Skip to content

GitLab

Unverified Commit 4173b984 authored by Rostyslav Geyyer's avatar Rostyslav Geyyer Committed by GitHub
Browse files

Merge branch 'develop' into lwpck-756

parents 6de7d10d 85e2e1e2
Hide whitespace changes
Inline Side-by-side
Showing with 840 additions and 255 deletions
library/include/ck/library/tensor_operation_instance/device_operation_instance_factory.hpp
View file @ 4173b984
......@@ -36,6 +36,7 @@ using F64_Tuple = ck::Tuple<F64>;
using F32_Tuple = ck::Tuple<F32>;
using I32_Tuple = ck::Tuple<I32>;
using I32_F32_Tuple = ck::Tuple<I32, F32>;
using I8_Tuple = ck::Tuple<I8>;
using F32_F32_Tuple = ck::Tuple<F32, F32>;
......
library/include/ck/library/tensor_operation_instance/gpu/gemm.hpp
View file @ 4173b984
......@@ -23,7 +23,7 @@ void add_device_gemm_dl_f16_f16_f16_km_kn_mn_instances(
DeviceGemm<Col, Row, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
instances);
void add_device_gemm_dl_dpp8_f16_f16_f16_km_kn_mn_instances(
void add_device_gemm_dpp_f16_f16_f16_km_kn_mn_instances(
std::vector<std::unique_ptr<
DeviceGemm<Col, Row, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
instances);
......@@ -38,7 +38,7 @@ void add_device_gemm_dl_f16_f16_f16_km_nk_mn_instances(
DeviceGemm<Col, Col, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
instances);
void add_device_gemm_dl_dpp8_f16_f16_f16_km_nk_mn_instances(
void add_device_gemm_dpp_f16_f16_f16_km_nk_mn_instances(
std::vector<std::unique_ptr<
DeviceGemm<Col, Col, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
instances);
......@@ -53,7 +53,7 @@ void add_device_gemm_dl_f16_f16_f16_mk_kn_mn_instances(
DeviceGemm<Row, Row, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
instances);
void add_device_gemm_dl_dpp8_f16_f16_f16_mk_kn_mn_instances(
void add_device_gemm_dpp_f16_f16_f16_mk_kn_mn_instances(
std::vector<std::unique_ptr<
DeviceGemm<Row, Row, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
instances);
......@@ -68,7 +68,7 @@ void add_device_gemm_dl_f16_f16_f16_mk_nk_mn_instances(
DeviceGemm<Row, Col, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
instances);
void add_device_gemm_dl_dpp8_f16_f16_f16_mk_nk_mn_instances(
void add_device_gemm_dpp_f16_f16_f16_mk_nk_mn_instances(
std::vector<std::unique_ptr<
DeviceGemm<Row, Col, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
instances);
......@@ -374,7 +374,7 @@ struct DeviceOperationInstanceFactory<
#ifdef DL_KERNELS
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_dpp8_f16_f16_f16_mk_kn_mn_instances(op_ptrs);
add_device_gemm_dpp_f16_f16_f16_mk_kn_mn_instances(op_ptrs);
#endif
add_device_gemm_xdl_c_shuffle_f16_f16_f16_mk_kn_mn_instances(op_ptrs);
}
......@@ -385,7 +385,7 @@ struct DeviceOperationInstanceFactory<
#ifdef DL_KERNELS
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_dpp8_f16_f16_f16_mk_nk_mn_instances(op_ptrs);
add_device_gemm_dpp_f16_f16_f16_mk_nk_mn_instances(op_ptrs);
#endif
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);
......@@ -397,7 +397,7 @@ struct DeviceOperationInstanceFactory<
#ifdef DL_KERNELS
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_dl_dpp8_f16_f16_f16_km_kn_mn_instances(op_ptrs);
add_device_gemm_dpp_f16_f16_f16_km_kn_mn_instances(op_ptrs);
#endif
add_device_gemm_xdl_c_shuffle_f16_f16_f16_km_kn_mn_instances(op_ptrs);
}
......@@ -408,7 +408,7 @@ struct DeviceOperationInstanceFactory<
#ifdef DL_KERNELS
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_dl_dpp8_f16_f16_f16_km_nk_mn_instances(op_ptrs);
add_device_gemm_dpp_f16_f16_f16_km_nk_mn_instances(op_ptrs);
#endif
add_device_gemm_xdl_c_shuffle_f16_f16_f16_km_nk_mn_instances(op_ptrs);
}
......
library/include/ck/library/tensor_operation_instance/gpu/gemm_bilinear.hpp
View file @ 4173b984
......@@ -69,6 +69,58 @@ void add_device_gemm_bilinear_xdl_c_shuffle_f16_f16_f16_f16_mk_nk_mn_mn_instance
PassThrough,
Bilinear>>>& instances);
void add_device_gemm_bilinear_wmma_c_shuffle_i8_i8_i8_i8_mk_kn_mn_mn_instances(
std::vector<std::unique_ptr<DeviceGemmMultipleD<Row,
Row,
Row_Tuple,
Row,
I8,
I8,
I8_Tuple,
I8,
PassThrough,
PassThrough,
Bilinear>>>& instances);
void add_device_gemm_bilinear_wmma_c_shuffle_i8_i8_i8_i8_mk_nk_mn_mn_instances(
std::vector<std::unique_ptr<DeviceGemmMultipleD<Row,
Col,
Row_Tuple,
Row,
I8,
I8,
I8_Tuple,
I8,
PassThrough,
PassThrough,
Bilinear>>>& instances);
void add_device_gemm_bilinear_wmma_c_shuffle_i8_i8_i8_i8_km_kn_mn_mn_instances(
std::vector<std::unique_ptr<DeviceGemmMultipleD<Col,
Row,
Row_Tuple,
Row,
I8,
I8,
I8_Tuple,
I8,
PassThrough,
PassThrough,
Bilinear>>>& instances);
void add_device_gemm_bilinear_wmma_c_shuffle_i8_i8_i8_i8_km_nk_mn_mn_instances(
std::vector<std::unique_ptr<DeviceGemmMultipleD<Col,
Col,
Row_Tuple,
Row,
I8,
I8,
I8_Tuple,
I8,
PassThrough,
PassThrough,
Bilinear>>>& instances);
// GEMM + Bilinear
template <typename ALayout,
typename BLayout,
......@@ -135,6 +187,30 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGemmMu
op_ptrs);
}
}
else if constexpr(is_same_v<ADataType, std::int8_t> && is_same_v<BDataType, std::int8_t> &&
is_same_v<DDataType, std::int8_t> && is_same_v<EDataType, std::int8_t>)
{
if constexpr(is_same_v<ALayout, Row> && is_same_v<BLayout, Row> &&
is_same_v<DLayout, Row> && is_same_v<ELayout, Row>)
{
add_device_gemm_bilinear_wmma_c_shuffle_i8_i8_i8_i8_mk_kn_mn_mn_instances(op_ptrs);
}
else if constexpr(is_same_v<ALayout, Row> && is_same_v<BLayout, Col> &&
is_same_v<DLayout, Row> && is_same_v<ELayout, Row>)
{
add_device_gemm_bilinear_wmma_c_shuffle_i8_i8_i8_i8_mk_nk_mn_mn_instances(op_ptrs);
}
else if constexpr(is_same_v<ALayout, Col> && is_same_v<BLayout, Row> &&
is_same_v<DLayout, Row> && is_same_v<ELayout, Row>)
{
add_device_gemm_bilinear_wmma_c_shuffle_i8_i8_i8_i8_km_kn_mn_mn_instances(op_ptrs);
}
else if constexpr(is_same_v<ALayout, Col> && is_same_v<BLayout, Col> &&
is_same_v<DLayout, Row> && is_same_v<ELayout, Row>)
{
add_device_gemm_bilinear_wmma_c_shuffle_i8_i8_i8_i8_km_nk_mn_mn_instances(op_ptrs);
}
}
return op_ptrs;
}
......
library/include/ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv2d_fwd_wmma_instance.hpp 0 → 100644
View file @ 4173b984
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
#include "ck/ck.hpp"
#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
#include "ck/tensor_operation/gpu/device/convolution_forward_specialization.hpp"
#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_wmma_cshuffle.hpp"
namespace ck {
namespace tensor_operation {
namespace device {
namespace instance {
using BF16 = ck::bhalf_t;
using F16 = ck::half_t;
using F32 = float;
using I8 = int8_t;
using I32 = int32_t;
using Empty_Tuple = ck::Tuple<>;
template <ck::index_t... Is>
using S = ck::Sequence<Is...>;
using NHWGC = ck::tensor_layout::convolution::NHWGC;
using GNHWC = ck::tensor_layout::convolution::GNHWC;
using GKYXC = ck::tensor_layout::convolution::GKYXC;
using NHWGK = ck::tensor_layout::convolution::NHWGK;
using GNHWK = ck::tensor_layout::convolution::GNHWK;
using PassThrough = ck::tensor_operation::element_wise::PassThrough;
static constexpr auto ConvFwdDefault =
ck::tensor_operation::device::ConvolutionForwardSpecialization::Default;
static constexpr auto ConvFwd1x1P0 =
ck::tensor_operation::device::ConvolutionForwardSpecialization::Filter1x1Pad0;
static constexpr auto ConvFwd1x1S1P0 =
ck::tensor_operation::device::ConvolutionForwardSpecialization::Filter1x1Stride1Pad0;
static constexpr auto ConvFwdOddC =
ck::tensor_operation::device::ConvolutionForwardSpecialization::OddC;
static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
template <typename ALayout,
typename BLayout,
typename DsLayout,
typename ELayout,
typename DsDatatype,
typename CDEElementOp,
ConvolutionForwardSpecialization ConvSpec>
using device_grouped_conv2d_fwd_wmma_f16_instances = std::tuple<
// clang-format off
//########################################| NumDim| A| B| Ds| E| AData| BData| Ds| EData| AccData| CShuffle| A| B| CDE| ConvForward| GEMM| Block| MPer| NPer| KPer| K1| MPer| NPer| MRepeat| NRepeat| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CShuffle| CShuffle| CBlockTransferClusterLengths| CBlockTransfer|
//########################################| Spatial| Layout| Layout| Layout| Layout| Type| Type| DataType| Type| Type| DataType| Elementwise| Elementwise| Elementwise| Specialization| Specialization| Size| Block| Block| Block| | WMMA| WMMA| | | ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave| _MBlock_MWaveMPerXdl| ScalarPerVector|
//########################################| | | | | | | | | | | | Operation| Operation| Operation| | | | | | | | | | | | Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| _NBlock_NWaveNPerXdl| _NWaveNPerXdl|
//########################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
// blocksize=256
DeviceGroupedConvFwdMultipleD_Wmma_CShuffle< 2, ALayout, BLayout, DsLayout, ELayout, F16, F16, DsDatatype, F16, F32, F16, PassThrough, PassThrough, CDEElementOp, ConvSpec, GemmMNKPadding, 256, 128, 128, 4, 8, 16, 16, 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>,
DeviceGroupedConvFwdMultipleD_Wmma_CShuffle< 2, ALayout, BLayout, DsLayout, ELayout, F16, F16, DsDatatype, F16, F32, F16, PassThrough, PassThrough, CDEElementOp, ConvSpec, GemmMNKPadding, 256, 64, 256, 4, 8, 16, 16, 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>,
DeviceGroupedConvFwdMultipleD_Wmma_CShuffle< 2, ALayout, BLayout, DsLayout, ELayout, F16, F16, DsDatatype, F16, F32, F16, PassThrough, PassThrough, CDEElementOp, ConvSpec, GemmMNKPadding, 256, 256, 64, 4, 8, 16, 16, 8, 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>,
DeviceGroupedConvFwdMultipleD_Wmma_CShuffle< 2, ALayout, BLayout, DsLayout, ELayout, F16, F16, DsDatatype, F16, F32, F16, PassThrough, PassThrough, CDEElementOp, ConvSpec, GemmMNKPadding, 256, 128, 128, 8, 8, 16, 16, 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>,
// blocksize=128
DeviceGroupedConvFwdMultipleD_Wmma_CShuffle< 2, ALayout, BLayout, DsLayout, ELayout, F16, F16, DsDatatype, F16, F32, F16, PassThrough, PassThrough, CDEElementOp, ConvSpec, GemmMNKPadding, 128, 64, 64, 4, 8, 16, 16, 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>,
DeviceGroupedConvFwdMultipleD_Wmma_CShuffle< 2, ALayout, BLayout, DsLayout, ELayout, F16, F16, DsDatatype, F16, F32, F16, PassThrough, PassThrough, CDEElementOp, ConvSpec, GemmMNKPadding, 128, 64, 64, 8, 8, 16, 16, 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>,
DeviceGroupedConvFwdMultipleD_Wmma_CShuffle< 2, ALayout, BLayout, DsLayout, ELayout, F16, F16, DsDatatype, F16, F32, F16, PassThrough, PassThrough, CDEElementOp, ConvSpec, GemmMNKPadding, 128, 64, 128, 4, 8, 16, 16, 2, 4, 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>,
DeviceGroupedConvFwdMultipleD_Wmma_CShuffle< 2, ALayout, BLayout, DsLayout, ELayout, F16, F16, DsDatatype, F16, F32, F16, PassThrough, PassThrough, CDEElementOp, ConvSpec, GemmMNKPadding, 128, 64, 128, 8, 8, 16, 16, 2, 4, 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>,
DeviceGroupedConvFwdMultipleD_Wmma_CShuffle< 2, ALayout, BLayout, DsLayout, ELayout, F16, F16, DsDatatype, F16, F32, F16, PassThrough, PassThrough, CDEElementOp, ConvSpec, GemmMNKPadding, 128, 128, 64, 4, 8, 16, 16, 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, 32, 1, 4>, 8>,
DeviceGroupedConvFwdMultipleD_Wmma_CShuffle< 2, ALayout, BLayout, DsLayout, ELayout, F16, F16, DsDatatype, F16, F32, F16, PassThrough, PassThrough, CDEElementOp, ConvSpec, GemmMNKPadding, 128, 128, 64, 8, 8, 16, 16, 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, 32, 1, 4>, 8>,
DeviceGroupedConvFwdMultipleD_Wmma_CShuffle< 2, ALayout, BLayout, DsLayout, ELayout, F16, F16, DsDatatype, F16, F32, F16, PassThrough, PassThrough, CDEElementOp, ConvSpec, GemmMNKPadding, 128, 32, 256, 4, 8, 16, 16, 1, 8, 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>,
DeviceGroupedConvFwdMultipleD_Wmma_CShuffle< 2, ALayout, BLayout, DsLayout, ELayout, F16, F16, DsDatatype, F16, F32, F16, PassThrough, PassThrough, CDEElementOp, ConvSpec, GemmMNKPadding, 128, 256, 32, 4, 8, 16, 16, 8, 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>,
// blocksize=64
DeviceGroupedConvFwdMultipleD_Wmma_CShuffle< 2, ALayout, BLayout, DsLayout, ELayout, F16, F16, DsDatatype, F16, F32, F16, PassThrough, PassThrough, CDEElementOp, ConvSpec, GemmMNKPadding, 64, 32, 64, 4, 8, 16, 16, 1, 4, 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, 32, 1, 2>, 8>,
DeviceGroupedConvFwdMultipleD_Wmma_CShuffle< 2, ALayout, BLayout, DsLayout, ELayout, F16, F16, DsDatatype, F16, F32, F16, PassThrough, PassThrough, CDEElementOp, ConvSpec, GemmMNKPadding, 64, 64, 32, 4, 8, 16, 16, 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, 32, 1, 2>, 8>,
DeviceGroupedConvFwdMultipleD_Wmma_CShuffle< 2, ALayout, BLayout, DsLayout, ELayout, F16, F16, DsDatatype, F16, F32, F16, PassThrough, PassThrough, CDEElementOp, ConvSpec, GemmMNKPadding, 64, 32, 32, 8, 8, 16, 16, 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, 32, 1, 2>, 8>,
DeviceGroupedConvFwdMultipleD_Wmma_CShuffle< 2, ALayout, BLayout, DsLayout, ELayout, F16, F16, DsDatatype, F16, F32, F16, PassThrough, PassThrough, CDEElementOp, ConvSpec, GemmMNKPadding, 64, 32, 128, 4, 8, 16, 16, 1, 8, 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, 32, 1, 2>, 8>,
// blocksize=32
DeviceGroupedConvFwdMultipleD_Wmma_CShuffle< 2, ALayout, BLayout, DsLayout, ELayout, F16, F16, DsDatatype, F16, F32, F16, PassThrough, PassThrough, CDEElementOp, ConvSpec, GemmMNKPadding, 32, 16, 64, 4, 8, 16, 16, 1, 4, S<2, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<2, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 2>, 8>,
DeviceGroupedConvFwdMultipleD_Wmma_CShuffle< 2, ALayout, BLayout, DsLayout, ELayout, F16, F16, DsDatatype, F16, F32, F16, PassThrough, PassThrough, CDEElementOp, ConvSpec, GemmMNKPadding, 32, 64, 16, 4, 8, 16, 16, 4, 1, S<2, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<2, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 2>, 8>,
DeviceGroupedConvFwdMultipleD_Wmma_CShuffle< 2, ALayout, BLayout, DsLayout, ELayout, F16, F16, DsDatatype, F16, F32, F16, PassThrough, PassThrough, CDEElementOp, ConvSpec, GemmMNKPadding, 32, 32, 32, 4, 8, 16, 16, 2, 2, S<2, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<2, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 2>, 8>,
DeviceGroupedConvFwdMultipleD_Wmma_CShuffle< 2, ALayout, BLayout, DsLayout, ELayout, F16, F16, DsDatatype, F16, F32, F16, PassThrough, PassThrough, CDEElementOp, ConvSpec, GemmMNKPadding, 32, 16, 16, 4, 8, 16, 16, 1, 1, S<2, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<2, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 2>, 8>
// clang-format on
>;
template <typename ALayout,
typename BLayout,
typename DsLayout,
typename ELayout,
typename DsDatatype,
typename CDEElementOp,
ConvolutionForwardSpecialization ConvSpec>
using device_grouped_conv2d_fwd_wmma_i8_instances = std::tuple<
// clang-format off
//########################################| NumDim| A| B| Ds| E| AData| BData| Ds| EData| AccData| CShuffle| A| B| CDE| ConvForward| GEMM| Block| MPer| NPer| KPer| K1| MPer| NPer| MRepeat| NRepeat| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CShuffle| CShuffle| CBlockTransferClusterLengths| CBlockTransfer|
//########################################| Spatial| Layout| Layout| Layout| Layout| Type| Type| DataType| Type| Type| DataType| Elementwise| Elementwise| Elementwise| Specialization| Specialization| Size| Block| Block| Block| | WMMA| WMMA| | | ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave| _MBlock_MWaveMPerXdl| ScalarPerVector|
//########################################| | | | | | | | | | | | Operation| Operation| Operation| | | | | | | | | | | | Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| _NBlock_NWaveNPerXdl| _NWaveNPerXdl|
//########################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
// blocksize=256
DeviceGroupedConvFwdMultipleD_Wmma_CShuffle< 2, ALayout, BLayout, DsLayout, ELayout, I8, I8, DsDatatype, I8, I32, I8, PassThrough, PassThrough, CDEElementOp, ConvSpec, GemmMNKPadding, 256, 128, 128, 4, 16, 16, 16, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 1, 1, 1, S<1, 32, 1, 8>, 8>,
DeviceGroupedConvFwdMultipleD_Wmma_CShuffle< 2, ALayout, BLayout, DsLayout, ELayout, I8, I8, DsDatatype, I8, I32, I8, PassThrough, PassThrough, CDEElementOp, ConvSpec, GemmMNKPadding, 256, 64, 256, 4, 16, 16, 16, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 1, 1, 1, S<1, 32, 1, 8>, 8>,
DeviceGroupedConvFwdMultipleD_Wmma_CShuffle< 2, ALayout, BLayout, DsLayout, ELayout, I8, I8, DsDatatype, I8, I32, I8, PassThrough, PassThrough, CDEElementOp, ConvSpec, GemmMNKPadding, 256, 256, 64, 4, 16, 16, 16, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 1, 1, 1, S<1, 32, 1, 8>, 8>,
DeviceGroupedConvFwdMultipleD_Wmma_CShuffle< 2, ALayout, BLayout, DsLayout, ELayout, I8, I8, DsDatatype, I8, I32, I8, PassThrough, PassThrough, CDEElementOp, ConvSpec, GemmMNKPadding, 256, 128, 128, 8, 16, 16, 16, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 1, 1, 1, S<1, 32, 1, 8>, 8>,
// blocksize=128
DeviceGroupedConvFwdMultipleD_Wmma_CShuffle< 2, ALayout, BLayout, DsLayout, ELayout, I8, I8, DsDatatype, I8, I32, I8, PassThrough, PassThrough, CDEElementOp, ConvSpec, GemmMNKPadding, 128, 64, 64, 4, 16, 16, 16, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 1, 1, 1, S<1, 32, 1, 4>, 8>,
DeviceGroupedConvFwdMultipleD_Wmma_CShuffle< 2, ALayout, BLayout, DsLayout, ELayout, I8, I8, DsDatatype, I8, I32, I8, PassThrough, PassThrough, CDEElementOp, ConvSpec, GemmMNKPadding, 128, 64, 64, 8, 16, 16, 16, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 1, 1, 1, S<1, 32, 1, 4>, 8>,
DeviceGroupedConvFwdMultipleD_Wmma_CShuffle< 2, ALayout, BLayout, DsLayout, ELayout, I8, I8, DsDatatype, I8, I32, I8, PassThrough, PassThrough, CDEElementOp, ConvSpec, GemmMNKPadding, 128, 64, 128, 4, 16, 16, 16, 2, 4, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 1, 1, 1, S<1, 32, 1, 4>, 8>,
DeviceGroupedConvFwdMultipleD_Wmma_CShuffle< 2, ALayout, BLayout, DsLayout, ELayout, I8, I8, DsDatatype, I8, I32, I8, PassThrough, PassThrough, CDEElementOp, ConvSpec, GemmMNKPadding, 128, 64, 128, 8, 16, 16, 16, 2, 4, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 1, 1, 1, S<1, 32, 1, 4>, 8>,
DeviceGroupedConvFwdMultipleD_Wmma_CShuffle< 2, ALayout, BLayout, DsLayout, ELayout, I8, I8, DsDatatype, I8, I32, I8, PassThrough, PassThrough, CDEElementOp, ConvSpec, GemmMNKPadding, 128, 128, 64, 4, 16, 16, 16, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 1, 1, 1, S<1, 32, 1, 4>, 8>,
DeviceGroupedConvFwdMultipleD_Wmma_CShuffle< 2, ALayout, BLayout, DsLayout, ELayout, I8, I8, DsDatatype, I8, I32, I8, PassThrough, PassThrough, CDEElementOp, ConvSpec, GemmMNKPadding, 128, 128, 64, 8, 16, 16, 16, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 1, 1, 1, S<1, 32, 1, 4>, 8>,
DeviceGroupedConvFwdMultipleD_Wmma_CShuffle< 2, ALayout, BLayout, DsLayout, ELayout, I8, I8, DsDatatype, I8, I32, I8, PassThrough, PassThrough, CDEElementOp, ConvSpec, GemmMNKPadding, 128, 32, 256, 4, 16, 16, 16, 1, 8, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 1, 1, 1, S<1, 32, 1, 4>, 8>,
DeviceGroupedConvFwdMultipleD_Wmma_CShuffle< 2, ALayout, BLayout, DsLayout, ELayout, I8, I8, DsDatatype, I8, I32, I8, PassThrough, PassThrough, CDEElementOp, ConvSpec, GemmMNKPadding, 128, 256, 32, 4, 16, 16, 16, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 1, 1, 1, S<1, 32, 1, 4>, 8>,
// blocksize=64
DeviceGroupedConvFwdMultipleD_Wmma_CShuffle< 2, ALayout, BLayout, DsLayout, ELayout, I8, I8, DsDatatype, I8, I32, I8, PassThrough, PassThrough, CDEElementOp, ConvSpec, GemmMNKPadding, 64, 32, 64, 4, 16, 16, 16, 1, 4, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 1, 1, 1, S<1, 32, 1, 2>, 8>,
DeviceGroupedConvFwdMultipleD_Wmma_CShuffle< 2, ALayout, BLayout, DsLayout, ELayout, I8, I8, DsDatatype, I8, I32, I8, PassThrough, PassThrough, CDEElementOp, ConvSpec, GemmMNKPadding, 64, 64, 32, 4, 16, 16, 16, 2, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 1, 1, 1, S<1, 32, 1, 2>, 8>,
DeviceGroupedConvFwdMultipleD_Wmma_CShuffle< 2, ALayout, BLayout, DsLayout, ELayout, I8, I8, DsDatatype, I8, I32, I8, PassThrough, PassThrough, CDEElementOp, ConvSpec, GemmMNKPadding, 64, 32, 32, 8, 16, 16, 16, 1, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 1, 1, 1, S<1, 32, 1, 2>, 8>,
DeviceGroupedConvFwdMultipleD_Wmma_CShuffle< 2, ALayout, BLayout, DsLayout, ELayout, I8, I8, DsDatatype, I8, I32, I8, PassThrough, PassThrough, CDEElementOp, ConvSpec, GemmMNKPadding, 64, 32, 128, 4, 16, 16, 16, 1, 8, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 1, 1, 1, S<1, 32, 1, 2>, 8>,
// blocksize=32
DeviceGroupedConvFwdMultipleD_Wmma_CShuffle< 2, ALayout, BLayout, DsLayout, ELayout, I8, I8, DsDatatype, I8, I32, I8, PassThrough, PassThrough, CDEElementOp, ConvSpec, GemmMNKPadding, 32, 16, 64, 4, 16, 16, 16, 1, 4, S<2, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 1, S<2, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 1, 1, 1, S<1, 16, 1, 2>, 8>,
DeviceGroupedConvFwdMultipleD_Wmma_CShuffle< 2, ALayout, BLayout, DsLayout, ELayout, I8, I8, DsDatatype, I8, I32, I8, PassThrough, PassThrough, CDEElementOp, ConvSpec, GemmMNKPadding, 32, 64, 16, 4, 16, 16, 16, 4, 1, S<2, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 1, S<2, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 1, 1, 1, S<1, 16, 1, 2>, 8>,
DeviceGroupedConvFwdMultipleD_Wmma_CShuffle< 2, ALayout, BLayout, DsLayout, ELayout, I8, I8, DsDatatype, I8, I32, I8, PassThrough, PassThrough, CDEElementOp, ConvSpec, GemmMNKPadding, 32, 32, 32, 4, 16, 16, 16, 2, 2, S<2, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 1, S<2, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 1, 1, 1, S<1, 16, 1, 2>, 8>,
DeviceGroupedConvFwdMultipleD_Wmma_CShuffle< 2, ALayout, BLayout, DsLayout, ELayout, I8, I8, DsDatatype, I8, I32, I8, PassThrough, PassThrough, CDEElementOp, ConvSpec, GemmMNKPadding, 32, 16, 16, 4, 16, 16, 16, 1, 1, S<2, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 1, S<2, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 1, 1, 1, S<1, 16, 1, 2>, 8>
// clang-format on
>;
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck
library/include/ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv_fwd_dl_instance.hpp
View file @ 4173b984
......@@ -53,6 +53,11 @@ using device_grouped_conv2d_fwd_dl_f16_instances = std::tuple<
// ########################################| Spatial| Type| Type| Type| Type| Type| | | Layout| | Elementwise| Elementwise| Elementwise| Forward| Spacialization| Size| Block| Block| Block| | ThreadM111| ThreadN111| Thread| ClusterM110Xs| ClusterN110Xs| ThreadSliceLengths| ThreadClusterLengths| ThreadCluster| SrcAccess| SrcVectorTensor| SrcVectorTensor| DstVectorTensor| ThreadSliceLengths| ThreadClusterLengths| ThreadCluster| SrcAccess| SrcVectorTensor| SrcVectorTensor| DstVectorTensor| SrcDstAccess| SrcDstVectorDim| DstScalarPerVector|
// ########################################| | | | | | | | | | | Operation| Operation| Operation| Specialization| | | | | | | | | | | | K0_M0_M1_K1| K0_M0_M1_K1| ArrangeOrder| Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder| Lengths_K0_M0_M1_K1| K0_N0_N1_K1| K0_N0_N1_K1| ArrangeOrder| Order| Lengths_K0_N0_N1_K1| ContiguousDimOrder| Lengths_K0_N0_N1_K1| Order| | |
// ########################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
// generic instances
// TODO: Change to ScalarPerVector = 1 when inner_product<half_t, half_t, float> will be supported
DeviceGroupedConvFwdDlMultipleD_NHWC_KYXC_NHWK< 2, F16, F16, DsDatatype, F16, F32, InLayout, WeiLayout, DsLayout, OutLayout, PassThrough, PassThrough, CDEElementOp, ConvSpec, GemmMNKPadding, 8, 16, 4, 2, 2, 1, 2, 1, S<4, 2>, S<1, 1>, S<2, 1, 2, 2>, S<1, 1, 8, 1>, S<1, 2, 0, 3>, S<1, 2, 0, 3>, S<1, 1, 1, 2>, S<1, 2, 0, 3>, S<1, 1, 1, 2>, S<1, 1, 1, 2>, S<2, 1, 4, 1>, S<1, 2, 0, 3>, S<1, 2, 0, 3>, S<1, 1, 1, 1>, S<1, 2, 0, 3>, S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>, 5, 2>,
DeviceGroupedConvFwdDlMultipleD_NHWC_KYXC_NHWK< 2, F16, F16, DsDatatype, F16, F32, InLayout, WeiLayout, DsLayout, OutLayout, PassThrough, PassThrough, CDEElementOp, ConvSpec, GemmMNKPadding, 256, 128, 128, 16, 2, 4, 4, 1, S<8, 2>, S<8, 2>, S<8, 1, 1, 2>, S<2, 1, 128, 1>, S<1, 2, 0, 3>, S<1, 2, 0, 3>, S<1, 1, 1, 2>, S<1, 2, 0, 3>, S<1, 1, 1, 2>, S<8, 1, 1, 2>, S<2, 1, 128, 1>, S<1, 2, 0, 3>, S<1, 2, 0, 3>, S<1, 1, 1, 1>, S<1, 2, 0, 3>, S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>, 5, 2>,
DeviceGroupedConvFwdDlMultipleD_NHWC_KYXC_NHWK< 2, F16, F16, DsDatatype, F16, F32, InLayout, WeiLayout, DsLayout, OutLayout, PassThrough, PassThrough, CDEElementOp, ConvSpec, GemmMNKPadding, 256, 128, 128, 16, 2, 4, 4, 1, S<8, 2>, S<8, 2>, S<8, 1, 1, 2>, S<2, 1, 128, 1>, S<1, 2, 0, 3>, S<1, 2, 0, 3>, S<4, 1, 1, 2>, S<1, 2, 0, 3>, S<1, 1, 1, 2>, S<8, 1, 1, 2>, S<2, 1, 128, 1>, S<1, 2, 0, 3>, S<1, 2, 0, 3>, S<4, 1, 1, 2>, S<1, 2, 0, 3>, S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>, 5, 4>
// clang-format on
>;
......@@ -71,6 +76,10 @@ using device_grouped_conv2d_fwd_dl_f32_instances = std::tuple<
// ########################################| Spatial| Type| Type| Type| Type| Type| | | Layout| | Elementwise| Elementwise| Elementwise| Forward| Spacialization| Size| Block| Block| Block| | ThreadM111| ThreadN111| Thread| ClusterM110Xs| ClusterN110Xs| ThreadSliceLengths| ThreadClusterLengths| ThreadCluster| SrcAccess| SrcVectorTensor| SrcVectorTensor| DstVectorTensor| ThreadSliceLengths| ThreadClusterLengths| ThreadCluster| SrcAccess| SrcVectorTensor| SrcVectorTensor| DstVectorTensor| SrcDstAccess| SrcDstVectorDim| DstScalarPerVector|
// ########################################| | | | | | | | | | | Operation| Operation| Operation| Specialization| | | | | | | | | | | | K0_M0_M1_K1| K0_M0_M1_K1| ArrangeOrder| Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder| Lengths_K0_M0_M1_K1| K0_N0_N1_K1| K0_N0_N1_K1| ArrangeOrder| Order| Lengths_K0_N0_N1_K1| ContiguousDimOrder| Lengths_K0_N0_N1_K1| Order| | |
// ########################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
// generic instances
DeviceGroupedConvFwdDlMultipleD_NHWC_KYXC_NHWK< 2, F32, F32, DsDatatype, F32, F32, InLayout, WeiLayout, DsLayout, OutLayout, PassThrough, PassThrough, CDEElementOp, ConvSpec, GemmMNKPadding, 8, 16, 4, 2, 1, 1, 2, 1, S<4, 2>, S<1, 1>, S<2, 1, 2, 1>, S<1, 1, 8, 1>, S<1, 2, 0, 3>, S<1, 2, 0, 3>, S<1, 1, 1, 1>, S<1, 2, 0, 3>, S<1, 1, 1, 1>, S<1, 1, 1, 1>, S<2, 1, 4, 1>, S<1, 2, 0, 3>, S<1, 2, 0, 3>, S<1, 1, 1, 1>, S<1, 2, 0, 3>, S<1, 1, 1, 1>, S<0, 1, 2, 3, 4, 5>, 5, 1>,
DeviceGroupedConvFwdDlMultipleD_NHWC_KYXC_NHWK< 2, F32, F32, DsDatatype, F32, F32, InLayout, WeiLayout, DsLayout, OutLayout, PassThrough, PassThrough, CDEElementOp, ConvSpec, GemmMNKPadding, 256, 128, 128, 16, 1, 4, 4, 1, S<8, 2>, S<8, 2>, S<8, 1, 1, 1>, S<2, 1, 128, 1>, S<1, 2, 0, 3>, S<1, 2, 0, 3>, S<1, 1, 1, 1>, S<1, 2, 0, 3>, S<1, 1, 1, 1>, S<8, 1, 1, 1>, S<2, 1, 128, 1>, S<1, 2, 0, 3>, S<1, 2, 0, 3>, S<1, 1, 1, 1>, S<1, 2, 0, 3>, S<1, 1, 1, 1>, S<0, 1, 2, 3, 4, 5>, 5, 1>,
DeviceGroupedConvFwdDlMultipleD_NHWC_KYXC_NHWK< 2, F32, F32, DsDatatype, F32, F32, InLayout, WeiLayout, DsLayout, OutLayout, PassThrough, PassThrough, CDEElementOp, ConvSpec, GemmMNKPadding, 256, 128, 128, 16, 1, 4, 4, 1, S<8, 2>, S<8, 2>, S<8, 1, 1, 1>, S<2, 1, 128, 1>, S<1, 2, 0, 3>, S<1, 2, 0, 3>, S<4, 1, 1, 1>, S<1, 2, 0, 3>, S<1, 1, 1, 1>, S<8, 1, 1, 1>, S<2, 1, 128, 1>, S<1, 2, 0, 3>, S<1, 2, 0, 3>, S<4, 1, 1, 1>, S<1, 2, 0, 3>, S<1, 1, 1, 1>, S<0, 1, 2, 3, 4, 5>, 5, 4>
// clang-format on
>;
......
library/include/ck/library/tensor_operation_instance/gpu/grouped_convolution_forward.hpp
View file @ 4173b984
......@@ -145,6 +145,60 @@ void add_device_grouped_conv2d_fwd_dl_gnhwc_gkyxc_gnhwk_f32_instances(
PassThrough,
PassThrough>>>& instances);
void add_device_grouped_conv2d_fwd_wmma_gnhwc_gkyxc_gnhwk_f16_instances(
std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
GNHWC,
GKYXC,
Empty_Tuple,
GNHWK,
F16,
F16,
Empty_Tuple,
F16,
PassThrough,
PassThrough,
PassThrough>>>& instances);
void add_device_grouped_conv2d_fwd_dl_nhwgc_gkyxc_nhwgk_f16_instances(
std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
NHWGC,
GKYXC,
Empty_Tuple,
NHWGK,
F16,
F16,
Empty_Tuple,
F16,
PassThrough,
PassThrough,
PassThrough>>>& instances);
void add_device_grouped_conv2d_fwd_wmma_gnhwc_gkyxc_gnhwk_i8_instances(
std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
GNHWC,
GKYXC,
Empty_Tuple,
GNHWK,
int8_t,
int8_t,
Empty_Tuple,
int8_t,
PassThrough,
PassThrough,
PassThrough>>>& instances);
void add_device_grouped_conv2d_fwd_dl_nhwgc_gkyxc_nhwgk_f32_instances(
std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
NHWGC,
GKYXC,
Empty_Tuple,
NHWGK,
F32,
F32,
Empty_Tuple,
F32,
PassThrough,
PassThrough,
PassThrough>>>& instances);
// grouped conv2d forward, NHWGC/GKYXC/NHWGK
void add_device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_bf16_instances(
std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
......@@ -379,6 +433,7 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
{
add_device_grouped_conv2d_fwd_xdl_gnhwc_gkyxc_gnhwk_f16_instances(op_ptrs);
add_device_grouped_conv2d_fwd_dl_gnhwc_gkyxc_gnhwk_f16_instances(op_ptrs);
add_device_grouped_conv2d_fwd_wmma_gnhwc_gkyxc_gnhwk_f16_instances(op_ptrs);
}
else if constexpr(is_same_v<InDataType, ck::bhalf_t> &&
is_same_v<WeiDataType, ck::bhalf_t> &&
......@@ -386,6 +441,11 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
{
add_device_grouped_conv1d_fwd_xdl_gnhwc_gkyxc_gnhwk_bf16_instances(op_ptrs);
}
else if constexpr(is_same_v<InDataType, int8_t> && is_same_v<WeiDataType, int8_t> &&
is_same_v<OutDataType, int8_t>)
{
add_device_grouped_conv2d_fwd_wmma_gnhwc_gkyxc_gnhwk_i8_instances(op_ptrs);
}
}
else if constexpr(NumDimSpatial == 2 && is_same_v<InLayout, NHWGC> &&
is_same_v<WeiLayout, GKYXC> && is_same_v<OutLayout, NHWGK>)
......@@ -394,11 +454,13 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
is_same_v<OutDataType, float>)
{
add_device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_f32_instances(op_ptrs);
add_device_grouped_conv2d_fwd_dl_nhwgc_gkyxc_nhwgk_f32_instances(op_ptrs);
}
else if constexpr(is_same_v<InDataType, half_t> && is_same_v<WeiDataType, half_t> &&
is_same_v<OutDataType, half_t>)
{
add_device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_f16_instances(op_ptrs);
add_device_grouped_conv2d_fwd_dl_nhwgc_gkyxc_nhwgk_f16_instances(op_ptrs);
}
else if constexpr(is_same_v<InDataType, ck::bhalf_t> &&
is_same_v<WeiDataType, ck::bhalf_t> &&
......
library/include/ck/library/tensor_operation_instance/gpu/image_to_column.hpp 0 → 100644
View file @ 4173b984
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
#pragma once
#include <vector>
#include <memory>
#include "ck/ck.hpp"
#include "ck/tensor_operation/gpu/device/device_image_to_column.hpp"
#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
#include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp"
namespace ck {
namespace tensor_operation {
namespace device {
namespace instance {
// nhwc, 1d
void add_device_image_to_column_nhwc_1d_bf16_instances(
std::vector<std::unique_ptr<DeviceImageToColumn<1, GNWC, BF16, BF16>>>& instances);
void add_device_image_to_column_nhwc_1d_f16_instances(
std::vector<std::unique_ptr<DeviceImageToColumn<1, GNWC, F16, F16>>>& instances);
void add_device_image_to_column_nhwc_1d_f32_instances(
std::vector<std::unique_ptr<DeviceImageToColumn<1, GNWC, F32, F32>>>& instances);
void add_device_image_to_column_nhwc_1d_i8_instances(
std::vector<std::unique_ptr<DeviceImageToColumn<1, GNWC, int8_t, int8_t>>>& instances);
// nhwc, 2d
void add_device_image_to_column_nhwc_2d_bf16_instances(
std::vector<std::unique_ptr<DeviceImageToColumn<2, GNHWC, BF16, BF16>>>& instances);
void add_device_image_to_column_nhwc_2d_f16_instances(
std::vector<std::unique_ptr<DeviceImageToColumn<2, GNHWC, F16, F16>>>& instances);
void add_device_image_to_column_nhwc_2d_f32_instances(
std::vector<std::unique_ptr<DeviceImageToColumn<2, GNHWC, F32, F32>>>& instances);
void add_device_image_to_column_nhwc_2d_i8_instances(
std::vector<std::unique_ptr<DeviceImageToColumn<2, GNHWC, int8_t, int8_t>>>& instances);
// nhwc, 3d
void add_device_image_to_column_nhwc_3d_bf16_instances(
std::vector<std::unique_ptr<DeviceImageToColumn<3, GNDHWC, BF16, BF16>>>& instances);
void add_device_image_to_column_nhwc_3d_f16_instances(
std::vector<std::unique_ptr<DeviceImageToColumn<3, GNDHWC, F16, F16>>>& instances);
void add_device_image_to_column_nhwc_3d_f32_instances(
std::vector<std::unique_ptr<DeviceImageToColumn<3, GNDHWC, F32, F32>>>& instances);
void add_device_image_to_column_nhwc_3d_i8_instances(
std::vector<std::unique_ptr<DeviceImageToColumn<3, GNDHWC, int8_t, int8_t>>>& instances);
template <ck::index_t NumDimSpatial, typename InLayout, typename InDataType, typename OutDataType>
struct DeviceOperationInstanceFactory<
ck::tensor_operation::device::
DeviceImageToColumn<NumDimSpatial, InLayout, InDataType, OutDataType>>
{
using DeviceOp = DeviceImageToColumn<NumDimSpatial, InLayout, InDataType, OutDataType>;
static auto GetInstances()
{
std::vector<std::unique_ptr<DeviceOp>> op_ptrs;
if constexpr(NumDimSpatial == 1 && is_same_v<InLayout, GNWC>)
{
if constexpr(is_same_v<InDataType, float> && is_same_v<OutDataType, float>)
{
add_device_image_to_column_nhwc_1d_f32_instances(op_ptrs);
}
else if constexpr(is_same_v<InDataType, half_t> && is_same_v<OutDataType, half_t>)
{
add_device_image_to_column_nhwc_1d_f16_instances(op_ptrs);
}
else if constexpr(is_same_v<InDataType, ck::bhalf_t> &&
is_same_v<OutDataType, ck::bhalf_t>)
{
add_device_image_to_column_nhwc_1d_bf16_instances(op_ptrs);
}
else if constexpr(is_same_v<InDataType, int8_t> && is_same_v<OutDataType, int8_t>)
{
add_device_image_to_column_nhwc_1d_i8_instances(op_ptrs);
}
}
else if constexpr(NumDimSpatial == 2 && is_same_v<InLayout, GNHWC>)
{
if constexpr(is_same_v<InDataType, float> && is_same_v<OutDataType, float>)
{
add_device_image_to_column_nhwc_2d_f32_instances(op_ptrs);
}
else if constexpr(is_same_v<InDataType, half_t> && is_same_v<OutDataType, half_t>)
{
add_device_image_to_column_nhwc_2d_f16_instances(op_ptrs);
}
else if constexpr(is_same_v<InDataType, ck::bhalf_t> &&
is_same_v<OutDataType, ck::bhalf_t>)
{
add_device_image_to_column_nhwc_2d_bf16_instances(op_ptrs);
}
else if constexpr(is_same_v<InDataType, int8_t> && is_same_v<OutDataType, int8_t>)
{
add_device_image_to_column_nhwc_2d_i8_instances(op_ptrs);
}
}
else if constexpr(NumDimSpatial == 3 && is_same_v<InLayout, GNDHWC>)
{
if constexpr(is_same_v<InDataType, float> && is_same_v<OutDataType, float>)
{
add_device_image_to_column_nhwc_3d_f32_instances(op_ptrs);
}
else if constexpr(is_same_v<InDataType, half_t> && is_same_v<OutDataType, half_t>)
{
add_device_image_to_column_nhwc_3d_f16_instances(op_ptrs);
}
else if constexpr(is_same_v<InDataType, ck::bhalf_t> &&
is_same_v<OutDataType, ck::bhalf_t>)
{
add_device_image_to_column_nhwc_3d_bf16_instances(op_ptrs);
}
else if constexpr(is_same_v<InDataType, int8_t> && is_same_v<OutDataType, int8_t>)
{
add_device_image_to_column_nhwc_3d_i8_instances(op_ptrs);
}
}
return op_ptrs;
}
};
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck
library/include/ck/library/tensor_operation_instance/gpu/image_to_column/device_image_to_column_instance.hpp 0 → 100644
View file @ 4173b984
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
#include "ck/ck.hpp"
#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
#include "ck/tensor_operation/gpu/device/impl/device_image_to_column_impl.hpp"
#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
namespace ck {
namespace tensor_operation {
namespace device {
namespace instance {
using namespace ck::tensor_layout::convolution;
using BF16 = ck::bhalf_t;
using F16 = ck::half_t;
using F32 = float;
template <ck::index_t... Is>
using S = ck::Sequence<Is...>;
template <ck::index_t NDimSpatial, typename InLayout>
using device_image_to_column_bf16_instances = std::tuple<
// clang-format off
//#####################| Num| InLayout| InDataType| OutDataType| Block| MPer| KPer| Thread| Scalar|
//#####################| Dim| | | | Size| Block| Block| Cluster| Per|
//#####################| Spatial| | | | | | | Lengths| Vector|
//#####################| | | | | | | | | |
DeviceImageToColumnImpl<NDimSpatial, InLayout, BF16, BF16, 64, 8, 8, S<8, 8>, 1>,
DeviceImageToColumnImpl<NDimSpatial, InLayout, BF16, BF16, 64, 16, 16, S<8, 8>, 1>,
DeviceImageToColumnImpl<NDimSpatial, InLayout, BF16, BF16, 64, 32, 32, S<8, 8>, 4>,
DeviceImageToColumnImpl<NDimSpatial, InLayout, BF16, BF16, 64, 64, 64, S<8, 8>, 8>,
DeviceImageToColumnImpl<NDimSpatial, InLayout, BF16, BF16, 128, 16, 16, S<8, 16>, 1>,
DeviceImageToColumnImpl<NDimSpatial, InLayout, BF16, BF16, 128, 64, 64, S<8, 16>, 1>,
DeviceImageToColumnImpl<NDimSpatial, InLayout, BF16, BF16, 128, 32, 64, S<8, 16>, 4>,
DeviceImageToColumnImpl<NDimSpatial, InLayout, BF16, BF16, 128, 64, 128, S<8, 16>, 8>,
DeviceImageToColumnImpl<NDimSpatial, InLayout, BF16, BF16, 256, 16, 16, S<16, 16>, 1>,
DeviceImageToColumnImpl<NDimSpatial, InLayout, BF16, BF16, 256, 64, 64, S<16, 16>, 1>,
DeviceImageToColumnImpl<NDimSpatial, InLayout, BF16, BF16, 256, 128, 128, S<16, 16>, 1>,
DeviceImageToColumnImpl<NDimSpatial, InLayout, BF16, BF16, 256, 64, 64, S<16, 16>, 4>,
DeviceImageToColumnImpl<NDimSpatial, InLayout, BF16, BF16, 256, 128, 128, S<16, 16>, 4>,
DeviceImageToColumnImpl<NDimSpatial, InLayout, BF16, BF16, 256, 128, 128, S<16, 16>, 8>
// clang-format on
>;
template <ck::index_t NDimSpatial, typename InLayout>
using device_image_to_column_f16_instances = std::tuple<
// clang-format off
//#####################| Num| InLayout| InDataType| OutDataType| Block| MPer| KPer| Thread| Scalar|
//#####################| Dim| | | | Size| Block| Block| Cluster| Per|
//#####################| Spatial| | | | | | | Lengths| Vector|
//#####################| | | | | | | | | |
DeviceImageToColumnImpl<NDimSpatial, InLayout, F16, F16, 64, 8, 8, S<8, 8>, 1>,
DeviceImageToColumnImpl<NDimSpatial, InLayout, F16, F16, 64, 16, 16, S<8, 8>, 1>,
DeviceImageToColumnImpl<NDimSpatial, InLayout, F16, F16, 64, 32, 32, S<8, 8>, 4>,
DeviceImageToColumnImpl<NDimSpatial, InLayout, F16, F16, 64, 64, 64, S<8, 8>, 8>,
DeviceImageToColumnImpl<NDimSpatial, InLayout, F16, F16, 128, 16, 16, S<8, 16>, 1>,
DeviceImageToColumnImpl<NDimSpatial, InLayout, F16, F16, 128, 64, 64, S<8, 16>, 1>,
DeviceImageToColumnImpl<NDimSpatial, InLayout, F16, F16, 128, 32, 64, S<8, 16>, 4>,
DeviceImageToColumnImpl<NDimSpatial, InLayout, F16, F16, 128, 64, 128, S<8, 16>, 8>,
DeviceImageToColumnImpl<NDimSpatial, InLayout, F16, F16, 256, 16, 16, S<16, 16>, 1>,
DeviceImageToColumnImpl<NDimSpatial, InLayout, F16, F16, 256, 64, 64, S<16, 16>, 1>,
DeviceImageToColumnImpl<NDimSpatial, InLayout, F16, F16, 256, 128, 128, S<16, 16>, 1>,
DeviceImageToColumnImpl<NDimSpatial, InLayout, F16, F16, 256, 64, 64, S<16, 16>, 4>,
DeviceImageToColumnImpl<NDimSpatial, InLayout, F16, F16, 256, 128, 128, S<16, 16>, 4>,
DeviceImageToColumnImpl<NDimSpatial, InLayout, F16, F16, 256, 128, 128, S<16, 16>, 8>
// clang-format on
>;
template <ck::index_t NDimSpatial, typename InLayout>
using device_image_to_column_f32_instances = std::tuple<
// clang-format off
//#####################| Num| InLayout| InDataType| OutDataType| Block| MPer| KPer| Thread| Scalar|
//#####################| Dim| | | | Size| Block| Block| Cluster| Per|
//#####################| Spatial| | | | | | | Lengths| Vector|
//#####################| | | | | | | | | |
DeviceImageToColumnImpl<NDimSpatial, InLayout, F32, F32, 64, 8, 8, S<8, 8>, 1>,
DeviceImageToColumnImpl<NDimSpatial, InLayout, F32, F32, 64, 16, 16, S<8, 8>, 1>,
DeviceImageToColumnImpl<NDimSpatial, InLayout, F32, F32, 64, 32, 32, S<8, 8>, 4>,
DeviceImageToColumnImpl<NDimSpatial, InLayout, F32, F32, 128, 16, 16, S<8, 16>, 1>,
DeviceImageToColumnImpl<NDimSpatial, InLayout, F32, F32, 128, 64, 64, S<8, 16>, 1>,
DeviceImageToColumnImpl<NDimSpatial, InLayout, F32, F32, 128, 32, 64, S<8, 16>, 4>,
DeviceImageToColumnImpl<NDimSpatial, InLayout, F32, F32, 256, 16, 16, S<16, 16>, 1>,
DeviceImageToColumnImpl<NDimSpatial, InLayout, F32, F32, 256, 64, 64, S<16, 16>, 1>,
DeviceImageToColumnImpl<NDimSpatial, InLayout, F32, F32, 256, 128, 128, S<16, 16>, 1>,
DeviceImageToColumnImpl<NDimSpatial, InLayout, F32, F32, 256, 64, 64, S<16, 16>, 4>,
DeviceImageToColumnImpl<NDimSpatial, InLayout, F32, F32, 256, 128, 128, S<16, 16>, 4>
// clang-format on
>;
template <ck::index_t NDimSpatial, typename InLayout>
using device_image_to_column_i8_instances = std::tuple<
// clang-format off
//#####################| Num| InLayout| InDataType| OutDataType| Block| MPer| KPer| Thread| Scalar|
//#####################| Dim| | | | Size| Block| Block| Cluster| Per|
//#####################| Spatial| | | | | | | Lengths| Vector|
//#####################| | | | | | | | | |
DeviceImageToColumnImpl<NDimSpatial, InLayout, int8_t, int8_t, 64, 8, 8, S<8, 8>, 1>,
DeviceImageToColumnImpl<NDimSpatial, InLayout, int8_t, int8_t, 64, 16, 16, S<8, 8>, 1>,
DeviceImageToColumnImpl<NDimSpatial, InLayout, int8_t, int8_t, 64, 32, 32, S<8, 8>, 4>,
DeviceImageToColumnImpl<NDimSpatial, InLayout, int8_t, int8_t, 64, 64, 64, S<8, 8>, 8>,
DeviceImageToColumnImpl<NDimSpatial, InLayout, int8_t, int8_t, 128, 16, 16, S<8, 16>, 1>,
DeviceImageToColumnImpl<NDimSpatial, InLayout, int8_t, int8_t, 128, 64, 64, S<8, 16>, 1>,
DeviceImageToColumnImpl<NDimSpatial, InLayout, int8_t, int8_t, 128, 32, 64, S<8, 16>, 4>,
DeviceImageToColumnImpl<NDimSpatial, InLayout, int8_t, int8_t, 128, 64, 128, S<8, 16>, 8>,
DeviceImageToColumnImpl<NDimSpatial, InLayout, int8_t, int8_t, 256, 16, 16, S<16, 16>, 1>,
DeviceImageToColumnImpl<NDimSpatial, InLayout, int8_t, int8_t, 256, 64, 64, S<16, 16>, 1>,
DeviceImageToColumnImpl<NDimSpatial, InLayout, int8_t, int8_t, 256, 128, 128, S<16, 16>, 1>,
DeviceImageToColumnImpl<NDimSpatial, InLayout, int8_t, int8_t, 256, 64, 64, S<16, 16>, 4>,
DeviceImageToColumnImpl<NDimSpatial, InLayout, int8_t, int8_t, 256, 128, 128, S<16, 16>, 4>,
DeviceImageToColumnImpl<NDimSpatial, InLayout, int8_t, int8_t, 256, 128, 128, S<16, 16>, 8>,
DeviceImageToColumnImpl<NDimSpatial, InLayout, int8_t, int8_t, 256, 256, 256, S<16, 16>, 16>
// clang-format on
>;
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck
library/include/ck/library/utility/host_tensor_generator.hpp
View file @ 4173b984
......@@ -95,6 +95,22 @@ struct GeneratorTensor_2<int8_t>
}
};
#if defined CK_ENABLE_FP8 || defined CK_ENABLE_BF8
template <>
struct GeneratorTensor_2<ck::f8_t>
{
int min_value = 0;
int max_value = 1;
template <typename... Is>
ck::f8_t operator()(Is...)
{
float tmp = (std::rand() % (max_value - min_value)) + min_value;
return ck::type_convert<ck::f8_t>(tmp);
}
};
#endif
template <typename T>
struct GeneratorTensor_3
{
......@@ -127,6 +143,25 @@ struct GeneratorTensor_3<ck::bhalf_t>
}
};
#if defined CK_ENABLE_FP8 || defined CK_ENABLE_BF8
template <>
struct GeneratorTensor_3<ck::f8_t>
{
float min_value = 0;
float max_value = 1;
template <typename... Is>
ck::f8_t operator()(Is...)
{
float tmp = float(std::rand()) / float(RAND_MAX);
float fp32_tmp = min_value + tmp * (max_value - min_value);
return ck::type_convert<ck::f8_t>(fp32_tmp);
}
};
#endif
template <typename T>
struct GeneratorTensor_4
{
......
library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_f16_f16_f16_gmk_gkn_gmn_instance.cpp
View file @ 4173b984
......@@ -25,6 +25,16 @@ using S = ck::Sequence<Is...>;
using PassThrough = ck::tensor_operation::element_wise::PassThrough;
using device_batched_gemm_xdl_f16_f16_f16_gmk_gkn_gmn_generic_instances = std::tuple<
// clang-format off
//#################| AData| BData| CData| AccData| ALayout| BLayout| CLayout| A| B| C| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer| NumGemmK| LoopScheduler| Pipeline|
//#################| Type| Type| Type| Type| | | | Elementwise| Elementwise| Elementwise| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| SrcDstVectorDim| DstScalar| Prefetch| | |
//#################| | | | | | | | Operation| Operation| Operation| | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector| Stage | | |
//#################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceBatchedGemmXdl< F16, F16, F16, F32, Row, Row, Row, PassThrough, PassThrough, PassThrough, 64, 16, 16, 4, 8, 16, 16, 1, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 8, true, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, true, 7, 1, 1, LoopScheduler::Default, PipelineVersion::v1>
// clang-format on
>;
// Compilation parameters for a[m, k] * b[k, n] = c[m, n]
using device_batched_gemm_xdl_f16_f16_f16_gmk_gkn_gmn_instances = std::tuple<
// clang-format off
......@@ -100,6 +110,8 @@ void add_device_batched_gemm_xdl_f16_f16_f16_gmk_gkn_gmn_instances(
DeviceBatchedGemm<Row, Row, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
instances)
{
add_device_operation_instances(
instances, device_batched_gemm_xdl_f16_f16_f16_gmk_gkn_gmn_generic_instances{});
add_device_operation_instances(instances,
device_batched_gemm_xdl_f16_f16_f16_gmk_gkn_gmn_instances{});
}
......
library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_f16_f16_f16_gmk_gnk_gmn_instance.cpp
View file @ 4173b984
......@@ -25,6 +25,16 @@ using S = ck::Sequence<Is...>;
using PassThrough = ck::tensor_operation::element_wise::PassThrough;
using device_batched_gemm_xdl_f16_f16_f16_gmk_gnk_gmn_generic_instances = std::tuple<
// clang-format off
//#################| AData| BData| CData| AccData| ALayout| BLayout| CLayout| A| B| C| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer| NumGemmK| LoopScheduler| Pipeline|
//#################| Type| Type| Type| Type| | | | Elementwise| Elementwise| Elementwise| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| SrcDstVectorDim| DstScalar| Prefetch| | |
//#################| | | | | | | | Operation| Operation| Operation| | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector| Stage | | |
//#################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceBatchedGemmXdl< F16, F16, F16, F32, Row, Col, Row, PassThrough, PassThrough, PassThrough, 64, 32, 64, 4, 8, 32, 32, 1, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 8, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 8, true, 7, 1, 1, LoopScheduler::Default, PipelineVersion::v1>
// clang-format on
>;
// Compilation parameters for a[m, k] * b[n, k] = c[m, n]
using device_batched_gemm_xdl_f16_f16_f16_gmk_gnk_gmn_instances = std::tuple<
// clang-format off
......@@ -88,6 +98,8 @@ void add_device_batched_gemm_xdl_f16_f16_f16_gmk_gnk_gmn_instances(
DeviceBatchedGemm<Row, Col, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
instances)
{
add_device_operation_instances(
instances, device_batched_gemm_xdl_f16_f16_f16_gmk_gnk_gmn_generic_instances{});
add_device_operation_instances(instances,
device_batched_gemm_xdl_f16_f16_f16_gmk_gnk_gmn_instances{});
}
......
library/src/tensor_operation_instance/gpu/gemm/CMakeLists.txt
View file @ 4173b984
......@@ -31,10 +31,10 @@ if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
list(APPEND GEMM_INSTANCES device_gemm_dl_f16_f16_f16_km_kn_mn_irregular_instance.cpp)
list(APPEND GEMM_INSTANCES device_gemm_dl_f16_f16_f16_km_nk_mn_instance.cpp)
list(APPEND GEMM_INSTANCES device_gemm_dl_f16_f16_f16_km_nk_mn_irregular_instance.cpp)
list(APPEND GEMM_INSTANCES device_gemm_dl_dpp8_f16_f16_f16_km_kn_mn_instance.cpp)
list(APPEND GEMM_INSTANCES device_gemm_dl_dpp8_f16_f16_f16_km_nk_mn_instance.cpp)
list(APPEND GEMM_INSTANCES device_gemm_dl_dpp8_f16_f16_f16_mk_kn_mn_instance.cpp)
list(APPEND GEMM_INSTANCES device_gemm_dl_dpp8_f16_f16_f16_mk_nk_mn_instance.cpp)
list(APPEND GEMM_INSTANCES device_gemm_dpp_f16_f16_f16_km_kn_mn_instance.cpp)
list(APPEND GEMM_INSTANCES device_gemm_dpp_f16_f16_f16_km_nk_mn_instance.cpp)
list(APPEND GEMM_INSTANCES device_gemm_dpp_f16_f16_f16_mk_kn_mn_instance.cpp)
list(APPEND GEMM_INSTANCES device_gemm_dpp_f16_f16_f16_mk_nk_mn_instance.cpp)
endif()
list(APPEND GEMM_INSTANCES device_gemm_xdl_c_shuffle_f16_f16_f16_mk_kn_mn_instance.cpp)
list(APPEND GEMM_INSTANCES device_gemm_xdl_c_shuffle_f16_f16_f16_mk_nk_mn_instance.cpp)
......
library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_dpp8_f16_f16_f16_km_kn_mn_instance.cpp deleted 100644 → 0
View file @ 6de7d10d
// 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_dl_dpp8.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 GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default;
// Compilation parameters for a[k, m] * b[k, n] = c[m, n]
using device_gemm_dl_dpp8_f16_f16_f16_km_kn_mn_instances = std::tuple<
// clang-format off
// ##########| AData| BData| CData| AccData| ALayout| BLayout| CLayout| A| B| C| GEMM| Block| MPer| NPer| K0Per| K1| M1Per| N1Per| KPer| M11N11Thread| M11N11Thread| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| CThreadTransfer| CThreadTransfer| CThreadTransfer|
// ##########| Type| Type| Type| Type| | | | Elementwise| Elementwise| Elementwise| Specialization| Size| Block| Block| Block| | ThreadM111| ThreadN111| Thread| ClusterM110Xs| ClusterN110Xs| ThreadSliceLengths| ThreadClusterLengths| ThreadCluster| SrcAccess| SrcVectorTensor| SrcVectorTensor| DstVectorTensor| ThreadSliceLengths| ThreadClusterLengths| ThreadCluster| SrcAccess| SrcVectorTensor| SrcVectorTensor| DstVectorTensor| SrcDstAccess| SrcDstVectorDim| DstScalarPerVector|
// ##########| | | | | | | | Operation| Operation| Operation| | | | | | | | | | | | K0_M0_M1_K1| K0_M0_M1_K1| ArrangeOrder| Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder| Lengths_K0_M0_M1_K1| K0_M0_M1_K1| K0_M0_M1_K1| ArrangeOrder| Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder| Lengths_K0_M0_M1_K1| Order| | |
// ##########| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceGemmDlDpp8< F16, F16, F16, F32, Col, Row, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 8, 8, 8, 4, 2, 1, 8, 1, S<1, 8>, S<1, 1>, S<1, 1, 4, 2>, S<4, 1, 2, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 4, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 2>, S<1, 1, 4, 2>, S<4, 1, 2, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 4, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>, 5, 4>,
DeviceGemmDlDpp8< F16, F16, F16, F32, Col, Row, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 32, 8, 64, 16, 2, 1, 8, 1, S<1, 8>, S<4, 1>, S<1, 1, 4, 2>, S<16, 1, 2, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 4, 1>, S<0, 3, 1, 2>, S<1, 1, 2, 2>, S<16, 1, 2, 2>, S<1, 1, 32, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 2, 1>, S<0, 3, 1, 2>, S<1, 1, 2, 2>, S<0, 1, 2, 3, 4, 5>, 5, 4>,
DeviceGemmDlDpp8< F16, F16, F16, F32, Col, Row, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 32, 8, 64, 16, 2, 1, 8, 1, S<1, 8>, S<4, 1>, S<1, 1, 4, 2>, S<16, 1, 2, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 4, 1>, S<0, 3, 1, 2>, S<1, 1, 2, 2>, S<4, 1, 8, 2>, S<4, 1, 8, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 4, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>, 5, 4>,
DeviceGemmDlDpp8< F16, F16, F16, F32, Col, Row, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 64, 8, 64, 16, 2, 1, 8, 1, S<1, 8>, S<8, 1>, S<1, 1, 2, 2>, S<16, 1, 4, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 2, 1>, S<0, 3, 1, 2>, S<1, 1, 2, 2>, S<4, 1, 4, 2>, S<4, 1, 16, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 4, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>, 5, 4>,
DeviceGemmDlDpp8< F16, F16, F16, F32, Col, Row, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 64, 64, 64, 8, 2, 4, 8, 1, S<2, 8>, S<4, 1>, S<2, 1, 4, 2>, S<4, 1, 16, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 4, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 2>, S<2, 1, 4, 2>, S<4, 1, 16, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 4, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>, 5, 4>,
DeviceGemmDlDpp8< F16, F16, F16, F32, Col, Row, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 64, 8, 64, 16, 2, 8, 1, 1, S<1, 1>, S<8, 8>, S<1, 1, 2, 2>, S<16, 1, 4, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 2, 1>, S<0, 3, 1, 2>, S<1, 1, 2, 2>, S<4, 1, 4, 2>, S<4, 1, 16, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 4, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>, 5, 1>,
DeviceGemmDlDpp8< F16, F16, F16, F32, Col, Row, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 128, 64, 64, 16, 2, 4, 8, 1, S<2, 8>, S<8, 1>, S<2, 1, 4, 2>, S<8, 1, 16, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 4, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 2>, S<2, 1, 4, 2>, S<8, 1, 16, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 4, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>, 5, 4>,
DeviceGemmDlDpp8< F16, F16, F16, F32, Col, Row, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 128, 128, 64, 16, 2, 1, 8, 8, S<4, 8>, S<4, 1>, S<4, 1, 4, 2>, S<4, 1, 32, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 4, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 2>, S<2, 1, 4, 2>, S<8, 1, 16, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 4, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>, 5, 4>,
DeviceGemmDlDpp8< F16, F16, F16, F32, Col, Row, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 256, 128, 128, 8, 2, 1, 8, 8, S<8, 8>, S<4, 1>, S<1, 1, 4, 2>, S<8, 1, 32, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 4, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 2>, S<1, 1, 4, 2>, S<8, 1, 32, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 4, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>, 5, 4>,
DeviceGemmDlDpp8< F16, F16, F16, F32, Col, Row, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 256, 128, 128, 16, 2, 4, 8, 8, S<2, 8>, S<16, 1>, S<2, 1, 4, 2>, S<8, 1, 32, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 4, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 2>, S<2, 1, 4, 2>, S<8, 1, 32, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 4, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>, 5, 4>
// clang-format on
>;
void add_device_gemm_dl_dpp8_f16_f16_f16_km_kn_mn_instances(
std::vector<std::unique_ptr<
DeviceGemm<Col, Row, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
instances)
{
add_device_operation_instances(instances, device_gemm_dl_dpp8_f16_f16_f16_km_kn_mn_instances{});
}
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck
library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_dpp8_f16_f16_f16_km_nk_mn_instance.cpp deleted 100644 → 0
View file @ 6de7d10d
// 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_dl_dpp8.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 GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default;
// Compilation parameters for a[k, m] * b[k, n] = c[m, n]
using device_gemm_dl_dpp8_f16_f16_f16_km_nk_mn_instances = std::tuple<
// clang-format off
// ##########| AData| BData| CData| AccData| ALayout| BLayout| CLayout| A| B| C| GEMM| Block| MPer| NPer| K0Per| K1| M1Per| N1Per| KPer| M11N11Thread| M11N11Thread| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| CThreadTransfer| CThreadTransfer| CThreadTransfer|
// ##########| Type| Type| Type| Type| | | | Elementwise| Elementwise| Elementwise| Specialization| Size| Block| Block| Block| | ThreadM111| ThreadN111| Thread| ClusterM110Xs| ClusterN110Xs| ThreadSliceLengths| ThreadClusterLengths| ThreadCluster| SrcAccess| SrcVectorTensor| SrcVectorTensor| DstVectorTensor| ThreadSliceLengths| ThreadClusterLengths| ThreadCluster| SrcAccess| SrcVectorTensor| SrcVectorTensor| DstVectorTensor| SrcDstAccess| SrcDstVectorDim| DstScalarPerVector|
// ##########| | | | | | | | Operation| Operation| Operation| | | | | | | | | | | | K0_M0_M1_K1| K0_M0_M1_K1| ArrangeOrder| Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder| Lengths_K0_M0_M1_K1| K0_M0_M1_K1| K0_M0_M1_K1| ArrangeOrder| Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder| Lengths_K0_M0_M1_K1| Order| | |
// ##########| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceGemmDlDpp8< F16, F16, F16, F32, Col, Col, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 8, 8, 8, 4, 2, 1, 8, 1, S<1, 8>, S<1, 1>, S<1, 1, 4, 2>, S<4, 1, 2, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 4, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 2>, S<4, 1, 1, 2>, S<1, 1, 8, 1>, S<1, 2, 0, 3>, S<1, 2, 0, 3>, S<4, 1, 1, 2>, S<1, 2, 0, 3>, S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>, 5, 4>,
DeviceGemmDlDpp8< F16, F16, F16, F32, Col, Col, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 32, 8, 64, 16, 2, 1, 8, 1, S<1, 8>, S<4, 1>, S<1, 1, 4, 2>, S<16, 1, 2, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 4, 1>, S<0, 3, 1, 2>, S<1, 1, 2, 2>, S<16, 1, 2, 2>, S<1, 1, 32, 1>, S<1, 2, 0, 3>, S<1, 2, 0, 3>, S<4, 1, 1, 2>, S<1, 2, 0, 3>, S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>, 5, 4>,
DeviceGemmDlDpp8< F16, F16, F16, F32, Col, Col, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 32, 8, 64, 16, 2, 1, 8, 1, S<1, 8>, S<4, 1>, S<1, 1, 4, 2>, S<16, 1, 2, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 4, 1>, S<0, 3, 1, 2>, S<1, 1, 2, 2>, S<4, 1, 8, 2>, S<4, 1, 8, 1>, S<1, 2, 0, 3>, S<1, 2, 0, 3>, S<4, 1, 1, 2>, S<1, 2, 0, 3>, S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>, 5, 4>,
DeviceGemmDlDpp8< F16, F16, F16, F32, Col, Col, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 64, 8, 64, 16, 2, 1, 8, 1, S<1, 8>, S<8, 1>, S<1, 1, 2, 2>, S<16, 1, 4, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 2, 1>, S<0, 3, 1, 2>, S<1, 1, 2, 2>, S<4, 1, 4, 2>, S<4, 1, 16, 1>, S<1, 2, 0, 3>, S<1, 2, 0, 3>, S<4, 1, 1, 2>, S<1, 2, 0, 3>, S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>, 5, 4>,
DeviceGemmDlDpp8< F16, F16, F16, F32, Col, Col, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 64, 64, 64, 8, 2, 4, 8, 1, S<2, 8>, S<4, 1>, S<2, 1, 4, 2>, S<4, 1, 16, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 4, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 2>, S<4, 1, 2, 2>, S<2, 1, 32, 1>, S<1, 2, 0, 3>, S<1, 2, 0, 3>, S<4, 1, 1, 2>, S<1, 2, 0, 3>, S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>, 5, 4>,
DeviceGemmDlDpp8< F16, F16, F16, F32, Col, Col, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 64, 8, 64, 16, 2, 8, 1, 1, S<1, 1>, S<8, 8>, S<1, 1, 2, 2>, S<16, 1, 4, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 2, 1>, S<0, 3, 1, 2>, S<1, 1, 2, 2>, S<4, 1, 4, 2>, S<4, 1, 16, 1>, S<1, 2, 0, 3>, S<1, 2, 0, 3>, S<4, 1, 1, 2>, S<1, 2, 0, 3>, S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>, 5, 1>,
DeviceGemmDlDpp8< F16, F16, F16, F32, Col, Col, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 128, 64, 64, 16, 2, 4, 8, 1, S<2, 8>, S<8, 1>, S<2, 1, 4, 2>, S<8, 1, 16, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 4, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 2>, S<4, 1, 2, 2>, S<4, 1, 32, 1>, S<1, 2, 0, 3>, S<1, 2, 0, 3>, S<4, 1, 1, 2>, S<1, 2, 0, 3>, S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>, 5, 4>,
DeviceGemmDlDpp8< F16, F16, F16, F32, Col, Col, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 128, 128, 64, 16, 2, 1, 8, 8, S<4, 8>, S<4, 1>, S<4, 1, 4, 2>, S<4, 1, 32, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 4, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 2>, S<4, 1, 2, 2>, S<4, 1, 32, 1>, S<1, 2, 0, 3>, S<1, 2, 0, 3>, S<4, 1, 1, 2>, S<1, 2, 0, 3>, S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>, 5, 4>,
DeviceGemmDlDpp8< F16, F16, F16, F32, Col, Col, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 256, 128, 128, 8, 2, 1, 8, 8, S<8, 8>, S<4, 1>, S<1, 1, 4, 2>, S<8, 1, 32, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 4, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 2>, S<4, 1, 1, 2>, S<2, 1, 128, 1>, S<1, 2, 0, 3>, S<1, 2, 0, 3>, S<4, 1, 1, 2>, S<1, 2, 0, 3>, S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>, 5, 4>,
DeviceGemmDlDpp8< F16, F16, F16, F32, Col, Col, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 256, 128, 128, 16, 2, 4, 8, 8, S<2, 8>, S<16, 1>, S<2, 1, 4, 2>, S<8, 1, 32, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 4, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 2>, S<4, 1, 2, 2>, S<4, 1, 64, 1>, S<1, 2, 0, 3>, S<1, 2, 0, 3>, S<4, 1, 1, 2>, S<1, 2, 0, 3>, S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>, 5, 4>
// clang-format on
>;
void add_device_gemm_dl_dpp8_f16_f16_f16_km_nk_mn_instances(
std::vector<std::unique_ptr<
DeviceGemm<Col, Col, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
instances)
{
add_device_operation_instances(instances, device_gemm_dl_dpp8_f16_f16_f16_km_nk_mn_instances{});
}
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck
library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_dpp8_f16_f16_f16_mk_kn_mn_instance.cpp deleted 100644 → 0
View file @ 6de7d10d
// 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_dl_dpp8.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;
template <ck::index_t... Is>
using S = ck::Sequence<Is...>;
using PassThrough = ck::tensor_operation::element_wise::PassThrough;
static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default;
// Compilation parameters for a[k, m] * b[k, n] = c[m, n]
using device_gemm_dl_dpp8_f16_f16_f16_mk_kn_mn_instances = std::tuple<
// clang-format off
// ##########| AData| BData| CData| AccData| ALayout| BLayout| CLayout| A| B| C| GEMM| Block| MPer| NPer| K0Per| K1| M1Per| N1Per| KPer| M11N11Thread| M11N11Thread| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| CThreadTransfer| CThreadTransfer| CThreadTransfer|
// ##########| Type| Type| Type| Type| | | | Elementwise| Elementwise| Elementwise| Specialization| Size| Block| Block| Block| | ThreadM111| ThreadN111| Thread| ClusterM110Xs| ClusterN110Xs| ThreadSliceLengths| ThreadClusterLengths| ThreadCluster| SrcAccess| SrcVectorTensor| SrcVectorTensor| DstVectorTensor| ThreadSliceLengths| ThreadClusterLengths| ThreadCluster| SrcAccess| SrcVectorTensor| SrcVectorTensor| DstVectorTensor| SrcDstAccess| SrcDstVectorDim| DstScalarPerVector|
// ##########| | | | | | | | Operation| Operation| Operation| | | | | | | | | | | | K0_M0_M1_K1| K0_M0_M1_K1| ArrangeOrder| Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder| Lengths_K0_M0_M1_K1| K0_M0_M1_K1| K0_M0_M1_K1| ArrangeOrder| Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder| Lengths_K0_M0_M1_K1| Order| | |
// ##########| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceGemmDlDpp8< F16, F16, F16, F32, Row, Row, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 8, 8, 8, 4, 2, 1, 8, 1, S<1, 8>, S<1, 1>, S<4, 1, 1, 2>, S<1, 1, 8, 1>, S<1, 2, 0 ,3>, S<1, 2, 0, 3>, S<4, 1, 1, 2>, S<1, 2, 0, 3>, S<1, 1, 1, 2>, S<1, 1, 4, 2>, S<4, 1, 2, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 4, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>, 5, 4>,
DeviceGemmDlDpp8< F16, F16, F16, F32, Row, Row, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 32, 8, 64, 16, 2, 1, 8, 1, S<1, 8>, S<4, 1>, S<4, 1, 1, 2>, S<4, 1, 8, 1>, S<1, 2, 0, 3>, S<1, 2, 0, 3>, S<4, 1, 1, 2>, S<1, 2, 0, 3>, S<1, 1, 1, 2>, S<16, 1, 2, 2>, S<1, 1, 32, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 2, 1>, S<0, 3, 1, 2>, S<1, 1, 2, 2>, S<0, 1, 2, 3, 4, 5>, 5, 4>,
DeviceGemmDlDpp8< F16, F16, F16, F32, Row, Row, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 32, 8, 64, 16, 2, 1, 8, 1, S<1, 8>, S<4, 1>, S<4, 1, 1, 2>, S<4, 1, 8, 1>, S<1, 2, 0, 3>, S<1, 2, 0, 3>, S<4, 1, 1, 2>, S<1, 2, 0, 3>, S<1, 1, 1, 2>, S<4, 1, 8, 2>, S<4, 1, 8, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 4, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>, 5, 4>,
DeviceGemmDlDpp8< F16, F16, F16, F32, Row, Row, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 64, 8, 64, 16, 2, 1, 8, 1, S<1, 8>, S<8, 1>, S<2, 1, 1, 2>, S<8, 1, 8, 1>, S<1, 2, 0, 3>, S<1, 2, 0, 3>, S<2, 1, 1, 2>, S<1, 2, 0, 3>, S<1, 1, 1, 2>, S<4, 1, 4, 2>, S<4, 1, 16, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 4, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>, 5, 4>,
DeviceGemmDlDpp8< F16, F16, F16, F32, Row, Row, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 64, 64, 64, 8, 2, 4, 8, 1, S<2, 8>, S<4, 1>, S<4, 1, 2, 2>, S<2, 1, 32, 1>, S<1, 2, 0, 3>, S<1, 2, 0, 3>, S<4, 1, 1, 2>, S<1, 2, 0, 3>, S<1, 1, 1, 2>, S<2, 1, 4, 2>, S<4, 1, 16, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 4, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>, 5, 4>,
DeviceGemmDlDpp8< F16, F16, F16, F32, Row, Row, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 64, 8, 64, 16, 2, 8, 1, 1, S<1, 1>, S<8, 8>, S<2, 1, 1, 2>, S<8, 1, 8, 1>, S<1, 2, 0, 3>, S<1, 2, 0, 3>, S<2, 1, 1, 2>, S<1, 2, 0, 3>, S<1, 1, 1, 2>, S<4, 1, 4, 2>, S<4, 1, 16, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 4, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>, 5, 1>,
DeviceGemmDlDpp8< F16, F16, F16, F32, Row, Row, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 128, 64, 64, 16, 2, 4, 8, 1, S<2, 8>, S<8, 1>, S<4, 1, 2, 2>, S<4, 1, 32, 1>, S<1, 2, 0, 3>, S<1, 2, 0, 3>, S<4, 1, 1, 2>, S<1, 2, 0, 3>, S<1, 1, 1, 2>, S<2, 1, 4, 2>, S<8, 1, 16, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 4, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>, 5, 4>,
DeviceGemmDlDpp8< F16, F16, F16, F32, Row, Row, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 128, 128, 64, 16, 2, 1, 8, 8, S<4, 8>, S<4, 1>, S<4, 1, 4, 2>, S<4, 1, 32, 1>, S<1, 2, 0, 3>, S<1, 2, 0, 3>, S<4, 1, 1, 2>, S<1, 2, 0, 3>, S<1, 1, 1, 2>, S<2, 1, 4, 2>, S<8, 1, 16, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 4, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>, 5, 4>,
DeviceGemmDlDpp8< F16, F16, F16, F32, Row, Row, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 256, 128, 128, 8, 2, 1, 8, 8, S<8, 8>, S<4, 1>, S<4, 1, 1, 2>, S<2, 1, 128, 1>, S<1, 2, 0, 3>, S<1, 2, 0, 3>, S<4, 1, 1, 2>, S<1, 2, 0, 3>, S<1, 1, 1, 2>, S<1, 1, 4, 2>, S<8, 1, 32, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 4, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>, 5, 4>,
DeviceGemmDlDpp8< F16, F16, F16, F32, Row, Row, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 256, 128, 128, 16, 2, 4, 8, 8, S<2, 8>, S<16, 1>, S<4, 1, 2, 2>, S<4, 1, 64, 1>, S<1, 2, 0, 3>, S<1, 2, 0, 3>, S<4, 1, 1, 2>, S<1, 2, 0, 3>, S<1, 1, 1, 2>, S<2, 1, 4, 2>, S<8, 1, 32, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 4, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>, 5, 4>
// clang-format on
>;
void add_device_gemm_dl_dpp8_f16_f16_f16_mk_kn_mn_instances(
std::vector<std::unique_ptr<
DeviceGemm<Row, Row, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
instances)
{
add_device_operation_instances(instances, device_gemm_dl_dpp8_f16_f16_f16_mk_kn_mn_instances{});
}
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck
library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_dpp8_f16_f16_f16_mk_nk_mn_instance.cpp deleted 100644 → 0
View file @ 6de7d10d
// 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_dl_dpp8.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 GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default;
// Compilation parameters for a[k, m] * b[k, n] = c[m, n]
using device_gemm_dl_dpp8_f16_f16_f16_mk_nk_mn_instances = std::tuple<
// clang-format off
// ##########| AData| BData| CData| AccData| ALayout| BLayout| CLayout| A| B| C| GEMM| Block| MPer| NPer| K0Per| K1| M1Per| N1Per| KPer| M11N11Thread| M11N11Thread| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| CThreadTransfer| CThreadTransfer| CThreadTransfer|
// ##########| Type| Type| Type| Type| | | | Elementwise| Elementwise| Elementwise| Specialization| Size| Block| Block| Block| | ThreadM111| ThreadN111| Thread| ClusterM110Xs| ClusterN110Xs| ThreadSliceLengths| ThreadClusterLengths| ThreadCluster| SrcAccess| SrcVectorTensor| SrcVectorTensor| DstVectorTensor| ThreadSliceLengths| ThreadClusterLengths| ThreadCluster| SrcAccess| SrcVectorTensor| SrcVectorTensor| DstVectorTensor| SrcDstAccess| SrcDstVectorDim| DstScalarPerVector|
// ##########| | | | | | | | Operation| Operation| Operation| | | | | | | | | | | | K0_M0_M1_K1| K0_M0_M1_K1| ArrangeOrder| Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder| Lengths_K0_M0_M1_K1| K0_M0_M1_K1| K0_M0_M1_K1| ArrangeOrder| Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder| Lengths_K0_M0_M1_K1| Order| | |
// ##########| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceGemmDlDpp8< F16, F16, F16, F32, Row, Col, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 8, 8, 8, 4, 2, 1, 8, 1, S<1, 8>, S<1, 1>, S<4, 1, 1, 2>, S<1, 1, 8, 1>, S<1, 2, 0 ,3>, S<1, 2, 0, 3>, S<4, 1, 1, 2>, S<1, 2, 0, 3>, S<1, 1, 1, 2>, S<4, 1, 1, 2>, S<1, 1, 8, 1>, S<1, 2, 0, 3>, S<1, 2, 0, 3>, S<4, 1, 1, 2>, S<1, 2, 0, 3>, S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>, 5, 4>,
DeviceGemmDlDpp8< F16, F16, F16, F32, Row, Col, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 32, 8, 64, 16, 2, 1, 8, 1, S<1, 8>, S<4, 1>, S<4, 1, 1, 2>, S<4, 1, 8, 1>, S<1, 2, 0, 3>, S<1, 2, 0, 3>, S<4, 1, 1, 2>, S<1, 2, 0, 3>, S<1, 1, 1, 2>, S<16, 1, 2, 2>, S<1, 1, 32, 1>, S<1, 2, 0, 3>, S<1, 2, 0, 3>, S<4, 1, 1, 2>, S<1, 2, 0, 3>, S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>, 5, 4>,
DeviceGemmDlDpp8< F16, F16, F16, F32, Row, Col, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 32, 8, 64, 16, 2, 1, 8, 1, S<1, 8>, S<4, 1>, S<4, 1, 1, 2>, S<4, 1, 8, 1>, S<1, 2, 0, 3>, S<1, 2, 0, 3>, S<4, 1, 1, 2>, S<1, 2, 0, 3>, S<1, 1, 1, 2>, S<4, 1, 8, 2>, S<4, 1, 8, 1>, S<1, 2, 0, 3>, S<1, 2, 0, 3>, S<4, 1, 1, 2>, S<1, 2, 0, 3>, S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>, 5, 4>,
DeviceGemmDlDpp8< F16, F16, F16, F32, Row, Col, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 64, 8, 64, 16, 2, 1, 8, 1, S<1, 8>, S<8, 1>, S<2, 1, 1, 2>, S<8, 1, 8, 1>, S<1, 2, 0, 3>, S<1, 2, 0, 3>, S<2, 1, 1, 2>, S<1, 2, 0, 3>, S<1, 1, 1, 2>, S<4, 1, 4, 2>, S<4, 1, 16, 1>, S<1, 2, 0, 3>, S<1, 2, 0, 3>, S<4, 1, 1, 2>, S<1, 2, 0, 3>, S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>, 5, 4>,
DeviceGemmDlDpp8< F16, F16, F16, F32, Row, Col, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 64, 64, 64, 8, 2, 4, 8, 1, S<2, 8>, S<4, 1>, S<4, 1, 2, 2>, S<2, 1, 32, 1>, S<1, 2, 0, 3>, S<1, 2, 0, 3>, S<4, 1, 1, 2>, S<1, 2, 0, 3>, S<1, 1, 1, 2>, S<4, 1, 2, 2>, S<2, 1, 32, 1>, S<1, 2, 0, 3>, S<1, 2, 0, 3>, S<4, 1, 1, 2>, S<1, 2, 0, 3>, S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>, 5, 4>,
DeviceGemmDlDpp8< F16, F16, F16, F32, Row, Col, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 64, 8, 64, 16, 2, 8, 1, 1, S<1, 1>, S<8, 8>, S<2, 1, 1, 2>, S<8, 1, 8, 1>, S<1, 2, 0, 3>, S<1, 2, 0, 3>, S<2, 1, 1, 2>, S<1, 2, 0, 3>, S<1, 1, 1, 2>, S<4, 1, 4, 2>, S<4, 1, 16, 1>, S<1, 2, 0, 3>, S<1, 2, 0, 3>, S<4, 1, 1, 2>, S<1, 2, 0, 3>, S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>, 5, 1>,
DeviceGemmDlDpp8< F16, F16, F16, F32, Row, Col, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 128, 64, 64, 16, 2, 4, 8, 1, S<2, 8>, S<8, 1>, S<4, 1, 2, 2>, S<4, 1, 32, 1>, S<1, 2, 0, 3>, S<1, 2, 0, 3>, S<4, 1, 1, 2>, S<1, 2, 0, 3>, S<1, 1, 1, 2>, S<4, 1, 2, 2>, S<4, 1, 32, 1>, S<1, 2, 0, 3>, S<1, 2, 0, 3>, S<4, 1, 1, 2>, S<1, 2, 0, 3>, S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>, 5, 4>,
DeviceGemmDlDpp8< F16, F16, F16, F32, Row, Col, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 128, 128, 64, 16, 2, 1, 8, 8, S<4, 8>, S<4, 1>, S<4, 1, 4, 2>, S<4, 1, 32, 1>, S<1, 2, 0, 3>, S<1, 2, 0, 3>, S<4, 1, 1, 2>, S<1, 2, 0, 3>, S<1, 1, 1, 2>, S<4, 1, 2, 2>, S<4, 1, 32, 1>, S<1, 2, 0, 3>, S<1, 2, 0, 3>, S<4, 1, 1, 2>, S<1, 2, 0, 3>, S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>, 5, 4>,
DeviceGemmDlDpp8< F16, F16, F16, F32, Row, Col, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 256, 128, 128, 8, 2, 1, 8, 8, S<8, 8>, S<4, 1>, S<4, 1, 1, 2>, S<2, 1, 128, 1>, S<1, 2, 0, 3>, S<1, 2, 0, 3>, S<4, 1, 1, 2>, S<1, 2, 0, 3>, S<1, 1, 1, 2>, S<4, 1, 1, 2>, S<2, 1, 128, 1>, S<1, 2, 0, 3>, S<1, 2, 0, 3>, S<4, 1, 1, 2>, S<1, 2, 0, 3>, S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>, 5, 4>,
DeviceGemmDlDpp8< F16, F16, F16, F32, Row, Col, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 256, 128, 128, 16, 2, 4, 8, 8, S<2, 8>, S<16, 1>, S<4, 1, 2, 2>, S<4, 1, 64, 1>, S<1, 2, 0, 3>, S<1, 2, 0, 3>, S<4, 1, 1, 2>, S<1, 2, 0, 3>, S<1, 1, 1, 2>, S<4, 1, 2, 2>, S<4, 1, 64, 1>, S<1, 2, 0, 3>, S<1, 2, 0, 3>, S<4, 1, 1, 2>, S<1, 2, 0, 3>, S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>, 5, 4>
// clang-format on
>;
void add_device_gemm_dl_dpp8_f16_f16_f16_mk_nk_mn_instances(
std::vector<std::unique_ptr<
DeviceGemm<Row, Col, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
instances)
{
add_device_operation_instances(instances, device_gemm_dl_dpp8_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_dpp_f16_f16_f16_km_kn_mn_instance.cpp 0 → 100644
View file @ 4173b984
// 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_dpp.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 GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default;
// Compilation parameters for a[k, m] * b[k, n] = c[m, n]
// clang-format off
using device_gemm_dpp_f16_f16_f16_km_kn_mn_instances = std::tuple<
// ########| AData| BData| CData| AccData| ALayout| BLayout| CLayout| A| B| C| GEMM| Block| MPer| NPer| KPer| AK1| BK1| MPer| NPer| MDpp| NDpp| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer|
// ########| Type| Type| Type| Type| | | | Elementwise| Elementwise| Elementwise| Specialization| Size| Block| Block| Block| | | Dpp| Dpp| PerWave| PerWave| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| SrcDstVectorDim| DstScalar|
// ########| | | | | | | | Operation| Operation| Operation| | | | | | | | | | | | Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector|
// ########| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceGemmDpp< F16, F16, F16, F32, Col, Row, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 256, 128, 128, 64, 4, 4, 16, 16, 2, 4, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 4, true, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 4, true, 5, 1>,
DeviceGemmDpp< F16, F16, F16, F32, Col, Row, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 256, 128, 128, 64, 4, 4, 32, 8, 2, 4, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 4, true, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 4, true, 5, 1>,
DeviceGemmDpp< F16, F16, F16, F32, Col, Row, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 128, 64, 64, 64, 4, 4, 32, 8, 2, 2, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 4, true, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 4, true, 5, 1>,
DeviceGemmDpp< F16, F16, F16, F32, Col, Row, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 128, 32, 32, 32, 4, 4, 32, 8, 1, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 4, true, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 4, true, 5, 1>,
DeviceGemmDpp< F16, F16, F16, F32, Col, Row, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 64, 64, 64, 64, 4, 4, 32, 8, 2, 4, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 4, true, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 4, true, 5, 1>,
DeviceGemmDpp< F16, F16, F16, F32, Col, Row, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 32, 32, 32, 32, 4, 4, 32, 8, 1, 4, S<4, 8, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 4, true, S<4, 8, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 4, true, 5, 1>,
DeviceGemmDpp< F16, F16, F16, F32, Col, Row, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 32, 16, 16, 16, 4, 4, 16, 16, 1, 1, S<2, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 4, true, S<2, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 4, true, 5, 1>
>;
// clang-format on
void add_device_gemm_dpp_f16_f16_f16_km_kn_mn_instances(
std::vector<std::unique_ptr<
DeviceGemm<Col, Row, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
instances)
{
add_device_operation_instances(instances, device_gemm_dpp_f16_f16_f16_km_kn_mn_instances{});
}
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck
library/src/tensor_operation_instance/gpu/gemm/device_gemm_dpp_f16_f16_f16_km_nk_mn_instance.cpp 0 → 100644
View file @ 4173b984
// 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_dpp.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 GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default;
// Compilation parameters for a[k, m] * b[n, k] = c[m, n]
// clang-format off
using device_gemm_dpp_f16_f16_f16_km_nk_mn_instances = std::tuple<
// ########| AData| BData| CData| AccData| ALayout| BLayout| CLayout| A| B| C| GEMM| Block| MPer| NPer| KPer| AK1| BK1| MPer| NPer| MDpp| NDpp| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer|
// ########| Type| Type| Type| Type| | | | Elementwise| Elementwise| Elementwise| Specialization| Size| Block| Block| Block| | | Dpp| Dpp| PerWave| PerWave| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| SrcDstVectorDim| DstScalar|
// ########| | | | | | | | Operation| Operation| Operation| | | | | | | | | | | | Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector|
// ########| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceGemmDpp< F16, F16, F16, F32, Col, Col, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 256, 128, 128, 64, 4, 8, 16, 16, 2, 4, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 4, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 5, 1>,
DeviceGemmDpp< F16, F16, F16, F32, Col, Col, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 256, 128, 128, 64, 4, 8, 32, 8, 2, 4, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 4, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 5, 1>,
DeviceGemmDpp< F16, F16, F16, F32, Col, Col, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 128, 64, 64, 64, 4, 8, 32, 8, 2, 2, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 4, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 5, 1>,
DeviceGemmDpp< F16, F16, F16, F32, Col, Col, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 128, 32, 32, 32, 4, 8, 32, 8, 1, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 4, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 5, 1>,
DeviceGemmDpp< F16, F16, F16, F32, Col, Col, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 64, 64, 64, 64, 4, 8, 32, 8, 2, 4, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 4, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 5, 1>,
DeviceGemmDpp< F16, F16, F16, F32, Col, Col, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 32, 32, 32, 32, 4, 8, 32, 8, 1, 4, S<4, 8, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 4, true, S<4, 8, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 5, 1>,
DeviceGemmDpp< F16, F16, F16, F32, Col, Col, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 32, 16, 16, 16, 4, 8, 16, 16, 1, 1, S<2, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 4, true, S<2, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 5, 1>
>;
// clang-format on
void add_device_gemm_dpp_f16_f16_f16_km_nk_mn_instances(
std::vector<std::unique_ptr<
DeviceGemm<Col, Col, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
instances)
{
add_device_operation_instances(instances, device_gemm_dpp_f16_f16_f16_km_nk_mn_instances{});
}
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck
library/src/tensor_operation_instance/gpu/gemm/device_gemm_dpp_f16_f16_f16_mk_kn_mn_instance.cpp 0 → 100644
View file @ 4173b984
// 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_dpp.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;
template <ck::index_t... Is>
using S = ck::Sequence<Is...>;
using PassThrough = ck::tensor_operation::element_wise::PassThrough;
static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default;
// Compilation parameters for a[m, k] * b[k, n] = c[m, n]
// clang-format off
using device_gemm_dpp_f16_f16_f16_mk_kn_mn_instances = std::tuple<
// ########| AData| BData| CData| AccData| ALayout| BLayout| CLayout| A| B| C| GEMM| Block| MPer| NPer| KPer| AK1| BK1| MPer| NPer| MDpp| NDpp| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer|
// ########| Type| Type| Type| Type| | | | Elementwise| Elementwise| Elementwise| Specialization| Size| Block| Block| Block| | | Dpp| Dpp| PerWave| PerWave| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| SrcDstVectorDim| DstScalar|
// ########| | | | | | | | Operation| Operation| Operation| | | | | | | | | | | | Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector|
// ########| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceGemmDpp< F16, F16, F16, F32, Row, Row, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 256, 128, 128, 64, 8, 4, 16, 16, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 4, true, 5, 1>,
DeviceGemmDpp< F16, F16, F16, F32, Row, Row, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 256, 128, 128, 64, 8, 4, 32, 8, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 4, true, 5, 1>,
DeviceGemmDpp< F16, F16, F16, F32, Row, Row, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 128, 64, 64, 64, 8, 4, 32, 8, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 4, true, 5, 1>,
DeviceGemmDpp< F16, F16, F16, F32, Row, Row, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 128, 32, 32, 32, 8, 4, 32, 8, 1, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 4, true, 5, 1>,
DeviceGemmDpp< F16, F16, F16, F32, Row, Row, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 64, 64, 64, 64, 8, 4, 32, 8, 2, 4, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 4, true, 5, 1>,
DeviceGemmDpp< F16, F16, F16, F32, Row, Row, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 32, 32, 32, 32, 8, 4, 32, 8, 1, 4, S<4, 8, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 8, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 4, true, 5, 1>,
DeviceGemmDpp< F16, F16, F16, F32, Row, Row, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 32, 16, 16, 16, 8, 4, 16, 16, 1, 1, S<2, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<2, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 4, true, 5, 1>
>;
// clang-format on
void add_device_gemm_dpp_f16_f16_f16_mk_kn_mn_instances(
std::vector<std::unique_ptr<
DeviceGemm<Row, Row, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
instances)
{
add_device_operation_instances(instances, device_gemm_dpp_f16_f16_f16_mk_kn_mn_instances{});
}
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck
library/src/tensor_operation_instance/gpu/gemm/device_gemm_dpp_f16_f16_f16_mk_nk_mn_instance.cpp 0 → 100644
View file @ 4173b984
// 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_dpp.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 GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default;
// Compilation parameters for a[m, k] * b[n, k] = c[m, n]
// clang-format off
using device_gemm_dpp_f16_f16_f16_mk_nk_mn_instances = std::tuple<
// ########| AData| BData| CData| AccData| ALayout| BLayout| CLayout| A| B| C| GEMM| Block| MPer| NPer| KPer| AK1| BK1| MPer| NPer| MDpp| NDpp| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer|
// ########| Type| Type| Type| Type| | | | Elementwise| Elementwise| Elementwise| Specialization| Size| Block| Block| Block| | | Dpp| Dpp| PerWave| PerWave| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| SrcDstVectorDim| DstScalar|
// ########| | | | | | | | Operation| Operation| Operation| | | | | | | | | | | | Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector|
// ########| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceGemmDpp< F16, F16, F16, F32, Row, Col, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 256, 128, 128, 64, 8, 8, 16, 16, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 5, 1>,
DeviceGemmDpp< F16, F16, F16, F32, Row, Col, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 256, 128, 128, 64, 8, 8, 32, 8, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 5, 1>,
DeviceGemmDpp< F16, F16, F16, F32, Row, Col, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 128, 64, 64, 64, 8, 8, 32, 8, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 5, 1>,
DeviceGemmDpp< F16, F16, F16, F32, Row, Col, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 128, 32, 32, 32, 8, 8, 32, 8, 1, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 5, 1>,
DeviceGemmDpp< F16, F16, F16, F32, Row, Col, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 64, 64, 64, 64, 8, 8, 32, 8, 2, 4, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 5, 1>,
DeviceGemmDpp< F16, F16, F16, F32, Row, Col, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 32, 32, 32, 32, 8, 8, 32, 8, 1, 4, S<4, 8, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 8, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 5, 1>,
DeviceGemmDpp< F16, F16, F16, F32, Row, Col, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 32, 16, 16, 16, 8, 8, 16, 16, 1, 1, S<2, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<2, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 5, 1>
>;
// clang-format on
void add_device_gemm_dpp_f16_f16_f16_mk_nk_mn_instances(
std::vector<std::unique_ptr<
DeviceGemm<Row, Col, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
instances)
{
add_device_operation_instances(instances, device_gemm_dpp_f16_f16_f16_mk_nk_mn_instances{});
}
} // 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