https://gcc.gnu.org/bugzilla/show_bug.cgi?id=91509
Richard Biener <rguenth at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Keywords| |lto, memory-hog
Target| |x86_64-*-*, arm-*-*,
| |aarch64-*-*, s390*-*-*
--- Comment #1 from Richard Biener <rguenth at gcc dot gnu.org> ---
We have loads of
Ahead 19, unroll factor 16, trip count 9223372036854775807
and the code-size growth is BIG. There's a single function in the
respective ltrans unit, 'integrate'.
You can also see the non-LTO binary size to double:
https://gcc.opensuse.org/gcc-old/SPEC/CFP/sb-megrez-head-64-2006/481_wrf-size_big.png
compared to the following increase which has prefetching disabled:
https://gcc.opensuse.org/gcc-old/SPEC/CFP/sb-czerny-head-64-2006/481_wrf-size_big.png
Not sure what to do about this. IIRC arm/aarch64/s390 also enable prefetching
in some circumstances.
