UPSTREAM PR #21452: metal : add GATED_LINEAR_ATTN op

[loci-dev]

Note

Source pull request: ggml-org/llama.cpp#21452

Add Metal backend support for GGML_OP_GATED_LINEAR_ATTN (GLA). Supports head_size 64 and 128, f32 only.

Tested with test-backend-ops (7/7 passed):

• M5 Max 128GB (Apple10)
• M2 Mac Mini (Apple8)

Overview

Adds Metal backend support for GGML_OP_GATED_LINEAR_ATTN, which currently falls back to CPU on Apple devices.

The kernel follows the RWKV WKV6 Metal structure: one thread per head element, threadgroup shared memory for k/q/gate, and a float4-vectorized inner loop.

supports_op restricts execution to F32 and head sizes 64 or 128.

No dedicated performance benchmarks are currently configured for this op.

Additional information

docs/ops.md and docs/ops/Metal.csv were regenerated using the required workflow (test-backend-ops support --output csv and ./scripts/create_ops_docs.py). The resulting diff includes updates to multiple ops due to recent changes in main; the only functional change in this PR is support for GATED_LINEAR_ATTN.

Mentions #14909

Requirements

• I have read and agree with the contributing guidelines
• AI usage disclosure: YES - AI used for research and guidance, all code manually reviewed and understood.

[loci-review]

No meaningful performance changes were detected across 125473 analyzed functions in the following binaries: build.bin.llama-tts, build.bin.llama-cvector-generator, build.bin.libllama.so, build.bin.libmtmd.so, build.bin.llama-bench, build.bin.libggml-cpu.so, build.bin.libggml-base.so, build.bin.libggml.so, build.bin.llama-tokenize, build.bin.llama-gemma3-cli, build.bin.llama-gguf-split, build.bin.llama-llava-cli, build.bin.llama-minicpmv-cli, build.bin.llama-quantize, build.bin.llama-qwen2vl-cli.

💬 Questions? Tag @loci-dev

[loci-review]

No meaningful performance changes were detected across 46867 analyzed functions in the following binaries: build.bin.libllama.so, build.bin.libmtmd.so, build.bin.libggml-cpu.so, build.bin.libggml.so, build.bin.libggml-base.so, build.bin.llama-tts, build.bin.llama-bench, build.bin.llama-cvector-generator, build.bin.llama-gemma3-cli, build.bin.llama-gguf-split, build.bin.llama-llava-cli, build.bin.llama-minicpmv-cli, build.bin.llama-quantize, build.bin.llama-qwen2vl-cli, build.bin.llama-tokenize.

💬 Questions? Tag @loci-dev