Add map for LogSoftmax
Signed-off-by: zhao.xia <zhao.xia@verisilicon.com>
This commit is contained in:
zhao.xia
Kainan Cha
parent
37f686c34d
commit
a1ba85691a
|
|
@ -0,0 +1,55 @@
|
|||
/****************************************************************************
|
||||
*
|
||||
* 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_LOG_SOFTMAX_H_
|
||||
#define TIM_VX_OPS_LOG_SOFTMAX_H_
|
||||
#include "tim/vx/operation.h"
|
||||
|
||||
namespace tim {
|
||||
namespace vx {
|
||||
namespace ops {
|
||||
|
||||
/**
|
||||
* ## LogSoftmax
|
||||
*
|
||||
* Computes the log softmax activation on the input tensor element-wise, per batch.
|
||||
*
|
||||
* ```
|
||||
* logsoftmax = logits - log(reduce_sum(exp(logits), axis))
|
||||
* ```
|
||||
*/
|
||||
|
||||
class LogSoftmax : public Operation {
|
||||
public:
|
||||
LogSoftmax(Graph* graph, int32_t axis, float beta = 1.f);
|
||||
|
||||
protected:
|
||||
int32_t axis_;
|
||||
float beta_;
|
||||
};
|
||||
|
||||
} // namespace ops
|
||||
} // namespace vx
|
||||
} // namespace tim
|
||||
|
||||
#endif /* TIM_VX_OPS_LOG_SOFTMAX_H_ */
|
||||
|
|
@ -119,7 +119,7 @@ ArgMin|ARGMIN|Mapped|[tf.math.argmin](https://tensorflow.google.cn/api_docs/pyth
|
|||
||GENERATE_PROPOSALS|Unmapped|[ANEURALNETWORKS_GENERATE_PROPOSALS](https://developer.android.com/ndk/reference/group/neural-networks#group___neural_networks_1ggaabbe492c60331b13038e39d4207940e0a72484020f2c41c814de0a7bf93dbbfd4)
|
||||
||DETECTION_POSTPROCESS|Unmapped|[ANEURALNETWORKS_DETECTION_POSTPROCESSING](https://developer.android.com/ndk/reference/group/neural-networks#group___neural_networks_1ggaabbe492c60331b13038e39d4207940e0abd6365933837275bb1f5cde1fd9b8234)
|
||||
||RANDOM_MULTINOMIAL|Unmapped|[ANEURALNETWORKS_RANDOM_MULTINOMIAL](https://developer.android.com/ndk/reference/group/neural-networks#group___neural_networks_1ggaabbe492c60331b13038e39d4207940e0a6cb5032c09d3c4b542d18495c247b5b4)
|
||||
||LOG_SOFTMAX|Unmapped|[tf.nn.log_softmax](https://tensorflow.google.cn/api_docs/python/tf/nn/log_softmax)
|
||||
LogSoftmax|LOG_SOFTMAX|Mapped|[tf.nn.log_softmax](https://tensorflow.google.cn/api_docs/python/tf/nn/log_softmax)
|
||||
||RELU_KERAS|Unmapped|[tf.keras.layers.ReLU](https://tensorflow.google.cn/api_docs/python/tf/keras/layers/ReLU)
|
||||
||GRU_OVXLIB|Unmapped
|
||||
||GRUCELL_OVXLIB|Unmapped
|
||||
|
|
|
|||
|
|
@ -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/logsoftmax.h"
|
||||
|
||||
#include "operation_private.h"
|
||||
#include "vsi_nn_pub.h"
|
||||
|
||||
namespace tim {
|
||||
namespace vx {
|
||||
namespace ops {
|
||||
|
||||
LogSoftmax::LogSoftmax(Graph* graph, int32_t axis, float beta)
|
||||
: Operation(graph, VSI_NN_OP_LOG_SOFTMAX), axis_(axis), beta_(beta) {
|
||||
this->impl()->node()->nn_param.log_softmax.betaValue = beta_;
|
||||
this->impl()->node()->nn_param.log_softmax.axis = axis_;
|
||||
}
|
||||
|
||||
} // namespace ops
|
||||
} // namespace vx
|
||||
} // namespace tim
|
||||
|
|
@ -0,0 +1,162 @@
|
|||
/****************************************************************************
|
||||
*
|
||||
* 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/context.h"
|
||||
#include "tim/vx/graph.h"
|
||||
#include "tim/vx/ops/logsoftmax.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){
|
||||
for (size_t i = 0; i < expected.size(); ++i){
|
||||
EXPECT_NEAR(expected[i], actual[i], abs_error) << "at index:" << i;
|
||||
}
|
||||
|
||||
return ::testing::AssertionSuccess();
|
||||
}
|
||||
}
|
||||
|
||||
TEST(LogSoftmax, shape_6_1_float_axis_0) {
|
||||
auto ctx = tim::vx::Context::Create();
|
||||
auto graph = ctx->CreateGraph();
|
||||
|
||||
tim::vx::ShapeType io_shape({6, 1});
|
||||
tim::vx::TensorSpec input_spec(tim::vx::DataType::FLOAT32,
|
||||
io_shape, tim::vx::TensorAttribute::INPUT);
|
||||
tim::vx::TensorSpec output_spec(tim::vx::DataType::FLOAT32,
|
||||
io_shape, tim::vx::TensorAttribute::OUTPUT);
|
||||
|
||||
auto input_tensor = graph->CreateTensor(input_spec);
|
||||
auto output_tensor = graph->CreateTensor(output_spec);
|
||||
|
||||
std::vector<float> in_data = {
|
||||
2, 3, 4, 5, 6, 7
|
||||
};
|
||||
std::vector<float> golden = {
|
||||
-5.4562, -4.4562, -3.4562, -2.4562, -1.4562, -0.4562,
|
||||
};
|
||||
|
||||
EXPECT_TRUE(input_tensor->CopyDataToTensor(in_data.data(), in_data.size() * sizeof(float)));
|
||||
|
||||
auto op = graph->CreateOperation<tim::vx::ops::LogSoftmax>(0);
|
||||
(*op).BindInputs({input_tensor}).BindOutputs({output_tensor});
|
||||
|
||||
EXPECT_TRUE(graph->Compile());
|
||||
EXPECT_TRUE(graph->Run());
|
||||
|
||||
std::vector<float> output(golden.size() * sizeof(float));
|
||||
EXPECT_TRUE(output_tensor->CopyDataFromTensor(output.data()));
|
||||
EXPECT_TRUE(ArraysMatch(golden, output, 1e-5f));
|
||||
}
|
||||
|
||||
TEST(LogSoftmax, shape_3_6_1_float_axis_1) {
|
||||
auto ctx = tim::vx::Context::Create();
|
||||
auto graph = ctx->CreateGraph();
|
||||
|
||||
tim::vx::ShapeType io_shape({3, 6, 1});
|
||||
tim::vx::TensorSpec input_spec(tim::vx::DataType::FLOAT32,
|
||||
io_shape, tim::vx::TensorAttribute::INPUT);
|
||||
tim::vx::TensorSpec output_spec(tim::vx::DataType::FLOAT32,
|
||||
io_shape, tim::vx::TensorAttribute::OUTPUT);
|
||||
|
||||
auto input_tensor = graph->CreateTensor(input_spec);
|
||||
auto output_tensor = graph->CreateTensor(output_spec);
|
||||
|
||||
std::vector<float> in_data = {
|
||||
-2.0000, 0.0000, 2.0000,
|
||||
-3.0000, 0.0000, 3.0000,
|
||||
-4.0000, 0.0000, 4.0000,
|
||||
-5.0000, 0.0000, 5.0000,
|
||||
-6.0000, 0.0000, 6.0000,
|
||||
-7.0000, 0.0000, 7.0000,
|
||||
};
|
||||
std::vector<float> golden = {
|
||||
-0.4561933, -1.7917595, -5.4561934,
|
||||
-1.4561933, -1.7917595, -4.4561934,
|
||||
-2.4561934, -1.7917595, -3.4561934,
|
||||
-3.4561934, -1.7917595, -2.4561934,
|
||||
-4.4561934, -1.7917595, -1.4561933,
|
||||
-5.4561934, -1.7917595, -0.4561933,
|
||||
};
|
||||
|
||||
EXPECT_TRUE(input_tensor->CopyDataToTensor(in_data.data(), in_data.size() * sizeof(float)));
|
||||
|
||||
auto op = graph->CreateOperation<tim::vx::ops::LogSoftmax>(1);
|
||||
(*op).BindInputs({input_tensor}).BindOutputs({output_tensor});
|
||||
|
||||
EXPECT_TRUE(graph->Compile());
|
||||
EXPECT_TRUE(graph->Run());
|
||||
|
||||
std::vector<float> output(golden.size() * sizeof(float));
|
||||
EXPECT_TRUE(output_tensor->CopyDataFromTensor(output.data()));
|
||||
EXPECT_TRUE(ArraysMatch(golden, output, 1e-5f));
|
||||
}
|
||||
|
||||
TEST(LogSoftmax, shape_3_6_1_uint8_axis_1) {
|
||||
auto ctx = tim::vx::Context::Create();
|
||||
auto graph = ctx->CreateGraph();
|
||||
|
||||
tim::vx::ShapeType io_shape({3, 6, 1});
|
||||
tim::vx::Quantization input_quant(tim::vx::QuantType::ASYMMETRIC, 1, 2);
|
||||
tim::vx::Quantization output_quant(tim::vx::QuantType::ASYMMETRIC, 1.7917595, 2);
|
||||
tim::vx::TensorSpec input_spec(tim::vx::DataType::UINT8,
|
||||
io_shape, tim::vx::TensorAttribute::INPUT, input_quant);
|
||||
tim::vx::TensorSpec output_spec(tim::vx::DataType::UINT8,
|
||||
io_shape, tim::vx::TensorAttribute::OUTPUT, output_quant);
|
||||
|
||||
auto input_tensor = graph->CreateTensor(input_spec);
|
||||
auto output_tensor = graph->CreateTensor(output_spec);
|
||||
|
||||
std::vector<uint8_t> in_data = {
|
||||
0, 2, 4,
|
||||
0, 2, 4,
|
||||
0, 2, 4,
|
||||
0, 2, 4,
|
||||
0, 2, 4,
|
||||
0, 2, 4,
|
||||
};
|
||||
std::vector<uint8_t> golden = {
|
||||
1, 1, 1,
|
||||
1, 1, 1,
|
||||
1, 1, 1,
|
||||
1, 1, 1,
|
||||
1, 1, 1,
|
||||
1, 1, 1,
|
||||
};
|
||||
|
||||
EXPECT_TRUE(input_tensor->CopyDataToTensor(in_data.data(), in_data.size()));
|
||||
|
||||
auto op = graph->CreateOperation<tim::vx::ops::LogSoftmax>(1);
|
||||
(*op).BindInputs({input_tensor}).BindOutputs({output_tensor});
|
||||
|
||||
EXPECT_TRUE(graph->Compile());
|
||||
EXPECT_TRUE(graph->Run());
|
||||
|
||||
std::vector<uint8_t> output(golden.size());
|
||||
EXPECT_TRUE(output_tensor->CopyDataFromTensor(output.data()));
|
||||
EXPECT_EQ(golden, output);
|
||||
}
|
||||
Loading…
Reference in New Issue