[NNAdapter] Add fill_range fuse pass、eliminate range op、move scale and expand_v2 op #7200
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
…into fix_fill_op
…into fix_fill_op
zhupengyang
reviewed
Oct 13, 2021
lite/backends/nnadapter/nnadapter/driver/huawei_ascend_npu/converter/all.h
Outdated
Show resolved
Hide resolved
shentanyue
force-pushed
the
fix_fill_op
branch
from
b511b70
to
ebaac47
Compare
shentanyue
force-pushed
the
fix_fill_op
branch
from
17df392
to
61bde22
Compare
| shape_data = reinterpret_cast<int32_t*>(shape_operand->buffer); \ | ||
| } else { \ | ||
| shape_count = shape_operand->type.dimensions.count; \ | ||
| shape_data = shape_operand->type.dimensions.data; \ |
There was a problem hiding this comment.
为什么要等于shape_operand的维度data?
这种情况可能是来自于一个temporary shape operand , 可以参考 fill 的实现
| @@ -34,10 +34,13 @@ int ConvertExpand(Converter* converter, hal::Operation* operation) { | |||
| if (!input_operator) { | |||
| input_operator = converter->ConvertOperand(input_operand); | |||
| } | |||
| auto shape_operator = converter->ConvertOperand(shape_operand); | |||
| auto expand_op = converter->AddOperator<ge::op::Expand>(output_operand); | |||
| auto expand_op = converter->AddOperator<ge::op::ExpandD>(output_operand); | |||
There was a problem hiding this comment.
为什么不用之前的ge::op::Expand支持shape作为输入，而是用只支持shape属性的ExpandD？
There was a problem hiding this comment.
因为在跑ppyolo模型时，expand算子出现了bug。后来参照训练那边的实现，同时改了两处地方修复了。
1、是expand支持为-1的情况
2、是改成ge::op::ExpandD
shentanyue
force-pushed
the
fix_fill_op
branch
from
9307d4c
to
30ad4ca
Compare
…into fix_fill_op test=huawei_ascend_npu
…into fix_fill_op test=huawei_ascend_npu
shentanyue
force-pushed
the
fix_fill_op
branch
from
ebc386e
to
e55d24f
Compare
| uint32_t shape_count; \ | ||
| int32_t* shape_data; \ | ||
| auto& shape_type = shape_operand->type; \ | ||
| if (IsConstantOperand(shape_operand)) { \ |
There was a problem hiding this comment.
不建议把这些判断放在.h里面，将其放在.cc的infershape里面吧，而且else部分的代码有问题，shape_operand->type.dimensions.data 指的是 shape_operand 自己的维度信息
修复PR：#7480 |
csy0225
pushed a commit
to csy0225/Paddle-Lite
that referenced
this pull request
Oct 29, 2021
zhupengyang
pushed a commit
that referenced
this pull request
Nov 1, 2021
* [NNAdapter] Move stack op to converter (#7332) * [NNAdapter] Migrate clip, batch_norm, reduce_mean, transpose and pow op bridges to the new converter framework (#7297) * [NNAdapter] Add fill_range fuse pass、eliminate range op、move scale and expand_v2 op (#7200) * [NNAdapter] Migrate cast, norm, pad, dropout converter test=develop,test=huawei_ascend_npu (#7455)
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.
No description provided.