[Bug target/121290] [16 regression] Regressions in TSVC s119, s3113, s312, s313, s314, s315, s316 since r16-2159-g3bf2aa834e1270

[ramana at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=121290

--- Comment #17 from Ramana Radhakrishnan <ramana at gcc dot gnu.org> ---



(In reply to Tamar Christina from comment #16)
> (In reply to Soumya AR from comment #13)
> > Hi Tamar,
> > 
> > Thanks for the fix.
> > 
> > This has now brought back performance for the mentioned kernels with -Ofast
> > but is now regressing with -O3 ...
> > 
> > Is this something you're still looking at? Just wanted to put it up here
> > anyway.
> > 
> > Example, for the inner loop in s314:
> > 
> 
> Looks like the ICE was compiler explorer caching the compile.
> 
> But trunk seems fine https://godbolt.org/z/1xhdM95sK
> 
> It does outer-loop vect as it's supposed to.
> 
> And VLA works correctly too https://godbolt.org/z/5of8Kq6hb
> 
> Am I missing some option?



As Soumya is out for a few days , I'll try and fill in the details - the
options from our nightly system (because of how it all munges options together
are)

"-O3 -Wl,-z,muldefs -lm -flto=auto -Wl,--sort-section=name -march=native
-mcpu=neoverse-v2"

I don't fully grok your last response in #c14 -  Is the expectation that GCC
trunk will no longer vectorize this loop at -O3 ? 


We also see a bunch of other regressions some of which are at -Ofast with
Rajaperf Apps_Mass3DEA for the options 

"-Ofast -Wl,-z,muldefs -lm -fpermissive -march=native -mcpu=neoverse-v2
-msve-vector-bits=128"
"-Ofast -Wl,-z,muldefs -lm -fpermissive -march=native -mcpu=neoverse-v2"

i.e. VLS Ofast and VLA Ofast. 

We'll try and reduce this and put it over here as well. 

regards
Ramana