 128, 32, 8, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<0, 1, 2, 3, 4, 5, 6, 7>, 7, 4, LoopSchedDefault, PipelineVersion::v1>,
DeviceGemm_Xdl_DirectCWriteOut< Row, Col, Row, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 64, 64, 64, 32, 8, 8, 32, 32, 2, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<0, 1, 2, 3, 4, 5, 6, 7>, 7, 4, LoopSchedDefault, PipelineVersion::v1>,
DeviceGemm_Xdl_DirectCWriteOut< Row, Col, Row, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 64, 64, 32, 32, 8, 8, 32, 32, 2, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<0, 1, 2, 3, 4, 5, 6, 7>, 7, 4, LoopSchedDefault, PipelineVersion::v1>,
DeviceGemm_Xdl_DirectCWriteOut< Row, Col, Row, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 256, 128, 128, 32, 8, 8, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<0, 1, 2, 3, 5, 4, 6, 7>, 7, 4, LoopSchedDefault, PipelineVersion::v1>,
DeviceGemm_Xdl_DirectCWriteOut< Row, Col, Row, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 256, 192, 64, 32, 8, 8, 32, 32, 3, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<0, 1, 2, 3, 5, 4, 6, 7>, 7, 4, LoopSchedDefault, PipelineVersion::v1>,
DeviceGemm_Xdl_DirectCWriteOut< Row, Col, Row, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 256, 64, 192, 32, 8, 8, 32, 32, 1, 3, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 48, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<0, 1, 2, 3, 5, 4, 6, 7>, 7, 4, LoopSchedDefault, PipelineVersion::v1>,
DeviceGemm_Xdl_DirectCWriteOut< Row, Col, Row, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 128, 128, 128, 32, 8, 8, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<0, 1, 2, 3, 5, 4, 6, 7>, 7, 4, LoopSchedDefault, PipelineVersion::v1>,
DeviceGemm_Xdl_DirectCWriteOut< Row, Col, Row, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 128, 128, 64, 32, 8, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<0, 1, 2, 3, 5, 4, 6, 7>, 7, 4, LoopSchedDefault, PipelineVersion::v1>,
DeviceGemm_Xdl_DirectCWriteOut< Row, Col, Row, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 128, 64, 128, 32, 8, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<0, 1, 2, 3, 5, 4, 6, 7>, 7, 4, LoopSchedDefault, PipelineVersion::v1>,
DeviceGemm_Xdl_DirectCWriteOut< Row, Col, Row, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 64, 64, 64, 32, 8, 8, 32, 32, 2, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<0, 1, 2, 3, 5, 4, 6, 7>, 7, 4, LoopSchedDefault, PipelineVersion::v1>,
DeviceGemm_Xdl_DirectCWriteOut< Row, Col, Row, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 64, 64, 32, 32, 8, 8, 32, 32, 2, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<0, 1, 2, 3, 5, 4, 6, 7>, 7, 4, LoopSchedDefault, PipelineVersion::v1>,
DeviceGemm_Xdl_DirectCWriteOut< Row, Col, Row, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 256, 128, 128, 32, 8, 8, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<1, 0, 2, 3, 5, 4, 6, 7>, 7, 4, LoopSchedDefault, PipelineVersion::v1>,
DeviceGemm_Xdl_DirectCWriteOut< Row, Col, Row, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 256, 192, 64, 32, 8, 8, 32, 32, 3, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<1, 0, 2, 3, 5, 4, 6, 7>, 7, 4, LoopSchedDefault, PipelineVersion::v1>,
DeviceGemm_Xdl_DirectCWriteOut< Row, Col, Row, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 256, 64, 192, 32, 8, 8, 32, 32, 1, 3, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 48, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<1, 0, 2, 3, 5, 4, 6, 7>, 7, 4, LoopSchedDefault, PipelineVersion::v1>,
DeviceGemm_Xdl_DirectCWriteOut< Row, Col, Row, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 128, 128, 128, 32, 8, 8, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<1, 0, 2, 3, 5, 4, 6, 7>, 7, 4, LoopSchedDefault, PipelineVersion::v1>,
DeviceGemm_Xdl_DirectCWriteOut< Row, Col, Row, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 128, 128, 64, 32, 8, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<1, 0, 2, 3, 5, 4, 6, 7>, 7, 4, LoopSchedDefault, PipelineVersion::v1>,
DeviceGemm_Xdl_DirectCWriteOut< Row, Col, Row, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 128, 64, 128, 32, 8, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<1, 0, 2, 3, 5, 4, 6, 7>, 7, 4, LoopSchedDefault, PipelineVersion::v1>,
DeviceGemm_Xdl_DirectCWriteOut< Row, Col, Row, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 64, 64, 64, 32, 8, 8, 32, 32, 2, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<1, 0, 2, 3, 5, 4, 6, 7>, 7, 4, LoopSchedDefault, PipelineVersion::v1>,
DeviceGemm_Xdl_DirectCWriteOut< Row, Col, Row, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 64, 64, 32, 32, 8, 8, 32, 32, 2, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<1, 0, 2, 3, 5, 4, 6, 7>, 7, 4, LoopSchedDefault, PipelineVersion::v1>,
DeviceGemm_Xdl_DirectCWriteOut< Row, Col, Row, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 256, 128, 128, 32, 8, 8, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<1, 0, 2, 3, 4, 5, 6, 7>, 7, 4, LoopSchedDefault, PipelineVersion::v1>,
DeviceGemm_Xdl_DirectCWriteOut< Row, Col, Row, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 256, 192, 64, 32, 8, 8, 32, 32, 3, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<1, 0, 2, 3, 4, 5, 6, 7>, 7, 4, LoopSchedDefault, PipelineVersion::v1>,
DeviceGemm_Xdl_DirectCWriteOut< Row, Col, Row, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 256, 64, 192, 32, 8, 8, 32, 32, 1, 3, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 48, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<1, 0, 2, 3, 4, 5, 6, 7>, 7, 4, LoopSchedDefault, PipelineVersion::v1>,
DeviceGemm_Xdl_DirectCWriteOut< Row, Col, Row, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 128, 128, 128, 32, 8, 8, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<1, 0, 2, 3, 4, 5, 6, 7>, 7, 4, LoopSchedDefault, PipelineVersion::v1>,
DeviceGemm_Xdl_DirectCWriteOut< Row, Col, Row, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 128, 128, 64, 32, 8, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<1, 0, 2, 3, 4, 5, 6, 7>, 7, 4, LoopSchedDefault, PipelineVersion::v1>,
DeviceGemm_Xdl_DirectCWriteOut< Row, Col, Row, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 128, 64, 128, 32, 8, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<1, 0, 2, 3, 4, 5, 6, 7>, 7, 4, LoopSchedDefault, PipelineVersion::v1>,
DeviceGemm_Xdl_DirectCWriteOut< Row, Col, Row, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 64, 64, 64, 32, 8, 8, 32, 32, 2, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<1, 0, 2, 3, 4, 5, 6, 7>, 7, 4, LoopSchedDefault, PipelineVersion::v1>,
DeviceGemm_Xdl_DirectCWriteOut< Row, Col, Row, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 64, 64, 32, 32, 8, 8, 32, 32, 2, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<1, 0, 2, 3, 4, 5, 6, 7>, 7, 4, LoopSchedDefault, PipelineVersion::v1>
// clang-format on
>;
void add_device_gemm_xdl_direct_c_write_out_f16_f16_f16_mk_nk_mn_instances_spv4(
std::vector<std::unique_ptr<
DeviceGemm<Row, Col, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
instances)
{
add_device_operation_instances(instances, device_gemm_xdl_f16_f16_f16_mk_nk_mn_instances{});
}
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck
library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_direct_c_write_out_f16_f16_f16_mk_nk_mn_instance.cpp → library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_direct_c_write_out_f16_f16_f16_mk_nk_mn_instance_spv8.cpp
View file @ d37ba7ea
...@@ -44,23 +44,39 @@ using device_gemm_xdl_f16_f16_f16_mk_nk_mn_instances = std::tuple< ...@@ -44,23 +44,39 @@ using device_gemm_xdl_f16_f16_f16_mk_nk_mn_instances = std::tuple<
DeviceGemm_Xdl_DirectCWriteOut< Row, Col, Row, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 128, 128, 64, 32, 8, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<0, 1, 2, 3, 4, 5, 6, 7>, 7, 8, LoopSchedDefault, PipelineVersion::v1>, DeviceGemm_Xdl_DirectCWriteOut< Row, Col, Row, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 128, 128, 64, 32, 8, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<0, 1, 2, 3, 4, 5, 6, 7>, 7, 8, LoopSchedDefault, PipelineVersion::v1>,
DeviceGemm_Xdl_DirectCWriteOut< Row, Col, Row, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 128, 64, 128, 32, 8, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<0, 1, 2, 3, 4, 5, 6, 7>, 7, 8, LoopSchedDefault, PipelineVersion::v1>, DeviceGemm_Xdl_DirectCWriteOut< Row, Col, Row, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 128, 64, 128, 32, 8, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<0, 1, 2, 3, 4, 5, 6, 7>, 7, 8, LoopSchedDefault, PipelineVersion::v1>,
DeviceGemm_Xdl_DirectCWriteOut< Row, Col, Row, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 64, 64, 64, 32, 8, 8, 32, 32, 2, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<0, 1, 2, 3, 4, 5, 6, 7>, 7, 8, LoopSchedDefault, PipelineVersion::v1>, DeviceGemm_Xdl_DirectCWriteOut< Row, Col, Row, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 64, 64, 64, 32, 8, 8, 32, 32, 2, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<0, 1, 2, 3, 4, 5, 6, 7>, 7, 8, LoopSchedDefault, PipelineVersion::v1>,
DeviceGemm_Xdl_DirectCWriteOut< Row, Col, Row, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 64, 64, 32, 32, 8, 8, 32, 32, 2, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<0, 1, 2, 3, 4, 5, 6, 7>, 7, 8, LoopSchedDefault, PipelineVersion::v1> DeviceGemm_Xdl_DirectCWriteOut< Row, Col, Row, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 64, 64, 32, 32, 8, 8, 32, 32, 2, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<0, 1, 2, 3, 4, 5, 6, 7>, 7, 8, LoopSchedDefault, PipelineVersion::v1>,
// #if CK_EXPERIMENTAL_PIPELINE_V2_INSTANCES
// // pipeline v2, 1 wave DeviceGemm_Xdl_DirectCWriteOut< Row, Col, Row, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 256, 128, 128, 32, 8, 8, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<0, 1, 2, 3, 5, 4, 6, 7>, 7, 8, LoopSchedDefault, PipelineVersion::v1>,
// , DeviceGemm_Xdl_DirectCWriteOut< Row, Col, Row, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 256, 192, 64, 32, 8, 8, 32, 32, 3, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<0, 1, 2, 3, 5, 4, 6, 7>, 7, 8, LoopSchedDefault, PipelineVersion::v1>,
// DeviceGemm_Xdl_DirectCWriteOut< Row, Col, Row, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 256, 128, 128, 32, 8, 8, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<0, 1, 2, 3, 4, 5, 6, 7>, 7, 8, LoopSchedDefault, PipelineVersion::v2>, DeviceGemm_Xdl_DirectCWriteOut< Row, Col, Row, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 256, 64, 192, 32, 8, 8, 32, 32, 1, 3, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 48, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<0, 1, 2, 3, 5, 4, 6, 7>, 7, 8, LoopSchedDefault, PipelineVersion::v1>,
// DeviceGemm_Xdl_DirectCWriteOut< Row, Col, Row, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 256, 192, 64, 32, 8, 8, 32, 32, 3, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<0, 1, 2, 3, 4, 5, 6, 7>, 7, 8, LoopSchedDefault, PipelineVersion::v2>, DeviceGemm_Xdl_DirectCWriteOut< Row, Col, Row, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 128, 128, 128, 32, 8, 8, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<0, 1, 2, 3, 5, 4, 6, 7>, 7, 8, LoopSchedDefault, PipelineVersion::v1>,
// DeviceGemm_Xdl_DirectCWriteOut< Row, Col, Row, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 256, 64, 192, 32, 8, 8, 32, 32, 1, 3, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 48, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<0, 1, 2, 3, 4, 5, 6, 7>, 7, 8, LoopSchedDefault, PipelineVersion::v2>, DeviceGemm_Xdl_DirectCWriteOut< Row, Col, Row, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 128, 128, 64, 32, 8, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<0, 1, 2, 3, 5, 4, 6, 7>, 7, 8, LoopSchedDefault, PipelineVersion::v1>,
// DeviceGemm_Xdl_DirectCWriteOut< Row, Col, Row, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 128, 128, 128, 32, 8, 8, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<0, 1, 2, 3, 4, 5, 6, 7>, 7, 8, LoopSchedDefault, PipelineVersion::v2>, DeviceGemm_Xdl_DirectCWriteOut< Row, Col, Row, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 128, 64, 128, 32, 8, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<0, 1, 2, 3, 5, 4, 6, 7>, 7, 8, LoopSchedDefault, PipelineVersion::v1>,
// DeviceGemm_Xdl_DirectCWriteOut< Row, Col, Row, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 128, 128, 64, 32, 8, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<0, 1, 2, 3, 4, 5, 6, 7>, 7, 8, LoopSchedDefault, PipelineVersion::v2>, DeviceGemm_Xdl_DirectCWriteOut< Row, Col, Row, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 64, 64, 64, 32, 8, 8, 32, 32, 2, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<0, 1, 2, 3, 5, 4, 6, 7>, 7, 8, LoopSchedDefault, PipelineVersion::v1>,
// DeviceGemm_Xdl_DirectCWriteOut< Row, Col, Row, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 128, 64, 128, 32, 8, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<0, 1, 2, 3, 4, 5, 6, 7>, 7, 8, LoopSchedDefault, PipelineVersion::v2>, DeviceGemm_Xdl_DirectCWriteOut< Row, Col, Row, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 64, 64, 32, 32, 8, 8, 32, 32, 2, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<0, 1, 2, 3, 5, 4, 6, 7>, 7, 8, LoopSchedDefault, PipelineVersion::v1>,
// DeviceGemm_Xdl_DirectCWriteOut< Row, Col, Row, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 64, 64, 64, 32, 8, 8, 32, 32, 2, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<0, 1, 2, 3, 4, 5, 6, 7>, 7, 8, LoopSchedDefault, PipelineVersion::v2>,
// DeviceGemm_Xdl_DirectCWriteOut< Row, Col, Row, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 64, 64, 32, 32, 8, 8, 32, 32, 2, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<0, 1, 2, 3, 4, 5, 6, 7>, 7, 8, LoopSchedDefault, PipelineVersion::v2>, DeviceGemm_Xdl_DirectCWriteOut< Row, Col, Row, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 256, 128, 128, 32, 8, 8, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<1, 0, 2, 3, 5, 4, 6, 7>, 7, 8, LoopSchedDefault, PipelineVersion::v1>,
// #endif DeviceGemm_Xdl_DirectCWriteOut< Row, Col, Row, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 256, 192, 64, 32, 8, 8, 32, 32, 3, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<1, 0, 2, 3, 5, 4, 6, 7>, 7, 8, LoopSchedDefault, PipelineVersion::v1>,
DeviceGemm_Xdl_DirectCWriteOut< Row, Col, Row, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 256, 64, 192, 32, 8, 8, 32, 32, 1, 3, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 48, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<1, 0, 2, 3, 5, 4, 6, 7>, 7, 8, LoopSchedDefault, PipelineVersion::v1>,
DeviceGemm_Xdl_DirectCWriteOut< Row, Col, Row, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 128, 128, 128, 32, 8, 8, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<1, 0, 2, 3, 5, 4, 6, 7>, 7, 8, LoopSchedDefault, PipelineVersion::v1>,
DeviceGemm_Xdl_DirectCWriteOut< Row, Col, Row, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 128, 128, 64, 32, 8, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<1, 0, 2, 3, 5, 4, 6, 7>, 7, 8, LoopSchedDefault, PipelineVersion::v1>,
DeviceGemm_Xdl_DirectCWriteOut< Row, Col, Row, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 128, 64, 128, 32, 8, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<1, 0, 2, 3, 5, 4, 6, 7>, 7, 8, LoopSchedDefault, PipelineVersion::v1>,
DeviceGemm_Xdl_DirectCWriteOut< Row, Col, Row, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 64, 64, 64, 32, 8, 8, 32, 32, 2, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<1, 0, 2, 3, 5, 4, 6, 7>, 7, 8, LoopSchedDefault, PipelineVersion::v1>,
DeviceGemm_Xdl_DirectCWriteOut< Row, Col, Row, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 64, 64, 32, 32, 8, 8, 32, 32, 2, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<1, 0, 2, 3, 5, 4, 6, 7>, 7, 8, LoopSchedDefault, PipelineVersion::v1>,
DeviceGemm_Xdl_DirectCWriteOut< Row, Col, Row, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 256, 128, 128, 32, 8, 8, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<1, 0, 2, 3, 4, 5, 6, 7>, 7, 8, LoopSchedDefault, PipelineVersion::v1>,
DeviceGemm_Xdl_DirectCWriteOut< Row, Col, Row, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 256, 192, 64, 32, 8, 8, 32, 32, 3, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<1, 0, 2, 3, 4, 5, 6, 7>, 7, 8, LoopSchedDefault, PipelineVersion::v1>,
DeviceGemm_Xdl_DirectCWriteOut< Row, Col, Row, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 256, 64, 192, 32, 8, 8, 32, 32, 1, 3, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 48, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<1, 0, 2, 3, 4, 5, 6, 7>, 7, 8, LoopSchedDefault, PipelineVersion::v1>,
DeviceGemm_Xdl_DirectCWriteOut< Row, Col, Row, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 128, 128, 128, 32, 8, 8, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<1, 0, 2, 3, 4, 5, 6, 7>, 7, 8, LoopSchedDefault, PipelineVersion::v1>,
DeviceGemm_Xdl_DirectCWriteOut< Row, Col, Row, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 128, 128, 64, 32, 8, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<1, 0, 2, 3, 4, 5, 6, 7>, 7, 8, LoopSchedDefault, PipelineVersion::v1>,
DeviceGemm_Xdl_DirectCWriteOut< Row, Col, Row, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 128, 64, 128, 32, 8, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<1, 0, 2, 3, 4, 5, 6, 7>, 7, 8, LoopSchedDefault, PipelineVersion::v1>,
DeviceGemm_Xdl_DirectCWriteOut< Row, Col, Row, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 64, 64, 64, 32, 8, 8, 32, 32, 2, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<1, 0, 2, 3, 4, 5, 6, 7>, 7, 8, LoopSchedDefault, PipelineVersion::v1>,
DeviceGemm_Xdl_DirectCWriteOut< Row, Col, Row, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 64, 64, 32, 32, 8, 8, 32, 32, 2, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<1, 0, 2, 3, 4, 5, 6, 7>, 7, 8, LoopSchedDefault, PipelineVersion::v1>
// clang-format on // clang-format on
>; >;
void add_device_gemm_xdl_direct_c_write_out_f16_f16_f16_mk_nk_mn_instances( void add_device_gemm_xdl_direct_c_write_out_f16_f16_f16_mk_nk_mn_instances_spv8(
std::vector<std::unique_ptr< std::vector<std::unique_ptr<
DeviceGemm<Row, Col, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>& DeviceGemm<Row, Col, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
instances) instances)
......
profiler/include/profiler/profile_gemm_impl.hpp
View file @ d37ba7ea
...@@ -173,8 +173,8 @@ int profile_gemm_impl(int do_verification, ...@@ -173,8 +173,8 @@ int profile_gemm_impl(int do_verification,
float gb_per_sec = num_btype / 1.E6 / avg_time; float gb_per_sec = num_btype / 1.E6 / avg_time;
std::cout << "Perf: " << std::setw(10) << avg_time << " ms, " << tflops << " TFlops, " std::cout << std::setw(10) << avg_time << "; " << tflops << "; " << gb_per_sec << "; "
<< gb_per_sec << " GB/s, " << op_name << std::endl; << op_name << std::endl;
if(tflops > best_tflops) if(tflops > best_tflops)
{ {
......
profiler/src/CMakeLists.txt
View file @ d37ba7ea
...@@ -2,38 +2,38 @@ ...@@ -2,38 +2,38 @@
set(PROFILER_SOURCES set(PROFILER_SOURCES
profiler.cpp profiler.cpp
profile_gemm.cpp profile_gemm.cpp
profile_gemm_splitk.cpp # profile_gemm_splitk.cpp
profile_gemm_bilinear.cpp # profile_gemm_bilinear.cpp
profile_gemm_bias_add_reduce.cpp # profile_gemm_bias_add_reduce.cpp
profile_gemm_add_add_fastgelu.cpp # profile_gemm_add_add_fastgelu.cpp
profile_gemm_add_multiply.cpp # profile_gemm_add_multiply.cpp
profile_gemm_add_fastgelu.cpp # profile_gemm_add_fastgelu.cpp
profile_gemm_add_relu_add_layernorm.cpp # profile_gemm_add_relu_add_layernorm.cpp
profile_gemm_fastgelu.cpp # profile_gemm_fastgelu.cpp
profile_gemm_reduce.cpp # profile_gemm_reduce.cpp
profile_batched_gemm.cpp # profile_batched_gemm.cpp
profile_batched_gemm_gemm.cpp # profile_batched_gemm_gemm.cpp
profile_batched_gemm_add_relu_gemm_add.cpp # profile_batched_gemm_add_relu_gemm_add.cpp
profile_batched_gemm_reduce.cpp # profile_batched_gemm_reduce.cpp
profile_grouped_gemm.cpp # profile_grouped_gemm.cpp
profile_conv_fwd.cpp # profile_conv_fwd.cpp
profile_conv_fwd_bias_relu.cpp # profile_conv_fwd_bias_relu.cpp
profile_conv_fwd_bias_relu_add.cpp # profile_conv_fwd_bias_relu_add.cpp
profile_conv_bwd_data.cpp # profile_conv_bwd_data.cpp
profile_grouped_conv_fwd.cpp # profile_grouped_conv_fwd.cpp
profile_grouped_conv_bwd_weight.cpp # profile_grouped_conv_bwd_weight.cpp
profile_reduce.cpp # profile_reduce.cpp
profile_groupnorm.cpp # profile_groupnorm.cpp
profile_layernorm.cpp # profile_layernorm.cpp
profile_avg_pool2d_fwd.cpp # profile_avg_pool2d_fwd.cpp
profile_max_pool3d_fwd.cpp # profile_max_pool3d_fwd.cpp
profile_softmax.cpp # profile_softmax.cpp
profile_batchnorm_fwd.cpp # profile_batchnorm_fwd.cpp
profile_batchnorm_bwd.cpp # profile_batchnorm_bwd.cpp
profile_batchnorm_infer.cpp # profile_batchnorm_infer.cpp
profile_grouped_gemm_fastgelu.cpp # profile_grouped_gemm_fastgelu.cpp
profile_contraction_bilinear.cpp # profile_contraction_bilinear.cpp
profile_contraction_scale.cpp # profile_contraction_scale.cpp
) )
set(PROFILER_EXECUTABLE ckProfiler) set(PROFILER_EXECUTABLE ckProfiler)
...@@ -43,39 +43,39 @@ target_compile_options(${PROFILER_EXECUTABLE} PRIVATE -Wno-global-constructors) ...@@ -43,39 +43,39 @@ target_compile_options(${PROFILER_EXECUTABLE} PRIVATE -Wno-global-constructors)
target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE utility) target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE utility)
target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_gemm_instance) target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_gemm_instance)
target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_gemm_splitk_instance) # target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_gemm_splitk_instance)
target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_gemm_bilinear_instance) # target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_gemm_bilinear_instance)
target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_gemm_add_add_fastgelu_instance) # target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_gemm_add_add_fastgelu_instance)
target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_gemm_add_multiply_instance) # target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_gemm_add_multiply_instance)
target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_gemm_add_fastgelu_instance) # target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_gemm_add_fastgelu_instance)
target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_gemm_fastgelu_instance) # target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_gemm_fastgelu_instance)
target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_gemm_add_relu_add_layernorm_instance) # target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_gemm_add_relu_add_layernorm_instance)
target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_gemm_reduce_instance) # target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_gemm_reduce_instance)
target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_gemm_bias_add_reduce_instance) # target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_gemm_bias_add_reduce_instance)
target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_batched_gemm_instance) # target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_batched_gemm_instance)
target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_batched_gemm_gemm_instance) # target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_batched_gemm_gemm_instance)
target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_batched_gemm_add_relu_gemm_add_instance) # target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_batched_gemm_add_relu_gemm_add_instance)
target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_batched_gemm_reduce_instance) # target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_batched_gemm_reduce_instance)
target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_grouped_gemm_instance) # target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_grouped_gemm_instance)
target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_conv2d_fwd_instance) # target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_conv2d_fwd_instance)
target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_grouped_conv1d_fwd_instance) # target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_grouped_conv1d_fwd_instance)
target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_grouped_conv2d_fwd_instance) # target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_grouped_conv2d_fwd_instance)
target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_grouped_conv3d_fwd_instance) # target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_grouped_conv3d_fwd_instance)
target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_conv1d_bwd_data_instance) # target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_conv1d_bwd_data_instance)
target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_conv2d_bwd_data_instance) # target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_conv2d_bwd_data_instance)
target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_conv3d_bwd_data_instance) # target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_conv3d_bwd_data_instance)
target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_grouped_conv1d_bwd_weight_instance) # target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_grouped_conv1d_bwd_weight_instance)
target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_grouped_conv2d_bwd_weight_instance) # target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_grouped_conv2d_bwd_weight_instance)
target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_grouped_conv3d_bwd_weight_instance) # target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_grouped_conv3d_bwd_weight_instance)
target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_conv2d_fwd_bias_relu_instance) # target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_conv2d_fwd_bias_relu_instance)
target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_conv2d_fwd_bias_relu_add_instance) # target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_conv2d_fwd_bias_relu_add_instance)
target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_normalization_instance) # target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_normalization_instance)
target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_softmax_instance) # target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_softmax_instance)
target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_reduce_instance) # target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_reduce_instance)
target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_batchnorm_instance) # target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_batchnorm_instance)
target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_grouped_gemm_fastgelu_instance) # target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_grouped_gemm_fastgelu_instance)
target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_contraction_bilinear_instance) # target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_contraction_bilinear_instance)
target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_contraction_scale_instance) # target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_contraction_scale_instance)
target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_pool_fwd_instance) # target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_pool_fwd_instance)
rocm_install(TARGETS ${PROFILER_EXECUTABLE} COMPONENT profiler) rocm_install(TARGETS ${PROFILER_EXECUTABLE} COMPONENT profiler)
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