[DFT] Introduce a cuFFT backend for the DFT interface

CMakeLists.txt

@@ -46,15 +46,24 @@ option(ENABLE_MKLGPU_BACKEND "" ON)
 if(ENABLE_MKLCPU_BACKEND)
   option(ENABLE_MKLCPU_THREAD_TBB "" ON)
 endif()
-option(ENABLE_CUBLAS_BACKEND "" OFF)
-
-option(ENABLE_CUSOLVER_BACKEND "" OFF)
 
+# blas
+option(ENABLE_CUBLAS_BACKEND "" OFF)
 option(ENABLE_ROCBLAS_BACKEND "" OFF)
+option(ENABLE_NETLIB_BACKEND "" OFF)
+
+# rand
 option(ENABLE_CURAND_BACKEND "" OFF)
 option(ENABLE_ROCRAND_BACKEND "" OFF)
+
+# lapack
+option(ENABLE_CUSOLVER_BACKEND "" OFF)
 option(ENABLE_ROCSOLVER_BACKEND "" OFF)
-option(ENABLE_NETLIB_BACKEND "" OFF)
+
+# dft
+option(ENABlE_CUFFT_BACKEND "enable the cuFFT backend for DFT" OFF)
+
+
 set(ONEMKL_SYCL_IMPLEMENTATION "dpc++" CACHE STRING "Name of the SYCL compiler")
 set(HIP_TARGETS "" CACHE STRING "Target HIP architectures")
 
@@ -89,7 +98,8 @@ if(ENABLE_MKLCPU_BACKEND
   list(APPEND DOMAINS_LIST "rng")
 endif()
 if(ENABLE_MKLGPU_BACKEND
-        OR ENABLE_MKLCPU_BACKEND)
+        OR ENABLE_MKLCPU_BACKEND
+        OR ENABLE_CUFFT_BACKEND)
   list(APPEND DOMAINS_LIST "dft")
 endif()
 
@@ -99,8 +109,8 @@ if(CMAKE_CXX_COMPILER OR NOT ONEMKL_SYCL_IMPLEMENTATION STREQUAL "dpc++")
     string(REPLACE "\\" "/" CMAKE_CXX_COMPILER ${CMAKE_CXX_COMPILER})
   endif()
 else()
-  if(ENABLE_CUBLAS_BACKEND OR ENABLE_CURAND_BACKEND OR ENABLE_ROCBLAS_BACKEND
-	  OR ENABLE_ROCRAND_BACKEND OR ENABLE_ROCSOLVER_BACKEND)
+  if(ENABLE_CUBLAS_BACKEND OR ENABLE_CURAND_BACKEND OR ENABlE_CUFFT_BACKEND
+   OR ENABLE_ROCBLAS_BACKEND OR ENABLE_ROCRAND_BACKEND OR ENABLE_ROCSOLVER_BACKEND)
     set(CMAKE_CXX_COMPILER "clang++")
   elseif(ENABLE_MKLGPU_BACKEND)
     if(UNIX)

include/oneapi/mkl/detail/backends.hpp

@@ -36,6 +36,7 @@ enum class backend {
     netlib,
     rocblas,
     rocrand,
+    cufft,
     unsupported
 };
 

include/oneapi/mkl/detail/backends_table.hpp

@@ -84,6 +84,12 @@ static std::map<domain, std::map<device, std::vector<const char*>>> libraries =
           {
 #ifdef ENABLE_MKLGPU_BACKEND
               LIB_NAME("dft_mklgpu")
+#endif
+          } },
+        { device::nvidiagpu,
+          {
+#ifdef ENABLE_CUFFT_BACKEND
+              LIB_NAME("dft_cufft")
 #endif
           } } } },
 

include/oneapi/mkl/dft/detail/commit_impl.hpp

@@ -52,8 +52,8 @@ class commit_impl {
     virtual void* get_handle() noexcept = 0;
 
 private:
-    mkl::backend backend_;
     sycl::queue queue_;
+    mkl::backend backend_;
 };
 
 } // namespace oneapi::mkl::dft::detail

include/oneapi/mkl/dft/detail/cufft/onemkl_dft_cufft.hpp

@@ -0,0 +1,48 @@
+/*******************************************************************************
+* Copyright 2023 Intel Corporation
+*
+* Licensed under the Apache License, Version 2.0 (the "License");
+* you may not use this file except in compliance with the License.
+* You may obtain a copy of the License at
+*
+* http://www.apache.org/licenses/LICENSE-2.0
+*
+* Unless required by applicable law or agreed to in writing,
+* software distributed under the License is distributed on an "AS IS" BASIS,
+* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+* See the License for the specific language governing permissions
+* and limitations under the License.
+*
+*
+* SPDX-License-Identifier: Apache-2.0
+*******************************************************************************/
+
+#ifndef _ONEMKL_DFT_CUFFT_HPP_
+#define _ONEMKL_DFT_CUFFT_HPP_
+
+#if __has_include(<sycl/sycl.hpp>)
+#include <sycl/sycl.hpp>
+#else
+#include <CL/sycl.hpp>
+#endif
+
+#include "oneapi/mkl/detail/export.hpp"
+#include "oneapi/mkl/dft/detail/types_impl.hpp"
+
+namespace oneapi::mkl::dft {
+
+namespace detail {
+// Forward declarations
+class commit_impl;
+
+template <precision prec, domain dom>
+class descriptor;
+} // namespace detail
+
+namespace cufft {
+#include "oneapi/mkl/dft/detail/dft_ct.hxx"
+} // namespace cufft
+
+} // namespace oneapi::mkl::dft
+
+#endif // _ONEMKL_DFT_CUFFT_HPP_

include/oneapi/mkl/dft/detail/descriptor_impl.hpp

@@ -68,6 +68,10 @@ class descriptor {
     void commit(backend_selector<backend::mklgpu> selector);
 #endif
 
+#ifdef ENABLE_CUFFT_BACKEND
+    void commit(backend_selector<backend::cufft> selector);
+#endif
+
     const dft_values<prec, dom>& get_values() const noexcept {
         return values_;
     };

src/config.hpp.in

@@ -22,6 +22,7 @@
 
 #cmakedefine ENABLE_CUBLAS_BACKEND
 #cmakedefine ENABLE_CUSOLVER_BACKEND
+#cmakedefine ENABLE_CUFFT_BACKEND
 #cmakedefine ENABLE_ROCBLAS_BACKEND
 #cmakedefine ENABLE_ROCRAND_BACKEND
 #cmakedefine ENABLE_ROCSOLVER_BACKEND

src/dft/backends/CMakeLists.txt

@@ -24,3 +24,7 @@ endif()
 if(ENABLE_MKLCPU_BACKEND)
   add_subdirectory(mklcpu)
 endif()
+
+if(ENABLE_CUFFT_BACKEND)
+  add_subdirectory(cufft)
+endif()

src/dft/backends/cufft/CMakeLists.txt

@@ -0,0 +1,74 @@
+#===============================================================================
+# Copyright 2023 Intel Corporation
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+# http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing,
+# software distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions
+# and limitations under the License.
+#
+#
+# SPDX-License-Identifier: Apache-2.0
+#===============================================================================
+
+set(LIB_NAME onemkl_dft_cufft)
+set(LIB_OBJ ${LIB_NAME}_obj)
+
+find_package(CUDAToolkit REQUIRED)
+
+add_library(${LIB_NAME})
+add_library(${LIB_OBJ} OBJECT
+  descriptor.cpp
+  commit.cpp
+  forward.cpp
+  backward.cpp
+  compute_signature.cpp
+  $<$<BOOL:${BUILD_SHARED_LIBS}>: mkl_dft_cufft_wrappers.cpp>
+)
+
+target_include_directories(${LIB_OBJ}
+  PRIVATE ${PROJECT_SOURCE_DIR}/include
+          ${PROJECT_SOURCE_DIR}/src
+          ${CMAKE_BINARY_DIR}/bin
+          ${MKL_INCLUDE}
+)
+
+target_compile_options(${LIB_OBJ} PRIVATE ${ONEMKL_BUILD_COPT} ${MKL_COPT} -Wall)# -Wextra -Wno-conversion -Wno-shadow)
+
+target_link_libraries(${LIB_OBJ} PRIVATE CUDA::cudart CUDA::cufft)
+
+target_link_libraries(${LIB_OBJ} PUBLIC ONEMKL::SYCL::SYCL ${MKL_LINK_SYCL})
+
+set_target_properties(${LIB_OBJ} PROPERTIES
+  POSITION_INDEPENDENT_CODE ON
+)
+target_link_libraries(${LIB_NAME} PUBLIC ${LIB_OBJ})
+
+#Set oneMKL libraries as not transitive for dynamic
+if(BUILD_SHARED_LIBS)
+  set_target_properties(${LIB_NAME} PROPERTIES
+    INTERFACE_LINK_LIBRARIES ONEMKL::SYCL::SYCL
+  )
+endif()
+
+# Add major version to the library
+set_target_properties(${LIB_NAME} PROPERTIES
+  SOVERSION ${PROJECT_VERSION_MAJOR}
+)
+
+# Add dependencies rpath to the library
+list(APPEND CMAKE_BUILD_RPATH $<TARGET_FILE_DIR:${LIB_NAME}>)
+
+# Add the library to install package
+install(TARGETS ${LIB_OBJ} EXPORT oneMKLTargets)
+install(TARGETS ${LIB_NAME} EXPORT oneMKLTargets
+  RUNTIME DESTINATION bin
+  ARCHIVE DESTINATION lib
+  LIBRARY DESTINATION lib
+)