From 0f3d9639a852cafa34071774a06b7ce7e702f23d Mon Sep 17 00:00:00 2001 From: Adam Osewski <19374865+aosewski@users.noreply.github.com> Date: Sun, 20 Nov 2022 05:08:26 +0100 Subject: [PATCH] Client examples AddFastGelu and FastGelu + instances. (#509) * FastGelu support for more data types. * AddFastGelu & FastGelu instances. * Client example. * clang-format * Remove unused stride variable. * Add new line at EOF. Co-authored-by: Adam Osewski [ROCm/composable_kernel commit: 43a889b72e3faabf04c16ff410d387ce28486c3e] --- .../02_gemm_add_add_fastgelu/CMakeLists.txt | 11 + .../gemm_add_fastgelu.cpp | 233 ++++++++++++++++++ .../gemm_fastgelu.cpp | 225 +++++++++++++++++ .../element/unary_element_wise_operation.hpp | 35 ++- .../device_operation_instance_factory.hpp | 2 + .../gpu/gemm_add_fastgelu.hpp | 145 +++++++++++ .../gpu/gemm_fastgelu.hpp | 138 +++++++++++ .../gpu/gemm_add_fastgelu/CMakeLists.txt | 6 + ...e_f16_f16_f16_f16_km_kn_mn_mn_instance.cpp | 109 ++++++++ ...e_f16_f16_f16_f16_km_nk_mn_mn_instance.cpp | 109 ++++++++ ...e_f16_f16_f16_f16_mk_kn_mn_mn_instance.cpp | 109 ++++++++ ...e_f16_f16_f16_f16_mk_nk_mn_mn_instance.cpp | 100 ++++++++ .../gpu/gemm_fastgelu/CMakeLists.txt | 6 + ..._shuffle_f16_f16_f16_km_kn_mn_instance.cpp | 109 ++++++++ ..._shuffle_f16_f16_f16_km_nk_mn_instance.cpp | 109 ++++++++ ..._shuffle_f16_f16_f16_mk_kn_mn_instance.cpp | 109 ++++++++ ..._shuffle_f16_f16_f16_mk_nk_mn_instance.cpp | 100 ++++++++ 17 files changed, 1645 insertions(+), 10 deletions(-) create mode 100644 client_example/02_gemm_add_add_fastgelu/gemm_add_fastgelu.cpp create mode 100644 client_example/02_gemm_add_add_fastgelu/gemm_fastgelu.cpp create mode 100644 library/include/ck/library/tensor_operation_instance/gpu/gemm_add_fastgelu.hpp create mode 100644 library/include/ck/library/tensor_operation_instance/gpu/gemm_fastgelu.hpp create mode 100644 library/src/tensor_operation_instance/gpu/gemm_add_fastgelu/CMakeLists.txt create mode 100644 library/src/tensor_operation_instance/gpu/gemm_add_fastgelu/device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_km_kn_mn_mn_instance.cpp create mode 100644 library/src/tensor_operation_instance/gpu/gemm_add_fastgelu/device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_km_nk_mn_mn_instance.cpp create mode 100644 library/src/tensor_operation_instance/gpu/gemm_add_fastgelu/device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_mk_kn_mn_mn_instance.cpp create mode 100644 library/src/tensor_operation_instance/gpu/gemm_add_fastgelu/device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_mk_nk_mn_mn_instance.cpp create mode 100644 library/src/tensor_operation_instance/gpu/gemm_fastgelu/CMakeLists.txt create mode 100644 library/src/tensor_operation_instance/gpu/gemm_fastgelu/device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_km_kn_mn_instance.cpp create mode 100644 library/src/tensor_operation_instance/gpu/gemm_fastgelu/device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_km_nk_mn_instance.cpp create mode 100644 library/src/tensor_operation_instance/gpu/gemm_fastgelu/device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_mk_kn_mn_instance.cpp create mode 100644 library/src/tensor_operation_instance/gpu/gemm_fastgelu/device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_mk_nk_mn_instance.cpp diff --git a/client_example/02_gemm_add_add_fastgelu/CMakeLists.txt b/client_example/02_gemm_add_add_fastgelu/CMakeLists.txt index 1064abc8fa..b7b724ccc4 100644 --- a/client_example/02_gemm_add_add_fastgelu/CMakeLists.txt +++ b/client_example/02_gemm_add_add_fastgelu/CMakeLists.txt @@ -1,2 +1,13 @@ +add_custom_target(client_gemm_fastgelu_examples) + add_executable(client_gemm_add_add_fastgelu gemm_add_add_fastgelu.cpp) target_link_libraries(client_gemm_add_add_fastgelu PRIVATE composable_kernel::device_operations) + +add_executable(client_gemm_add_fastgelu gemm_add_fastgelu.cpp) +target_link_libraries(client_gemm_add_fastgelu PRIVATE composable_kernel::device_operations) + +add_executable(client_gemm_fastgelu gemm_fastgelu.cpp) +target_link_libraries(client_gemm_fastgelu PRIVATE composable_kernel::device_operations) + +add_dependencies(client_gemm_fastgelu_examples client_gemm_add_add_fastgelu client_gemm_add_fastgelu + client_gemm_fastgelu) diff --git a/client_example/02_gemm_add_add_fastgelu/gemm_add_fastgelu.cpp b/client_example/02_gemm_add_add_fastgelu/gemm_add_fastgelu.cpp new file mode 100644 index 0000000000..512555f978 --- /dev/null +++ b/client_example/02_gemm_add_add_fastgelu/gemm_add_fastgelu.cpp @@ -0,0 +1,233 @@ +// SPDX-License-Identifier: MIT +// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved. + +#include +#include +#include + +#include "ck/ck.hpp" +#include "ck/tensor_operation/gpu/device/tensor_layout.hpp" +#include "ck/tensor_operation/gpu/device/device_gemm_multiple_d.hpp" +#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp" + +#include "ck/library/tensor_operation_instance/gpu/gemm_add_fastgelu.hpp" + +using F16 = ck::half_t; +using F32 = float; + +using Row = ck::tensor_layout::gemm::RowMajor; +using Col = ck::tensor_layout::gemm::ColumnMajor; + +using PassThrough = ck::tensor_operation::element_wise::PassThrough; +using AddFastGelu = ck::tensor_operation::element_wise::AddFastGelu; + +using AElementOp = PassThrough; +using BElementOp = PassThrough; +using CDEElementOp = AddFastGelu; + +using ADataType = F16; +using BDataType = F16; +using D0DataType = F16; +using EDataType = F16; + +using ALayout = Row; +using BLayout = Col; +using D0Layout = Row; +using ELayout = Row; + +struct SimpleDeviceMem +{ + SimpleDeviceMem() = delete; + + SimpleDeviceMem(std::size_t mem_size) : p_mem_{} + { + (void)hipMalloc(static_cast(&p_mem_), mem_size); + } + + void* GetDeviceBuffer() { return p_mem_; } + + ~SimpleDeviceMem() { (void)hipFree(p_mem_); } + + void* p_mem_; +}; + +int main(int argc, char* argv[]) +{ + // GEMM shape + ck::index_t M = 3840; + ck::index_t N = 4096; + ck::index_t K = 4096; + + ck::index_t StrideA = 4096; + ck::index_t StrideB = 4096; + ck::index_t StrideD0 = 0; + ck::index_t StrideE = 4096; + + if(argc == 1) + { + // use default case + } + else if(argc == 8) + { + M = std::stoi(argv[1]); + N = std::stoi(argv[2]); + K = std::stoi(argv[3]); + + StrideA = std::stoi(argv[4]); + StrideB = std::stoi(argv[5]); + StrideD0 = std::stoi(argv[6]); + StrideE = std::stoi(argv[8]); + } + else + { + printf("arg1 to 7: M, N, K, StrideA, StrideB, StrideD0, StrideE\n"); + exit(0); + } + + auto f_matrix_space_size = + [](std::size_t nRow, std::size_t nCol, std::size_t stride, auto layout) { + using Layout = decltype(layout); + + if(std::is_same::value) + { + return (nRow - 1) * stride + nCol; + } + else + { + return (nCol - 1) * stride + nRow; + } + }; + + SimpleDeviceMem a_device_buf(sizeof(ADataType) * f_matrix_space_size(M, K, StrideA, ALayout{})); + SimpleDeviceMem b_device_buf(sizeof(BDataType) * f_matrix_space_size(K, N, StrideB, BLayout{})); + SimpleDeviceMem d0_m_n_device_buf(sizeof(D0DataType) * + f_matrix_space_size(M, N, StrideD0, D0Layout{})); + SimpleDeviceMem e_device_buf(sizeof(EDataType) * f_matrix_space_size(M, N, StrideE, ELayout{})); + + using DeviceOp = ck::tensor_operation::device::DeviceGemmMultipleD< + ALayout, + BLayout, + ck::Tuple, + ELayout, + ADataType, + BDataType, + ck::Tuple, + EDataType, + ck::tensor_operation::element_wise::PassThrough, + ck::tensor_operation::element_wise::PassThrough, + ck::tensor_operation::element_wise::AddFastGelu>; + + // get device op instances + const auto op_ptrs = ck::tensor_operation::device::instance::DeviceOperationInstanceFactory< + DeviceOp>::GetInstances(); + + std::cout << "found " << op_ptrs.size() << " instances" << std::endl; + + const auto a_element_op = AElementOp{}; + const auto b_element_op = BElementOp{}; + const auto cde_element_op = CDEElementOp{}; + + std::string best_op_name; + bool found = false; + int best_op_id = -1; + float best_ave_time = 0; + float best_tflops = 0; + float best_gb_per_sec = 0; + + // profile device operation instances + std::cout << "Run all instances and do timing" << std::endl; + + for(int i = 0; i < op_ptrs.size(); ++i) + { + auto& op_ptr = op_ptrs[i]; + + auto argument_ptr = op_ptr->MakeArgumentPointer( + a_device_buf.GetDeviceBuffer(), + b_device_buf.GetDeviceBuffer(), + std::array{d0_m_n_device_buf.GetDeviceBuffer()}, + e_device_buf.GetDeviceBuffer(), + M, + N, + K, + StrideA, + StrideB, + std::array{StrideD0}, + StrideE, + a_element_op, + b_element_op, + cde_element_op); + + auto invoker_ptr = op_ptr->MakeInvokerPointer(); + + std::string op_name = op_ptr->GetTypeString(); + + if(op_ptr->IsSupportedArgument(argument_ptr.get())) + { + float ave_time = invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, true}); + + std::size_t flop = std::size_t(2) * M * N * K; + + std::size_t num_btype = + sizeof(ADataType) * M * K + sizeof(BDataType) * K * N + sizeof(EDataType) * M * N; + + float tflops = static_cast(flop) / 1.E9 / ave_time; + + float gb_per_sec = num_btype / 1.E6 / ave_time; + + std::cout << "Perf: " << std::setw(10) << ave_time << " ms, " << tflops << " TFlops, " + << gb_per_sec << " GB/s, " << op_name << std::endl; + + if(tflops > best_tflops) + { + found = true; + best_op_id = i; + best_op_name = op_name; + best_tflops = tflops; + best_ave_time = ave_time; + best_gb_per_sec = gb_per_sec; + } + } + else + { + std::cout << op_name << " does not support this problem" << std::endl; + } + } + + std::cout << "Best Perf: " << best_ave_time << " ms, " << best_tflops << " TFlops, " + << best_gb_per_sec << " GB/s, " << best_op_name << std::endl; + + // run the best intance + { + auto& op_ptr = op_ptrs[best_op_id]; + + std::cout << "Run the best instance without timing: " << op_ptr->GetTypeString() + << std::endl; + + auto argument_ptr = op_ptr->MakeArgumentPointer( + a_device_buf.GetDeviceBuffer(), + b_device_buf.GetDeviceBuffer(), + std::array{d0_m_n_device_buf.GetDeviceBuffer()}, + e_device_buf.GetDeviceBuffer(), + M, + N, + K, + StrideA, + StrideB, + std::array{StrideD0}, + StrideE, + a_element_op, + b_element_op, + cde_element_op); + + auto invoker_ptr = op_ptr->MakeInvokerPointer(); + + if(op_ptr->IsSupportedArgument(argument_ptr.get())) + { + invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, false}); + } + + std::cout << "Done" << std::endl; + } + + return 0; +} diff --git a/client_example/02_gemm_add_add_fastgelu/gemm_fastgelu.cpp b/client_example/02_gemm_add_add_fastgelu/gemm_fastgelu.cpp new file mode 100644 index 0000000000..7237231032 --- /dev/null +++ b/client_example/02_gemm_add_add_fastgelu/gemm_fastgelu.cpp @@ -0,0 +1,225 @@ +// SPDX-License-Identifier: MIT +// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved. + +#include +#include +#include + +#include "ck/ck.hpp" +#include "ck/tensor_operation/gpu/device/tensor_layout.hpp" +#include "ck/tensor_operation/gpu/device/device_gemm_multiple_d.hpp" +#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp" + +#include "ck/library/tensor_operation_instance/gpu/gemm_fastgelu.hpp" + +using F16 = ck::half_t; +using F32 = float; + +using Row = ck::tensor_layout::gemm::RowMajor; +using Col = ck::tensor_layout::gemm::ColumnMajor; + +using PassThrough = ck::tensor_operation::element_wise::PassThrough; +using FastGelu = ck::tensor_operation::element_wise::FastGelu; + +using AElementOp = PassThrough; +using BElementOp = PassThrough; +using CDEElementOp = FastGelu; + +using ADataType = F16; +using BDataType = F16; +using EDataType = F16; + +using ALayout = Row; +using BLayout = Col; +using ELayout = Row; + +struct SimpleDeviceMem +{ + SimpleDeviceMem() = delete; + + SimpleDeviceMem(std::size_t mem_size) : p_mem_{} + { + (void)hipMalloc(static_cast(&p_mem_), mem_size); + } + + void* GetDeviceBuffer() { return p_mem_; } + + ~SimpleDeviceMem() { (void)hipFree(p_mem_); } + + void* p_mem_; +}; + +int main(int argc, char* argv[]) +{ + // GEMM shape + ck::index_t M = 3840; + ck::index_t N = 4096; + ck::index_t K = 4096; + + ck::index_t StrideA = 4096; + ck::index_t StrideB = 4096; + ck::index_t StrideE = 4096; + + if(argc == 1) + { + // use default case + } + else if(argc == 7) + { + M = std::stoi(argv[1]); + N = std::stoi(argv[2]); + K = std::stoi(argv[3]); + + StrideA = std::stoi(argv[4]); + StrideB = std::stoi(argv[5]); + StrideE = std::stoi(argv[8]); + } + else + { + printf("arg1 to 6: M, N, K, StrideA, StrideB, StrideE\n"); + exit(0); + } + + auto f_matrix_space_size = + [](std::size_t nRow, std::size_t nCol, std::size_t stride, auto layout) { + using Layout = decltype(layout); + + if(std::is_same::value) + { + return (nRow - 1) * stride + nCol; + } + else + { + return (nCol - 1) * stride + nRow; + } + }; + + SimpleDeviceMem a_device_buf(sizeof(ADataType) * f_matrix_space_size(M, K, StrideA, ALayout{})); + SimpleDeviceMem b_device_buf(sizeof(BDataType) * f_matrix_space_size(K, N, StrideB, BLayout{})); + SimpleDeviceMem e_device_buf(sizeof(EDataType) * f_matrix_space_size(M, N, StrideE, ELayout{})); + + using DeviceOp = ck::tensor_operation::device::DeviceGemmMultipleD< + ALayout, + BLayout, + ck::Tuple<>, + ELayout, + ADataType, + BDataType, + ck::Tuple<>, + EDataType, + ck::tensor_operation::element_wise::PassThrough, + ck::tensor_operation::element_wise::PassThrough, + ck::tensor_operation::element_wise::FastGelu>; + + // get device op instances + const auto op_ptrs = ck::tensor_operation::device::instance::DeviceOperationInstanceFactory< + DeviceOp>::GetInstances(); + + std::cout << "found " << op_ptrs.size() << " instances" << std::endl; + + const auto a_element_op = AElementOp{}; + const auto b_element_op = BElementOp{}; + const auto cde_element_op = CDEElementOp{}; + + std::string best_op_name; + bool found = false; + int best_op_id = -1; + float best_ave_time = 0; + float best_tflops = 0; + float best_gb_per_sec = 0; + + // profile device operation instances + std::cout << "Run all instances and do timing" << std::endl; + + for(int i = 0; i < op_ptrs.size(); ++i) + { + auto& op_ptr = op_ptrs[i]; + + auto argument_ptr = op_ptr->MakeArgumentPointer(a_device_buf.GetDeviceBuffer(), + b_device_buf.GetDeviceBuffer(), + {}, + e_device_buf.GetDeviceBuffer(), + M, + N, + K, + StrideA, + StrideB, + {}, + StrideE, + a_element_op, + b_element_op, + cde_element_op); + + auto invoker_ptr = op_ptr->MakeInvokerPointer(); + + std::string op_name = op_ptr->GetTypeString(); + + if(op_ptr->IsSupportedArgument(argument_ptr.get())) + { + float ave_time = invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, true}); + + std::size_t flop = std::size_t(2) * M * N * K; + + std::size_t num_btype = + sizeof(ADataType) * M * K + sizeof(BDataType) * K * N + sizeof(EDataType) * M * N; + + float tflops = static_cast(flop) / 1.E9 / ave_time; + + float gb_per_sec = num_btype / 1.E6 / ave_time; + + std::cout << "Perf: " << std::setw(10) << ave_time << " ms, " << tflops << " TFlops, " + << gb_per_sec << " GB/s, " << op_name << std::endl; + + if(tflops > best_tflops) + { + found = true; + best_op_id = i; + best_op_name = op_name; + best_tflops = tflops; + best_ave_time = ave_time; + best_gb_per_sec = gb_per_sec; + } + } + else + { + std::cout << op_name << " does not support this problem" << std::endl; + } + } + + std::cout << "Best Perf: " << best_ave_time << " ms, " << best_tflops << " TFlops, " + << best_gb_per_sec << " GB/s, " << best_op_name << std::endl; + + // run the best intance + { + auto& op_ptr = op_ptrs[best_op_id]; + + std::cout << "Run the best instance without timing: " << op_ptr->GetTypeString() + << std::endl; + + auto argument_ptr = op_ptr->MakeArgumentPointer(a_device_buf.GetDeviceBuffer(), + b_device_buf.GetDeviceBuffer(), + {}, + e_device_buf.GetDeviceBuffer(), + M, + N, + K, + StrideA, + StrideB, + {}, + StrideE, + a_element_op, + b_element_op, + cde_element_op); + + auto invoker_ptr = op_ptr->MakeInvokerPointer(); + + if(op_ptr->IsSupportedArgument(argument_ptr.get())) + { + invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, false}); + } + + std::cout << "Done" << std::endl; + } + + return 0; +} diff --git a/include/ck/tensor_operation/gpu/element/unary_element_wise_operation.hpp b/include/ck/tensor_operation/gpu/element/unary_element_wise_operation.hpp index dcf80b9439..fbdfe9262f 100644 --- a/include/ck/tensor_operation/gpu/element/unary_element_wise_operation.hpp +++ b/include/ck/tensor_operation/gpu/element/unary_element_wise_operation.hpp @@ -194,21 +194,36 @@ struct Relu } }; -// https://paperswithcode.com/method/gelu -// y = 0.5*x*(1+tanh(sqrt(2/pi)*(x+0.044715*x^3))) +// Y = FastGelu(X) struct FastGelu { - template - __host__ __device__ void operator()(Y& y, const X& x) const; - - template <> - __host__ __device__ void operator()(float& y, const float& x) const + // Fast GeLU + // https://paperswithcode.com/method/gelu + // y = 0.5*x*(1+tanh(sqrt(2/pi)*(x+0.044715*x^3))) + __host__ __device__ static constexpr float GetFastGeLU(float x) { - const float u = float(2) * x * (float(0.035677) * x * x + float(0.797885)); + const float u = 2.f * x * (0.035677f * x * x + 0.797885f); const float emu = exp(-u); - const float cdf = float(0.5) + float(0.5) * (float(2) / (float(1) + emu) - float(1)); + const float cdf = 0.5f + 0.5f * (2.f / (1.f + emu) - 1.f); + return x * cdf; + } - y = x * cdf; + template + static inline constexpr bool is_valid_param_type_v = + std::is_same_v || std::is_same_v || std::is_same_v || + std::is_same_v || std::is_same_v +#ifdef CK_EXPERIMENTAL_BIT_INT_EXTENSION_INT4 + || std::is_same_v +#endif + ; + + template + __host__ __device__ void operator()(Y& y, const X& x) const + { + static_assert(is_valid_param_type_v && is_valid_param_type_v); + + const float tmp_y = GetFastGeLU(type_convert(x)); + y = type_convert(tmp_y); } }; diff --git a/library/include/ck/library/tensor_operation_instance/device_operation_instance_factory.hpp b/library/include/ck/library/tensor_operation_instance/device_operation_instance_factory.hpp index f8d408dfff..785d5510f3 100644 --- a/library/include/ck/library/tensor_operation_instance/device_operation_instance_factory.hpp +++ b/library/include/ck/library/tensor_operation_instance/device_operation_instance_factory.hpp @@ -87,6 +87,8 @@ using Relu = ck::tensor_operation::element_wise::Relu; using Scale = ck::tensor_operation::element_wise::Scale; using Bilinear = ck::tensor_operation::element_wise::Bilinear; using AddAddFastGelu = ck::tensor_operation::element_wise::AddAddFastGelu; +using AddFastGelu = ck::tensor_operation::element_wise::AddFastGelu; +using FastGelu = ck::tensor_operation::element_wise::FastGelu; template using Activation_Mul_Clamp = ck::tensor_operation::element_wise::Activation_Mul_Clamp; diff --git a/library/include/ck/library/tensor_operation_instance/gpu/gemm_add_fastgelu.hpp b/library/include/ck/library/tensor_operation_instance/gpu/gemm_add_fastgelu.hpp new file mode 100644 index 0000000000..554437f490 --- /dev/null +++ b/library/include/ck/library/tensor_operation_instance/gpu/gemm_add_fastgelu.hpp @@ -0,0 +1,145 @@ +// SPDX-License-Identifier: MIT +// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved. + +#pragma once + +#include +#include +#include + +#include "ck/ck.hpp" +#include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp" +#include "ck/tensor_operation/gpu/device/device_gemm_multiple_d.hpp" + +namespace ck { +namespace tensor_operation { +namespace device { +namespace instance { + +void add_device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_mk_kn_mn_mn_instances( + std::vector>>&); + +void add_device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_mk_nk_mn_mn_instances( + std::vector>>&); + +void add_device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_km_kn_mn_mn_instances( + std::vector>>&); + +void add_device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_km_nk_mn_mn_instances( + std::vector>>&); + +// GEMM + Add + FastGelu +template +struct DeviceOperationInstanceFactory< + ck::tensor_operation::device::DeviceGemmMultipleD, + ELayout, + ADataType, + BDataType, + ck::Tuple, + EDataType, + PassThrough, + PassThrough, + AddFastGelu>> +{ + using DeviceOp = DeviceGemmMultipleD, + ELayout, + ADataType, + BDataType, + ck::Tuple, + EDataType, + PassThrough, + PassThrough, + AddFastGelu>; + + static auto GetInstances() + { + std::vector> op_ptrs; + + if constexpr(is_same_v && is_same_v && + is_same_v && is_same_v) + { + if constexpr(is_same_v && is_same_v && + is_same_v && is_same_v) + { + add_device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_mk_kn_mn_mn_instances( + op_ptrs); + } + else if constexpr(is_same_v && is_same_v && + is_same_v && is_same_v) + { + add_device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_mk_nk_mn_mn_instances( + op_ptrs); + } + else if constexpr(is_same_v && is_same_v && + is_same_v && is_same_v) + { + add_device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_km_kn_mn_mn_instances( + op_ptrs); + } + else if constexpr(is_same_v && is_same_v && + is_same_v && is_same_v) + { + add_device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_km_nk_mn_mn_instances( + op_ptrs); + } + } + + return op_ptrs; + } +}; + +} // namespace instance +} // namespace device +} // namespace tensor_operation +} // namespace ck diff --git a/library/include/ck/library/tensor_operation_instance/gpu/gemm_fastgelu.hpp b/library/include/ck/library/tensor_operation_instance/gpu/gemm_fastgelu.hpp new file mode 100644 index 0000000000..fbc5df98a4 --- /dev/null +++ b/library/include/ck/library/tensor_operation_instance/gpu/gemm_fastgelu.hpp @@ -0,0 +1,138 @@ +// SPDX-License-Identifier: MIT +// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved. + +#pragma once + +#include +#include +#include + +#include "ck/ck.hpp" +#include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp" +#include "ck/tensor_operation/gpu/device/device_gemm_multiple_d.hpp" + +namespace ck { +namespace tensor_operation { +namespace device { +namespace instance { + +void add_device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_mk_kn_mn_instances( + std::vector>>&); + +void add_device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_mk_nk_mn_instances( + std::vector>>&); + +void add_device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_km_kn_mn_instances( + std::vector>>&); + +void add_device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_km_nk_mn_instances( + std::vector>>&); + +// GEMM + FastGelu +template +struct DeviceOperationInstanceFactory> +{ + using DeviceOp = DeviceGemmMultipleD; + + static auto GetInstances() + { + std::vector> op_ptrs; + + if constexpr(is_same_v && is_same_v && + is_same_v) + { + if constexpr(is_same_v && is_same_v && + is_same_v) + { + add_device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_mk_kn_mn_instances(op_ptrs); + } + else if constexpr(is_same_v && is_same_v && + is_same_v) + { + add_device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_mk_nk_mn_instances(op_ptrs); + } + else if constexpr(is_same_v && is_same_v && + is_same_v) + { + add_device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_km_kn_mn_instances(op_ptrs); + } + else if constexpr(is_same_v && is_same_v && + is_same_v) + { + add_device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_km_nk_mn_instances(op_ptrs); + } + } + + return op_ptrs; + } +}; + +} // namespace instance +} // namespace device +} // namespace tensor_operation +} // namespace ck diff --git a/library/src/tensor_operation_instance/gpu/gemm_add_fastgelu/CMakeLists.txt b/library/src/tensor_operation_instance/gpu/gemm_add_fastgelu/CMakeLists.txt new file mode 100644 index 0000000000..0beb10e379 --- /dev/null +++ b/library/src/tensor_operation_instance/gpu/gemm_add_fastgelu/CMakeLists.txt @@ -0,0 +1,6 @@ +add_instance_library(device_gemm_add_fastgelu_instance + device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_km_kn_mn_mn_instance.cpp + device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_km_nk_mn_mn_instance.cpp + device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_mk_kn_mn_mn_instance.cpp + device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_mk_nk_mn_mn_instance.cpp +) diff --git a/library/src/tensor_operation_instance/gpu/gemm_add_fastgelu/device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_km_kn_mn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_add_fastgelu/device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_km_kn_mn_mn_instance.cpp new file mode 100644 index 0000000000..b99ccd3263 --- /dev/null +++ b/library/src/tensor_operation_instance/gpu/gemm_add_fastgelu/device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_km_kn_mn_mn_instance.cpp @@ -0,0 +1,109 @@ +// SPDX-License-Identifier: MIT +// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved. + +#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp" +#include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp" +#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp" +#include "ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_xdl_cshuffle.hpp" +#include "ck/utility/sequence.hpp" + +namespace ck { +namespace tensor_operation { +namespace device { +namespace instance { + +template +using S = ck::Sequence; + +static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default; + +// e = elementwise((a * b), d0) +// outout: e[m, n] +// input: a[k, m], b[k, n], d0[m, n] +using device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_km_kn_mn_mn_instances = std::tuple< + // clang-format off + //##############################| A| B| Ds| E| AData| BData| AccData| CShuffle| DsData| EData| A| B| CDE| GEMM| NumGemmK| Block| MPer| NPer| KPer| AK1| BK1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CShuffle| CShuffle| CBlockTransferClusterLengths| CBlockTransfer| LoopScheduler| Pipeline| + //##############################| Layout| Layout| Layout| Layout| Type| Type| Type| DataType| Type| Type| Elementwise| Elementwise| Elementwise| Specialization| Prefetch| Size| Block| Block| Block| | | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave| _MBlock_MWaveMPerXdl| ScalarPerVector| | | + //##############################| | | | | | | | | | | Operation| Operation| Operation| | Stage| | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| _NBlock_NWaveNPerXdl| _NWaveNPerXdl| | | + //##############################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | + // pipeline v1, 1 wave + DeviceGemmMultipleD_Xdl_CShuffle< Col, Row, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 256, 256, 128, 32, 2, 2, 32, 32, 4, 2, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, S<8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Row, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 256, 256, 128, 32, 8, 8, 32, 32, 4, 2, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 8, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Row, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 256, 128, 256, 32, 2, 2, 32, 32, 2, 4, S<8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Row, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 256, 128, 256, 32, 8, 8, 32, 32, 2, 4, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Row, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 128, 128, 128, 32, 2, 2, 32, 32, 4, 2, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, 1, 1, S<1, 16, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Row, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 128, 128, 128, 32, 8, 8, 32, 32, 4, 2, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 8, 1, 1, 1, S<1, 16, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Row, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 256, 128, 128, 32, 2, 2, 32, 32, 2, 2, S<8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, S<8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Row, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 256, 128, 128, 32, 8, 8, 32, 32, 2, 2, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Row, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 128, 128, 64, 32, 2, 2, 32, 32, 2, 2, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, 1, 1, S<1, 32, 1, 4>, 8, LoopScheduler::Default, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Row, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 128, 128, 64, 32, 8, 8, 32, 32, 2, 2, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, 1, 1, 1, S<1, 32, 1, 4>, 8, LoopScheduler::Default, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Row, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 128, 64, 128, 32, 2, 2, 32, 32, 2, 2, S<8, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, 1, 1, S<1, 16, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Row, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 128, 64, 128, 32, 8, 8, 32, 32, 2, 2, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 8, 1, 1, 1, S<1, 16, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Row, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 256, 128, 64, 32, 2, 2, 32, 32, 2, 1, S<8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, S<16,16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Row, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 256, 128, 64, 32, 8, 8, 32, 32, 2, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Row, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 256, 64, 128, 32, 2, 2, 32, 32, 1, 2, S<16,16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, S<8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Row, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 256, 64, 128, 32, 8, 8, 32, 32, 1, 2, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1> +#if CK_EXPERIMENTAL_INTER_WAVE_INSTANCES + // pipeline v1, 2 waves + , + DeviceGemmMultipleD_Xdl_CShuffle< Col, Row, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 256, 256, 128, 32, 2, 2, 32, 32, 4, 2, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, S<8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Interwave, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Row, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 256, 256, 128, 32, 8, 8, 32, 32, 4, 2, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 8, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Interwave, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Row, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 256, 128, 256, 32, 2, 2, 32, 32, 2, 4, S<8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Interwave, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Row, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 256, 128, 256, 32, 8, 8, 32, 32, 2, 4, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Interwave, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Row, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 128, 128, 128, 32, 2, 2, 32, 32, 4, 2, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, 1, 1, S<1, 16, 1, 8>, 8, LoopScheduler::Interwave, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Row, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 128, 128, 128, 32, 8, 8, 32, 32, 4, 2, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 8, 1, 1, 1, S<1, 16, 1, 8>, 8, LoopScheduler::Interwave, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Row, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 256, 128, 128, 32, 2, 2, 32, 32, 2, 2, S<8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, S<8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Interwave, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Row, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 256, 128, 128, 32, 8, 8, 32, 32, 2, 2, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Interwave, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Row, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 128, 128, 64, 32, 2, 2, 32, 32, 2, 2, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, 1, 1, S<1, 32, 1, 4>, 8, LoopScheduler::Interwave, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Row, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 128, 128, 64, 32, 8, 8, 32, 32, 2, 2, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, 1, 1, 1, S<1, 32, 1, 4>, 8, LoopScheduler::Interwave, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Row, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 128, 64, 128, 32, 2, 2, 32, 32, 2, 2, S<8, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, 1, 1, S<1, 16, 1, 8>, 8, LoopScheduler::Interwave, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Row, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 128, 64, 128, 32, 8, 8, 32, 32, 2, 2, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 8, 1, 1, 1, S<1, 16, 1, 8>, 8, LoopScheduler::Interwave, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Row, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 256, 128, 64, 32, 2, 2, 32, 32, 2, 1, S<8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, S<16,16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Interwave, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Row, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 256, 128, 64, 32, 8, 8, 32, 32, 2, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Interwave, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Row, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 256, 64, 128, 32, 2, 2, 32, 32, 1, 2, S<16,16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, S<8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Interwave, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Row, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 256, 64, 128, 32, 8, 8, 32, 32, 1, 2, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Interwave, PipelineVersion::v1> +#endif +#if CK_EXPERIMENTAL_PIPELINE_V2_INSTANCES + // pipeline v2, 1 wave + , + DeviceGemmMultipleD_Xdl_CShuffle< Col, Row, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 256, 256, 128, 32, 2, 2, 32, 32, 4, 2, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, S<8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v2>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Row, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 256, 256, 128, 32, 8, 8, 32, 32, 4, 2, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 8, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v2>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Row, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 256, 128, 256, 32, 2, 2, 32, 32, 2, 4, S<8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v2>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Row, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 256, 128, 256, 32, 8, 8, 32, 32, 2, 4, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v2>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Row, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 128, 128, 128, 32, 2, 2, 32, 32, 4, 2, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, 1, 1, S<1, 16, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v2>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Row, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 128, 128, 128, 32, 8, 8, 32, 32, 4, 2, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 8, 1, 1, 1, S<1, 16, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v2>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Row, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 256, 128, 128, 32, 2, 2, 32, 32, 2, 2, S<8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, S<8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v2>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Row, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 256, 128, 128, 32, 8, 8, 32, 32, 2, 2, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v2>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Row, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 128, 128, 64, 32, 2, 2, 32, 32, 2, 2, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, 1, 1, S<1, 32, 1, 4>, 8, LoopScheduler::Default, PipelineVersion::v2>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Row, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 128, 128, 64, 32, 8, 8, 32, 32, 2, 2, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, 1, 1, 1, S<1, 32, 1, 4>, 8, LoopScheduler::Default, PipelineVersion::v2>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Row, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 128, 64, 128, 32, 2, 2, 32, 32, 2, 2, S<8, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, 1, 1, S<1, 16, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v2>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Row, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 128, 64, 128, 32, 8, 8, 32, 32, 2, 2, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 8, 1, 1, 1, S<1, 16, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v2>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Row, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 256, 128, 64, 32, 2, 2, 32, 32, 2, 1, S<8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, S<16,16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v2>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Row, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 256, 128, 64, 32, 8, 8, 32, 32, 2, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v2>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Row, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 256, 64, 128, 32, 2, 2, 32, 32, 1, 2, S<16,16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, S<8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v2>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Row, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 256, 64, 128, 32, 8, 8, 32, 32, 1, 2, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v2> +#endif + // clang-format on + >; + +void add_device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_km_kn_mn_mn_instances( + std::vector>>& instances) +{ + add_device_operation_instances( + instances, device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_km_kn_mn_mn_instances{}); +} + +} // namespace instance +} // namespace device +} // namespace tensor_operation +} // namespace ck diff --git a/library/src/tensor_operation_instance/gpu/gemm_add_fastgelu/device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_km_nk_mn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_add_fastgelu/device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_km_nk_mn_mn_instance.cpp new file mode 100644 index 0000000000..6391d43f40 --- /dev/null +++ b/library/src/tensor_operation_instance/gpu/gemm_add_fastgelu/device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_km_nk_mn_mn_instance.cpp @@ -0,0 +1,109 @@ +// SPDX-License-Identifier: MIT +// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved. + +#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp" +#include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp" +#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp" +#include "ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_xdl_cshuffle.hpp" +#include "ck/utility/sequence.hpp" + +namespace ck { +namespace tensor_operation { +namespace device { +namespace instance { + +template +using S = ck::Sequence; + +static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default; + +// e = elementwise((a * b), d0, d1) +// outout: e[m, n] +// input: a[k, m], b[n, k], d0[m, n], d1[m, n] +using device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_km_nk_mn_mn_instances = std::tuple< + // clang-format off + //##############################| A| B| Ds| E| AData| BData| AccData| CShuffle| DsData| EData| A| B| CDE| GEMM| NumGemmK| Block| MPer| NPer| KPer| AK1| BK1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CShuffle| CShuffle| CBlockTransferClusterLengths| CBlockTransfer| LoopScheduler| Pipeline| + //##############################| Layout| Layout| Layout| Layout| Type| Type| Type| DataType| Type| Type| Elementwise| Elementwise| Elementwise| Specialization| Prefetch| Size| Block| Block| Block| | | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave| _MBlock_MWaveMPerXdl| ScalarPerVector| | | + //##############################| | | | | | | | | | | Operation| Operation| Operation| | Stage| | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| _NBlock_NWaveNPerXdl| _NWaveNPerXdl| | | + //##############################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | + // pipeline v1, 1 wave + DeviceGemmMultipleD_Xdl_CShuffle< Col, Col, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 256, 256, 128, 32, 2, 8, 32, 32, 4, 2, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Col, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 256, 256, 128, 32, 8, 8, 32, 32, 4, 2, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Col, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 256, 128, 256, 32, 2, 8, 32, 32, 2, 4, S<8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Col, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 256, 128, 256, 32, 8, 8, 32, 32, 2, 4, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Col, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 128, 128, 128, 32, 2, 8, 32, 32, 4, 2, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Col, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 128, 128, 128, 32, 8, 8, 32, 32, 4, 2, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Col, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 256, 128, 128, 32, 2, 8, 32, 32, 2, 2, S<8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Col, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 256, 128, 128, 32, 8, 8, 32, 32, 2, 2, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Col, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 128, 128, 64, 32, 2, 8, 32, 32, 2, 2, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 4>, 8, LoopScheduler::Default, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Col, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 128, 128, 64, 32, 8, 8, 32, 32, 2, 2, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 4>, 8, LoopScheduler::Default, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Col, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 128, 64, 128, 32, 2, 8, 32, 32, 2, 2, S<8, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Col, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 128, 64, 128, 32, 8, 8, 32, 32, 2, 2, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Col, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 256, 128, 64, 32, 2, 8, 32, 32, 2, 1, S<8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Col, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 256, 128, 64, 32, 8, 8, 32, 32, 2, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Col, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 256, 64, 128, 32, 2, 8, 32, 32, 1, 2, S<16,16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Col, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 256, 64, 128, 32, 8, 8, 32, 32, 1, 2, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1> +#if CK_EXPERIMENTAL_INTER_WAVE_INSTANCES + // pipeline v1, 2 waves + , + DeviceGemmMultipleD_Xdl_CShuffle< Col, Col, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 256, 256, 128, 32, 2, 8, 32, 32, 4, 2, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Interwave, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Col, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 256, 256, 128, 32, 8, 8, 32, 32, 4, 2, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Interwave, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Col, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 256, 128, 256, 32, 2, 8, 32, 32, 2, 4, S<8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Interwave, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Col, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 256, 128, 256, 32, 8, 8, 32, 32, 2, 4, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Interwave, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Col, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 128, 128, 128, 32, 2, 8, 32, 32, 4, 2, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 8>, 8, LoopScheduler::Interwave, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Col, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 128, 128, 128, 32, 8, 8, 32, 32, 4, 2, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 8>, 8, LoopScheduler::Interwave, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Col, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 256, 128, 128, 32, 2, 8, 32, 32, 2, 2, S<8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Interwave, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Col, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 256, 128, 128, 32, 8, 8, 32, 32, 2, 2, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Interwave, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Col, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 128, 128, 64, 32, 2, 8, 32, 32, 2, 2, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 4>, 8, LoopScheduler::Interwave, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Col, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 128, 128, 64, 32, 8, 8, 32, 32, 2, 2, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 4>, 8, LoopScheduler::Interwave, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Col, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 128, 64, 128, 32, 2, 8, 32, 32, 2, 2, S<8, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 8>, 8, LoopScheduler::Interwave, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Col, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 128, 64, 128, 32, 8, 8, 32, 32, 2, 2, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 8>, 8, LoopScheduler::Interwave, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Col, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 256, 128, 64, 32, 2, 8, 32, 32, 2, 1, S<8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Interwave, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Col, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 256, 128, 64, 32, 8, 8, 32, 32, 2, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Interwave, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Col, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 256, 64, 128, 32, 2, 8, 32, 32, 1, 2, S<16,16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Interwave, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Col, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 256, 64, 128, 32, 8, 8, 32, 32, 1, 2, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Interwave, PipelineVersion::v1> +#endif +#if CK_EXPERIMENTAL_PIPELINE_V2_INSTANCES + // pipeline v2, 1 wave + , + DeviceGemmMultipleD_Xdl_CShuffle< Col, Col, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 256, 256, 128, 32, 2, 8, 32, 32, 4, 2, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v2>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Col, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 256, 256, 128, 32, 8, 8, 32, 32, 4, 2, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v2>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Col, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 256, 128, 256, 32, 2, 8, 32, 32, 2, 4, S<8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v2>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Col, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 256, 128, 256, 32, 8, 8, 32, 32, 2, 4, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v2>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Col, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 128, 128, 128, 32, 2, 8, 32, 32, 4, 2, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v2>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Col, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 128, 128, 128, 32, 8, 8, 32, 32, 4, 2, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v2>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Col, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 256, 128, 128, 32, 2, 8, 32, 32, 2, 2, S<8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v2>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Col, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 256, 128, 128, 32, 8, 8, 32, 32, 2, 2, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v2>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Col, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 128, 128, 64, 32, 2, 8, 32, 32, 2, 2, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 4>, 8, LoopScheduler::Default, PipelineVersion::v2>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Col, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 128, 128, 64, 32, 8, 8, 32, 32, 2, 2, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 4>, 8, LoopScheduler::Default, PipelineVersion::v2>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Col, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 128, 64, 128, 32, 2, 8, 32, 32, 2, 2, S<8, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v2>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Col, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 128, 64, 128, 32, 8, 8, 32, 32, 2, 2, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v2>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Col, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 256, 128, 64, 32, 2, 8, 32, 32, 2, 1, S<8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v2>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Col, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 256, 128, 64, 32, 8, 8, 32, 32, 2, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v2>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Col, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 256, 64, 128, 32, 2, 8, 32, 32, 1, 2, S<16,16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v2>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Col, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 256, 64, 128, 32, 8, 8, 32, 32, 1, 2, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v2> +#endif + // clang-format on + >; + +void add_device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_km_nk_mn_mn_instances( + std::vector>>& instances) +{ + add_device_operation_instances( + instances, device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_km_nk_mn_mn_instances{}); +} + +} // namespace instance +} // namespace device +} // namespace tensor_operation +} // namespace ck diff --git a/library/src/tensor_operation_instance/gpu/gemm_add_fastgelu/device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_mk_kn_mn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_add_fastgelu/device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_mk_kn_mn_mn_instance.cpp new file mode 100644 index 0000000000..b43bbed60a --- /dev/null +++ b/library/src/tensor_operation_instance/gpu/gemm_add_fastgelu/device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_mk_kn_mn_mn_instance.cpp @@ -0,0 +1,109 @@ +// SPDX-License-Identifier: MIT +// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved. + +#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp" +#include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp" +#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp" +#include "ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_xdl_cshuffle.hpp" +#include "ck/utility/sequence.hpp" + +namespace ck { +namespace tensor_operation { +namespace device { +namespace instance { + +template +using S = ck::Sequence; + +static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default; + +// e = elementwise((a * b), d0, d1) +// outout: e[m, n] +// input: a[m, k], b[k, n], d0[m, n], d1[m, n] +using device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_mk_kn_mn_mn_instances = std::tuple< + // clang-format off + //##############################| A| B| Ds| E| AData| BData| AccData| CShuffle| DsData| EData| A| B| CDE| GEMM| NumGemmK| Block| MPer| NPer| KPer| AK1| BK1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CShuffle| CShuffle| CBlockTransferClusterLengths| CBlockTransfer| LoopScheduler| Pipeline| + //##############################| Layout| Layout| Layout| Layout| Type| Type| Type| DataType| Type| Type| Elementwise| Elementwise| Elementwise| Specialization| Prefetch| Size| Block| Block| Block| | | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave| _MBlock_MWaveMPerXdl| ScalarPerVector| | | + //##############################| | | | | | | | | | | Operation| Operation| Operation| | Stage| | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| _NBlock_NWaveNPerXdl| _NWaveNPerXdl| | | + //##############################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | + // pipeline v1, 1 wave + DeviceGemmMultipleD_Xdl_CShuffle< Row, Row, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 256, 256, 128, 32, 8, 2, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Row, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 256, 256, 128, 32, 8, 8, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Row, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 256, 128, 256, 32, 8, 2, 32, 32, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Row, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 256, 128, 256, 32, 8, 8, 32, 32, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Row, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 128, 128, 128, 32, 8, 2, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, 1, 1, S<1, 16, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Row, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 128, 128, 128, 32, 8, 8, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 8, 1, 1, 1, S<1, 16, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Row, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 256, 128, 128, 32, 8, 2, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Row, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 256, 128, 128, 32, 8, 8, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Row, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 128, 128, 64, 32, 8, 2, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<8, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, 1, 1, S<1, 32, 1, 4>, 8, LoopScheduler::Default, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Row, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 128, 128, 64, 32, 8, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, 1, 1, 1, S<1, 32, 1, 4>, 8, LoopScheduler::Default, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Row, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 128, 64, 128, 32, 8, 2, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, 1, 1, S<1, 16, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Row, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 128, 64, 128, 32, 8, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 8, 1, 1, 1, S<1, 16, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Row, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 256, 128, 64, 32, 8, 2, 32, 32, 2, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<16,16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Row, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 256, 128, 64, 32, 8, 8, 32, 32, 2, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Row, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 256, 64, 128, 32, 8, 2, 32, 32, 1, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Row, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 256, 64, 128, 32, 8, 8, 32, 32, 1, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1> +#if CK_EXPERIMENTAL_INTER_WAVE_INSTANCES + // pipeline v1, 2 waves + , + DeviceGemmMultipleD_Xdl_CShuffle< Row, Row, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 256, 256, 128, 32, 8, 2, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Interwave, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Row, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 256, 256, 128, 32, 8, 8, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Interwave, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Row, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 256, 128, 256, 32, 8, 2, 32, 32, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Interwave, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Row, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 256, 128, 256, 32, 8, 8, 32, 32, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Interwave, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Row, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 128, 128, 128, 32, 8, 2, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, 1, 1, S<1, 16, 1, 8>, 8, LoopScheduler::Interwave, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Row, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 128, 128, 128, 32, 8, 8, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 8, 1, 1, 1, S<1, 16, 1, 8>, 8, LoopScheduler::Interwave, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Row, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 256, 128, 128, 32, 8, 2, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Interwave, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Row, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 256, 128, 128, 32, 8, 8, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Interwave, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Row, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 128, 128, 64, 32, 8, 2, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<8, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, 1, 1, S<1, 32, 1, 4>, 8, LoopScheduler::Interwave, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Row, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 128, 128, 64, 32, 8, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, 1, 1, 1, S<1, 32, 1, 4>, 8, LoopScheduler::Interwave, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Row, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 128, 64, 128, 32, 8, 2, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, 1, 1, S<1, 16, 1, 8>, 8, LoopScheduler::Interwave, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Row, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 128, 64, 128, 32, 8, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 8, 1, 1, 1, S<1, 16, 1, 8>, 8, LoopScheduler::Interwave, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Row, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 256, 128, 64, 32, 8, 2, 32, 32, 2, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<16,16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Interwave, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Row, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 256, 128, 64, 32, 8, 8, 32, 32, 2, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Interwave, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Row, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 256, 64, 128, 32, 8, 2, 32, 32, 1, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Interwave, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Row, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 256, 64, 128, 32, 8, 8, 32, 32, 1, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Interwave, PipelineVersion::v1> +#endif +#if CK_EXPERIMENTAL_PIPELINE_V2_INSTANCES + // pipeline v2, 1 wave + , + DeviceGemmMultipleD_Xdl_CShuffle< Row, Row, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 256, 256, 128, 32, 8, 2, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v2>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Row, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 256, 256, 128, 32, 8, 8, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v2>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Row, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 256, 128, 256, 32, 8, 2, 32, 32, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v2>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Row, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 256, 128, 256, 32, 8, 8, 32, 32, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v2>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Row, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 128, 128, 128, 32, 8, 2, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, 1, 1, S<1, 16, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v2>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Row, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 128, 128, 128, 32, 8, 8, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 8, 1, 1, 1, S<1, 16, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v2>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Row, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 256, 128, 128, 32, 8, 2, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v2>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Row, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 256, 128, 128, 32, 8, 8, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v2>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Row, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 128, 128, 64, 32, 8, 2, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<8, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, 1, 1, S<1, 32, 1, 4>, 8, LoopScheduler::Default, PipelineVersion::v2>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Row, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 128, 128, 64, 32, 8, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, 1, 1, 1, S<1, 32, 1, 4>, 8, LoopScheduler::Default, PipelineVersion::v2>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Row, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 128, 64, 128, 32, 8, 2, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, 1, 1, S<1, 16, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v2>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Row, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 128, 64, 128, 32, 8, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 8, 1, 1, 1, S<1, 16, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v2>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Row, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 256, 128, 64, 32, 8, 2, 32, 32, 2, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<16,16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v2>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Row, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 256, 128, 64, 32, 8, 8, 32, 32, 2, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v2>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Row, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 256, 64, 128, 32, 8, 2, 32, 32, 1, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v2>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Row, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 256, 64, 128, 32, 8, 8, 32, 32, 1, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v2> +#endif + // clang-format on + >; + +void add_device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_mk_kn_mn_mn_instances( + std::vector>>& instances) +{ + add_device_operation_instances( + instances, device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_mk_kn_mn_mn_instances{}); +} + +} // namespace instance +} // namespace device +} // namespace tensor_operation +} // namespace ck diff --git a/library/src/tensor_operation_instance/gpu/gemm_add_fastgelu/device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_mk_nk_mn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_add_fastgelu/device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_mk_nk_mn_mn_instance.cpp new file mode 100644 index 0000000000..8411e9d08c --- /dev/null +++ b/library/src/tensor_operation_instance/gpu/gemm_add_fastgelu/device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_mk_nk_mn_mn_instance.cpp @@ -0,0 +1,100 @@ +// SPDX-License-Identifier: MIT +// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved. + +#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp" +#include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp" +#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp" +#include "ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_xdl_cshuffle.hpp" +#include "ck/utility/sequence.hpp" + +namespace ck { +namespace tensor_operation { +namespace device { +namespace instance { + +template +using S = ck::Sequence; + +static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default; + +// e = elementwise((a * b), d0, d1) +// outout: e[m, n] +// input: a[m, k], b[n, k], d0[m, n], d1[m ,n] +using device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_mk_nk_mn_mn_instances = std::tuple< + // clang-format off + //##############################| A| B| Ds| E| AData| BData| AccData| CShuffle| DsData| EData| A| B| CDE| GEMM| NumGemmK| Block| MPer| NPer| KPer| AK1| BK1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CShuffle| CShuffle| CBlockTransferClusterLengths| CBlockTransfer| LoopScheduler| Pipeline| + //##############################| Layout| Layout| Layout| Layout| Type| Type| Type| DataType| Type| Type| Elementwise| Elementwise| Elementwise| Specialization| Prefetch| Size| Block| Block| Block| | | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave| _MBlock_MWaveMPerXdl| ScalarPerVector| | | + //##############################| | | | | | | | | | | Operation| Operation| Operation| | Stage| | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| _NBlock_NWaveNPerXdl| _NWaveNPerXdl| | | + //##############################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | + // pipeline v1, 1 wave + DeviceGemmMultipleD_Xdl_CShuffle< Row, Col, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 256, 256, 128, 32, 8, 8, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Col, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 256, 128, 256, 32, 8, 8, 32, 32, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Col, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 128, 128, 128, 32, 8, 8, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Col, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 256, 128, 128, 32, 8, 8, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Col, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 128, 128, 64, 32, 8, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 4>, 8, LoopScheduler::Default, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Col, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 128, 64, 128, 32, 8, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Col, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 64, 64, 64, 32, 8, 8, 32, 32, 2, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 4>, 8, LoopScheduler::Default, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Col, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 256, 128, 64, 32, 8, 8, 32, 32, 2, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Col, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 256, 64, 128, 32, 8, 8, 32, 32, 1, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Col, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 128, 128, 32, 32, 8, 8, 32, 32, 2, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 4>, 8, LoopScheduler::Default, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Col, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 128, 32, 128, 32, 8, 8, 32, 32, 1, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Col, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 64, 64, 32, 32, 8, 8, 32, 32, 2, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 4>, 8, LoopScheduler::Default, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Col, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 64, 32, 64, 32, 8, 8, 32, 32, 1, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 4>, 8, LoopScheduler::Default, PipelineVersion::v1> +#if CK_EXPERIMENTAL_INTER_WAVE_INSTANCES + // pipeline v1, 2 waves + , + DeviceGemmMultipleD_Xdl_CShuffle< Row, Col, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 256, 256, 128, 32, 8, 8, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Interwave, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Col, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 256, 128, 256, 32, 8, 8, 32, 32, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Interwave, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Col, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 128, 128, 128, 32, 8, 8, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 8>, 8, LoopScheduler::Interwave, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Col, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 256, 128, 128, 32, 8, 8, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Interwave, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Col, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 128, 128, 64, 32, 8, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 4>, 8, LoopScheduler::Interwave, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Col, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 128, 64, 128, 32, 8, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 8>, 8, LoopScheduler::Interwave, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Col, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 64, 64, 64, 32, 8, 8, 32, 32, 2, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 4>, 8, LoopScheduler::Interwave, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Col, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 256, 128, 64, 32, 8, 8, 32, 32, 2, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Interwave, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Col, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 256, 64, 128, 32, 8, 8, 32, 32, 1, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Interwave, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Col, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 128, 128, 32, 32, 8, 8, 32, 32, 2, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 4>, 8, LoopScheduler::Interwave, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Col, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 128, 32, 128, 32, 8, 8, 32, 32, 1, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 8>, 8, LoopScheduler::Interwave, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Col, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 64, 64, 32, 32, 8, 8, 32, 32, 2, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 4>, 8, LoopScheduler::Interwave, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Col, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 64, 32, 64, 32, 8, 8, 32, 32, 1, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 4>, 8, LoopScheduler::Interwave, PipelineVersion::v1> +#endif +#if CK_EXPERIMENTAL_PIPELINE_V2_INSTANCES + // pipeline v2, 1 wave + , + DeviceGemmMultipleD_Xdl_CShuffle< Row, Col, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 256, 256, 128, 32, 8, 8, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v2>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Col, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 256, 128, 256, 32, 8, 8, 32, 32, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v2>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Col, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 128, 128, 128, 32, 8, 8, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v2>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Col, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 256, 128, 128, 32, 8, 8, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v2>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Col, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 128, 128, 64, 32, 8, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 4>, 8, LoopScheduler::Default, PipelineVersion::v2>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Col, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 128, 64, 128, 32, 8, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v2>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Col, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 64, 64, 64, 32, 8, 8, 32, 32, 2, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 4>, 8, LoopScheduler::Default, PipelineVersion::v2>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Col, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 256, 128, 64, 32, 8, 8, 32, 32, 2, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v2>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Col, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 256, 64, 128, 32, 8, 8, 32, 32, 1, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v2>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Col, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 128, 128, 32, 32, 8, 8, 32, 32, 2, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 4>, 8, LoopScheduler::Default, PipelineVersion::v2>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Col, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 128, 32, 128, 32, 8, 8, 32, 32, 1, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v2>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Col, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 64, 64, 32, 32, 8, 8, 32, 32, 2, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 4>, 8, LoopScheduler::Default, PipelineVersion::v2>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Col, Row_Tuple, Row, F16, F16, F32, F32, F16_Tuple, F16, PassThrough, PassThrough, AddFastGelu, GemmDefault, 1, 64, 32, 64, 32, 8, 8, 32, 32, 1, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 4>, 8, LoopScheduler::Default, PipelineVersion::v2> +#endif + // clang-format on + >; + +void add_device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_mk_nk_mn_mn_instances( + std::vector>>& instances) +{ + add_device_operation_instances( + instances, device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_mk_nk_mn_mn_instances{}); +} + +} // namespace instance +} // namespace device +} // namespace tensor_operation +} // namespace ck diff --git a/library/src/tensor_operation_instance/gpu/gemm_fastgelu/CMakeLists.txt b/library/src/tensor_operation_instance/gpu/gemm_fastgelu/CMakeLists.txt new file mode 100644 index 0000000000..17d27ab150 --- /dev/null +++ b/library/src/tensor_operation_instance/gpu/gemm_fastgelu/CMakeLists.txt @@ -0,0 +1,6 @@ +add_instance_library(device_gemm_fastgelu_instance + device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_km_kn_mn_instance.cpp + device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_km_nk_mn_instance.cpp + device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_mk_kn_mn_instance.cpp + device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_mk_nk_mn_instance.cpp +) diff --git a/library/src/tensor_operation_instance/gpu/gemm_fastgelu/device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_km_kn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_fastgelu/device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_km_kn_mn_instance.cpp new file mode 100644 index 0000000000..fb472ce369 --- /dev/null +++ b/library/src/tensor_operation_instance/gpu/gemm_fastgelu/device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_km_kn_mn_instance.cpp @@ -0,0 +1,109 @@ +// SPDX-License-Identifier: MIT +// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved. + +#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp" +#include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp" +#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp" +#include "ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_xdl_cshuffle.hpp" +#include "ck/utility/sequence.hpp" + +namespace ck { +namespace tensor_operation { +namespace device { +namespace instance { + +template +using S = ck::Sequence; + +static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default; + +// e = elementwise((a * b)) +// outout: e[m, n] +// input: a[k, m], b[k, n] +using device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_km_kn_mn_instances = std::tuple< + // clang-format off + //##############################| A| B| Ds| E| AData| BData| AccData| CShuffle| DsData| EData| A| B| CDE| GEMM| NumGemmK| Block| MPer| NPer| KPer| AK1| BK1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CShuffle| CShuffle| CBlockTransferClusterLengths| CBlockTransfer| LoopScheduler| Pipeline| + //##############################| Layout| Layout| Layout| Layout| Type| Type| Type| DataType| Type| Type| Elementwise| Elementwise| Elementwise| Specialization| Prefetch| Size| Block| Block| Block| | | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave| _MBlock_MWaveMPerXdl| ScalarPerVector| | | + //##############################| | | | | | | | | | | Operation| Operation| Operation| | Stage| | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| _NBlock_NWaveNPerXdl| _NWaveNPerXdl| | | + //##############################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | + // pipeline v1, 1 wave + DeviceGemmMultipleD_Xdl_CShuffle< Col, Row, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 256, 256, 128, 32, 2, 2, 32, 32, 4, 2, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, S<8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Row, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 256, 256, 128, 32, 8, 8, 32, 32, 4, 2, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 8, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Row, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 256, 128, 256, 32, 2, 2, 32, 32, 2, 4, S<8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Row, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 256, 128, 256, 32, 8, 8, 32, 32, 2, 4, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Row, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 128, 128, 128, 32, 2, 2, 32, 32, 4, 2, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, 1, 1, S<1, 16, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Row, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 128, 128, 128, 32, 8, 8, 32, 32, 4, 2, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 8, 1, 1, 1, S<1, 16, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Row, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 256, 128, 128, 32, 2, 2, 32, 32, 2, 2, S<8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, S<8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Row, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 256, 128, 128, 32, 8, 8, 32, 32, 2, 2, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Row, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 128, 128, 64, 32, 2, 2, 32, 32, 2, 2, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, 1, 1, S<1, 32, 1, 4>, 8, LoopScheduler::Default, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Row, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 128, 128, 64, 32, 8, 8, 32, 32, 2, 2, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, 1, 1, 1, S<1, 32, 1, 4>, 8, LoopScheduler::Default, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Row, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 128, 64, 128, 32, 2, 2, 32, 32, 2, 2, S<8, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, 1, 1, S<1, 16, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Row, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 128, 64, 128, 32, 8, 8, 32, 32, 2, 2, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 8, 1, 1, 1, S<1, 16, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Row, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 256, 128, 64, 32, 2, 2, 32, 32, 2, 1, S<8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, S<16,16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Row, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 256, 128, 64, 32, 8, 8, 32, 32, 2, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Row, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 256, 64, 128, 32, 2, 2, 32, 32, 1, 2, S<16,16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, S<8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Row, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 256, 64, 128, 32, 8, 8, 32, 32, 1, 2, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1> +#if CK_EXPERIMENTAL_INTER_WAVE_INSTANCES + // pipeline v1, 2 waves + , + DeviceGemmMultipleD_Xdl_CShuffle< Col, Row, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 256, 256, 128, 32, 2, 2, 32, 32, 4, 2, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, S<8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Interwave, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Row, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 256, 256, 128, 32, 8, 8, 32, 32, 4, 2, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 8, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Interwave, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Row, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 256, 128, 256, 32, 2, 2, 32, 32, 2, 4, S<8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Interwave, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Row, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 256, 128, 256, 32, 8, 8, 32, 32, 2, 4, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Interwave, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Row, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 128, 128, 128, 32, 2, 2, 32, 32, 4, 2, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, 1, 1, S<1, 16, 1, 8>, 8, LoopScheduler::Interwave, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Row, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 128, 128, 128, 32, 8, 8, 32, 32, 4, 2, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 8, 1, 1, 1, S<1, 16, 1, 8>, 8, LoopScheduler::Interwave, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Row, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 256, 128, 128, 32, 2, 2, 32, 32, 2, 2, S<8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, S<8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Interwave, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Row, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 256, 128, 128, 32, 8, 8, 32, 32, 2, 2, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Interwave, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Row, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 128, 128, 64, 32, 2, 2, 32, 32, 2, 2, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, 1, 1, S<1, 32, 1, 4>, 8, LoopScheduler::Interwave, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Row, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 128, 128, 64, 32, 8, 8, 32, 32, 2, 2, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, 1, 1, 1, S<1, 32, 1, 4>, 8, LoopScheduler::Interwave, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Row, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 128, 64, 128, 32, 2, 2, 32, 32, 2, 2, S<8, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, 1, 1, S<1, 16, 1, 8>, 8, LoopScheduler::Interwave, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Row, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 128, 64, 128, 32, 8, 8, 32, 32, 2, 2, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 8, 1, 1, 1, S<1, 16, 1, 8>, 8, LoopScheduler::Interwave, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Row, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 256, 128, 64, 32, 2, 2, 32, 32, 2, 1, S<8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, S<16,16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Interwave, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Row, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 256, 128, 64, 32, 8, 8, 32, 32, 2, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Interwave, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Row, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 256, 64, 128, 32, 2, 2, 32, 32, 1, 2, S<16,16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, S<8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Interwave, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Row, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 256, 64, 128, 32, 8, 8, 32, 32, 1, 2, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Interwave, PipelineVersion::v1> +#endif +#if CK_EXPERIMENTAL_PIPELINE_V2_INSTANCES + // pipeline v2, 1 wave + , + DeviceGemmMultipleD_Xdl_CShuffle< Col, Row, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 256, 256, 128, 32, 2, 2, 32, 32, 4, 2, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, S<8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v2>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Row, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 256, 256, 128, 32, 8, 8, 32, 32, 4, 2, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 8, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v2>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Row, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 256, 128, 256, 32, 2, 2, 32, 32, 2, 4, S<8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v2>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Row, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 256, 128, 256, 32, 8, 8, 32, 32, 2, 4, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v2>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Row, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 128, 128, 128, 32, 2, 2, 32, 32, 4, 2, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, 1, 1, S<1, 16, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v2>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Row, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 128, 128, 128, 32, 8, 8, 32, 32, 4, 2, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 8, 1, 1, 1, S<1, 16, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v2>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Row, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 256, 128, 128, 32, 2, 2, 32, 32, 2, 2, S<8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, S<8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v2>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Row, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 256, 128, 128, 32, 8, 8, 32, 32, 2, 2, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v2>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Row, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 128, 128, 64, 32, 2, 2, 32, 32, 2, 2, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, 1, 1, S<1, 32, 1, 4>, 8, LoopScheduler::Default, PipelineVersion::v2>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Row, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 128, 128, 64, 32, 8, 8, 32, 32, 2, 2, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, 1, 1, 1, S<1, 32, 1, 4>, 8, LoopScheduler::Default, PipelineVersion::v2>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Row, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 128, 64, 128, 32, 2, 2, 32, 32, 2, 2, S<8, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, 1, 1, S<1, 16, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v2>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Row, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 128, 64, 128, 32, 8, 8, 32, 32, 2, 2, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 8, 1, 1, 1, S<1, 16, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v2>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Row, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 256, 128, 64, 32, 2, 2, 32, 32, 2, 1, S<8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, S<16,16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v2>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Row, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 256, 128, 64, 32, 8, 8, 32, 32, 2, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v2>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Row, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 256, 64, 128, 32, 2, 2, 32, 32, 1, 2, S<16,16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, S<8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v2>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Row, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 256, 64, 128, 32, 8, 8, 32, 32, 1, 2, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v2> +#endif + // clang-format on + >; + +void add_device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_km_kn_mn_instances( + std::vector>>& instances) +{ + add_device_operation_instances( + instances, device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_km_kn_mn_instances{}); +} + +} // namespace instance +} // namespace device +} // namespace tensor_operation +} // namespace ck diff --git a/library/src/tensor_operation_instance/gpu/gemm_fastgelu/device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_km_nk_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_fastgelu/device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_km_nk_mn_instance.cpp new file mode 100644 index 0000000000..71f15e3157 --- /dev/null +++ b/library/src/tensor_operation_instance/gpu/gemm_fastgelu/device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_km_nk_mn_instance.cpp @@ -0,0 +1,109 @@ +// SPDX-License-Identifier: MIT +// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved. + +#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp" +#include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp" +#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp" +#include "ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_xdl_cshuffle.hpp" +#include "ck/utility/sequence.hpp" + +namespace ck { +namespace tensor_operation { +namespace device { +namespace instance { + +template +using S = ck::Sequence; + +static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default; + +// e = elementwise((a * b)) +// outout: e[m, n] +// input: a[k, m], b[k, n] +using device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_km_nk_mn_instances = std::tuple< + // clang-format off + //##############################| A| B| Ds| E| AData| BData| AccData| CShuffle| DsData| EData| A| B| CDE| GEMM| NumGemmK| Block| MPer| NPer| KPer| AK1| BK1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CShuffle| CShuffle| CBlockTransferClusterLengths| CBlockTransfer| LoopScheduler| Pipeline| + //##############################| Layout| Layout| Layout| Layout| Type| Type| Type| DataType| Type| Type| Elementwise| Elementwise| Elementwise| Specialization| Prefetch| Size| Block| Block| Block| | | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave| _MBlock_MWaveMPerXdl| ScalarPerVector| | | + //##############################| | | | | | | | | | | Operation| Operation| Operation| | Stage| | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| _NBlock_NWaveNPerXdl| _NWaveNPerXdl| | | + //##############################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | + // pipeline v1, 1 wave + DeviceGemmMultipleD_Xdl_CShuffle< Col, Col, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 256, 256, 128, 32, 2, 8, 32, 32, 4, 2, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Col, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 256, 256, 128, 32, 8, 8, 32, 32, 4, 2, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Col, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 256, 128, 256, 32, 2, 8, 32, 32, 2, 4, S<8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Col, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 256, 128, 256, 32, 8, 8, 32, 32, 2, 4, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Col, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 128, 128, 128, 32, 2, 8, 32, 32, 4, 2, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Col, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 128, 128, 128, 32, 8, 8, 32, 32, 4, 2, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Col, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 256, 128, 128, 32, 2, 8, 32, 32, 2, 2, S<8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Col, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 256, 128, 128, 32, 8, 8, 32, 32, 2, 2, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Col, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 128, 128, 64, 32, 2, 8, 32, 32, 2, 2, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 4>, 8, LoopScheduler::Default, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Col, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 128, 128, 64, 32, 8, 8, 32, 32, 2, 2, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 4>, 8, LoopScheduler::Default, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Col, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 128, 64, 128, 32, 2, 8, 32, 32, 2, 2, S<8, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Col, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 128, 64, 128, 32, 8, 8, 32, 32, 2, 2, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Col, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 256, 128, 64, 32, 2, 8, 32, 32, 2, 1, S<8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Col, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 256, 128, 64, 32, 8, 8, 32, 32, 2, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Col, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 256, 64, 128, 32, 2, 8, 32, 32, 1, 2, S<16,16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Col, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 256, 64, 128, 32, 8, 8, 32, 32, 1, 2, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1> +#if CK_EXPERIMENTAL_INTER_WAVE_INSTANCES + // pipeline v1, 2 waves + , + DeviceGemmMultipleD_Xdl_CShuffle< Col, Col, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 256, 256, 128, 32, 2, 8, 32, 32, 4, 2, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Interwave, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Col, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 256, 256, 128, 32, 8, 8, 32, 32, 4, 2, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Interwave, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Col, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 256, 128, 256, 32, 2, 8, 32, 32, 2, 4, S<8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Interwave, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Col, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 256, 128, 256, 32, 8, 8, 32, 32, 2, 4, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Interwave, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Col, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 128, 128, 128, 32, 2, 8, 32, 32, 4, 2, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 8>, 8, LoopScheduler::Interwave, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Col, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 128, 128, 128, 32, 8, 8, 32, 32, 4, 2, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 8>, 8, LoopScheduler::Interwave, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Col, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 256, 128, 128, 32, 2, 8, 32, 32, 2, 2, S<8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Interwave, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Col, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 256, 128, 128, 32, 8, 8, 32, 32, 2, 2, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Interwave, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Col, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 128, 128, 64, 32, 2, 8, 32, 32, 2, 2, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 4>, 8, LoopScheduler::Interwave, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Col, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 128, 128, 64, 32, 8, 8, 32, 32, 2, 2, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 4>, 8, LoopScheduler::Interwave, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Col, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 128, 64, 128, 32, 2, 8, 32, 32, 2, 2, S<8, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 8>, 8, LoopScheduler::Interwave, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Col, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 128, 64, 128, 32, 8, 8, 32, 32, 2, 2, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 8>, 8, LoopScheduler::Interwave, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Col, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 256, 128, 64, 32, 2, 8, 32, 32, 2, 1, S<8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Interwave, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Col, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 256, 128, 64, 32, 8, 8, 32, 32, 2, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Interwave, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Col, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 256, 64, 128, 32, 2, 8, 32, 32, 1, 2, S<16,16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Interwave, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Col, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 256, 64, 128, 32, 8, 8, 32, 32, 1, 2, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Interwave, PipelineVersion::v1> +#endif +#if CK_EXPERIMENTAL_PIPELINE_V2_INSTANCES + // pipeline v2, 1 wave + , + DeviceGemmMultipleD_Xdl_CShuffle< Col, Col, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 256, 256, 128, 32, 2, 8, 32, 32, 4, 2, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v2>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Col, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 256, 256, 128, 32, 8, 8, 32, 32, 4, 2, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v2>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Col, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 256, 128, 256, 32, 2, 8, 32, 32, 2, 4, S<8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v2>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Col, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 256, 128, 256, 32, 8, 8, 32, 32, 2, 4, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v2>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Col, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 128, 128, 128, 32, 2, 8, 32, 32, 4, 2, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v2>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Col, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 128, 128, 128, 32, 8, 8, 32, 32, 4, 2, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v2>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Col, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 256, 128, 128, 32, 2, 8, 32, 32, 2, 2, S<8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v2>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Col, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 256, 128, 128, 32, 8, 8, 32, 32, 2, 2, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v2>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Col, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 128, 128, 64, 32, 2, 8, 32, 32, 2, 2, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 4>, 8, LoopScheduler::Default, PipelineVersion::v2>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Col, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 128, 128, 64, 32, 8, 8, 32, 32, 2, 2, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 4>, 8, LoopScheduler::Default, PipelineVersion::v2>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Col, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 128, 64, 128, 32, 2, 8, 32, 32, 2, 2, S<8, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v2>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Col, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 128, 64, 128, 32, 8, 8, 32, 32, 2, 2, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v2>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Col, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 256, 128, 64, 32, 2, 8, 32, 32, 2, 1, S<8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v2>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Col, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 256, 128, 64, 32, 8, 8, 32, 32, 2, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v2>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Col, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 256, 64, 128, 32, 2, 8, 32, 32, 1, 2, S<16,16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v2>, + DeviceGemmMultipleD_Xdl_CShuffle< Col, Col, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 256, 64, 128, 32, 8, 8, 32, 32, 1, 2, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v2> +#endif + // clang-format on + >; + +void add_device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_km_nk_mn_instances( + std::vector>>& instances) +{ + add_device_operation_instances( + instances, device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_km_nk_mn_instances{}); +} + +} // namespace instance +} // namespace device +} // namespace tensor_operation +} // namespace ck diff --git a/library/src/tensor_operation_instance/gpu/gemm_fastgelu/device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_mk_kn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_fastgelu/device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_mk_kn_mn_instance.cpp new file mode 100644 index 0000000000..7579d9b469 --- /dev/null +++ b/library/src/tensor_operation_instance/gpu/gemm_fastgelu/device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_mk_kn_mn_instance.cpp @@ -0,0 +1,109 @@ +// SPDX-License-Identifier: MIT +// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved. + +#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp" +#include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp" +#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp" +#include "ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_xdl_cshuffle.hpp" +#include "ck/utility/sequence.hpp" + +namespace ck { +namespace tensor_operation { +namespace device { +namespace instance { + +template +using S = ck::Sequence; + +static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default; + +// e = elementwise((a * b)) +// outout: e[m, n] +// input: a[m, k], b[k, n] +using device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_mk_kn_mn_instances = std::tuple< + // clang-format off + //##############################| A| B| Ds| E| AData| BData| AccData| CShuffle| DsData| EData| A| B| CDE| GEMM| NumGemmK| Block| MPer| NPer| KPer| AK1| BK1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CShuffle| CShuffle| CBlockTransferClusterLengths| CBlockTransfer| LoopScheduler| Pipeline| + //##############################| Layout| Layout| Layout| Layout| Type| Type| Type| DataType| Type| Type| Elementwise| Elementwise| Elementwise| Specialization| Prefetch| Size| Block| Block| Block| | | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave| _MBlock_MWaveMPerXdl| ScalarPerVector| | | + //##############################| | | | | | | | | | | Operation| Operation| Operation| | Stage| | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| _NBlock_NWaveNPerXdl| _NWaveNPerXdl| | | + //##############################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | + // pipeline v1, 1 wave + DeviceGemmMultipleD_Xdl_CShuffle< Row, Row, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 256, 256, 128, 32, 8, 2, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Row, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 256, 256, 128, 32, 8, 8, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Row, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 256, 128, 256, 32, 8, 2, 32, 32, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Row, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 256, 128, 256, 32, 8, 8, 32, 32, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Row, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 128, 128, 128, 32, 8, 2, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, 1, 1, S<1, 16, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Row, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 128, 128, 128, 32, 8, 8, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 8, 1, 1, 1, S<1, 16, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Row, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 256, 128, 128, 32, 8, 2, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Row, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 256, 128, 128, 32, 8, 8, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Row, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 128, 128, 64, 32, 8, 2, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<8, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, 1, 1, S<1, 32, 1, 4>, 8, LoopScheduler::Default, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Row, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 128, 128, 64, 32, 8, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, 1, 1, 1, S<1, 32, 1, 4>, 8, LoopScheduler::Default, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Row, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 128, 64, 128, 32, 8, 2, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, 1, 1, S<1, 16, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Row, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 128, 64, 128, 32, 8, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 8, 1, 1, 1, S<1, 16, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Row, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 256, 128, 64, 32, 8, 2, 32, 32, 2, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<16,16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Row, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 256, 128, 64, 32, 8, 8, 32, 32, 2, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Row, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 256, 64, 128, 32, 8, 2, 32, 32, 1, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Row, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 256, 64, 128, 32, 8, 8, 32, 32, 1, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1> +#if CK_EXPERIMENTAL_INTER_WAVE_INSTANCES + // pipeline v1, 2 waves + , + DeviceGemmMultipleD_Xdl_CShuffle< Row, Row, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 256, 256, 128, 32, 8, 2, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Interwave, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Row, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 256, 256, 128, 32, 8, 8, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Interwave, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Row, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 256, 128, 256, 32, 8, 2, 32, 32, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Interwave, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Row, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 256, 128, 256, 32, 8, 8, 32, 32, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Interwave, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Row, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 128, 128, 128, 32, 8, 2, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, 1, 1, S<1, 16, 1, 8>, 8, LoopScheduler::Interwave, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Row, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 128, 128, 128, 32, 8, 8, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 8, 1, 1, 1, S<1, 16, 1, 8>, 8, LoopScheduler::Interwave, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Row, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 256, 128, 128, 32, 8, 2, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Interwave, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Row, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 256, 128, 128, 32, 8, 8, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Interwave, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Row, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 128, 128, 64, 32, 8, 2, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<8, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, 1, 1, S<1, 32, 1, 4>, 8, LoopScheduler::Interwave, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Row, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 128, 128, 64, 32, 8, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, 1, 1, 1, S<1, 32, 1, 4>, 8, LoopScheduler::Interwave, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Row, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 128, 64, 128, 32, 8, 2, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, 1, 1, S<1, 16, 1, 8>, 8, LoopScheduler::Interwave, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Row, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 128, 64, 128, 32, 8, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 8, 1, 1, 1, S<1, 16, 1, 8>, 8, LoopScheduler::Interwave, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Row, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 256, 128, 64, 32, 8, 2, 32, 32, 2, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<16,16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Interwave, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Row, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 256, 128, 64, 32, 8, 8, 32, 32, 2, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Interwave, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Row, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 256, 64, 128, 32, 8, 2, 32, 32, 1, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Interwave, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Row, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 256, 64, 128, 32, 8, 8, 32, 32, 1, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Interwave, PipelineVersion::v1> +#endif +#if CK_EXPERIMENTAL_PIPELINE_V2_INSTANCES + // pipeline v2, 1 wave + , + DeviceGemmMultipleD_Xdl_CShuffle< Row, Row, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 256, 256, 128, 32, 8, 2, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v2>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Row, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 256, 256, 128, 32, 8, 8, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v2>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Row, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 256, 128, 256, 32, 8, 2, 32, 32, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v2>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Row, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 256, 128, 256, 32, 8, 8, 32, 32, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v2>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Row, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 128, 128, 128, 32, 8, 2, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, 1, 1, S<1, 16, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v2>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Row, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 128, 128, 128, 32, 8, 8, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 8, 1, 1, 1, S<1, 16, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v2>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Row, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 256, 128, 128, 32, 8, 2, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v2>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Row, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 256, 128, 128, 32, 8, 8, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v2>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Row, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 128, 128, 64, 32, 8, 2, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<8, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, 1, 1, S<1, 32, 1, 4>, 8, LoopScheduler::Default, PipelineVersion::v2>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Row, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 128, 128, 64, 32, 8, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, 1, 1, 1, S<1, 32, 1, 4>, 8, LoopScheduler::Default, PipelineVersion::v2>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Row, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 128, 64, 128, 32, 8, 2, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, 1, 1, S<1, 16, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v2>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Row, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 128, 64, 128, 32, 8, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 8, 1, 1, 1, S<1, 16, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v2>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Row, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 256, 128, 64, 32, 8, 2, 32, 32, 2, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<16,16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v2>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Row, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 256, 128, 64, 32, 8, 8, 32, 32, 2, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v2>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Row, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 256, 64, 128, 32, 8, 2, 32, 32, 1, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v2>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Row, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 256, 64, 128, 32, 8, 8, 32, 32, 1, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v2> +#endif + // clang-format on + >; + +void add_device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_mk_kn_mn_instances( + std::vector>>& instances) +{ + add_device_operation_instances( + instances, device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_mk_kn_mn_instances{}); +} + +} // namespace instance +} // namespace device +} // namespace tensor_operation +} // namespace ck diff --git a/library/src/tensor_operation_instance/gpu/gemm_fastgelu/device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_mk_nk_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_fastgelu/device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_mk_nk_mn_instance.cpp new file mode 100644 index 0000000000..fbaa3b9417 --- /dev/null +++ b/library/src/tensor_operation_instance/gpu/gemm_fastgelu/device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_mk_nk_mn_instance.cpp @@ -0,0 +1,100 @@ +// SPDX-License-Identifier: MIT +// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved. + +#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp" +#include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp" +#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp" +#include "ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_xdl_cshuffle.hpp" +#include "ck/utility/sequence.hpp" + +namespace ck { +namespace tensor_operation { +namespace device { +namespace instance { + +template +using S = ck::Sequence; + +static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default; + +// e = elementwise((a * b)) +// outout: e[m, n] +// input: a[m, k], b[n, k] +using device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_mk_nk_mn_instances = std::tuple< + // clang-format off + //##############################| A| B| Ds| E| AData| BData| AccData| CShuffle| DsData| EData| A| B| CDE| GEMM| NumGemmK| Block| MPer| NPer| KPer| AK1| BK1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CShuffle| CShuffle| CBlockTransferClusterLengths| CBlockTransfer| LoopScheduler| Pipeline| + //##############################| Layout| Layout| Layout| Layout| Type| Type| Type| DataType| Type| Type| Elementwise| Elementwise| Elementwise| Specialization| Prefetch| Size| Block| Block| Block| | | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave| _MBlock_MWaveMPerXdl| ScalarPerVector| | | + //##############################| | | | | | | | | | | Operation| Operation| Operation| | Stage| | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| _NBlock_NWaveNPerXdl| _NWaveNPerXdl| | | + //##############################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | + // pipeline v1, 1 wave + DeviceGemmMultipleD_Xdl_CShuffle< Row, Col, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 256, 256, 128, 32, 8, 8, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Col, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 256, 128, 256, 32, 8, 8, 32, 32, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Col, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 128, 128, 128, 32, 8, 8, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Col, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 256, 128, 128, 32, 8, 8, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Col, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 128, 128, 64, 32, 8, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 4>, 8, LoopScheduler::Default, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Col, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 128, 64, 128, 32, 8, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Col, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 64, 64, 64, 32, 8, 8, 32, 32, 2, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 4>, 8, LoopScheduler::Default, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Col, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 256, 128, 64, 32, 8, 8, 32, 32, 2, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Col, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 256, 64, 128, 32, 8, 8, 32, 32, 1, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Col, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 128, 128, 32, 32, 8, 8, 32, 32, 2, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 4>, 8, LoopScheduler::Default, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Col, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 128, 32, 128, 32, 8, 8, 32, 32, 1, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Col, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 64, 64, 32, 32, 8, 8, 32, 32, 2, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 4>, 8, LoopScheduler::Default, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Col, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 64, 32, 64, 32, 8, 8, 32, 32, 1, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 4>, 8, LoopScheduler::Default, PipelineVersion::v1> +#if CK_EXPERIMENTAL_INTER_WAVE_INSTANCES + // pipeline v1, 2 waves + , + DeviceGemmMultipleD_Xdl_CShuffle< Row, Col, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 256, 256, 128, 32, 8, 8, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Interwave, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Col, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 256, 128, 256, 32, 8, 8, 32, 32, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Interwave, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Col, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 128, 128, 128, 32, 8, 8, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 8>, 8, LoopScheduler::Interwave, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Col, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 256, 128, 128, 32, 8, 8, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Interwave, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Col, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 128, 128, 64, 32, 8, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 4>, 8, LoopScheduler::Interwave, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Col, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 128, 64, 128, 32, 8, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 8>, 8, LoopScheduler::Interwave, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Col, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 64, 64, 64, 32, 8, 8, 32, 32, 2, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 4>, 8, LoopScheduler::Interwave, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Col, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 256, 128, 64, 32, 8, 8, 32, 32, 2, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Interwave, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Col, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 256, 64, 128, 32, 8, 8, 32, 32, 1, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Interwave, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Col, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 128, 128, 32, 32, 8, 8, 32, 32, 2, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 4>, 8, LoopScheduler::Interwave, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Col, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 128, 32, 128, 32, 8, 8, 32, 32, 1, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 8>, 8, LoopScheduler::Interwave, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Col, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 64, 64, 32, 32, 8, 8, 32, 32, 2, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 4>, 8, LoopScheduler::Interwave, PipelineVersion::v1>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Col, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 64, 32, 64, 32, 8, 8, 32, 32, 1, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 4>, 8, LoopScheduler::Interwave, PipelineVersion::v1> +#endif +#if CK_EXPERIMENTAL_PIPELINE_V2_INSTANCES + // pipeline v2, 1 wave + , + DeviceGemmMultipleD_Xdl_CShuffle< Row, Col, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 256, 256, 128, 32, 8, 8, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v2>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Col, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 256, 128, 256, 32, 8, 8, 32, 32, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v2>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Col, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 128, 128, 128, 32, 8, 8, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v2>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Col, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 256, 128, 128, 32, 8, 8, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v2>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Col, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 128, 128, 64, 32, 8, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 4>, 8, LoopScheduler::Default, PipelineVersion::v2>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Col, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 128, 64, 128, 32, 8, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v2>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Col, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 64, 64, 64, 32, 8, 8, 32, 32, 2, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 4>, 8, LoopScheduler::Default, PipelineVersion::v2>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Col, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 256, 128, 64, 32, 8, 8, 32, 32, 2, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v2>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Col, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 256, 64, 128, 32, 8, 8, 32, 32, 1, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v2>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Col, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 128, 128, 32, 32, 8, 8, 32, 32, 2, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 4>, 8, LoopScheduler::Default, PipelineVersion::v2>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Col, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 128, 32, 128, 32, 8, 8, 32, 32, 1, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 8>, 8, LoopScheduler::Default, PipelineVersion::v2>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Col, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 64, 64, 32, 32, 8, 8, 32, 32, 2, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 4>, 8, LoopScheduler::Default, PipelineVersion::v2>, + DeviceGemmMultipleD_Xdl_CShuffle< Row, Col, Empty_Tuple, Row, F16, F16, F32, F32, Empty_Tuple, F16, PassThrough, PassThrough, FastGelu, GemmDefault, 1, 64, 32, 64, 32, 8, 8, 32, 32, 1, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 4>, 8, LoopScheduler::Default, PipelineVersion::v2> +#endif + // clang-format on + >; + +void add_device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_mk_nk_mn_instances( + std::vector>>& instances) +{ + add_device_operation_instances( + instances, device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_mk_nk_mn_instances{}); +} + +} // namespace instance +} // namespace device +} // namespace tensor_operation +} // namespace ck