Skip to content

GitLab

Commit 1c1da090 authored by rtmadduri's avatar rtmadduri
Browse files

fixing sequence length

parent a6f99d8f
Hide whitespace changes
Inline Side-by-side
Showing with 199 additions and 200 deletions
example/15_grouped_gemm/grouped_gemm_xdl_splitk_fp16.cpp
View file @ 1c1da090
......@@ -54,7 +54,7 @@ using DeviceGemmInstance = ck::tensor_operation::device::DeviceGroupedGemmXdlSpl
//######| | | | | Type| Type| Type| DataType| Type| Type| Elementwise| Elementwise| Elementwise| Spacialization| 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| | | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| _NBlock_NWaveNPerXdl| _NWaveNPerXdl|
//######| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
< ALayout, BLayout, DsLayout, ELayout, ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementOp, BElementOp, CDEElementOp, GemmDefault, 256, 256, 128, 32, 8, 8, 32, 32, 4, 2, S<1, 4, 64, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, S<1, 4, 64, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8>;
< ALayout, BLayout, DsLayout, ELayout, ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementOp, BElementOp, CDEElementOp, GemmDefault, 256, 256, 128, 32, 8, 8, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, 1, 1, S<32, 1, 8>, 8>;
// clang-format on
#include "run_grouped_gemm_example.inc"
......
include/ck/tensor_operation/gpu/device/impl/device_grouped_gemm_xdl_splitk_cshuffle.hpp
View file @ 1c1da090
......@@ -200,9 +200,7 @@ struct DeviceGroupedGemmXdlSplitKCShuffle : public DeviceGroupedGemmSplitK<ALayo
ComputeTypeA,
ComputeTypeB>;
// Block2CTileMap configuration parameter.
using Block2ETileMap = typename GridwiseGemm::Block2CTileMap;
;
using GroupedGemmBlock2ETileMap = OffsettedBlockToCTileMap<Block2ETileMap>;
using KernelArgument = typename GridwiseGemm::Argument;
......@@ -279,14 +277,15 @@ struct DeviceGroupedGemmXdlSplitKCShuffle : public DeviceGroupedGemmSplitK<ALayo
const index_t stride_b = gemm_descs[i].stride_B_;
const index_t stride_c = gemm_descs[i].stride_C_;
const index_t m_padded = GridwiseGemm::CalculateMPadded(M);
const index_t n_padded = GridwiseGemm::CalculateNPadded(N);
// const index_t m_padded = GridwiseGemm::CalculateMPadded(M);
// const index_t n_padded = GridwiseGemm::CalculateNPadded(N);
const auto c_grid_desc_m_n =
GridwiseGemm::MakeCGridDescriptor_M_N(M, m_padded, N, n_padded, stride_c);
index_t gdx, gdy, gdz;
std::tie(gdx, gdy, gdz) = GridwiseGemm::CalculateGridSize(M, N, K_BATCH);
const auto local_b2c_tile_map = Block2ETileMap{c_grid_desc_m_n, B2E_M01, K_BATCH};
const index_t grid_size_grp = local_b2c_tile_map.CalculateGridSize(c_grid_desc_m_n);
const auto local_b2c_tile_map = Block2ETileMap{gdx, gdy, gdz};
// const index_t grid_size_grp = local_b2c_tile_map.CalculateGridSize(c_grid_desc_m_n);
const index_t grid_size_grp = gdx * gdy * gdz;
const index_t block_start = grid_size_;
const index_t block_end = grid_size_ + grid_size_grp;
......@@ -328,21 +327,21 @@ struct DeviceGroupedGemmXdlSplitKCShuffle : public DeviceGroupedGemmSplitK<ALayo
auto& karg = gemm_kernel_args_[i].karg_;
const index_t m_padded = GridwiseGemm::CalculateMPadded(karg.M);
const index_t n_padded = GridwiseGemm::CalculateNPadded(karg.N);
// const index_t m_padded = GridwiseGemm::CalculateMPadded(karg.M);
// const index_t n_padded = GridwiseGemm::CalculateNPadded(karg.N);
const auto c_grid_desc_m_n = GridwiseGemm::MakeCGridDescriptor_M_N(
karg.M, karg.N, m_padded, n_padded, karg.StrideC);
index_t gdx, gdy, gdz;
std::tie(gdx, gdy, gdz) = GridwiseGemm::CalculateGridSize(karg.M, karg.N, karg.KBatch);
const auto local_b2c_tile_map = Block2ETileMap{c_grid_desc_m_n, B2E_M01, K_BATCH};
const index_t grid_size_grp = local_b2c_tile_map.CalculateGridSize(c_grid_desc_m_n);
const auto local_b2c_tile_map = Block2ETileMap{gdx, gdy, gdz};
// const index_t grid_size_grp = local_b2c_tile_map.CalculateGridSize(c_grid_desc_m_n);
const index_t grid_size_grp = gdx * gdy * gdz;
const index_t block_start = grid_size_;
const index_t block_end = grid_size_ + grid_size_grp;
grid_size_ += grid_size_grp;
// block-to-e-tile map
auto grouped_block_2_ctile_map =
GroupedGemmBlock2ETileMap(local_b2c_tile_map, block_start);
......@@ -440,9 +439,9 @@ struct DeviceGroupedGemmXdlSplitKCShuffle : public DeviceGroupedGemmSplitK<ALayo
0,
cast_pointer_to_constant_address_space(arg.p_workspace_),
arg.gemm_kernel_args_.size(),
AElementwiseOperation,
BElementwiseOperation,
CDEElementwiseOperation);
PassThrough{},
PassThrough{},
PassThrough{});
};
constexpr index_t minimum_occupancy =
......
library/include/ck/library/tensor_operation_instance/gpu/grouped_gemm/device_grouped_gemm_xdl_splitk_instance.hpp
View file @ 1c1da090
......@@ -48,26 +48,26 @@ using device_grouped_gemm_xdl_splitk_2Bt_rrr_instances = std::tuple<
//################################| Layout| Layout| Layout| Layout| Type| Type| Type| DataType| Type| Type| Elementwise| Elementwise| Elementwise| Spacialization| 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| Version | Scheduler |
//################################| | | | | | | | | | | Operation| Operation| Operation| | | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| _NBlock_NWaveNPerXdl| _NWaveNPerXdl| | |
//################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, T, T, F32, T, Empty_Tuple, T, PassThrough, PassThrough, PassThrough, GemmSpec, 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, Pipeline, Scheduler>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, T, T, F32, T, Empty_Tuple, T, PassThrough, PassThrough, PassThrough, GemmSpec, 256, 128, 256, 32, 8, 8, 32, 32, 2, 4, S<1, 4, 64, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, S<1, 4, 64, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 2, 4, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, Pipeline, Scheduler>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, T, T, F32, T, Empty_Tuple, T, PassThrough, PassThrough, PassThrough, GemmSpec, 256, 192, 64, 32, 8, 8, 32, 32, 3, 1, S<1, 4, 64, 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, 8>, 8, Pipeline, Scheduler>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, T, T, F32, T, Empty_Tuple, T, PassThrough, PassThrough, PassThrough, GemmSpec, 256, 64, 192, 32, 8, 8, 32, 32, 1, 3, S<1, 4, 64, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, S<1, 4, 48, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 2, 2, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, Pipeline, Scheduler>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, T, T, F32, T, Empty_Tuple, T, PassThrough, PassThrough, PassThrough, GemmSpec, 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, 2, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, Pipeline, Scheduler>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, T, T, F32, T, Empty_Tuple, T, PassThrough, PassThrough, PassThrough, GemmSpec, 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, Pipeline, Scheduler>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, T, T, F32, T, Empty_Tuple, T, PassThrough, PassThrough, PassThrough, GemmSpec, 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, Pipeline, Scheduler>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, T, T, F32, T, Empty_Tuple, T, PassThrough, PassThrough, PassThrough, GemmSpec, 128, 128, 128, 32, 8, 8, 32, 32, 4, 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, Pipeline, Scheduler>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, T, T, F32, T, Empty_Tuple, T, PassThrough, PassThrough, PassThrough, GemmSpec, 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, Pipeline, Scheduler>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, T, T, F32, T, Empty_Tuple, T, PassThrough, PassThrough, PassThrough, GemmSpec, 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, Pipeline, Scheduler>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, T, T, F32, T, Empty_Tuple, T, PassThrough, PassThrough, PassThrough, GemmSpec, 128, 32, 192, 32, 8, 8, 32, 32, 1, 3, S<1, 4, 32, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, S<1, 4, 24, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 8>, 8, Pipeline, Scheduler>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, T, T, F32, T, Empty_Tuple, T, PassThrough, PassThrough, PassThrough, GemmSpec, 128, 192, 32, 32, 8, 8, 32, 32, 3, 1, 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, 1, 8, 1, 1, 1, S<1, 32, 1, 4>, 8, Pipeline, Scheduler>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, T, T, F32, T, Empty_Tuple, T, PassThrough, PassThrough, PassThrough, GemmSpec, 128, 32, 64, 32, 8, 8, 32, 32, 1, 1, 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, Pipeline, Scheduler>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, T, T, F32, T, Empty_Tuple, T, PassThrough, PassThrough, PassThrough, GemmSpec, 128, 64, 32, 32, 8, 8, 32, 32, 1, 1, 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, 1, 8, 1, 1, 1, S<1, 32, 1, 4>, 8, Pipeline, Scheduler>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, T, T, F32, T, Empty_Tuple, T, PassThrough, PassThrough, PassThrough, GemmSpec, 128, 32, 128, 32, 8, 8, 32, 32, 1, 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, Pipeline, Scheduler>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, T, T, F32, T, Empty_Tuple, T, PassThrough, PassThrough, PassThrough, GemmSpec, 128, 128, 32, 32, 8, 8, 32, 32, 2, 1, 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, 1, 8, 1, 1, 1, S<1, 32, 1, 4>, 8, Pipeline, Scheduler>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, T, T, F32, T, Empty_Tuple, T, PassThrough, PassThrough, PassThrough, GemmSpec, 128, 64, 64, 32, 8, 8, 32, 32, 2, 1, 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, 16, 1, 8>, 8, Pipeline, Scheduler>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, T, T, F32, T, Empty_Tuple, T, PassThrough, PassThrough, PassThrough, GemmSpec, 64, 64, 64, 32, 8, 8, 32, 32, 2, 2, S<1, 4, 16, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, S<1, 4, 16, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 2, 2, 8, 1, 1, 1, S<1, 16, 1, 4>, 8, Pipeline, Scheduler>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, T, T, F32, T, Empty_Tuple, T, PassThrough, PassThrough, PassThrough, GemmSpec, 64, 64, 32, 32, 8, 8, 32, 32, 2, 1, S<1, 4, 16, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, S<1, 4, 16, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 2, 2, 8, 1, 1, 1, S<1, 16, 1, 4>, 8, Pipeline, Scheduler>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, T, T, F32, T, Empty_Tuple, T, PassThrough, PassThrough, PassThrough, GemmSpec, 64, 32, 64, 32, 8, 8, 32, 32, 1, 2, S<1, 4, 16, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, S<1, 4, 16, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 2, 2, 8, 1, 1, 1, S<1, 16, 1, 4>, 8, Pipeline, Scheduler>
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, T, T, F32, T, Empty_Tuple, T, PassThrough, PassThrough, PassThrough, GemmSpec, 256, 256, 128, 32, 8, 8, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 2, 8, 1, 1, 1, S<32, 1, 8>, 8, Pipeline, Scheduler>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, T, T, F32, T, Empty_Tuple, T, PassThrough, PassThrough, PassThrough, GemmSpec, 256, 128, 256, 32, 8, 8, 32, 32, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 4, 8, 1, 1, 1, S<32, 1, 8>, 8, Pipeline, Scheduler>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, T, T, F32, T, Empty_Tuple, T, PassThrough, PassThrough, PassThrough, GemmSpec, 256, 192, 64, 32, 8, 8, 32, 32, 3, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 2, 8, 1, 1, 1, S<32, 1, 8>, 8, Pipeline, Scheduler>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, T, T, F32, T, Empty_Tuple, T, PassThrough, PassThrough, PassThrough, GemmSpec, 256, 64, 192, 32, 8, 8, 32, 32, 1, 3, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, S<4, 48, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 2, 8, 1, 1, 1, S<32, 1, 8>, 8, Pipeline, Scheduler>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, T, T, F32, T, Empty_Tuple, T, PassThrough, PassThrough, PassThrough, GemmSpec, 256, 128, 128, 32, 8, 8, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 2, 8, 1, 1, 1, S<32, 1, 8>, 8, Pipeline, Scheduler>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, T, T, F32, T, Empty_Tuple, T, PassThrough, PassThrough, PassThrough, GemmSpec, 256, 128, 64, 32, 8, 8, 32, 32, 2, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 1, 8, 1, 1, 1, S<32, 1, 8>, 8, Pipeline, Scheduler>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, T, T, F32, T, Empty_Tuple, T, PassThrough, PassThrough, PassThrough, GemmSpec, 256, 64, 128, 32, 8, 8, 32, 32, 1, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 2, 8, 1, 1, 1, S<32, 1, 8>, 8, Pipeline, Scheduler>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, T, T, F32, T, Empty_Tuple, T, PassThrough, PassThrough, PassThrough, GemmSpec, 128, 128, 128, 32, 8, 8, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 4, 8, 1, 1, 1, S<16, 1, 8>, 8, Pipeline, Scheduler>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, T, T, F32, T, Empty_Tuple, T, PassThrough, PassThrough, PassThrough, GemmSpec, 128, 128, 64, 32, 8, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 2, 8, 1, 1, 1, S<32, 1, 4>, 8, Pipeline, Scheduler>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, T, T, F32, T, Empty_Tuple, T, PassThrough, PassThrough, PassThrough, GemmSpec, 128, 64, 128, 32, 8, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 4, 8, 1, 1, 1, S<16, 1, 8>, 8, Pipeline, Scheduler>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, T, T, F32, T, Empty_Tuple, T, PassThrough, PassThrough, PassThrough, GemmSpec, 128, 32, 192, 32, 8, 8, 32, 32, 1, 3, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, S<4, 24, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 8, 8, 1, 1, 1, S<16, 1, 8>, 8, Pipeline, Scheduler>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, T, T, F32, T, Empty_Tuple, T, PassThrough, PassThrough, PassThrough, GemmSpec, 128, 192, 32, 32, 8, 8, 32, 32, 3, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 1, 8, 1, 1, 1, S<32, 1, 4>, 8, Pipeline, Scheduler>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, T, T, F32, T, Empty_Tuple, T, PassThrough, PassThrough, PassThrough, GemmSpec, 128, 32, 64, 32, 8, 8, 32, 32, 1, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 2, 8, 1, 1, 1, S<32, 1, 4>, 8, Pipeline, Scheduler>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, T, T, F32, T, Empty_Tuple, T, PassThrough, PassThrough, PassThrough, GemmSpec, 128, 64, 32, 32, 8, 8, 32, 32, 1, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 1, 8, 1, 1, 1, S<32, 1, 4>, 8, Pipeline, Scheduler>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, T, T, F32, T, Empty_Tuple, T, PassThrough, PassThrough, PassThrough, GemmSpec, 128, 32, 128, 32, 8, 8, 32, 32, 1, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 4, 8, 1, 1, 1, S<16, 1, 8>, 8, Pipeline, Scheduler>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, T, T, F32, T, Empty_Tuple, T, PassThrough, PassThrough, PassThrough, GemmSpec, 128, 128, 32, 32, 8, 8, 32, 32, 2, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 1, 8, 1, 1, 1, S<32, 1, 4>, 8, Pipeline, Scheduler>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, T, T, F32, T, Empty_Tuple, T, PassThrough, PassThrough, PassThrough, GemmSpec, 128, 64, 64, 32, 8, 8, 32, 32, 2, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 2, 8, 1, 1, 1, S<16, 1, 8>, 8, Pipeline, Scheduler>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, T, T, F32, T, Empty_Tuple, T, PassThrough, PassThrough, PassThrough, GemmSpec, 64, 64, 64, 32, 8, 8, 32, 32, 2, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 2, 8, 1, 1, 1, S<16, 1, 4>, 8, Pipeline, Scheduler>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, T, T, F32, T, Empty_Tuple, T, PassThrough, PassThrough, PassThrough, GemmSpec, 64, 64, 32, 32, 8, 8, 32, 32, 2, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 2, 8, 1, 1, 1, S<16, 1, 4>, 8, Pipeline, Scheduler>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, T, T, F32, T, Empty_Tuple, T, PassThrough, PassThrough, PassThrough, GemmSpec, 64, 32, 64, 32, 8, 8, 32, 32, 1, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 2, 8, 1, 1, 1, S<16, 1, 4>, 8, Pipeline, Scheduler>
// clang-format on
>;
......@@ -78,23 +78,23 @@ template <typename T,
enable_if_t<sizeof(T) == 2, bool> = false>
using device_grouped_gemm_xdl_splitk_2Bt_rcr_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| Pipeline | Loop |
//################################| 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| Version | Scheduler |
//################################| | | | | | | | | | | 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| | |
//################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceGroupedGemmXdlSplitKCShuffle< Row, Col, Empty_Tuple, Row, T, T, F32, T, Empty_Tuple, T, PassThrough, PassThrough, PassThrough, GemmSpec, 256, 256, 128, 32, 8, 8, 32, 32, 4, 2, S<1, 4, 64, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, S<1, 4, 64, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, Pipeline, Scheduler>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Col, Empty_Tuple, Row, T, T, F32, T, Empty_Tuple, T, PassThrough, PassThrough, PassThrough, GemmSpec, 256, 128, 256, 32, 8, 8, 32, 32, 2, 4, S<1, 4, 64, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, S<1, 4, 64, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, Pipeline, Scheduler>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Col, Empty_Tuple, Row, T, T, F32, T, Empty_Tuple, T, PassThrough, PassThrough, PassThrough, GemmSpec, 128, 128, 128, 32, 8, 8, 32, 32, 4, 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, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, 1, 1, S<1, 16, 1, 8>, 8, Pipeline, Scheduler>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Col, Empty_Tuple, Row, T, T, F32, T, Empty_Tuple, T, PassThrough, PassThrough, PassThrough, GemmSpec, 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, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, Pipeline, Scheduler>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Col, Empty_Tuple, Row, T, T, F32, T, Empty_Tuple, T, PassThrough, PassThrough, PassThrough, GemmSpec, 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, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, 1, 1, S<1, 32, 1, 4>, 8, Pipeline, Scheduler>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Col, Empty_Tuple, Row, T, T, F32, T, Empty_Tuple, T, PassThrough, PassThrough, PassThrough, GemmSpec, 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, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, 1, 1, S<1, 16, 1, 8>, 8, Pipeline, Scheduler>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Col, Empty_Tuple, Row, T, T, F32, T, Empty_Tuple, T, PassThrough, PassThrough, PassThrough, GemmSpec, 64, 64, 64, 32, 8, 8, 32, 32, 2, 2, S<1, 4, 16, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, S<1, 4, 16, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, 1, 1, S<1, 16, 1, 4>, 8, Pipeline, Scheduler>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Col, Empty_Tuple, Row, T, T, F32, T, Empty_Tuple, T, PassThrough, PassThrough, PassThrough, GemmSpec, 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, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, Pipeline, Scheduler>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Col, Empty_Tuple, Row, T, T, F32, T, Empty_Tuple, T, PassThrough, PassThrough, PassThrough, GemmSpec, 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, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, Pipeline, Scheduler>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Col, Empty_Tuple, Row, T, T, F32, T, Empty_Tuple, T, PassThrough, PassThrough, PassThrough, GemmSpec, 128, 128, 32, 32, 8, 8, 32, 32, 2, 1, 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, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, 1, 1, S<1, 32, 1, 4>, 8, Pipeline, Scheduler>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Col, Empty_Tuple, Row, T, T, F32, T, Empty_Tuple, T, PassThrough, PassThrough, PassThrough, GemmSpec, 128, 32, 128, 32, 8, 8, 32, 32, 1, 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, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, 1, 1, S<1, 16, 1, 8>, 8, Pipeline, Scheduler>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Col, Empty_Tuple, Row, T, T, F32, T, Empty_Tuple, T, PassThrough, PassThrough, PassThrough, GemmSpec, 64, 64, 32, 32, 8, 8, 32, 32, 2, 1, S<1, 4, 16, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, S<1, 4, 16, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, 1, 1, S<1, 16, 1, 4>, 8, Pipeline, Scheduler>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Col, Empty_Tuple, Row, T, T, F32, T, Empty_Tuple, T, PassThrough, PassThrough, PassThrough, GemmSpec, 64, 32, 64, 32, 8, 8, 32, 32, 1, 2, S<1, 4, 16, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, S<1, 4, 16, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, 1, 1, S<1, 16, 1, 4>, 8, Pipeline, Scheduler>
//################################| A| B| Ds| E| AData| BData| AccData| CShuffle| DsData| EData| A| B| C| GEMM| 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| Pipeline | Loop |
//################################| Layout| Layout| Layout| Layout| Type| Type| Type| DataType| Type| Type| Elementwise| Elementwise| Elementwise| Spacialization| 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| Version | Scheduler |
//################################| | | | | | | | | | | Operation| Operation| Operation| | | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| _NBlock_NWaveNPerXdl| _NWaveNPerXdl| | |
//################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceGroupedGemmXdlSplitKCShuffle< Row, Col, Empty_Tuple, Row, T, T, F32, T, Empty_Tuple, T, PassThrough, PassThrough, PassThrough, GemmSpec, 256, 256, 128, 32, 8, 8, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, 1, 1, S<32, 1, 8>, 8, Pipeline, Scheduler>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Col, Empty_Tuple, Row, T, T, F32, T, Empty_Tuple, T, PassThrough, PassThrough, PassThrough, GemmSpec, 256, 128, 256, 32, 8, 8, 32, 32, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, 1, 1, S<32, 1, 8>, 8, Pipeline, Scheduler>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Col, Empty_Tuple, Row, T, T, F32, T, Empty_Tuple, T, PassThrough, PassThrough, PassThrough, GemmSpec, 128, 128, 128, 32, 8, 8, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, 1, 1, S<16, 1, 8>, 8, Pipeline, Scheduler>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Col, Empty_Tuple, Row, T, T, F32, T, Empty_Tuple, T, PassThrough, PassThrough, PassThrough, GemmSpec, 256, 128, 128, 32, 8, 8, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, 1, 1, S<32, 1, 8>, 8, Pipeline, Scheduler>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Col, Empty_Tuple, Row, T, T, F32, T, Empty_Tuple, T, PassThrough, PassThrough, PassThrough, GemmSpec, 128, 128, 64, 32, 8, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, 1, 1, S<32, 1, 4>, 8, Pipeline, Scheduler>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Col, Empty_Tuple, Row, T, T, F32, T, Empty_Tuple, T, PassThrough, PassThrough, PassThrough, GemmSpec, 128, 64, 128, 32, 8, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, 1, 1, S<16, 1, 8>, 8, Pipeline, Scheduler>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Col, Empty_Tuple, Row, T, T, F32, T, Empty_Tuple, T, PassThrough, PassThrough, PassThrough, GemmSpec, 64, 64, 64, 32, 8, 8, 32, 32, 2, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, 1, 1, S<16, 1, 4>, 8, Pipeline, Scheduler>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Col, Empty_Tuple, Row, T, T, F32, T, Empty_Tuple, T, PassThrough, PassThrough, PassThrough, GemmSpec, 256, 128, 64, 32, 8, 8, 32, 32, 2, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, 1, 1, S<32, 1, 8>, 8, Pipeline, Scheduler>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Col, Empty_Tuple, Row, T, T, F32, T, Empty_Tuple, T, PassThrough, PassThrough, PassThrough, GemmSpec, 256, 64, 128, 32, 8, 8, 32, 32, 1, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, 1, 1, S<32, 1, 8>, 8, Pipeline, Scheduler>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Col, Empty_Tuple, Row, T, T, F32, T, Empty_Tuple, T, PassThrough, PassThrough, PassThrough, GemmSpec, 128, 128, 32, 32, 8, 8, 32, 32, 2, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, 1, 1, S<32, 1, 4>, 8, Pipeline, Scheduler>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Col, Empty_Tuple, Row, T, T, F32, T, Empty_Tuple, T, PassThrough, PassThrough, PassThrough, GemmSpec, 128, 32, 128, 32, 8, 8, 32, 32, 1, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, 1, 1, S<16, 1, 8>, 8, Pipeline, Scheduler>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Col, Empty_Tuple, Row, T, T, F32, T, Empty_Tuple, T, PassThrough, PassThrough, PassThrough, GemmSpec, 64, 64, 32, 32, 8, 8, 32, 32, 2, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, 1, 1, S<16, 1, 4>, 8, Pipeline, Scheduler>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Col, Empty_Tuple, Row, T, T, F32, T, Empty_Tuple, T, PassThrough, PassThrough, PassThrough, GemmSpec, 64, 32, 64, 32, 8, 8, 32, 32, 1, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, 1, 1, S<16, 1, 4>, 8, Pipeline, Scheduler>
// clang-format on
>;
......@@ -105,30 +105,30 @@ template <typename T,
enable_if_t<sizeof(T) == 2, bool> = false>
using device_grouped_gemm_xdl_splitk_2Bt_crr_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| Pipeline | Loop |
//################################| 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| Version | Scheduler |
//################################| | | | | | | | | | | 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| | |
//################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceGroupedGemmXdlSplitKCShuffle< Col, Row, Empty_Tuple, Row, T, T, F32, T, Empty_Tuple, T, PassThrough, PassThrough, PassThrough, GemmSpec, 256, 256, 128, 32, 8, 8, 32, 32, 4, 2, S<1, 4, 64, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 2, 4, 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, Pipeline, Scheduler>,
DeviceGroupedGemmXdlSplitKCShuffle< Col, Row, Empty_Tuple, Row, T, T, F32, T, Empty_Tuple, T, PassThrough, PassThrough, PassThrough, GemmSpec, 256, 128, 256, 32, 8, 8, 32, 32, 2, 4, S<1, 4, 64, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 2, 2, 8, 1, S<1, 4, 64, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 2, 4, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, Pipeline, Scheduler>,
DeviceGroupedGemmXdlSplitKCShuffle< Col, Row, Empty_Tuple, Row, T, T, F32, T, Empty_Tuple, T, PassThrough, PassThrough, PassThrough, GemmSpec, 256, 192, 64, 32, 8, 8, 32, 32, 3, 1, S<1, 4, 48, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 2, 4, 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, 8>, 8, Pipeline, Scheduler>,
DeviceGroupedGemmXdlSplitKCShuffle< Col, Row, Empty_Tuple, Row, T, T, F32, T, Empty_Tuple, T, PassThrough, PassThrough, PassThrough, GemmSpec, 256, 64, 192, 32, 8, 8, 32, 32, 1, 3, S<1, 4, 64, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 2, 1, 8, 1, S<1, 4, 48, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 2, 2, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, Pipeline, Scheduler>,
DeviceGroupedGemmXdlSplitKCShuffle< Col, Row, Empty_Tuple, Row, T, T, F32, T, Empty_Tuple, T, PassThrough, PassThrough, PassThrough, GemmSpec, 256, 128, 128, 32, 8, 8, 32, 32, 2, 2, S<1, 4, 32, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 2, 4, 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, Pipeline, Scheduler>,
DeviceGroupedGemmXdlSplitKCShuffle< Col, Row, Empty_Tuple, Row, T, T, F32, T, Empty_Tuple, T, PassThrough, PassThrough, PassThrough, GemmSpec, 256, 128, 64, 32, 8, 8, 32, 32, 2, 1, S<1, 4, 32, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 2, 4, 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, Pipeline, Scheduler>,
DeviceGroupedGemmXdlSplitKCShuffle< Col, Row, Empty_Tuple, Row, T, T, F32, T, Empty_Tuple, T, PassThrough, PassThrough, PassThrough, GemmSpec, 256, 64, 128, 32, 8, 8, 32, 32, 1, 2, S<1, 4, 64, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 2, 1, 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, Pipeline, Scheduler>,
DeviceGroupedGemmXdlSplitKCShuffle< Col, Row, Empty_Tuple, Row, T, T, F32, T, Empty_Tuple, T, PassThrough, PassThrough, PassThrough, GemmSpec, 128, 128, 128, 32, 8, 8, 32, 32, 4, 2, S<1, 4, 32, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 2, 4, 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, Pipeline, Scheduler>,
DeviceGroupedGemmXdlSplitKCShuffle< Col, Row, Empty_Tuple, Row, T, T, F32, T, Empty_Tuple, T, PassThrough, PassThrough, PassThrough, GemmSpec, 128, 128, 64, 32, 8, 8, 32, 32, 2, 2, S<1, 4, 32, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 2, 4, 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, Pipeline, Scheduler>,
DeviceGroupedGemmXdlSplitKCShuffle< Col, Row, Empty_Tuple, Row, T, T, F32, T, Empty_Tuple, T, PassThrough, PassThrough, PassThrough, GemmSpec, 128, 64, 128, 32, 8, 8, 32, 32, 2, 2, S<1, 4, 32, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 2, 2, 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, Pipeline, Scheduler>,
DeviceGroupedGemmXdlSplitKCShuffle< Col, Row, Empty_Tuple, Row, T, T, F32, T, Empty_Tuple, T, PassThrough, PassThrough, PassThrough, GemmSpec, 128, 32, 192, 32, 8, 8, 32, 32, 1, 3, S<1, 4, 32, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 2, 1, 8, 1, S<1, 4, 24, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 8>, 8, Pipeline, Scheduler>,
DeviceGroupedGemmXdlSplitKCShuffle< Col, Row, Empty_Tuple, Row, T, T, F32, T, Empty_Tuple, T, PassThrough, PassThrough, PassThrough, GemmSpec, 128, 192, 32, 32, 8, 8, 32, 32, 3, 1, S<1, 2, 48, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 2, 4, 8, 1, S<1, 4, 32, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 2, 1, 8, 1, 1, 1, S<1, 32, 1, 4>, 8, Pipeline, Scheduler>,
DeviceGroupedGemmXdlSplitKCShuffle< Col, Row, Empty_Tuple, Row, T, T, F32, T, Empty_Tuple, T, PassThrough, PassThrough, PassThrough, GemmSpec, 128, 32, 64, 32, 8, 8, 32, 32, 1, 1, S<1, 4, 32, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 2, 1, 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, Pipeline, Scheduler>,
DeviceGroupedGemmXdlSplitKCShuffle< Col, Row, Empty_Tuple, Row, T, T, F32, T, Empty_Tuple, T, PassThrough, PassThrough, PassThrough, GemmSpec, 128, 64, 32, 32, 8, 8, 32, 32, 1, 1, S<1, 4, 32, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 2, 2, 8, 1, S<1, 4, 32, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 2, 1, 8, 1, 1, 1, S<1, 32, 1, 4>, 8, Pipeline, Scheduler>,
DeviceGroupedGemmXdlSplitKCShuffle< Col, Row, Empty_Tuple, Row, T, T, F32, T, Empty_Tuple, T, PassThrough, PassThrough, PassThrough, GemmSpec, 128, 32, 128, 32, 8, 8, 32, 32, 1, 2, S<1, 4, 32, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 2, 1, 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, Pipeline, Scheduler>,
DeviceGroupedGemmXdlSplitKCShuffle< Col, Row, Empty_Tuple, Row, T, T, F32, T, Empty_Tuple, T, PassThrough, PassThrough, PassThrough, GemmSpec, 128, 128, 32, 32, 8, 8, 32, 32, 2, 1, S<1, 4, 32, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 2, 4, 8, 1, S<1, 4, 32, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 2, 1, 8, 1, 1, 1, S<1, 32, 1, 4>, 8, Pipeline, Scheduler>,
DeviceGroupedGemmXdlSplitKCShuffle< Col, Row, Empty_Tuple, Row, T, T, F32, T, Empty_Tuple, T, PassThrough, PassThrough, PassThrough, GemmSpec, 128, 64, 64, 32, 8, 8, 32, 32, 2, 1, S<1, 4, 32, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 2, 2, 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, 16, 1, 8>, 8, Pipeline, Scheduler>,
DeviceGroupedGemmXdlSplitKCShuffle< Col, Row, Empty_Tuple, Row, T, T, F32, T, Empty_Tuple, T, PassThrough, PassThrough, PassThrough, GemmSpec, 64, 64, 64, 32, 8, 8, 32, 32, 2, 2, S<1, 4, 16, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 2, 4, 8, 1, S<1, 4, 16, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 2, 2, 8, 1, 1, 1, S<1, 16, 1, 4>, 8, Pipeline, Scheduler>,
DeviceGroupedGemmXdlSplitKCShuffle< Col, Row, Empty_Tuple, Row, T, T, F32, T, Empty_Tuple, T, PassThrough, PassThrough, PassThrough, GemmSpec, 64, 64, 32, 32, 8, 8, 32, 32, 2, 1, S<1, 4, 16, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 2, 4, 8, 1, S<1, 4, 16, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 2, 2, 8, 1, 1, 1, S<1, 16, 1, 4>, 8, Pipeline, Scheduler>,
DeviceGroupedGemmXdlSplitKCShuffle< Col, Row, Empty_Tuple, Row, T, T, F32, T, Empty_Tuple, T, PassThrough, PassThrough, PassThrough, GemmSpec, 64, 32, 64, 32, 8, 8, 32, 32, 1, 2, S<1, 4, 16, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 2, 2, 8, 1, S<1, 4, 16, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 2, 2, 8, 1, 1, 1, S<1, 16, 1, 4>, 8, Pipeline, Scheduler>
//################################| A| B| Ds| E| AData| BData| AccData| CShuffle| DsData| EData| A| B| C| GEMM| 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| Pipeline | Loop |
//################################| Layout| Layout| Layout| Layout| Type| Type| Type| DataType| Type| Type| Elementwise| Elementwise| Elementwise| Spacialization| 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| Version | Scheduler |
//################################| | | | | | | | | | | Operation| Operation| Operation| | | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| _NBlock_NWaveNPerXdl| _NWaveNPerXdl| | |
//################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceGroupedGemmXdlSplitKCShuffle< Col, Row, Empty_Tuple, Row, T, T, F32, T, Empty_Tuple, T, PassThrough, PassThrough, PassThrough, GemmSpec, 256, 256, 128, 32, 8, 8, 32, 32, 4, 2, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 4, 8, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 2, 8, 1, 1, 1, S<32, 1, 8>, 8, Pipeline, Scheduler>,
DeviceGroupedGemmXdlSplitKCShuffle< Col, Row, Empty_Tuple, Row, T, T, F32, T, Empty_Tuple, T, PassThrough, PassThrough, PassThrough, GemmSpec, 256, 128, 256, 32, 8, 8, 32, 32, 2, 4, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 2, 8, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 4, 8, 1, 1, 1, S<32, 1, 8>, 8, Pipeline, Scheduler>,
DeviceGroupedGemmXdlSplitKCShuffle< Col, Row, Empty_Tuple, Row, T, T, F32, T, Empty_Tuple, T, PassThrough, PassThrough, PassThrough, GemmSpec, 256, 192, 64, 32, 8, 8, 32, 32, 3, 1, S<4, 48, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 4, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 2, 8, 1, 1, 1, S<32, 1, 8>, 8, Pipeline, Scheduler>,
DeviceGroupedGemmXdlSplitKCShuffle< Col, Row, Empty_Tuple, Row, T, T, F32, T, Empty_Tuple, T, PassThrough, PassThrough, PassThrough, GemmSpec, 256, 64, 192, 32, 8, 8, 32, 32, 1, 3, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 1, 8, 1, S<4, 48, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 2, 8, 1, 1, 1, S<32, 1, 8>, 8, Pipeline, Scheduler>,
DeviceGroupedGemmXdlSplitKCShuffle< Col, Row, Empty_Tuple, Row, T, T, F32, T, Empty_Tuple, T, PassThrough, PassThrough, PassThrough, GemmSpec, 256, 128, 128, 32, 8, 8, 32, 32, 2, 2, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 4, 8, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 2, 8, 1, 1, 1, S<32, 1, 8>, 8, Pipeline, Scheduler>,
DeviceGroupedGemmXdlSplitKCShuffle< Col, Row, Empty_Tuple, Row, T, T, F32, T, Empty_Tuple, T, PassThrough, PassThrough, PassThrough, GemmSpec, 256, 128, 64, 32, 8, 8, 32, 32, 2, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 4, 8, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 1, 8, 1, 1, 1, S<32, 1, 8>, 8, Pipeline, Scheduler>,
DeviceGroupedGemmXdlSplitKCShuffle< Col, Row, Empty_Tuple, Row, T, T, F32, T, Empty_Tuple, T, PassThrough, PassThrough, PassThrough, GemmSpec, 256, 64, 128, 32, 8, 8, 32, 32, 1, 2, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 1, 8, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 2, 8, 1, 1, 1, S<32, 1, 8>, 8, Pipeline, Scheduler>,
DeviceGroupedGemmXdlSplitKCShuffle< Col, Row, Empty_Tuple, Row, T, T, F32, T, Empty_Tuple, T, PassThrough, PassThrough, PassThrough, GemmSpec, 128, 128, 128, 32, 8, 8, 32, 32, 4, 2, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 4, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 4, 8, 1, 1, 1, S<16, 1, 8>, 8, Pipeline, Scheduler>,
DeviceGroupedGemmXdlSplitKCShuffle< Col, Row, Empty_Tuple, Row, T, T, F32, T, Empty_Tuple, T, PassThrough, PassThrough, PassThrough, GemmSpec, 128, 128, 64, 32, 8, 8, 32, 32, 2, 2, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 4, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 2, 8, 1, 1, 1, S<32, 1, 4>, 8, Pipeline, Scheduler>,
DeviceGroupedGemmXdlSplitKCShuffle< Col, Row, Empty_Tuple, Row, T, T, F32, T, Empty_Tuple, T, PassThrough, PassThrough, PassThrough, GemmSpec, 128, 64, 128, 32, 8, 8, 32, 32, 2, 2, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 2, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 4, 8, 1, 1, 1, S<16, 1, 8>, 8, Pipeline, Scheduler>,
DeviceGroupedGemmXdlSplitKCShuffle< Col, Row, Empty_Tuple, Row, T, T, F32, T, Empty_Tuple, T, PassThrough, PassThrough, PassThrough, GemmSpec, 128, 32, 192, 32, 8, 8, 32, 32, 1, 3, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 1, 8, 1, S<4, 24, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 8, 8, 1, 1, 1, S<16, 1, 8>, 8, Pipeline, Scheduler>,
DeviceGroupedGemmXdlSplitKCShuffle< Col, Row, Empty_Tuple, Row, T, T, F32, T, Empty_Tuple, T, PassThrough, PassThrough, PassThrough, GemmSpec, 128, 192, 32, 32, 8, 8, 32, 32, 3, 1, S<2, 48, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 4, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 1, 8, 1, 1, 1, S<32, 1, 4>, 8, Pipeline, Scheduler>,
DeviceGroupedGemmXdlSplitKCShuffle< Col, Row, Empty_Tuple, Row, T, T, F32, T, Empty_Tuple, T, PassThrough, PassThrough, PassThrough, GemmSpec, 128, 32, 64, 32, 8, 8, 32, 32, 1, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 1, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 2, 8, 1, 1, 1, S<32, 1, 4>, 8, Pipeline, Scheduler>,
DeviceGroupedGemmXdlSplitKCShuffle< Col, Row, Empty_Tuple, Row, T, T, F32, T, Empty_Tuple, T, PassThrough, PassThrough, PassThrough, GemmSpec, 128, 64, 32, 32, 8, 8, 32, 32, 1, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 2, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 1, 8, 1, 1, 1, S<32, 1, 4>, 8, Pipeline, Scheduler>,
DeviceGroupedGemmXdlSplitKCShuffle< Col, Row, Empty_Tuple, Row, T, T, F32, T, Empty_Tuple, T, PassThrough, PassThrough, PassThrough, GemmSpec, 128, 32, 128, 32, 8, 8, 32, 32, 1, 2, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 1, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 4, 8, 1, 1, 1, S<16, 1, 8>, 8, Pipeline, Scheduler>,
DeviceGroupedGemmXdlSplitKCShuffle< Col, Row, Empty_Tuple, Row, T, T, F32, T, Empty_Tuple, T, PassThrough, PassThrough, PassThrough, GemmSpec, 128, 128, 32, 32, 8, 8, 32, 32, 2, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 4, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 1, 8, 1, 1, 1, S<32, 1, 4>, 8, Pipeline, Scheduler>,
DeviceGroupedGemmXdlSplitKCShuffle< Col, Row, Empty_Tuple, Row, T, T, F32, T, Empty_Tuple, T, PassThrough, PassThrough, PassThrough, GemmSpec, 128, 64, 64, 32, 8, 8, 32, 32, 2, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 2, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 2, 8, 1, 1, 1, S<16, 1, 8>, 8, Pipeline, Scheduler>,
DeviceGroupedGemmXdlSplitKCShuffle< Col, Row, Empty_Tuple, Row, T, T, F32, T, Empty_Tuple, T, PassThrough, PassThrough, PassThrough, GemmSpec, 64, 64, 64, 32, 8, 8, 32, 32, 2, 2, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 4, 8, 1, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 2, 8, 1, 1, 1, S<16, 1, 4>, 8, Pipeline, Scheduler>,
DeviceGroupedGemmXdlSplitKCShuffle< Col, Row, Empty_Tuple, Row, T, T, F32, T, Empty_Tuple, T, PassThrough, PassThrough, PassThrough, GemmSpec, 64, 64, 32, 32, 8, 8, 32, 32, 2, 1, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 4, 8, 1, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 2, 8, 1, 1, 1, S<16, 1, 4>, 8, Pipeline, Scheduler>,
DeviceGroupedGemmXdlSplitKCShuffle< Col, Row, Empty_Tuple, Row, T, T, F32, T, Empty_Tuple, T, PassThrough, PassThrough, PassThrough, GemmSpec, 64, 32, 64, 32, 8, 8, 32, 32, 1, 2, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 2, 8, 1, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 2, 8, 1, 1, 1, S<16, 1, 4>, 8, Pipeline, Scheduler>
// clang-format on
>;
......
library/src/tensor_operation_instance/gpu/grouped_gemm/device_grouped_gemm_xdl_splitk_f16_f8_f16_mk_kn_mn_irregular_instance.cpp
View file @ 1c1da090
......@@ -36,68 +36,68 @@ using device_grouped_gemm_xdl_splitk_f16_f8_f16_mk_kn_mn_irregular_tile_instance
//################################| Layout| Layout| Layout| Layout| Type| Type| Type| DataType| Type| Type| Elementwise| Elementwise| Elementwise| Spacialization| 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| | | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| _NBlock_NWaveNPerXdl| _NWaveNPerXdl|
//################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F16, F8, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 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, BlockGemmPipelineVersion::v1>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F16, F8, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 256, 128, 256, 32, 8, 8, 32, 32, 2, 4, S<1, 4, 64, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, S<1, 4, 64, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 2, 4, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, BlockGemmPipelineVersion::v1>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F16, F8, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 256, 192, 64, 32, 8, 8, 32, 32, 3, 1, S<1, 4, 64, 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, 8>, 8, BlockGemmPipelineVersion::v1>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F16, F8, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 256, 64, 192, 32, 8, 8, 32, 32, 1, 3, S<1, 4, 64, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, S<1, 4, 48, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 2, 2, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, BlockGemmPipelineVersion::v1>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F16, F8, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 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, 2, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, BlockGemmPipelineVersion::v1>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F16, F8, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 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, BlockGemmPipelineVersion::v1>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F16, F8, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 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, BlockGemmPipelineVersion::v1>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F16, F8, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 128, 128, 128, 32, 8, 8, 32, 32, 4, 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, BlockGemmPipelineVersion::v1>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F16, F8, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 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, BlockGemmPipelineVersion::v1>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F16, F8, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 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, BlockGemmPipelineVersion::v1>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F16, F8, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 128, 32, 192, 32, 8, 8, 32, 32, 1, 3, S<1, 4, 32, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, S<1, 4, 24, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 8>, 8, BlockGemmPipelineVersion::v1>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F16, F8, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 128, 192, 32, 32, 8, 8, 32, 32, 3, 1, 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, 1, 8, 1, 1, 1, S<1, 32, 1, 4>, 8, BlockGemmPipelineVersion::v1>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F16, F8, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 128, 32, 64, 32, 8, 8, 32, 32, 1, 1, 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, BlockGemmPipelineVersion::v1>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F16, F8, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 128, 64, 32, 32, 8, 8, 32, 32, 1, 1, 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, 1, 8, 1, 1, 1, S<1, 32, 1, 4>, 8, BlockGemmPipelineVersion::v1>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F16, F8, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 128, 32, 128, 32, 8, 8, 32, 32, 1, 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, BlockGemmPipelineVersion::v1>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F16, F8, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 128, 128, 32, 32, 8, 8, 32, 32, 2, 1, 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, 1, 8, 1, 1, 1, S<1, 32, 1, 4>, 8, BlockGemmPipelineVersion::v1>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F16, F8, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 128, 64, 64, 32, 8, 8, 32, 32, 2, 1, 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, 16, 1, 8>, 8, BlockGemmPipelineVersion::v1>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F16, F8, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 64, 64, 64, 32, 8, 8, 32, 32, 2, 2, S<1, 4, 16, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, S<1, 4, 16, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 2, 2, 8, 1, 1, 1, S<1, 16, 1, 4>, 8, BlockGemmPipelineVersion::v1>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F16, F8, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 64, 64, 32, 32, 8, 8, 32, 32, 2, 1, S<1, 4, 16, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, S<1, 4, 16, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 2, 2, 8, 1, 1, 1, S<1, 16, 1, 4>, 8, BlockGemmPipelineVersion::v1>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F16, F8, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 64, 32, 64, 32, 8, 8, 32, 32, 1, 2, S<1, 4, 16, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, S<1, 4, 16, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 2, 2, 8, 1, 1, 1, S<1, 16, 1, 4>, 8, BlockGemmPipelineVersion::v1>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F16, F8, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 256, 256, 128, 32, 8, 8, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 2, 8, 1, 1, 1, S<32, 1, 8>, 8, BlockGemmPipelineVersion::v1>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F16, F8, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 256, 128, 256, 32, 8, 8, 32, 32, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 4, 8, 1, 1, 1, S<32, 1, 8>, 8, BlockGemmPipelineVersion::v1>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F16, F8, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 256, 192, 64, 32, 8, 8, 32, 32, 3, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 2, 8, 1, 1, 1, S<32, 1, 8>, 8, BlockGemmPipelineVersion::v1>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F16, F8, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 256, 64, 192, 32, 8, 8, 32, 32, 1, 3, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, S<4, 48, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 2, 8, 1, 1, 1, S<32, 1, 8>, 8, BlockGemmPipelineVersion::v1>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F16, F8, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 256, 128, 128, 32, 8, 8, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 2, 8, 1, 1, 1, S<32, 1, 8>, 8, BlockGemmPipelineVersion::v1>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F16, F8, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 256, 128, 64, 32, 8, 8, 32, 32, 2, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 1, 8, 1, 1, 1, S<32, 1, 8>, 8, BlockGemmPipelineVersion::v1>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F16, F8, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 256, 64, 128, 32, 8, 8, 32, 32, 1, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 2, 8, 1, 1, 1, S<32, 1, 8>, 8, BlockGemmPipelineVersion::v1>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F16, F8, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 128, 128, 128, 32, 8, 8, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 4, 8, 1, 1, 1, S<16, 1, 8>, 8, BlockGemmPipelineVersion::v1>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F16, F8, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 128, 128, 64, 32, 8, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 2, 8, 1, 1, 1, S<32, 1, 4>, 8, BlockGemmPipelineVersion::v1>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F16, F8, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 128, 64, 128, 32, 8, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 4, 8, 1, 1, 1, S<16, 1, 8>, 8, BlockGemmPipelineVersion::v1>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F16, F8, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 128, 32, 192, 32, 8, 8, 32, 32, 1, 3, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, S<4, 24, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 8, 8, 1, 1, 1, S<16, 1, 8>, 8, BlockGemmPipelineVersion::v1>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F16, F8, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 128, 192, 32, 32, 8, 8, 32, 32, 3, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 1, 8, 1, 1, 1, S<32, 1, 4>, 8, BlockGemmPipelineVersion::v1>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F16, F8, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 128, 32, 64, 32, 8, 8, 32, 32, 1, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 2, 8, 1, 1, 1, S<32, 1, 4>, 8, BlockGemmPipelineVersion::v1>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F16, F8, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 128, 64, 32, 32, 8, 8, 32, 32, 1, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 1, 8, 1, 1, 1, S<32, 1, 4>, 8, BlockGemmPipelineVersion::v1>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F16, F8, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 128, 32, 128, 32, 8, 8, 32, 32, 1, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 4, 8, 1, 1, 1, S<16, 1, 8>, 8, BlockGemmPipelineVersion::v1>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F16, F8, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 128, 128, 32, 32, 8, 8, 32, 32, 2, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 1, 8, 1, 1, 1, S<32, 1, 4>, 8, BlockGemmPipelineVersion::v1>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F16, F8, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 128, 64, 64, 32, 8, 8, 32, 32, 2, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 2, 8, 1, 1, 1, S<16, 1, 8>, 8, BlockGemmPipelineVersion::v1>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F16, F8, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 64, 64, 64, 32, 8, 8, 32, 32, 2, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 2, 8, 1, 1, 1, S<16, 1, 4>, 8, BlockGemmPipelineVersion::v1>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F16, F8, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 64, 64, 32, 32, 8, 8, 32, 32, 2, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 2, 8, 1, 1, 1, S<16, 1, 4>, 8, BlockGemmPipelineVersion::v1>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F16, F8, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 64, 32, 64, 32, 8, 8, 32, 32, 1, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 2, 8, 1, 1, 1, S<16, 1, 4>, 8, BlockGemmPipelineVersion::v1>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F16, F8, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 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, BlockGemmPipelineVersion::v1, BlockGemmPipelineScheduler::Interwave>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F16, F8, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 256, 128, 256, 32, 8, 8, 32, 32, 2, 4, S<1, 4, 64, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, S<1, 4, 64, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 2, 4, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, BlockGemmPipelineVersion::v1, BlockGemmPipelineScheduler::Interwave>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F16, F8, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 256, 192, 64, 32, 8, 8, 32, 32, 3, 1, S<1, 4, 64, 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, 8>, 8, BlockGemmPipelineVersion::v1, BlockGemmPipelineScheduler::Interwave>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F16, F8, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 256, 64, 192, 32, 8, 8, 32, 32, 1, 3, S<1, 4, 64, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, S<1, 4, 48, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 2, 2, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, BlockGemmPipelineVersion::v1, BlockGemmPipelineScheduler::Interwave>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F16, F8, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 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, 2, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, BlockGemmPipelineVersion::v1, BlockGemmPipelineScheduler::Interwave>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F16, F8, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 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, BlockGemmPipelineVersion::v1, BlockGemmPipelineScheduler::Interwave>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F16, F8, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 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, BlockGemmPipelineVersion::v1, BlockGemmPipelineScheduler::Interwave>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F16, F8, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 128, 128, 128, 32, 8, 8, 32, 32, 4, 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, BlockGemmPipelineVersion::v1, BlockGemmPipelineScheduler::Interwave>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F16, F8, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 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, BlockGemmPipelineVersion::v1, BlockGemmPipelineScheduler::Interwave>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F16, F8, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 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, BlockGemmPipelineVersion::v1, BlockGemmPipelineScheduler::Interwave>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F16, F8, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 128, 32, 192, 32, 8, 8, 32, 32, 1, 3, S<1, 4, 32, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, S<1, 4, 24, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 8>, 8, BlockGemmPipelineVersion::v1, BlockGemmPipelineScheduler::Interwave>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F16, F8, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 128, 192, 32, 32, 8, 8, 32, 32, 3, 1, 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, 1, 8, 1, 1, 1, S<1, 32, 1, 4>, 8, BlockGemmPipelineVersion::v1, BlockGemmPipelineScheduler::Interwave>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F16, F8, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 128, 32, 64, 32, 8, 8, 32, 32, 1, 1, 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, BlockGemmPipelineVersion::v1, BlockGemmPipelineScheduler::Interwave>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F16, F8, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 128, 64, 32, 32, 8, 8, 32, 32, 1, 1, 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, 1, 8, 1, 1, 1, S<1, 32, 1, 4>, 8, BlockGemmPipelineVersion::v1, BlockGemmPipelineScheduler::Interwave>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F16, F8, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 128, 32, 128, 32, 8, 8, 32, 32, 1, 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, BlockGemmPipelineVersion::v1, BlockGemmPipelineScheduler::Interwave>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F16, F8, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 128, 128, 32, 32, 8, 8, 32, 32, 2, 1, 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, 1, 8, 1, 1, 1, S<1, 32, 1, 4>, 8, BlockGemmPipelineVersion::v1, BlockGemmPipelineScheduler::Interwave>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F16, F8, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 128, 64, 64, 32, 8, 8, 32, 32, 2, 1, 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, 16, 1, 8>, 8, BlockGemmPipelineVersion::v1, BlockGemmPipelineScheduler::Interwave>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F16, F8, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 64, 64, 64, 32, 8, 8, 32, 32, 2, 2, S<1, 4, 16, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, S<1, 4, 16, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 2, 2, 8, 1, 1, 1, S<1, 16, 1, 4>, 8, BlockGemmPipelineVersion::v1, BlockGemmPipelineScheduler::Interwave>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F16, F8, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 64, 64, 32, 32, 8, 8, 32, 32, 2, 1, S<1, 4, 16, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, S<1, 4, 16, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 2, 2, 8, 1, 1, 1, S<1, 16, 1, 4>, 8, BlockGemmPipelineVersion::v1, BlockGemmPipelineScheduler::Interwave>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F16, F8, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 64, 32, 64, 32, 8, 8, 32, 32, 1, 2, S<1, 4, 16, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, S<1, 4, 16, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 2, 2, 8, 1, 1, 1, S<1, 16, 1, 4>, 8, BlockGemmPipelineVersion::v1, BlockGemmPipelineScheduler::Interwave>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F16, F8, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 256, 256, 128, 32, 8, 8, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 2, 8, 1, 1, 1, S<32, 1, 8>, 8, BlockGemmPipelineVersion::v1, BlockGemmPipelineScheduler::Interwave>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F16, F8, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 256, 128, 256, 32, 8, 8, 32, 32, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 4, 8, 1, 1, 1, S<32, 1, 8>, 8, BlockGemmPipelineVersion::v1, BlockGemmPipelineScheduler::Interwave>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F16, F8, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 256, 192, 64, 32, 8, 8, 32, 32, 3, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 2, 8, 1, 1, 1, S<32, 1, 8>, 8, BlockGemmPipelineVersion::v1, BlockGemmPipelineScheduler::Interwave>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F16, F8, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 256, 64, 192, 32, 8, 8, 32, 32, 1, 3, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, S<4, 48, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 2, 8, 1, 1, 1, S<32, 1, 8>, 8, BlockGemmPipelineVersion::v1, BlockGemmPipelineScheduler::Interwave>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F16, F8, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 256, 128, 128, 32, 8, 8, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 2, 8, 1, 1, 1, S<32, 1, 8>, 8, BlockGemmPipelineVersion::v1, BlockGemmPipelineScheduler::Interwave>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F16, F8, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 256, 128, 64, 32, 8, 8, 32, 32, 2, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 1, 8, 1, 1, 1, S<32, 1, 8>, 8, BlockGemmPipelineVersion::v1, BlockGemmPipelineScheduler::Interwave>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F16, F8, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 256, 64, 128, 32, 8, 8, 32, 32, 1, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 2, 8, 1, 1, 1, S<32, 1, 8>, 8, BlockGemmPipelineVersion::v1, BlockGemmPipelineScheduler::Interwave>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F16, F8, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 128, 128, 128, 32, 8, 8, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 4, 8, 1, 1, 1, S<16, 1, 8>, 8, BlockGemmPipelineVersion::v1, BlockGemmPipelineScheduler::Interwave>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F16, F8, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 128, 128, 64, 32, 8, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 2, 8, 1, 1, 1, S<32, 1, 4>, 8, BlockGemmPipelineVersion::v1, BlockGemmPipelineScheduler::Interwave>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F16, F8, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 128, 64, 128, 32, 8, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 4, 8, 1, 1, 1, S<16, 1, 8>, 8, BlockGemmPipelineVersion::v1, BlockGemmPipelineScheduler::Interwave>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F16, F8, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 128, 32, 192, 32, 8, 8, 32, 32, 1, 3, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, S<4, 24, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 8, 8, 1, 1, 1, S<16, 1, 8>, 8, BlockGemmPipelineVersion::v1, BlockGemmPipelineScheduler::Interwave>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F16, F8, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 128, 192, 32, 32, 8, 8, 32, 32, 3, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 1, 8, 1, 1, 1, S<32, 1, 4>, 8, BlockGemmPipelineVersion::v1, BlockGemmPipelineScheduler::Interwave>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F16, F8, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 128, 32, 64, 32, 8, 8, 32, 32, 1, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 2, 8, 1, 1, 1, S<32, 1, 4>, 8, BlockGemmPipelineVersion::v1, BlockGemmPipelineScheduler::Interwave>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F16, F8, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 128, 64, 32, 32, 8, 8, 32, 32, 1, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 1, 8, 1, 1, 1, S<32, 1, 4>, 8, BlockGemmPipelineVersion::v1, BlockGemmPipelineScheduler::Interwave>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F16, F8, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 128, 32, 128, 32, 8, 8, 32, 32, 1, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 4, 8, 1, 1, 1, S<16, 1, 8>, 8, BlockGemmPipelineVersion::v1, BlockGemmPipelineScheduler::Interwave>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F16, F8, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 128, 128, 32, 32, 8, 8, 32, 32, 2, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 1, 8, 1, 1, 1, S<32, 1, 4>, 8, BlockGemmPipelineVersion::v1, BlockGemmPipelineScheduler::Interwave>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F16, F8, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 128, 64, 64, 32, 8, 8, 32, 32, 2, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 2, 8, 1, 1, 1, S<16, 1, 8>, 8, BlockGemmPipelineVersion::v1, BlockGemmPipelineScheduler::Interwave>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F16, F8, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 64, 64, 64, 32, 8, 8, 32, 32, 2, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 2, 8, 1, 1, 1, S<16, 1, 4>, 8, BlockGemmPipelineVersion::v1, BlockGemmPipelineScheduler::Interwave>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F16, F8, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 64, 64, 32, 32, 8, 8, 32, 32, 2, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 2, 8, 1, 1, 1, S<16, 1, 4>, 8, BlockGemmPipelineVersion::v1, BlockGemmPipelineScheduler::Interwave>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F16, F8, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 64, 32, 64, 32, 8, 8, 32, 32, 1, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 2, 8, 1, 1, 1, S<16, 1, 4>, 8, BlockGemmPipelineVersion::v1, BlockGemmPipelineScheduler::Interwave>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F16, F8, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 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, BlockGemmPipelineVersion::v2>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F16, F8, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 256, 128, 256, 32, 8, 8, 32, 32, 2, 4, S<1, 4, 64, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, S<1, 4, 64, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 2, 4, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, BlockGemmPipelineVersion::v2>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F16, F8, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 256, 192, 64, 32, 8, 8, 32, 32, 3, 1, S<1, 4, 64, 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, 8>, 8, BlockGemmPipelineVersion::v2>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F16, F8, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 256, 64, 192, 32, 8, 8, 32, 32, 1, 3, S<1, 4, 64, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, S<1, 4, 48, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 2, 2, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, BlockGemmPipelineVersion::v2>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F16, F8, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 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, 2, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, BlockGemmPipelineVersion::v2>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F16, F8, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 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, BlockGemmPipelineVersion::v2>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F16, F8, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 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, BlockGemmPipelineVersion::v2>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F16, F8, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 128, 128, 128, 32, 8, 8, 32, 32, 4, 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, BlockGemmPipelineVersion::v2>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F16, F8, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 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, BlockGemmPipelineVersion::v2>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F16, F8, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 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, BlockGemmPipelineVersion::v2>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F16, F8, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 128, 32, 192, 32, 8, 8, 32, 32, 1, 3, S<1, 4, 32, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, S<1, 4, 24, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 8>, 8, BlockGemmPipelineVersion::v2>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F16, F8, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 128, 192, 32, 32, 8, 8, 32, 32, 3, 1, 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, 1, 8, 1, 1, 1, S<1, 32, 1, 4>, 8, BlockGemmPipelineVersion::v2>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F16, F8, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 128, 32, 64, 32, 8, 8, 32, 32, 1, 1, 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, BlockGemmPipelineVersion::v2>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F16, F8, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 128, 64, 32, 32, 8, 8, 32, 32, 1, 1, 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, 1, 8, 1, 1, 1, S<1, 32, 1, 4>, 8, BlockGemmPipelineVersion::v2>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F16, F8, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 128, 32, 128, 32, 8, 8, 32, 32, 1, 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, BlockGemmPipelineVersion::v2>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F16, F8, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 128, 128, 32, 32, 8, 8, 32, 32, 2, 1, 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, 1, 8, 1, 1, 1, S<1, 32, 1, 4>, 8, BlockGemmPipelineVersion::v2>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F16, F8, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 128, 64, 64, 32, 8, 8, 32, 32, 2, 1, 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, 16, 1, 8>, 8, BlockGemmPipelineVersion::v2>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F16, F8, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 64, 64, 64, 32, 8, 8, 32, 32, 2, 2, S<1, 4, 16, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, S<1, 4, 16, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 2, 2, 8, 1, 1, 1, S<1, 16, 1, 4>, 8, BlockGemmPipelineVersion::v2>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F16, F8, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 64, 64, 32, 32, 8, 8, 32, 32, 2, 1, S<1, 4, 16, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, S<1, 4, 16, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 2, 2, 8, 1, 1, 1, S<1, 16, 1, 4>, 8, BlockGemmPipelineVersion::v2>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F16, F8, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 64, 32, 64, 32, 8, 8, 32, 32, 1, 2, S<1, 4, 16, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, S<1, 4, 16, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 2, 2, 8, 1, 1, 1, S<1, 16, 1, 4>, 8, BlockGemmPipelineVersion::v2>
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F16, F8, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 256, 256, 128, 32, 8, 8, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 2, 8, 1, 1, 1, S<32, 1, 8>, 8, BlockGemmPipelineVersion::v2>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F16, F8, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 256, 128, 256, 32, 8, 8, 32, 32, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 4, 8, 1, 1, 1, S<32, 1, 8>, 8, BlockGemmPipelineVersion::v2>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F16, F8, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 256, 192, 64, 32, 8, 8, 32, 32, 3, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 2, 8, 1, 1, 1, S<32, 1, 8>, 8, BlockGemmPipelineVersion::v2>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F16, F8, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 256, 64, 192, 32, 8, 8, 32, 32, 1, 3, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, S<4, 48, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 2, 8, 1, 1, 1, S<32, 1, 8>, 8, BlockGemmPipelineVersion::v2>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F16, F8, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 256, 128, 128, 32, 8, 8, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 2, 8, 1, 1, 1, S<32, 1, 8>, 8, BlockGemmPipelineVersion::v2>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F16, F8, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 256, 128, 64, 32, 8, 8, 32, 32, 2, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 1, 8, 1, 1, 1, S<32, 1, 8>, 8, BlockGemmPipelineVersion::v2>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F16, F8, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 256, 64, 128, 32, 8, 8, 32, 32, 1, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 2, 8, 1, 1, 1, S<32, 1, 8>, 8, BlockGemmPipelineVersion::v2>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F16, F8, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 128, 128, 128, 32, 8, 8, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 4, 8, 1, 1, 1, S<16, 1, 8>, 8, BlockGemmPipelineVersion::v2>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F16, F8, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 128, 128, 64, 32, 8, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 2, 8, 1, 1, 1, S<32, 1, 4>, 8, BlockGemmPipelineVersion::v2>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F16, F8, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 128, 64, 128, 32, 8, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 4, 8, 1, 1, 1, S<16, 1, 8>, 8, BlockGemmPipelineVersion::v2>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F16, F8, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 128, 32, 192, 32, 8, 8, 32, 32, 1, 3, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, S<4, 24, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 8, 8, 1, 1, 1, S<16, 1, 8>, 8, BlockGemmPipelineVersion::v2>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F16, F8, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 128, 192, 32, 32, 8, 8, 32, 32, 3, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 1, 8, 1, 1, 1, S<32, 1, 4>, 8, BlockGemmPipelineVersion::v2>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F16, F8, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 128, 32, 64, 32, 8, 8, 32, 32, 1, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 2, 8, 1, 1, 1, S<32, 1, 4>, 8, BlockGemmPipelineVersion::v2>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F16, F8, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 128, 64, 32, 32, 8, 8, 32, 32, 1, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 1, 8, 1, 1, 1, S<32, 1, 4>, 8, BlockGemmPipelineVersion::v2>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F16, F8, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 128, 32, 128, 32, 8, 8, 32, 32, 1, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 4, 8, 1, 1, 1, S<16, 1, 8>, 8, BlockGemmPipelineVersion::v2>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F16, F8, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 128, 128, 32, 32, 8, 8, 32, 32, 2, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 1, 8, 1, 1, 1, S<32, 1, 4>, 8, BlockGemmPipelineVersion::v2>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F16, F8, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 128, 64, 64, 32, 8, 8, 32, 32, 2, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 2, 8, 1, 1, 1, S<16, 1, 8>, 8, BlockGemmPipelineVersion::v2>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F16, F8, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 64, 64, 64, 32, 8, 8, 32, 32, 2, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 2, 8, 1, 1, 1, S<16, 1, 4>, 8, BlockGemmPipelineVersion::v2>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F16, F8, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 64, 64, 32, 32, 8, 8, 32, 32, 2, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 2, 8, 1, 1, 1, S<16, 1, 4>, 8, BlockGemmPipelineVersion::v2>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F16, F8, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 64, 32, 64, 32, 8, 8, 32, 32, 1, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 2, 8, 1, 1, 1, S<16, 1, 4>, 8, BlockGemmPipelineVersion::v2>
// clang-format on
>;
......
library/src/tensor_operation_instance/gpu/grouped_gemm/device_grouped_gemm_xdl_splitk_f8_f16_f16_mk_kn_mn_irregular_instance.cpp
View file @ 1c1da090
......@@ -36,68 +36,68 @@ using device_grouped_gemm_xdl_splitk_f8_f16_f16_mk_kn_mn_irregular_tile_instance
//################################| Layout| Layout| Layout| Layout| Type| Type| Type| DataType| Type| Type| Elementwise| Elementwise| Elementwise| Spacialization| 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| | | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| _NBlock_NWaveNPerXdl| _NWaveNPerXdl|
//################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F8, F16, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 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, BlockGemmPipelineVersion::v1>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F8, F16, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 256, 128, 256, 32, 8, 8, 32, 32, 2, 4, S<1, 4, 64, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, S<1, 4, 64, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 2, 4, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, BlockGemmPipelineVersion::v1>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F8, F16, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 256, 192, 64, 32, 8, 8, 32, 32, 3, 1, S<1, 4, 64, 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, 8>, 8, BlockGemmPipelineVersion::v1>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F8, F16, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 256, 64, 192, 32, 8, 8, 32, 32, 1, 3, S<1, 4, 64, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, S<1, 4, 48, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 2, 2, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, BlockGemmPipelineVersion::v1>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F8, F16, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 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, 2, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, BlockGemmPipelineVersion::v1>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F8, F16, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 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, BlockGemmPipelineVersion::v1>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F8, F16, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 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, BlockGemmPipelineVersion::v1>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F8, F16, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 128, 128, 128, 32, 8, 8, 32, 32, 4, 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, BlockGemmPipelineVersion::v1>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F8, F16, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 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, BlockGemmPipelineVersion::v1>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F8, F16, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 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, BlockGemmPipelineVersion::v1>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F8, F16, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 128, 32, 192, 32, 8, 8, 32, 32, 1, 3, S<1, 4, 32, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, S<1, 4, 24, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 8>, 8, BlockGemmPipelineVersion::v1>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F8, F16, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 128, 192, 32, 32, 8, 8, 32, 32, 3, 1, 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, 1, 8, 1, 1, 1, S<1, 32, 1, 4>, 8, BlockGemmPipelineVersion::v1>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F8, F16, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 128, 32, 64, 32, 8, 8, 32, 32, 1, 1, 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, BlockGemmPipelineVersion::v1>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F8, F16, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 128, 64, 32, 32, 8, 8, 32, 32, 1, 1, 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, 1, 8, 1, 1, 1, S<1, 32, 1, 4>, 8, BlockGemmPipelineVersion::v1>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F8, F16, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 128, 32, 128, 32, 8, 8, 32, 32, 1, 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, BlockGemmPipelineVersion::v1>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F8, F16, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 128, 128, 32, 32, 8, 8, 32, 32, 2, 1, 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, 1, 8, 1, 1, 1, S<1, 32, 1, 4>, 8, BlockGemmPipelineVersion::v1>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F8, F16, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 128, 64, 64, 32, 8, 8, 32, 32, 2, 1, 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, 16, 1, 8>, 8, BlockGemmPipelineVersion::v1>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F8, F16, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 64, 64, 64, 32, 8, 8, 32, 32, 2, 2, S<1, 4, 16, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, S<1, 4, 16, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 2, 2, 8, 1, 1, 1, S<1, 16, 1, 4>, 8, BlockGemmPipelineVersion::v1>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F8, F16, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 64, 64, 32, 32, 8, 8, 32, 32, 2, 1, S<1, 4, 16, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, S<1, 4, 16, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 2, 2, 8, 1, 1, 1, S<1, 16, 1, 4>, 8, BlockGemmPipelineVersion::v1>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F8, F16, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 64, 32, 64, 32, 8, 8, 32, 32, 1, 2, S<1, 4, 16, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, S<1, 4, 16, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 2, 2, 8, 1, 1, 1, S<1, 16, 1, 4>, 8, BlockGemmPipelineVersion::v1>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F8, F16, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 256, 256, 128, 32, 8, 8, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 2, 8, 1, 1, 1, S<32, 1, 8>, 8, BlockGemmPipelineVersion::v1>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F8, F16, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 256, 128, 256, 32, 8, 8, 32, 32, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 4, 8, 1, 1, 1, S<32, 1, 8>, 8, BlockGemmPipelineVersion::v1>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F8, F16, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 256, 192, 64, 32, 8, 8, 32, 32, 3, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 2, 8, 1, 1, 1, S<32, 1, 8>, 8, BlockGemmPipelineVersion::v1>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F8, F16, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 256, 64, 192, 32, 8, 8, 32, 32, 1, 3, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, S<4, 48, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 2, 8, 1, 1, 1, S<32, 1, 8>, 8, BlockGemmPipelineVersion::v1>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F8, F16, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 256, 128, 128, 32, 8, 8, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 2, 8, 1, 1, 1, S<32, 1, 8>, 8, BlockGemmPipelineVersion::v1>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F8, F16, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 256, 128, 64, 32, 8, 8, 32, 32, 2, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 1, 8, 1, 1, 1, S<32, 1, 8>, 8, BlockGemmPipelineVersion::v1>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F8, F16, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 256, 64, 128, 32, 8, 8, 32, 32, 1, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 2, 8, 1, 1, 1, S<32, 1, 8>, 8, BlockGemmPipelineVersion::v1>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F8, F16, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 128, 128, 128, 32, 8, 8, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 4, 8, 1, 1, 1, S<16, 1, 8>, 8, BlockGemmPipelineVersion::v1>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F8, F16, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 128, 128, 64, 32, 8, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 2, 8, 1, 1, 1, S<32, 1, 4>, 8, BlockGemmPipelineVersion::v1>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F8, F16, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 128, 64, 128, 32, 8, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 4, 8, 1, 1, 1, S<16, 1, 8>, 8, BlockGemmPipelineVersion::v1>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F8, F16, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 128, 32, 192, 32, 8, 8, 32, 32, 1, 3, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, S<4, 24, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 8, 8, 1, 1, 1, S<16, 1, 8>, 8, BlockGemmPipelineVersion::v1>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F8, F16, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 128, 192, 32, 32, 8, 8, 32, 32, 3, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 1, 8, 1, 1, 1, S<32, 1, 4>, 8, BlockGemmPipelineVersion::v1>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F8, F16, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 128, 32, 64, 32, 8, 8, 32, 32, 1, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 2, 8, 1, 1, 1, S<32, 1, 4>, 8, BlockGemmPipelineVersion::v1>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F8, F16, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 128, 64, 32, 32, 8, 8, 32, 32, 1, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 1, 8, 1, 1, 1, S<32, 1, 4>, 8, BlockGemmPipelineVersion::v1>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F8, F16, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 128, 32, 128, 32, 8, 8, 32, 32, 1, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 4, 8, 1, 1, 1, S<16, 1, 8>, 8, BlockGemmPipelineVersion::v1>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F8, F16, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 128, 128, 32, 32, 8, 8, 32, 32, 2, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 1, 8, 1, 1, 1, S<32, 1, 4>, 8, BlockGemmPipelineVersion::v1>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F8, F16, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 128, 64, 64, 32, 8, 8, 32, 32, 2, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 2, 8, 1, 1, 1, S<16, 1, 8>, 8, BlockGemmPipelineVersion::v1>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F8, F16, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 64, 64, 64, 32, 8, 8, 32, 32, 2, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 2, 8, 1, 1, 1, S<16, 1, 4>, 8, BlockGemmPipelineVersion::v1>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F8, F16, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 64, 64, 32, 32, 8, 8, 32, 32, 2, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 2, 8, 1, 1, 1, S<16, 1, 4>, 8, BlockGemmPipelineVersion::v1>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F8, F16, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 64, 32, 64, 32, 8, 8, 32, 32, 1, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 2, 8, 1, 1, 1, S<16, 1, 4>, 8, BlockGemmPipelineVersion::v1>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F8, F16, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 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, BlockGemmPipelineVersion::v1, BlockGemmPipelineScheduler::Interwave>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F8, F16, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 256, 128, 256, 32, 8, 8, 32, 32, 2, 4, S<1, 4, 64, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, S<1, 4, 64, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 2, 4, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, BlockGemmPipelineVersion::v1, BlockGemmPipelineScheduler::Interwave>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F8, F16, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 256, 192, 64, 32, 8, 8, 32, 32, 3, 1, S<1, 4, 64, 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, 8>, 8, BlockGemmPipelineVersion::v1, BlockGemmPipelineScheduler::Interwave>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F8, F16, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 256, 64, 192, 32, 8, 8, 32, 32, 1, 3, S<1, 4, 64, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, S<1, 4, 48, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 2, 2, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, BlockGemmPipelineVersion::v1, BlockGemmPipelineScheduler::Interwave>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F8, F16, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 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, 2, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, BlockGemmPipelineVersion::v1, BlockGemmPipelineScheduler::Interwave>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F8, F16, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 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, BlockGemmPipelineVersion::v1, BlockGemmPipelineScheduler::Interwave>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F8, F16, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 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, BlockGemmPipelineVersion::v1, BlockGemmPipelineScheduler::Interwave>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F8, F16, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 128, 128, 128, 32, 8, 8, 32, 32, 4, 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, BlockGemmPipelineVersion::v1, BlockGemmPipelineScheduler::Interwave>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F8, F16, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 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, BlockGemmPipelineVersion::v1, BlockGemmPipelineScheduler::Interwave>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F8, F16, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 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, BlockGemmPipelineVersion::v1, BlockGemmPipelineScheduler::Interwave>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F8, F16, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 128, 32, 192, 32, 8, 8, 32, 32, 1, 3, S<1, 4, 32, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, S<1, 4, 24, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 8>, 8, BlockGemmPipelineVersion::v1, BlockGemmPipelineScheduler::Interwave>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F8, F16, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 128, 192, 32, 32, 8, 8, 32, 32, 3, 1, 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, 1, 8, 1, 1, 1, S<1, 32, 1, 4>, 8, BlockGemmPipelineVersion::v1, BlockGemmPipelineScheduler::Interwave>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F8, F16, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 128, 32, 64, 32, 8, 8, 32, 32, 1, 1, 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, BlockGemmPipelineVersion::v1, BlockGemmPipelineScheduler::Interwave>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F8, F16, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 128, 64, 32, 32, 8, 8, 32, 32, 1, 1, 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, 1, 8, 1, 1, 1, S<1, 32, 1, 4>, 8, BlockGemmPipelineVersion::v1, BlockGemmPipelineScheduler::Interwave>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F8, F16, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 128, 32, 128, 32, 8, 8, 32, 32, 1, 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, BlockGemmPipelineVersion::v1, BlockGemmPipelineScheduler::Interwave>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F8, F16, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 128, 128, 32, 32, 8, 8, 32, 32, 2, 1, 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, 1, 8, 1, 1, 1, S<1, 32, 1, 4>, 8, BlockGemmPipelineVersion::v1, BlockGemmPipelineScheduler::Interwave>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F8, F16, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 128, 64, 64, 32, 8, 8, 32, 32, 2, 1, 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, 16, 1, 8>, 8, BlockGemmPipelineVersion::v1, BlockGemmPipelineScheduler::Interwave>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F8, F16, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 64, 64, 64, 32, 8, 8, 32, 32, 2, 2, S<1, 4, 16, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, S<1, 4, 16, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 2, 2, 8, 1, 1, 1, S<1, 16, 1, 4>, 8, BlockGemmPipelineVersion::v1, BlockGemmPipelineScheduler::Interwave>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F8, F16, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 64, 64, 32, 32, 8, 8, 32, 32, 2, 1, S<1, 4, 16, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, S<1, 4, 16, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 2, 2, 8, 1, 1, 1, S<1, 16, 1, 4>, 8, BlockGemmPipelineVersion::v1, BlockGemmPipelineScheduler::Interwave>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F8, F16, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 64, 32, 64, 32, 8, 8, 32, 32, 1, 2, S<1, 4, 16, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, S<1, 4, 16, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 2, 2, 8, 1, 1, 1, S<1, 16, 1, 4>, 8, BlockGemmPipelineVersion::v1, BlockGemmPipelineScheduler::Interwave>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F8, F16, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 256, 256, 128, 32, 8, 8, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 2, 8, 1, 1, 1, S<32, 1, 8>, 8, BlockGemmPipelineVersion::v1, BlockGemmPipelineScheduler::Interwave>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F8, F16, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 256, 128, 256, 32, 8, 8, 32, 32, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 4, 8, 1, 1, 1, S<32, 1, 8>, 8, BlockGemmPipelineVersion::v1, BlockGemmPipelineScheduler::Interwave>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F8, F16, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 256, 192, 64, 32, 8, 8, 32, 32, 3, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 2, 8, 1, 1, 1, S<32, 1, 8>, 8, BlockGemmPipelineVersion::v1, BlockGemmPipelineScheduler::Interwave>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F8, F16, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 256, 64, 192, 32, 8, 8, 32, 32, 1, 3, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, S<4, 48, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 2, 8, 1, 1, 1, S<32, 1, 8>, 8, BlockGemmPipelineVersion::v1, BlockGemmPipelineScheduler::Interwave>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F8, F16, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 256, 128, 128, 32, 8, 8, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 2, 8, 1, 1, 1, S<32, 1, 8>, 8, BlockGemmPipelineVersion::v1, BlockGemmPipelineScheduler::Interwave>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F8, F16, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 256, 128, 64, 32, 8, 8, 32, 32, 2, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 1, 8, 1, 1, 1, S<32, 1, 8>, 8, BlockGemmPipelineVersion::v1, BlockGemmPipelineScheduler::Interwave>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F8, F16, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 256, 64, 128, 32, 8, 8, 32, 32, 1, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 2, 8, 1, 1, 1, S<32, 1, 8>, 8, BlockGemmPipelineVersion::v1, BlockGemmPipelineScheduler::Interwave>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F8, F16, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 128, 128, 128, 32, 8, 8, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 4, 8, 1, 1, 1, S<16, 1, 8>, 8, BlockGemmPipelineVersion::v1, BlockGemmPipelineScheduler::Interwave>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F8, F16, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 128, 128, 64, 32, 8, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 2, 8, 1, 1, 1, S<32, 1, 4>, 8, BlockGemmPipelineVersion::v1, BlockGemmPipelineScheduler::Interwave>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F8, F16, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 128, 64, 128, 32, 8, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 4, 8, 1, 1, 1, S<16, 1, 8>, 8, BlockGemmPipelineVersion::v1, BlockGemmPipelineScheduler::Interwave>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F8, F16, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 128, 32, 192, 32, 8, 8, 32, 32, 1, 3, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, S<4, 24, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 8, 8, 1, 1, 1, S<16, 1, 8>, 8, BlockGemmPipelineVersion::v1, BlockGemmPipelineScheduler::Interwave>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F8, F16, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 128, 192, 32, 32, 8, 8, 32, 32, 3, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 1, 8, 1, 1, 1, S<32, 1, 4>, 8, BlockGemmPipelineVersion::v1, BlockGemmPipelineScheduler::Interwave>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F8, F16, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 128, 32, 64, 32, 8, 8, 32, 32, 1, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 2, 8, 1, 1, 1, S<32, 1, 4>, 8, BlockGemmPipelineVersion::v1, BlockGemmPipelineScheduler::Interwave>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F8, F16, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 128, 64, 32, 32, 8, 8, 32, 32, 1, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 1, 8, 1, 1, 1, S<32, 1, 4>, 8, BlockGemmPipelineVersion::v1, BlockGemmPipelineScheduler::Interwave>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F8, F16, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 128, 32, 128, 32, 8, 8, 32, 32, 1, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 4, 8, 1, 1, 1, S<16, 1, 8>, 8, BlockGemmPipelineVersion::v1, BlockGemmPipelineScheduler::Interwave>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F8, F16, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 128, 128, 32, 32, 8, 8, 32, 32, 2, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 1, 8, 1, 1, 1, S<32, 1, 4>, 8, BlockGemmPipelineVersion::v1, BlockGemmPipelineScheduler::Interwave>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F8, F16, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 128, 64, 64, 32, 8, 8, 32, 32, 2, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 2, 8, 1, 1, 1, S<16, 1, 8>, 8, BlockGemmPipelineVersion::v1, BlockGemmPipelineScheduler::Interwave>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F8, F16, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 64, 64, 64, 32, 8, 8, 32, 32, 2, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 2, 8, 1, 1, 1, S<16, 1, 4>, 8, BlockGemmPipelineVersion::v1, BlockGemmPipelineScheduler::Interwave>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F8, F16, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 64, 64, 32, 32, 8, 8, 32, 32, 2, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 2, 8, 1, 1, 1, S<16, 1, 4>, 8, BlockGemmPipelineVersion::v1, BlockGemmPipelineScheduler::Interwave>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F8, F16, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 64, 32, 64, 32, 8, 8, 32, 32, 1, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 2, 8, 1, 1, 1, S<16, 1, 4>, 8, BlockGemmPipelineVersion::v1, BlockGemmPipelineScheduler::Interwave>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F8, F16, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 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, BlockGemmPipelineVersion::v2>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F8, F16, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 256, 128, 256, 32, 8, 8, 32, 32, 2, 4, S<1, 4, 64, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, S<1, 4, 64, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 2, 4, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, BlockGemmPipelineVersion::v2>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F8, F16, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 256, 192, 64, 32, 8, 8, 32, 32, 3, 1, S<1, 4, 64, 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, 8>, 8, BlockGemmPipelineVersion::v2>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F8, F16, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 256, 64, 192, 32, 8, 8, 32, 32, 1, 3, S<1, 4, 64, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, S<1, 4, 48, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 2, 2, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, BlockGemmPipelineVersion::v2>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F8, F16, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 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, 2, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, BlockGemmPipelineVersion::v2>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F8, F16, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 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, BlockGemmPipelineVersion::v2>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F8, F16, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 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, BlockGemmPipelineVersion::v2>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F8, F16, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 128, 128, 128, 32, 8, 8, 32, 32, 4, 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, BlockGemmPipelineVersion::v2>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F8, F16, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 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, BlockGemmPipelineVersion::v2>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F8, F16, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 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, BlockGemmPipelineVersion::v2>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F8, F16, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 128, 32, 192, 32, 8, 8, 32, 32, 1, 3, S<1, 4, 32, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, S<1, 4, 24, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 8>, 8, BlockGemmPipelineVersion::v2>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F8, F16, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 128, 192, 32, 32, 8, 8, 32, 32, 3, 1, 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, 1, 8, 1, 1, 1, S<1, 32, 1, 4>, 8, BlockGemmPipelineVersion::v2>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F8, F16, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 128, 32, 64, 32, 8, 8, 32, 32, 1, 1, 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, BlockGemmPipelineVersion::v2>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F8, F16, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 128, 64, 32, 32, 8, 8, 32, 32, 1, 1, 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, 1, 8, 1, 1, 1, S<1, 32, 1, 4>, 8, BlockGemmPipelineVersion::v2>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F8, F16, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 128, 32, 128, 32, 8, 8, 32, 32, 1, 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, BlockGemmPipelineVersion::v2>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F8, F16, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 128, 128, 32, 32, 8, 8, 32, 32, 2, 1, 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, 1, 8, 1, 1, 1, S<1, 32, 1, 4>, 8, BlockGemmPipelineVersion::v2>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F8, F16, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 128, 64, 64, 32, 8, 8, 32, 32, 2, 1, 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, 16, 1, 8>, 8, BlockGemmPipelineVersion::v2>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F8, F16, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 64, 64, 64, 32, 8, 8, 32, 32, 2, 2, S<1, 4, 16, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, S<1, 4, 16, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 2, 2, 8, 1, 1, 1, S<1, 16, 1, 4>, 8, BlockGemmPipelineVersion::v2>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F8, F16, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 64, 64, 32, 32, 8, 8, 32, 32, 2, 1, S<1, 4, 16, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, S<1, 4, 16, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 2, 2, 8, 1, 1, 1, S<1, 16, 1, 4>, 8, BlockGemmPipelineVersion::v2>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F8, F16, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 64, 32, 64, 32, 8, 8, 32, 32, 1, 2, S<1, 4, 16, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, S<1, 4, 16, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 2, 2, 8, 1, 1, 1, S<1, 16, 1, 4>, 8, BlockGemmPipelineVersion::v2>
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F8, F16, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 256, 256, 128, 32, 8, 8, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 2, 8, 1, 1, 1, S<32, 1, 8>, 8, BlockGemmPipelineVersion::v2>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F8, F16, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 256, 128, 256, 32, 8, 8, 32, 32, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 4, 8, 1, 1, 1, S<32, 1, 8>, 8, BlockGemmPipelineVersion::v2>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F8, F16, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 256, 192, 64, 32, 8, 8, 32, 32, 3, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 2, 8, 1, 1, 1, S<32, 1, 8>, 8, BlockGemmPipelineVersion::v2>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F8, F16, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 256, 64, 192, 32, 8, 8, 32, 32, 1, 3, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, S<4, 48, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 2, 8, 1, 1, 1, S<32, 1, 8>, 8, BlockGemmPipelineVersion::v2>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F8, F16, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 256, 128, 128, 32, 8, 8, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 2, 8, 1, 1, 1, S<32, 1, 8>, 8, BlockGemmPipelineVersion::v2>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F8, F16, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 256, 128, 64, 32, 8, 8, 32, 32, 2, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 1, 8, 1, 1, 1, S<32, 1, 8>, 8, BlockGemmPipelineVersion::v2>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F8, F16, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 256, 64, 128, 32, 8, 8, 32, 32, 1, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 2, 8, 1, 1, 1, S<32, 1, 8>, 8, BlockGemmPipelineVersion::v2>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F8, F16, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 128, 128, 128, 32, 8, 8, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 4, 8, 1, 1, 1, S<16, 1, 8>, 8, BlockGemmPipelineVersion::v2>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F8, F16, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 128, 128, 64, 32, 8, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 2, 8, 1, 1, 1, S<32, 1, 4>, 8, BlockGemmPipelineVersion::v2>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F8, F16, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 128, 64, 128, 32, 8, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 4, 8, 1, 1, 1, S<16, 1, 8>, 8, BlockGemmPipelineVersion::v2>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F8, F16, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 128, 32, 192, 32, 8, 8, 32, 32, 1, 3, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, S<4, 24, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 8, 8, 1, 1, 1, S<16, 1, 8>, 8, BlockGemmPipelineVersion::v2>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F8, F16, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 128, 192, 32, 32, 8, 8, 32, 32, 3, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 1, 8, 1, 1, 1, S<32, 1, 4>, 8, BlockGemmPipelineVersion::v2>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F8, F16, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 128, 32, 64, 32, 8, 8, 32, 32, 1, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 2, 8, 1, 1, 1, S<32, 1, 4>, 8, BlockGemmPipelineVersion::v2>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F8, F16, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 128, 64, 32, 32, 8, 8, 32, 32, 1, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 1, 8, 1, 1, 1, S<32, 1, 4>, 8, BlockGemmPipelineVersion::v2>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F8, F16, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 128, 32, 128, 32, 8, 8, 32, 32, 1, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 4, 8, 1, 1, 1, S<16, 1, 8>, 8, BlockGemmPipelineVersion::v2>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F8, F16, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 128, 128, 32, 32, 8, 8, 32, 32, 2, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 1, 8, 1, 1, 1, S<32, 1, 4>, 8, BlockGemmPipelineVersion::v2>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F8, F16, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 128, 64, 64, 32, 8, 8, 32, 32, 2, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 2, 8, 1, 1, 1, S<16, 1, 8>, 8, BlockGemmPipelineVersion::v2>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F8, F16, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 64, 64, 64, 32, 8, 8, 32, 32, 2, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 2, 8, 1, 1, 1, S<16, 1, 4>, 8, BlockGemmPipelineVersion::v2>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F8, F16, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 64, 64, 32, 32, 8, 8, 32, 32, 2, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 2, 8, 1, 1, 1, S<16, 1, 4>, 8, BlockGemmPipelineVersion::v2>,
DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F8, F16, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 64, 32, 64, 32, 8, 8, 32, 32, 1, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 3, 8, 8, 1, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 2, 2, 8, 1, 1, 1, S<16, 1, 4>, 8, BlockGemmPipelineVersion::v2>
// clang-format on
>;
......
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