From 714fc59909e26290de9a337bf89e148c4e454964 Mon Sep 17 00:00:00 2001 From: ltqin Date: Tue, 25 Jul 2023 09:08:15 +0800 Subject: [PATCH] Add bias scalar vectorload = 1 for gemm bias gemm (#791) * first change bias load * add bias dim and scalervector parameter * make CDE0BlockTransferSrcVectorDim not work * changse toinstance * add limit for CDE0BlockTransferSrcScalarPerVector [ROCm/composable_kernel commit: 50643dd5551e988425b6297d5dbe6d200bd63735] --- ...ed_gemm_add_add_relu_gemm_add_xdl_fp16.cpp | 4 +- ...ultiple_d_gemm_multiple_d_xdl_cshuffle.hpp | 4 + ...iple_d_gemm_multiple_d_xdl_cshuffle_v1.hpp | 114 ++++++++---------- ...6_f16_f16_f16_gmk_gnk_gno_gmo_instance.cpp | 30 ++--- ...6_f16_f16_f16_gmk_gnk_gon_gmo_instance.cpp | 32 ++--- 5 files changed, 91 insertions(+), 93 deletions(-) diff --git a/example/37_batched_gemm_add_add_relu_gemm_add/batched_gemm_add_add_relu_gemm_add_xdl_fp16.cpp b/example/37_batched_gemm_add_add_relu_gemm_add/batched_gemm_add_add_relu_gemm_add_xdl_fp16.cpp index b3d0ab6bf7..36dcf58d70 100644 --- a/example/37_batched_gemm_add_add_relu_gemm_add/batched_gemm_add_add_relu_gemm_add_xdl_fp16.cpp +++ b/example/37_batched_gemm_add_add_relu_gemm_add/batched_gemm_add_add_relu_gemm_add_xdl_fp16.cpp @@ -173,6 +173,8 @@ using DeviceGemmInstance = 8, 8, true, + 9, // D0sTransferSrcVectorDim + 4, // D0sTransferSrcScalaerPerVector S<8, 32, 1>, // B1BlockTransfer S<0, 2, 1>, S<0, 2, 1>, @@ -189,7 +191,7 @@ int main(int argc, char* argv[]) { bool do_verification = true; int init_method = 1; - bool time_kernel = false; + bool time_kernel = true; // GEMM shape ck::index_t M = 1024; diff --git a/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_multiple_d_gemm_multiple_d_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_multiple_d_gemm_multiple_d_xdl_cshuffle.hpp index d259ebffae..d92865a9ec 100644 --- a/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_multiple_d_gemm_multiple_d_xdl_cshuffle.hpp +++ b/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_multiple_d_gemm_multiple_d_xdl_cshuffle.hpp @@ -196,6 +196,8 @@ template {}, Number{}, n2, n4)); + static_assert(CDE0BlockTransferSrcScalarPerVector <= n4, + "vector load must be not greater than n4"); + static_assert(n4 % CDE0BlockTransferSrcScalarPerVector == 0); auto d0s_threadwise_copy = generate_tuple( [&](auto i) { @@ -742,10 +745,19 @@ struct GridwiseBatchedGemmMultipleDGemmMultipleD_Xdl_CShuffle A0B0B1DataType, decltype(d0s_griddesc_m0_n0_m1_n1_m2_n2_m3_n3_n4_n5[i]), decltype(d0_thread_desc_m0_n0_m1_n1_m2_n2_m3_n3_n4_n5), - Sequence, + Sequence, Sequence<0, 1, 2, 3, 4, 5, 6, 7, 8, 9>, - 9, - n4, + 9, // CDE0BlockTransferSrcVectorDim + CDE0BlockTransferSrcScalarPerVector, 1, false>(d0s_griddesc_m0_n0_m1_n1_m2_n2_m3_n3_n4_n5[i], make_multi_index(block_work_idx[I0], // MBlockId @@ -898,66 +910,42 @@ struct GridwiseBatchedGemmMultipleDGemmMultipleD_Xdl_CShuffle blockwise_gemm0, acc0_thread_buf, num_k_block_main_loop); - // bias+gelu + // multiple d + if constexpr(NumD0Tensor) { - static_for<0, Gemm0MXdlPerWave, 1>{}([&](auto mr) { - static_for<0, Gemm0NXdlPerWave, 1>{}([&](auto nr) { - static_for<0, n2, 1>{}([&](auto groupid) { - static_for<0, NumD0Tensor, 1>{}([&](auto i) { - d0s_threadwise_copy(i).Run( - d0s_griddesc_m0_n0_m1_n1_m2_n2_m3_n3_n4_n5[i], - d0s_grid_buf[i], - d0_thread_desc_m0_n0_m1_n1_m2_n2_m3_n3_n4_n5, - make_tuple(I0, I0, I0, I0, I0, I0, I0, I0, I0, I0), - d0s_thread_buf(i)); - }); + static_for<0, NumD0Tensor, 1>{}([&](auto i) { + d0s_threadwise_copy(i).Run(d0s_griddesc_m0_n0_m1_n1_m2_n2_m3_n3_n4_n5[i], + d0s_grid_buf[i], + d0_thread_desc_m0_n0_m1_n1_m2_n2_m3_n3_n4_n5, + make_tuple(I0, I0, I0, I0, I0, I0, I0, I0, I0, I0), + d0s_thread_buf(i)); + }); + static_for<0, m0 * n0 * n2 * n4, 1>{}([&](auto i) { + // get reference to src data + const auto src_data_refs = generate_tie( + // return type should be lvalue + [&](auto iSrc) -> const auto& { return d0s_thread_buf[iSrc][i]; }, + Number{}); - static_for<0, n4, 1>{}([&](auto i) { - constexpr index_t c_offset = acc0_thread_desc.CalculateOffset( - make_tuple(mr, nr, groupid, i)); + // get reference to dst data + auto dst_data_refs = generate_tie( + // return type should be lvalue + [&](auto) -> auto& { return acc0_thread_buf(i); }, + Number<2>{}); - // get reference to src data - const auto src_data_refs = generate_tie( - // return type should be lvalue - [&](auto iSrc) -> const auto& { - return d0s_thread_buf[iSrc][i]; - }, - Number{}); - - // get reference to dst data - auto dst_data_refs = generate_tie( - // return type should be lvalue - [&](auto) -> auto& { - return acc0_thread_buf(Number{}); - }, - Number<2>{}); - - unpack2(cde0_element_op, dst_data_refs, src_data_refs); - }); - static_for<0, NumD0Tensor, 1>{}([&](auto i) { - d0s_threadwise_copy(i).MoveSrcSliceWindow( - d0s_griddesc_m0_n0_m1_n1_m2_n2_m3_n3_n4_n5[i], - make_multi_index(0, 0, 0, 0, 0, 0, 0, 1, 0, 0)); - }); - }); - static_for<0, NumD0Tensor, 1>{}([&](auto i) { - d0s_threadwise_copy(i).MoveSrcSliceWindow( - d0s_griddesc_m0_n0_m1_n1_m2_n2_m3_n3_n4_n5[i], - make_multi_index(0, 0, 0, 1, 0, 0, 0, -n2.value, 0, 0)); - }); - }); - static_for<0, NumD0Tensor, 1>{}([&](auto i) { - d0s_threadwise_copy(i).MoveSrcSliceWindow( - d0s_griddesc_m0_n0_m1_n1_m2_n2_m3_n3_n4_n5[i], - make_multi_index(0, 0, 1, -Gemm0NXdlPerWave, 0, 0, 0, 0, 0, 0)); - }); + unpack2(cde0_element_op, dst_data_refs, src_data_refs); }); static_for<0, NumD0Tensor, 1>{}([&](auto i) { d0s_threadwise_copy(i).MoveSrcSliceWindow( d0s_griddesc_m0_n0_m1_n1_m2_n2_m3_n3_n4_n5[i], - make_multi_index(0, 1, -Gemm0MXdlPerWave, 0, 0, 0, 0, 0, 0, 0)); + make_multi_index(0, 1, 0, 0, 0, 0, 0, 0, 0, 0)); }); } + else + { + static_for<0, acc0_thread_buf.Size(), 1>{}( + [&](auto i) { cde0_element_op(acc_thread_buf(i), acc0_thread_buf[i]); }); + } // gemm1 { // TODO: explore using dynamic buffer for a1 thread buffer diff --git a/library/src/tensor_operation_instance/gpu/batched_gemm_add_relu_gemm_add/device_batched_gemm_add_relu_gemm_add_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gno_gmo_instance.cpp b/library/src/tensor_operation_instance/gpu/batched_gemm_add_relu_gemm_add/device_batched_gemm_add_relu_gemm_add_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gno_gmo_instance.cpp index 4db05589b1..e750f18cc7 100644 --- a/library/src/tensor_operation_instance/gpu/batched_gemm_add_relu_gemm_add/device_batched_gemm_add_relu_gemm_add_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gno_gmo_instance.cpp +++ b/library/src/tensor_operation_instance/gpu/batched_gemm_add_relu_gemm_add/device_batched_gemm_add_relu_gemm_add_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gno_gmo_instance.cpp @@ -31,22 +31,24 @@ using CDE1ElementOp = ck::tensor_operation::element_wise::Add; using device_batched_gemm_add_relu_gemm_add_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gno_gmo_instances = std::tuple< // clang-format off - //##################################################| A0Layout| B0Layout| D0Layout| B1Layout| D1sLayout| E1Layout| A0Data| B0Data| Acc0DataType| D0DataType| B1Data| Acc1CData| CShuffle| D1sData| E1Data| A0| B0| CDE0| B1| CDE1| PadGemm0M| PadGemm0N| PadGemm0K| PadGemm1N| PadGemm1K|NumGemm0K| Block| Gemm0| Gemm0| Gemm0| Gemm1| Gemm1|A0K1|B0K1| B1K1| MPer| NPer| Gemm0| Gemm0| Gemm1|A0BlockTransfer|A0BlockTransfer|A0BlockTransfer|A0BlockTransfer|A0BlockTransfer|A0BlockTransfer|A0BlockLds| B0BlockTransfer| B0BlockTransfer| B0BlockTransfer| B0BlockTransfer| B0BlockTransfer| B0BlockTransfer| B0BlockLds| B1BlockTransfer| B1BlockTransfer| B1BlockTransfer| B1BlockTransfer| B1BlockTransfer| B1BlockTransfer| B1BlockLds| C1Shuffle| C1Shuffle| CDE1BlockTransferClusterLengths| CDE1BlockTransfer| - //##################################################| | | | | | | Type| Type| Type| Type| Type| Type| DataType| Type| Type| Elementwise| Elementwise| Elementwise| Elementwise| Elementwise| | | | | | Prefetch| Size| MPer| NPer| KPer| NPer| KPer| | | | XDL| XDL| MXdl| NXdl| NXdl| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave| _MBlock_MWaveMPerXdl| ScalarPerVector| - //##################################################| | | | | | | | | | | | | | | | Operation| Operation| Operation| Operation| Operation| | | | | | Stage| | Block| Block| Block| Block| Block| | | | | | Per| Per| Per|Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_AK1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| _NBlock_NWaveNPerXdl| _NWaveNPerXdl| - //##################################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | Wave| Wave| Wave| | | | | | | | | | | | | | | | | | | | | | | | | | + //##################################################| A0Layout| B0Layout| D0Layout| B1Layout| D1sLayout| E1Layout| A0Data| B0Data| Acc0DataType| D0DataType| B1Data| Acc1CData| CShuffle| D1sData| E1Data| A0| B0| CDE0| B1| CDE1| PadGemm0M| PadGemm0N| PadGemm0K| PadGemm1N| PadGemm1K|NumGemm0K| Block| Gemm0| Gemm0| Gemm0| Gemm1| Gemm1|A0K1|B0K1| B1K1| MPer| NPer| Gemm0| Gemm0| Gemm1|A0BlockTransfer|A0BlockTransfer|A0BlockTransfer|A0BlockTransfer|A0BlockTransfer|A0BlockTransfer|A0BlockLds| B0BlockTransfer| B0BlockTransfer| B0BlockTransfer| B0BlockTransfer| B0BlockTransfer| B0BlockTransfer| B0BlockLds| CDE0BlockTransfer| CDE0BlockTransfer| B1BlockTransfer| B1BlockTransfer| B1BlockTransfer| B1BlockTransfer| B1BlockTransfer| B1BlockTransfer| B1BlockLds| C1Shuffle| C1Shuffle| CDE1BlockTransferClusterLengths| CDE1BlockTransfer| + //##################################################| | | | | | | Type| Type| Type| Type| Type| Type| DataType| Type| Type| Elementwise| Elementwise| Elementwise| Elementwise| Elementwise| | | | | | Prefetch| Size| MPer| NPer| KPer| NPer| KPer| | | | XDL| XDL| MXdl| NXdl| NXdl| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| SrcVectorDim| SrcScalar| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave| _MBlock_MWaveMPerXdl| ScalarPerVector| + //##################################################| | | | | | | | | | | | | | | | Operation| Operation| Operation| Operation| Operation| | | | | | Stage| | Block| Block| Block| Block| Block| | | | | | Per| Per| Per|Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_AK1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector| Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| _NBlock_NWaveNPerXdl| _NWaveNPerXdl| + //##################################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | Wave| Wave| Wave| | | | | | | | | | | | | | | | | | | | | | | | | | | | + //generic + DeviceBatchedGemmMultipleDGemmMultipleD_Xdl_CShuffle< Row, Col, ck::Tuple, Row, ck::Tuple, Row, F16, F16, F32, ck::Tuple, F16, F32, F32, ck::Tuple, F16, PassThrough, PassThrough, CDE0ElementOp, PassThrough, CDE1ElementOp, true, true, true, true, true, 1, 256, 128, 64, 32, 128, 32, 8, 8, 2, 32, 32, 1, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 9, 1, S< 8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, false, 1, 2, S<1, 32, 1, 8>, 8>, // no padding - DeviceBatchedGemmMultipleDGemmMultipleD_Xdl_CShuffle< Row, Col, ck::Tuple, Row, ck::Tuple, Row, F16, F16, F32, ck::Tuple, F16, F32, F32, ck::Tuple, F16, PassThrough, PassThrough, CDE0ElementOp, PassThrough, CDE1ElementOp, false, false, false, false, false, 1, 256, 128, 128, 64, 64, 32, 8, 8, 2, 32, 32, 1, 4, 2, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, false, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, false, S<16, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, false, 1, 2, S<1, 32, 1, 8>, 8>, - DeviceBatchedGemmMultipleDGemmMultipleD_Xdl_CShuffle< Row, Col, ck::Tuple, Row, ck::Tuple, Row, F16, F16, F32, ck::Tuple, F16, F32, F32, ck::Tuple, F16, PassThrough, PassThrough, CDE0ElementOp, PassThrough, CDE1ElementOp, false, false, false, false, false, 1, 256, 128, 128, 32, 64, 32, 8, 8, 2, 32, 32, 1, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<16, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, false, 1, 2, S<1, 32, 1, 8>, 8>, - DeviceBatchedGemmMultipleDGemmMultipleD_Xdl_CShuffle< Row, Col, ck::Tuple, Row, ck::Tuple, Row, F16, F16, F32, ck::Tuple, F16, F32, F32, ck::Tuple, F16, PassThrough, PassThrough, CDE0ElementOp, PassThrough, CDE1ElementOp, false, false, false, false, false, 1, 256, 128, 128, 64, 128, 32, 8, 8, 2, 32, 32, 1, 4, 4, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, false, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, false, S< 8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, false, 1, 2, S<1, 32, 1, 8>, 8>, - DeviceBatchedGemmMultipleDGemmMultipleD_Xdl_CShuffle< Row, Col, ck::Tuple, Row, ck::Tuple, Row, F16, F16, F32, ck::Tuple, F16, F32, F32, ck::Tuple, F16, PassThrough, PassThrough, CDE0ElementOp, PassThrough, CDE1ElementOp, false, false, false, false, false, 1, 256, 128, 128, 32, 128, 32, 8, 8, 2, 32, 32, 1, 4, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S< 8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, false, 1, 2, S<1, 32, 1, 8>, 8>, - DeviceBatchedGemmMultipleDGemmMultipleD_Xdl_CShuffle< Row, Col, ck::Tuple, Row, ck::Tuple, Row, F16, F16, F32, ck::Tuple, F16, F32, F32, ck::Tuple, F16, PassThrough, PassThrough, CDE0ElementOp, PassThrough, CDE1ElementOp, false, false, false, false, false, 1, 256, 64, 256, 32, 128, 32, 8, 8, 2, 16, 16, 1, 16, 8, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S< 8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, false, 1, 8, S<1, 16, 1,16>, 8>, - DeviceBatchedGemmMultipleDGemmMultipleD_Xdl_CShuffle< Row, Col, ck::Tuple, Row, ck::Tuple, Row, F16, F16, F32, ck::Tuple, F16, F32, F32, ck::Tuple, F16, PassThrough, PassThrough, CDE0ElementOp, PassThrough, CDE1ElementOp, false, false, false, false, false, 1, 256, 64, 256, 32, 64, 32, 8, 8, 2, 16, 16, 1, 16, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<16, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, false, 1, 4, S<1, 32, 1, 8>, 8>, - DeviceBatchedGemmMultipleDGemmMultipleD_Xdl_CShuffle< Row, Col, ck::Tuple, Row, ck::Tuple, Row, F16, F16, F32, ck::Tuple, F16, F32, F32, ck::Tuple, F16, PassThrough, PassThrough, CDE0ElementOp, PassThrough, CDE1ElementOp, false, false, false, false, false, 1, 256, 64, 256, 64, 128, 32, 8, 8, 2, 16, 16, 1, 16, 8, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S< 8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, false, 1, 8, S<1, 16, 1,16>, 8>, - DeviceBatchedGemmMultipleDGemmMultipleD_Xdl_CShuffle< Row, Col, ck::Tuple, Row, ck::Tuple, Row, F16, F16, F32, ck::Tuple, F16, F32, F32, ck::Tuple, F16, PassThrough, PassThrough, CDE0ElementOp, PassThrough, CDE1ElementOp, false, false, false, false, false, 1, 256, 64, 256, 64, 64, 32, 8, 8, 2, 16, 16, 1, 16, 4, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<16, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, false, 1, 4, S<1, 32, 1, 8>, 8>, + DeviceBatchedGemmMultipleDGemmMultipleD_Xdl_CShuffle< Row, Col, ck::Tuple, Row, ck::Tuple, Row, F16, F16, F32, ck::Tuple, F16, F32, F32, ck::Tuple, F16, PassThrough, PassThrough, CDE0ElementOp, PassThrough, CDE1ElementOp, false, false, false, false, false, 1, 256, 128, 128, 64, 64, 32, 8, 8, 2, 32, 32, 1, 4, 2, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, false, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, false, 9, 4, S<16, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, false, 1, 2, S<1, 32, 1, 8>, 8>, + DeviceBatchedGemmMultipleDGemmMultipleD_Xdl_CShuffle< Row, Col, ck::Tuple, Row, ck::Tuple, Row, F16, F16, F32, ck::Tuple, F16, F32, F32, ck::Tuple, F16, PassThrough, PassThrough, CDE0ElementOp, PassThrough, CDE1ElementOp, false, false, false, false, false, 1, 256, 128, 128, 32, 64, 32, 8, 8, 2, 32, 32, 1, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 9, 4, S<16, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, false, 1, 2, S<1, 32, 1, 8>, 8>, + DeviceBatchedGemmMultipleDGemmMultipleD_Xdl_CShuffle< Row, Col, ck::Tuple, Row, ck::Tuple, Row, F16, F16, F32, ck::Tuple, F16, F32, F32, ck::Tuple, F16, PassThrough, PassThrough, CDE0ElementOp, PassThrough, CDE1ElementOp, false, false, false, false, false, 1, 256, 128, 128, 64, 128, 32, 8, 8, 2, 32, 32, 1, 4, 4, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, false, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, false, 9, 4, S< 8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, false, 1, 2, S<1, 32, 1, 8>, 8>, + DeviceBatchedGemmMultipleDGemmMultipleD_Xdl_CShuffle< Row, Col, ck::Tuple, Row, ck::Tuple, Row, F16, F16, F32, ck::Tuple, F16, F32, F32, ck::Tuple, F16, PassThrough, PassThrough, CDE0ElementOp, PassThrough, CDE1ElementOp, false, false, false, false, false, 1, 256, 128, 128, 32, 128, 32, 8, 8, 2, 32, 32, 1, 4, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 9, 4, S< 8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, false, 1, 2, S<1, 32, 1, 8>, 8>, + DeviceBatchedGemmMultipleDGemmMultipleD_Xdl_CShuffle< Row, Col, ck::Tuple, Row, ck::Tuple, Row, F16, F16, F32, ck::Tuple, F16, F32, F32, ck::Tuple, F16, PassThrough, PassThrough, CDE0ElementOp, PassThrough, CDE1ElementOp, false, false, false, false, false, 1, 256, 64, 256, 32, 128, 32, 8, 8, 2, 16, 16, 1, 16, 8, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 9, 4, S< 8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, false, 1, 8, S<1, 16, 1,16>, 8>, + DeviceBatchedGemmMultipleDGemmMultipleD_Xdl_CShuffle< Row, Col, ck::Tuple, Row, ck::Tuple, Row, F16, F16, F32, ck::Tuple, F16, F32, F32, ck::Tuple, F16, PassThrough, PassThrough, CDE0ElementOp, PassThrough, CDE1ElementOp, false, false, false, false, false, 1, 256, 64, 256, 32, 64, 32, 8, 8, 2, 16, 16, 1, 16, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 9, 4, S<16, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, false, 1, 4, S<1, 32, 1, 8>, 8>, + DeviceBatchedGemmMultipleDGemmMultipleD_Xdl_CShuffle< Row, Col, ck::Tuple, Row, ck::Tuple, Row, F16, F16, F32, ck::Tuple, F16, F32, F32, ck::Tuple, F16, PassThrough, PassThrough, CDE0ElementOp, PassThrough, CDE1ElementOp, false, false, false, false, false, 1, 256, 64, 256, 64, 128, 32, 8, 8, 2, 16, 16, 1, 16, 8, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 9, 4, S< 8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, false, 1, 8, S<1, 16, 1,16>, 8>, + DeviceBatchedGemmMultipleDGemmMultipleD_Xdl_CShuffle< Row, Col, ck::Tuple, Row, ck::Tuple, Row, F16, F16, F32, ck::Tuple, F16, F32, F32, ck::Tuple, F16, PassThrough, PassThrough, CDE0ElementOp, PassThrough, CDE1ElementOp, false, false, false, false, false, 1, 256, 64, 256, 64, 64, 32, 8, 8, 2, 16, 16, 1, 16, 4, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 9, 4, S<16, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, false, 1, 4, S<1, 32, 1, 8>, 8>, // Padded fallback kernel - DeviceBatchedGemmMultipleDGemmMultipleD_Xdl_CShuffle< Row, Col, ck::Tuple, Row, ck::Tuple, Row, F16, F16, F32, ck::Tuple, F16, F32, F32, ck::Tuple, F16, PassThrough, PassThrough, CDE0ElementOp, PassThrough, CDE1ElementOp, true, true, true, true, true, 1, 256, 128, 128, 64, 128, 32, 8, 8, 2, 32, 32, 1, 4, 4, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, false, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, false, S< 8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, false, 1, 2, S<1, 32, 1, 8>, 8>, - DeviceBatchedGemmMultipleDGemmMultipleD_Xdl_CShuffle< Row, Col, ck::Tuple, Row, ck::Tuple, Row, F16, F16, F32, ck::Tuple, F16, F32, F32, ck::Tuple, F16, PassThrough, PassThrough, CDE0ElementOp, PassThrough, CDE1ElementOp, true, true, true, true, true, 1, 256, 128, 64, 32, 128, 32, 8, 8, 2, 32, 32, 1, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S< 8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, false, 1, 2, S<1, 32, 1, 8>, 8> + DeviceBatchedGemmMultipleDGemmMultipleD_Xdl_CShuffle< Row, Col, ck::Tuple, Row, ck::Tuple, Row, F16, F16, F32, ck::Tuple, F16, F32, F32, ck::Tuple, F16, PassThrough, PassThrough, CDE0ElementOp, PassThrough, CDE1ElementOp, true, true, true, true, true, 1, 256, 128, 128, 64, 128, 32, 8, 8, 2, 32, 32, 1, 4, 4, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, false, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, false, 9, 4, S< 8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, false, 1, 2, S<1, 32, 1, 8>, 8>, + DeviceBatchedGemmMultipleDGemmMultipleD_Xdl_CShuffle< Row, Col, ck::Tuple, Row, ck::Tuple, Row, F16, F16, F32, ck::Tuple, F16, F32, F32, ck::Tuple, F16, PassThrough, PassThrough, CDE0ElementOp, PassThrough, CDE1ElementOp, true, true, true, true, true, 1, 256, 128, 64, 32, 128, 32, 8, 8, 2, 32, 32, 1, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 9, 4, S< 8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, false, 1, 2, S<1, 32, 1, 8>, 8> // clang-format on >; diff --git a/library/src/tensor_operation_instance/gpu/batched_gemm_add_relu_gemm_add/device_batched_gemm_add_relu_gemm_add_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gon_gmo_instance.cpp b/library/src/tensor_operation_instance/gpu/batched_gemm_add_relu_gemm_add/device_batched_gemm_add_relu_gemm_add_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gon_gmo_instance.cpp index e25f903a82..37342e5f45 100644 --- a/library/src/tensor_operation_instance/gpu/batched_gemm_add_relu_gemm_add/device_batched_gemm_add_relu_gemm_add_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gon_gmo_instance.cpp +++ b/library/src/tensor_operation_instance/gpu/batched_gemm_add_relu_gemm_add/device_batched_gemm_add_relu_gemm_add_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gon_gmo_instance.cpp @@ -31,23 +31,25 @@ using CDE1ElementOp = ck::tensor_operation::element_wise::Add; using device_batched_gemm_add_relu_gemm_add_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gon_gmo_instances = std::tuple< // clang-format off - //##################################################| A0Layout| B0Layout| D0Layout| B1Layout| D1sLayout| E1Layout| A0Data| B0Data| Acc0DataType| D0DataType| B1Data| Acc1CData| CShuffle| D1sData| E1Data| A0| B0| CDE0| B1| CDE1| PadGemm0M| PadGemm0N| PadGemm0K| PadGemm1N| PadGemm1K| NumGemm0K| Block| Gemm0| Gemm0| Gemm0| Gemm1| Gemm1| A0K1| B0K1|B1K1| MPer| NPer| Gemm0| Gemm0| Gemm1| A0BlockTransfer|A0BlockTransfer|A0BlockTransfer|A0BlockTransfer|A0BlockTransfer|A0BlockTransfer|A0BlockLds| B0BlockTransfer| B0BlockTransfer| B0BlockTransfer| B0BlockTransfer| B0BlockTransfer| B0BlockTransfer| B0BlockLds| B1BlockTransfer| B1BlockTransfer| B1BlockTransfer| B1BlockTransfer| B1BlockTransfer| B1BlockTransfer| B1BlockLds| C1Shuffle| C1Shuffle| CDE1BlockTransferClusterLengths| CDE1BlockTransfer| - //##################################################| | | | | | | Type| Type| Type| Type| Type| Type| DataType| Type| Type| Elementwise| Elementwise| Elementwise| Elementwise| Elementwise| | | | | | Prefetch| Size| MPer| NPer| KPer| NPer| KPer| | | | XDL| XDL| MXdl| NXdl| NXdl| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave| _MBlock_MWaveMPerXdl| ScalarPerVector| - //##################################################| | | | | | | | | | | | | | | | Operation| Operation| Operation| Operation| Operation| | | | | | Stage| | Block| Block| Block| Block| Block| | | | | | Per| Per| Per| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_AK1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| _NBlock_NWaveNPerXdl| _NWaveNPerXdl| - //##################################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | Wave| Wave| Wave| | | | | | | | | | | | | | | | | | | | | | | | | | + //##################################################| A0Layout| B0Layout| D0Layout| B1Layout| D1sLayout| E1Layout| A0Data| B0Data| Acc0DataType| D0DataType| B1Data| Acc1CData| CShuffle| D1sData| E1Data| A0| B0| CDE0| B1| CDE1| PadGemm0M| PadGemm0N| PadGemm0K| PadGemm1N| PadGemm1K| NumGemm0K| Block| Gemm0| Gemm0| Gemm0| Gemm1| Gemm1| A0K1| B0K1|B1K1| MPer| NPer| Gemm0| Gemm0| Gemm1| A0BlockTransfer|A0BlockTransfer|A0BlockTransfer|A0BlockTransfer|A0BlockTransfer|A0BlockTransfer|A0BlockLds| B0BlockTransfer| B0BlockTransfer| B0BlockTransfer| B0BlockTransfer| B0BlockTransfer| B0BlockTransfer| B0BlockLds| CDE0BlockTransfer| CDE0BlockTransfer| B1BlockTransfer| B1BlockTransfer| B1BlockTransfer| B1BlockTransfer| B1BlockTransfer| B1BlockTransfer| B1BlockLds| C1Shuffle| C1Shuffle| CDE1BlockTransferClusterLengths| CDE1BlockTransfer| + //##################################################| | | | | | | Type| Type| Type| Type| Type| Type| DataType| Type| Type| Elementwise| Elementwise| Elementwise| Elementwise| Elementwise| | | | | | Prefetch| Size| MPer| NPer| KPer| NPer| KPer| | | | XDL| XDL| MXdl| NXdl| NXdl| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| SrcVectorDim| SrcScalar| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave| _MBlock_MWaveMPerXdl| ScalarPerVector| + //##################################################| | | | | | | | | | | | | | | | Operation| Operation| Operation| Operation| Operation| | | | | | Stage| | Block| Block| Block| Block| Block| | | | | | Per| Per| Per| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_AK1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector| Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| _NBlock_NWaveNPerXdl| _NWaveNPerXdl| + //##################################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | Wave| Wave| Wave| | | | | | | | | | | | | | | | | | | | | | | | | | | | + //generic + DeviceBatchedGemmMultipleDGemmMultipleD_Xdl_CShuffle< Row, Col, ck::Tuple, Col, ck::Tuple, Row, F16, F16, F32, ck::Tuple, F16, F32, F32, ck::Tuple, F16, PassThrough, PassThrough, CDE0ElementOp, PassThrough, CDE1ElementOp, true, true, true, true, true, 1, 256, 128, 64, 32, 128, 32, 8, 8, 4, 32, 32, 1, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 9, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 1, 2, S<1, 32, 1, 8>, 8>, // no padding - DeviceBatchedGemmMultipleDGemmMultipleD_Xdl_CShuffle< Row, Col, ck::Tuple, Col, ck::Tuple, Row, F16, F16, F32, ck::Tuple, F16, F32, F32, ck::Tuple, F16, PassThrough, PassThrough, CDE0ElementOp, PassThrough, CDE1ElementOp, false, false, false, false, false, 1, 256, 256, 128, 32, 128, 32, 8, 8, 4, 32, 32, 2, 4, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 1, 2, S<1, 32, 1, 8>, 8>, - DeviceBatchedGemmMultipleDGemmMultipleD_Xdl_CShuffle< Row, Col, ck::Tuple, Col, ck::Tuple, Row, F16, F16, F32, ck::Tuple, F16, F32, F32, ck::Tuple, F16, PassThrough, PassThrough, CDE0ElementOp, PassThrough, CDE1ElementOp, false, false, false, false, false, 1, 256, 128, 128, 64, 64, 32, 8, 8, 4, 32, 32, 1, 4, 2, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, false, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, false, 1, 2, S<1, 32, 1, 8>, 8>, - DeviceBatchedGemmMultipleDGemmMultipleD_Xdl_CShuffle< Row, Col, ck::Tuple, Col, ck::Tuple, Row, F16, F16, F32, ck::Tuple, F16, F32, F32, ck::Tuple, F16, PassThrough, PassThrough, CDE0ElementOp, PassThrough, CDE1ElementOp, false, false, false, false, false, 1, 256, 128, 128, 32, 64, 32, 8, 8, 4, 32, 32, 1, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 1, 2, S<1, 32, 1, 8>, 8>, - DeviceBatchedGemmMultipleDGemmMultipleD_Xdl_CShuffle< Row, Col, ck::Tuple, Col, ck::Tuple, Row, F16, F16, F32, ck::Tuple, F16, F32, F32, ck::Tuple, F16, PassThrough, PassThrough, CDE0ElementOp, PassThrough, CDE1ElementOp, false, false, false, false, false, 1, 256, 128, 128, 64, 128, 32, 8, 8, 4, 32, 32, 1, 4, 4, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, false, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, false, 1, 2, S<1, 32, 1, 8>, 8>, - DeviceBatchedGemmMultipleDGemmMultipleD_Xdl_CShuffle< Row, Col, ck::Tuple, Col, ck::Tuple, Row, F16, F16, F32, ck::Tuple, F16, F32, F32, ck::Tuple, F16, PassThrough, PassThrough, CDE0ElementOp, PassThrough, CDE1ElementOp, false, false, false, false, false, 1, 256, 128, 128, 32, 128, 32, 8, 8, 4, 32, 32, 1, 4, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 1, 2, S<1, 32, 1, 8>, 8>, - DeviceBatchedGemmMultipleDGemmMultipleD_Xdl_CShuffle< Row, Col, ck::Tuple, Col, ck::Tuple, Row, F16, F16, F32, ck::Tuple, F16, F32, F32, ck::Tuple, F16, PassThrough, PassThrough, CDE0ElementOp, PassThrough, CDE1ElementOp, false, false, false, false, false, 1, 256, 64, 256, 32, 128, 32, 8, 8, 4, 16, 16, 1, 16, 8, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 1, 8, S<1, 16, 1,16>, 8>, - DeviceBatchedGemmMultipleDGemmMultipleD_Xdl_CShuffle< Row, Col, ck::Tuple, Col, ck::Tuple, Row, F16, F16, F32, ck::Tuple, F16, F32, F32, ck::Tuple, F16, PassThrough, PassThrough, CDE0ElementOp, PassThrough, CDE1ElementOp, false, false, false, false, false, 1, 256, 64, 256, 32, 64, 32, 8, 8, 4, 16, 16, 1, 16, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 1, 4, S<1, 32, 1, 8>, 8>, - DeviceBatchedGemmMultipleDGemmMultipleD_Xdl_CShuffle< Row, Col, ck::Tuple, Col, ck::Tuple, Row, F16, F16, F32, ck::Tuple, F16, F32, F32, ck::Tuple, F16, PassThrough, PassThrough, CDE0ElementOp, PassThrough, CDE1ElementOp, false, false, false, false, false, 1, 256, 64, 256, 64, 128, 32, 8, 8, 4, 16, 16, 1, 16, 8, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 1, 8, S<1, 16, 1,16>, 8>, - DeviceBatchedGemmMultipleDGemmMultipleD_Xdl_CShuffle< Row, Col, ck::Tuple, Col, ck::Tuple, Row, F16, F16, F32, ck::Tuple, F16, F32, F32, ck::Tuple, F16, PassThrough, PassThrough, CDE0ElementOp, PassThrough, CDE1ElementOp, false, false, false, false, false, 1, 256, 64, 256, 64, 64, 32, 8, 8, 4, 16, 16, 1, 16, 4, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 1, 4, S<1, 32, 1, 8>, 8>, + DeviceBatchedGemmMultipleDGemmMultipleD_Xdl_CShuffle< Row, Col, ck::Tuple, Col, ck::Tuple, Row, F16, F16, F32, ck::Tuple, F16, F32, F32, ck::Tuple, F16, PassThrough, PassThrough, CDE0ElementOp, PassThrough, CDE1ElementOp, false, false, false, false, false, 1, 256, 256, 128, 32, 128, 32, 8, 8, 4, 32, 32, 2, 4, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 9, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 1, 2, S<1, 32, 1, 8>, 8>, + DeviceBatchedGemmMultipleDGemmMultipleD_Xdl_CShuffle< Row, Col, ck::Tuple, Col, ck::Tuple, Row, F16, F16, F32, ck::Tuple, F16, F32, F32, ck::Tuple, F16, PassThrough, PassThrough, CDE0ElementOp, PassThrough, CDE1ElementOp, false, false, false, false, false, 1, 256, 128, 128, 64, 64, 32, 8, 8, 4, 32, 32, 1, 4, 2, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, false, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 9, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, false, 1, 2, S<1, 32, 1, 8>, 8>, + DeviceBatchedGemmMultipleDGemmMultipleD_Xdl_CShuffle< Row, Col, ck::Tuple, Col, ck::Tuple, Row, F16, F16, F32, ck::Tuple, F16, F32, F32, ck::Tuple, F16, PassThrough, PassThrough, CDE0ElementOp, PassThrough, CDE1ElementOp, false, false, false, false, false, 1, 256, 128, 128, 32, 64, 32, 8, 8, 4, 32, 32, 1, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 9, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 1, 2, S<1, 32, 1, 8>, 8>, + DeviceBatchedGemmMultipleDGemmMultipleD_Xdl_CShuffle< Row, Col, ck::Tuple, Col, ck::Tuple, Row, F16, F16, F32, ck::Tuple, F16, F32, F32, ck::Tuple, F16, PassThrough, PassThrough, CDE0ElementOp, PassThrough, CDE1ElementOp, false, false, false, false, false, 1, 256, 128, 128, 64, 128, 32, 8, 8, 4, 32, 32, 1, 4, 4, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, false, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 9, 4, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, false, 1, 2, S<1, 32, 1, 8>, 8>, + DeviceBatchedGemmMultipleDGemmMultipleD_Xdl_CShuffle< Row, Col, ck::Tuple, Col, ck::Tuple, Row, F16, F16, F32, ck::Tuple, F16, F32, F32, ck::Tuple, F16, PassThrough, PassThrough, CDE0ElementOp, PassThrough, CDE1ElementOp, false, false, false, false, false, 1, 256, 128, 128, 32, 128, 32, 8, 8, 4, 32, 32, 1, 4, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 9, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 1, 2, S<1, 32, 1, 8>, 8>, + DeviceBatchedGemmMultipleDGemmMultipleD_Xdl_CShuffle< Row, Col, ck::Tuple, Col, ck::Tuple, Row, F16, F16, F32, ck::Tuple, F16, F32, F32, ck::Tuple, F16, PassThrough, PassThrough, CDE0ElementOp, PassThrough, CDE1ElementOp, false, false, false, false, false, 1, 256, 64, 256, 32, 128, 32, 8, 8, 4, 16, 16, 1, 16, 8, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 9, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 1, 8, S<1, 16, 1,16>, 8>, + DeviceBatchedGemmMultipleDGemmMultipleD_Xdl_CShuffle< Row, Col, ck::Tuple, Col, ck::Tuple, Row, F16, F16, F32, ck::Tuple, F16, F32, F32, ck::Tuple, F16, PassThrough, PassThrough, CDE0ElementOp, PassThrough, CDE1ElementOp, false, false, false, false, false, 1, 256, 64, 256, 32, 64, 32, 8, 8, 4, 16, 16, 1, 16, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 9, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 1, 4, S<1, 32, 1, 8>, 8>, + DeviceBatchedGemmMultipleDGemmMultipleD_Xdl_CShuffle< Row, Col, ck::Tuple, Col, ck::Tuple, Row, F16, F16, F32, ck::Tuple, F16, F32, F32, ck::Tuple, F16, PassThrough, PassThrough, CDE0ElementOp, PassThrough, CDE1ElementOp, false, false, false, false, false, 1, 256, 64, 256, 64, 128, 32, 8, 8, 4, 16, 16, 1, 16, 8, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 9, 4, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 1, 8, S<1, 16, 1,16>, 8>, + DeviceBatchedGemmMultipleDGemmMultipleD_Xdl_CShuffle< Row, Col, ck::Tuple, Col, ck::Tuple, Row, F16, F16, F32, ck::Tuple, F16, F32, F32, ck::Tuple, F16, PassThrough, PassThrough, CDE0ElementOp, PassThrough, CDE1ElementOp, false, false, false, false, false, 1, 256, 64, 256, 64, 64, 32, 8, 8, 4, 16, 16, 1, 16, 4, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 9, 4, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 1, 4, S<1, 32, 1, 8>, 8>, // Padded fallback kernel - DeviceBatchedGemmMultipleDGemmMultipleD_Xdl_CShuffle< Row, Col, ck::Tuple, Col, ck::Tuple, Row, F16, F16, F32, ck::Tuple, F16, F32, F32, ck::Tuple, F16, PassThrough, PassThrough, CDE0ElementOp, PassThrough, CDE1ElementOp, true, true, true, true, true, 1, 256, 128, 128, 64, 128, 32, 8, 8, 4, 32, 32, 1, 4, 4, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, false, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, false, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, false, 1, 2, S<1, 32, 1, 8>, 8>, - DeviceBatchedGemmMultipleDGemmMultipleD_Xdl_CShuffle< Row, Col, ck::Tuple, Col, ck::Tuple, Row, F16, F16, F32, ck::Tuple, F16, F32, F32, ck::Tuple, F16, PassThrough, PassThrough, CDE0ElementOp, PassThrough, CDE1ElementOp, true, true, true, true, true, 1, 256, 128, 64, 32, 128, 32, 8, 8, 4, 32, 32, 1, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 1, 2, S<1, 32, 1, 8>, 8> + DeviceBatchedGemmMultipleDGemmMultipleD_Xdl_CShuffle< Row, Col, ck::Tuple, Col, ck::Tuple, Row, F16, F16, F32, ck::Tuple, F16, F32, F32, ck::Tuple, F16, PassThrough, PassThrough, CDE0ElementOp, PassThrough, CDE1ElementOp, true, true, true, true, true, 1, 256, 128, 128, 64, 128, 32, 8, 8, 4, 32, 32, 1, 4, 4, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, false, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, false, 9, 4, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, false, 1, 2, S<1, 32, 1, 8>, 8>, + DeviceBatchedGemmMultipleDGemmMultipleD_Xdl_CShuffle< Row, Col, ck::Tuple, Col, ck::Tuple, Row, F16, F16, F32, ck::Tuple, F16, F32, F32, ck::Tuple, F16, PassThrough, PassThrough, CDE0ElementOp, PassThrough, CDE1ElementOp, true, true, true, true, true, 1, 256, 128, 64, 32, 128, 32, 8, 8, 4, 32, 32, 1, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 9, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 1, 2, S<1, 32, 1, 8>, 8> // clang-format on >;