|
Packit |
5c3484 |
dnl AMD64 mpn_hamdist -- hamming distance.
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
dnl Copyright 2008, 2010-2012 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 AMD K8,K9 n/a
|
|
Packit |
5c3484 |
C AMD K10 2
|
|
Packit |
5c3484 |
C Intel P4 n/a
|
|
Packit |
5c3484 |
C Intel core2 n/a
|
|
Packit |
5c3484 |
C Intel corei 2.05
|
|
Packit |
5c3484 |
C Intel atom n/a
|
|
Packit |
5c3484 |
C VIA nano n/a
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
C This is very straightforward 2-way unrolled code.
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
C TODO
|
|
Packit |
5c3484 |
C * Write something less basic. It should not be hard to reach 1.5 c/l with
|
|
Packit |
5c3484 |
C 4-way unrolling.
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
define(`ap', `%rdi')
|
|
Packit |
5c3484 |
define(`bp', `%rsi')
|
|
Packit |
5c3484 |
define(`n', `%rdx')
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
ABI_SUPPORT(DOS64)
|
|
Packit |
5c3484 |
ABI_SUPPORT(STD64)
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
ASM_START()
|
|
Packit |
5c3484 |
TEXT
|
|
Packit |
5c3484 |
ALIGN(32)
|
|
Packit |
5c3484 |
PROLOGUE(mpn_hamdist)
|
|
Packit |
5c3484 |
FUNC_ENTRY(3)
|
|
Packit |
5c3484 |
mov (ap), %r8
|
|
Packit |
5c3484 |
xor (bp), %r8
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
lea (ap,n,8), ap C point at A operand end
|
|
Packit |
5c3484 |
lea (bp,n,8), bp C point at B operand end
|
|
Packit |
5c3484 |
neg n
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
bt $0, R32(n)
|
|
Packit |
5c3484 |
jnc L(2)
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
L(1): .byte 0xf3,0x49,0x0f,0xb8,0xc0 C popcnt %r8, %rax
|
|
Packit |
5c3484 |
xor R32(%r10), R32(%r10)
|
|
Packit |
5c3484 |
add $1, n
|
|
Packit |
5c3484 |
js L(top)
|
|
Packit |
5c3484 |
FUNC_EXIT()
|
|
Packit |
5c3484 |
ret
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
ALIGN(16)
|
|
Packit |
5c3484 |
L(2): mov 8(ap,n,8), %r9
|
|
Packit |
5c3484 |
.byte 0xf3,0x49,0x0f,0xb8,0xc0 C popcnt %r8, %rax
|
|
Packit |
5c3484 |
xor 8(bp,n,8), %r9
|
|
Packit |
5c3484 |
.byte 0xf3,0x4d,0x0f,0xb8,0xd1 C popcnt %r9, %r10
|
|
Packit |
5c3484 |
add $2, n
|
|
Packit |
5c3484 |
js L(top)
|
|
Packit |
5c3484 |
lea (%r10, %rax), %rax
|
|
Packit |
5c3484 |
FUNC_EXIT()
|
|
Packit |
5c3484 |
ret
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
ALIGN(16)
|
|
Packit |
5c3484 |
L(top): mov (ap,n,8), %r8
|
|
Packit |
5c3484 |
lea (%r10, %rax), %rax
|
|
Packit |
5c3484 |
mov 8(ap,n,8), %r9
|
|
Packit |
5c3484 |
xor (bp,n,8), %r8
|
|
Packit |
5c3484 |
xor 8(bp,n,8), %r9
|
|
Packit |
5c3484 |
.byte 0xf3,0x49,0x0f,0xb8,0xc8 C popcnt %r8, %rcx
|
|
Packit |
5c3484 |
lea (%rcx, %rax), %rax
|
|
Packit |
5c3484 |
.byte 0xf3,0x4d,0x0f,0xb8,0xd1 C popcnt %r9, %r10
|
|
Packit |
5c3484 |
add $2, n
|
|
Packit |
5c3484 |
js L(top)
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
lea (%r10, %rax), %rax
|
|
Packit |
5c3484 |
FUNC_EXIT()
|
|
Packit |
5c3484 |
ret
|
|
Packit |
5c3484 |
EPILOGUE()
|
|
rpm-build |
01f633 |
CF_PROT
|