2006241 - llama.cpp needs to include <algorithm> for std::max when compiled as C++23

Status: RESOLVED FIXED

(Core :: Machine Learning: On Device, task)

Description

[Chris Peterson [:cpeterson]]

When compiling with -std=c++23, clang defines _LIBCPP_REMOVE_TRANSITIVE_INCLUDES, causing the <algorithm> header to no longer be included in llama-hparams.cpp. Until this issue is fixed in upstream llama.cpp, we can include the <algorithm> header in moz-overrides.h.

https://libcxx.llvm.org/DesignDocs/HeaderRemovalPolicy.html

llama.cpp/src/llama-hparams.cpp:99:20: error: no member named 'max' in namespace 'std'
99 | val = std::max(val, n_embd_k_gqa(il));
| ~~~~~^
llama.cpp/src/llama-hparams.cpp:108:20: error: no member named 'max' in namespace 'std'
108 | val = std::max(val, n_embd_v_gqa(il));
| ~~~~~^

Comment 1

[Chris Peterson [:cpeterson]]

Attachment: Bug 2006241 - llama.cpp needs to include <algorithm> for std::max when compiled as C++23. r?tarek

When compiling with -std=c++23, clang defines _LIBCPP_REMOVE_TRANSITIVE_INCLUDES, causing the <algorithm> header to no longer be included in llama-hparams.cpp. Until this issue is fixed in upstream llama.cpp, we can include the <algorithm> header in moz-overrides.h.

https://libcxx.llvm.org/DesignDocs/HeaderRemovalPolicy.html

llama.cpp/src/llama-hparams.cpp:99:20: error: no member named 'max' in namespace 'std'
99 | val = std::max(val, n_embd_k_gqa(il));
| ~~~~~^
llama.cpp/src/llama-hparams.cpp:108:20: error: no member named 'max' in namespace 'std'
108 | val = std::max(val, n_embd_v_gqa(il));
| ~~~~~^

Comment 2

[Chris Peterson [:cpeterson]]

Since this is a fix that could benefit other llama.cpp users, I submitted a PR to llama.cpp upstream:

https://github.com/ggml-org/llama.cpp/pull/18078

Comment 3

[[:sergesanspaille]]

Attachment: Bug 2006241 - Fix llama implicit includes r=cpeterson

Backported from https://github.com/ggml-org/llama.cpp/commit/2aa45ef9e31069ea0a7d0fef7ce858facdf25218

Comment 4

[[:sergesanspaille]]

@cpeterson : patch backported, and https://treeherder.mozilla.org/jobs?repo=try&revision=5b560c7596fed92c7e1e25bd8022b1e251a72d96 looks good, Bug 2005997 seems unblocked!

Comment 5

[Pulsebot]

Pushed by [e-mail]:
https://github.com/mozilla-firefox/firefox/commit/67b2fbad227e
https://hg.mozilla.org/integration/autoland/rev/5fe7c54d51ca
Fix llama implicit includes r=cpeterson,padenot

Comment 6

[pstanciu]

https://hg.mozilla.org/mozilla-central/rev/5fe7c54d51ca