Skip to content

GitLab

Commit 235903ed authored by Adam Osewski's avatar Adam Osewski
Browse files

More memory-bound instances.

parent 73adb83d
Hide whitespace changes
Inline Side-by-side
Showing with 533 additions and 194 deletions
library/include/ck/library/tensor_operation_instance/gpu/grouped_gemm.hpp
View file @ 235903ed
......@@ -17,83 +17,83 @@ namespace instance {
#if defined(CK_USE_XDL)
#if defined(CK_ENABLE_FP16)
void add_device_grouped_gemm_xdl_f16_f16_f16_mk_kn_mn_instances(
std::vector<std::unique_ptr<DeviceGroupedGemm<Row,
Row,
Empty_Tuple,
Row,
F16,
F16,
Empty_Tuple,
F16,
PassThrough,
PassThrough,
PassThrough>>>& instances);
// void add_device_grouped_gemm_xdl_f16_f16_f16_mk_kn_mn_instances(
// std::vector<std::unique_ptr<DeviceGroupedGemm<Row,
// Row,
// Empty_Tuple,
// Row,
// F16,
// F16,
// Empty_Tuple,
// F16,
// PassThrough,
// PassThrough,
// PassThrough>>>& instances);
void add_device_grouped_gemm_xdl_f16_f16_f16_mk_nk_mn_instances(
std::vector<std::unique_ptr<DeviceGroupedGemm<Row,
Col,
Empty_Tuple,
Row,
F16,
F16,
Empty_Tuple,
F16,
PassThrough,
PassThrough,
PassThrough>>>& instances);
// void add_device_grouped_gemm_xdl_f16_f16_f16_mk_nk_mn_instances(
// std::vector<std::unique_ptr<DeviceGroupedGemm<Row,
// Col,
// Empty_Tuple,
// Row,
// F16,
// F16,
// Empty_Tuple,
// F16,
// PassThrough,
// PassThrough,
// PassThrough>>>& instances);
void add_device_grouped_gemm_xdl_f16_f16_f16_km_kn_mn_instances(
std::vector<std::unique_ptr<DeviceGroupedGemm<Col,
Row,
Empty_Tuple,
Row,
F16,
F16,
Empty_Tuple,
F16,
PassThrough,
PassThrough,
PassThrough>>>& instances);
// void add_device_grouped_gemm_xdl_f16_f16_f16_km_kn_mn_instances(
// std::vector<std::unique_ptr<DeviceGroupedGemm<Col,
// Row,
// Empty_Tuple,
// Row,
// F16,
// F16,
// Empty_Tuple,
// F16,
// PassThrough,
// PassThrough,
// PassThrough>>>& instances);
void add_device_grouped_gemm_xdl_f16_f16_f16_km_nk_mn_instances(
std::vector<std::unique_ptr<DeviceGroupedGemm<Col,
Col,
Empty_Tuple,
Row,
F16,
F16,
Empty_Tuple,
F16,
PassThrough,
PassThrough,
PassThrough>>>& instances);
// void add_device_grouped_gemm_xdl_f16_f16_f16_km_nk_mn_instances(
// std::vector<std::unique_ptr<DeviceGroupedGemm<Col,
// Col,
// Empty_Tuple,
// Row,
// F16,
// F16,
// Empty_Tuple,
// F16,
// PassThrough,
// PassThrough,
// PassThrough>>>& instances);
void add_device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_nk_mn_instances(
std::vector<std::unique_ptr<DeviceGroupedGemm<Row,
Col,
Empty_Tuple,
Row,
F16,
F16,
Empty_Tuple,
F16,
PassThrough,
PassThrough,
PassThrough>>>& instances);
// void add_device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_nk_mn_instances(
// std::vector<std::unique_ptr<DeviceGroupedGemm<Row,
// Col,
// Empty_Tuple,
// Row,
// F16,
// F16,
// Empty_Tuple,
// F16,
// PassThrough,
// PassThrough,
// PassThrough>>>& instances);
void add_device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_instances(
std::vector<std::unique_ptr<DeviceGroupedGemm<Row,
Row,
Empty_Tuple,
Row,
F16,
F16,
Empty_Tuple,
F16,
PassThrough,
PassThrough,
PassThrough>>>& instances);
// void add_device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_instances(
// std::vector<std::unique_ptr<DeviceGroupedGemm<Row,
// Row,
// Empty_Tuple,
// Row,
// F16,
// F16,
// Empty_Tuple,
// F16,
// PassThrough,
// PassThrough,
// PassThrough>>>& instances);
void add_device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_nk_mn_irregular_instances(
std::vector<std::unique_ptr<DeviceGroupedGemm<Row,
......@@ -108,116 +108,116 @@ void add_device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_nk_mn_irregular_instances
PassThrough,
PassThrough>>>& instances);
void add_device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_irregular_instances(
std::vector<std::unique_ptr<DeviceGroupedGemm<Row,
Row,
Empty_Tuple,
Row,
F16,
F16,
Empty_Tuple,
F16,
PassThrough,
PassThrough,
PassThrough>>>& instances);
// void add_device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_irregular_instances(
// std::vector<std::unique_ptr<DeviceGroupedGemm<Row,
// Row,
// Empty_Tuple,
// Row,
// F16,
// F16,
// Empty_Tuple,
// F16,
// PassThrough,
// PassThrough,
// PassThrough>>>& instances);
void add_device_grouped_gemm_multiple_d_xdl_two_stage_f16_f16_f16_mk_kn_mn_instances(
std::vector<std::unique_ptr<DeviceGroupedGemm<Row,
Row,
Empty_Tuple,
Row,
F16,
F16,
Empty_Tuple,
F16,
PassThrough,
PassThrough,
PassThrough>>>& instances);
// void add_device_grouped_gemm_multiple_d_xdl_two_stage_f16_f16_f16_mk_kn_mn_instances(
// std::vector<std::unique_ptr<DeviceGroupedGemm<Row,
// Row,
// Empty_Tuple,
// Row,
// F16,
// F16,
// Empty_Tuple,
// F16,
// PassThrough,
// PassThrough,
// PassThrough>>>& instances);
#endif
#if defined(CK_ENABLE_FP16) && defined(CK_ENABLE_FP8)
void add_device_grouped_gemm_xdl_splitk_f16_f8_f16_mk_kn_mn_irregular_instances(
std::vector<std::unique_ptr<DeviceGroupedGemm<Row,
Row,
Empty_Tuple,
Row,
F16,
F8,
Empty_Tuple,
F16,
PassThrough,
PassThrough,
PassThrough>>>& instances);
// #if defined(CK_ENABLE_FP16) && defined(CK_ENABLE_FP8)
// void add_device_grouped_gemm_xdl_splitk_f16_f8_f16_mk_kn_mn_irregular_instances(
// std::vector<std::unique_ptr<DeviceGroupedGemm<Row,
// Row,
// Empty_Tuple,
// Row,
// F16,
// F8,
// Empty_Tuple,
// F16,
// PassThrough,
// PassThrough,
// PassThrough>>>& instances);
void add_device_grouped_gemm_xdl_splitk_f8_f16_f16_mk_kn_mn_irregular_instances(
std::vector<std::unique_ptr<DeviceGroupedGemm<Row,
Row,
Empty_Tuple,
Row,
F8,
F16,
Empty_Tuple,
F16,
PassThrough,
PassThrough,
PassThrough>>>& instances);
#endif
// void add_device_grouped_gemm_xdl_splitk_f8_f16_f16_mk_kn_mn_irregular_instances(
// std::vector<std::unique_ptr<DeviceGroupedGemm<Row,
// Row,
// Empty_Tuple,
// Row,
// F8,
// F16,
// Empty_Tuple,
// F16,
// PassThrough,
// PassThrough,
// PassThrough>>>& instances);
// #endif
#if defined(CK_ENABLE_BF16)
void add_device_grouped_gemm_multiple_d_xdl_two_stage_bf16_bf16_bf16_mk_kn_mn_instances(
std::vector<std::unique_ptr<DeviceGroupedGemm<Row,
Row,
Empty_Tuple,
Row,
BF16,
BF16,
Empty_Tuple,
BF16,
PassThrough,
PassThrough,
PassThrough>>>& instances);
// #if defined(CK_ENABLE_BF16)
// void add_device_grouped_gemm_multiple_d_xdl_two_stage_bf16_bf16_bf16_mk_kn_mn_instances(
// std::vector<std::unique_ptr<DeviceGroupedGemm<Row,
// Row,
// Empty_Tuple,
// Row,
// BF16,
// BF16,
// Empty_Tuple,
// BF16,
// PassThrough,
// PassThrough,
// PassThrough>>>& instances);
void add_device_grouped_gemm_multiple_d_xdl_two_stage_bf16_bf16_bf16_mk_nk_mn_instances(
std::vector<std::unique_ptr<DeviceGroupedGemm<Row,
Col,
Empty_Tuple,
Row,
BF16,
BF16,
Empty_Tuple,
BF16,
PassThrough,
PassThrough,
PassThrough>>>& instances);
#endif
// void add_device_grouped_gemm_multiple_d_xdl_two_stage_bf16_bf16_bf16_mk_nk_mn_instances(
// std::vector<std::unique_ptr<DeviceGroupedGemm<Row,
// Col,
// Empty_Tuple,
// Row,
// BF16,
// BF16,
// Empty_Tuple,
// BF16,
// PassThrough,
// PassThrough,
// PassThrough>>>& instances);
// #endif
#if defined(CK_ENABLE_BF16) && defined(CK_ENABLE_INT8)
void add_device_grouped_gemm_multiple_d_xdl_two_stage_bf16_i8_bf16_mk_kn_mn_instances(
std::vector<std::unique_ptr<DeviceGroupedGemm<Row,
Row,
Empty_Tuple,
Row,
BF16,
I8,
Empty_Tuple,
BF16,
PassThrough,
PassThrough,
PassThrough>>>& instances);
// #if defined(CK_ENABLE_BF16) && defined(CK_ENABLE_INT8)
// void add_device_grouped_gemm_multiple_d_xdl_two_stage_bf16_i8_bf16_mk_kn_mn_instances(
// std::vector<std::unique_ptr<DeviceGroupedGemm<Row,
// Row,
// Empty_Tuple,
// Row,
// BF16,
// I8,
// Empty_Tuple,
// BF16,
// PassThrough,
// PassThrough,
// PassThrough>>>& instances);
void add_device_grouped_gemm_multiple_d_xdl_two_stage_bf16_i8_bf16_mk_nk_mn_instances(
std::vector<std::unique_ptr<DeviceGroupedGemm<Row,
Col,
Empty_Tuple,
Row,
BF16,
I8,
Empty_Tuple,
BF16,
PassThrough,
PassThrough,
PassThrough>>>& instances);
#endif
// void add_device_grouped_gemm_multiple_d_xdl_two_stage_bf16_i8_bf16_mk_nk_mn_instances(
// std::vector<std::unique_ptr<DeviceGroupedGemm<Row,
// Col,
// Empty_Tuple,
// Row,
// BF16,
// I8,
// Empty_Tuple,
// BF16,
// PassThrough,
// PassThrough,
// PassThrough>>>& instances);
// #endif
#endif // CK_USE_XDL
template <typename ALayout,
typename BLayout,
......@@ -260,30 +260,30 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
if constexpr(is_same_v<ALayout, Row> && is_same_v<BLayout, Row> &&
is_same_v<ELayout, Row>)
{
add_device_grouped_gemm_xdl_f16_f16_f16_mk_kn_mn_instances(op_ptrs);
add_device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_instances(op_ptrs);
add_device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_irregular_instances(
op_ptrs);
add_device_grouped_gemm_multiple_d_xdl_two_stage_f16_f16_f16_mk_kn_mn_instances(
op_ptrs);
// add_device_grouped_gemm_xdl_f16_f16_f16_mk_kn_mn_instances(op_ptrs);
// add_device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_instances(op_ptrs);
// add_device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_irregular_instances(
// op_ptrs);
// add_device_grouped_gemm_multiple_d_xdl_two_stage_f16_f16_f16_mk_kn_mn_instances(
// op_ptrs);
}
else if constexpr(is_same_v<ALayout, Row> && is_same_v<BLayout, Col> &&
is_same_v<ELayout, Row>)
{
add_device_grouped_gemm_xdl_f16_f16_f16_mk_nk_mn_instances(op_ptrs);
add_device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_nk_mn_instances(op_ptrs);
// add_device_grouped_gemm_xdl_f16_f16_f16_mk_nk_mn_instances(op_ptrs);
// add_device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_nk_mn_instances(op_ptrs);
add_device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_nk_mn_irregular_instances(
op_ptrs);
}
else if constexpr(is_same_v<ALayout, Col> && is_same_v<BLayout, Row> &&
is_same_v<ELayout, Row>)
{
add_device_grouped_gemm_xdl_f16_f16_f16_km_kn_mn_instances(op_ptrs);
// add_device_grouped_gemm_xdl_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<ELayout, Row>)
{
add_device_grouped_gemm_xdl_f16_f16_f16_km_nk_mn_instances(op_ptrs);
// add_device_grouped_gemm_xdl_f16_f16_f16_km_nk_mn_instances(op_ptrs);
}
}
#endif
......@@ -294,7 +294,7 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
if constexpr(is_same_v<ALayout, Row> && is_same_v<BLayout, Row> &&
is_same_v<ELayout, Row>)
{
add_device_grouped_gemm_xdl_splitk_f16_f8_f16_mk_kn_mn_irregular_instances(op_ptrs);
// add_device_grouped_gemm_xdl_splitk_f16_f8_f16_mk_kn_mn_irregular_instances(op_ptrs);
}
}
else if constexpr(is_same_v<ADataType, f8_t> && is_same_v<BDataType, half_t> &&
......@@ -303,7 +303,7 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
if constexpr(is_same_v<ALayout, Row> && is_same_v<BLayout, Row> &&
is_same_v<ELayout, Row>)
{
add_device_grouped_gemm_xdl_splitk_f8_f16_f16_mk_kn_mn_irregular_instances(op_ptrs);
// add_device_grouped_gemm_xdl_splitk_f8_f16_f16_mk_kn_mn_irregular_instances(op_ptrs);
}
}
#endif
......@@ -314,14 +314,14 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
if constexpr(is_same_v<ALayout, Row> && is_same_v<BLayout, Row> &&
is_same_v<ELayout, Row>)
{
add_device_grouped_gemm_multiple_d_xdl_two_stage_bf16_i8_bf16_mk_kn_mn_instances(
op_ptrs);
// add_device_grouped_gemm_multiple_d_xdl_two_stage_bf16_i8_bf16_mk_kn_mn_instances(
// op_ptrs);
}
else if constexpr(is_same_v<ALayout, Row> && is_same_v<BLayout, Col> &&
is_same_v<ELayout, Row>)
{
add_device_grouped_gemm_multiple_d_xdl_two_stage_bf16_i8_bf16_mk_nk_mn_instances(
op_ptrs);
// add_device_grouped_gemm_multiple_d_xdl_two_stage_bf16_i8_bf16_mk_nk_mn_instances(
// op_ptrs);
}
}
#endif
......@@ -332,14 +332,14 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
if constexpr(is_same_v<ALayout, Row> && is_same_v<BLayout, Row> &&
is_same_v<ELayout, Row>)
{
add_device_grouped_gemm_multiple_d_xdl_two_stage_bf16_bf16_bf16_mk_kn_mn_instances(
op_ptrs);
// add_device_grouped_gemm_multiple_d_xdl_two_stage_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<ELayout, Row>)
{
add_device_grouped_gemm_multiple_d_xdl_two_stage_bf16_bf16_bf16_mk_nk_mn_instances(
op_ptrs);
// add_device_grouped_gemm_multiple_d_xdl_two_stage_bf16_bf16_bf16_mk_nk_mn_instances(
// op_ptrs);
}
}
#endif
......
library/include/ck/library/tensor_operation_instance/gpu/grouped_gemm_multiple_d.hpp
View file @ 235903ed
......@@ -31,6 +31,58 @@ void add_device_grouped_gemm_multi_d_splitk_cshuffle_f16_f16_f16_mk_nk_mn_irregu
PassThrough,
PassThrough>>>& instances);
void add_device_grouped_gemm_multi_d_splitk_cshuffle_f16_f16_f16_mk_nk_mn_mem_pv1_pf1_instances(
std::vector<std::unique_ptr<DeviceGroupedGemm<Row,
Col,
Empty_Tuple,
Row,
F16,
F16,
Empty_Tuple,
F16,
PassThrough,
PassThrough,
PassThrough>>>& instances);
void add_device_grouped_gemm_multi_d_splitk_cshuffle_f16_f16_f16_mk_nk_mn_mem_pv1_pf2_instances(
std::vector<std::unique_ptr<DeviceGroupedGemm<Row,
Col,
Empty_Tuple,
Row,
F16,
F16,
Empty_Tuple,
F16,
PassThrough,
PassThrough,
PassThrough>>>& instances);
void add_device_grouped_gemm_multi_d_splitk_cshuffle_f16_f16_f16_mk_nk_mn_mem_pv2_pf1_instances(
std::vector<std::unique_ptr<DeviceGroupedGemm<Row,
Col,
Empty_Tuple,
Row,
F16,
F16,
Empty_Tuple,
F16,
PassThrough,
PassThrough,
PassThrough>>>& instances);
void add_device_grouped_gemm_multi_d_splitk_cshuffle_f16_f16_f16_mk_nk_mn_mem_pv2_pf2_instances(
std::vector<std::unique_ptr<DeviceGroupedGemm<Row,
Col,
Empty_Tuple,
Row,
F16,
F16,
Empty_Tuple,
F16,
PassThrough,
PassThrough,
PassThrough>>>& instances);
void add_device_grouped_gemm_multi_d_splitk_cshuffle_f16_f16_f16_mk_kn_mn_irregular_instances_pipeline_v1(
std::vector<std::unique_ptr<DeviceGroupedGemm<Row,
Row,
......@@ -130,6 +182,14 @@ struct DeviceOperationInstanceFactory<
#if defined(CK_ENABLE_FP16)
add_device_grouped_gemm_multi_d_splitk_cshuffle_f16_f16_f16_mk_nk_mn_irregular_instances(
op_ptrs);
add_device_grouped_gemm_multi_d_splitk_cshuffle_f16_f16_f16_mk_nk_mn_mem_pv1_pf1_instances(
op_ptrs);
add_device_grouped_gemm_multi_d_splitk_cshuffle_f16_f16_f16_mk_nk_mn_mem_pv1_pf2_instances(
op_ptrs);
add_device_grouped_gemm_multi_d_splitk_cshuffle_f16_f16_f16_mk_nk_mn_mem_pv2_pf1_instances(
op_ptrs);
add_device_grouped_gemm_multi_d_splitk_cshuffle_f16_f16_f16_mk_nk_mn_mem_pv2_pf2_instances(
op_ptrs);
#endif
}
}
......
library/include/ck/library/tensor_operation_instance/gpu/grouped_gemm_multiple_d/device_grouped_gemm_multiple_d_splitk_xdl_cshuffle_f16_f16_f16_mk_kn_mn_irregular_instance.hpp
View file @ 235903ed
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
#pragma once
#include <cstdlib>
......
library/include/ck/library/tensor_operation_instance/gpu/grouped_gemm_multiple_d/device_grouped_gemm_multiple_d_splitk_xdl_cshuffle_f16_f16_f16_mk_nk_mn_instance.hpp 0 → 100644
View file @ 235903ed
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
#pragma once
#include <cstdlib>
#include "ck/ck.hpp"
#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
#include "ck/tensor_operation/gpu/device/impl/device_grouped_gemm_multiple_d_splitk_xdl_cshuffle_tile_loop.hpp"
#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
#include "ck/tensor_operation/gpu/grid/gridwise_gemm_pipeline_selector.hpp"
#include "ck/utility/loop_scheduler.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 Empty_Tuple = ck::Tuple<>;
using ck::tensor_operation::device::GemmSpecialization;
using PassThrough = ck::tensor_operation::element_wise::PassThrough;
template <GemmSpecialization GemmSpec,
ck::index_t NumPrefetch,
ck::PipelineVersion Pipeline,
LoopScheduler Scheduler = LoopScheduler::Default>
using device_ggemm_md_splitk_xdl_cshuffle_f16_f16_f16_mk_kn_mn_memory_instances = std::tuple<
// clang-format off
//#########################################| A| B| Ds| E| AData| BData| AccData| CShuffle| DsData| EData| 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| CShuffle| CShuffle| CBlockTransferClusterLengths| CBlockTransfer| PipelineVersion| LoopScheduler|
//#########################################| Layout| Layout| Layout| Layout| Type| Type| Type| DataType| 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| MXdlPerWave| NXdlPerWave| _MBlock_MWaveMPerXdl| ScalarPerVector| | |
//#########################################| | | | | | | | | | | 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| | |
//#########################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
// Memory friendly
DeviceGroupedGemmMultipleDSplitKXdlCShuffle< Row, Col, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmSpec, NumPrefetch, 256, 256, 32, 64, 8, 8, 32, 32, 2, 1, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 4, Pipeline, Scheduler>,
DeviceGroupedGemmMultipleDSplitKXdlCShuffle< Row, Col, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmSpec, NumPrefetch, 256, 256, 16, 64, 8, 8, 16, 16, 4, 1, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<8, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 2, Pipeline, Scheduler>,
DeviceGroupedGemmMultipleDSplitKXdlCShuffle< Row, Col, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmSpec, NumPrefetch, 128, 128, 32, 64, 8, 8, 32, 32, 2, 1, S<8, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<8, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 8>, 4, Pipeline, Scheduler>,
DeviceGroupedGemmMultipleDSplitKXdlCShuffle< Row, Col, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmSpec, NumPrefetch, 128, 128, 16, 64, 8, 8, 16, 16, 4, 1, S<8, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<8, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 8>, 2, Pipeline, Scheduler>,
DeviceGroupedGemmMultipleDSplitKXdlCShuffle< Row, Col, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmSpec, NumPrefetch, 128, 64, 32, 64, 8, 8, 32, 32, 1, 1, S<8, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<8, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 8>, 4, Pipeline, Scheduler>,
DeviceGroupedGemmMultipleDSplitKXdlCShuffle< Row, Col, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmSpec, NumPrefetch, 128, 64, 16, 64, 8, 8, 16, 16, 2, 1, S<8, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<8, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 8>, 2, Pipeline, Scheduler>,
DeviceGroupedGemmMultipleDSplitKXdlCShuffle< Row, Col, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmSpec, NumPrefetch, 128, 32, 16, 64, 8, 8, 16, 16, 1, 1, S<8, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<8, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 8>, 2, Pipeline, Scheduler>,
DeviceGroupedGemmMultipleDSplitKXdlCShuffle< Row, Col, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmSpec, NumPrefetch, 64, 16, 16, 128, 8, 8, 16, 16, 1, 1, S<16, 4, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<16, 4, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 4>, 4, Pipeline, Scheduler>,
DeviceGroupedGemmMultipleDSplitKXdlCShuffle< Row, Col, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmSpec, NumPrefetch, 64, 16, 16, 64, 8, 8, 16, 16, 1, 1, S<8, 8, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<8, 8, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 4>, 4, Pipeline, Scheduler>,
DeviceGroupedGemmMultipleDSplitKXdlCShuffle< Row, Col, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmSpec, NumPrefetch, 128, 16, 32, 64, 8, 8, 16, 16, 1, 1, S<8, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<8, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 8>, 4, Pipeline, Scheduler>,
DeviceGroupedGemmMultipleDSplitKXdlCShuffle< Row, Col, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmSpec, NumPrefetch, 128, 16, 64, 64, 8, 8, 16, 16, 1, 2, S<8, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<8, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 8>, 4, Pipeline, Scheduler>,
DeviceGroupedGemmMultipleDSplitKXdlCShuffle< Row, Col, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmSpec, NumPrefetch, 128, 32, 64, 64, 8, 8, 32, 32, 1, 1, S<8, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<8, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 8>, 8, Pipeline, Scheduler>,
DeviceGroupedGemmMultipleDSplitKXdlCShuffle< Row, Col, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmSpec, NumPrefetch, 128, 16, 128, 64, 8, 8, 16, 16, 1, 4, S<8, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<8, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 8>, 4, Pipeline, Scheduler>,
DeviceGroupedGemmMultipleDSplitKXdlCShuffle< Row, Col, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmSpec, NumPrefetch, 128, 32, 128, 64, 8, 8, 32, 32, 1, 2, S<8, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<8, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 8>, 8, Pipeline, Scheduler>,
DeviceGroupedGemmMultipleDSplitKXdlCShuffle< Row, Col, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmSpec, NumPrefetch, 256, 16, 256, 64, 8, 8, 16, 16, 1, 4, S<8, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<8, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 16>, 4, Pipeline, Scheduler>,
DeviceGroupedGemmMultipleDSplitKXdlCShuffle< Row, Col, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmSpec, NumPrefetch, 256, 32, 256, 64, 8, 8, 32, 32, 1, 2, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 16>, 8, Pipeline, Scheduler>
>;
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck
library/src/tensor_operation_instance/gpu/grouped_gemm_multiple_d/CMakeLists.txt
View file @ 235903ed
# ONLY XDL_KERNELS
add_instance_library(device_grouped_gemm_multiple_d_instance
device_grouped_gemm_multiple_d_splitk_xdl_cshuffle_f16_f16_f16_mk_nk_mn_irregular_instance.cpp
device_grouped_gemm_multiple_d_splitk_xdl_cshuffle_f16_f16_f16_mk_nk_mn_memory_pv1_pf1_instance.cpp
device_grouped_gemm_multiple_d_splitk_xdl_cshuffle_f16_f16_f16_mk_nk_mn_memory_pv1_pf2_instance.cpp
device_grouped_gemm_multiple_d_splitk_xdl_cshuffle_f16_f16_f16_mk_nk_mn_memory_pv2_pf1_instance.cpp
device_grouped_gemm_multiple_d_splitk_xdl_cshuffle_f16_f16_f16_mk_nk_mn_memory_pv2_pf2_instance.cpp
device_grouped_gemm_multiple_d_splitk_xdl_cshuffle_f16_f16_f16_mk_kn_mn_irregular_instance_pipeline_v1.cpp
device_grouped_gemm_multiple_d_splitk_xdl_cshuffle_f16_f16_f16_mk_kn_mn_irregular_instance_pipeline_v1_interwave.cpp
device_grouped_gemm_multiple_d_splitk_xdl_cshuffle_f16_f16_f16_mk_kn_mn_irregular_instance_pipeline_v2.cpp
......
library/src/tensor_operation_instance/gpu/grouped_gemm_multiple_d/device_grouped_gemm_multiple_d_splitk_xdl_cshuffle_f16_f16_f16_mk_nk_mn_memory_pv1_pf1_instance.cpp 0 → 100644
View file @ 235903ed
// SPDX-License-Identifier: MIT
// Copyright (c) 2024, Advanced Micro Devices, Inc. All rights reserved.
#include <cstdlib>
#include "ck/ck.hpp"
#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
#include "ck/library/tensor_operation_instance/gpu/grouped_gemm_multiple_d/device_grouped_gemm_multiple_d_splitk_xdl_cshuffle_f16_f16_f16_mk_nk_mn_instance.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_grouped_gemm_multiple_d_splitk_xdl_cshuffle_tile_loop.hpp"
#include "ck/tensor_operation/gpu/grid/gridwise_gemm_pipeline_selector.hpp"
namespace ck {
namespace tensor_operation {
namespace device {
namespace instance {
using F16 = ck::half_t;
using Row = ck::tensor_layout::gemm::RowMajor;
using Col = ck::tensor_layout::gemm::ColumnMajor;
using Empty_Tuple = ck::Tuple<>;
using PassThrough = ck::tensor_operation::element_wise::PassThrough;
static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
static constexpr ck::index_t NumPrefetchK = 1;
void add_device_grouped_gemm_multi_d_splitk_cshuffle_f16_f16_f16_mk_nk_mn_mem_pv1_pf1_instances(
std::vector<std::unique_ptr<DeviceGroupedGemm<Row,
Col,
Empty_Tuple,
Row,
F16,
F16,
Empty_Tuple,
F16,
PassThrough,
PassThrough,
PassThrough>>>& instances)
{
add_device_operation_instances(
instances,
device_ggemm_md_splitk_xdl_cshuffle_f16_f16_f16_mk_kn_mn_memory_instances<
GemmMNKPadding,
NumPrefetchK,
ck::PipelineVersion::v1>{});
}
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck
library/src/tensor_operation_instance/gpu/grouped_gemm_multiple_d/device_grouped_gemm_multiple_d_splitk_xdl_cshuffle_f16_f16_f16_mk_nk_mn_memory_pv1_pf2_instance.cpp 0 → 100644
View file @ 235903ed
// SPDX-License-Identifier: MIT
// Copyright (c) 2024, Advanced Micro Devices, Inc. All rights reserved.
#include <cstdlib>
#include "ck/ck.hpp"
#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
#include "ck/library/tensor_operation_instance/gpu/grouped_gemm_multiple_d/device_grouped_gemm_multiple_d_splitk_xdl_cshuffle_f16_f16_f16_mk_nk_mn_instance.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_grouped_gemm_multiple_d_splitk_xdl_cshuffle_tile_loop.hpp"
#include "ck/tensor_operation/gpu/grid/gridwise_gemm_pipeline_selector.hpp"
namespace ck {
namespace tensor_operation {
namespace device {
namespace instance {
using F16 = ck::half_t;
using Row = ck::tensor_layout::gemm::RowMajor;
using Col = ck::tensor_layout::gemm::ColumnMajor;
using Empty_Tuple = ck::Tuple<>;
using PassThrough = ck::tensor_operation::element_wise::PassThrough;
static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
static constexpr ck::index_t NumPrefetchK = 2;
void add_device_grouped_gemm_multi_d_splitk_cshuffle_f16_f16_f16_mk_nk_mn_mem_pv1_pf2_instances(
std::vector<std::unique_ptr<DeviceGroupedGemm<Row,
Col,
Empty_Tuple,
Row,
F16,
F16,
Empty_Tuple,
F16,
PassThrough,
PassThrough,
PassThrough>>>& instances)
{
add_device_operation_instances(
instances,
device_ggemm_md_splitk_xdl_cshuffle_f16_f16_f16_mk_kn_mn_memory_instances<
GemmMNKPadding,
NumPrefetchK,
ck::PipelineVersion::v1>{});
}
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck
library/src/tensor_operation_instance/gpu/grouped_gemm_multiple_d/device_grouped_gemm_multiple_d_splitk_xdl_cshuffle_f16_f16_f16_mk_nk_mn_memory_pv2_pf1_instance.cpp 0 → 100644
View file @ 235903ed
// SPDX-License-Identifier: MIT
// Copyright (c) 2024, Advanced Micro Devices, Inc. All rights reserved.
#include <cstdlib>
#include "ck/ck.hpp"
#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
#include "ck/library/tensor_operation_instance/gpu/grouped_gemm_multiple_d/device_grouped_gemm_multiple_d_splitk_xdl_cshuffle_f16_f16_f16_mk_nk_mn_instance.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_grouped_gemm_multiple_d_splitk_xdl_cshuffle_tile_loop.hpp"
#include "ck/tensor_operation/gpu/grid/gridwise_gemm_pipeline_selector.hpp"
namespace ck {
namespace tensor_operation {
namespace device {
namespace instance {
using F16 = ck::half_t;
using Row = ck::tensor_layout::gemm::RowMajor;
using Col = ck::tensor_layout::gemm::ColumnMajor;
using Empty_Tuple = ck::Tuple<>;
using PassThrough = ck::tensor_operation::element_wise::PassThrough;
static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
static constexpr ck::index_t NumPrefetchK = 1;
void add_device_grouped_gemm_multi_d_splitk_cshuffle_f16_f16_f16_mk_nk_mn_mem_pv2_pf1_instances(
std::vector<std::unique_ptr<DeviceGroupedGemm<Row,
Col,
Empty_Tuple,
Row,
F16,
F16,
Empty_Tuple,
F16,
PassThrough,
PassThrough,
PassThrough>>>& instances)
{
add_device_operation_instances(
instances,
device_ggemm_md_splitk_xdl_cshuffle_f16_f16_f16_mk_kn_mn_memory_instances<
GemmMNKPadding,
NumPrefetchK,
ck::PipelineVersion::v2>{});
}
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck
library/src/tensor_operation_instance/gpu/grouped_gemm_multiple_d/device_grouped_gemm_multiple_d_splitk_xdl_cshuffle_f16_f16_f16_mk_nk_mn_memory_pv2_pf2_instance.cpp 0 → 100644
View file @ 235903ed
// SPDX-License-Identifier: MIT
// Copyright (c) 2024, Advanced Micro Devices, Inc. All rights reserved.
#include <cstdlib>
#include "ck/ck.hpp"
#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
#include "ck/library/tensor_operation_instance/gpu/grouped_gemm_multiple_d/device_grouped_gemm_multiple_d_splitk_xdl_cshuffle_f16_f16_f16_mk_nk_mn_instance.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_grouped_gemm_multiple_d_splitk_xdl_cshuffle_tile_loop.hpp"
#include "ck/tensor_operation/gpu/grid/gridwise_gemm_pipeline_selector.hpp"
namespace ck {
namespace tensor_operation {
namespace device {
namespace instance {
using F16 = ck::half_t;
using Row = ck::tensor_layout::gemm::RowMajor;
using Col = ck::tensor_layout::gemm::ColumnMajor;
using Empty_Tuple = ck::Tuple<>;
using PassThrough = ck::tensor_operation::element_wise::PassThrough;
static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
static constexpr ck::index_t NumPrefetchK = 2;
void add_device_grouped_gemm_multi_d_splitk_cshuffle_f16_f16_f16_mk_nk_mn_mem_pv2_pf2_instances(
std::vector<std::unique_ptr<DeviceGroupedGemm<Row,
Col,
Empty_Tuple,
Row,
F16,
F16,
Empty_Tuple,
F16,
PassThrough,
PassThrough,
PassThrough>>>& instances)
{
add_device_operation_instances(
instances,
device_ggemm_md_splitk_xdl_cshuffle_f16_f16_f16_mk_kn_mn_memory_instances<
GemmMNKPadding,
NumPrefetchK,
ck::PipelineVersion::v2>{});
}
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck
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