Skip to content

ggml-webgpu: Fix dequantization helpers to not pass in pointers#21872

Merged
reeselevine merged 3 commits intoggml-org:masterfrom
reeselevine:fix-dequant-loaders
Apr 15, 2026
Merged

ggml-webgpu: Fix dequantization helpers to not pass in pointers#21872
reeselevine merged 3 commits intoggml-org:masterfrom
reeselevine:fix-dequant-loaders

Conversation

@reeselevine
Copy link
Copy Markdown
Contributor

Overview

Unfortunately, one of the changes in #21521 to pass buffers as pointers into helper loader functions ended up breaking the WebGPU backend on Firefox, as it does not yet seem to support the WGSL unrestricted_pointer_parameters language extension. So, this PR efactors the helpers back to specialized loader functions for different buffer names for now.

Requirements

@reeselevine reeselevine requested a review from a team as a code owner April 13, 2026 22:31
@github-actions github-actions Bot added ggml changes relating to the ggml tensor library for machine learning WebGPU labels Apr 13, 2026
@reeselevine reeselevine changed the title Fix dequantization helpers to not pass in pointers ggml-webgpu: Fix dequantization helpers to not pass in pointers Apr 13, 2026
@reeselevine reeselevine added the merge ready A maintainer can use this label to indicate that they consider the changes final and ready to merge. label Apr 14, 2026
@ggerganov ggerganov removed the merge ready A maintainer can use this label to indicate that they consider the changes final and ready to merge. label Apr 14, 2026
@reeselevine reeselevine added the merge ready A maintainer can use this label to indicate that they consider the changes final and ready to merge. label Apr 15, 2026
@reeselevine
Copy link
Copy Markdown
Contributor Author

Fixed merge conflicts, also increased precision of XIELU f16 operations to fix failure in recently re-enabled mac CI: https://github.com/ggml-org/llama.cpp/actions/runs/24407947694/job/71296362836#step:4:18460

@CISC
Copy link
Copy Markdown
Member

CISC commented Apr 15, 2026

Fixed merge conflicts, also increased precision of XIELU f16 operations to fix failure in recently re-enabled mac CI

Ooops, snap (I'll close #21940). :)

@reeselevine
Copy link
Copy Markdown
Contributor Author

@ggml-org/maintainers can I get another review on this PR?

@reeselevine reeselevine merged commit 20d3bc2 into ggml-org:master Apr 15, 2026
47 checks passed
mengqin pushed a commit to mengqin/llama.cpp that referenced this pull request Apr 20, 2026
…-org#21872)

* Fix dequantization helpers to not pass in pointers

* Increase XIELU precision
ArberSephirotheca pushed a commit to ArberSephirotheca/llama.cpp that referenced this pull request Apr 21, 2026
…-org#21872)

* Fix dequantization helpers to not pass in pointers

* Increase XIELU precision
arthw pushed a commit to arthw/llama.cpp that referenced this pull request Apr 23, 2026
…-org#21872)

* Fix dequantization helpers to not pass in pointers

* Increase XIELU precision
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

ggml changes relating to the ggml tensor library for machine learning merge ready A maintainer can use this label to indicate that they consider the changes final and ready to merge. WebGPU

Projects

None yet

Development

Successfully merging this pull request may close these issues.

4 participants