https://gcc.gnu.org/bugzilla/show_bug.cgi?id=102929
Bug ID: 102929
Summary: [missed optimization] two ways to
rounddown-to-next-multiple
Product: gcc
Version: 11.2.1
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: tree-optimization
Assignee: unassigned at gcc dot gnu.org
Reporter: jengelh at inai dot de
Target Milestone: ---
Input
=====
unsigned long calc(unsigned long x, unsigned long y)
{
return x/y*y;
}
unsigned long calc2(unsigned long x, unsigned long y)
{
return x - x % y;
}
Observed
========
ยป g++ -O3 -c x.c; objdump -Mintel -d x.o
gcc version 11.2.1 20210816 [revision 056e324ce46a7924b5cf10f61010cf9dd2ca10e9]
(SUSE Linux) x86_64
0000000000000000 <_Z4calcmm>:
0: 48 89 f8 mov rax,rdi
3: 31 d2 xor edx,edx
5: 48 f7 f6 div rsi
8: 48 0f af c6 imul rax,rsi
c: c3 ret
d: 0f 1f 00 nop DWORD PTR [rax]
0000000000000010 <_Z5calc2mm>:
10: 48 89 f8 mov rax,rdi
13: 31 d2 xor edx,edx
15: 48 f7 f6 div rsi
18: 48 89 f8 mov rax,rdi
1b: 48 29 d0 sub rax,rdx
1e: c3 ret
Expected
========
I do not see any obvious differences in the outcome of the two C functions, so
I would expect that, ideally, both should lead to the same asm. (Either by
making calc use div-mov-sub, or by making calc2 using div-imul; whichever
happens to be determined more beneficial as per the machine descriptions).
