NetBSD/sys/lib/libkern/arch/mips/memcpy.S

208 lines
3.5 KiB
ArmAsm

/* $NetBSD: memcpy.S,v 1.9 2005/02/26 22:58:56 perry Exp $ */
/* XXXX We need to define this in a way which supports multiple architectures */
#include <machine/cdefs.h> /* Get SZREG correct */
#include <mips/asm.h>
#include <machine/endian.h>
#if defined(_LP64)
#define ADDU daddu
#define SUBU dsubu
#else
#define ADDU addu
#define SUBU subu
#endif
.set push
.set noreorder
.text
/*
* {ov}bcopy(from, to, len)
*/
LEAF(bcopy)
XLEAF(ovbcopy)
move v0, a0 # swap from and to for compat interface.
move a0, a1
move a1, v0
/*
* memcpy(to, from, len)
*/
XLEAF(memmove)
XLEAF(memcpy)
.set noat
move v0, a1 # save dst for return value.
/*
* Make sure we can copy forwards.
*/
sltu t0,a1,a0 # t0 == a1 < a0
ADDU a3,a1,a2 # a3 == end of source
sltu t1,a0,a3 # t1 == a0 < a1+a2
and t2,t0,t1 # overlap -- copy backwards
bne t2,zero,backcopy
/*
* There are four alignment cases (with frequency)
* (Based on measurements taken with a DECstation 5000/200
* inside a Mach kernel.)
*
* aligned -> aligned (mostly)
* unaligned -> aligned (sometimes)
* aligned,unaligned -> unaligned (almost never)
*
* Note that we could add another case that checks if
* the destination and source are unaligned but the
* copy is alignable. eg if src and dest are both
* on a halfword boundary.
*/
andi t1,a0,3 # BDSLOT: get last 2 bits of dest
bne t1,zero,bytecopy
andi t0,a1,3 # BDSLOT: get last 2 bits of src
bne t0,zero,destaligned
/*
* Forward aligned->aligned copy, 8*4 bytes at a time.
*/
li AT,-32 # BDSLOT
and t0,a2,AT # count truncated to multiple of 32
ADDU a3,a1,t0 # run fast loop up to this address
sltu AT,a1,a3 # any work to do?
beq AT,zero,wordcopy
SUBU a2,t0 # BDSLOT
/*
* loop body
*/
#if SZREG == 8
.set mips3
or t3,a0,a1 # 4 or 8 byte aligned?
andi t3,t3,7
bne t3,zero,cp
nop # BDSLOT
cp8:
ld t3,0(a1)
ld v1,8(a1)
ld t0,16(a1)
ld t1,24(a1)
ADDU a1,32
sd t3,0(a0)
sd v1,8(a0)
sd t0,16(a0)
sd t1,24(a0)
bne a1,a3,cp8
ADDU a0,32 # BDSLOT
b wordcopy
nop # BDSLOT
#endif
cp:
lw t3,0(a1)
lw v1,4(a1)
lw t0,8(a1)
lw t1,12(a1)
ADDU a1,32
sw t3,0(a0)
sw v1,4(a0)
sw t0,8(a0)
sw t1,12(a0)
lw t1,-4(a1)
lw t0,-8(a1)
lw v1,-12(a1)
lw t3,-16(a1)
ADDU a0,32
sw t1,-4(a0)
sw t0,-8(a0)
sw v1,-12(a0)
bne a1,a3,cp
sw t3,-16(a0) # BDSLOT
/*
* Copy a word at a time, no loop unrolling.
*/
wordcopy:
andi t2,a2,3 # get byte count / 4
SUBU t2,a2,t2 # t2 = number of words to copy * 4
beq t2,zero,bytecopy
ADDU t0,a1,t2 # BDSLOT stop at t0
SUBU a2,a2,t2
1:
lw t3,0(a1)
ADDU a1,4
sw t3,0(a0)
#ifdef MIPS3_5900
nop
nop
#endif
bne a1,t0,1b
ADDU a0,4 # BDSLOT
bytecopy:
beq a2,zero,copydone # nothing left to do?
nop
2:
lb t3,0(a1)
ADDU a1,1
sb t3,0(a0)
SUBU a2,1
#ifdef MIPS3_5900
nop
#endif
bgtz a2,2b
ADDU a0,1 # BDSLOT
copydone:
j ra
nop
/*
* Copy from unaligned source to aligned dest.
*/
destaligned:
andi t0,a2,3 # t0 = bytecount mod 4
SUBU a3,a2,t0 # number of words to transfer
beq a3,zero,bytecopy
ADDU a3,a1,a3 # BDSLOT: stop point for destaligned
move a2,t0 # this many to do after we are done
3:
LWHI t3,0(a1)
LWLO t3,3(a1)
addi a1,4
sw t3,0(a0)
#ifdef MIPS3_5900
nop
#endif
bne a1,a3,3b
addi a0,4 # BDSLOT
j bytecopy
nop
/*
* Copy by bytes backwards.
*/
backcopy:
blez a2,copydone # nothing left to do?
ADDU t0,a1,a2 # BDSLOT: end of source
ADDU t1,a0,a2 # end of destination
4:
lb t3,-1(t0)
SUBU t0,1
sb t3,-1(t1)
#ifdef MIPS3_5900
nop
nop
#endif
bne t0,a1,4b
SUBU t1,1 # BDSLOT
j ra
nop
.set at
END(bcopy)
.set pop