https://gcc.gnu.org/bugzilla/show_bug.cgi?id=88916
Bug ID: 88916
Summary: [x86] suboptimal code generated for integer
comparisons joined with boolean operators
Product: gcc
Version: 8.2.0
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: target
Assignee: unassigned at gcc dot gnu.org
Reporter: wojciech_mula at poczta dot onet.pl
Target Milestone: ---
Let's consider these two simple, yet pretty useful functions:
--test.c---
int both_nonnegative(long a, long b) {
return (a >= 0) && (b >= 0);
}
int both_nonzero(unsigned long a, unsigned long b) {
return (a > 0) && (b > 0);
}
---eof--
$ gcc --version
gcc (Debian 8.2.0-13) 8.2.0
$ gcc -O3 test.c -march=skylake -S
$ cat test.s
both_nonnegative:
notq %rdi
movq %rdi, %rax
notq %rsi
shrq $63, %rax
shrq $63, %rsi
andl %esi, %eax
ret
both_nonzero:
testq %rdi, %rdi
setne %al
xorl %edx, %edx
testq %rsi, %rsi
setne %dl
andl %edx, %eax
ret
I checked different target machines (haswell, broadwell and cannonlake),
however the result remained the same. Also GCC trunk on godbolt.org
produces the same assembly code.
The first function, `both_nonnegative`, can be rewritten as:
(((unsigned long)(a) | (unsigned long)(b)) >> 63) ^ 1
yielding something like this:
both_nonnegative:
orq %rsi, %rdi
movq %rdi, %rax
shrq $63, %rax
xorl $1, %eax
ret
It's also possible to use this expression:
(long)(unsigned long)a | (unsigned long)b) < 0,
but the assembly output is almost the same.
The condition from `both_nonzero` can be expressed as:
((unsigned long)a | (unsigned long)b) != 0
GCC compiles it to:
both_nonzero:
xorl %eax, %eax
orq %rsi, %rdi
setne %al
retq
