https://gcc.gnu.org/bugzilla/show_bug.cgi?id=62147
Bug ID: 62147
Summary: missed loop counter based optimization
Product: gcc
Version: 5.0
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: target
Assignee: unassigned at gcc dot gnu.org
Reporter: carrot at google dot com
Target: powerpc64le
Compile following source code with options -m64 -mcpu=power8 -O2
typedef struct {
int l;
int b[258];
} S;
void clear (S* s )
{
int i;
int len = s->l + 1; // int len = s->l;
for (i = 0; i <= len; i++)
s->b[i] = 0;
}
Trunk compiler generates:
clear:
lwz 9,0(3)
cmpwi 7,9,-1
bltlr- 7
addi 9,9,1
li 10,0
rldicl 9,9,0,32
addi 9,9,1
sldi 9,9,2
add 9,3,9
.p2align 4,,15
.L3:
stwu 10,4(3)
cmpld 7,3,9
bne+ 7,.L3
blr
It uses cmp/jmp instructions to construct the loop, a better code sequence
should use the bdnz instruction.
If I change the source code as in the comment, gcc can generate bdnz
instruction to form the loop
clear:
lwz 9,0(3)
cmpwi 7,9,0
extsw 9,9
bltlr- 7
sldi 9,9,2
li 10,0
srdi 9,9,2
addi 9,9,1
mtctr 9
.p2align 4,,15
.L3:
stwu 10,4(3)
bdnz .L3
blr