metal : request Metal 4.0 language version for the tensor API

[JamesFranc]

Overview

• Requests Metal 4.0 when the device reports tensor support using has_tensor so unsupported hardware is untouched
• New guard clears has_tensor when the library came from a pre-compiled metallib.

Additional information

This issue was encountered when running main on an M5 Max. During some prefill and decode benchmarks I was noticing that prefill speeds seemed curiously slow.

The prefill tests used to generate the data above build a prompt of 6,000, random, six-letter words which should prevent cache hits.

This seems to be caused by MTLCompileOptions defaulting to an older Metal language version that doesn't expose tensor API headers (metal_tensor, MetalPerformancePrimitives) to the shader compiler. Both tensor API probes in ggml_metal_device_init() therefore fail:

ggml_metal_library_init_from_source: error compiling source
ggml_metal_device_init: - the tensor API is not supported in this environment - disabling

The compiler error behind it (visible only with verbose = true):

program_source:6:17: error: use of undeclared identifier 'mpp'
using namespace mpp::tensor_ops;
                ^
program_source:7:46: error: use of undeclared identifier 'dextents'

As a result has_tensor is disabled on every device, and prefill runs matmuls on general-purpose shader ALUs.

This fix will only impact prefill on Apple Silicon-based hardware

Requirements

• I have read and agree with the contributing guidelines
• AI usage disclosure: Yes, AI was used to generate the code, evaluate model performance based on human defined evals and implement the solution

[ggml-gh-bot]

Hi @JamesFranc, thanks for your contribution!

Per our contribution guidelines, the automated PR checker found the following issue(s) that need your attention:

• PR Template not respected: Please respect the template when creating a new pull request. Make sure to fill out all required sections.

• AI-generated content: While code is allowed to be generated by AI, please write the PR description and commit messages on your own without the help of AI.

---

Please note that maintainers reserve the right to make final decisions on PRs. If you believe there is a mistake, please comment below.