|
Packit |
5c3484 |
dnl PowerPC-64 mpn_and_n, mpn_andn_n, mpn_nand_n, mpn_ior_n, mpn_iorn_n,
|
|
Packit |
5c3484 |
dnl mpn_nior_n, mpn_xor_n, mpn_xnor_n -- mpn bitwise logical operations.
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
dnl Copyright 2003-2005 Free Software Foundation, Inc.
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
dnl This file is part of the GNU MP Library.
|
|
Packit |
5c3484 |
dnl
|
|
Packit |
5c3484 |
dnl The GNU MP Library is free software; you can redistribute it and/or modify
|
|
Packit |
5c3484 |
dnl it under the terms of either:
|
|
Packit |
5c3484 |
dnl
|
|
Packit |
5c3484 |
dnl * the GNU Lesser General Public License as published by the Free
|
|
Packit |
5c3484 |
dnl Software Foundation; either version 3 of the License, or (at your
|
|
Packit |
5c3484 |
dnl option) any later version.
|
|
Packit |
5c3484 |
dnl
|
|
Packit |
5c3484 |
dnl or
|
|
Packit |
5c3484 |
dnl
|
|
Packit |
5c3484 |
dnl * the GNU General Public License as published by the Free Software
|
|
Packit |
5c3484 |
dnl Foundation; either version 2 of the License, or (at your option) any
|
|
Packit |
5c3484 |
dnl later version.
|
|
Packit |
5c3484 |
dnl
|
|
Packit |
5c3484 |
dnl or both in parallel, as here.
|
|
Packit |
5c3484 |
dnl
|
|
Packit |
5c3484 |
dnl The GNU MP Library is distributed in the hope that it will be useful, but
|
|
Packit |
5c3484 |
dnl WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
|
|
Packit |
5c3484 |
dnl or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
|
|
Packit |
5c3484 |
dnl for more details.
|
|
Packit |
5c3484 |
dnl
|
|
Packit |
5c3484 |
dnl You should have received copies of the GNU General Public License and the
|
|
Packit |
5c3484 |
dnl GNU Lesser General Public License along with the GNU MP Library. If not,
|
|
Packit |
5c3484 |
dnl see https://www.gnu.org/licenses/.
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
include(`../config.m4')
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
C cycles/limb
|
|
Packit |
5c3484 |
C POWER3/PPC630 1.75
|
|
Packit |
5c3484 |
C POWER4/PPC970 2.10
|
|
Packit |
5c3484 |
C POWER5 ?
|
|
Packit |
5c3484 |
C POWER6 ?
|
|
Packit |
5c3484 |
C POWER7 1.75
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
C n POWER3/PPC630 POWER4/PPC970
|
|
Packit |
5c3484 |
C 1 15.00 15.33
|
|
Packit |
5c3484 |
C 2 7.50 7.99
|
|
Packit |
5c3484 |
C 3 5.33 6.00
|
|
Packit |
5c3484 |
C 4 4.50 4.74
|
|
Packit |
5c3484 |
C 5 4.20 4.39
|
|
Packit |
5c3484 |
C 6 3.50 3.99
|
|
Packit |
5c3484 |
C 7 3.14 3.64
|
|
Packit |
5c3484 |
C 8 3.00 3.36
|
|
Packit |
5c3484 |
C 9 3.00 3.36
|
|
Packit |
5c3484 |
C 10 2.70 3.25
|
|
Packit |
5c3484 |
C 11 2.63 3.11
|
|
Packit |
5c3484 |
C 12 2.58 3.00
|
|
Packit |
5c3484 |
C 13 2.61 3.02
|
|
Packit |
5c3484 |
C 14 2.42 2.82
|
|
Packit |
5c3484 |
C 15 2.40 2.79
|
|
Packit |
5c3484 |
C 50 2.08 2.67
|
|
Packit |
5c3484 |
C 100 1.85 2.31
|
|
Packit |
5c3484 |
C 200 1.80 2.18
|
|
Packit |
5c3484 |
C 400 1.77 2.14
|
|
Packit |
5c3484 |
C 1000 1.76 2.10#
|
|
Packit |
5c3484 |
C 2000 1.75# 2.13
|
|
Packit |
5c3484 |
C 4000 2.30 2.57
|
|
Packit |
5c3484 |
C 8000 2.62 2.58
|
|
Packit |
5c3484 |
C 16000 2.52 4.25
|
|
Packit |
5c3484 |
C 32000 2.49 16.25
|
|
Packit |
5c3484 |
C 64000 2.66 18.76
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
ifdef(`OPERATION_and_n',
|
|
Packit |
5c3484 |
` define(`func',`mpn_and_n')
|
|
Packit |
5c3484 |
define(`logop', `and')')
|
|
Packit |
5c3484 |
ifdef(`OPERATION_andn_n',
|
|
Packit |
5c3484 |
` define(`func',`mpn_andn_n')
|
|
Packit |
5c3484 |
define(`logop', `andc')')
|
|
Packit |
5c3484 |
ifdef(`OPERATION_nand_n',
|
|
Packit |
5c3484 |
` define(`func',`mpn_nand_n')
|
|
Packit |
5c3484 |
define(`logop', `nand')')
|
|
Packit |
5c3484 |
ifdef(`OPERATION_ior_n',
|
|
Packit |
5c3484 |
` define(`func',`mpn_ior_n')
|
|
Packit |
5c3484 |
define(`logop', `or')')
|
|
Packit |
5c3484 |
ifdef(`OPERATION_iorn_n',
|
|
Packit |
5c3484 |
` define(`func',`mpn_iorn_n')
|
|
Packit |
5c3484 |
define(`logop', `orc')')
|
|
Packit |
5c3484 |
ifdef(`OPERATION_nior_n',
|
|
Packit |
5c3484 |
` define(`func',`mpn_nior_n')
|
|
Packit |
5c3484 |
define(`logop', `nor')')
|
|
Packit |
5c3484 |
ifdef(`OPERATION_xor_n',
|
|
Packit |
5c3484 |
` define(`func',`mpn_xor_n')
|
|
Packit |
5c3484 |
define(`logop', `xor')')
|
|
Packit |
5c3484 |
ifdef(`OPERATION_xnor_n',
|
|
Packit |
5c3484 |
` define(`func',`mpn_xnor_n')
|
|
Packit |
5c3484 |
define(`logop', `eqv')')
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
C INPUT PARAMETERS
|
|
Packit |
5c3484 |
C rp r3
|
|
Packit |
5c3484 |
C up r4
|
|
Packit |
5c3484 |
C vp r5
|
|
Packit |
5c3484 |
C n r6
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
MULFUNC_PROLOGUE(mpn_and_n mpn_andn_n mpn_nand_n mpn_ior_n mpn_iorn_n mpn_nior_n mpn_xor_n mpn_xnor_n)
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
ASM_START()
|
|
Packit |
5c3484 |
PROLOGUE(func)
|
|
Packit |
5c3484 |
ld r8, 0(r4) C read lowest u limb
|
|
Packit |
5c3484 |
ld r9, 0(r5) C read lowest v limb
|
|
Packit |
5c3484 |
addi r6, r6, 3 C compute branch count (1)
|
|
Packit |
5c3484 |
rldic. r0, r6, 3, 59 C r0 = (n-1 & 3) << 3; cr0 = (n == 4(t+1))?
|
|
Packit |
5c3484 |
cmpldi cr6, r0, 16 C cr6 = (n cmp 4t + 3)
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
ifdef(`HAVE_ABI_mode32',
|
|
Packit |
5c3484 |
` rldicl r6, r6, 62,34', C ...branch count
|
|
Packit |
5c3484 |
` rldicl r6, r6, 62, 2') C ...branch count
|
|
Packit |
5c3484 |
mtctr r6
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
ld r6, 0(r4) C read lowest u limb (again)
|
|
Packit |
5c3484 |
ld r7, 0(r5) C read lowest v limb (again)
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
add r5, r5, r0 C offset vp
|
|
Packit |
5c3484 |
add r4, r4, r0 C offset up
|
|
Packit |
5c3484 |
add r3, r3, r0 C offset rp
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
beq cr0, L(L01)
|
|
Packit |
5c3484 |
blt cr6, L(L10)
|
|
Packit |
5c3484 |
beq cr6, L(L11)
|
|
Packit |
5c3484 |
b L(L00)
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
L(oop): ld r8, -24(r4)
|
|
Packit |
5c3484 |
ld r9, -24(r5)
|
|
Packit |
5c3484 |
logop r10, r6, r7
|
|
Packit |
5c3484 |
std r10, -32(r3)
|
|
Packit |
5c3484 |
L(L00): ld r6, -16(r4)
|
|
Packit |
5c3484 |
ld r7, -16(r5)
|
|
Packit |
5c3484 |
logop r10, r8, r9
|
|
Packit |
5c3484 |
std r10, -24(r3)
|
|
Packit |
5c3484 |
L(L11): ld r8, -8(r4)
|
|
Packit |
5c3484 |
ld r9, -8(r5)
|
|
Packit |
5c3484 |
logop r10, r6, r7
|
|
Packit |
5c3484 |
std r10, -16(r3)
|
|
Packit |
5c3484 |
L(L10): ld r6, 0(r4)
|
|
Packit |
5c3484 |
ld r7, 0(r5)
|
|
Packit |
5c3484 |
logop r10, r8, r9
|
|
Packit |
5c3484 |
std r10, -8(r3)
|
|
Packit |
5c3484 |
L(L01): addi r5, r5, 32
|
|
Packit |
5c3484 |
addi r4, r4, 32
|
|
Packit |
5c3484 |
addi r3, r3, 32
|
|
Packit |
5c3484 |
bdnz L(oop)
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
logop r10, r6, r7
|
|
Packit |
5c3484 |
std r10, -32(r3)
|
|
Packit |
5c3484 |
blr
|
|
Packit |
5c3484 |
EPILOGUE()
|