d/freebsd-nq
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

amd64: tidy up memset to have rax set earlier for small sizes

Browse Source
This commit is contained in:
Mateusz Guzik 2018-10-21 10:46:00 +00:00
parent 146ebc766c
commit bbf3607b86
1 changed files with 17 additions and 15 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

32
sys/amd64/amd64/support.S
View File

@ -452,11 +452,11 @@ END(memcpy_erms)
*/
.macro MEMSET erms
PUSH_FRAME_POINTER
movq %rdi,%r9
movq %rdi,%rax
movq %rdx,%rcx
movzbq %sil,%r8
movabs $0x0101010101010101,%rax
imulq %r8,%rax
movabs $0x0101010101010101,%r10
imulq %r8,%r10
cmpq $32,%rcx
jb 1016f
@ -465,10 +465,10 @@ END(memcpy_erms)
ja 1256f
1032:
movq %rax,(%rdi)
movq %rax,8(%rdi)
movq %rax,16(%rdi)
movq %rax,24(%rdi)
movq %r10,(%rdi)
movq %r10,8(%rdi)
movq %r10,16(%rdi)
movq %r10,24(%rdi)
leaq 32(%rdi),%rdi
subq $32,%rcx
cmpq $32,%rcx
@ -478,54 +478,56 @@ END(memcpy_erms)
1016:
cmpb $16,%cl
jl 1008f
movq %rax,(%rdi)
movq %rax,8(%rdi)
movq %r10,(%rdi)
movq %r10,8(%rdi)
subb $16,%cl
jz 1000f
leaq 16(%rdi),%rdi
1008:
cmpb $8,%cl
jl 1004f
movq %rax,(%rdi)
movq %r10,(%rdi)
subb $8,%cl
jz 1000f
leaq 8(%rdi),%rdi
1004:
cmpb $4,%cl
jl 1002f
movl %eax,(%rdi)
movl %r10d,(%rdi)
subb $4,%cl
jz 1000f
leaq 4(%rdi),%rdi
1002:
cmpb $2,%cl
jl 1001f
movw %ax,(%rdi)
movw %r10w,(%rdi)
subb $2,%cl
jz 1000f
leaq 2(%rdi),%rdi
1001:
cmpb $1,%cl
jl 1000f
movb %al,(%rdi)
movb %r10b,(%rdi)
1000:
movq %r9,%rax
POP_FRAME_POINTER
ret
ALIGN_TEXT
1256:
movq %rdi,%r9
movq %r10,%rax
.if \erms == 1
rep
stosb
movq %r9,%rax
.else
shrq $3,%rcx
rep
stosq
movq %r9,%rax
movq %rdx,%rcx
andb $7,%cl
jne 1004b
.endif
movq %r9,%rax
POP_FRAME_POINTER
ret
.endm