diff --git a/library/include/ck/library/tensor_operation_instance/gpu/grouped_gemm_fixed_nk.hpp b/library/include/ck/library/tensor_operation_instance/gpu/grouped_gemm_fixed_nk.hpp index a90fe14603..850183c513 100644 --- a/library/include/ck/library/tensor_operation_instance/gpu/grouped_gemm_fixed_nk.hpp +++ b/library/include/ck/library/tensor_operation_instance/gpu/grouped_gemm_fixed_nk.hpp @@ -126,6 +126,35 @@ void add_device_grouped_gemm_xdl_fixed_nk_bf16_i8_bf16_mk_nk_mn_instances( PassThrough>>>& instances); #endif +// bf16_inputA bf16_inputB +#if defined(CK_ENABLE_BF16) +void add_device_grouped_gemm_xdl_fixed_nk_bf16_bf16_bf16_mk_kn_mn_instances( + std::vector>>& instances); + +void add_device_grouped_gemm_xdl_fixed_nk_bf16_bf16_bf16_mk_nk_mn_instances( + std::vector>>& instances); +#endif // CK_ENABLE_BF16 + template && is_same_v && + is_same_v) + { + if constexpr(is_same_v && is_same_v && + is_same_v) + { + add_device_grouped_gemm_xdl_fixed_nk_bf16_bf16_bf16_mk_kn_mn_instances(op_ptrs); + } + if constexpr(is_same_v && is_same_v && + is_same_v) + { + add_device_grouped_gemm_xdl_fixed_nk_bf16_bf16_bf16_mk_nk_mn_instances(op_ptrs); + } + } +#endif // CK_ENABLE_BF16 + return op_ptrs; } }; diff --git a/library/src/tensor_operation_instance/gpu/grouped_gemm_fixed_nk/CMakeLists.txt b/library/src/tensor_operation_instance/gpu/grouped_gemm_fixed_nk/CMakeLists.txt index 1ee3d0add4..bc9c711d3a 100644 --- a/library/src/tensor_operation_instance/gpu/grouped_gemm_fixed_nk/CMakeLists.txt +++ b/library/src/tensor_operation_instance/gpu/grouped_gemm_fixed_nk/CMakeLists.txt @@ -8,6 +8,8 @@ list(APPEND GROUPED_GEMM_FIXED_NK_INSTANCES device_grouped_gemm_xdl_fixed_nk_f16 device_grouped_gemm_xdl_fixed_nk_f16_i8_f16_mk_kn_mn_instance.cpp device_grouped_gemm_xdl_fixed_nk_f16_i8_f16_mk_nk_mn_instance.cpp device_grouped_gemm_xdl_fixed_nk_bf16_i8_bf16_mk_kn_mn_instance.cpp - device_grouped_gemm_xdl_fixed_nk_bf16_i8_bf16_mk_nk_mn_instance.cpp) + device_grouped_gemm_xdl_fixed_nk_bf16_i8_bf16_mk_nk_mn_instance.cpp + device_grouped_gemm_xdl_fixed_nk_bf16_bf16_bf16_mk_kn_mn_instance.cpp + device_grouped_gemm_xdl_fixed_nk_bf16_bf16_bf16_mk_nk_mn_instance.cpp) -add_instance_library(device_grouped_gemm_fixed_nk_instance ${GROUPED_GEMM_FIXED_NK_INSTANCES}) +add_instance_library(device_grouped_gemm_fixed_nk_instance ${GROUPED_GEMM_FIXED_NK_INSTANCES}) \ No newline at end of file diff --git a/library/src/tensor_operation_instance/gpu/grouped_gemm_fixed_nk/device_grouped_gemm_xdl_fixed_nk_bf16_bf16_bf16_mk_kn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_gemm_fixed_nk/device_grouped_gemm_xdl_fixed_nk_bf16_bf16_bf16_mk_kn_mn_instance.cpp new file mode 100644 index 0000000000..5fdde2bfbc --- /dev/null +++ b/library/src/tensor_operation_instance/gpu/grouped_gemm_fixed_nk/device_grouped_gemm_xdl_fixed_nk_bf16_bf16_bf16_mk_kn_mn_instance.cpp @@ -0,0 +1,73 @@ +// SPDX-License-Identifier: MIT +// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved. + +#include + +#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_grouped_gemm_xdl_fixed_nk.hpp" + +#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp" + +namespace ck { +namespace tensor_operation { +namespace device { +namespace instance { + +using BF16 = ck::bhalf_t; +using F32 = float; + +using Row = ck::tensor_layout::gemm::RowMajor; +using Col = ck::tensor_layout::gemm::ColumnMajor; + +template +using S = ck::Sequence; + +using DsDataType = ck::Tuple<>; +using DsLayout = ck::Tuple<>; +using PassThrough = ck::tensor_operation::element_wise::PassThrough; +static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecialization::MNKPadding; + +using device_grouped_gemm_xdl_fixed_nk_bf16_bf16_bf16_mk_kn_mn_irregular_tile_instances = + std::tuple< + // clang-format off + //############################| A| B| Ds| E| AData| BData| AccData| CShuffle| DsData| EData| A| B| C| GEMM| NumGemmK| Block| MPer| NPer| KPer| AK1| BK1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CShuffle| CShuffle| CBlockTransferClusterLengths| CBlockTransfer| + //############################| Layout| Layout| Layout| Layout| Type| Type| Type| DataType| Type| Type| Elementwise| Elementwise| Elementwise| Spacialization| Prefetch| Size| Block| Block| Block| | | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave| _MBlock_MWaveMPerXdl| ScalarPerVector| + //############################| | | | | | | | | | | Operation| Operation| Operation| | Stage| | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| _NBlock_NWaveNPerXdl| _NWaveNPerXdl| + //############################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | + DeviceGroupedGemm_Xdl_Fixed_NK< Row, Row, DsLayout, Row, BF16, BF16, F32, F32, DsDataType, BF16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 256, 256, 128, 32, 8, 8, 32, 32, 4, 2, S< 1, 4, 64, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, S<1, 4, 64, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 2, 2, 8, 1, 1, 1, S<1, 32, 1, 8>, 8>, + DeviceGroupedGemm_Xdl_Fixed_NK< Row, Row, DsLayout, Row, BF16, BF16, F32, F32, DsDataType, BF16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 256, 128, 128, 32, 8, 8, 32, 32, 2, 2, S< 1, 4, 64, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, S<1, 4, 64, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 2, 1, 8, 1, 1, 1, S<1, 32, 1, 8>, 8>, + DeviceGroupedGemm_Xdl_Fixed_NK< Row, Row, DsLayout, Row, BF16, BF16, F32, F32, DsDataType, BF16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 256, 128, 64, 32, 8, 2, 32, 32, 2, 1, S< 1, 4, 64, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, S<1, 16,16, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 2, 4, 2, 0, 1, 1, S<1, 32, 1, 8>, 8>, + DeviceGroupedGemm_Xdl_Fixed_NK< Row, Row, DsLayout, Row, BF16, BF16, F32, F32, DsDataType, BF16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 256, 128, 64, 32, 8, 8, 32, 32, 2, 1, S< 1, 4, 64, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, S<1, 4, 64, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 2, 1, 8, 1, 1, 1, S<1, 32, 1, 8>, 8>, + DeviceGroupedGemm_Xdl_Fixed_NK< Row, Row, DsLayout, Row, BF16, BF16, F32, F32, DsDataType, BF16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 256, 64, 128, 32, 8, 2, 32, 32, 1, 2, S< 1, 4, 64, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, S<1, 8, 32, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 2, 4, 2, 0, 1, 1, S<1, 32, 1, 8>, 8>, + DeviceGroupedGemm_Xdl_Fixed_NK< Row, Row, DsLayout, Row, BF16, BF16, F32, F32, DsDataType, BF16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 256, 64, 128, 32, 8, 8, 32, 32, 1, 2, S< 1, 4, 64, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, S<1, 4, 64, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 2, 2, 8, 1, 1, 1, S<1, 32, 1, 8>, 8>, + DeviceGroupedGemm_Xdl_Fixed_NK< Row, Row, DsLayout, Row, BF16, BF16, F32, F32, DsDataType, BF16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 128, 128, 64, 32, 8, 2, 32, 32, 2, 2, S< 1, 4, 32, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, S<1, 8, 16, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 2, 4, 2, 0, 1, 1, S<1, 32, 1, 4>, 8>, + DeviceGroupedGemm_Xdl_Fixed_NK< Row, Row, DsLayout, Row, BF16, BF16, F32, F32, DsDataType, BF16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 128, 128, 64, 32, 8, 8, 32, 32, 2, 2, S< 1, 4, 32, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, S<1, 4, 32, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 2, 2, 8, 1, 1, 1, S<1, 32, 1, 4>, 8>, + DeviceGroupedGemm_Xdl_Fixed_NK< Row, Row, DsLayout, Row, BF16, BF16, F32, F32, DsDataType, BF16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 128, 64, 128, 32, 8, 2, 32, 32, 2, 2, S< 1, 4, 32, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, S<1, 4, 32, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 2, 4, 2, 0, 1, 1, S<1, 16, 1, 8>, 8>, + DeviceGroupedGemm_Xdl_Fixed_NK< Row, Row, DsLayout, Row, BF16, BF16, F32, F32, DsDataType, BF16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 128, 64, 128, 32, 8, 8, 32, 32, 2, 2, S< 1, 4, 32, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, S<1, 4, 32, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 2, 4, 8, 1, 1, 1, S<1, 16, 1, 8>, 8> + // clang-format on + >; + +void add_device_grouped_gemm_xdl_fixed_nk_bf16_bf16_bf16_mk_kn_mn_instances( + std::vector>>& instances) +{ + add_device_operation_instances( + instances, + device_grouped_gemm_xdl_fixed_nk_bf16_bf16_bf16_mk_kn_mn_irregular_tile_instances{}); +} + +} // namespace instance +} // namespace device +} // namespace tensor_operation +} // namespace ck diff --git a/library/src/tensor_operation_instance/gpu/grouped_gemm_fixed_nk/device_grouped_gemm_xdl_fixed_nk_bf16_bf16_bf16_mk_nk_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_gemm_fixed_nk/device_grouped_gemm_xdl_fixed_nk_bf16_bf16_bf16_mk_nk_mn_instance.cpp new file mode 100644 index 0000000000..32db426cb2 --- /dev/null +++ b/library/src/tensor_operation_instance/gpu/grouped_gemm_fixed_nk/device_grouped_gemm_xdl_fixed_nk_bf16_bf16_bf16_mk_nk_mn_instance.cpp @@ -0,0 +1,76 @@ +// SPDX-License-Identifier: MIT +// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved. + +#include + +#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_grouped_gemm_xdl_fixed_nk.hpp" + +#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp" + +namespace ck { +namespace tensor_operation { +namespace device { +namespace instance { + +using BF16 = ck::bhalf_t; +using F32 = float; + +using Row = ck::tensor_layout::gemm::RowMajor; +using Col = ck::tensor_layout::gemm::ColumnMajor; + +template +using S = ck::Sequence; + +using DsDataType = ck::Tuple<>; +using DsLayout = ck::Tuple<>; +using PassThrough = ck::tensor_operation::element_wise::PassThrough; +static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecialization::MNKPadding; + +using device_grouped_gemm_xdl_fixed_nk_bf16_bf16_bf16_mk_nk_mn_irregular_tile_instances = + std::tuple< + // clang-format off + //############################| A| B| Ds| E| AData| BData| AccData| CShuffle| DsData| EData| A| B| CDE| GEMM| NumGemmK| Block| MPer| NPer| KPer| AK1| BK1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CShuffle| CShuffle| CBlockTransferClusterLengths| CBlockTransfer| + //############################| Layout| Layout| Layout| Layout| Type| Type| Type| DataType| Type| Type| Elementwise| Elementwise| Elementwise| Spacialization| Prefetch| Size| Block| Block| Block| | | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave| _MBlock_MWaveMPerXdl| ScalarPerVector| + //############################| | | | | | | | | | | Operation| Operation| Operation| | Stage| | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| _NBlock_NWaveNPerXdl| _NWaveNPerXdl| + //############################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | + DeviceGroupedGemm_Xdl_Fixed_NK< Row, Col, DsLayout, Row, BF16, BF16, F32, F32, DsDataType, BF16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 256, 128, 256, 64, 8, 8, 32, 32, 2, 4, S<1, 8, 32, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, S<1, 8, 32, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8>, + DeviceGroupedGemm_Xdl_Fixed_NK< Row, Col, DsLayout, Row, BF16, BF16, F32, F32, DsDataType, BF16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 256, 128, 128, 64, 8, 8, 32, 32, 2, 2, S<1, 8, 32, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, S<1, 8, 32, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8>, + DeviceGroupedGemm_Xdl_Fixed_NK< Row, Col, DsLayout, Row, BF16, BF16, F32, F32, DsDataType, BF16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 256, 128, 64, 64, 8, 8, 32, 32, 2, 1, S<1, 8, 32, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, S<1, 8, 32, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8>, + DeviceGroupedGemm_Xdl_Fixed_NK< Row, Col, DsLayout, Row, BF16, BF16, F32, F32, DsDataType, BF16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 256, 64, 128, 64, 8, 8, 32, 32, 1, 2, S<1, 8, 32, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, S<1, 8, 32, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8>, + DeviceGroupedGemm_Xdl_Fixed_NK< Row, Col, DsLayout, Row, BF16, BF16, F32, F32, DsDataType, BF16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 128, 128, 128, 64, 8, 8, 32, 32, 4, 2, S<1, 8, 16, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, S<1, 8, 16, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, 1, 1, S<1, 16, 1, 8>, 8>, + DeviceGroupedGemm_Xdl_Fixed_NK< Row, Col, DsLayout, Row, BF16, BF16, F32, F32, DsDataType, BF16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 128, 128, 64, 64, 8, 8, 32, 32, 2, 2, S<1, 8, 16, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, S<1, 8, 16, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, 1, 1, S<1, 32, 1, 4>, 8>, + DeviceGroupedGemm_Xdl_Fixed_NK< Row, Col, DsLayout, Row, BF16, BF16, F32, F32, DsDataType, BF16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 128, 64, 128, 64, 8, 8, 32, 32, 2, 2, S<1, 8, 16, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, S<1, 8, 16, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, 1, 1, S<1, 16, 1, 8>, 8>, + DeviceGroupedGemm_Xdl_Fixed_NK< Row, Col, DsLayout, Row, BF16, BF16, F32, F32, DsDataType, BF16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 128, 128, 32, 64, 8, 8, 32, 32, 2, 1, S<1, 8, 16, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, S<1, 8, 16, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, 1, 1, S<1, 32, 1, 4>, 8>, + DeviceGroupedGemm_Xdl_Fixed_NK< Row, Col, DsLayout, Row, BF16, BF16, F32, F32, DsDataType, BF16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 128, 32, 128, 64, 8, 8, 32, 32, 1, 2, S<1, 8, 16, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, S<1, 8, 16, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, 1, 1, S<1, 16, 1, 8>, 8>, + DeviceGroupedGemm_Xdl_Fixed_NK< Row, Col, DsLayout, Row, BF16, BF16, F32, F32, DsDataType, BF16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 128, 32, 256, 64, 8, 8, 32, 32, 1, 4, S<1, 8, 16, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, S<1, 8, 16, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, 1, 1, S<1, 16, 1, 8>, 8>, + DeviceGroupedGemm_Xdl_Fixed_NK< Row, Col, DsLayout, Row, BF16, BF16, F32, F32, DsDataType, BF16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 64, 64, 64, 64, 8, 8, 32, 32, 2, 2, S<1, 8, 8, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, S<1, 8, 8, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, 1, 1, S<1, 16, 1, 4>, 8>, + DeviceGroupedGemm_Xdl_Fixed_NK< Row, Col, DsLayout, Row, BF16, BF16, F32, F32, DsDataType, BF16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 64, 64, 32, 64, 8, 8, 32, 32, 2, 1, S<1, 8, 8, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, S<1, 8, 8, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, 1, 1, S<1, 16, 1, 4>, 8>, + DeviceGroupedGemm_Xdl_Fixed_NK< Row, Col, DsLayout, Row, BF16, BF16, F32, F32, DsDataType, BF16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 64, 32, 64, 64, 8, 8, 32, 32, 1, 2, S<1, 8, 8, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, S<1, 8, 8, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, 1, 1, S<1, 16, 1, 4>, 8> + // clang-format on + >; + +void add_device_grouped_gemm_xdl_fixed_nk_bf16_bf16_bf16_mk_nk_mn_instances( + std::vector>>& instances) +{ + add_device_operation_instances( + instances, + device_grouped_gemm_xdl_fixed_nk_bf16_bf16_bf16_mk_nk_mn_irregular_tile_instances{}); +} + +} // namespace instance +} // namespace device +} // namespace tensor_operation +} // namespace ck diff --git a/profiler/src/profile_grouped_gemm_fixed_nk.cpp b/profiler/src/profile_grouped_gemm_fixed_nk.cpp index 093557e7f3..827866ce69 100644 --- a/profiler/src/profile_grouped_gemm_fixed_nk.cpp +++ b/profiler/src/profile_grouped_gemm_fixed_nk.cpp @@ -17,10 +17,11 @@ enum struct GemmMatrixLayout enum struct GemmDataType { - BF16_I8_BF16, // 0 - F16_F16_F16, // 1 - F16_F8_F16, // 2 - F16_I8_F16, // 3 + BF16_I8_BF16, // 0 + F16_F16_F16, // 1 + F16_F8_F16, // 2 + F16_I8_F16, // 3 + BF16_BF16_BF16 // 4 }; #define OP_NAME "grouped_gemm_fixed_nk" @@ -182,7 +183,7 @@ int profile_grouped_gemm_fixed_nk(int argc, char* argv[]) n_warmup, n_iter); } -#endif +#endif // CK_ENABLE_FP8 #if defined(CK_ENABLE_INT8) else if(data_type == GemmDataType::F16_I8_F16 && layout == GemmMatrixLayout::MK_KN_MN) { @@ -226,12 +227,58 @@ int profile_grouped_gemm_fixed_nk(int argc, char* argv[]) StrideAs, StrideBs, StrideCs, - 1, + kbatch, n_warmup, n_iter); } -#endif +#endif // CK_ENABLE_INT8 #if defined(CK_ENABLE_BF16) + else if(data_type == GemmDataType::BF16_BF16_BF16 && layout == GemmMatrixLayout::MK_KN_MN) + { + ck::profiler::profile_grouped_gemm_fixed_nk_impl( + do_verification, + init_method, + do_log, + time_kernel, + Ms, + Ns, + Ks, + StrideAs, + StrideBs, + StrideCs, + kbatch, + n_warmup, + n_iter); + } + else if(data_type == GemmDataType::BF16_BF16_BF16 && layout == GemmMatrixLayout::MK_NK_MN) + { + ck::profiler::profile_grouped_gemm_fixed_nk_impl( + do_verification, + init_method, + do_log, + time_kernel, + Ms, + Ns, + Ks, + StrideAs, + StrideBs, + StrideCs, + kbatch, + n_warmup, + n_iter); + } #if defined(CK_ENABLE_INT8) else if(data_type == GemmDataType::BF16_I8_BF16 && layout == GemmMatrixLayout::MK_KN_MN) { @@ -279,8 +326,8 @@ int profile_grouped_gemm_fixed_nk(int argc, char* argv[]) n_warmup, n_iter); } -#endif -#endif +#endif // CK_ENABLE_INT8 +#endif // CK_ENABLE_BF16 else { throw std::runtime_error("wrong! this GEMM data_type & layout is not implemented");