Add map for moments

Signed-off-by: zhao.xia <zhao.xia@verisilicon.com>
This commit is contained in:
zhao.xia 2021-06-29 13:25:50 +08:00 committed by Kainan Cha
parent 3c59694025
commit 3fa2bf519a
4 changed files with 298 additions and 0 deletions

View File

@ -0,0 +1,56 @@
/****************************************************************************
*
* Copyright (c) 2021 Vivante Corporation
*
* Permission is hereby granted, free of charge, to any person obtaining a
* copy of this software and associated documentation files (the "Software"),
* to deal in the Software without restriction, including without limitation
* the rights to use, copy, modify, merge, publish, distribute, sublicense,
* and/or sell copies of the Software, and to permit persons to whom the
* Software is furnished to do so, subject to the following conditions:
*
* The above copyright notice and this permission notice shall be included in
* all copies or substantial portions of the Software.
*
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
* AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
* FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
* DEALINGS IN THE SOFTWARE.
*
*****************************************************************************/
#ifndef TIM_VX_OPS_MOMENTS_H_
#define TIM_VX_OPS_MOMENTS_H_
#include "tim/vx/operation.h"
namespace tim {
namespace vx {
namespace ops {
/**
* ## Moments
*
* The mean and variance are calculated by aggregating the contents of x across axes.
* If x is 1-D and axes = [0] this is just the mean and variance of a vector.
*
* - axes : Axes along which to compute mean and variance.
* - keep_dims : Produce moments with the same dimensionality as input.
*/
class Moments : public Operation {
public:
Moments(Graph* graph, const std::vector<int32_t>& axes,
bool keep_dims = false);
protected:
const std::vector<int32_t> axes_;
const bool keep_dims_;
};
} // namespace ops
} // namespace vx
} // namespace tim
#endif /* TIM_VX_OPS_MEMONTS_H_ */

View File

@ -83,6 +83,7 @@ LogSoftmax|LOG_SOFTMAX|Mapped|[tf.nn.log_softmax](https://tensorflow.google.cn/a
HardSwish|SWISH|Mapped|[tf.keras.activations.swish](https://tensorflow.google.cn/api_docs/python/tf/keras/activations/swish)
GatherNd|GATHER_ND|Mapped|[tf.gather_nd](https://tensorflow.google.cn/api_docs/python/tf/gather_nd)
Cast|CAST|Mapped|[tf.cast](https://tensorflow.google.cn/api_docs/python/tf/cast)
Moments|MOMENTS|Mapped|[tf.moments](https://tensorflow.google.cn/api_docs/python/tf/nn/moments)
Squeeze|SQUEEZE|Mapped|[tf.squeeze](https://tensorflow.google.cn/api_docs/python/tf/squeeze)
HardSigmoid|HARD_SIGMOID|Mapped|[tf.keras.activations.hard_sigmoid](https://tensorflow.google.cn/api_docs/python/tf/keras/activations/hard_sigmoid)
Mish|MISH|Mapped|[tfa.activations.mish](https://tensorflow.google.cn/addons/api_docs/python/tfa/activations/mish)

41
src/tim/vx/ops/mements.cc Normal file
View File

@ -0,0 +1,41 @@
/****************************************************************************
*
* Copyright (c) 2021 Vivante Corporation
*
* Permission is hereby granted, free of charge, to any person obtaining a
* copy of this software and associated documentation files (the "Software"),
* to deal in the Software without restriction, including without limitation
* the rights to use, copy, modify, merge, publish, distribute, sublicense,
* and/or sell copies of the Software, and to permit persons to whom the
* Software is furnished to do so, subject to the following conditions:
*
* The above copyright notice and this permission notice shall be included in
* all copies or substantial portions of the Software.
*
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
* AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
* FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
* DEALINGS IN THE SOFTWARE.
*
*****************************************************************************/
#include "tim/vx/ops/moments.h"
#include "operation_private.h"
#include "type_utils.h"
#include "vsi_nn_pub.h"
namespace tim {
namespace vx {
namespace ops {
Moments::Moments(Graph* graph, const std::vector<int32_t>& axes, bool keep_dims)
: Operation(graph, VSI_NN_OP_MOMENTS), axes_(axes), keep_dims_(keep_dims) {
this->impl()->node()->nn_param.moments.axis = axes_.data();
this->impl()->node()->nn_param.moments.axis_num = axes_.size();
this->impl()->node()->nn_param.moments.keep_dim = ToVxBool(keep_dims_);
}
} // namespace ops
} // namespace vx
} // namespace tim

View File

@ -0,0 +1,200 @@
/****************************************************************************
*
* Copyright (c) 2021 Vivante Corporation
*
* Permission is hereby granted, free of charge, to any person obtaining a
* copy of this software and associated documentation files (the "Software"),
* to deal in the Software without restriction, including without limitation
* the rights to use, copy, modify, merge, publish, distribute, sublicense,
* and/or sell copies of the Software, and to permit persons to whom the
* Software is furnished to do so, subject to the following conditions:
*
* The above copyright notice and this permission notice shall be included in
* all copies or substantial portions of the Software.
*
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
* AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
* FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
* DEALINGS IN THE SOFTWARE.
*
*****************************************************************************/
#include <string>
#include "tim/vx/context.h"
#include "tim/vx/graph.h"
#include "tim/vx/ops/moments.h"
#include "gtest/gtest.h"
namespace {
template<typename T>
::testing::AssertionResult ArraysMatch(const std::vector<T>& expected,
const std::vector<T>& actual,
T abs_error,
const std::string& msg){
for (size_t i = 0; i < expected.size(); ++i){
EXPECT_NEAR(expected[i], actual[i], abs_error) << msg << " at index:" << i;
}
return ::testing::AssertionSuccess();
}
}
TEST(Moments, shape_6_3_1_float_axes_0_1) {
auto ctx = tim::vx::Context::Create();
auto graph = ctx->CreateGraph();
tim::vx::ShapeType input_shape({6, 3, 1});
tim::vx::ShapeType output_shape({1});
tim::vx::TensorSpec input_spec(tim::vx::DataType::FLOAT32,
input_shape, tim::vx::TensorAttribute::INPUT);
tim::vx::TensorSpec output_spec(tim::vx::DataType::FLOAT32,
output_shape, tim::vx::TensorAttribute::OUTPUT);
auto input_tensor = graph->CreateTensor(input_spec);
auto mean = graph->CreateTensor(output_spec);
auto variance = graph->CreateTensor(output_spec);
std::vector<float> in_data = {
-2, 0, 2,
-3, 0, 3,
-4, 0, 4,
-5, 0, 5,
-6, 0, 6,
-7, 0, 7 };
std::vector<float> mean_golden = {
0
};
std::vector<float> variance_golden = {
15.444444
};
EXPECT_TRUE(input_tensor->CopyDataToTensor(in_data.data(), in_data.size() * sizeof(float)));
std::vector<int32_t> axes = { 0, 1 };
auto op = graph->CreateOperation<tim::vx::ops::Moments>(axes);
(*op).BindInputs({input_tensor}).BindOutputs({mean, variance});
EXPECT_TRUE(graph->Compile());
EXPECT_TRUE(graph->Run());
std::vector<float> mean_output(mean_golden);
std::vector<float> variance_output(variance_golden);
EXPECT_TRUE(mean->CopyDataFromTensor(mean_output.data()));
EXPECT_TRUE(variance->CopyDataFromTensor(variance_output.data()));
EXPECT_TRUE(ArraysMatch(mean_golden, mean_output, 1e-5f, "mean output"));
EXPECT_TRUE(ArraysMatch(variance_golden, variance_output, 1e-5f, "variance output"));
}
TEST(Moments, shape_3_6_1_float_axes_1_keepdims) {
auto ctx = tim::vx::Context::Create();
auto graph = ctx->CreateGraph();
tim::vx::ShapeType input_shape({3, 6, 1});
tim::vx::ShapeType output_shape({3,1,1});
tim::vx::TensorSpec input_spec(tim::vx::DataType::FLOAT32,
input_shape, tim::vx::TensorAttribute::INPUT);
tim::vx::TensorSpec mean_spec(tim::vx::DataType::FLOAT32,
output_shape, tim::vx::TensorAttribute::OUTPUT);
tim::vx::TensorSpec variance_spec(tim::vx::DataType::FLOAT32,
output_shape, tim::vx::TensorAttribute::OUTPUT);
auto input_tensor = graph->CreateTensor(input_spec);
auto mean = graph->CreateTensor(mean_spec);
auto variance = graph->CreateTensor(variance_spec);
std::vector<float> in_data = {
-2, 0, 2,
-3, 0, 3,
-4, 0, 4,
-5, 0, 5,
-6, 0, 6,
-7, 0, 7
};
std::vector<float> mean_golden = {
-4.5, 0, 4.5
};
std::vector<float> variance_golden = {
2.916666, 0, 2.916666
};
EXPECT_TRUE(input_tensor->CopyDataToTensor(
in_data.data(), in_data.size() * sizeof(float)));
std::vector<int32_t> axes = { 1 };
auto op = graph->CreateOperation<tim::vx::ops::Moments>(axes, true);
(*op).BindInputs({input_tensor}).BindOutputs({mean, variance});
EXPECT_TRUE(graph->Compile());
EXPECT_TRUE(graph->Run());
std::vector<float> mean_output(mean_golden);
std::vector<float> variance_output(variance_golden);
EXPECT_TRUE(mean->CopyDataFromTensor(mean_output.data()));
EXPECT_TRUE(variance->CopyDataFromTensor(variance_output.data()));
EXPECT_TRUE(ArraysMatch(mean_golden, mean_output, 1e-5f, "mean output"));
EXPECT_TRUE(ArraysMatch(variance_golden, variance_output, 1e-5f, "variance output"));
}
#if 0
// TODO: Support uint8
TEST(Moments, shape_3_6_1_uint8_axes_1_keepdims) {
auto ctx = tim::vx::Context::Create();
auto graph = ctx->CreateGraph();
tim::vx::ShapeType input_shape({3, 6, 1});
tim::vx::ShapeType output_shape({3,1,1});
tim::vx::Quantization input_quant(tim::vx::QuantType::ASYMMETRIC, 1, 7);
tim::vx::Quantization mean_quant(tim::vx::QuantType::ASYMMETRIC, 0.5, 9);
tim::vx::Quantization variance_quant(tim::vx::QuantType::ASYMMETRIC, 2.916666, 0);
tim::vx::TensorSpec input_spec(tim::vx::DataType::UINT8,
input_shape, tim::vx::TensorAttribute::INPUT, input_quant);
tim::vx::TensorSpec mean_spec(tim::vx::DataType::UINT8,
output_shape, tim::vx::TensorAttribute::OUTPUT, mean_quant);
tim::vx::TensorSpec variance_spec(tim::vx::DataType::UINT8,
output_shape, tim::vx::TensorAttribute::OUTPUT, variance_quant);
auto input_tensor = graph->CreateTensor(input_spec);
auto mean = graph->CreateTensor(mean_spec);
auto variance = graph->CreateTensor(variance_spec);
std::vector<uint8_t> in_data = {
5, 7, 9,
4, 7, 10,
3, 7, 11,
2, 7, 12,
1, 7, 13,
0, 7, 14,
};
std::vector<uint8_t> mean_golden = {
0, 9, 18
};
std::vector<uint8_t> variance_golden = {
1, 0, 1
};
EXPECT_TRUE(input_tensor->CopyDataToTensor(in_data.data(), in_data.size()));
std::vector<int32_t> axes = { 1 };
auto op = graph->CreateOperation<tim::vx::ops::Moments>(axes, true);
(*op).BindInputs({input_tensor}).BindOutputs({mean, variance});
EXPECT_TRUE(graph->Compile());
EXPECT_TRUE(graph->Run());
std::vector<uint8_t> mean_output(mean_golden);
std::vector<uint8_t> variance_output(variance_golden);
EXPECT_TRUE(mean->CopyDataFromTensor(mean_output.data()));
EXPECT_TRUE(variance->CopyDataFromTensor(variance_output.data()));
EXPECT_EQ(mean_golden, mean_output);
EXPECT_EQ(variance_golden, variance_output);
}
#endif