https://gcc.gnu.org/bugzilla/show_bug.cgi?id=89081
Bug ID: 89081
Summary: [x86] suboptimal code generated for condition
expression returning negation
Product: gcc
Version: 9.0
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 this trivial function:
---clamp.c---
#include <stdint.h>
uint64_t clamp1(int64_t x) {
return (x < 0) ? -x : 0;
}
---eof---
$ gcc --version
gcc (GCC) 9.0.0 20190117 (experimental)
$ gcc -O3 -march=skylake clamp.c -c -S && cat clamp.s
clamp1:
movq %rdi, %rax
negq %rax
movl $0, %edx
testq %rdi, %rdi
cmovns %rdx, %rax
ret
This procedure can be way shorter, like this
clamp1:
xorq %rax, %rax # res = 0
negq %rdi # -x, sets SF
cmovns %rdi, %rax
ret
One thing I observed recently when looking at assembly is that GCC
never modifies input registers %rdi or %rsi, always makes their
copies -- -thus the proposed shorter version is not possible.
However, clang modifies these registers, so seems the ABI allows this.
