Compile bug: vulkan set_rows seems broken

[414n]

Git commit

a118d80

Operating systems

Linux

GGML backends

Vulkan

Problem description & steps to reproduce

I'm trying to compile llama.cpp with vulkan support on a Slackware64-15.0 system with updated kernel (6.14.9), mesa (25.1.2) and vulkan-sdk (1.4.313).

CMakeCache.txt

First Bad Commit

b3ad3a0

Compile command

make -C build

Relevant log output

/data/toshi/ai/llama.cpp/ggml/src/ggml-vulkan/ggml-vulkan.cpp:11403:37: error: too few arguments to function ‘ggml_tensor* ggml_set_rows(ggml_context*, ggml_tensor*, ggml_tensor*, ggml_tensor*)’
11403 |         tensor_clone = ggml_set_rows(ggml_ctx, src_clone[0], src_clone[1]);
      |                        ~~~~~~~~~~~~~^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
In file included from /data/toshi/ai/llama.cpp/ggml/src/../include/ggml-vulkan.h:3,
                 from /data/toshi/ai/llama.cpp/ggml/src/ggml-vulkan/ggml-vulkan.cpp:1:
/data/toshi/ai/llama.cpp/ggml/src/../include/ggml.h:1514:35: note: declared here
 1514 |     GGML_API struct ggml_tensor * ggml_set_rows(
      |                                   ^~~~~~~~~~~~~

[jeffbolznv]

This is being resolved in #14899, but shouldn't affect "normal" builds.

[414n]

I can confirm that applying the changes from #14899 fixes the original compilation issue.
Can't vouch for its correctness, though.

[0cc4m]

Why are you compiling with debug flags?

[414n]

I'm sorry, what debug flags are you referring to?

[jeffbolznv]

That code is only built if you enable the GGML_VULKAN_CHECK_RESULTS cmake setting, which is only meant for debugging (and is extremely slow)

[414n]

You’re right, I likely enabled that build option by mistake and forgot about it.

Now that it’s turned off, everything seems to be working fine on my system.

[0cc4m]

Great!