diff options
| author | Pablo Tello <pablo.tello@arm.com> | 2017-12-15 09:48:59 +0000 |
|---|---|---|
| committer | Anthony Barbier <anthony.barbier@arm.com> | 2018-11-02 16:42:33 +0000 |
| commit | 02541fb21eca5574fcce012973774a6f213877ee (patch) | |
| tree | fb3b2652cfab65119f9814f4167a10b030e5f307 /src/core/NEON/kernels/NEWinogradLayerKernel.cpp | |
| parent | d064389293e4a71781984b2b24f3d44964812949 (diff) | |
| download | ComputeLibrary-02541fb21eca5574fcce012973774a6f213877ee.tar.gz | |
COMPMID-719: NEWinogradLayer reordering using NEPermute.
Input reordering from NCHW to NHWC
Output reordering from NHWC to NCHW
Weights reordering from [Ofm x Ifm x Height x Width] to [Height x Width x Ifm x Ofm]
Change-Id: I85aabedb1f9c13700bc4919eb3130f4d4bd0b465
Reviewed-on: https://eu-gerrit-1.euhpc.arm.com/113631
Reviewed-by: Anthony Barbier <anthony.barbier@arm.com>
Tested-by: Jenkins <bsgcomp@arm.com>
Diffstat (limited to 'src/core/NEON/kernels/NEWinogradLayerKernel.cpp')
| -rw-r--r-- | src/core/NEON/kernels/NEWinogradLayerKernel.cpp | 19 |
1 files changed, 7 insertions, 12 deletions
diff --git a/src/core/NEON/kernels/NEWinogradLayerKernel.cpp b/src/core/NEON/kernels/NEWinogradLayerKernel.cpp index fe633368c0..eaf77e6253 100644 --- a/src/core/NEON/kernels/NEWinogradLayerKernel.cpp +++ b/src/core/NEON/kernels/NEWinogradLayerKernel.cpp @@ -108,30 +108,25 @@ size_t NEWinogradLayerKernel::get_kernel_transform_working_size(const KernelShap } NEWinogradLayerKernel::NEWinogradLayerKernel() - : _convolver(nullptr), _output(nullptr) + : _convolver(nullptr) { } -void NEWinogradLayerKernel::configure(ITensor *output, Winograd3x3F32 *convolver) +void NEWinogradLayerKernel::configure(Winograd3x3F32 *convolver) { - ARM_COMPUTE_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(output, 1, DataType::F32); + ARM_COMPUTE_ERROR_ON_NULLPTR(convolver); _convolver = convolver; - Window win = calculate_max_window(*output->info()); + Window win; + win.set(Window::DimX, Window::Dimension(0, 15, 1)); INEKernel::configure(win); } void NEWinogradLayerKernel::run(const Window &window, const ThreadInfo &info) { - ARM_COMPUTE_UNUSED(window); ARM_COMPUTE_UNUSED(info); ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(this); - ARM_COMPUTE_ERROR_ON_INVALID_SUBWINDOW(INEKernel::window(), window); - ARM_COMPUTE_ERROR_ON(info.num_threads < 1); - const size_t tid = info.thread_id; - const size_t num_threads = std::min(info.num_threads, 16); - const size_t num_gemms_per_thread = 16 / num_threads; - const size_t first_gemm = tid * num_gemms_per_thread; - const size_t last_gemm = (tid == (num_threads - 1)) ? 15 : first_gemm + num_gemms_per_thread - 1; + const size_t first_gemm = window.x().start(); + const size_t last_gemm = window.x().end(); _convolver->_pimpl->convolver.execute(first_gemm, last_gemm); } } // namespace arm_compute |