Request: ARM SME support (for Apple M4)..

OpenMathLib / OpenBLAS

OpenBLAS is an optimized BLAS library based on GotoBLAS2 1.13 BSD version.

http://www.openblas.net

BSD 3-Clause "New" or "Revised" License

6.32k stars 1.49k forks source link

Request: ARM SME support (for Apple M4).. #4715

Open oscarbg opened 4 months ago

oscarbg commented 4 months ago

No need for unofficial Apple AMX intruction set on M4.. 2tflops possible..

martin-frbg commented 4 months ago

PRs welcome... do you have the hardware to test ?

oscarbg commented 4 months ago

Not yet.. waiting for a mac mini m4..

brada4 commented 4 months ago

I think this can be closed, zero chance to run your cpuid on ipad, and normal computers release year later. AMX is NOT ISA , it is a co-processor with prefixed instructions emitted from main cpu. Like FPU on 80386 or crypto accelerators nowadays. There is no public documentation outsude accelerate cblas using it.

martin-frbg commented 4 months ago

try reading that again, it's about SME...

brada4 commented 4 months ago

SME

Which is rumoured on some sites to be present....

Mousius commented 4 months ago

You can develop and test using the Fixed Virtual Platform (FVP): https://github.com/apache/tvm/pull/16755 https://github.com/apache/tvm/pull/16749

brada4 commented 4 months ago

Gcc11+ can compile it, the question is whether itvis supported on particular cpu.

martin-frbg commented 1 month ago

some implementation hints there: https://scalable.uni-jena.de/opt/sme/index.html