arm_compute v18.05
diff --git a/examples/graph_vgg16.cpp b/examples/graph_vgg16.cpp
index c8cc5b2..72e7240 100644
--- a/examples/graph_vgg16.cpp
+++ b/examples/graph_vgg16.cpp
@@ -21,8 +21,7 @@
* OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
* SOFTWARE.
*/
-#include "arm_compute/graph/Graph.h"
-#include "arm_compute/graph/Nodes.h"
+#include "arm_compute/graph.h"
#include "support/ToolchainSupport.h"
#include "utils/GraphUtils.h"
#include "utils/Utils.h"
@@ -30,27 +29,13 @@
#include <cstdlib>
using namespace arm_compute::utils;
-using namespace arm_compute::graph;
+using namespace arm_compute::graph::frontend;
using namespace arm_compute::graph_utils;
-namespace
-{
-/** This function checks if we can use GEMM-based convolution trying to allocate a memory of size "size_in_bytes"
- *
- * @param[in] size_in_bytes Memory size in bytes needed for VGG-16
- *
- * @return The convolution layer hint
- */
-ConvolutionMethodHint convolution_hint_vgg16(size_t size_in_bytes)
-{
- return ((get_mem_free_from_meminfo() * 1024) >= size_in_bytes) ? ConvolutionMethodHint::GEMM : ConvolutionMethodHint::DIRECT;
-}
-} // namespace
-
/** Example demonstrating how to implement VGG16's network using the Compute Library's graph API
*
* @param[in] argc Number of arguments
- * @param[in] argv Arguments ( [optional] Target (0 = NEON, 1 = OpenCL), [optional] Path to the weights folder, [optional] image, [optional] labels )
+ * @param[in] argv Arguments ( [optional] Target (0 = NEON, 1 = OpenCL, 2 = OpenCL with Tuner), [optional] Path to the weights folder, [optional] image, [optional] labels, [optional] Fast math for convolution layer (0 = DISABLED, 1 = ENABLED) )
*/
class GraphVGG16Example : public Example
{
@@ -66,168 +51,198 @@
std::unique_ptr<IPreprocessor> preprocessor = arm_compute::support::cpp14::make_unique<CaffePreproccessor>(mean_rgb);
// Set target. 0 (NEON), 1 (OpenCL), 2 (OpenCL with Tuner). By default it is NEON
- const int int_target_hint = argc > 1 ? std::strtol(argv[1], nullptr, 10) : 0;
- TargetHint target_hint = set_target_hint(int_target_hint);
+ const int target = argc > 1 ? std::strtol(argv[1], nullptr, 10) : 0;
+ Target target_hint = set_target_hint(target);
+ const bool is_opencl = target_hint == Target::CL;
- // Check if we can use GEMM-based convolutions evaluating if the platform has at least 1.8 GB of available memory
- const size_t memory_required = 1932735283L;
- ConvolutionMethodHint convolution_hint = convolution_hint_vgg16(memory_required);
+ ConvolutionMethod first_convolution3x3_hint = is_opencl ? ConvolutionMethod::DIRECT : ConvolutionMethod::GEMM;
+ ConvolutionMethod convolution3x3_hint = ConvolutionMethod::DEFAULT;
+ FastMathHint fast_math_hint = FastMathHint::DISABLED;
// Parse arguments
if(argc < 2)
{
// Print help
- std::cout << "Usage: " << argv[0] << " [target] [path_to_data] [image] [labels]\n\n";
+ std::cout << "Usage: " << argv[0] << " [target] [path_to_data] [image] [labels] [fast_math_hint]\n\n";
std::cout << "No data folder provided: using random values\n\n";
}
else if(argc == 2)
{
- std::cout << "Usage: " << argv[0] << " " << argv[1] << " [path_to_data] [image] [labels]\n\n";
+ std::cout << "Usage: " << argv[0] << " " << argv[1] << " [path_to_data] [image] [labels] [fast_math_hint]\n\n";
std::cout << "No data folder provided: using random values\n\n";
}
else if(argc == 3)
{
data_path = argv[2];
- std::cout << "Usage: " << argv[0] << " " << argv[1] << " " << argv[2] << " [image] [labels]\n\n";
+ std::cout << "Usage: " << argv[0] << " " << argv[1] << " " << argv[2] << " [image] [labels] [fast_math_hint]\n\n";
std::cout << "No image provided: using random values\n\n";
}
else if(argc == 4)
{
data_path = argv[2];
image = argv[3];
- std::cout << "Usage: " << argv[0] << " " << argv[1] << " " << argv[2] << " " << argv[3] << " [labels]\n\n";
+ std::cout << "Usage: " << argv[0] << " " << argv[1] << " " << argv[2] << " " << argv[3] << " [labels] [fast_math_hint]\n\n";
std::cout << "No text file with labels provided: skipping output accessor\n\n";
}
- else
+ else if(argc == 5)
{
data_path = argv[2];
image = argv[3];
label = argv[4];
+ std::cout << "Usage: " << argv[0] << " " << argv[1] << " " << argv[2] << " " << argv[3] << " " << argv[4] << " [fast_math_hint]\n\n";
+ std::cout << "No fast math info provided: disabling fast math\n\n";
+ }
+ else
+ {
+ data_path = argv[2];
+ image = argv[3];
+ label = argv[4];
+ fast_math_hint = (std::strtol(argv[5], nullptr, 1) == 0) ? FastMathHint::DISABLED : FastMathHint::ENABLED;
}
graph << target_hint
- << convolution_hint
- << Tensor(TensorInfo(TensorShape(224U, 224U, 3U, 1U), 1, DataType::F32),
- get_input_accessor(image, std::move(preprocessor)))
+ << fast_math_hint
+ << first_convolution3x3_hint
+ << InputLayer(TensorDescriptor(TensorShape(224U, 224U, 3U, 1U), DataType::F32),
+ get_input_accessor(image, std::move(preprocessor)))
// Layer 1
<< ConvolutionLayer(
3U, 3U, 64U,
get_weights_accessor(data_path, "/cnn_data/vgg16_model/conv1_1_w.npy"),
get_weights_accessor(data_path, "/cnn_data/vgg16_model/conv1_1_b.npy"),
PadStrideInfo(1, 1, 1, 1))
- << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU))
+ .set_name("conv1_1")
+ << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU)).set_name("conv1_1/Relu")
+ << convolution3x3_hint
// Layer 2
<< ConvolutionLayer(
3U, 3U, 64U,
get_weights_accessor(data_path, "/cnn_data/vgg16_model/conv1_2_w.npy"),
get_weights_accessor(data_path, "/cnn_data/vgg16_model/conv1_2_b.npy"),
PadStrideInfo(1, 1, 1, 1))
- << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU))
- << PoolingLayer(PoolingLayerInfo(PoolingType::MAX, 2, PadStrideInfo(2, 2, 0, 0)))
+ .set_name("conv1_2")
+ << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU)).set_name("conv1_2/Relu")
+ << PoolingLayer(PoolingLayerInfo(PoolingType::MAX, 2, PadStrideInfo(2, 2, 0, 0))).set_name("pool1")
// Layer 3
<< ConvolutionLayer(
3U, 3U, 128U,
get_weights_accessor(data_path, "/cnn_data/vgg16_model/conv2_1_w.npy"),
get_weights_accessor(data_path, "/cnn_data/vgg16_model/conv2_1_b.npy"),
PadStrideInfo(1, 1, 1, 1))
- << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU))
+ .set_name("conv2_1")
+ << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU)).set_name("conv2_1/Relu")
// Layer 4
<< ConvolutionLayer(
3U, 3U, 128U,
get_weights_accessor(data_path, "/cnn_data/vgg16_model/conv2_2_w.npy"),
get_weights_accessor(data_path, "/cnn_data/vgg16_model/conv2_2_b.npy"),
PadStrideInfo(1, 1, 1, 1))
- << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU))
- << PoolingLayer(PoolingLayerInfo(PoolingType::MAX, 2, PadStrideInfo(2, 2, 0, 0)))
+ .set_name("conv2_2")
+ << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU)).set_name("conv2_2/Relu")
+ << PoolingLayer(PoolingLayerInfo(PoolingType::MAX, 2, PadStrideInfo(2, 2, 0, 0))).set_name("pool2")
// Layer 5
<< ConvolutionLayer(
3U, 3U, 256U,
get_weights_accessor(data_path, "/cnn_data/vgg16_model/conv3_1_w.npy"),
get_weights_accessor(data_path, "/cnn_data/vgg16_model/conv3_1_b.npy"),
PadStrideInfo(1, 1, 1, 1))
- << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU))
+ .set_name("conv3_1")
+ << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU)).set_name("conv3_1/Relu")
// Layer 6
<< ConvolutionLayer(
3U, 3U, 256U,
get_weights_accessor(data_path, "/cnn_data/vgg16_model/conv3_2_w.npy"),
get_weights_accessor(data_path, "/cnn_data/vgg16_model/conv3_2_b.npy"),
PadStrideInfo(1, 1, 1, 1))
- << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU))
+ .set_name("conv3_2")
+ << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU)).set_name("conv3_2/Relu")
// Layer 7
<< ConvolutionLayer(
3U, 3U, 256U,
get_weights_accessor(data_path, "/cnn_data/vgg16_model/conv3_3_w.npy"),
get_weights_accessor(data_path, "/cnn_data/vgg16_model/conv3_3_b.npy"),
PadStrideInfo(1, 1, 1, 1))
- << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU))
- << PoolingLayer(PoolingLayerInfo(PoolingType::MAX, 2, PadStrideInfo(2, 2, 0, 0)))
+ .set_name("conv3_3")
+ << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU)).set_name("conv3_3/Relu")
+ << PoolingLayer(PoolingLayerInfo(PoolingType::MAX, 2, PadStrideInfo(2, 2, 0, 0))).set_name("pool3")
// Layer 8
<< ConvolutionLayer(
3U, 3U, 512U,
get_weights_accessor(data_path, "/cnn_data/vgg16_model/conv4_1_w.npy"),
get_weights_accessor(data_path, "/cnn_data/vgg16_model/conv4_1_b.npy"),
PadStrideInfo(1, 1, 1, 1))
- << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU))
+ .set_name("conv4_1")
+ << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU)).set_name("conv4_1/Relu")
// Layer 9
<< ConvolutionLayer(
3U, 3U, 512U,
get_weights_accessor(data_path, "/cnn_data/vgg16_model/conv4_2_w.npy"),
get_weights_accessor(data_path, "/cnn_data/vgg16_model/conv4_2_b.npy"),
PadStrideInfo(1, 1, 1, 1))
- << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU))
+ .set_name("conv4_2")
+ << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU)).set_name("conv4_2/Relu")
// Layer 10
<< ConvolutionLayer(
3U, 3U, 512U,
get_weights_accessor(data_path, "/cnn_data/vgg16_model/conv4_3_w.npy"),
get_weights_accessor(data_path, "/cnn_data/vgg16_model/conv4_3_b.npy"),
PadStrideInfo(1, 1, 1, 1))
- << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU))
- << PoolingLayer(PoolingLayerInfo(PoolingType::MAX, 2, PadStrideInfo(2, 2, 0, 0)))
+ .set_name("conv4_3")
+ << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU)).set_name("conv4_3/Relu")
+ << PoolingLayer(PoolingLayerInfo(PoolingType::MAX, 2, PadStrideInfo(2, 2, 0, 0))).set_name("pool4")
// Layer 11
<< ConvolutionLayer(
3U, 3U, 512U,
get_weights_accessor(data_path, "/cnn_data/vgg16_model/conv5_1_w.npy"),
get_weights_accessor(data_path, "/cnn_data/vgg16_model/conv5_1_b.npy"),
PadStrideInfo(1, 1, 1, 1))
- << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU))
+ .set_name("conv5_1")
+ << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU)).set_name("conv5_1/Relu")
// Layer 12
<< ConvolutionLayer(
3U, 3U, 512U,
get_weights_accessor(data_path, "/cnn_data/vgg16_model/conv5_2_w.npy"),
get_weights_accessor(data_path, "/cnn_data/vgg16_model/conv5_2_b.npy"),
PadStrideInfo(1, 1, 1, 1))
- << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU))
+ .set_name("conv5_2")
+ << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU)).set_name("conv5_2/Relu")
// Layer 13
<< ConvolutionLayer(
3U, 3U, 512U,
get_weights_accessor(data_path, "/cnn_data/vgg16_model/conv5_3_w.npy"),
get_weights_accessor(data_path, "/cnn_data/vgg16_model/conv5_3_b.npy"),
PadStrideInfo(1, 1, 1, 1))
- << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU))
- << PoolingLayer(PoolingLayerInfo(PoolingType::MAX, 2, PadStrideInfo(2, 2, 0, 0)))
+ .set_name("conv5_3")
+ << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU)).set_name("conv5_3/Relu")
+ << PoolingLayer(PoolingLayerInfo(PoolingType::MAX, 2, PadStrideInfo(2, 2, 0, 0))).set_name("pool5")
// Layer 14
<< FullyConnectedLayer(
4096U,
get_weights_accessor(data_path, "/cnn_data/vgg16_model/fc6_w.npy"),
get_weights_accessor(data_path, "/cnn_data/vgg16_model/fc6_b.npy"))
- << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU))
+ .set_name("fc6")
+ << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU)).set_name("Relu")
// Layer 15
<< FullyConnectedLayer(
4096U,
get_weights_accessor(data_path, "/cnn_data/vgg16_model/fc7_w.npy"),
get_weights_accessor(data_path, "/cnn_data/vgg16_model/fc7_b.npy"))
- << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU))
+ .set_name("fc7")
+ << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU)).set_name("Relu_1")
// Layer 16
<< FullyConnectedLayer(
1000U,
get_weights_accessor(data_path, "/cnn_data/vgg16_model/fc8_w.npy"),
get_weights_accessor(data_path, "/cnn_data/vgg16_model/fc8_b.npy"))
+ .set_name("fc8")
// Softmax
- << SoftmaxLayer()
- << Tensor(get_output_accessor(label, 5));
+ << SoftmaxLayer().set_name("prob")
+ << OutputLayer(get_output_accessor(label, 5));
- // In order to enable the OpenCL tuner, graph_init() has to be called only when all nodes have been instantiated
- graph.graph_init(int_target_hint == 2);
+ // Finalize graph
+ GraphConfig config;
+ config.use_tuner = (target == 2);
+ graph.finalize(target_hint, config);
}
void do_run() override
{
@@ -236,13 +251,13 @@
}
private:
- Graph graph{};
+ Stream graph{ 0, "VGG16" };
};
/** Main program for VGG16
*
* @param[in] argc Number of arguments
- * @param[in] argv Arguments ( [optional] Target (0 = NEON, 1 = OpenCL), [optional] Path to the weights folder, [optional] image, [optional] labels )
+ * @param[in] argv Arguments ( [optional] Target (0 = NEON, 1 = OpenCL, 2 = OpenCL with Tuner), [optional] Path to the weights folder, [optional] image, [optional] labels, [optional] Fast math for convolution layer (0 = DISABLED, 1 = ENABLED) )
*/
int main(int argc, char **argv)
{