Lephenixnoir
/
gint
10
14
Fork 8
Code Issues 4 Pull Requests Releases Wiki Activity

libc: faster memcpy for on-chip memory 

An optimization suggested by TSWilliamson, which pushes not only RAM,
but also on-chip memory and the CPU pipeline to their limits.
This commit is contained in:
Lephe 2020-07-27 22:46:17 +02:00
parent 7b4eb078c4
commit 492f61f7b2
Signed by: Lephenixnoir
GPG Key ID: 1BBA026E13FC0495
1 changed files with 30 additions and 3 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

33
src/std/memcpy.s
View File

@ -24,7 +24,7 @@ _memcpy_align_dst:
/* If source is 4-aligned, use mov.l */
tst r2, r5
bt/s .aligned4
bt/s .aligned4_32
mov #4, r2
/* If unaligned but SH4, use movua.l */
@ -42,13 +42,40 @@ _memcpy_align_dst:
bra _naive_memcpy
nop
.aligned4:
.aligned4_32:
mov #36, r2
/* Copy 32 bytes at a time until at most 32 bytes are left */
mov.l @r5+, r0
mov.l @r5+, r1
mov.l @r5+, r7
mov.l r0, @r4
mov.l r1, @(4,r4)
mov.l r7, @(8,r4)
mov.l @r5+, r0
mov.l @r5+, r1
mov.l @r5+, r7
mov.l r0, @(12,r4)
mov.l r1, @(16,r4)
mov.l r7, @(20,r4)
mov.l @r5+, r0
mov.l @r5+, r1
add #-32, r6
mov.l r0, @(24,r4)
mov.l r1, @(28,r4)
cmp/ge r6, r2
bf/s .aligned4_32
add #32, r4
.aligned4_4:
mov #4, r2
/* Copy 4 bytes at a time until at most 4 bytes are left */
mov.l @r5+, r0
mov.l r0, @r4
add #-4, r6
cmp/ge r6, r2
bf/s .aligned4
bf/s .aligned4_4
add #4, r4
bra _naive_memcpy