Add Unstack

Signed-off-by: zhao.xia <zhao.xia@verisilicon.com>
This commit is contained in:
zhao.xia 2021-06-03 15:28:09 +08:00 committed by Kainan Cha
parent aa1137c568
commit 748658e47d
4 changed files with 233 additions and 1 deletions

View File

@ -0,0 +1,52 @@
/****************************************************************************
*
* 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_UNSTACK_H_
#define TIM_VX_OPS_UNSTACK_H_
#include "tim/vx/operation.h"
namespace tim {
namespace vx {
namespace ops {
/**
* ## Unstack
*
* Unpacks the given dimension of a rank-R tensor into rank-(R-1) tensors.
* - axis : An int. The axis to unstack along. Defaults to the first dimension.
* Negative values wrap around, so the valid range is [-R, R).
*/
class Unstack : public Operation {
public:
Unstack(Graph* graph, int32_t axis, uint32_t output_num);
protected:
int32_t axis_;
};
} // namespace ops
} // namespace vx
} // namespace tim
#endif /* TIM_VX_OPS_UNSTACK_H_ */

View File

@ -92,7 +92,7 @@ Linear|LINEAR|Mapped|[tf.keras.activations.linear](https://www.tensorflow.org/ap
ScatterND|SCATTER_ND|Mapped|[tf.scatter_nd](https://tensorflow.google.cn/api_docs/python/tf/scatter_nd)
||MOMENTS|Planned 21Q2|[tf.moments](https://tensorflow.google.cn/api_docs/python/tf/nn/moments)
||MATRIXMUL|Planned 21Q2|[tf.experimental.numpy.matmul](https://www.tensorflow.org/api_docs/python/tf/experimental/numpy/matmul)
||UNSTACK|Planned 21Q2|[tf.unstack](https://tensorflow.google.cn/api_docs/python/tf/unstack)
Unstack|UNSTACK|Mapped|[tf.unstack](https://tensorflow.google.cn/api_docs/python/tf/unstack)
||TILE|Planned 21Q2|[tf.tile](https://tensorflow.google.cn/api_docs/python/tf/tile)
||TOPK|Planned 21Q2|[tf.math.top_k](https://tensorflow.google.cn/api_docs/python/tf/math/top_k)
||GROUPED_CONV2D|Planned 21Q2|[ANEURALNETWORKS_GROUPED_CONV_2D](https://developer.android.com/ndk/reference/group/neural-networks#group___neural_networks_1ggaabbe492c60331b13038e39d4207940e0a847acf8d9f3d2343328c3dbe6d447c50)

40
src/tim/vx/ops/unstack.cc Normal file
View File

@ -0,0 +1,40 @@
/****************************************************************************
*
* 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/unstack.h"
#include "operation_private.h"
#include "vsi_nn_pub.h"
namespace tim {
namespace vx {
namespace ops {
Unstack::Unstack(Graph* graph, int32_t axis, uint32_t output_num)
: Operation(graph, VSI_NN_OP_UNSTACK, 1, output_num), axis_(axis) {
this->impl()->node()->nn_param.unstack.axis = axis_;
}
} // namespace ops
} // namespace vx
} // namespace tim

View File

@ -0,0 +1,140 @@
/****************************************************************************
*
* 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/unstack.h"
#include "gtest/gtest.h"
TEST(Unstack, shape_4_3_axis_0) {
auto ctx = tim::vx::Context::Create();
auto graph = ctx->CreateGraph();
tim::vx::ShapeType input_shape({4,3});
tim::vx::ShapeType output_shape({3});
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 output1_tensor = graph->CreateTensor(output_spec);
auto output2_tensor = graph->CreateTensor(output_spec);
auto output3_tensor = graph->CreateTensor(output_spec);
auto output4_tensor = graph->CreateTensor(output_spec);
std::vector<float> in_data = {
1,2,3,4,
5,6,7,8,
9,10,11,12,
};
std::vector<float> golden1 = {
1,5,9
};
std::vector<float> golden2 = {
2,6,10
};
std::vector<float> golden3 = {
3,7,11
};
std::vector<float> golden4 = {
4,8,12
};
EXPECT_TRUE(input_tensor->CopyDataToTensor(
in_data.data(), in_data.size() * sizeof(float)));
auto op = graph->CreateOperation<tim::vx::ops::Unstack>(0, 4);
(*op).BindInputs({input_tensor}).BindOutputs(
{output1_tensor, output2_tensor, output3_tensor, output4_tensor});
EXPECT_TRUE(graph->Compile());
EXPECT_TRUE(graph->Run());
std::vector<float> output1(golden1.size());
std::vector<float> output2(golden2.size());
std::vector<float> output3(golden3.size());
std::vector<float> output4(golden4.size());
EXPECT_TRUE(output1_tensor->CopyDataFromTensor(output1.data()));
EXPECT_TRUE(output2_tensor->CopyDataFromTensor(output2.data()));
EXPECT_TRUE(output3_tensor->CopyDataFromTensor(output3.data()));
EXPECT_TRUE(output4_tensor->CopyDataFromTensor(output4.data()));
EXPECT_EQ(golden1, output1);
EXPECT_EQ(golden2, output2);
EXPECT_EQ(golden3, output3);
EXPECT_EQ(golden4, output4);
}
TEST(Unstack, shape_4_3_axis_1) {
auto ctx = tim::vx::Context::Create();
auto graph = ctx->CreateGraph();
tim::vx::ShapeType input_shape({4,3});
tim::vx::ShapeType output_shape({4});
tim::vx::Quantization quant(tim::vx::QuantType::ASYMMETRIC, 0.5, 0);
tim::vx::TensorSpec input_spec(tim::vx::DataType::UINT8,
input_shape, tim::vx::TensorAttribute::INPUT, quant);
tim::vx::TensorSpec output_spec(tim::vx::DataType::UINT8,
output_shape, tim::vx::TensorAttribute::OUTPUT, quant);
auto input_tensor = graph->CreateTensor(input_spec);
auto output1_tensor = graph->CreateTensor(output_spec);
auto output2_tensor = graph->CreateTensor(output_spec);
auto output3_tensor = graph->CreateTensor(output_spec);
std::vector<uint8_t> in_data = {
2,4,6,8,
10,12,14,16,
18,20,22,24,
};
std::vector<uint8_t> golden1 = {
2,4,6,8
};
std::vector<uint8_t> golden2 = {
10,12,14,16,
};
std::vector<uint8_t> golden3 = {
18,20,22,24,
};
EXPECT_TRUE(input_tensor->CopyDataToTensor(
in_data.data(), in_data.size() * sizeof(float)));
auto op = graph->CreateOperation<tim::vx::ops::Unstack>(1, 3);
(*op).BindInputs({input_tensor}).BindOutputs(
{output1_tensor, output2_tensor, output3_tensor});
EXPECT_TRUE(graph->Compile());
EXPECT_TRUE(graph->Run());
std::vector<uint8_t> output1(golden1.size());
std::vector<uint8_t> output2(golden2.size());
std::vector<uint8_t> output3(golden3.size());
EXPECT_TRUE(output1_tensor->CopyDataFromTensor(output1.data()));
EXPECT_TRUE(output2_tensor->CopyDataFromTensor(output2.data()));
EXPECT_TRUE(output3_tensor->CopyDataFromTensor(output3.data()));
EXPECT_EQ(golden1, output1);
EXPECT_EQ(golden2, output2);
EXPECT_EQ(golden3, output3);
}