Invoke AMD specific kernel reorder_batched_ad_indices_kernel_vec #4412
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
✅ Deploy Preview for pytorch-fbgemm-docs ready!
To edit notification comments on pull requests, go to your Netlify project configuration. |
|
This pull request was exported from Phabricator. Differential Revision: D77459476 |
ghq24int
force-pushed
the
export-D77459476
branch
from
a28c5fc to
f8f2d34
Compare
|
This pull request was exported from Phabricator. Differential Revision: D77459476 |
ghq24int
added a commit
to ghq24int/FBGEMM
that referenced
this pull request
Jun 29, 2025
…orch#4412) Summary: X-link: facebookresearch/FBGEMM#1483 For the benchmark in the codebase, the larger the profuct of length and num-ads is, the better performance. Two optimization: 1. Vector loading in a warp. 2. The product of batch-size and table-size determines the # of thread blocks (https://www.internalfb.com/code/fbsource/[cecfed562b79afad0eb9c44259141f50352da342]/fbcode/deeplearning/fbgemm/fbgemm_gpu/src/sparse_ops/sparse_reorder_batched_ad.cu?lines=361). In MRS models, we expect more thread blocks in our user cases. As such, we shrink the block size to achieve more thread blocks, thus improving compute utilization. Performance results and local test benchmarks: D77066925 Differential Revision: D77459476
ghq24int
force-pushed
the
export-D77459476
branch
from
f8f2d34 to
2b8ecc2
Compare
ghq24int
added a commit
to ghq24int/FBGEMM
that referenced
this pull request
Jun 29, 2025
…orch#4412) Summary: X-link: facebookresearch/FBGEMM#1483 For the benchmark in the codebase, the larger the profuct of length and num-ads is, the better performance. Two optimization: 1. Vector loading in a warp. 2. The product of batch-size and table-size determines the # of thread blocks (https://www.internalfb.com/code/fbsource/[cecfed562b79afad0eb9c44259141f50352da342]/fbcode/deeplearning/fbgemm/fbgemm_gpu/src/sparse_ops/sparse_reorder_batched_ad.cu?lines=361). In MRS models, we expect more thread blocks in our user cases. As such, we shrink the block size to achieve more thread blocks, thus improving compute utilization. Performance results and local test benchmarks: D77066925 Differential Revision: D77459476
ghq24int
force-pushed
the
export-D77459476
branch
from
2b8ecc2 to
2746017
Compare
|
This pull request was exported from Phabricator. Differential Revision: D77459476 |
ghq24int
added a commit
to ghq24int/FBGEMM
that referenced
this pull request
Jun 29, 2025
…orch#4412) Summary: Pull Request resolved: pytorch#4412 X-link: facebookresearch/FBGEMM#1483 For the benchmark in the codebase, the larger the profuct of length and num-ads is, the better performance. Two optimization: 1. Vector loading in a warp. 2. The product of batch-size and table-size determines the # of thread blocks (https://www.internalfb.com/code/fbsource/[cecfed562b79afad0eb9c44259141f50352da342]/fbcode/deeplearning/fbgemm/fbgemm_gpu/src/sparse_ops/sparse_reorder_batched_ad.cu?lines=361). In MRS models, we expect more thread blocks in our user cases. As such, we shrink the block size to achieve more thread blocks, thus improving compute utilization. Performance results and local test benchmarks: D77066925 Differential Revision: D77459476
ghq24int
force-pushed
the
export-D77459476
branch
from
2746017 to
0f2a7ad
Compare
|
This pull request was exported from Phabricator. Differential Revision: D77459476 |
1 similar comment
|
This pull request was exported from Phabricator. Differential Revision: D77459476 |
ghq24int
force-pushed
the
export-D77459476
branch
from
0f2a7ad to
0f1e843
Compare
ghq24int
added a commit
to ghq24int/FBGEMM
that referenced
this pull request
Jun 29, 2025
…orch#4412) Summary: Pull Request resolved: pytorch#4412 X-link: facebookresearch/FBGEMM#1483 For the benchmark in the codebase, the larger the profuct of length and num-ads is, the better performance. Two optimization: 1. Vector loading in a warp. 2. The product of batch-size and table-size determines the # of thread blocks (https://www.internalfb.com/code/fbsource/[cecfed562b79afad0eb9c44259141f50352da342]/fbcode/deeplearning/fbgemm/fbgemm_gpu/src/sparse_ops/sparse_reorder_batched_ad.cu?lines=361). In MRS models, we expect more thread blocks in our user cases. As such, we shrink the block size to achieve more thread blocks, thus improving compute utilization. Performance results and local test benchmarks: D77066925 Differential Revision: D77459476
|
This pull request was exported from Phabricator. Differential Revision: D77459476 |
ghq24int
force-pushed
the
export-D77459476
branch
from
0f1e843 to
9933084
Compare
ghq24int
added a commit
to ghq24int/FBGEMM
that referenced
this pull request
Jul 8, 2025
…orch#4412) Summary: X-link: facebookresearch/FBGEMM#1483 For the benchmark in the codebase, the larger the profuct of length and num-ads is, the better performance. Two optimization: 1. Vector loading in a warp. 2. The product of batch-size and table-size determines the # of thread blocks (https://www.internalfb.com/code/fbsource/[cecfed562b79afad0eb9c44259141f50352da342]/fbcode/deeplearning/fbgemm/fbgemm_gpu/src/sparse_ops/sparse_reorder_batched_ad.cu?lines=361). In MRS models, we expect more thread blocks in our user cases. As such, we shrink the block size to achieve more thread blocks, thus improving compute utilization. Performance results and local test benchmarks: D77066925 Differential Revision: D77459476
ghq24int
added a commit
to ghq24int/FBGEMM
that referenced
this pull request
Jul 8, 2025
…orch#4412) Summary: X-link: facebookresearch/FBGEMM#1483 For the benchmark in the codebase, the larger the profuct of length and num-ads is, the better performance. Two optimization: 1. Vector loading in a warp. 2. The product of batch-size and table-size determines the # of thread blocks (https://www.internalfb.com/code/fbsource/[cecfed562b79afad0eb9c44259141f50352da342]/fbcode/deeplearning/fbgemm/fbgemm_gpu/src/sparse_ops/sparse_reorder_batched_ad.cu?lines=361). In MRS models, we expect more thread blocks in our user cases. As such, we shrink the block size to achieve more thread blocks, thus improving compute utilization. Performance results and local test benchmarks: D77066925 Differential Revision: D77459476
ghq24int
force-pushed
the
export-D77459476
branch
from
9933084 to
e529709
Compare
|
This pull request was exported from Phabricator. Differential Revision: D77459476 |
ghq24int
added a commit
to ghq24int/FBGEMM
that referenced
this pull request
Jul 8, 2025
…orch#4412) Summary: Pull Request resolved: pytorch#4412 X-link: facebookresearch/FBGEMM#1483 For the benchmark in the codebase, the larger the profuct of length and num-ads is, the better performance. Two optimization: 1. Vector loading in a warp. 2. The product of batch-size and table-size determines the # of thread blocks (https://www.internalfb.com/code/fbsource/[cecfed562b79afad0eb9c44259141f50352da342]/fbcode/deeplearning/fbgemm/fbgemm_gpu/src/sparse_ops/sparse_reorder_batched_ad.cu?lines=361). In MRS models, we expect more thread blocks in our user cases. As such, we shrink the block size to achieve more thread blocks, thus improving compute utilization. Performance results and local test benchmarks: D77066925 Differential Revision: D77459476
ghq24int
force-pushed
the
export-D77459476
branch
from
e529709 to
7e89b4f
Compare
|
This pull request was exported from Phabricator. Differential Revision: D77459476 |
1 similar comment
|
This pull request was exported from Phabricator. Differential Revision: D77459476 |
ghq24int
added a commit
to ghq24int/FBGEMM
that referenced
this pull request
Jul 8, 2025
…orch#4412) Summary: Pull Request resolved: pytorch#4412 X-link: facebookresearch/FBGEMM#1483 For the benchmark in the codebase, the larger the profuct of length and num-ads is, the better performance. Two optimization: 1. Vector loading in a warp. 2. The product of batch-size and table-size determines the # of thread blocks (https://www.internalfb.com/code/fbsource/[cecfed562b79afad0eb9c44259141f50352da342]/fbcode/deeplearning/fbgemm/fbgemm_gpu/src/sparse_ops/sparse_reorder_batched_ad.cu?lines=361). In MRS models, we expect more thread blocks in our user cases. As such, we shrink the block size to achieve more thread blocks, thus improving compute utilization. Performance results and local test benchmarks: D77066925 Differential Revision: D77459476
ghq24int
force-pushed
the
export-D77459476
branch
from
7e89b4f to
827c4f2
Compare
|
This pull request was exported from Phabricator. Differential Revision: D77459476 |
ghq24int
force-pushed
the
export-D77459476
branch
from
827c4f2 to
1260a57
Compare
ghq24int
added a commit
to ghq24int/FBGEMM
that referenced
this pull request
Jul 9, 2025
…orch#4412) Summary: X-link: facebookresearch/FBGEMM#1483 For the benchmark in the codebase, the larger the profuct of length and num-ads is, the better performance. Two optimization: 1. Vector loading in a warp. 2. The product of batch-size and table-size determines the # of thread blocks (https://www.internalfb.com/code/fbsource/[cecfed562b79afad0eb9c44259141f50352da342]/fbcode/deeplearning/fbgemm/fbgemm_gpu/src/sparse_ops/sparse_reorder_batched_ad.cu?lines=361). In MRS models, we expect more thread blocks in our user cases. As such, we shrink the block size to achieve more thread blocks, thus improving compute utilization. Performance results and local test benchmarks: D77066925 Differential Revision: D77459476
|
This pull request was exported from Phabricator. Differential Revision: D77459476 |
ghq24int
force-pushed
the
export-D77459476
branch
from
1260a57 to
4b93b15
Compare
ghq24int
added a commit
to ghq24int/FBGEMM
that referenced
this pull request
Jul 10, 2025
…orch#4412) Summary: X-link: facebookresearch/FBGEMM#1483 For the benchmark in the codebase, the larger the profuct of length and num-ads is, the better performance. Two optimization: 1. Vector loading in a warp. 2. The product of batch-size and table-size determines the # of thread blocks (https://www.internalfb.com/code/fbsource/[cecfed562b79afad0eb9c44259141f50352da342]/fbcode/deeplearning/fbgemm/fbgemm_gpu/src/sparse_ops/sparse_reorder_batched_ad.cu?lines=361). In MRS models, we expect more thread blocks in our user cases. As such, we shrink the block size to achieve more thread blocks, thus improving compute utilization. Performance results and local test benchmarks: D77066925 Differential Revision: D77459476
|
This pull request was exported from Phabricator. Differential Revision: D77459476 |
ghq24int
force-pushed
the
export-D77459476
branch
from
4b93b15 to
fa3f37b
Compare
ghq24int
added a commit
to ghq24int/FBGEMM
that referenced
this pull request
Jul 10, 2025
…orch#4412) Summary: X-link: facebookresearch/FBGEMM#1483 For the benchmark in the codebase, the larger the profuct of length and num-ads is, the better performance. Two optimization: 1. Vector loading in a warp. 2. The product of batch-size and table-size determines the # of thread blocks (https://www.internalfb.com/code/fbsource/[cecfed562b79afad0eb9c44259141f50352da342]/fbcode/deeplearning/fbgemm/fbgemm_gpu/src/sparse_ops/sparse_reorder_batched_ad.cu?lines=361). In MRS models, we expect more thread blocks in our user cases. As such, we shrink the block size to achieve more thread blocks, thus improving compute utilization. Performance results and local test benchmarks: D77066925 Differential Revision: D77459476
ghq24int
added a commit
to ghq24int/FBGEMM
that referenced
this pull request
Jul 10, 2025
…orch#4412) Summary: X-link: facebookresearch/FBGEMM#1483 For the benchmark in the codebase, the larger the profuct of length and num-ads is, the better performance. Two optimization: 1. Vector loading in a warp. 2. The product of batch-size and table-size determines the # of thread blocks (https://www.internalfb.com/code/fbsource/[cecfed562b79afad0eb9c44259141f50352da342]/fbcode/deeplearning/fbgemm/fbgemm_gpu/src/sparse_ops/sparse_reorder_batched_ad.cu?lines=361). In MRS models, we expect more thread blocks in our user cases. As such, we shrink the block size to achieve more thread blocks, thus improving compute utilization. Performance results and local test benchmarks: D77066925 Differential Revision: D77459476
ghq24int
force-pushed
the
export-D77459476
branch
from
fa3f37b to
564312a
Compare
|
This pull request was exported from Phabricator. Differential Revision: D77459476 |
1 similar comment
|
This pull request was exported from Phabricator. Differential Revision: D77459476 |
ghq24int
added a commit
to ghq24int/FBGEMM
that referenced
this pull request
Jul 10, 2025
…orch#4412) Summary: Pull Request resolved: pytorch#4412 X-link: facebookresearch/FBGEMM#1483 For the benchmark in the codebase, the larger the profuct of length and num-ads is, the better performance. Two optimization: 1. Vector loading in a warp. 2. The product of batch-size and table-size determines the # of thread blocks (https://www.internalfb.com/code/fbsource/[cecfed562b79afad0eb9c44259141f50352da342]/fbcode/deeplearning/fbgemm/fbgemm_gpu/src/sparse_ops/sparse_reorder_batched_ad.cu?lines=361). In MRS models, we expect more thread blocks in our user cases. As such, we shrink the block size to achieve more thread blocks, thus improving compute utilization. Performance results and local test benchmarks: D77066925 Differential Revision: D77459476
ghq24int
force-pushed
the
export-D77459476
branch
from
564312a to
27be2c0
Compare
…orch#4412) Summary: Pull Request resolved: pytorch#4412 X-link: facebookresearch/FBGEMM#1483 For the benchmark in the codebase, the larger the profuct of length and num-ads is, the better performance. Two optimization: 1. Vector loading in a warp. 2. The product of batch-size and table-size determines the # of thread blocks (https://www.internalfb.com/code/fbsource/[cecfed562b79afad0eb9c44259141f50352da342]/fbcode/deeplearning/fbgemm/fbgemm_gpu/src/sparse_ops/sparse_reorder_batched_ad.cu?lines=361). In MRS models, we expect more thread blocks in our user cases. As such, we shrink the block size to achieve more thread blocks, thus improving compute utilization. Performance results and local test benchmarks: D77066925 Differential Revision: D77459476
|
This pull request was exported from Phabricator. Differential Revision: D77459476 |
ghq24int
force-pushed
the
export-D77459476
branch
from
27be2c0 to
35dd366
Compare
|
This pull request was exported from Phabricator. Differential Revision: D77459476 |
|
This pull request has been merged in 3571258. |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 participants
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Summary:
For the benchmark in the codebase, the larger the profuct of length and num-ads is, the better performance.
Two optimization:
Performance results and local test benchmarks: D77066925
Differential Revision: D77459476