[ExecuTorch] Dramatically improve op_clamp build time #5784

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

swolchok wants to merge 5 commits into gh/swolchok/104/base from gh/swolchok/104/head

Contributor

swolchok commented Oct 1, 2024 •

edited

Loading

Stack from ghstack (oldest at bottom):

Instead of building O(|CTYPE_IN| * |CTYPE_MIN| * |CTYPE_MAX| * |CTYPE_OUT|) kernel code (where |T| means the number of possibilities for type T), we build O((|CTYPE_IN| + |CTYPE_MIN| + |CTYPE_MAX| + |CTYPE_COMMON|) * |CTYPE_OUT|) kernel code. (Concretely, ET_SWITCH_REALHB_TYPES has 9 possibilities, so I estimate that we went from 9**4 = 6561 template instantiations to 9 * 4 * 9 = 324
instantiations, or a 20x reduction.)

Differential Revision: D63681034


          [ExecuTorch] Dramatically improve op_clamp build time (2/2)

a00c741

Instead of building `O(|CTYPE_IN| * |CTYPE_MIN| * |CTYPE_MAX|
* |CTYPE_OUT|)` kernel code (where |T| means the number of
possibilities for type T), we build `O((|CTYPE_IN| + |CTYPE_MIN| +
|CTYPE_MAX|) * |CTYPE_OUT|)` kernel code. (Concretely,
`ET_SWITCH_REALHB_TYPES` has 9 possibilities, so I estimate that we
went from 9**4 = 6561 template instantiations to 9 * 3 * 9 = 243
kernels, or a 27x reduction.)

Differential Revision: [D63681034](https://our.internmc.facebook.com/intern/diff/D63681034/)

[ghstack-poisoned]

pytorch-bot bot commented Oct 1, 2024 •

edited

Loading

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/5784

📄 Preview Python docs built from this PR

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

✅ You can merge normally! (1 Unrelated Failure)

As of commit fdc1fc1 with merge base 92d1d1e ():

BROKEN TRUNK - The following job failed but were present on the merge base:

👉 Rebase onto the `viable/strict` branch to avoid these failures

pull / unittest-arm (buck2) / linux-job (gh) (trunk failure)
../../dev/ops/test_sigmoid.py::TestSigmoid::test_sigmoid_add_tosa_BI

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

facebook-github-bot added the CLA Signed label

swolchok mentioned this pull request

[ExecuTorch] Rework apply_ternary_elementwise_fn to allow fixing op_clamp build time (1/2) #5783

Closed

swolchok added a commit that referenced this pull request


          [ExecuTorch] Dramatically improve op_clamp build time (2/2)

af3e8ac

Instead of building `O(|CTYPE_IN| * |CTYPE_MIN| * |CTYPE_MAX|
* |CTYPE_OUT|)` kernel code (where |T| means the number of
possibilities for type T), we build `O((|CTYPE_IN| + |CTYPE_MIN| +
|CTYPE_MAX|) * |CTYPE_OUT|)` kernel code. (Concretely,
`ET_SWITCH_REALHB_TYPES` has 9 possibilities, so I estimate that we
went from 9**4 = 6561 template instantiations to 9 * 3 * 9 = 243
kernels, or a 27x reduction.)

Differential Revision: [D63681034](https://our.internmc.facebook.com/intern/diff/D63681034/)

ghstack-source-id: 245613707
Pull Request resolved: #5784

Contributor

facebook-github-bot commented Oct 1, 2024

This pull request was exported from Phabricator. Differential Revision: D63681034

facebook-github-bot added the fb-exported label

swolchok requested a review from manuelcandales

October 1, 2024 06:11


          slight improvement on "[ExecuTorch] Dramatically improve op_clamp bui…

69461cb

…ld time (2/2)"


Instead of building `O(|CTYPE_IN| * |CTYPE_MIN| * |CTYPE_MAX|* |CTYPE_OUT|)` kernel code (where |T| means the number of possibilities for type T), we build `O((|CTYPE_IN| + |CTYPE_MIN| + |CTYPE_MAX|) * |CTYPE_OUT|)` kernel code. (Concretely, `ET_SWITCH_REALHB_TYPES` has 9 possibilities, so I estimate that we went from 9**4 = 6561 template instantiations to 9 * 3 * 9 = 243 instantiations, or a 27x reduction.)

Differential Revision: [D63681034](https://our.internmc.facebook.com/intern/diff/D63681034/)

[ghstack-poisoned]

swolchok added a commit that referenced this pull request


          [ExecuTorch] Dramatically improve op_clamp build time (2/2)

6f5cd29

Pull Request resolved: #5784

Instead of building `O(|CTYPE_IN| * |CTYPE_MIN| * |CTYPE_MAX| * |CTYPE_OUT|)` kernel code (where |T| means the number of possibilities for type T), we build `O((|CTYPE_IN| + |CTYPE_MIN| + |CTYPE_MAX|) * |CTYPE_OUT|)` kernel code. (Concretely, `ET_SWITCH_REALHB_TYPES` has 9 possibilities, so I estimate that we went from 9**4 = 6561 template instantiations to 9 * 3 * 9 = 243
instantiations, or a 27x reduction.)

Differential Revision: [D63681034](https://our.internmc.facebook.com/intern/diff/D63681034/)
ghstack-source-id: 245741505

Contributor

facebook-github-bot commented Oct 1, 2024

This pull request was exported from Phabricator. Differential Revision: D63681034

Contributor Author

swolchok commented Oct 1, 2024 •

edited

Loading

updated description to match rebase atop #5798 despite not having exported yet; will export when #5798 lands

swolchok changed the title ~~[ExecuTorch] Dramatically improve op_clamp build time (2/2)~~ [ExecuTorch] Dramatically improve op_clamp build time


          Update on "[ExecuTorch] Dramatically improve op_clamp build time"

efd6c08

Instead of building `O(|CTYPE_IN| * |CTYPE_MIN| * |CTYPE_MAX| * |CTYPE_OUT|)` kernel code (where |T| means the number of possibilities for type T), we build `O((|CTYPE_IN| + |CTYPE_MIN| + |CTYPE_MAX| + |CTYPE_COMMON|) * |CTYPE_OUT|)` kernel code. (Concretely, `ET_SWITCH_REALHB_TYPES` has 9 possibilities, so I estimate that we went from 9**4 = 6561 template instantiations to 9 * 4 * 9 = 324
instantiations, or a 20x reduction.)


Differential Revision: [D63681034](https://our.internmc.facebook.com/intern/diff/D63681034/)

[ghstack-poisoned]

Contributor

facebook-github-bot commented Oct 2, 2024

This pull request was exported from Phabricator. Differential Revision: D63681034

swolchok added a commit that referenced this pull request


          [ExecuTorch] Dramatically improve op_clamp build time

3c5b670

Pull Request resolved: #5784

Instead of building `O(|CTYPE_IN| * |CTYPE_MIN| * |CTYPE_MAX| * |CTYPE_OUT|)` kernel code (where |T| means the number of possibilities for type T), we build `O((|CTYPE_IN| + |CTYPE_MIN| + |CTYPE_MAX| + |CTYPE_COMMON|) * |CTYPE_OUT|)` kernel code. (Concretely, `ET_SWITCH_REALHB_TYPES` has 9 possibilities, so I estimate that we went from 9**4 = 6561 template instantiations to 9 * 4 * 9 = 324
instantiations, or a 20x reduction.)
ghstack-source-id: 245912988

Differential Revision: [D63681034](https://our.internmc.facebook.com/intern/diff/D63681034/)


          Update on "[ExecuTorch] Dramatically improve op_clamp build time"

c49f48a

Instead of building `O(|CTYPE_IN| * |CTYPE_MIN| * |CTYPE_MAX| * |CTYPE_OUT|)` kernel code (where |T| means the number of possibilities for type T), we build `O((|CTYPE_IN| + |CTYPE_MIN| + |CTYPE_MAX| + |CTYPE_COMMON|) * |CTYPE_OUT|)` kernel code. (Concretely, `ET_SWITCH_REALHB_TYPES` has 9 possibilities, so I estimate that we went from 9**4 = 6561 template instantiations to 9 * 4 * 9 = 324
instantiations, or a 20x reduction.)


Differential Revision: [D63681034](https://our.internmc.facebook.com/intern/diff/D63681034/)

[ghstack-poisoned]

Contributor

facebook-github-bot commented Oct 2, 2024

This pull request was exported from Phabricator. Differential Revision: D63681034

swolchok added a commit that referenced this pull request


          [ExecuTorch] Dramatically improve op_clamp build time

9fc0d7b

Pull Request resolved: #5784

Instead of building `O(|CTYPE_IN| * |CTYPE_MIN| * |CTYPE_MAX| * |CTYPE_OUT|)` kernel code (where |T| means the number of possibilities for type T), we build `O((|CTYPE_IN| + |CTYPE_MIN| + |CTYPE_MAX| + |CTYPE_COMMON|) * |CTYPE_OUT|)` kernel code. (Concretely, `ET_SWITCH_REALHB_TYPES` has 9 possibilities, so I estimate that we went from 9**4 = 6561 template instantiations to 9 * 4 * 9 = 324
instantiations, or a 20x reduction.)
ghstack-source-id: 246018181

Differential Revision: [D63681034](https://our.internmc.facebook.com/intern/diff/D63681034/)

This was referenced Oct 2, 2024

[ExecuTorch] Simplify function pointers for apply_ternary_elementwise_fn #5833

Closed

[ExecuTorch] Just pass SupportedTensorDtypes for each tensor to apply_ternary_elementwise_fn #5834

Closed


          Update on "[ExecuTorch] Dramatically improve op_clamp build time"

fdc1fc1

Instead of building `O(|CTYPE_IN| * |CTYPE_MIN| * |CTYPE_MAX| * |CTYPE_OUT|)` kernel code (where |T| means the number of possibilities for type T), we build `O((|CTYPE_IN| + |CTYPE_MIN| + |CTYPE_MAX| + |CTYPE_COMMON|) * |CTYPE_OUT|)` kernel code. (Concretely, `ET_SWITCH_REALHB_TYPES` has 9 possibilities, so I estimate that we went from 9**4 = 6561 template instantiations to 9 * 4 * 9 = 324
instantiations, or a 20x reduction.)


Differential Revision: [D63681034](https://our.internmc.facebook.com/intern/diff/D63681034/)

[ghstack-poisoned]

Contributor

facebook-github-bot commented Oct 3, 2024

This pull request was exported from Phabricator. Differential Revision: D63681034

manuelcandales approved these changes

View reviewed changes

facebook-github-bot closed this in

aa8a93c

facebook-github-bot added the Merged label

Contributor

facebook-github-bot commented Oct 3, 2024

This pull request has been merged in aa8a93c.

kedarnath03 pushed a commit to kedarnath03/executorch that referenced this pull request


          [ExecuTorch] Dramatically improve op_clamp build time

eec8006

Pull Request resolved: pytorch/executorch#5784

Instead of building `O(|CTYPE_IN| * |CTYPE_MIN| * |CTYPE_MAX| * |CTYPE_OUT|)` kernel code (where |T| means the number of possibilities for type T), we build `O((|CTYPE_IN| + |CTYPE_MIN| + |CTYPE_MAX| + |CTYPE_COMMON|) * |CTYPE_OUT|)` kernel code. (Concretely, `ET_SWITCH_REALHB_TYPES` has 9 possibilities, so I estimate that we went from 9**4 = 6561 template instantiations to 9 * 4 * 9 = 324
instantiations, or a 20x reduction.)
ghstack-source-id: 246132260

Differential Revision: [D63681034](https://our.internmc.facebook.com/intern/diff/D63681034/)

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

Labels

CLA Signed fb-exported Merged