https://gcc.gnu.org/bugzilla/show_bug.cgi?id=60145
Georg-Johann Lay <gjl at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
See Also| |https://gcc.gnu.org/bugzill
| |a/show_bug.cgi?id=84211
--- Comment #5 from Georg-Johann Lay <gjl at gcc dot gnu.org> ---
(In reply to Matthijs Kooijman from comment #0)
> uint32_t join4 (uint8_t a, uint8_t b, uint8_t c, uint8_t d)
> {
> return ((uint32_t)a << 24) | ((uint32_t)b << 16) | ((uint32_t)c <<
> 8) | d;
> }
With PR84211 in place, this test case improves a bit in v15:
Previous (without PR84211 resp -O3 -mno-fuse-move):
join4:
push r14 ; 50 [c=4 l=1] pushqi1/0
push r15 ; 51 [c=4 l=1] pushqi1/0
push r16 ; 52 [c=4 l=1] pushqi1/0
push r17 ; 53 [c=4 l=1] pushqi1/0
/* prologue: function */
mov r25,r20 ; 34 [c=4 l=1] movqi_insn/0
ldi r19,0 ; 40 [c=4 l=1] movqi_insn/0
ldi r20,0 ; 41 [c=4 l=2] *movhi/1
ldi r21,0
or r19,r25 ; 43 [c=4 l=1] *iorqi3/0
or r20,r22 ; 45 [c=4 l=1] *iorqi3/0
movw r14,r18 ; 46 [c=4 l=2] *movsi/0
movw r16,r20
or r17,r24 ; 48 [c=4 l=1] *iorqi3/0
movw r24,r16 ; 49 [c=4 l=2] *movsi/0
movw r22,r14
/* epilogue start */
pop r17 ; 56 [c=4 l=1] popqi
pop r16 ; 57 [c=4 l=1] popqi
pop r15 ; 58 [c=4 l=1] popqi
pop r14 ; 59 [c=4 l=1] popqi
ret ; 60 [c=0 l=1] return_from_epilogue
With PR84211 / -mfuse-move:
join4:
push r14 ; 50 [c=4 l=1] pushqi1/0
push r15 ; 51 [c=4 l=1] pushqi1/0
push r16 ; 52 [c=4 l=1] pushqi1/0
push r17 ; 53 [c=4 l=1] pushqi1/0
/* prologue: function */
mov r19,r20 ; 66 [c=4 l=1] movqi_insn/0
mov r16,r22 ; 70 [c=4 l=1] movqi_insn/0
mov r17,r24 ; 72 [c=4 l=1] movqi_insn/0
movw r22,r18 ; 78 [c=4 l=1] *movhi/0
movw r24,r16 ; 79 [c=4 l=1] *movhi/0
/* epilogue start */
pop r17 ; 56 [c=4 l=1] popqi
pop r16 ; 57 [c=4 l=1] popqi
pop r15 ; 58 [c=4 l=1] popqi
pop r14 ; 59 [c=4 l=1] popqi
ret ; 60 [c=0 l=1] return_from_epilogue
So the body shinks from 11 instructions to 5.
Unfortunately, IRA does only a sub-optimal job at register allocation. And
PR84211 runs after reg-alloc, so the register pressure / prologue-epilogue size
won't go down :-(
