[MXNET-483] C++ tests for mkldnn convolution/deconvolution operator #11778

azai91 · 2018-07-17T07:02:48Z

Description

Add tests for MKLDNN convolution / deconvolution operator. Need to convert MKLDNN conv/deconv inputs if they are views (#12303).

Checklist

Essentials

Please feel free to remove inapplicable items for your PR.

The PR title starts with [MXNET-$JIRA_ID], where $JIRA_ID refers to the relevant JIRA issue created (except PRs with tiny changes)
Changes are complete (i.e. I finished coding on this PR)
All changes have test coverage:
Unit tests are added for small changes to verify correctness (e.g. adding a new operator)
Nightly tests are added for complicated/long-running ones (e.g. changing distributed kvstore)
Build tests will be added for build configuration changes (e.g. adding a new build option with NCCL)
Code is well-documented:
For user-facing API changes, API doc string has been updated.
For new C++ functions in header files, their functionalities and arguments are documented.
For new examples, README.md is added to explain the what the example does, the source of the dataset, expected performance on test set and reference to the original paper if applicable
Check the API doc at http://mxnet-ci-doc.s3-accelerate.dualstack.amazonaws.com/PR-$PR_ID/$BUILD_ID/index.html
To the my best knowledge, examples are either not affected by this change, or have been fixed to be compatible with this change

Changes

Add tests for MKLDNN convolution / deconvolution operator
Reorder inputs of conv / deconv mkldnn inputs if MKLDNN and view.

Comments

If this change is a backward incompatible change, why must this change be made.
Interesting edge cases to note here

azai91 · 2018-07-19T16:41:31Z

@zheng-da can you review

zheng-da · 2018-07-20T07:19:51Z

can you test both conv and deconv? these two operators have almost the same inputs.

azai91 · 2018-08-01T06:25:47Z

@mseth10 please take a look as well

zheng-da · 2018-08-01T06:41:30Z

tests/cpp/operator/mkldnn.cc

+      backwards_ex_outputs[2] = &tmp_bias2;
+
+      for (int i = 0; i < backwards_attrs.num_outputs; i++)
+        back_req[0] = kWriteTo;


only WriteTo? I think you should test AddTo as well.

regular fcompute for deconvolution/convolution does not support kAddTo currently
https://github.com/azai91/incubator-mxnet/blob/d79e1ad3294837cac653478045023fd312ceed78/src/operator/nn/convolution-inl.h#L178

https://github.com/azai91/incubator-mxnet/blob/46e47cbc6183d2812a2e405851f0b209383e72ad/src/operator/nn/deconvolution-inl.h#L223

larroy · 2018-08-02T21:29:10Z

tests/cpp/operator/mkldnn.cc

      continue;
-    shape[dim] = shape[dim] * num_inputs;
+
+    for (int dim = 0; dim < scale.size(); dim++)


right type is size_t, ++dim (pre-increment) is standard C++ idiom

larroy · 2018-08-02T21:29:20Z

tests/cpp/operator/mkldnn.cc


    // Type 1.
    NDArray arr(shape, Context());
    in_arrs.emplace_back(arr, "Normal NDArray");
    InitDefaultArray(&in_arrs.back().arr, rand);
    for (auto pd : pds) {
-      if (num_inputs > 1) {
+      for (int dim = 0; dim < scale.size(); dim++) {


size_t, otherwise signed vs unsigned comparison warning

eric-haibin-lin · 2018-11-13T23:08:15Z

@TaoLv this PR already got many reviews and one approval. Is it okay to merge this PR, or you have further unaddressed comments?

TaoLv · 2018-11-14T01:46:25Z

@eric-haibin-lin, need @azai91 's inputs about the difference of this PR and #13084.

azai91 · 2018-11-14T02:00:22Z

#13084 is a branch off of this PR that includes the BN unit tests. we should merge this one first and then we can more easily review #13084. @TaoLv @eric-haibin-lin

TaoLv · 2018-11-14T02:17:52Z

tests/cpp/include/test_mkldnn.h

+
+inline NDArray CreateKernelNDArray(TShape kernel, int num_filters, TShape input,
+    bool is_deconv = false) {
+  CHECK(kernel.ndim() == 2) << "mkldnn only supports 2d filters on 4d inputs";


TaoLv · 2018-11-14T02:26:02Z

Please make sure that weight format NDAarray is not send to MKL-DNN operators as input data.
Take care of the performance. I have the same concern as @zheng-da .
Please clean compile warnings of MKL-DNN cpp tests.

cc @pengzhao-intel

azai91 · 2018-11-15T23:03:31Z

@TaoLv I checked that this PR does not cause regression.

azai91 · 2018-11-15T23:04:26Z

what MKL-DNN warnings? are they introduced in this PR or were they always present? if they were already present, then let's file an issue and tackle in a separate PR.

TaoLv · 2018-11-18T11:46:10Z

Thanks, @azai91 . I noticed that there were many warnings when built mkldnn cpp test with Makefile. It would be great if you can take a look and fix them. Not necessarily in this PR.

lupesko · 2018-11-19T08:28:33Z

@TaoLv and @pengzhao-intel if you are good with this PR, please approve.
This is critical to allow @azai91 to finally get this merged.

anirudh2290 · 2018-11-19T22:11:51Z

src/operator/nn/mkldnn/mkldnn_deconvolution.cc

    } else {
      weight_mem = weight.GetMKLDNNData();
      CHECK(weight_mem->get_primitive_desc() == fwd_pd.weights_primitive_desc());
    }
+    weight_mem = weight.GetMKLDNNData();
+    CHECK(weight_mem->get_primitive_desc() == fwd_pd.weights_primitive_desc());


do we need else if these two lines are going to execute irrespective

anirudh2290 · 2018-11-19T22:31:30Z

The warnings are related to batch norm test and implicit_gemm in mshadow. Since they are not being modified in this PR, I think its fine if they are not fixed in this PR.

pengzhao-intel · 2018-11-20T04:30:05Z

@azai91 @anirudh2290 thanks for the help and great works to improve the quality of MKL-DNN.

…erator (#11778)" This reverts commit 91c536d.

azai91 requested a review from anirudh2290 as a code owner July 17, 2018 07:02

azai91 force-pushed the test/mkldnn-conv-op branch from 0ffb0fe to 3d33273 Compare July 18, 2018 22:13

azai91 changed the title ~~[MXNET-483] C++ tests for convolution operator~~ [MXNET-483] C++ tests for mkldnn convolution operator Jul 19, 2018

azai91 force-pushed the test/mkldnn-conv-op branch 5 times, most recently from 1cb5052 to 6750e75 Compare July 31, 2018 00:55

zheng-da reviewed Aug 1, 2018

View reviewed changes

azai91 force-pushed the test/mkldnn-conv-op branch 2 times, most recently from 0bba2a0 to d1a6fdb Compare August 2, 2018 21:06

larroy reviewed Aug 2, 2018

View reviewed changes

azai91 force-pushed the test/mkldnn-conv-op branch 4 times, most recently from 56f575d to 6cfd95f Compare August 3, 2018 02:36

azai91 added 10 commits August 3, 2018 12:58

add conv test

f41d6b1

remove pool type

82859ad

gettestinput can receive scale input

43195f8

create kernels and bias arrays

888fb95

fix shape of kernel / bias

e95fef4

fix format

efd41aa

bias is 1dim

5eb5231

fix output shape

736698b

fix backwards input

64c4ba6

fix var name to backwards_ex_outputs

825c9ef

azai91 added 7 commits November 8, 2018 12:52

merge from master

e0a36c9

fix lint

2658953

retrigger

68713e6

Merge branch 'master' into test/mkldnn-conv-op

3995a16

remove data format

24323f6

retrigger

d2428aa

retrigger

272fed4

zheng-da approved these changes Nov 13, 2018

View reviewed changes

TaoLv reviewed Nov 14, 2018

View reviewed changes

azai91 added 4 commits November 14, 2018 10:40

use check_eq

bfdc3d5

Merge branch 'master' into test/mkldnn-conv-op

d9ca2a1

retrriger

5cec3ce

merge from master

bebe81f

anirudh2290 reviewed Nov 19, 2018

View reviewed changes

refactor deconv if/else block

ce97f14

anirudh2290 approved these changes Nov 19, 2018

View reviewed changes

anirudh2290 merged commit 91c536d into apache:master Nov 20, 2018

lanking520 added a commit that referenced this pull request Nov 21, 2018

Revert "[MXNET-483] C++ tests for mkldnn convolution/deconvolution op…

51d5e45

…erator (#11778)" This reverts commit 91c536d.

TaoLv mentioned this pull request Nov 27, 2018

MKL-DNN deconvolution runs into crash #13421

Closed

azai91 mentioned this pull request Nov 28, 2018

Fix deconvolution / PR 13421 #13433

Merged

6 tasks

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

[MXNET-483] C++ tests for mkldnn convolution/deconvolution operator #11778

[MXNET-483] C++ tests for mkldnn convolution/deconvolution operator #11778

azai91 commented Jul 17, 2018 •

edited

Loading

azai91 commented Jul 19, 2018 •

edited

Loading

zheng-da commented Jul 20, 2018

azai91 commented Aug 1, 2018 •

edited

Loading

zheng-da Aug 1, 2018

azai91 Aug 2, 2018

larroy Aug 2, 2018

larroy Aug 2, 2018 •

edited

Loading

eric-haibin-lin commented Nov 13, 2018

TaoLv commented Nov 14, 2018

azai91 commented Nov 14, 2018 •

edited

Loading

TaoLv Nov 14, 2018

TaoLv commented Nov 14, 2018

azai91 commented Nov 15, 2018

azai91 commented Nov 15, 2018

TaoLv commented Nov 18, 2018

lupesko commented Nov 19, 2018

anirudh2290 Nov 19, 2018

anirudh2290 commented Nov 19, 2018

pengzhao-intel commented Nov 20, 2018

[MXNET-483] C++ tests for mkldnn convolution/deconvolution operator #11778

[MXNET-483] C++ tests for mkldnn convolution/deconvolution operator #11778

Conversation

azai91 commented Jul 17, 2018 • edited Loading

Description

Checklist

Essentials

Changes

Comments

azai91 commented Jul 19, 2018 • edited Loading

zheng-da commented Jul 20, 2018

azai91 commented Aug 1, 2018 • edited Loading

zheng-da Aug 1, 2018

Choose a reason for hiding this comment

azai91 Aug 2, 2018

Choose a reason for hiding this comment

larroy Aug 2, 2018

Choose a reason for hiding this comment

larroy Aug 2, 2018 • edited Loading

Choose a reason for hiding this comment

eric-haibin-lin commented Nov 13, 2018

TaoLv commented Nov 14, 2018

azai91 commented Nov 14, 2018 • edited Loading

TaoLv Nov 14, 2018

Choose a reason for hiding this comment

TaoLv commented Nov 14, 2018

azai91 commented Nov 15, 2018

azai91 commented Nov 15, 2018

TaoLv commented Nov 18, 2018

lupesko commented Nov 19, 2018

anirudh2290 Nov 19, 2018

Choose a reason for hiding this comment

anirudh2290 commented Nov 19, 2018

pengzhao-intel commented Nov 20, 2018

azai91 commented Jul 17, 2018 •

edited

Loading

azai91 commented Jul 19, 2018 •

edited

Loading

azai91 commented Aug 1, 2018 •

edited

Loading

larroy Aug 2, 2018 •

edited

Loading

azai91 commented Nov 14, 2018 •

edited

Loading