GraphExecutor: Fix wild pointer assign when input and output are reshape #15121
Conversation
|
Thanks for contributing to TVM! Please refer to the contributing guidelines https://tvm.apache.org/docs/contribute/ for useful information and tips. Please request code reviews from Reviewers by @-ing them in a comment.
Generated by tvm-bot |
echuraev
left a comment
echuraev
left a comment
There was a problem hiding this comment.
Could you please add correspondent unit tests?
|
|
|
||
| if (param.func_name == "__nop") { | ||
| return {[]() {}, arg_ptr}; | ||
| return {[arg_ptr]() {}, arg_ptr}; |
There was a problem hiding this comment.
What is the reason of capturing arg_ptr in empty lambda?
There was a problem hiding this comment.
IMHO, I would say that better use zero instead of 0 in the name of the file, e.g test_reshape_zero_copy.py.
Also, probably you can move this test to the file with existing tests on zero_copy.
echuraev
left a comment
echuraev
left a comment
There was a problem hiding this comment.
In general LGTM. Just a couple of moments which were described in previous comment.
|
oh also @vinx13 was involved sorry, forgot to mention |
|
@corehalt The PR needs to resolve the conflict. Feel free to take over this PR and create a new one. |
|
Yes it's fine. |
6 participants
If reshape in our relay expression, like reshape + matmul expr, when we use set_input_zero_copy or set_output_zero_copy，the result goes wrong.