This is the mail archive of the libc-help@sourceware.org mailing list for the glibc project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: disable avx instructions at runtime


>> > it is a bit tricky: my debugger provides a replay engine, which prbly
>> > maps cpuid to the physical instruction, but it does not understand avx
>> > instructions. so i'd like to to disable the avx-specific code,
>> > *although* cpuid would allow them.
>> >
>> > otherwise, is it possible to compile glibc, disabling the runtime
>> > dispatching?
> I know of no way to easily disable AVX support. The framework that
> enables the support was not designed to be disabled (though
> it should).
> 
> The community is in the middle of a discussion right now about runtime
> library tuning and what kinds of things we might expose to the users
> to control. Perhaps there is an additional use case in allowing users
> to tune the supported features used by the runtime.
> 
> Could you tell us a little bit more about your use case? Is this a temporary
> thing where AVX support will eventually be added e.g. where hacking
> glibc/sysdeps/x86_64/multiarch/init-firc.c to disable AVX is not an
> option?

i'm using the totalview debugger, a proprietary c++ debugger. it
probably uses similar technologies as valgrind to implement back-in-time
debugging.
it is probably not a common use case, as it is a rather specialized
program. it may eventually support avx instructions, but no idea when it
will be ready.

in a way, it would be nice if the runtime dispatching could be
controlled via environment variables or the like.

for now, i'll try to hack the architecture initialization to disable the
avx detection...

thanks, tim


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]