https://gcc.gnu.org/bugzilla/show_bug.cgi?id=125189
Richard Biener <rguenth at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Ever confirmed|0 |1
Blocks| |53947
Status|UNCONFIRMED |NEW
Last reconfirmed| |2026-05-05
Keywords| |missed-optimization
Target| |x86_64-*-*
--- Comment #1 from Richard Biener <rguenth at gcc dot gnu.org> ---
Confirmed. The issue is that affine data dependence analysis has no support
to represent that y[a[i]%16] is an (invariant in a vector iteration)
permutation
of the same contiguous part of y[]. And since we cannot do a runtime
check on x[] vs. y[] aliasing we fail (we actually could, given y[] has
a fixed size but we don't implement that).
You do not provide a vector form for
for (int k=0; k<16; ++k) j[k] = int(a[i+k])%16;
but I suppose the modulo 16 can be done as bitwise and and a[] is accessed
contiguously.
Practically it's difficult to fit this in nicely. Pattern-matching
the gather somehow might work.
Referenced Bugs:
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=53947
[Bug 53947] [meta-bug] vectorizer missed-optimizations
