## diffname carrera/memmove.s 1994/0209
## diff -e /dev/null /n/fornaxdump/1994/0209/sys/src/brazil/carrera/memmove.s
0a
/*
* R4000 double-word version. Only works in kernel mode.
* Requires low-core support to save R5 and R6 in a 64-bit hole.
*/
/*
* R4000 instructions
*/
#define LD(offset, base, rt) WORD $((067<<26)|((base)<<21)|((rt)<<16)|((offset)&0xFFFF))
#define STD(rt, offset, base) WORD $((077<<26)|((base)<<21)|((rt)<<16)|((offset)&0xFFFF))
TEXT memmove(SB), $0
JMP move
TEXT memcpy(SB), $0
move:
MOVW R1, s1+0(FP)
MOVW n+8(FP), R3 /* R3 is count */
MOVW R1, R4 /* R4 is to-pointer */
SGT R0, R3, R10
BEQ R10, ok
MOVW (R0), R0 /* abort if negative count */
ok:
MOVW s2+4(FP), R10 /* R10 is from-pointer */
ADDU R3,R10, R7 /* R7 is end from-pointer */
ADDU R3,R4, R11 /* R11 is end to-pointer */
/*
* easiest test is copy backwards if
* destination string has higher mem address
*/
SGT $8,R3, R2
SGTU R4,R10, R1
BNE R1, back
/*
* if not at least 8 chars,
* dont even mess around.
* 7 chars to guarantee any
* rounding up to a doubleword
* boundary and 8 characters
* to get at least maybe one
* full doubleword store.
*/
BNE R2, fout
/*
* test if both pointers
* are similarly word aligned
*/
XOR R4,R10, R1
AND $3, R1
BNE R1, fout
/*
* test if both pointers
* are similarly double-word aligned
*/
XOR R4,R10, R1
AND $7, R1
BNE R1, f1
/*
* double-word aligned; start by byte at a time until aligned
*/
fd1:
AND $7,R4, R1
BEQ R1, fd2
MOVB 0(R10), R8
ADDU $1, R10
MOVB R8, 0(R4)
ADDU $1, R4
JMP fd1
/*
* turn R3 into to-end pointer-31
* copy 32 at a time while there's room.
* R11 is smaller than R7 --
* there are problems if R7 is 0.
*/
fd2:
ADDU $-31,R11, R3
fd3:
SGTU R3,R4, R1
BEQ R1, f4
LD (0,(10), 5) /* MOVW 0(R10), R8 */
LD (8,(10), 6) /* MOVW 4(R10), R9 */
STD (5, 0,(4)) /* MOVW R8, 0(R4) */
LD (16,(10), 5) /* MOVW 8(R10), R8 */
STD (6, 8,(4)) /* MOVW R9, 4(R4) */
LD (24,(10), 6) /* MOVW 12(R10), R9 */
ADDU $32, R10
STD (5, 16,(4)) /* MOVW R8, 8(R4) */
STD (6, 24,(4)) /* MOVW R9, 12(R4) */
ADDU $32, R4
JMP fd3
/*
* byte at a time to word align
*/
f1:
AND $3,R4, R1
BEQ R1, f2
MOVB 0(R10), R8
ADDU $1, R10
MOVB R8, 0(R4)
ADDU $1, R4
JMP f1
/*
* turn R3 into to-end pointer-15
* copy 16 at a time while there's room.
* R11 is smaller than R7 --
* there are problems if R7 is 0.
*/
f2:
ADDU $-15,R11, R3
f3:
SGTU R3,R4, R1
BEQ R1, f4
MOVW 0(R10), R8
MOVW 4(R10), R9
MOVW R8, 0(R4)
MOVW 8(R10), R8
MOVW R9, 4(R4)
MOVW 12(R10), R9
ADDU $16, R10
MOVW R8, 8(R4)
MOVW R9, 12(R4)
ADDU $16, R4
JMP f3
/*
* turn R3 into to-end pointer-3
* copy 4 at a time while theres room
*/
f4:
ADDU $-3,R11, R3
f5:
SGTU R3,R4, R1
BEQ R1, fout
MOVW 0(R10), R8
ADDU $4, R10
MOVW R8, 0(R4)
ADDU $4, R4
JMP f5
/*
* last loop, copy byte at a time
*/
fout:
BEQ R7,R10, ret
MOVB 0(R10), R8
ADDU $1, R10
MOVB R8, 0(R4)
ADDU $1, R4
JMP fout
/*
* whole thing repeated for backwards
*/
back:
BNE R2, bout
XOR R11,R7, R1
AND $3, R1
BNE R1, bout
XOR R11,R7, R1
AND $7, R1
BNE R1, b1
/* double words */
bd1:
AND $7,R7, R1
BEQ R1, bd2
MOVB -1(R7), R8
ADDU $-1, R7
MOVB R8, -1(R11)
ADDU $-1, R11
JMP bd1
bd2:
ADDU $31,R10, R3
bd3:
SGTU R7,R3, R1
BEQ R1, b4
LD (-8,(7), 5) /* MOVW -4(R7), R8 */
LD (-16,(7), 6) /* MOVW -8(R7), R9 */
STD (5, -8,(11)) /* MOVW R8, -4(R11) */
LD (-24,(7), 5) /* MOVW -12(R7), R8 */
STD (6, -16,(11)) /* MOVW R9, -8(R11) */
LD (-32,(7), 6) /* MOVW -16(R7), R9 */
ADDU $-32, R7
STD (5, -24,(11)) /* MOVW R8, -12(R11) */
STD (6, -32,(11)) /* MOVW R9, -16(R11) */
ADDU $-32, R11
JMP bd3
/* regular words */
b1:
AND $3,R7, R1
BEQ R1, b2
MOVB -1(R7), R8
ADDU $-1, R7
MOVB R8, -1(R11)
ADDU $-1, R11
JMP b1
b2:
ADDU $15,R10, R3
b3:
SGTU R7,R3, R1
BEQ R1, b4
MOVW -4(R7), R8
MOVW -8(R7), R9
MOVW R8, -4(R11)
MOVW -12(R7), R8
MOVW R9, -8(R11)
MOVW -16(R7), R9
ADDU $-16, R7
MOVW R8, -12(R11)
MOVW R9, -16(R11)
ADDU $-16, R11
JMP b3
b4:
ADDU $3,R10, R3
b5:
SGTU R7,R3, R1
BEQ R1, bout
MOVW -4(R7), R8
ADDU $-4, R7
MOVW R8, -4(R11)
ADDU $-4, R11
JMP b5
bout:
BEQ R7,R10, ret
MOVB -1(R7), R8
ADDU $-1, R7
MOVB R8, -1(R11)
ADDU $-1, R11
JMP bout
ret:
MOVW s1+0(FP), R1
RET
END
.
## diffname carrera/memmove.s 1997/1101
## diff -e /n/fornaxdump/1994/0209/sys/src/brazil/carrera/memmove.s /n/emeliedump/1997/1101/sys/src/brazil/carrera/memmove.s
240,241c
MOVB R8, -1(R6)
ADDU $-1, R6
.
237c
BEQ R7,R5, ret
.
235a
bun2:
ADDU $31,R5, R3
bun3:
SGTU R7,R3, R1
BEQ R1, bun4
MOVVL -8(R7), R8
MOVVR -1(R7), R8
MOVVL -16(R7), R9
MOVVR -9(R7), R9
MOVV R8, -8(R6)
MOVVL -24(R7), R8
MOVVR -17(R7), R8
MOVV R9, -16(R6)
MOVVL -32(R7), R9
MOVVR -25(R7), R9
ADDU $-32, R7
MOVV R8, -24(R6)
MOVV R9, -32(R6)
ADDU $-32, R6
JMP bun3
bun4:
ADDU $3,R5, R3
bun5:
SGTU R7,R3, R1
BEQ R1, bout
MOVWL -4(R7), R8
MOVWR -1(R7), R8
ADDU $-4, R7
MOVW R8, -4(R6)
ADDU $-4, R6
JMP bun5
.
232,233c
MOVW R8, -4(R6)
ADDU $-4, R6
.
226c
ADDU $3,R5, R3
.
214,223c
MOVV -8(R7), R8
MOVV -16(R7), R9
MOVV R8, -8(R6)
MOVV -24(R7), R8
MOVV R9, -16(R6)
MOVV -32(R7), R9
ADDU $-32, R7
MOVV R8, -24(R6)
MOVV R9, -32(R6)
ADDU $-32, R6
.
210c
AND $7, R7, R1
BNE R1, bun2
ADDU $31,R5, R3
.
208a
.
206,207c
MOVB R8, -1(R6)
ADDU $-1, R6
.
202c
AND $7,R6, R1
.
166,200d
154,156c
BEQ R7,R5, ret
MOVB 0(R5), R8
ADDU $1, R5
.
148c
JMP fun5
.
144,145c
MOVWL 0(R5), R8
MOVWR 3(R5), R8
ADDU $4, R5
.
139,141c
fun4:
ADDU $-3,R6, R3
fun5:
.
122,133c
BEQ R1, fun4
MOVVL 0(R5), R8
MOVVR 7(R5), R8
MOVVL 8(R5), R9
MOVVR 15(R5), R9
MOVV R8, 0(R4)
MOVVL 16(R5), R8
MOVVR 23(R5), R8
MOVV R9, 8(R4)
MOVVL 24(R5), R9
MOVVR 31(R5), R9
ADDU $32, R5
MOVV R8, 16(R4)
MOVV R9, 24(R4)
ADDU $32, R4
JMP fun3
.
118,120c
fun2:
ADDU $-31,R6, R3
fun3:
.
113,115c
* forward copy, unaligned
* turn R3 into to-end pointer-31
* copy 32 at a time while theres room.
* R6 is smaller than R7 --
.
103,110c
f4:
ADDU $-3,R6, R3
f5:
SGTU R3,R4, R1
BEQ R1, fout
MOVW 0(R5), R8
ADDU $4, R5
MOVW R8, 0(R4)
ADDU $4, R4
JMP f5
.
101c
* turn R3 into to-end pointer-3
* copy 4 at a time while theres room
.
98c
JMP f3
.
88,96c
MOVV 0(R5), R8
MOVV 8(R5), R9
MOVV R8, 0(R4)
MOVV 16(R5), R8
MOVV R9, 8(R4)
MOVV 24(R5), R9
ADDU $32, R5
MOVV R8, 16(R4)
MOVV R9, 24(R4)
.
83,85c
ADDU $-31,R6, R3
f3:
.
79,80c
* copy 32 at a time while theres room.
* R6 is smaller than R7 --
.
77a
* test if source is now doubleword aligned
*/
f2:
AND $7, R5, R1
BNE R1, fun2
/*
.
75c
JMP f1
.
70,72c
BEQ R1, f2
MOVB 0(R5), R8
ADDU $1, R5
.
53,68c
f1:
.
50,51c
* byte at a time to doubleword align destination
.
45c
* doubleword store.
.
39,40c
* if not at least 15 chars,
* don't even mess around.
.
34,35c
SGT $15,R3, R2
SGTU R4,R5, R1
.
26,28c
MOVW s2+4(FP), R5 /* R5 is from-pointer */
ADDU R3,R5, R7 /* R7 is end from-pointer */
ADDU R3,R4, R6 /* R6 is end to-pointer */
.
22,23c
SGT R0, R3, R5
BEQ R5, ok
.
13d
5,11d
2,3c
* R4000-compatible memmove uses V instructions.
* The kernel uses the 3000 compilers, but we can
* use this procedure for speed.
.
## diffname carrera/memmove.s 2001/0527 # deleted
## diff -e /n/emeliedump/1997/1101/sys/src/brazil/carrera/memmove.s /n/emeliedump/2001/0527/sys/src/9/carrera/memmove.s
1,241d
|