http://gcc.gnu.org/bugzilla/show_bug.cgi?id=49687
Summary: AVR: Missed optimization for widening MUL
Product: gcc
Version: 4.6.1
Status: UNCONFIRMED
Keywords: missed-optimization
Severity: normal
Priority: P3
Component: target
AssignedTo: unassig...@gcc.gnu.org
ReportedBy: g...@gcc.gnu.org
Target: avr
With avr-gcc 4.6.1 and
-Os -mmcu=atmega88 -S -dp
int mul8_55 (char x)
{
return x * 55;
}
gets compiled to
mul8_55:
ldi r25,lo8(55) ; 7 *movqi/2 [length = 1]
muls r24,r25 ; 8 mulqihi3 [length = 3]
movw r24,r0
clr r1
ret ; 26 return [length = 1]
which is fine. If the constant is 126, however, the result is bloated and
might be related to PR36467, i.e. MUL* is better than shift.
mul8_126:
clr r25 ; 7 extendqihi2/1 [length = 3]
sbrc r24,7
com r25
movw r18,r24 ; 28 *movhi/1 [length = 1]
lsl r18 ; 33 *ashlhi3_const/2 [length = 2]
rol r19
lsr r25 ; 34 *ashlhi3_const/5 [length = 5]
mov r25,r24
clr r24
ror r25
ror r24
sub r24,r18 ; 12 subhi3/1 [length = 2]
sbc r25,r19
ret ; 31 return [length = 1]
If the constant is 155, a MUL is invented, but a widening MUL would be smarter:
mul8_155:
mov r20,r24 ; 6 extendqihi2/2 [length = 4]
clr r21
sbrc r20,7
com r21
ldi r18,lo8(155) ; 7 *movhi/4 [length = 2]
ldi r19,hi8(155)
mul r20,r18 ; 8 *mulhi3_enh [length = 7]
movw r24,r0
mul r20,r19
add r25,r0
mul r21,r18
add r25,r0
clr r1
ret ; 26 return [length = 1]
