[Bug tree-optimization/87859] [8/9 Regression] store-merging miscompilation of mesa

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=87859

--- Comment #10 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
Created attachment 44991
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=44991&action=edit
gcc9-pr87859.patch

This is the approach I had my mind.  For *_24.c it makes no difference, but for
*_25.c, instead of:
New sequence of 2 stores to replace old one of 14 stores
New sequence of 1 stores to replace old one of 6 stores
it now emits:
New sequence of 1 stores to replace old one of 8 stores
New sequence of 2 stores to replace old one of 10 stores
New sequence of 1 stores to replace old one of 6 stores
Resulting assembly is one insn larger.  So, something that would need to be
analyzed.