https://gcc.gnu.org/bugzilla/show_bug.cgi?id=84067
Bug ID: 84067
Summary: [8 regression] gcc.dg/wmul-1.c regression on aarch64
after r257077
Product: gcc
Version: 8.0
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: middle-end
Assignee: unassigned at gcc dot gnu.org
Reporter: clyon at gcc dot gnu.org
Target Milestone: ---
After r257077, I have noticed a regression on aarch64:
FAIL: gcc.dg/wmul-1.c scan-tree-dump-not widening_mul "WIDEN_MULT_PLUS_EXPR"
Richi said:
Late forwprop does
@@ -75,7 +21,7 @@
_1 = (long unsigned int) Idx_6(D);
_2 = _1 * 40;
_12 = _1 * 4;
- _17 = _2 + _12;
+ _17 = _1 * 44;
_13 = Arr_7(D) + _17;
MEM[(int[10] *)_13] = 1;
_4 = _2 + 400;
_18 = _4 + _12;
_16 = Arr_7(D) + _18;
MEM[(int[10] *)_16] = 2;
which I'm not sure ends up profitable at the end. It makes _12
dead so the total number of multiplications stays the same.
For this we then apply the WIDEN_MULT_PLUS_EXPR two
times given _2 is no longer used multiple times.
The reason the above transform happens is the match.pd :s
behavior which "ignores" :s in case the resulting expression
is "simple".
