Enable roofline for mx4 <-> fp32 #599
Open
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
Summary:
Adding gbps metric and setting kernels as memory bound for `mx4_to_fp32` and `fp32_to_mx4`. On H100, we see
```
<tritonbench run> --op mx4_to_fp32 --device cuda --metrics gbps,hw_roofline,latency
```
prints
```
(Size, Group Size, ebits, mbits) hw_roofline fbgemm_mx4_to_fp32-gbps fbgemm_mx4_to_fp32-latency
---------------------------------- ------------- ------------------------- ----------------------------
(6392, 32, 2, 1) 2000 8.6048 0.006336 (±7.07%)
(278528, 32, 2, 1) 2000 335.928 0.007072 (±8.14%)
(17825792, 32, 2, 1) 2000 1874.3 0.081120 (±1.70%)
(17825809, 32, 2, 1) 2000 1879.5 0.080896 (±1.90%)
```
```
<tritonbench run> --op fp32_to_mx4 --device cuda --metrics gbps,hw_roofline,latency
```
prints
```
(Size, Group Size, ebits, mbits, rounding_mode, stochastic_casting) hw_roofline fbgemm_fp32_to_mx4-gbps fbgemm_fp32_to_mx4-latency
--------------------------------------------------------------------- ------------- ------------------------- ----------------------------
(24048, 32, 2, 1, <RoundingMode.even: 2>, False) 2000 15.6924 0.006944 (±6.91%)
(1048576, 32, 2, 1, <RoundingMode.even: 2>, False) 2000 412.444 0.011520 (±5.00%)
(67108864, 32, 2, 1, <RoundingMode.even: 2>, False) 2000 1734.07 0.175360 (±0.53%)
(67108880, 32, 2, 1, <RoundingMode.even: 2>, False) 2000 1733.76 0.175392 (±0.64%)
```
Differential Revision: D85782147
bremerm31
temporarily deployed
to
docker-s3-upload
GitHub Actions
Inactive
bremerm31
temporarily deployed
to
docker-s3-upload
GitHub Actions
Inactive
bremerm31
temporarily deployed
to
docker-s3-upload
GitHub Actions
Inactive
|
@bremerm31 has exported this pull request. If you are a Meta employee, you can view the originating Diff in D85782147. |
xuzhao9
approved these changes
Oct 29, 2025
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 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:
Adding gbps metric and setting kernels as memory bound for
mx4_to_fp32andfp32_to_mx4. On H100, we seeprints
prints
Differential Revision: D85782147