Backend rocfft
Collaboration diagram for Backend rocfft:
Classes | |
| struct | heffte::backend::default_backend< tag::gpu > |
| In CUDA mode, the default GPU backend is cufft. More... | |
| struct | heffte::backend::data_manipulator< tag::gpu > |
| Specialization for the data operations in CUDA mode. More... | |
| struct | heffte::backend::is_enabled< rocfft > |
| Indicate that the rocFFT backend has been enabled. More... | |
| struct | heffte::backend::is_enabled< rocfft_cos > |
| Indicate that the rocFFT backend has been enabled. More... | |
| struct | heffte::backend::is_enabled< rocfft_sin > |
| Indicate that the rocFFT backend has been enabled. More... | |
| struct | heffte::backend::device_instance< tag::gpu > |
| The CUDA backend uses a CUDA stream. More... | |
| struct | heffte::backend::buffer_traits< rocfft > |
| Defines the location type-tag and the cuda container. More... | |
| struct | heffte::backend::buffer_traits< rocfft_cos > |
| Defines the location type-tag and the cuda container. More... | |
| struct | heffte::backend::buffer_traits< rocfft_sin > |
| Defines the location type-tag and the cuda container. More... | |
| struct | heffte::plan_rocfft< precision_type, dir > |
| Plan for the r2c single precision transform. More... | |
| struct | heffte::plan_rocfft< std::complex< precision_type >, dir > |
| Plan for the single precision complex transform. More... | |
| class | heffte::rocfft_executor |
| Wrapper around the rocFFT API. More... | |
| class | heffte::rocfft_executor_r2c |
| Wrapper to rocFFT API for real-to-complex transform with shortening of the data. More... | |
| struct | heffte::one_dim_backend< backend::rocfft > |
| Helper struct that defines the types and creates instances of one-dimensional executors. More... | |
| struct | heffte::one_dim_backend< backend::rocfft_cos > |
| Helper struct that defines the types and creates instances of one-dimensional executors. More... | |
| struct | heffte::one_dim_backend< backend::rocfft_sin > |
| Helper struct that defines the types and creates instances of one-dimensional executors. More... | |
| struct | heffte::default_plan_options< backend::rocfft > |
| Sets the default options for the cufft backend. More... | |
| struct | heffte::default_plan_options< backend::rocfft_cos > |
| Sets the default options for the cufft backend. More... | |
| struct | heffte::default_plan_options< backend::rocfft_sin > |
| Sets the default options for the cufft backend. More... | |
| struct | heffte::backend::rocfft |
| Type-tag for the rocFFT backend. More... | |
| struct | heffte::backend::rocfft_cos |
| Type-tag for the Cosine Transform using the rocFFT backend. More... | |
| struct | heffte::backend::rocfft_sin |
| Type-tag for the Sine Transform using the rocFFT backend. More... | |
Functions | |
| void | heffte::rocm::check_error (hipError_t status, const char *function_name) |
| Checks the status of a ROCm command and in case of a failure, converts it to a C++ exception. | |
| void | heffte::rocm::check_error (rocfft_status status, const char *function_name) |
| Checks the status of a cufft command and in case of a failure, converts it to a C++ exception. | |
| template<typename precision_type , typename index > | |
| void | heffte::rocm::convert (hipStream_t stream, index num_entries, precision_type const source[], std::complex< precision_type > destination[]) |
| Convert real numbers to complex when both are located on the GPU device. More... | |
| template<typename precision_type , typename index > | |
| void | heffte::rocm::convert (hipStream_t stream, index num_entries, std::complex< precision_type > const source[], precision_type destination[]) |
| Convert complex numbers to real when both are located on the GPU device. More... | |
| template<typename scalar_type , typename index > | |
| void | heffte::rocm::scale_data (hipStream_t stream, index num_entries, scalar_type *data, double scale_factor) |
| Scales real data (double or float) by the scaling factor. | |
| template<typename scalar_type , typename index > | |
| void | heffte::rocm::direct_pack (hipStream_t stream, index nfast, index nmid, index nslow, index line_stride, index plane_stide, scalar_type const source[], scalar_type destination[]) |
| Performs a direct-pack operation for data sitting on the GPU device. More... | |
| template<typename scalar_type , typename index > | |
| void | heffte::rocm::direct_unpack (hipStream_t stream, index nfast, index nmid, index nslow, index line_stride, index plane_stide, scalar_type const source[], scalar_type destination[]) |
| Performs a direct-unpack operation for data sitting on the GPU device. More... | |
| template<typename scalar_type , typename index > | |
| void | heffte::rocm::transpose_unpack (hipStream_t stream, index nfast, index nmid, index nslow, index line_stride, index plane_stide, index buff_line_stride, index buff_plane_stride, int map0, int map1, int map2, scalar_type const source[], scalar_type destination[]) |
| Performs a transpose-unpack operation for data sitting on the GPU device. More... | |
| template<> | |
| std::string | heffte::backend::name< rocfft > () |
| Returns the human readable name of the rocFFT backend. | |
| template<> | |
| std::string | heffte::backend::name< rocfft_cos > () |
| Returns the human readable name of the rocFFT backend. | |
| template<> | |
| std::string | heffte::backend::name< rocfft_sin > () |
| Returns the human readable name of the rocFFT backend. | |
| plan_options | heffte::force_reorder (plan_options opts) |
| Forces the reorder logic for the ROCM r2c variant. | |
Detailed Description
Wrappers and template specializations related to the rocFFT backend. Requires CMake option:
-D Heffte_ENABLE_ROCM=ON
In addition to the rocFFT wrappers, this also includes a series of kernels for packing/unpacking/scaling the data, as well as a simple container that wraps around ROCM arrays for RAII style of resource management.
Function Documentation
◆ convert() [1/2]
template<typename precision_type , typename index >
| void heffte::rocm::convert | ( | hipStream_t | stream, |
| index | num_entries, | ||
| precision_type const | source[], | ||
| std::complex< precision_type > | destination[] | ||
| ) |
Convert real numbers to complex when both are located on the GPU device.
Launches a ROCM kernel.
◆ convert() [2/2]
template<typename precision_type , typename index >
| void heffte::rocm::convert | ( | hipStream_t | stream, |
| index | num_entries, | ||
| std::complex< precision_type > const | source[], | ||
| precision_type | destination[] | ||
| ) |
Convert complex numbers to real when both are located on the GPU device.
Launches a ROCM kernel.
◆ direct_pack()
template<typename scalar_type , typename index >
| void heffte::rocm::direct_pack | ( | hipStream_t | stream, |
| index | nfast, | ||
| index | nmid, | ||
| index | nslow, | ||
| index | line_stride, | ||
| index | plane_stide, | ||
| scalar_type const | source[], | ||
| scalar_type | destination[] | ||
| ) |
Performs a direct-pack operation for data sitting on the GPU device.
Launches a HIP kernel.
◆ direct_unpack()
template<typename scalar_type , typename index >
| void heffte::rocm::direct_unpack | ( | hipStream_t | stream, |
| index | nfast, | ||
| index | nmid, | ||
| index | nslow, | ||
| index | line_stride, | ||
| index | plane_stide, | ||
| scalar_type const | source[], | ||
| scalar_type | destination[] | ||
| ) |
Performs a direct-unpack operation for data sitting on the GPU device.
Launches a HIP kernel.
◆ transpose_unpack()
template<typename scalar_type , typename index >
| void heffte::rocm::transpose_unpack | ( | hipStream_t | stream, |
| index | nfast, | ||
| index | nmid, | ||
| index | nslow, | ||
| index | line_stride, | ||
| index | plane_stide, | ||
| index | buff_line_stride, | ||
| index | buff_plane_stride, | ||
| int | map0, | ||
| int | map1, | ||
| int | map2, | ||
| scalar_type const | source[], | ||
| scalar_type | destination[] | ||
| ) |
Performs a transpose-unpack operation for data sitting on the GPU device.
Launches a HIP kernel.
