https://gcc.gnu.org/bugzilla/show_bug.cgi?id=93742
Bug ID: 93742
Summary: Optimization request: pattern-match typical addition
overflow checks
Product: gcc
Version: 10.0
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: middle-end
Assignee: unassigned at gcc dot gnu.org
Reporter: tavianator at gmail dot com
Target Milestone: ---
It would be nice if GCC could pattern-match the typical pattern for
overflow-safe addition and optimize it to just branch on the overflow flag.
Right now, this:
_Bool add_overflow(int a, int b, int *res) {
if ((a > 0 && b > INT_MAX - a) || (a < 0 && b < INT_MIN - a)) {
return 1;
} else {
*res = a + b;
return 0;
}
}
compiles to this:
testl %edi, %edi
jle .L2
movl $2147483647, %ecx
movl $1, %eax
subl %edi, %ecx
cmpl %esi, %ecx
jl .L12
.L4:
addl %esi, %edi
xorl %eax, %eax
movl %edi, (%rdx)
ret
.L12:
ret
.L2:
je .L4
movl $-2147483648, %ecx
movl $1, %eax
subl %edi, %ecx
cmpl %esi, %ecx
jle .L4
ret
#48580 is similar but about multiplication, for which the overflow-safe pattern
is a lot more complicated. But the addition one above is a lot simpler and
pretty widespread I think. For example, it's what CERT recommends:
https://wiki.sei.cmu.edu/confluence/display/c/INT32-C.+Ensure+that+operations+on+signed+integers+do+not+result+in+overflow
