KEMBAR78
[Pytorch][Vulkan] Add bmm op by tina134 · Pull Request #109360 · pytorch/pytorch · GitHub
Skip to content

[Pytorch][Vulkan] Add bmm op #109360

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom
Closed
wants to merge 1 commit into from
Closed

[Pytorch][Vulkan] Add bmm op #109360

wants to merge 1 commit into from

Conversation

@tina134
Copy link
Contributor

@tina134 tina134 commented Sep 15, 2023

Summary:
BMM is developed on top of MM methodology, the main difference is the 1st input matrix changed from a standard Vulkan 2D tensor to a Vulkan 3D tenser, so the indexing changed quite differently. The matrices of a batch are appended on z-dimension and the channel of size 4 (texel).

The 2nd input matrix remains the same as a packed format (fit a H*W matrix into a H/2 * W/2 * 1 3D image texture by utilizing all 4 values in the texel), but appends more matrices in the batch on z-dimension (only has 1 element in the case of MM).

Vulkan 2D Basic (1st input of MM & output):

ivec3 pos(j, i, 0);
float v = texelFetch(uInput, pos, 0)[0];
# no batch

Vulkan 3D Basic (1st input of BMM & output):

ivec3 pos(k, j, i/4);
float v = texelFetch(uInput, pos, 0)[i % 4];
# i as batch id

Packed weights (2nd input of MM):

ivec3 pos(k_, j_, 0);
float v = texelFetch(uInput, pos, 0)
# v.xyzw are 4 numbers in one matrix
# no batch
# k_, j_ has only 1/4 of the range as the original matrix size (H*W matrix i=> H/2*W/2*1 3D Image).

Packed weights (2nd input of BMM):

ivec3 pos(k_, j_, i);
float v = texelFetch(uInput, pos, 0)
# v.xyzw are 4 numbers in one matrix
# i as batch id

Based on the different indexing of MM & BMM. I modified the MM methodology to produce the desired output image.

Test Plan:

[ttingchulin@27298.od /data/sandcastle/boxes/fbsource (bmm)]$ LD_LIBRARY_PATH=third-party/swiftshader/lib/linux-x64/ buck run fbcode/mode/dev-nosan //xplat/caffe2:pt_vulkan_api_test_bin -- --gtest_filter="*<test>*" eg.  -- --gtest_filter="*mm*"
Building: finished in 0.1 sec (100%) 328/3361 jobs, 0/3361 updated
  Total time: 0.1 sec
BUILD SUCCEEDED
Running main() from xplat/third-party/gmock/googletest-1.12.1/googletest/src/gtest_main.cc
Note: Google Test filter = *mm*
[==========] Running 8 tests from 1 test suite.
[----------] Global test environment set-up.
[----------] 8 tests from VulkanAPITest
[ RUN      ] VulkanAPITest.addmm
[       OK ] VulkanAPITest.addmm (125 ms)
[ RUN      ] VulkanAPITest.addmm_expand
[       OK ] VulkanAPITest.addmm_expand (76 ms)
[ RUN      ] VulkanAPITest.addmm_expand2
[       OK ] VulkanAPITest.addmm_expand2 (0 ms)
[ RUN      ] VulkanAPITest.bmm
[       OK ] VulkanAPITest.bmm (152 ms)
[ RUN      ] VulkanAPITest.bmm_large
[       OK ] VulkanAPITest.bmm_large (4818 ms)
[ RUN      ] VulkanAPITest.bmm_small
[       OK ] VulkanAPITest.bmm_small (4 ms)
[ RUN      ] VulkanAPITest.bmm_one
[       OK ] VulkanAPITest.bmm_one (0 ms)
[ RUN      ] VulkanAPITest.mm
[       OK ] VulkanAPITest.mm (55 ms)
[----------] 8 tests from VulkanAPITest (5233 ms total)


[----------] Global test environment tear-down
[==========] 8 tests from 1 test suite ran. (5233 ms total)
[  PASSED  ] 8 tests.

Differential Revision: D49306279

@pytorch-bot
Copy link

pytorch-bot bot commented Sep 15, 2023 •
edited
Loading

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/109360

Note: Links to docs will display an error until the docs builds have been completed.

✅ You can merge normally! (2 Unrelated Failures)

As of commit 6108108 with merge base 238fb66 (image):

FLAKY - The following job failed but was likely due to flakiness present on trunk:

UNSTABLE - The following job failed but was likely due to flakiness present on trunk and has been marked as unstable:

This comment was automatically generated by Dr. CI and updates every 15 minutes.

@pytorch-bot pytorch-bot bot added ciflow/periodic Trigger jobs ran periodically on master (periodic.yml) on the PR module: vulkan release notes: vulkan release notes category labels Sep 15, 2023
@facebook-github-bot
Copy link
Contributor

This pull request was exported from Phabricator. Differential Revision: D49306279

@facebook-github-bot
Copy link
Contributor

This pull request was exported from Phabricator. Differential Revision: D49306279

@facebook-github-bot
Copy link
Contributor

This pull request was exported from Phabricator. Differential Revision: D49306279

tina134 added a commit to tina134/pytorch that referenced this pull request Sep 18, 2023
@tina134 @facebook-github-bot
[Pytorch][Vulkan] Add bmm op (pytorch#109360)
dc72f99 
Summary:
Pull Request resolved: pytorch#109360

BMM is developed on top of MM methodology, the main difference is the 1st input matrix changed from a standard Vulkan 2D tensor to a Vulkan 3D tenser, so the indexing changed quite differently. The matrices of a batch are appended on z-dimension and the channel of size 4 (texel).

The 2nd input matrix remains the same as a packed format (fit a `H*W` matrix into a `H/2 * W/2 * 1` 3D image texture by utilizing all 4 values in the texel), but appends more matrices in the batch on z-dimension (only has 1 element in the case of MM).

**Vulkan 2D Basic (1st input of MM & output):**
```
ivec3 pos(j, i, 0);
float v = texelFetch(uInput, pos, 0)[0];
# no batch
```

**Vulkan 3D Basic (1st input of BMM & output):**
```
ivec3 pos(k, j, i/4);
float v = texelFetch(uInput, pos, 0)[i % 4];
# i as batch id
```

**Packed weights (2nd input of MM):**
```
ivec3 pos(k_, j_, 0);
float v = texelFetch(uInput, pos, 0)
# v.xyzw are 4 numbers in one matrix
# no batch
# k_, j_ has only 1/4 of the range as the original matrix size (H*W matrix i=> H/2*W/2*1 3D Image).
```

**Packed weights (2nd input of BMM):**
```
ivec3 pos(k_, j_, i);
float v = texelFetch(uInput, pos, 0)
# v.xyzw are 4 numbers in one matrix
# i as batch id
```

Based on the different indexing of MM & BMM. I modified the MM methodology to produce the desired output image.

Test Plan:
```
[ttingchulin@27298.od /data/sandcastle/boxes/fbsource (bmm)]$ LD_LIBRARY_PATH=third-party/swiftshader/lib/linux-x64/ buck run fbcode/mode/dev-nosan //xplat/caffe2:pt_vulkan_api_test_bin -- --gtest_filter="*<test>*" eg.  -- --gtest_filter="*mm*"
Building: finished in 0.1 sec (100%) 328/3361 jobs, 0/3361 updated
  Total time: 0.1 sec
BUILD SUCCEEDED
Running main() from xplat/third-party/gmock/googletest-1.12.1/googletest/src/gtest_main.cc
Note: Google Test filter = *mm*
[==========] Running 8 tests from 1 test suite.
[----------] Global test environment set-up.
[----------] 8 tests from VulkanAPITest
[ RUN      ] VulkanAPITest.addmm
[       OK ] VulkanAPITest.addmm (125 ms)
[ RUN      ] VulkanAPITest.addmm_expand
[       OK ] VulkanAPITest.addmm_expand (76 ms)
[ RUN      ] VulkanAPITest.addmm_expand2
[       OK ] VulkanAPITest.addmm_expand2 (0 ms)
[ RUN      ] VulkanAPITest.bmm
[       OK ] VulkanAPITest.bmm (152 ms)
[ RUN      ] VulkanAPITest.bmm_large
[       OK ] VulkanAPITest.bmm_large (4818 ms)
[ RUN      ] VulkanAPITest.bmm_small
[       OK ] VulkanAPITest.bmm_small (4 ms)
[ RUN      ] VulkanAPITest.bmm_one
[       OK ] VulkanAPITest.bmm_one (0 ms)
[ RUN      ] VulkanAPITest.mm
[       OK ] VulkanAPITest.mm (55 ms)
[----------] 8 tests from VulkanAPITest (5233 ms total)

[----------] Global test environment tear-down
[==========] 8 tests from 1 test suite ran. (5233 ms total)
[  PASSED  ] 8 tests.

```

Differential Revision: D49306279

fbshipit-source-id: 46558c105873f60743c22a3fa52368084cc11c0a
@facebook-github-bot
Copy link
Contributor

This pull request was exported from Phabricator. Differential Revision: D49306279

tina134 added a commit to tina134/pytorch that referenced this pull request Sep 19, 2023
@tina134 @facebook-github-bot
[Pytorch][Vulkan] Add bmm op (pytorch#109360)
b13eb1b 
Summary:
Pull Request resolved: pytorch#109360

BMM is developed on top of MM methodology, the main difference is the 1st input matrix changed from a standard Vulkan 2D tensor to a Vulkan 3D tenser, so the indexing changed quite differently. The matrices of a batch are appended on z-dimension and the channel of size 4 (texel).

The 2nd input matrix remains the same as a packed format (fit a `H*W` matrix into a `H/2 * W/2 * 1` 3D image texture by utilizing all 4 values in the texel), but appends more matrices in the batch on z-dimension (only has 1 element in the case of MM).

**Vulkan 2D Basic (1st input of MM & output):**
```
ivec3 pos(j, i, 0);
float v = texelFetch(uInput, pos, 0)[0];
# no batch
```

**Vulkan 3D Basic (1st input of BMM & output):**
```
ivec3 pos(k, j, i/4);
float v = texelFetch(uInput, pos, 0)[i % 4];
# i as batch id
```

**Packed weights (2nd input of MM):**
```
ivec3 pos(k_, j_, 0);
float v = texelFetch(uInput, pos, 0)
# v.xyzw are 4 numbers in one matrix
# no batch
# k_, j_ has only 1/4 of the range as the original matrix size (H*W matrix i=> H/2*W/2*1 3D Image).
```

**Packed weights (2nd input of BMM):**
```
ivec3 pos(k_, j_, i);
float v = texelFetch(uInput, pos, 0)
# v.xyzw are 4 numbers in one matrix
# i as batch id
```

Based on the different indexing of MM & BMM. I modified the MM methodology to produce the desired output image.

Test Plan:
```
[ttingchulin@27298.od /data/sandcastle/boxes/fbsource (bmm)]$ LD_LIBRARY_PATH=third-party/swiftshader/lib/linux-x64/ buck run fbcode/mode/dev-nosan //xplat/caffe2:pt_vulkan_api_test_bin -- --gtest_filter="*<test>*" eg.  -- --gtest_filter="*mm*"
Building: finished in 0.1 sec (100%) 328/3361 jobs, 0/3361 updated
  Total time: 0.1 sec
BUILD SUCCEEDED
Running main() from xplat/third-party/gmock/googletest-1.12.1/googletest/src/gtest_main.cc
Note: Google Test filter = *mm*
[==========] Running 8 tests from 1 test suite.
[----------] Global test environment set-up.
[----------] 8 tests from VulkanAPITest
[ RUN      ] VulkanAPITest.addmm
[       OK ] VulkanAPITest.addmm (125 ms)
[ RUN      ] VulkanAPITest.addmm_expand
[       OK ] VulkanAPITest.addmm_expand (76 ms)
[ RUN      ] VulkanAPITest.addmm_expand2
[       OK ] VulkanAPITest.addmm_expand2 (0 ms)
[ RUN      ] VulkanAPITest.bmm
[       OK ] VulkanAPITest.bmm (152 ms)
[ RUN      ] VulkanAPITest.bmm_large
[       OK ] VulkanAPITest.bmm_large (4818 ms)
[ RUN      ] VulkanAPITest.bmm_small
[       OK ] VulkanAPITest.bmm_small (4 ms)
[ RUN      ] VulkanAPITest.bmm_one
[       OK ] VulkanAPITest.bmm_one (0 ms)
[ RUN      ] VulkanAPITest.mm
[       OK ] VulkanAPITest.mm (55 ms)
[----------] 8 tests from VulkanAPITest (5233 ms total)

[----------] Global test environment tear-down
[==========] 8 tests from 1 test suite ran. (5233 ms total)
[  PASSED  ] 8 tests.

```

Differential Revision: D49306279

fbshipit-source-id: f906be557cd446d011d9e9deaea4315352521312
@tina134 @facebook-github-bot
[Pytorch][Vulkan] Add bmm op (pytorch#109360)
6108108 
Summary:

BMM is developed on top of MM methodology, the main difference is the 1st input matrix changed from a standard Vulkan 2D tensor to a Vulkan 3D tenser, so the indexing changed quite differently. The matrices of a batch are appended on z-dimension and the channel of size 4 (texel).

The 2nd input matrix remains the same as a packed format (fit a `H*W` matrix into a `H/2 * W/2 * 1` 3D image texture by utilizing all 4 values in the texel), but appends more matrices in the batch on z-dimension (only has 1 element in the case of MM).

**Vulkan 2D Basic (1st input of MM & output):**
```
ivec3 pos(j, i, 0);
float v = texelFetch(uInput, pos, 0)[0];
# no batch
```


**Vulkan 3D Basic (1st input of BMM & output):**
```
ivec3 pos(k, j, i/4);
float v = texelFetch(uInput, pos, 0)[i % 4];
# i as batch id
```


**Packed weights (2nd input of MM):**
```
ivec3 pos(k_, j_, 0);
float v = texelFetch(uInput, pos, 0)
# v.xyzw are 4 numbers in one matrix
# no batch
# k_, j_ has only 1/4 of the range as the original matrix size (H*W matrix i=> H/2*W/2*1 3D Image).
```


**Packed weights (2nd input of BMM):**
```
ivec3 pos(k_, j_, i);
float v = texelFetch(uInput, pos, 0)
# v.xyzw are 4 numbers in one matrix
# i as batch id
```


Based on the different indexing of MM & BMM. I modified the MM methodology to produce the desired output image.

Test Plan:
```
[ttingchulin@27298.od /data/sandcastle/boxes/fbsource (bmm)]$ LD_LIBRARY_PATH=third-party/swiftshader/lib/linux-x64/ buck run fbcode/mode/dev-nosan //xplat/caffe2:pt_vulkan_api_test_bin -- --gtest_filter="*<test>*" eg.  -- --gtest_filter="*mm*"
Building: finished in 0.1 sec (100%) 328/3361 jobs, 0/3361 updated
  Total time: 0.1 sec
BUILD SUCCEEDED
Running main() from xplat/third-party/gmock/googletest-1.12.1/googletest/src/gtest_main.cc
Note: Google Test filter = *mm*
[==========] Running 8 tests from 1 test suite.
[----------] Global test environment set-up.
[----------] 8 tests from VulkanAPITest
[ RUN      ] VulkanAPITest.addmm
[       OK ] VulkanAPITest.addmm (125 ms)
[ RUN      ] VulkanAPITest.addmm_expand
[       OK ] VulkanAPITest.addmm_expand (76 ms)
[ RUN      ] VulkanAPITest.addmm_expand2
[       OK ] VulkanAPITest.addmm_expand2 (0 ms)
[ RUN      ] VulkanAPITest.bmm
[       OK ] VulkanAPITest.bmm (152 ms)
[ RUN      ] VulkanAPITest.bmm_large
[       OK ] VulkanAPITest.bmm_large (4818 ms)
[ RUN      ] VulkanAPITest.bmm_small
[       OK ] VulkanAPITest.bmm_small (4 ms)
[ RUN      ] VulkanAPITest.bmm_one
[       OK ] VulkanAPITest.bmm_one (0 ms)
[ RUN      ] VulkanAPITest.mm
[       OK ] VulkanAPITest.mm (55 ms)
[----------] 8 tests from VulkanAPITest (5233 ms total)


[----------] Global test environment tear-down
[==========] 8 tests from 1 test suite ran. (5233 ms total)
[  PASSED  ] 8 tests.

```

Reviewed By: yipjustin

Differential Revision: D49306279
@facebook-github-bot
Copy link
Contributor

This pull request was exported from Phabricator. Differential Revision: D49306279

@yipjustin yipjustin self-requested a review September 20, 2023 23:02
@facebook-github-bot
Copy link
Contributor

@pytorchbot merge

(Initiating merge automatically since Phabricator Diff has merged)

@pytorch-bot pytorch-bot bot added the ciflow/trunk Trigger trunk jobs on your pull request label Sep 22, 2023
@pytorchmergebot
Copy link
Collaborator

Merge started

Your change will be merged once all checks pass (ETA 0-4 Hours).

Learn more about merging in the wiki.

Questions? Feedback? Please reach out to the PyTorch DevX Team

Advanced Debugging
Check the merge workflow status
here

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

@yipjustin yipjustin yipjustin approved these changes

Assignees

No one assigned

Labels

ciflow/periodic Trigger jobs ran periodically on master (periodic.yml) on the PR ciflow/trunk Trigger trunk jobs on your pull request fb-exported Merged module: vulkan release notes: vulkan release notes category

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

4 participants

@tina134 @facebook-github-bot @pytorchmergebot @yipjustin