[PATCH] x86-64: Optimize e_expf with FMA [BZ #21912]

Szabolcs Nagy szabolcs.nagy@arm.com
Wed Aug 16 15:14:00 GMT 2017


On 16/08/17 15:59, Arjan van de Ven wrote:
> On 8/16/2017 7:56 AM, Szabolcs Nagy wrote:
>> based on my benchmarks portable c code can
>> easily beat the hand written sse asm
> 
> the hand written sse asm is not great on skylake or similar new cpus.
> (few serious glass jaws)
> 
> the avx one is a lot better... but I'll welcome the challenge.
> 
> if you have a version in C, can you give an URL or something to it;
> I'll just hook it into my local testbench and compare
> 

you can wait for a few days when i clean up the patches
for glibc, this one has inline error handling and
various config options that may not be relevant for glibc

https://github.com/ARM-software/optimized-routines/blob/master/math/e_expf.c
https://github.com/ARM-software/optimized-routines/blob/master/math/e_exp2f_data.c



More information about the Libc-alpha mailing list