UPSTREAM PR #19226: opencl: refactor some ops, concat, repeat, tanh and scale

[loci-dev]

Note

Source pull request: ggml-org/llama.cpp#19226

Gemma-3n-E2B and Gemma-3n-E4B have been producing weird (not really gibberish but apparently not correct) output. Ended up refactoring these ops and the issue is now fixed. In addition, this refactor also improves perf a bit.

On X Elite,

gemma-3n-E2B-it-Q8_0,

before,

common_perf_print: prompt eval time =    2522.36 ms /   235 tokens (   10.73 ms per token,    93.17 tokens per second)
common_perf_print:        eval time =   24209.42 ms /   256 runs   (   94.57 ms per token,    10.57 tokens per second)

after,

common_perf_print: prompt eval time =    1473.28 ms /   235 tokens (    6.27 ms per token,   159.51 tokens per second)
common_perf_print:        eval time =   15944.91 ms /   256 runs   (   62.28 ms per token,    16.06 tokens per second)

[loci-review]

No meaningful performance changes were detected across 115327 analyzed functions in the following binaries: build.bin.llama-cvector-generator, build.bin.llama-tts, build.bin.libllama.so, build.bin.libmtmd.so, 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.libggml-base.so, build.bin.libggml-cpu.so, build.bin.libggml.so, build.bin.llama-tokenize, build.bin.llama-qwen2vl-cli, build.bin.llama-bench.

🔎 Full breakdown: Loci Inspector.
💬 Questions? Tag @loci-dev.