https://gcc.gnu.org/bugzilla/show_bug.cgi?id=88603
Richard Biener <rguenth at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Status|UNCONFIRMED |NEW
Last reconfirmed| |2019-01-02
CC| |rguenth at gcc dot gnu.org
Version|unknown |8.2.1
Ever confirmed|0 |1
--- Comment #1 from Richard Biener <rguenth at gcc dot gnu.org> ---
Confirmed. GCC "optimized" the code to
<bb 2> [local count: 1073741825]:
_1 = (long unsigned int) a_4(D);
_2 = (long unsigned int) b_5(D);
tmp_6 = _1 + _2;
if (tmp_6 > 4294967295)
goto <bb 4>; [21.72%]
else
goto <bb 3>; [78.28%]
<bb 3> [local count: 840525100]:
_7 = a_4(D) + b_5(D);
<bb 4> [local count: 1073741825]:
# _3 = PHI <_7(3), 4294967295(2)>
return _3;
forwprop1 does this already by means of narrowing patterns in match.pd:
/* Narrowing of arithmetic and logical operations.
These are conceptually similar to the transformations performed for
the C/C++ front-ends by shorten_binary_op and shorten_compare. Long
term we want to move all that code out of the front-ends into here. */
/* If we have a narrowing conversion of an arithmetic operation where
both operands are widening conversions from the same type as the outer
narrowing conversion. Then convert the innermost operands to a suitable
unsigned type (to avoid introducing undefined behavior), perform the
operation and convert the result to the desired type. */
(for op (plus minus)
(simplify
(convert (op:s (convert@2 @0) (convert?@3 @1)))
...
(convert (op (convert:utype @0)
(convert:utype @1))))))))
here :s is not effective for your testcase since there's no extra
conversions in the end. Disabling this yields
<bb 2> [local count: 1073741825]:
_1 = (long unsigned int) a_4(D);
_2 = (long unsigned int) b_5(D);
tmp_6 = _1 + _2;
_10 = MIN_EXPR <tmp_6, 4294967295>;
_3 = (uint32_t) _10;
and not much better code in the end:
saturation_add:
.LFB4:
.cfi_startproc
movl %edi, %edi
movl %esi, %eax
movl $4294967295, %edx
addq %rdi, %rax
cmpq %rdx, %rax
cmova %rdx, %rax
ret
