Skip to content

vulkan : check src0 type in GGML_OP_SET_ROWS to avoid failures due to unimplemented f16 support#25351

Merged
am17an merged 2 commits into
ggml-org:masterfrom
fairydreaming:set-rows-f16-vulkan-no-support
Jul 7, 2026
Merged

vulkan : check src0 type in GGML_OP_SET_ROWS to avoid failures due to unimplemented f16 support#25351
am17an merged 2 commits into
ggml-org:masterfrom
fairydreaming:set-rows-f16-vulkan-no-support

Conversation

@fairydreaming

Copy link
Copy Markdown
Collaborator

Overview

This PR adds src0 type check for GGML_OP_SET_ROWS in ggml_backend_vk_device_supports_op() to handle case with non-f32 src0 as unsupported (for now).

Additional information

Needed to avoid Vulkan test-backend-op failures when #25344 is merged.

 Failing tests:
  SET_ROWS(type_src=f16,type_dst=f16,type_idx=i64,ne=[1,8,1,3],nr23=[1,1],r=2,v=0)
  SET_ROWS(type_src=f16,type_dst=f16,type_idx=i32,ne=[1,8,1,3],nr23=[1,1],r=2,v=0)
  SET_ROWS(type_src=f16,type_dst=f16,type_idx=i64,ne=[1,8,1,3],nr23=[1,1],r=2,v=1)
  SET_ROWS(type_src=f16,type_dst=f16,type_idx=i32,ne=[1,8,1,3],nr23=[1,1],r=2,v=1)
  Backend Vulkan0: FAIL

Requirements

@fairydreaming
fairydreaming requested a review from a team as a code owner July 6, 2026 11:11
@github-actions github-actions Bot added Vulkan Issues specific to the Vulkan backend ggml changes relating to the ggml tensor library for machine learning labels Jul 6, 2026
Comment thread ggml/src/ggml-vulkan/ggml-vulkan.cpp Outdated
default:
return false;
}
} else {

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

The else isn't needed.

@0cc4m

0cc4m commented Jul 6, 2026

Copy link
Copy Markdown
Contributor

@ggerganov Do you know what's up with the ubuntu-llvmpipe runner? It consistently fails with ILLEGAL errors, e.g.

The following tests FAILED:
	 25 - test-llama-archs (ILLEGAL)                        main
	 34 - test-thread-safety (ILLEGAL)                      main

I don't think we caused that.

@ggerganov

Copy link
Copy Markdown
Member

@ggerganov Do you know what's up with the ubuntu-llvmpipe runner? It consistently fails with ILLEGAL errors, e.g.

The following tests FAILED:
	 25 - test-llama-archs (ILLEGAL)                        main
	 34 - test-thread-safety (ILLEGAL)                      main

I don't think we caused that.

No, there is nothing obvious. Have you tried deleting the ccache?

@0cc4m

0cc4m commented Jul 6, 2026

Copy link
Copy Markdown
Contributor

I'm not familiar with the runner, where is the cache controlled?

@ggerganov

Copy link
Copy Markdown
Member

Not sure if you have access to this page, but it's here:

https://github.com/ggml-org/llama.cpp/actions/caches?query=key%3Accache-vulkan-ubuntu+

You can simply delete all the caches and see if it helps.

@am17an
am17an merged commit a8cfdbb into ggml-org:master Jul 7, 2026
28 of 29 checks passed
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 Vulkan Issues specific to the Vulkan backend

Projects

None yet

Development

Successfully merging this pull request may close these issues.

5 participants