[NPU] Support npu op index_select #34611
Merged
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| # case 1: | ||
| with program_guard(Program(), Program()): | ||
| x = fluid.layers.data(name='x', shape=[-1, 4], dtype='float32') | ||
| index = fluid.layers.data( |
There was a problem hiding this comment.
目前不推荐使用fluid下的API了，例如，fluid.layers.data替换为paddle.static.data，下面其他的接口都类似，查阅下文档
There was a problem hiding this comment.
Done. 所有fluid相关API已remove并替换。
| index = fluid.layers.data( | ||
| name='index', shape=[3], dtype='int32', append_batch_size=False) | ||
| z = paddle.index_select(x, index, axis=1) | ||
| #exe = fluid.Executor(fluid.CPUPlace()) |
|
|
||
| import paddle.fluid as fluid | ||
| import paddle.fluid.core as core | ||
| from paddle.fluid import Program, program_guard |
There was a problem hiding this comment.
使用paddle.static.program_guard
| self.input_data() | ||
|
|
||
| # case 1: | ||
| with fluid.dygraph.guard(): |
There was a problem hiding this comment.
目前框架默认使用动态图执行，可以通过enable_static或者disable_static接口来控制。别用fluid.dygraph.guard()了
| # case 1: | ||
| with fluid.dygraph.guard(): | ||
| x = fluid.dygraph.to_variable(self.data_x) | ||
| index = fluid.dygraph.to_variable(self.data_index) |
There was a problem hiding this comment.
动态图组网参考下paddle.index_select，to_variable -> to_tensor
qili93
requested changes
Aug 5, 2021
| #include <memory> | ||
| #include <string> | ||
| #include <vector> | ||
| #include "paddle/fluid/framework/tensor_util.h" |
|
|
||
| def config(self): | ||
| self.x_shape = (100, 4, 5) | ||
| self.x_type = np.float64 |
There was a problem hiding this comment.
确认下你的index_select算子没有注册double类型的数据，为什么这里设置的x_type是float64的情况下单测可以跑过呢？看看是不是其实fall back到CPU算子上了？
There was a problem hiding this comment.
笔误，已改正，float64->float32。PR描述中的测试结果是对应float32的测试，没有问题。
limin2021
force-pushed
the
add_npu_index_select
branch
from
d9f9a33 to
705514d
Compare
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
PR types
New features
PR changes
OPs
Describe
[NPU] Support npu op index_select (forward op)
说明：
1.关于数据类型
cpu端的index_select算子支持的数据类型有float, double, int, int64；相比cpu端，npu端没有支持double，因为index_select前向是通过调用CANN GatherV2 API 实现的，该 API 不支持double类型。
2.该PR在npu端的单测覆盖了cpu端的所有测试情况。
3.本PR实现了index_select的前向计算，添加反向时，可参考代码中标注'todo'的注释。
运行结果
单测运行结果
调用 index_select npu kernel
