https://gcc.gnu.org/bugzilla/show_bug.cgi?id=94945
Andrew Pinski <pinskia at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Resolution|--- |FIXED
Status|UNCONFIRMED |RESOLVED
See Also| |https://gcc.gnu.org/bugzill
| |a/show_bug.cgi?id=94913,
| |https://gcc.gnu.org/bugzill
| |a/show_bug.cgi?id=97387
Target Milestone|--- |11.0
--- Comment #2 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
GCC 11+ can produce:
.L3:
movl %eax, %r10d
addb $-1, %r9b
leal 1(%rax), %r9d
movq (%rdi,%r10,8), %rdx
adcq (%rsi,%r10,8), %rdx
movq %rdx, (%r8,%r10,8)
movq (%rdi,%r9,8), %rdx
leal 3(%rax), %r10d
adcq (%rsi,%r9,8), %rdx
movq %rdx, (%r8,%r9,8)
leal 2(%rax), %r9d
movq (%rdi,%r9,8), %rdx
adcq (%rsi,%r9,8), %rdx
movq %rdx, (%r8,%r9,8)
movq (%rdi,%r10,8), %rdx
adcq (%rsi,%r10,8), %rdx
setc %r9b
addl $4, %eax
movq %rdx, (%r8,%r10,8)
cmpl %eax, %ecx
ja .L3
With a single iteration and -funroll-loops GCC 11 gives this for the inner
loop:
.L12:
movq (%r8,%r10), %r11
addb $-1, %dl
adcq (%rdi,%r10), %r11
movq %r11, (%rsi,%r10)
movq 8(%r8,%r10), %r9
adcq 8(%rdi,%r10), %r9
movq %r9, 8(%rsi,%r10)
movq 16(%r8,%r10), %rax
adcq 16(%rdi,%r10), %rax
movq %rax, 16(%rsi,%r10)
movq 24(%r8,%r10), %rdx
adcq 24(%rdi,%r10), %rdx
movq %rdx, 24(%rsi,%r10)
movq 32(%r8,%r10), %r11
adcq 32(%rdi,%r10), %r11
movq %r11, 32(%rsi,%r10)
movq 40(%r8,%r10), %r9
adcq 40(%rdi,%r10), %r9
movq %r9, 40(%rsi,%r10)
movq 48(%r8,%r10), %rax
adcq 48(%rdi,%r10), %rax
movq %rax, 48(%rsi,%r10)
movq 56(%r8,%r10), %r11
adcq 56(%rdi,%r10), %r11
movq %r11, 56(%rsi,%r10)
setc %dl
addq $64, %r10
cmpq %rcx, %r10
jne .L12
So Fixed in GCC 11.
Note was most likely fixed by r11-145 (which was pushed around the time you
filed the bug) and r11-3882 (later on last year).
