* 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/SubGraph.h"
+#include "arm_compute/graph.h"
#include "support/ToolchainSupport.h"
#include "utils/GraphUtils.h"
#include "utils/Utils.h"
#include <tuple>
using namespace arm_compute::utils;
-using namespace arm_compute::graph;
+using namespace arm_compute::graph::frontend;
using namespace arm_compute::graph_utils;
-using namespace arm_compute::logging;
namespace
{
/** Example demonstrating how to implement Squeezenet's v1.1 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 GraphSqueezenet_v1_1Example : public Example
{
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);
+ 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
- << Tensor(TensorInfo(TensorShape(227U, 227U, 3U, 1U), 1, DataType::F32),
- get_input_accessor(image, std::move(preprocessor)))
+ << fast_math_hint
+ << InputLayer(TensorDescriptor(TensorShape(227U, 227U, 3U, 1U), DataType::F32),
+ get_input_accessor(image, std::move(preprocessor)))
+ << ConvolutionMethod::DIRECT
<< ConvolutionLayer(
3U, 3U, 64U,
get_weights_accessor(data_path, "/cnn_data/squeezenet_v1_1_model/conv1_w.npy"),
PadStrideInfo(2, 2, 0, 0))
<< ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU))
<< PoolingLayer(PoolingLayerInfo(PoolingType::MAX, 3, PadStrideInfo(2, 2, 0, 0, DimensionRoundingType::CEIL)))
+ << ConvolutionMethod::DEFAULT
<< ConvolutionLayer(
1U, 1U, 16U,
get_weights_accessor(data_path, "/cnn_data/squeezenet_v1_1_model/fire2_squeeze1x1_w.npy"),
get_weights_accessor(data_path, "/cnn_data/squeezenet_v1_1_model/fire2_squeeze1x1_b.npy"),
PadStrideInfo(1, 1, 0, 0))
- << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU))
- << get_expand_fire_node(data_path, "fire2", 64U, 64U)
- << ConvolutionLayer(
+ << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU));
+ graph << get_expand_fire_node(data_path, "fire2", 64U, 64U);
+ graph << ConvolutionLayer(
1U, 1U, 16U,
get_weights_accessor(data_path, "/cnn_data/squeezenet_v1_1_model/fire3_squeeze1x1_w.npy"),
get_weights_accessor(data_path, "/cnn_data/squeezenet_v1_1_model/fire3_squeeze1x1_b.npy"),
PadStrideInfo(1, 1, 0, 0))
- << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU))
- << get_expand_fire_node(data_path, "fire3", 64U, 64U)
- << PoolingLayer(PoolingLayerInfo(PoolingType::MAX, 3, PadStrideInfo(2, 2, 0, 0, DimensionRoundingType::CEIL)))
+ << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU));
+ graph << get_expand_fire_node(data_path, "fire3", 64U, 64U);
+ graph << PoolingLayer(PoolingLayerInfo(PoolingType::MAX, 3, PadStrideInfo(2, 2, 0, 0, DimensionRoundingType::CEIL)))
<< ConvolutionLayer(
1U, 1U, 32U,
get_weights_accessor(data_path, "/cnn_data/squeezenet_v1_1_model/fire4_squeeze1x1_w.npy"),
get_weights_accessor(data_path, "/cnn_data/squeezenet_v1_1_model/fire4_squeeze1x1_b.npy"),
PadStrideInfo(1, 1, 0, 0))
- << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU))
- << get_expand_fire_node(data_path, "fire4", 128U, 128U)
- << ConvolutionLayer(
+ << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU));
+ graph << get_expand_fire_node(data_path, "fire4", 128U, 128U);
+ graph << ConvolutionLayer(
1U, 1U, 32U,
get_weights_accessor(data_path, "/cnn_data/squeezenet_v1_1_model/fire5_squeeze1x1_w.npy"),
get_weights_accessor(data_path, "/cnn_data/squeezenet_v1_1_model/fire5_squeeze1x1_b.npy"),
PadStrideInfo(1, 1, 0, 0))
- << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU))
- << get_expand_fire_node(data_path, "fire5", 128U, 128U)
- << PoolingLayer(PoolingLayerInfo(PoolingType::MAX, 3, PadStrideInfo(2, 2, 0, 0, DimensionRoundingType::CEIL)))
+ << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU));
+ graph << get_expand_fire_node(data_path, "fire5", 128U, 128U);
+ graph << PoolingLayer(PoolingLayerInfo(PoolingType::MAX, 3, PadStrideInfo(2, 2, 0, 0, DimensionRoundingType::CEIL)))
<< ConvolutionLayer(
1U, 1U, 48U,
get_weights_accessor(data_path, "/cnn_data/squeezenet_v1_1_model/fire6_squeeze1x1_w.npy"),
get_weights_accessor(data_path, "/cnn_data/squeezenet_v1_1_model/fire6_squeeze1x1_b.npy"),
PadStrideInfo(1, 1, 0, 0))
- << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU))
- << get_expand_fire_node(data_path, "fire6", 192U, 192U)
- << ConvolutionLayer(
+ << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU));
+ graph << get_expand_fire_node(data_path, "fire6", 192U, 192U);
+ graph << ConvolutionLayer(
1U, 1U, 48U,
get_weights_accessor(data_path, "/cnn_data/squeezenet_v1_1_model/fire7_squeeze1x1_w.npy"),
get_weights_accessor(data_path, "/cnn_data/squeezenet_v1_1_model/fire7_squeeze1x1_b.npy"),
PadStrideInfo(1, 1, 0, 0))
- << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU))
- << get_expand_fire_node(data_path, "fire7", 192U, 192U)
- << ConvolutionLayer(
+ << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU));
+ graph << get_expand_fire_node(data_path, "fire7", 192U, 192U);
+ graph << ConvolutionLayer(
1U, 1U, 64U,
get_weights_accessor(data_path, "/cnn_data/squeezenet_v1_1_model/fire8_squeeze1x1_w.npy"),
get_weights_accessor(data_path, "/cnn_data/squeezenet_v1_1_model/fire8_squeeze1x1_b.npy"),
PadStrideInfo(1, 1, 0, 0))
- << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU))
- << get_expand_fire_node(data_path, "fire8", 256U, 256U)
- << ConvolutionLayer(
+ << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU));
+ graph << get_expand_fire_node(data_path, "fire8", 256U, 256U);
+ graph << ConvolutionLayer(
1U, 1U, 64U,
get_weights_accessor(data_path, "/cnn_data/squeezenet_v1_1_model/fire9_squeeze1x1_w.npy"),
get_weights_accessor(data_path, "/cnn_data/squeezenet_v1_1_model/fire9_squeeze1x1_b.npy"),
PadStrideInfo(1, 1, 0, 0))
- << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU))
- << get_expand_fire_node(data_path, "fire9", 256U, 256U)
- << ConvolutionLayer(
+ << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU));
+ graph << get_expand_fire_node(data_path, "fire9", 256U, 256U);
+ graph << ConvolutionLayer(
1U, 1U, 1000U,
get_weights_accessor(data_path, "/cnn_data/squeezenet_v1_1_model/conv10_w.npy"),
get_weights_accessor(data_path, "/cnn_data/squeezenet_v1_1_model/conv10_b.npy"),
<< PoolingLayer(PoolingLayerInfo(PoolingType::AVG))
<< FlattenLayer()
<< SoftmaxLayer()
- << Tensor(get_output_accessor(label, 5));
+ << 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
{
}
private:
- Graph graph{};
+ Stream graph{ 0, "SqueezeNetV1.1" };
BranchLayer get_expand_fire_node(const std::string &data_path, std::string &¶m_path, unsigned int expand1_filt, unsigned int expand3_filt)
{
std::string total_path = "/cnn_data/squeezenet_v1_1_model/" + param_path + "_";
- SubGraph i_a;
+ SubStream i_a(graph);
i_a << ConvolutionLayer(
1U, 1U, expand1_filt,
get_weights_accessor(data_path, total_path + "expand1x1_w.npy"),
PadStrideInfo(1, 1, 0, 0))
<< ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU));
- SubGraph i_b;
+ SubStream i_b(graph);
i_b << ConvolutionLayer(
3U, 3U, expand3_filt,
get_weights_accessor(data_path, total_path + "expand3x3_w.npy"),
/** Main program for Squeezenet v1.1
*
* @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)
{
projects / platform / upstream / armcl.git / blobdiff