Makes use_fast_accum configurable.
#3829
Closed
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
|
This pull request was exported from Phabricator. Differential Revision: D71290596 |
✅ Deploy Preview for pytorch-fbgemm-docs ready!
To edit notification comments on pull requests, go to your Netlify site configuration. |
levendlee
added a commit
to levendlee/FBGEMM
that referenced
this pull request
Mar 17, 2025
Summary: X-link: facebookresearch/FBGEMM#913 [Public to OSS] Thanks htyu for pointing out the issue. Looking forward to warp specialization support on Nvidia! - Exposes fast accumulation as a configurable. - Not enable it by default. No change in default behavior. - No additional tuning regarding to `use_fast_accum=True`. W/ HIP backend, the semantics of `c += tl.dot(a, b)` and `c = tl.dot(a,b,c)` seems to be the same. Differential Revision: D71290596
levendlee
force-pushed
the
export-D71290596
branch
from
af74814 to
83859ec
Compare
levendlee
added a commit
to levendlee/FBGEMM
that referenced
this pull request
Mar 17, 2025
Summary: X-link: facebookresearch/FBGEMM#913 [Public to OSS] Thanks htyu for pointing out the issue. Looking forward to warp specialization support on Nvidia! - Exposes fast accumulation as a configurable. - Not enable it by default. No change in default behavior. - No additional tuning regarding to `use_fast_accum=True`. W/ HIP backend, the semantics of `c += tl.dot(a, b)` and `c = tl.dot(a,b,c)` seems to be the same. Differential Revision: D71290596
levendlee
force-pushed
the
export-D71290596
branch
from
83859ec to
378e98d
Compare
|
This pull request was exported from Phabricator. Differential Revision: D71290596 |
levendlee
added a commit
to levendlee/FBGEMM
that referenced
this pull request
Mar 17, 2025
Summary: Pull Request resolved: pytorch#3829 X-link: facebookresearch/FBGEMM#913 [Public to OSS] Thanks htyu for pointing out the issue. Looking forward to warp specialization support on Nvidia! - Exposes fast accumulation as a configurable. - Not enable it by default. No change in default behavior. - No additional tuning regarding to `use_fast_accum=True`. W/ HIP backend, the semantics of `c += tl.dot(a, b)` and `c = tl.dot(a,b,c)` seems to be the same. Differential Revision: D71290596
levendlee
force-pushed
the
export-D71290596
branch
from
378e98d to
f55a6ef
Compare
|
This pull request was exported from Phabricator. Differential Revision: D71290596 |
levendlee
added a commit
to levendlee/FBGEMM
that referenced
this pull request
Mar 17, 2025
Summary: Pull Request resolved: pytorch#3829 X-link: facebookresearch/FBGEMM#913 [Public to OSS] Thanks htyu for pointing out the issue. Looking forward to warp specialization support on Nvidia! - Exposes fast accumulation as a configurable. - Not enable it by default. No change in default behavior. - No additional tuning regarding to `use_fast_accum=True`. W/ HIP backend, the semantics of `c += tl.dot(a, b)` and `c = tl.dot(a,b,c)` seems to be the same. Differential Revision: D71290596
levendlee
force-pushed
the
export-D71290596
branch
from
f55a6ef to
cd135b9
Compare
levendlee
added a commit
to levendlee/FBGEMM
that referenced
this pull request
Mar 17, 2025
Summary: X-link: facebookresearch/FBGEMM#913 [Public to OSS] Thanks htyu for pointing out the issue. Looking forward to warp specialization support on Nvidia! - Exposes fast accumulation as a configurable. - Not enable it by default. No change in default behavior. - No additional tuning regarding to `use_fast_accum=True`. W/ HIP backend, the semantics of `c += tl.dot(a, b)` and `c = tl.dot(a,b,c)` seems to be the same. Differential Revision: D71290596
levendlee
force-pushed
the
export-D71290596
branch
from
cd135b9 to
ca943ac
Compare
levendlee
added a commit
to levendlee/FBGEMM
that referenced
this pull request
Mar 17, 2025
Summary: X-link: facebookresearch/FBGEMM#913 [Public to OSS] Thanks htyu for pointing out the issue. Looking forward to warp specialization support on Nvidia! - Exposes fast accumulation as a configurable. - Not enable it by default. No change in default behavior. - No additional tuning regarding to `use_fast_accum=True`. W/ HIP backend, the semantics of `c += tl.dot(a, b)` and `c = tl.dot(a,b,c)` seems to be the same. Differential Revision: D71290596
levendlee
force-pushed
the
export-D71290596
branch
from
ca943ac to
5eef915
Compare
|
This pull request was exported from Phabricator. Differential Revision: D71290596 |
levendlee
added a commit
to levendlee/FBGEMM
that referenced
this pull request
Mar 17, 2025
Summary: Pull Request resolved: pytorch#3829 X-link: facebookresearch/FBGEMM#913 [Public to OSS] Thanks htyu for pointing out the issue. Looking forward to warp specialization support on Nvidia! - Exposes fast accumulation as a configurable. - Not enable it by default. No change in default behavior. - No additional tuning regarding to `use_fast_accum=True`. W/ HIP backend, the semantics of `c += tl.dot(a, b)` and `c = tl.dot(a,b,c)` seems to be the same. Differential Revision: D71290596
levendlee
force-pushed
the
export-D71290596
branch
from
5eef915 to
eaa7e90
Compare
|
This pull request was exported from Phabricator. Differential Revision: D71290596 |
levendlee
added a commit
to levendlee/FBGEMM
that referenced
this pull request
Mar 17, 2025
Summary: Pull Request resolved: pytorch#3829 X-link: facebookresearch/FBGEMM#913 [Public to OSS] Thanks htyu for pointing out the issue. Looking forward to warp specialization support on Nvidia! - Exposes fast accumulation as a configurable. - Not enable it by default. No change in default behavior. - No additional tuning regarding to `use_fast_accum=True`. W/ HIP backend, the semantics of `c += tl.dot(a, b)` and `c = tl.dot(a,b,c)` seems to be the same. Differential Revision: D71290596
levendlee
force-pushed
the
export-D71290596
branch
2 times, most recently
from
dc8074c to
58ba3e8
Compare
levendlee
added a commit
to levendlee/FBGEMM
that referenced
this pull request
Mar 17, 2025
Summary: X-link: facebookresearch/FBGEMM#913 [Public to OSS] Thanks htyu for pointing out the issue. Looking forward to warp specialization support on Nvidia! - Exposes fast accumulation as a configurable. - Not enable it by default. No change in default behavior. - No additional tuning regarding to `use_fast_accum=True`. W/ HIP backend, the semantics of `c += tl.dot(a, b)` and `c = tl.dot(a,b,c)` seems to be the same. Reviewed By: htyu Differential Revision: D71290596
levendlee
added a commit
to levendlee/FBGEMM
that referenced
this pull request
Mar 17, 2025
Summary: X-link: facebookresearch/FBGEMM#913 [Public to OSS] Thanks htyu for pointing out the issue. Looking forward to warp specialization support on Nvidia! - Exposes fast accumulation as a configurable. - Not enable it by default. No change in default behavior. - No additional tuning regarding to `use_fast_accum=True`. W/ HIP backend, the semantics of `c += tl.dot(a, b)` and `c = tl.dot(a,b,c)` seems to be the same. Reviewed By: htyu Differential Revision: D71290596
levendlee
force-pushed
the
export-D71290596
branch
from
58ba3e8 to
ab6d092
Compare
|
This pull request was exported from Phabricator. Differential Revision: D71290596 |
levendlee
added a commit
to levendlee/FBGEMM
that referenced
this pull request
Mar 17, 2025
Summary: Pull Request resolved: pytorch#3829 X-link: facebookresearch/FBGEMM#913 [Public to OSS] Thanks htyu for pointing out the issue. Looking forward to warp specialization support on Nvidia! - Exposes fast accumulation as a configurable. - Not enable it by default. No change in default behavior. - No additional tuning regarding to `use_fast_accum=True`. W/ HIP backend, the semantics of `c += tl.dot(a, b)` and `c = tl.dot(a,b,c)` seems to be the same. Reviewed By: htyu Differential Revision: D71290596
levendlee
force-pushed
the
export-D71290596
branch
from
ab6d092 to
1f12df4
Compare
Summary: Pull Request resolved: pytorch#3829 X-link: facebookresearch/FBGEMM#913 [Public to OSS] Thanks htyu for pointing out the issue. Looking forward to warp specialization support on Nvidia! - Exposes fast accumulation as a configurable. - Not enable it by default. No change in default behavior. - No additional tuning regarding to `use_fast_accum=True`. W/ HIP backend, the semantics of `c += tl.dot(a, b)` and `c = tl.dot(a,b,c)` seems to be the same. Reviewed By: htyu Differential Revision: D71290596
|
This pull request was exported from Phabricator. Differential Revision: D71290596 |
levendlee
force-pushed
the
export-D71290596
branch
from
1f12df4 to
c848446
Compare
|
This pull request has been merged in 067b63c. |
liligwu
pushed a commit
to ROCm/FBGEMM
that referenced
this pull request
Mar 19, 2025
Summary: Pull Request resolved: pytorch#3829 X-link: https://github.com/facebookresearch/FBGEMM/pull/913 [Public to OSS] Thanks htyu for pointing out the issue. Looking forward to warp specialization support on Nvidia! - Exposes fast accumulation as a configurable. - Not enable it by default. No change in default behavior. - No additional tuning regarding to `use_fast_accum=True`. W/ HIP backend, the semantics of `c += tl.dot(a, b)` and `c = tl.dot(a,b,c)` seems to be the same. Reviewed By: htyu Differential Revision: D71290596 fbshipit-source-id: 8e2a20899f301f861d8d72f6290e573e23288e63
q10
pushed a commit
to q10/FBGEMM
that referenced
this pull request
Apr 10, 2025
Summary: X-link: pytorch#3829 Pull Request resolved: facebookresearch/FBGEMM#913 [Public to OSS] Thanks htyu for pointing out the issue. Looking forward to warp specialization support on Nvidia! - Exposes fast accumulation as a configurable. - Not enable it by default. No change in default behavior. - No additional tuning regarding to `use_fast_accum=True`. W/ HIP backend, the semantics of `c += tl.dot(a, b)` and `c = tl.dot(a,b,c)` seems to be the same. Reviewed By: htyu Differential Revision: D71290596 fbshipit-source-id: 8e2a20899f301f861d8d72f6290e573e23288e63
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
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:
[Public to OSS]
Thanks htyu for pointing out the issue. Looking forward to warp specialization support on Nvidia!
use_fast_accum=True.Differential Revision: D71290596