https://gcc.gnu.org/bugzilla/show_bug.cgi?id=125204

--- Comment #2 from Hunter X <bug_hunters at yeah dot net> ---
(In reply to Richard Biener from comment #1)
> It does sth for me:
> 
> obj-aarch64-g/gcc> ./xgcc -B. t.c -I include -S -O3 -march=armv9-a+sve
> -fopt-info-vec -fno-vect-cost-model
> t.c:13:27: optimized: loop vectorized using 16 byte vectors and unroll
> factor 4
> t.c:13:27: optimized:  loop versioned for vectorization to enhance alignment
> t.c:16:18: optimized: basic block part vectorized using 16 byte vectors

Thanks for checking.

I tried the same option `-fno-vect-cost-model` on Godbolt with GCC 16.1.0, but
it still does not vectorize the loop on my side. The diagnostic still reports:

  missed: unsupported SLP instances
  Analysis failed with vector mode VNx2DI

Could this be due to a difference between the GCC revision used on Godbolt and
your local trunk build?

Reply via email to