diff options
author | Georgios Pinitas <georgios.pinitas@arm.com> | 2017-09-15 19:06:47 +0100 |
---|---|---|
committer | Anthony Barbier <anthony.barbier@arm.com> | 2018-11-02 16:35:24 +0000 |
commit | 8a94e7cec7b09a417a278425e2b56e7af5bf45d9 (patch) | |
tree | e952f39903d4624bbd6445c9cc6c7dbcc1114026 /src/runtime/CL/functions/CLConvolution.cpp | |
parent | 658039bc4e06be34272eccf559a516a6b52f75f5 (diff) | |
download | ComputeLibrary-8a94e7cec7b09a417a278425e2b56e7af5bf45d9.tar.gz |
COMPMID-534: Add MemoryManager support in OpenCL functions
Adds support for:
-CLConvolution
-CLGEMM
-CLGEMMLowp
-CLHOGDescriptor
-CLHOGGradient
-CLHOGMultiDetection
-CLL2Normalize
-CLLocallyConnectedLayer
-CLOpticalFlow
-CLReductionOperation
Change-Id: Ib13354d274ccf32ae933f3fbbad3ac3896cfd3bd
Reviewed-on: http://mpd-gerrit.cambridge.arm.com/87938
Tested-by: Kaizen <jeremy.johnson+kaizengerrit@arm.com>
Reviewed-by: Pablo Tello <pablo.tello@arm.com>
Diffstat (limited to 'src/runtime/CL/functions/CLConvolution.cpp')
-rw-r--r-- | src/runtime/CL/functions/CLConvolution.cpp | 11 |
1 files changed, 9 insertions, 2 deletions
diff --git a/src/runtime/CL/functions/CLConvolution.cpp b/src/runtime/CL/functions/CLConvolution.cpp index 641044451d..a9b086773c 100644 --- a/src/runtime/CL/functions/CLConvolution.cpp +++ b/src/runtime/CL/functions/CLConvolution.cpp @@ -47,8 +47,8 @@ void CLConvolution3x3::configure(ICLTensor *input, ICLTensor *output, const int1 } template <unsigned int matrix_size> -CLConvolutionSquare<matrix_size>::CLConvolutionSquare() - : _tmp(), _is_separable(false), _kernel_hor(), _kernel_vert(), _kernel(), _border_handler() +CLConvolutionSquare<matrix_size>::CLConvolutionSquare(std::shared_ptr<IMemoryManager> memory_manager) + : _memory_group(std::move(memory_manager)), _tmp(), _is_separable(false), _kernel_hor(), _kernel_vert(), _kernel(), _border_handler() { } @@ -66,6 +66,9 @@ void CLConvolutionSquare<matrix_size>::configure(ICLTensor *input, ICLTensor *ou std::pair<DataType, DataType> type_pair = data_type_for_convolution(conv_col, conv_row, matrix_size); _tmp.allocator()->init(TensorInfo(input->info()->tensor_shape(), 1, type_pair.first)); + // Manage intermediate buffers + _memory_group.manage(&_tmp); + if(scale == 0) { scale = calculate_matrix_scale(conv, matrix_size); @@ -92,8 +95,12 @@ void CLConvolutionSquare<matrix_size>::run() if(_is_separable) { + _memory_group.acquire(); + CLScheduler::get().enqueue(_kernel_hor, false); CLScheduler::get().enqueue(_kernel_vert); + + _memory_group.release(); } else { |