search

lamikr / rocm_sdk_builder

Other
115 stars 8 forks source link

Improve amd-fftw patch (re-fixes #74) #89

Open jeroen-mostert opened 6 days ago

jeroen-mostert commented 6 days ago

Add inline copies of ifunc resolvers to fix the segfault. This allows us to turn --enable-dynamic-dispatcher back on (in case it ever does anyone any good).

I've tested this against a rebuild of amd-fftw and seeing if torio loads and gives back decoders, but an independent build is of course always welcome.

I do not intend to submit this patch upstream, though I hereby freely give permission for anyone else to do so; the AMD repo seems extremely inactive and I'm not sure they aren't going to overhaul this whole ifunc business if/when they finally do get down to it (as they should).

lamikr commented 6 days ago

Thanks, I heard that it should be able to get linux vm with epyc cpu and amd gpu from azure cloud that could be a good place to test this.