https://gcc.gnu.org/bugzilla/show_bug.cgi?id=92492
Richard Biener <rguenth at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Keywords| |missed-optimization
Status|UNCONFIRMED |NEW
Last reconfirmed| |2019-11-13
Blocks| |53947
Ever confirmed|0 |1
--- Comment #2 from Richard Biener <rguenth at gcc dot gnu.org> ---
t.c:12:14: missed: not vectorized: relevant stmt not supported: _4 = (int)
_3;
# x_25 = PHI <x_13(7), 0(15)>
# ivtmp_29 = PHI <ivtmp_28(7), 16(15)>
_1 = (sizetype) x_25;
_2 = src_9(D) + _1;
_3 = *_2;
_4 = (int) _3;
_5 = dst_10(D) + _1;
_11 = _4 & -64;
_14 = -_4;
_15 = _14 >> 7;
iftmp.0_16 = (unsigned char) _15;
iftmp.0_17 = _11 == 0 ? _3 : iftmp.0_16;
*_5 = iftmp.0_17;
x_13 = x_25 + 1;
ivtmp_28 = ivtmp_29 - 1;
if (ivtmp_28 != 0)
goto <bb 7>; [93.75%]
else
goto <bb 6>; [6.25%]
was the if-converted loop. This requires unpacking of char to int so this
all boils down to failure to narrow all operations to 'short'. Might
be doable with a match.pd pattern simplifying ((int) X) & -64 == 0
to X & -64 == 0 which we maybe already have but there's other uses of _4.
ICC also uses effectively two vector sizes, v8qi and v8hi AFAICS? But
why does it use %ymm then...
Referenced Bugs:
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=53947
[Bug 53947] [meta-bug] vectorizer missed-optimizations
