1 dnl PowerPC
-32/VMX
and PowerPC
-64/VMX mpn_copyd.
3 dnl Copyright
2006 Free Software Foundation
, Inc.
5 dnl
This file is part of the GNU MP Library.
7 dnl The GNU MP Library is free software
; you can redistribute it and/or modify
8 dnl it under the terms of
either:
10 dnl
* the GNU Lesser General
Public License as published by the Free
11 dnl Software Foundation
; either version 3 of the License, or (at your
12 dnl option
) any later version.
16 dnl
* the GNU General
Public License as published by the Free Software
17 dnl Foundation
; either version 2 of the License, or (at your option) any
20 dnl
or both
in parallel
, as here.
22 dnl The GNU MP Library is distributed
in the hope that it will be useful
, but
23 dnl WITHOUT ANY WARRANTY
; without even the implied warranty of MERCHANTABILITY
24 dnl
or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General
Public License
27 dnl You should have received copies of the GNU General
Public License
and the
28 dnl GNU Lesser General
Public License along with the GNU MP Library. If
not,
29 dnl see
https://www.gnu.
org/licenses
/.
31 include(`..
/config.m4
')
33 C 16-byte coaligned unaligned
34 C cycles/limb cycles/limb
35 C 7400,7410 (G4): 0.5 0.64
36 C 744x,745x (G4+): 0.75 0.82
37 C 970 (G5): 0.78 1.02 (64-bit limbs)
40 C * Works for all sizes and alignments.
43 C * Optimize unaligned case. Some basic tests with 2-way and 4-way unrolling
44 C indicate that we can reach 0.56 c/l for 7400, 0.75 c/l for 745x, and 0.80
46 C * Consider using VMX instructions also for head and tail, by using some
47 C read-modify-write tricks.
48 C * The VMX code is used from the smallest sizes it handles, but measurements
49 C show a large speed bump at the cutoff points. Small copying (perhaps
50 C using some read-modify-write technique) should be optimized.
51 C * Make an mpn_com based on this code.
53 define(`GMP_LIMB_BYTES', eval
(GMP_LIMB_BITS
/8))
54 define
(`LIMBS_PER_VR
', eval(16/GMP_LIMB_BYTES))
55 define(`LIMBS_PER_2VR', eval
(32/GMP_LIMB_BYTES
))
58 ifelse
(GMP_LIMB_BITS
,32,`
59 define
(`LIMB32
',` $1')
63 define(`LIMB64',`
$1')
77 LIMB32
(`slwi. r0
, n
, 2 ')
78 LIMB64(`sldi. r0, n, 3 ')
82 LIMB32
(`cmpi cr7
, n
, 11 ')
83 LIMB64(`cmpdi cr7, n, 5 ')
88 C Handle
small cases with plain operations
91 LIMB32
(`lwz r0
, -4(up
) ')
92 LIMB64(`ld r0, -8(up) ')
93 addi up
, up
, -GMP_LIMB_BYTES
94 LIMB32
(`stw r0
, -4(rp
) ')
95 LIMB64(`std r0, -8(rp) ')
96 addi rp
, rp
, -GMP_LIMB_BYTES
100 C Handle
large cases with VMX operations
105 oris r0
, r12
, 0xf800 C Set VRSAVE bit
0-4
108 LIMB32
(`rlwinm. r7
, rp
, 30,30,31') C (rp >> 2) mod 4
109 LIMB64(`rlwinm. r7, rp, 29,31,31') C
(rp
>> 3) mod 2
114 LIMB32
(`lwz r0
, 12(up
) ')
115 LIMB64(`ld r0, 8(up) ')
116 addi up
, up
, -GMP_LIMB_BYTES
117 LIMB32
(`addic. r7
, r7
, -1 ')
118 LIMB32(`stw r0, 12(rp) ')
119 LIMB64
(`
std r0
, 8(rp
) ')
120 addi rp, rp, -GMP_LIMB_BYTES
121 LIMB32(`bne L(top0) ')
125 LIMB32
(`rlwinm. r0
, up
, 30,30,31') C (up >> 2) mod 4
126 LIMB64(`rlwinm. r0, up, 29,31,31') C
(up
>> 3) mod 2
128 LIMB64
(`srdi r7
, n
, 2 ') C loop count corresponding to n
129 LIMB32(`srwi r7, n, 3 ') C
loop count corresponding to n
130 mtctr r7 C copy n to count register
139 LIMB32
(`andi. r0
, n
, 0x4 ')
140 LIMB64(`andi. r0, n, 0x2 ')
154 L
(lpu
): lvx v0
, 0, up
168 LIMB32
(`andi. r0
, n
, 0x4 ')
169 LIMB64(`andi. r0, n, 0x2 ')
178 L
(lpa
): lvx v0
, 0, up
188 LIMB32
(`rlwinm. r7
, n
, 0,30,31 ') C r7 = n mod 4
189 LIMB64(`rlwinm. r7, n, 0,31,31 ') C r7
= n
mod 2
191 LIMB32
(`li r10
, 12 ')
193 LIMB32(`lwzx r0, r10, up ')
194 LIMB64
(`ld r0
, 8(up
) ')
195 LIMB32(`addic. r7, r7, -1 ')
196 LIMB32
(`stwx r0
, r10
, rp
')
197 LIMB64(`std r0, 8(rp) ')
198 LIMB32
(`addi r10
, r10
, -GMP_LIMB_BYTES
')
199 LIMB32(`bne L(top2) ')
201 L
(ret): mtspr
256, r12