ggml: fix zero division in ‘dne’ calculation in CUDA COUNT_EQUAL operator when ‘ne’ is small #10213
Conversation
This file contains 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
…ator when ‘ne’ is small
JohannesGaessler
approved these changes
Nov 8, 2024
|
The test-backend-ops with CUDA enabled failed, which is obvious. |
Nexesenex
added a commit
to Nexesenex/croco.cpp
that referenced
this pull request
Nov 9, 2024
…rator when ‘ne’ is small ggerganov#10213 LCPP PR by SongXiaoXi
|
Sorry, I thought the threshold at which this bug would occur with the test was 500 streaming multiprocessors (which would be much larger than any NVIDIA GPU currently available) but it's actually 125 which is less than is e.g. on an RTX 4090. Thanks for the patch. |
arthw
pushed a commit
to arthw/llama.cpp
that referenced
this pull request
Nov 15, 2024
…ator when ‘ne’ is small (ggerganov#10213)
arthw
pushed a commit
to arthw/llama.cpp
that referenced
this pull request
Nov 18, 2024
…ator when ‘ne’ is small (ggerganov#10213)
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Outdated suggestions cannot be applied.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.
When the total number of elements ne is smaller than 4 * nsm , the integer division ne / (4 * nsm) in the COUNT_EQUAL operator results in zero. This causes dne, which determines the data chunk size per thread block, to be zero. As a result, the CUDA kernel for the COUNT_EQUAL operator doesn’t execute correctly, leading to incorrect computation results.