From 72219330fd85b1271e714d4ba894d6d8e26340c9 Mon Sep 17 00:00:00 2001 From: Georgios Pinitas Date: Tue, 5 Jun 2018 14:56:06 +0100 Subject: COMPMID-1145: (API) Introduce prepare() stage (NEON/CL/GLES) Change-Id: I5b46764f9c3154ec3e3b9c951cc9e6dfbcb81dfb Reviewed-on: https://eu-gerrit-1.euhpc.arm.com/134255 Reviewed-by: Anthony Barbier Tested-by: Jenkins Reviewed-by: Pablo Tello Reviewed-by: Michele DiGiorgio --- .../CL/functions/CLLocallyConnectedLayer.cpp | 33 +++++++++++++--------- 1 file changed, 19 insertions(+), 14 deletions(-) (limited to 'src/runtime/CL/functions/CLLocallyConnectedLayer.cpp') diff --git a/src/runtime/CL/functions/CLLocallyConnectedLayer.cpp b/src/runtime/CL/functions/CLLocallyConnectedLayer.cpp index 31d5cd5a7e..d15e5dfa3d 100644 --- a/src/runtime/CL/functions/CLLocallyConnectedLayer.cpp +++ b/src/runtime/CL/functions/CLLocallyConnectedLayer.cpp @@ -73,7 +73,7 @@ void calculate_shapes(const ITensorInfo *input, const ITensorInfo *weights, cons CLLocallyConnectedLayer::CLLocallyConnectedLayer(std::shared_ptr memory_manager) : _memory_group(std::move(memory_manager)), _input_im2col_kernel(), _weights_reshape_kernel(), _mm_kernel(), _output_col2im_kernel(), _input_im2col_reshaped(), _weights_reshaped(), _gemm_output(), - _is_first_run(false), _original_weights(nullptr) + _is_prepared(false), _original_weights(nullptr) { } @@ -128,7 +128,7 @@ void CLLocallyConnectedLayer::configure(const ICLTensor *input, const ICLTensor bool _has_bias = (biases != nullptr); _original_weights = weights; - _is_first_run = true; + _is_prepared = false; const unsigned int kernel_width = weights->info()->dimension(0); const unsigned int kernel_height = weights->info()->dimension(1); @@ -160,7 +160,6 @@ void CLLocallyConnectedLayer::configure(const ICLTensor *input, const ICLTensor _output_col2im_kernel.configure(&_gemm_output, output, std::make_pair(conv_w, conv_h)); // Allocate intermediate tensors - _weights_reshaped.allocator()->allocate(); _input_im2col_reshaped.allocator()->allocate(); _gemm_output.allocator()->allocate(); @@ -169,17 +168,7 @@ void CLLocallyConnectedLayer::configure(const ICLTensor *input, const ICLTensor void CLLocallyConnectedLayer::run() { - // Run weights reshaping (Runs once for every configure) - if(_is_first_run) - { - ARM_COMPUTE_ERROR_ON(!_original_weights->is_used()); - - _is_first_run = false; - CLScheduler::get().enqueue(_weights_reshape_kernel); - - // Mark original weights tensor as unused - _original_weights->mark_as_unused(); - } + prepare(); _memory_group.acquire(); @@ -194,3 +183,19 @@ void CLLocallyConnectedLayer::run() _memory_group.release(); } + +void CLLocallyConnectedLayer::prepare() +{ + if(!_is_prepared) + { + ARM_COMPUTE_ERROR_ON(!_original_weights->is_used()); + + // Run weights reshaping and mark original weights tensor as unused + _weights_reshaped.allocator()->allocate(); + CLScheduler::get().enqueue(_weights_reshape_kernel); + _original_weights->mark_as_unused(); + + CLScheduler::get().queue().finish(); + _is_prepared = true; + } +} -- cgit v1.2.1