Adding quantized::conv2d function for pytorch mobile in c10 #26152

supriyar · 2019-09-13T03:25:29Z

Stack from ghstack:

Unify Quantization APIs for add, pool and relu #26335 Unify Quantization APIs for add, pool and relu
Changes to support int8 weight and fp32 bias in QNNPACK #26307 Changes to support int8 weight and fp32 bias in QNNPACK
Add support to call unpack for pytorch mobile quantized FC and Conv #26211 Add support to call unpack for pytorch mobile quantized FC and Conv
Adding quantized::conv2d function for pytorch mobile in c10 #26152 Adding quantized::conv2d function for pytorch mobile in c10

Summary:
This change adds the support to call QNNPACK using the refactored API for Conv2d operators

Test Plan:
python test/test_quantized.py TestQNNPackOps.test_qconv_qnnpack

Reviewers:

Subscribers:

Tasks:

Tags:

Differential Revision: D17459892

Summary: This change adds the support to call QNNPACK using the refactored API for Conv2d operators Test Plan: python test/test_quantized.py TestQNNPackOps.test_qconv_qnnpack Reviewers: Subscribers: Tasks: Tags:

Summary: This change adds the support to call QNNPACK using the refactored API for Conv2d operators Test Plan: python test/test_quantized.py TestQNNPackOps.test_qconv_qnnpack Reviewers: Subscribers: Tasks: Tags: ghstack-source-id: b390f37 Pull Request resolved: #26152

Summary: This change adds the support to call QNNPACK using the refactored API for Conv2d operators Test Plan: python test/test_quantized.py TestQNNPackOps.test_qconv_qnnpack Reviewers: Subscribers: Tasks: Tags:

Summary: This change adds the support to call QNNPACK using the refactored API for Conv2d operators Test Plan: python test/test_quantized.py TestQNNPackOps.test_qconv_qnnpack Reviewers: Subscribers: Tasks: Tags: ghstack-source-id: 01cc178 Pull Request resolved: #26152

Summary: This change adds the support to call QNNPACK using the refactored API for Conv2d operators Test Plan: python test/test_quantized.py TestQNNPackOps.test_qconv_qnnpack Reviewers: Subscribers: Tasks: Tags:

Summary: This change adds the support to call QNNPACK using the refactored API for Conv2d operators Test Plan: python test/test_quantized.py TestQNNPackOps.test_qconv_qnnpack Reviewers: Subscribers: Tasks: Tags: ghstack-source-id: aa6a36d Pull Request resolved: #26152

test/test_quantized.py

aten/src/ATen/native/quantized/cpu/qconv_prepack.cpp

dzhulgakov

looking good, some comments

aten/src/ATen/native/quantized/cpu/qconv.cpp

aten/src/ATen/native/quantized/cpu/qconv_prepack.cpp

test/test_quantized.py

Summary: This change adds the support to call QNNPACK using the refactored API for Conv2d operators Test Plan: python test/test_quantized.py TestQNNPackOps.test_qconv_qnnpack Reviewers: Subscribers: Tasks: Tags:

test/test_quantized.py

aten/src/ATen/native/quantized/cpu/qconv.cpp

Summary: This change adds the support to call QNNPACK using the refactored API for Conv2d operators Test Plan: python test/test_quantized.py TestQNNPackOps.test_qconv_qnnpack Reviewers: Subscribers: Tasks: Tags: ghstack-source-id: b33ea74 Pull Request resolved: pytorch#26152

Summary: This change adds the support to call QNNPACK using the refactored API for Conv2d operators Test Plan: python test/test_quantized.py TestQNNPackOps.test_qconv_qnnpack Reviewers: Subscribers: Tasks: Tags:

Summary: This change adds the support to call QNNPACK using the refactored API for Conv2d operators Test Plan: python test/test_quantized.py TestQNNPackOps.test_qconv_qnnpack Reviewers: Subscribers: Tasks: Tags: [ghstack-poisoned]

Summary: This change adds the support to call QNNPACK using the refactored API for Conv2d operators Test Plan: python test/test_quantized.py TestQNNPackOps.test_qconv_qnnpack Reviewers: Subscribers: Tasks: Tags: Differential Revision: [D17422828](https://our.internmc.facebook.com/intern/diff/D17422828) [ghstack-poisoned]

dzhulgakov

I guess one of us would need to rebase depending on whether permute PR lands first or not

dzhulgakov · 2019-09-18T06:41:24Z

test/test_quantized.py

@@ -1323,6 +1322,129 @@ def test_qlinear_qnnpack(self, batch_size, input_channels, output_channels, use_
            np.testing.assert_equal(
                Y_q_ref2.int_repr().numpy(), Y_q.int_repr().numpy())

+    @given(batch_size=st.integers(1, 3),


why do we need separate tests for qnnpack? can't we just use the same set of tests as for fbgemm? (and iterate over available engines in the beginning)

same applies to linear tests above

Ensuring numeric consistency of QNNPACK and FBGEMM is also a requirement for quantization aware training to work, so it will be good to check that both kernels pass the same tests. For QNNPACK we should also add tests that allow for full use of dynamic range of wts and activations

Discussed offline. Will merge tests in a different PR.

dzhulgakov · 2019-09-18T06:42:42Z

test/test_quantized.py

+            Y_scale,
+            Y_zp,
+            use_relu):
+        with enable_mobile_quantized_engine():


nit: I'd rather rename it to a function taking qengine as an argument, but since it's a test - it's minor

aten/src/ATen/native/quantized/cpu/qconv_prepack.cpp

raghuramank100

Few comments/questions. Thanks!

Summary: This change adds the support to call QNNPACK using the refactored API for Conv2d operators Test Plan: python test/test_quantized.py TestQNNPackOps.test_qconv_qnnpack Reviewers: Subscribers: Tasks: Tags: Differential Revision: [D17459892](https://our.internmc.facebook.com/intern/diff/D17459892) [ghstack-poisoned]

Summary: Pull Request resolved: pytorch/pytorch#26152 This change adds the support to call QNNPACK using the refactored API for Conv2d operators Test Plan: python test/test_quantized.py TestQNNPackOps.test_qconv_qnnpack Imported from OSS Differential Revision: D17459892 fbshipit-source-id: d20b3e8b81dd403541cb2b9164731448ca229695

facebook-github-bot · 2019-09-19T01:09:06Z

This pull request has been merged in b23be95.

Adding quantized::conv2d function for pytorch mobile in c10

23a1e4a

Summary: This change adds the support to call QNNPACK using the refactored API for Conv2d operators Test Plan: python test/test_quantized.py TestQNNPackOps.test_qconv_qnnpack Reviewers: Subscribers: Tasks: Tags:

pytorchbot added module: operators oncall: quantization Quantization support in PyTorch labels Sep 13, 2019

This was referenced Sep 13, 2019

[WIP] Integrate forked QNNPACK into mobile PyTorch builds. #26134

Closed

Adding quantized::linear function for pytorch mobile in c10 #26135

Closed

Update on "Adding quantized::conv2d function for pytorch mobile in c10"

ffc8df2

Summary: This change adds the support to call QNNPACK using the refactored API for Conv2d operators Test Plan: python test/test_quantized.py TestQNNPackOps.test_qconv_qnnpack Reviewers: Subscribers: Tasks: Tags:

Update on "Adding quantized::conv2d function for pytorch mobile in c10"

bdf3893

Summary: This change adds the support to call QNNPACK using the refactored API for Conv2d operators Test Plan: python test/test_quantized.py TestQNNPackOps.test_qconv_qnnpack Reviewers: Subscribers: Tasks: Tags:

supriyar requested review from ljk53, raghuramank100 and dskhudia September 13, 2019 15:13

raghuramank100 reviewed Sep 13, 2019

View reviewed changes

test/test_quantized.py Show resolved Hide resolved

supriyar mentioned this pull request Sep 13, 2019

Add support to call unpack for pytorch mobile quantized FC and Conv #26211

Closed

dzhulgakov reviewed Sep 13, 2019

View reviewed changes

aten/src/ATen/native/quantized/cpu/qconv_prepack.cpp Show resolved Hide resolved

dzhulgakov requested changes Sep 14, 2019

View reviewed changes

Update on "Adding quantized::conv2d function for pytorch mobile in c10"

4878501

Summary: This change adds the support to call QNNPACK using the refactored API for Conv2d operators Test Plan: python test/test_quantized.py TestQNNPackOps.test_qconv_qnnpack Reviewers: Subscribers: Tasks: Tags:

supriyar requested a review from dzhulgakov September 15, 2019 00:31

dzhulgakov reviewed Sep 15, 2019

View reviewed changes

test/test_quantized.py Outdated Show resolved Hide resolved

aten/src/ATen/native/quantized/cpu/qconv.cpp Outdated Show resolved Hide resolved

Update on "Adding quantized::conv2d function for pytorch mobile in c10"

481c6b4

Summary: This change adds the support to call QNNPACK using the refactored API for Conv2d operators Test Plan: python test/test_quantized.py TestQNNPackOps.test_qconv_qnnpack Reviewers: Subscribers: Tasks: Tags:

supriyar requested a review from raghuramank100 September 16, 2019 05:20

supriyar mentioned this pull request Sep 16, 2019

Changes to support int8 weight and fp32 bias in QNNPACK #26307

Closed

supriyar mentioned this pull request Sep 17, 2019

Unify Quantization APIs for add, pool and relu #26335

Closed

supriyar added 4 commits September 16, 2019 22:07

supriyar added 2 commits September 17, 2019 13:24

supriyar requested a review from dzhulgakov September 17, 2019 20:58

dzhulgakov approved these changes Sep 18, 2019

View reviewed changes

raghuramank100 reviewed Sep 18, 2019

View reviewed changes

aten/src/ATen/native/quantized/cpu/qconv_prepack.cpp Show resolved Hide resolved

raghuramank100 suggested changes Sep 18, 2019

View reviewed changes

facebook-github-bot closed this in b23be95 Sep 18, 2019

facebook-github-bot added the merged label Sep 19, 2019

facebook-github-bot deleted the gh/supriyar/15/head branch October 28, 2019 22:20

mruberry added the Merged label Oct 28, 2020

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

Adding quantized::conv2d function for pytorch mobile in c10 #26152

Adding quantized::conv2d function for pytorch mobile in c10 #26152

Uh oh!

supriyar commented Sep 13, 2019 •

edited

Loading

Uh oh!

Uh oh!

Uh oh!

dzhulgakov left a comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

dzhulgakov left a comment

Uh oh!

dzhulgakov Sep 18, 2019

Uh oh!

raghuramank100 Sep 18, 2019

Uh oh!

supriyar Sep 18, 2019

Uh oh!

dzhulgakov Sep 18, 2019

Uh oh!

Uh oh!

raghuramank100 left a comment

Uh oh!

facebook-github-bot commented Sep 19, 2019

Uh oh!

Uh oh!

Adding quantized::conv2d function for pytorch mobile in c10 #26152

Adding quantized::conv2d function for pytorch mobile in c10 #26152

Uh oh!

Conversation

supriyar commented Sep 13, 2019 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Uh oh!

Uh oh!

dzhulgakov left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

dzhulgakov left a comment

Choose a reason for hiding this comment

Uh oh!

dzhulgakov Sep 18, 2019

Choose a reason for hiding this comment

Uh oh!

raghuramank100 Sep 18, 2019

Choose a reason for hiding this comment

Uh oh!

supriyar Sep 18, 2019

Choose a reason for hiding this comment

Uh oh!

dzhulgakov Sep 18, 2019

Choose a reason for hiding this comment

Uh oh!

Uh oh!

raghuramank100 left a comment

Choose a reason for hiding this comment

Uh oh!

facebook-github-bot commented Sep 19, 2019

Uh oh!

Uh oh!

supriyar commented Sep 13, 2019 •

edited

Loading