/*
- * Copyright (c) 2016, 2017 ARM Limited.
+ * Copyright (c) 2016, 2018 ARM Limited.
*
* SPDX-License-Identifier: MIT
*
#include "arm_compute/runtime/NEON/NEFunctions.h"
#include "arm_compute/core/Types.h"
+#include "arm_compute/runtime/Allocator.h"
+#include "arm_compute/runtime/BlobLifetimeManager.h"
+#include "arm_compute/runtime/MemoryManagerOnDemand.h"
+#include "arm_compute/runtime/PoolManager.h"
#include "utils/Utils.h"
using namespace arm_compute;
using namespace utils;
-void main_cnn(int argc, const char **argv)
+class NEONCNNExample : public Example
{
- ARM_COMPUTE_UNUSED(argc);
- ARM_COMPUTE_UNUSED(argv);
+public:
+ void do_setup(int argc, char **argv) override
+ {
+ ARM_COMPUTE_UNUSED(argc);
+ ARM_COMPUTE_UNUSED(argv);
- // The src tensor should contain the input image
- Tensor src;
-
- // The weights and biases tensors should be initialized with the values inferred with the training
- Tensor weights0;
- Tensor weights1;
- Tensor weights2;
- Tensor biases0;
- Tensor biases1;
- Tensor biases2;
-
- Tensor out_conv0;
- Tensor out_conv1;
- Tensor out_act0;
- Tensor out_act1;
- Tensor out_act2;
- Tensor out_pool0;
- Tensor out_pool1;
- Tensor out_fc0;
- Tensor out_softmax;
-
- NEConvolutionLayer conv0;
- NEConvolutionLayer conv1;
- NEPoolingLayer pool0;
- NEPoolingLayer pool1;
- NEFullyConnectedLayer fc0;
- NEActivationLayer act0;
- NEActivationLayer act1;
- NEActivationLayer act2;
- NESoftmaxLayer softmax;
+ // Create memory manager components
+ // We need 2 memory managers: 1 for handling the tensors within the functions (mm_layers) and 1 for handling the input and output tensors of the functions (mm_transitions))
+ auto lifetime_mgr0 = std::make_shared<BlobLifetimeManager>(); // Create lifetime manager
+ auto lifetime_mgr1 = std::make_shared<BlobLifetimeManager>(); // Create lifetime manager
+ auto pool_mgr0 = std::make_shared<PoolManager>(); // Create pool manager
+ auto pool_mgr1 = std::make_shared<PoolManager>(); // Create pool manager
+ auto mm_layers = std::make_shared<MemoryManagerOnDemand>(lifetime_mgr0, pool_mgr0); // Create the memory manager
+ auto mm_transitions = std::make_shared<MemoryManagerOnDemand>(lifetime_mgr1, pool_mgr1); // Create the memory manager
- /* [Initialize tensors] */
+ // The weights and biases tensors should be initialized with the values inferred with the training
- // Initialize src tensor
- constexpr unsigned int width_src_image = 32;
- constexpr unsigned int height_src_image = 32;
- constexpr unsigned int ifm_src_img = 1;
+ // Set memory manager where allowed to manage internal memory requirements
+ conv0 = arm_compute::support::cpp14::make_unique<NEConvolutionLayer>(mm_layers);
+ conv1 = arm_compute::support::cpp14::make_unique<NEConvolutionLayer>(mm_layers);
+ fc0 = arm_compute::support::cpp14::make_unique<NEFullyConnectedLayer>(mm_layers);
+ softmax = arm_compute::support::cpp14::make_unique<NESoftmaxLayer>(mm_layers);
- const TensorShape src_shape(width_src_image, height_src_image, ifm_src_img);
- src.allocator()->init(TensorInfo(src_shape, 1, DataType::F32));
+ /* [Initialize tensors] */
- // Initialize tensors of conv0
- constexpr unsigned int kernel_x_conv0 = 5;
- constexpr unsigned int kernel_y_conv0 = 5;
- constexpr unsigned int ofm_conv0 = 8;
+ // Initialize src tensor
+ constexpr unsigned int width_src_image = 32;
+ constexpr unsigned int height_src_image = 32;
+ constexpr unsigned int ifm_src_img = 1;
- const TensorShape weights_shape_conv0(kernel_x_conv0, kernel_y_conv0, src_shape.z(), ofm_conv0);
- const TensorShape biases_shape_conv0(weights_shape_conv0[3]);
- const TensorShape out_shape_conv0(src_shape.x(), src_shape.y(), weights_shape_conv0[3]);
-
- weights0.allocator()->init(TensorInfo(weights_shape_conv0, 1, DataType::F32));
- biases0.allocator()->init(TensorInfo(biases_shape_conv0, 1, DataType::F32));
- out_conv0.allocator()->init(TensorInfo(out_shape_conv0, 1, DataType::F32));
-
- // Initialize tensor of act0
- out_act0.allocator()->init(TensorInfo(out_shape_conv0, 1, DataType::F32));
-
- // Initialize tensor of pool0
- TensorShape out_shape_pool0 = out_shape_conv0;
- out_shape_pool0.set(0, out_shape_pool0.x() / 2);
- out_shape_pool0.set(1, out_shape_pool0.y() / 2);
- out_pool0.allocator()->init(TensorInfo(out_shape_pool0, 1, DataType::F32));
-
- // Initialize tensors of conv1
- constexpr unsigned int kernel_x_conv1 = 3;
- constexpr unsigned int kernel_y_conv1 = 3;
- constexpr unsigned int ofm_conv1 = 16;
+ const TensorShape src_shape(width_src_image, height_src_image, ifm_src_img);
+ src.allocator()->init(TensorInfo(src_shape, 1, DataType::F32));
- const TensorShape weights_shape_conv1(kernel_x_conv1, kernel_y_conv1, out_shape_pool0.z(), ofm_conv1);
+ // Initialize tensors of conv0
+ constexpr unsigned int kernel_x_conv0 = 5;
+ constexpr unsigned int kernel_y_conv0 = 5;
+ constexpr unsigned int ofm_conv0 = 8;
- const TensorShape biases_shape_conv1(weights_shape_conv1[3]);
- const TensorShape out_shape_conv1(out_shape_pool0.x(), out_shape_pool0.y(), weights_shape_conv1[3]);
-
- weights1.allocator()->init(TensorInfo(weights_shape_conv1, 1, DataType::F32));
- biases1.allocator()->init(TensorInfo(biases_shape_conv1, 1, DataType::F32));
- out_conv1.allocator()->init(TensorInfo(out_shape_conv1, 1, DataType::F32));
-
- // Initialize tensor of act1
- out_act1.allocator()->init(TensorInfo(out_shape_conv1, 1, DataType::F32));
-
- // Initialize tensor of pool1
- TensorShape out_shape_pool1 = out_shape_conv1;
- out_shape_pool1.set(0, out_shape_pool1.x() / 2);
- out_shape_pool1.set(1, out_shape_pool1.y() / 2);
- out_pool1.allocator()->init(TensorInfo(out_shape_pool1, 1, DataType::F32));
-
- // Initialize tensor of fc0
- constexpr unsigned int num_labels = 128;
-
- const TensorShape weights_shape_fc0(out_shape_pool1.x() * out_shape_pool1.y() * out_shape_pool1.z(), num_labels);
- const TensorShape biases_shape_fc0(num_labels);
- const TensorShape out_shape_fc0(num_labels);
-
- weights2.allocator()->init(TensorInfo(weights_shape_fc0, 1, DataType::F32));
- biases2.allocator()->init(TensorInfo(biases_shape_fc0, 1, DataType::F32));
- out_fc0.allocator()->init(TensorInfo(out_shape_fc0, 1, DataType::F32));
-
- // Initialize tensor of act2
- out_act2.allocator()->init(TensorInfo(out_shape_fc0, 1, DataType::F32));
-
- // Initialize tensor of softmax
- const TensorShape out_shape_softmax(out_shape_fc0.x());
- out_softmax.allocator()->init(TensorInfo(out_shape_softmax, 1, DataType::F32));
-
- /* -----------------------End: [Initialize tensors] */
-
- /* [Configure functions] */
-
- // in:32x32x1: 5x5 convolution, 8 output features maps (OFM)
- conv0.configure(&src, &weights0, &biases0, &out_conv0, PadStrideInfo());
+ const TensorShape weights_shape_conv0(kernel_x_conv0, kernel_y_conv0, src_shape.z(), ofm_conv0);
+ const TensorShape biases_shape_conv0(weights_shape_conv0[3]);
+ const TensorShape out_shape_conv0(src_shape.x(), src_shape.y(), weights_shape_conv0[3]);
- // in:32x32x8, out:32x32x8, Activation function: relu
- act0.configure(&out_conv0, &out_act0, ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU));
+ weights0.allocator()->init(TensorInfo(weights_shape_conv0, 1, DataType::F32));
+ biases0.allocator()->init(TensorInfo(biases_shape_conv0, 1, DataType::F32));
+ out_conv0.allocator()->init(TensorInfo(out_shape_conv0, 1, DataType::F32));
- // in:32x32x8, out:16x16x8 (2x2 pooling), Pool type function: Max
- pool0.configure(&out_act0, &out_pool0, PoolingLayerInfo(PoolingType::MAX, 2));
+ // Initialize tensor of act0
+ out_act0.allocator()->init(TensorInfo(out_shape_conv0, 1, DataType::F32));
- // in:16x16x8: 3x3 convolution, 16 output features maps (OFM)
- conv1.configure(&out_pool0, &weights1, &biases1, &out_conv1, PadStrideInfo());
+ // Initialize tensor of pool0
+ TensorShape out_shape_pool0 = out_shape_conv0;
+ out_shape_pool0.set(0, out_shape_pool0.x() / 2);
+ out_shape_pool0.set(1, out_shape_pool0.y() / 2);
+ out_pool0.allocator()->init(TensorInfo(out_shape_pool0, 1, DataType::F32));
- // in:16x16x16, out:16x16x16, Activation function: relu
- act1.configure(&out_conv1, &out_act1, ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU));
+ // Initialize tensors of conv1
+ constexpr unsigned int kernel_x_conv1 = 3;
+ constexpr unsigned int kernel_y_conv1 = 3;
+ constexpr unsigned int ofm_conv1 = 16;
- // in:16x16x16, out:8x8x16 (2x2 pooling), Pool type function: Average
- pool1.configure(&out_act1, &out_pool1, PoolingLayerInfo(PoolingType::AVG, 2));
+ const TensorShape weights_shape_conv1(kernel_x_conv1, kernel_y_conv1, out_shape_pool0.z(), ofm_conv1);
- // in:8x8x16, out:128
- fc0.configure(&out_pool1, &weights2, &biases2, &out_fc0);
+ const TensorShape biases_shape_conv1(weights_shape_conv1[3]);
+ const TensorShape out_shape_conv1(out_shape_pool0.x(), out_shape_pool0.y(), weights_shape_conv1[3]);
+
+ weights1.allocator()->init(TensorInfo(weights_shape_conv1, 1, DataType::F32));
+ biases1.allocator()->init(TensorInfo(biases_shape_conv1, 1, DataType::F32));
+ out_conv1.allocator()->init(TensorInfo(out_shape_conv1, 1, DataType::F32));
+
+ // Initialize tensor of act1
+ out_act1.allocator()->init(TensorInfo(out_shape_conv1, 1, DataType::F32));
- // in:128, out:128, Activation function: relu
- act2.configure(&out_fc0, &out_act2, ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU));
+ // Initialize tensor of pool1
+ TensorShape out_shape_pool1 = out_shape_conv1;
+ out_shape_pool1.set(0, out_shape_pool1.x() / 2);
+ out_shape_pool1.set(1, out_shape_pool1.y() / 2);
+ out_pool1.allocator()->init(TensorInfo(out_shape_pool1, 1, DataType::F32));
- // in:128, out:128
- softmax.configure(&out_act2, &out_softmax);
+ // Initialize tensor of fc0
+ constexpr unsigned int num_labels = 128;
- /* -----------------------End: [Configure functions] */
+ const TensorShape weights_shape_fc0(out_shape_pool1.x() * out_shape_pool1.y() * out_shape_pool1.z(), num_labels);
+ const TensorShape biases_shape_fc0(num_labels);
+ const TensorShape out_shape_fc0(num_labels);
- /* [Allocate tensors] */
+ weights2.allocator()->init(TensorInfo(weights_shape_fc0, 1, DataType::F32));
+ biases2.allocator()->init(TensorInfo(biases_shape_fc0, 1, DataType::F32));
+ out_fc0.allocator()->init(TensorInfo(out_shape_fc0, 1, DataType::F32));
- // Now that the padding requirements are known we can allocate the images:
- src.allocator()->allocate();
- weights0.allocator()->allocate();
- weights1.allocator()->allocate();
- weights2.allocator()->allocate();
- biases0.allocator()->allocate();
- biases1.allocator()->allocate();
- biases2.allocator()->allocate();
- out_conv0.allocator()->allocate();
- out_conv1.allocator()->allocate();
- out_act0.allocator()->allocate();
- out_act1.allocator()->allocate();
- out_act2.allocator()->allocate();
- out_pool0.allocator()->allocate();
- out_pool1.allocator()->allocate();
- out_fc0.allocator()->allocate();
- out_softmax.allocator()->allocate();
+ // Initialize tensor of act2
+ out_act2.allocator()->init(TensorInfo(out_shape_fc0, 1, DataType::F32));
- /* -----------------------End: [Allocate tensors] */
+ // Initialize tensor of softmax
+ const TensorShape out_shape_softmax(out_shape_fc0.x());
+ out_softmax.allocator()->init(TensorInfo(out_shape_softmax, 1, DataType::F32));
- /* [Initialize weights and biases tensors] */
+ /* -----------------------End: [Initialize tensors] */
- // Once the tensors have been allocated, the src, weights and biases tensors can be initialized
- // ...
+ /* [Configure functions] */
- /* -----------------------[Initialize weights and biases tensors] */
+ // in:32x32x1: 5x5 convolution, 8 output features maps (OFM)
+ conv0->configure(&src, &weights0, &biases0, &out_conv0, PadStrideInfo(1 /* stride_x */, 1 /* stride_y */, 2 /* pad_x */, 2 /* pad_y */));
- /* [Execute the functions] */
+ // in:32x32x8, out:32x32x8, Activation function: relu
+ act0.configure(&out_conv0, &out_act0, ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU));
- conv0.run();
- act0.run();
- pool0.run();
- conv1.run();
- act1.run();
- pool1.run();
- fc0.run();
- act2.run();
- softmax.run();
+ // in:32x32x8, out:16x16x8 (2x2 pooling), Pool type function: Max
+ pool0.configure(&out_act0, &out_pool0, PoolingLayerInfo(PoolingType::MAX, 2, PadStrideInfo(2 /* stride_x */, 2 /* stride_y */)));
- /* -----------------------End: [Execute the functions] */
-}
+ // in:16x16x8: 3x3 convolution, 16 output features maps (OFM)
+ conv1->configure(&out_pool0, &weights1, &biases1, &out_conv1, PadStrideInfo(1 /* stride_x */, 1 /* stride_y */, 1 /* pad_x */, 1 /* pad_y */));
+
+ // in:16x16x16, out:16x16x16, Activation function: relu
+ act1.configure(&out_conv1, &out_act1, ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU));
+
+ // in:16x16x16, out:8x8x16 (2x2 pooling), Pool type function: Average
+ pool1.configure(&out_act1, &out_pool1, PoolingLayerInfo(PoolingType::AVG, 2, PadStrideInfo(2 /* stride_x */, 2 /* stride_y */)));
+
+ // in:8x8x16, out:128
+ fc0->configure(&out_pool1, &weights2, &biases2, &out_fc0);
+
+ // in:128, out:128, Activation function: relu
+ act2.configure(&out_fc0, &out_act2, ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU));
+
+ // in:128, out:128
+ softmax->configure(&out_act2, &out_softmax);
+
+ /* -----------------------End: [Configure functions] */
+
+ /*[ Add tensors to memory manager ]*/
+
+ // We need 2 memory groups for handling the input and output
+ // We call explicitly allocate after manage() in order to avoid overlapping lifetimes
+ memory_group0 = arm_compute::support::cpp14::make_unique<MemoryGroup>(mm_transitions);
+ memory_group1 = arm_compute::support::cpp14::make_unique<MemoryGroup>(mm_transitions);
+
+ memory_group0->manage(&out_conv0);
+ out_conv0.allocator()->allocate();
+ memory_group1->manage(&out_act0);
+ out_act0.allocator()->allocate();
+ memory_group0->manage(&out_pool0);
+ out_pool0.allocator()->allocate();
+ memory_group1->manage(&out_conv1);
+ out_conv1.allocator()->allocate();
+ memory_group0->manage(&out_act1);
+ out_act1.allocator()->allocate();
+ memory_group1->manage(&out_pool1);
+ out_pool1.allocator()->allocate();
+ memory_group0->manage(&out_fc0);
+ out_fc0.allocator()->allocate();
+ memory_group1->manage(&out_act2);
+ out_act2.allocator()->allocate();
+ memory_group0->manage(&out_softmax);
+ out_softmax.allocator()->allocate();
+
+ /* -----------------------End: [ Add tensors to memory manager ] */
+
+ /* [Allocate tensors] */
+
+ // Now that the padding requirements are known we can allocate all tensors
+ src.allocator()->allocate();
+ weights0.allocator()->allocate();
+ weights1.allocator()->allocate();
+ weights2.allocator()->allocate();
+ biases0.allocator()->allocate();
+ biases1.allocator()->allocate();
+ biases2.allocator()->allocate();
+
+ /* -----------------------End: [Allocate tensors] */
+
+ // Finalize layers memory manager
+
+ // Set allocator that the memory manager will use
+ mm_layers->set_allocator(&allocator);
+
+ // Number of pools that the manager will create. This specifies how many layers you want to run in parallel
+ mm_layers->set_num_pools(1);
+
+ // Finalize the manager. (Validity checks, memory allocations etc)
+ mm_layers->finalize();
+
+ // Finalize transitions memory manager
+
+ // Set allocator that the memory manager will use
+ mm_transitions->set_allocator(&allocator);
+
+ // Number of pools that the manager will create. This specifies how many models we can run in parallel.
+ // Setting to 2 as we need one for the input and one for the output at any given time
+ mm_transitions->set_num_pools(2);
+
+ // Finalize the manager. (Validity checks, memory allocations etc)
+ mm_transitions->finalize();
+ }
+ void do_run() override
+ {
+ // Acquire memory for the memory groups
+ memory_group0->acquire();
+ memory_group1->acquire();
+
+ conv0->run();
+ act0.run();
+ pool0.run();
+ conv1->run();
+ act1.run();
+ pool1.run();
+ fc0->run();
+ act2.run();
+ softmax->run();
+
+ // Release memory
+ memory_group0->release();
+ memory_group1->release();
+ }
+
+private:
+ // The src tensor should contain the input image
+ Tensor src{};
+
+ // Intermediate tensors used
+ Tensor weights0{};
+ Tensor weights1{};
+ Tensor weights2{};
+ Tensor biases0{};
+ Tensor biases1{};
+ Tensor biases2{};
+ Tensor out_conv0{};
+ Tensor out_conv1{};
+ Tensor out_act0{};
+ Tensor out_act1{};
+ Tensor out_act2{};
+ Tensor out_pool0{};
+ Tensor out_pool1{};
+ Tensor out_fc0{};
+ Tensor out_softmax{};
+
+ // NEON allocator
+ Allocator allocator{};
+
+ // Memory groups
+ std::unique_ptr<MemoryGroup> memory_group0{};
+ std::unique_ptr<MemoryGroup> memory_group1{};
+
+ // Layers
+ std::unique_ptr<NEConvolutionLayer> conv0{};
+ std::unique_ptr<NEConvolutionLayer> conv1{};
+ std::unique_ptr<NEFullyConnectedLayer> fc0{};
+ std::unique_ptr<NESoftmaxLayer> softmax{};
+ NEPoolingLayer pool0{};
+ NEPoolingLayer pool1{};
+ NEActivationLayer act0{};
+ NEActivationLayer act1{};
+ NEActivationLayer act2{};
+};
/** Main program for cnn test
*
* @param[in] argc Number of arguments
* @param[in] argv Arguments
*/
-int main(int argc, const char **argv)
+int main(int argc, char **argv)
{
- return utils::run_example(argc, argv, main_cnn);
-}
\ No newline at end of file
+ return utils::run_example<NEONCNNExample>(argc, argv);
+}
projects / platform / upstream / armcl.git / blobdiff