Aren
/
NetBSD
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Just use a store/load pair to get the pattern into the FP reg. This eliminates 

a bunch of cruft and avoids using a v9a instruction.
In addition, eliminate 8 of the fmovda's, which we are not using the result of
anyway.
Net result is that this should be faster in all cases.
This commit is contained in:
mycroft 2000-07-23 20:34:07 +00:00
parent 601d3aa149
commit e740400671
1 changed files with 4 additions and 24 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

28
lib/libc/arch/sparc64/string/bzero.S
View File

@ -1,4 +1,4 @@
/* $NetBSD: bzero.S,v 1.3 2000/07/08 16:02:15 eeh Exp $ */
/* $NetBSD: bzero.S,v 1.4 2000/07/23 20:34:07 mycroft Exp $ */
/*
* Copyright (c) 1992, 1993, 1999
@ -51,7 +51,7 @@
#if 0
.asciz "@(#)bzero.s 8.1 (Berkeley) 6/4/93"
#else
RCSID("$NetBSD: bzero.S,v 1.3 2000/07/08 16:02:15 eeh Exp $")
RCSID("$NetBSD: bzero.S,v 1.4 2000/07/23 20:34:07 mycroft Exp $")
#endif
#endif /* LIBC_SCCS and not lint */
@ -132,7 +132,7 @@ Lbzero_internal:
bl,a,pn %icc, Lbzero_cleanup ! Less than 8 left
dec 8, %o1 ! Fixup count -8
2:
!! Now we're 64-bit aligned
!! Now we are 64-bit aligned
cmp %o1, 256 ! Use block clear if len > 256
bge,pt %xcc, Lbzero_block ! use block store insns
deccc 8, %o1
@ -223,22 +223,10 @@ Lbzero_block:
dec 8, %i1
2:
brz,pt %i2, 4f ! Do we have a pattern to load?
fzero %f0 ! Set up FPU
btst 1, %fp
bnz,pt %icc, 3f ! 64-bit stack?
nop
stw %i2, [%fp + 0x28] ! Flush this puppy to RAM
membar #StoreLoad
ld [%fp + 0x28], %f0
ba,pt %icc, 4f
fmovsa %icc, %f0, %f1
3:
stx %i2, [%fp + BIAS + 0x50] ! Flush this puppy to RAM
membar #StoreLoad
ldd [%fp + BIAS + 0x50], %f0
4:
fmovda %icc, %f0, %f2 ! Duplicate the pattern
fmovda %icc, %f0, %f4
fmovda %icc, %f0, %f6
@ -246,14 +234,6 @@ Lbzero_block:
fmovda %icc, %f0, %f10
fmovda %icc, %f0, %f12
fmovda %icc, %f0, %f14
fmovda %icc, %f0, %f16 ! And second bank
fmovda %icc, %f0, %f18
fmovda %icc, %f0, %f20
fmovda %icc, %f0, %f22
fmovda %icc, %f0, %f24
fmovda %icc, %f0, %f26
fmovda %icc, %f0, %f28
fmovda %icc, %f0, %f30
!! Remember: we were 8 bytes too far
dec 56, %i1 ! Go one iteration too far