dequantize_fp8_cache_kernel: Move D=128 device-side-assertion check to host #4869
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. |
|
@ColinPeppler has exported this pull request. If you are a Meta employee, you can view the originating diff in D82320518. |
ColinPeppler
force-pushed
the
export-D82320518
branch
from
26f24c9 to
cd565eb
Compare
ColinPeppler
added a commit
to ColinPeppler/FBGEMM
that referenced
this pull request
Sep 18, 2025
…o host (pytorch#4869) Summary: X-link: facebookresearch/FBGEMM#1891 ## What Move the device-side assertions to the host since all the kernels share the same assertion. ## Why When running evals with symmetric quantization, I ran into the following error. > CUDA error: too many resources requested for launch It failed with this launch configuration: blockDim = (32, 32) = 1024 threads per block. - `$ cuobjdump --dump-resource-usage kv_cache.cu.pic.o.sm_90.cubin | c++filt | grep -A 1 'dequantize_fp8_cache_kernel'` gives me - `void fbgemm_gpu::dequantize_fp8_cache_kernel<true, true>... REG:66` - P1908720668 - That means one threadblock has 66 * 1024 = 67584 registers which exceeds the limit of 65,536. Differential Revision: D82320518
…rch#4868) Summary: X-link: facebookresearch/FBGEMM#1890 Differential Revision: D82320500
…o host (pytorch#4869) Summary: X-link: facebookresearch/FBGEMM#1891 ## What Move the device-side assertions to the host since all the kernels share the same assertion. ## Why When running evals with symmetric quantization, I ran into the following error. > CUDA error: too many resources requested for launch It failed with this launch configuration: blockDim = (32, 32) = 1024 threads per block. - `$ cuobjdump --dump-resource-usage kv_cache.cu.pic.o.sm_90.cubin | c++filt | grep -A 1 'dequantize_fp8_cache_kernel'` gives me - `void fbgemm_gpu::dequantize_fp8_cache_kernel<true, true>... REG:66` - P1908720668 - That means one threadblock has 66 * 1024 = 67584 registers which exceeds the limit of 65,536. Differential Revision: D82320518
|
@ColinPeppler has exported this pull request. If you are a Meta employee, you can view the originating diff in D82320518. |
ColinPeppler
force-pushed
the
export-D82320518
branch
from
cd565eb to
7cf05c6
Compare
|
@ColinPeppler has exported this pull request. If you are a Meta employee, you can view the originating diff in D82320518. |
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:
What
Move the device-side assertions to the host since all the kernels share the same assertion.
Why
When running evals with symmetric quantization, I ran into the following error.
It failed with this launch configuration: blockDim = (32, 32) = 1024 threads per block.
$ cuobjdump --dump-resource-usage kv_cache.cu.pic.o.sm_90.cubin | c++filt | grep -A 1 'dequantize_fp8_cache_kernel'gives mevoid fbgemm_gpu::dequantize_fp8_cache_kernel<true, true>... REG:66Differential Revision: D82320518