new api trunc, test=develop #33371

zhangbo9674 · 2021-06-07T03:19:45Z

PR types

New features

PR changes

APIs

Describe

Create a new paddle API: paddle.trunc(x, name=none).

API描述：
对一个输入tensor，返回一个新的tensor，其中包含输入的截断整数值。支持的数据类型包括：int32、int64、float、double。

2.新增内容综述：
（1）新增两个Op：正向Op：trunc、对应的反向Op：trunc_grad；
（2）新增python API：paddle.trunc()；
（3）新增单测文件：test_trunc_op.py；

3.API使用示例：

import paddle

input = paddle.rand([2,2],'float32')
print(input)
# Tensor(shape=[2, 2], dtype=float32, place=CUDAPlace(0), stop_gradient=True,
#         [[0.02331470, 0.42374918],
#         [0.79647720, 0.74970269]])

output = paddle.trunc(input)
print(output)
# Tensor(shape=[2, 2], dtype=float32, place=CUDAPlace(0), stop_gradient=True,
#         [[0., 0.],
#         [0., 0.]]))

4.文档预览截图：

paddle-bot-old · 2021-06-07T03:19:48Z

Thanks for your contribution!
Please wait for the result of CI firstly. See Paddle CI Manual for details.

zhiqiu · 2021-06-07T03:30:52Z

paddle/fluid/operators/trunc_op.cu

+__global__ void TruncGrad(const T* dout, T* dx, int N) {
+  CUDA_KERNEL_LOOP(index, N) { dx[index] = 0.0; }
+}


The input argument dx is not needed

Done, thanks!

zhiqiu · 2021-06-07T03:31:57Z

paddle/fluid/operators/trunc_op.cc

+
+  void Apply(GradOpPtr<T> retv) const override {
+    retv->SetType("trunc_grad");
+    retv->SetInput(framework::GradVarName("Out"), this->OutputGrad("Out"));


You can register NoNeedBufferVars for X@GRAD to save memory, see https://www.paddlepaddle.org.cn/documentation/docs/zh/guides/07_new_op/op_notes_cn.html#id6 for details.

Done, thanks!

zhiqiu · 2021-06-07T03:32:56Z

paddle/fluid/operators/trunc_op.cu

+    const T* x_data = x->data<T>();
+    T* out_data = out->mutable_data<T>(context.GetPlace());
+
+    int numel = x->numel();


int -> int64_t, since there are tensors of huge size.

Done, thanks!

zhiqiu · 2021-06-07T03:34:14Z

python/paddle/tensor/math.py

+            #        [0., 0.]])
+    '''
+    if in_dygraph_mode():
+        out = _varbase_creator(dtype=x.dtype)


Unused out.

Done, thanks!

zhiqiu · 2021-06-07T03:35:06Z

python/paddle/tensor/math.py

+        attrs = {}
+
+        helper = LayerHelper("trunc", **locals())
+        check_variable_and_dtype(x, 'X', ['float16', 'float32', 'float64'], 'trunc')


It seems float16 is not supported in c++ code.

Done, thanks!

zhiqiu · 2021-06-07T03:35:54Z

python/paddle/tensor/math.py

+    Examples:
+        .. code-block:: python
+            import paddle
+            import numpy as np


You can use paddle.rand() directly instead of numpy.random.random

Done, thanks!

zhiqiu · 2021-06-07T03:37:05Z

python/paddle/fluid/tests/unittests/test_trunc_op.py

+        self.outputs = {'Out': (np.trunc(self.inputs['X']))}
+
+    def init_dtype_type(self):
+        pass


Why not put self.dtype = np.float64 here?

Done, thanks!

zhiqiu · 2021-06-07T03:38:46Z

paddle/fluid/operators/trunc_op.h

+/* Copyright (c) 2021 PaddlePaddle Authors. All Rights Reserved.
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+    http://www.apache.org/licenses/LICENSE-2.0
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License. */


There are blank lines in copyright.

Done, thanks!

is it changed?

Now it is changed, done, thanks!

zhiqiu · 2021-06-09T03:48:50Z

paddle/fluid/operators/trunc_op.cu

+namespace operators {
+
+template <typename T>
+class truncFunctor {


Suggested change

class truncFunctor {

class TruncFunctor {

Done, thanks!

zhiqiu · 2021-06-09T03:51:41Z

paddle/fluid/operators/trunc_op.cu

+ public:
+  __device__ truncFunctor(const T x) : _x(x) {}
+  __device__ T operator()() { return trunc(_x); }
+  const T _x;


x -> x, plz refer to google code stype: https://google.github.io/styleguide/cppguide.html.

Done, thanks

zhiqiu · 2021-06-09T03:53:56Z

paddle/fluid/operators/trunc_op.cu

+
+template <typename T>
+__global__ void TruncGrad(T* dx, int64_t N) {
+  CUDA_KERNEL_LOOP(index, N) { dx[index] = 0.0; }


Suggested change

CUDA_KERNEL_LOOP(index, N) { dx[index] = 0.0; }

CUDA_KERNEL_LOOP(index, N) {

dx[index] = static_cast<T>(0.0);

}

Done, thanks

zhiqiu · 2021-06-09T04:00:06Z

paddle/fluid/operators/trunc_op.cu

+    dim3 blockSize(256);
+    dim3 gridSize((numel + blockSize.x - 1) / blockSize.x);


theads = platform::PADDLE_CUDA_NUM_THREADS;
blocks = (numel + PADDLE_CUDA_NUM_THREADS - 1) / PADDLE_CUDA_NUM_THREADS

Done, thanks!

zhiqiu · 2021-06-09T04:00:50Z

paddle/fluid/operators/trunc_op.h

+/* Copyright (c) 2021 PaddlePaddle Authors. All Rights Reserved.
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+    http://www.apache.org/licenses/LICENSE-2.0
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License. */


is it changed?

lanxianghit · 2021-06-09T08:50:46Z

python/paddle/tensor/math.py

@@ -855,6 +855,44 @@ def add_n(inputs, name=None):
    return out


+def trunc(input, name=None):
+    '''


新增API需要文档预览截图，并且需要同时提交中文文档

（1）新增API文档预览截图：

（2）中文文档pr：
PaddlePaddle/docs#3585

lanxianghit · 2021-06-09T08:54:07Z

python/paddle/tensor/math.py

+
+            paddle.set_device('cpu')
+            input = paddle.rand([2,2],'float32')
+            output = paddle.trunc(input)


没有给出原始tensor内容，可能会影响用户对行为的理解

已添加原始tensor内容，谢谢！

lanxianghit · 2021-06-09T08:54:24Z

python/paddle/tensor/math.py

+
+            import paddle
+
+            paddle.set_device('cpu')


为什么要特意设置CPU？

没有必要特意设置CPU，已删除，谢谢！

lanxianghit · 2021-06-09T09:00:18Z

python/paddle/tensor/math.py

+            paddle.set_device('cpu')
+            input = paddle.rand([2,2],'float32')
+            output = paddle.trunc(input)
+            print(output)


注释给出的输出与代码不符，可以改为print(output.numpy())，或者把注释信息改为tensor的输出

已将注释信息改为tensor的输出，谢谢！

…dev/trunc

zhiqiu

LGTM

lanxianghit

LGTM

TCChenlong · 2021-06-11T01:33:22Z

python/paddle/tensor/math.py

+    This API is used to returns a new tensor with the truncated integer values of input.
+
+    Args:
+        input (Tensor): The input tensor, it's data type should be int, float, double.


data type 应具体一点，如 int8、int16、int32、float32 等；

Done, thanks!

TCChenlong

LGTM

… dev/trunc

zhiqiu

LGTM

new api trunc, test=develop

35aebd5

zhiqiu reviewed Jun 7, 2021

View reviewed changes

zhangbo9674 closed this Jun 7, 2021

zhangbo9674 reopened this Jun 7, 2021

zhangbo9674 added 3 commits June 7, 2021 05:54

new api trunc, test=develop

cd716f2

new api paddle.trunc, test=develop

33cc89b

new api paddle.trunc, test=develop

b577296

zhiqiu reviewed Jun 9, 2021

View reviewed changes

zhangbo9674 added 4 commits June 9, 2021 05:16

new api paddle.trunc, test=develop

9a6272b

new api paddle.trunc, test=develop

de107a3

new api paddle.trunc, test=develop

58b19cb

new api paddle.trunc, test=develop

b7b5803

lanxianghit reviewed Jun 9, 2021

View reviewed changes

zhangbo9674 added 2 commits June 9, 2021 09:21

new api paddle.trunc, test=develop

f76b32e

new api paddle.trunc, test=develop

8158a58

zhangbo9674 added 2 commits June 10, 2021 02:15

Merge branch 'develop' of https://github.com/zhangbo9674/Paddle into …

feaeaa4

…dev/trunc

new api paddle.trunc, test=develop

b7f18f9

zhangbo9674 force-pushed the dev/trunc branch from cc37cfd to b7f18f9 Compare June 10, 2021 02:45

zhiqiu previously approved these changes Jun 10, 2021

View reviewed changes

lanxianghit previously approved these changes Jun 10, 2021

View reviewed changes

TCChenlong reviewed Jun 11, 2021

View reviewed changes

new api paddle.trunc, test=develop

2783560

zhangbo9674 dismissed stale reviews from lanxianghit and zhiqiu via 2783560 June 11, 2021 02:05

TCChenlong previously approved these changes Jun 11, 2021

View reviewed changes

zhiqiu previously approved these changes Jun 11, 2021

View reviewed changes

lanxianghit previously approved these changes Jun 11, 2021

View reviewed changes

zhangbo9674 dismissed lanxianghit’s stale review via 982b57b June 11, 2021 06:45

zhangbo9674 dismissed stale reviews from zhiqiu and TCChenlong via 982b57b June 11, 2021 06:45

zhiqiu previously approved these changes Jun 15, 2021

View reviewed changes

zhangbo9674 dismissed zhiqiu’s stale review via c60830a June 15, 2021 02:59

Merge branch 'develop' of https://github.com/PaddlePaddle/Paddle into…

df79e49

… dev/trunc

zhangbo9674 force-pushed the dev/trunc branch from c60830a to df79e49 Compare June 15, 2021 05:54

zhiqiu approved these changes Jun 15, 2021

View reviewed changes

TCChenlong approved these changes Jun 16, 2021

View reviewed changes

lanxianghit approved these changes Jun 16, 2021

View reviewed changes

zhiqiu merged commit 72d3697 into PaddlePaddle:develop Jun 16, 2021

TCChenlong mentioned this pull request Mar 8, 2022

【PaddlePaddle Hackathon 2】27、为 Paddle 新增 frac 数学计算API #40306

Closed

zhangbo9674 deleted the dev/trunc branch September 14, 2022 02:22

		dim3 blockSize(256);
		dim3 gridSize((numel + blockSize.x - 1) / blockSize.x);

new api trunc, test=develop #33371

new api trunc, test=develop #33371

Conversation

zhangbo9674 commented Jun 7, 2021 • edited Loading

PR types

PR changes

Describe

paddle-bot-old bot commented Jun 7, 2021

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

zhangbo9674 Jun 10, 2021 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

zhiqiu left a comment

Choose a reason for hiding this comment

lanxianghit left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

TCChenlong left a comment

Choose a reason for hiding this comment

zhiqiu left a comment

Choose a reason for hiding this comment

zhangbo9674 commented Jun 7, 2021 •

edited

Loading

zhangbo9674 Jun 10, 2021 •

edited

Loading