From bef7fa27b0d231a8649952f60808132d109b6345 Mon Sep 17 00:00:00 2001 From: Sang-Hoon Park Date: Wed, 21 Oct 2020 15:58:54 +0100 Subject: COMPMID-3639: (3RDPARTY_UPDATE) Move CL kernels to src Change-Id: I10d27db788e5086adae1841e3e2441cd9b76ef84 Signed-off-by: Sang-Hoon Park Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/4310 Reviewed-by: Georgios Pinitas Comments-Addressed: Arm Jenkins Tested-by: Arm Jenkins --- src/runtime/CL/functions/CLLSTMLayer.cpp | 57 +++++++++++++++++++++----------- 1 file changed, 38 insertions(+), 19 deletions(-) (limited to 'src/runtime/CL/functions/CLLSTMLayer.cpp') diff --git a/src/runtime/CL/functions/CLLSTMLayer.cpp b/src/runtime/CL/functions/CLLSTMLayer.cpp index 058b6027c2..b095c06535 100644 --- a/src/runtime/CL/functions/CLLSTMLayer.cpp +++ b/src/runtime/CL/functions/CLLSTMLayer.cpp @@ -29,6 +29,22 @@ #include "arm_compute/core/utils/misc/ShapeCalculator.h" #include "arm_compute/core/utils/quantization/AsymmHelpers.h" #include "arm_compute/runtime/CL/CLScheduler.h" +#include "src/core/CL/kernels/CLCopyKernel.h" +#include "src/core/CL/kernels/CLDepthConvertLayerKernel.h" +#include "src/core/CL/kernels/CLFillBorderKernel.h" +#include "src/core/CL/kernels/CLGEMMLowpMatrixMultiplyNativeKernel.h" +#include "src/core/CL/kernels/CLGEMMLowpMatrixMultiplyReshapedOnlyRHSKernel.h" +#include "src/core/CL/kernels/CLGEMMLowpOffsetContributionKernel.h" +#include "src/core/CL/kernels/CLGEMMLowpOffsetContributionOutputStageKernel.h" +#include "src/core/CL/kernels/CLGEMMLowpReductionKernel.h" +#include "src/core/CL/kernels/CLGEMMMatrixMultiplyKernel.h" +#include "src/core/CL/kernels/CLGEMMMatrixMultiplyReshapedKernel.h" +#include "src/core/CL/kernels/CLGEMMMatrixMultiplyReshapedOnlyRHSKernel.h" +#include "src/core/CL/kernels/CLGEMMReshapeLHSMatrixKernel.h" +#include "src/core/CL/kernels/CLGEMMReshapeRHSMatrixKernel.h" +#include "src/core/CL/kernels/CLMemsetKernel.h" +#include "src/core/CL/kernels/CLTransposeKernel.h" +#include "support/MemorySupport.h" namespace arm_compute { @@ -37,20 +53,23 @@ using namespace arm_compute::utils::info_helpers; CLLSTMLayer::CLLSTMLayer(std::shared_ptr memory_manager) : _memory_group(std::move(memory_manager)), _fully_connected_input_gate(), _accum_input_gate1(), _subtract_input_gate(), _pixelwise_mul_input_gate(), _activation_input_gate(), - _fully_connected_forget_gate(), _accum_forget_gate1(), _pixelwise_mul_forget_gate(), _activation_forget_gate(), _fully_connected_cell_state(), _gemm_cell_state1(), _transpose_cell_state(), - _accum_cell_state1(), _accum_cell_state2(), _pixelwise_mul_cell_state1(), _activation_cell_state(), _cell_clip(), _pixelwise_mul_cell_state2(), _fully_connected_output(), - _pixelwise_mul_output_state1(), _accum_output1(), _activation_output(), _activation_output_state(), _pixelwise_mul_output_state2(), _fully_connected_output_state(), _projection_clip(), - _copy_cell_state(), _copy_output(), _concat_scratch_buffer(), _concat_inputs_forget_gate(), _concat_weights_forget_gate(), _concat_weights_input_gate(), _concat_weights_output(), - _ones_memset_kernel(), _mean_std_norm_input_gate(), _pixelwise_mul_input_gate_coeff(), _accum_input_gate_bias(), _mean_std_norm_forget_gate(), _pixelwise_mul_forget_gate_coeff(), - _accum_forget_gate_bias(), _mean_std_norm_cell_gate(), _pixelwise_mul_cell_gate_coeff(), _accum_cell_gate_bias(), _mean_std_norm_output_gate(), _pixelwise_mul_output_gate_coeff(), - _accum_output_gate_bias(), _input_gate_out1(), _input_gate_out2(), _input_gate_out3(), _input_gate_out4(), _forget_gate_out1(), _forget_gate_out2(), _forget_gate_out3(), _forget_gate_out4(), - _forget_gate_out5(), _forget_gate_out6(), _cell_state_out1(), _cell_state_out2(), _cell_state_out3(), _cell_state_out4(), _cell_state_out5(), _output1(), _output2(), _output3(), _output4(), - _cell_state_activation(), _output_state1(), _ones(), _input_layer_norm_out1(), _input_layer_norm_out2(), _forget_layer_norm_out1(), _forget_layer_norm_out2(), _cell_layer_norm_out1(), - _cell_layer_norm_out2(), _output_layer_norm_out1(), _output_layer_norm_out2(), _run_peephole_opt(false), _run_cifg_opt(false), _perform_cell_clipping(false), _has_projection_weights(false), - _perform_projection_clipping(false), _is_prepared(false), _is_layer_norm_lstm(false) + _fully_connected_forget_gate(), _accum_forget_gate1(), _pixelwise_mul_forget_gate(), _activation_forget_gate(), _fully_connected_cell_state(), _gemm_cell_state1(), + _transpose_cell_state(support::cpp14::make_unique()), _accum_cell_state1(), _accum_cell_state2(), _pixelwise_mul_cell_state1(), _activation_cell_state(), _cell_clip(), + _pixelwise_mul_cell_state2(), _fully_connected_output(), _pixelwise_mul_output_state1(), _accum_output1(), _activation_output(), _activation_output_state(), _pixelwise_mul_output_state2(), + _fully_connected_output_state(), _projection_clip(), _copy_cell_state(support::cpp14::make_unique()), _copy_output(support::cpp14::make_unique()), _concat_scratch_buffer(), + _concat_inputs_forget_gate(), _concat_weights_forget_gate(), _concat_weights_input_gate(), _concat_weights_output(), _ones_memset_kernel(support::cpp14::make_unique()), + _mean_std_norm_input_gate(), _pixelwise_mul_input_gate_coeff(), _accum_input_gate_bias(), _mean_std_norm_forget_gate(), _pixelwise_mul_forget_gate_coeff(), _accum_forget_gate_bias(), + _mean_std_norm_cell_gate(), _pixelwise_mul_cell_gate_coeff(), _accum_cell_gate_bias(), _mean_std_norm_output_gate(), _pixelwise_mul_output_gate_coeff(), _accum_output_gate_bias(), _input_gate_out1(), + _input_gate_out2(), _input_gate_out3(), _input_gate_out4(), _forget_gate_out1(), _forget_gate_out2(), _forget_gate_out3(), _forget_gate_out4(), _forget_gate_out5(), _forget_gate_out6(), + _cell_state_out1(), _cell_state_out2(), _cell_state_out3(), _cell_state_out4(), _cell_state_out5(), _output1(), _output2(), _output3(), _output4(), _cell_state_activation(), _output_state1(), _ones(), + _input_layer_norm_out1(), _input_layer_norm_out2(), _forget_layer_norm_out1(), _forget_layer_norm_out2(), _cell_layer_norm_out1(), _cell_layer_norm_out2(), _output_layer_norm_out1(), + _output_layer_norm_out2(), _run_peephole_opt(false), _run_cifg_opt(false), _perform_cell_clipping(false), _has_projection_weights(false), _perform_projection_clipping(false), _is_prepared(false), + _is_layer_norm_lstm(false) { } +CLLSTMLayer::~CLLSTMLayer() = default; + void CLLSTMLayer::configure(const ICLTensor *input, const ICLTensor *input_to_forget_weights, const ICLTensor *input_to_cell_weights, const ICLTensor *input_to_output_weights, const ICLTensor *recurrent_to_forget_weights, const ICLTensor *recurrent_to_cell_weights, const ICLTensor *recurrent_to_output_weights, @@ -172,7 +191,7 @@ void CLLSTMLayer::configure(const CLCompileContext &compile_context, const ICLTe { _memory_group.manage(&_input_gate_out1); _ones.allocator()->init(TensorInfo(cell_state_shape, 1, input->info()->data_type())); - _ones_memset_kernel.configure(compile_context, &_ones, PixelValue(1, _ones.info()->data_type())); + _ones_memset_kernel->configure(compile_context, &_ones, PixelValue(1, _ones.info()->data_type())); _subtract_input_gate.configure(compile_context, &_ones, forget_gate_out, &_input_gate_out1, ConvertPolicy::SATURATE); _ones.allocator()->allocate(); _run_cifg_opt = true; @@ -241,7 +260,7 @@ void CLLSTMLayer::configure(const CLCompileContext &compile_context, const ICLTe _memory_group.manage(&_cell_state_out1); _fully_connected_cell_state.configure(compile_context, input, input_to_cell_weights, (_is_layer_norm_lstm) ? nullptr : cell_bias, &_cell_state_out1); _memory_group.manage(&_cell_state_out2); - _transpose_cell_state.configure(compile_context, recurrent_to_cell_weights, &_cell_state_out2); + _transpose_cell_state->configure(compile_context, recurrent_to_cell_weights, &_cell_state_out2); _memory_group.manage(&_cell_state_out3); _gemm_cell_state1.configure(compile_context, output_state_in, &_cell_state_out2, nullptr, &_cell_state_out3, 1.f, 0.f); _cell_state_out2.allocator()->allocate(); @@ -367,8 +386,8 @@ void CLLSTMLayer::configure(const CLCompileContext &compile_context, const ICLTe } // Copy cell state and output - _copy_cell_state.configure(compile_context, &_cell_state_out1, cell_state_out); - _copy_output.configure(compile_context, output_state_out, output); + _copy_cell_state->configure(compile_context, &_cell_state_out1, cell_state_out); + _copy_output->configure(compile_context, output_state_out, output); // Vector for holding the tensors to store in scratch buffer std::vector scratch_inputs; @@ -642,7 +661,7 @@ void CLLSTMLayer::run() if(_run_cifg_opt) { - CLScheduler::get().enqueue(_ones_memset_kernel); + CLScheduler::get().enqueue(*_ones_memset_kernel); _subtract_input_gate.run(); } else @@ -665,7 +684,7 @@ void CLLSTMLayer::run() } _fully_connected_cell_state.run(); - CLScheduler::get().enqueue(_transpose_cell_state); + CLScheduler::get().enqueue(*_transpose_cell_state); _gemm_cell_state1.run(); _accum_cell_state1.run(); if(_is_layer_norm_lstm) @@ -711,8 +730,8 @@ void CLLSTMLayer::run() } } - CLScheduler::get().enqueue(_copy_cell_state); - CLScheduler::get().enqueue(_copy_output); + CLScheduler::get().enqueue(*_copy_cell_state); + CLScheduler::get().enqueue(*_copy_output); _concat_scratch_buffer.run(); } -- cgit v1.2.1