[caffe2] Support group filter in IDEEP fusion#9767
[caffe2] Support group filter in IDEEP fusion#9767gujinghui wants to merge 3 commits intopytorch:masterfrom
Conversation
|
We need to update mkl-dnn module, no? |
|
@yinghai |
|
@yinghai MKL-DNN supports CONV+SUM fusion for all conv kernels except for GEMM conv. We reported the issue and they are fixing the problem now. |
|
@gujinghui What's the status of this PR now? |
add886f to
a2cb870
Compare
|
@yinghai |
facebook-github-bot
left a comment
There was a problem hiding this comment.
@yinghai has imported this pull request. If you are a Facebook employee, you can view this diff on Phabricator.
yinghai
left a comment
There was a problem hiding this comment.
Do we have any unit tests for this?
a2cb870 to
cf6b479
Compare
|
@yinghai failed test cases should be not relevant. |
facebook-github-bot
left a comment
There was a problem hiding this comment.
@yinghai has imported this pull request. If you are a Facebook employee, you can view this diff on Phabricator.
|
@yinghai Thanks a lot. |
c3f89c2 to
d40784e
Compare
|
@yinghai |
|
case failed due to this PR. will fix it soon. |
|
root-caused. need time to figure out how to fix it. :( |
d40784e to
49f36de
Compare
|
@yinghai
|
| #include "caffe2/opt/fusion.h" | ||
|
|
||
| #ifdef CAFFE2_USE_MKLDNN | ||
| #include <cpuinfo.h> |
There was a problem hiding this comment.
Where does this come from?
There was a problem hiding this comment.
the src of cpuinfo can be found under third_party/cpuinfo.
it is static built in to caffe2 lib by default.
it is introduced here to check if the avx512 is supported on current machine.
because the conv + sum fusion is only available on avx512 path so far.
facebook-github-bot
left a comment
There was a problem hiding this comment.
@yinghai has imported this pull request. If you are a Facebook employee, you can view this diff on Phabricator.
facebook-github-bot
left a comment
There was a problem hiding this comment.
@yinghai has imported this pull request. If you are a Facebook employee, you can view this diff on Phabricator.
yinghai
left a comment
There was a problem hiding this comment.
It fails ASAN tests in our CI.
|
hi @yinghai |
|
Yeah, the ASAN msg is not very obvious, complaining wide pointer access. Somewhere in the Ideep code it's corrupting the memory. |
Signed-off-by: Gu, Jinghui <jinghui.gu@intel.com>
…port Signed-off-by: Gu, Jinghui <jinghui.gu@intel.com>
Signed-off-by: Gu, Jinghui <jinghui.gu@intel.com>
The grouped filter has been supported in mkldnn.
This commit removed the check to enable it in IDEEP fusion.