https://gcc.gnu.org/bugzilla/show_bug.cgi?id=104116
Bug ID: 104116
Summary: Optimize {FLOOR,CEIL,ROUND}_{DIV,MOD}_EXPR in
tree-vect-patterns.cc
Product: gcc
Version: 12.0
Status: UNCONFIRMED
Keywords: ice-on-valid-code
Severity: normal
Priority: P3
Component: tree-optimization
Assignee: unassigned at gcc dot gnu.org
Reporter: jakub at gcc dot gnu.org
CC: asolokha at gmx dot com, bergner at gcc dot gnu.org,
jakub at gcc dot gnu.org, linkw at gcc dot gnu.org,
luoxhu at gcc dot gnu.org, rguenth at gcc dot gnu.org,
seurer at gcc dot gnu.org, wschmidt at gcc dot gnu.org
Depends on: 102860
Blocks: 102938
Target Milestone: ---
Host: powerpc64le-linux-gnu
Target: powerpc64le-linux-gnu
Build: powerpc64le-linux-gnu
+++ This bug was initially created as a clone of Bug #102860 +++
As mentioned in https://gcc.gnu.org/bugzilla/show_bug.cgi?id=102860#c7, we
should
try to pattern recognize these operations and turn them into
TRUNC_{DIV,MOD}_EXPR
if the optab is supported (note, for constant last operand build on top of the
current vect_recog_divmod_pattern, so that we handle it even if the optab isn't
supported etc.).
For signed operands,
r = x %[fl] y;
is
r = x % y; if (r && (x ^ y) < 0) r += y;
and
d = x /[fl] y;
is
r = x % y; d = x / y; if (r && (x ^ y) < 0) --d;
and
r = x %[cl] y;
is
r = x % y; if (r && (x ^ y) >= 0) r -= y;
and
d = /[cl] y;
is
r = x % y; d = x / y; if (r && (x ^ y) >= 0) ++d;
(too lazy to write rounding case or do the unsigned case of %[cl] or /[cl] -
([fl] isn't needed for unsigned)). Of course, in vect-patterns the ifs should
be written as COND_EXPRs and probably the conditional increments or decrements
should be done using corresponding unsigned type with casts.
Referenced Bugs:
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=102860
[Bug 102860] [12 regression] libgomp.fortran/simd2.f90 ICEs after r12-4526
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=102938
[Bug 102938] [12 regression] ICE in fortran test cases after r12-4240
