|
Packit |
5c3484 |
dnl AMD64 mpn_com optimised for CPUs with fast SSE.
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
dnl Copyright 2003, 2005, 2007, 2011, 2012, 2015 Free Software Foundation,
|
|
Packit |
5c3484 |
dnl Inc.
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
dnl Contributed to the GNU project by Torbjorn Granlund.
|
|
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 cycles/limb cycles/limb good
|
|
Packit |
5c3484 |
C aligned unaligned best seen for cpu?
|
|
Packit |
5c3484 |
C AMD K8,K9 2.0 2.0 N
|
|
Packit |
5c3484 |
C AMD K10 0.85 1.3 Y/N
|
|
Packit |
5c3484 |
C AMD bull 1.40 1.40 Y
|
|
Packit |
5c3484 |
C AMD pile 0.9-1.4 0.9-1.4 Y
|
|
Packit |
5c3484 |
C AMD steam
|
|
Packit |
5c3484 |
C AMD excavator
|
|
Packit |
5c3484 |
C AMD bobcat 3.1 3.1 N
|
|
Packit |
5c3484 |
C AMD jaguar 0.91 0.91 opt/opt Y
|
|
Packit |
5c3484 |
C Intel P4 2.28 illop Y
|
|
Packit |
5c3484 |
C Intel core2 1.02 1.02 N
|
|
Packit |
5c3484 |
C Intel NHM 0.53 0.68 Y
|
|
Packit |
5c3484 |
C Intel SBR 0.51 0.75 opt/0.65 Y/N
|
|
Packit |
5c3484 |
C Intel IBR 0.50 0.57 opt/opt Y
|
|
Packit |
5c3484 |
C Intel HWL 0.51 0.64 opt/0.58 Y
|
|
Packit |
5c3484 |
C Intel BWL 0.61 0.65 0.57/opt Y
|
|
Packit |
5c3484 |
C Intel atom 3.68 3.68 N
|
|
Packit |
5c3484 |
C Intel SLM 1.09 1.35 N
|
|
Packit |
5c3484 |
C VIA nano 1.17 5.09 Y/N
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
C We try to do as many 16-byte operations as possible. The top-most and
|
|
Packit |
5c3484 |
C bottom-most writes might need 8-byte operations. We can always write using
|
|
Packit |
5c3484 |
C aligned 16-byte operations, we read with both aligned and unaligned 16-byte
|
|
Packit |
5c3484 |
C operations.
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
C Instead of having separate loops for reading aligned and unaligned, we read
|
|
Packit |
5c3484 |
C using MOVDQU. This seems to work great except for core2; there performance
|
|
Packit |
5c3484 |
C doubles when reading using MOVDQA (for aligned source). It is unclear how to
|
|
Packit |
5c3484 |
C best handle the unaligned case there.
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
C INPUT PARAMETERS
|
|
Packit |
5c3484 |
define(`rp', `%rdi')
|
|
Packit |
5c3484 |
define(`up', `%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(16)
|
|
Packit |
5c3484 |
PROLOGUE(mpn_com)
|
|
Packit |
5c3484 |
FUNC_ENTRY(3)
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
pcmpeqb %xmm7, %xmm7 C set to 111...111
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
test $8, R8(rp) C is rp 16-byte aligned?
|
|
Packit |
5c3484 |
jz L(ali) C jump if rp aligned
|
|
Packit |
5c3484 |
mov (up), %rax
|
|
Packit |
5c3484 |
lea 8(up), up
|
|
Packit |
5c3484 |
not %rax
|
|
Packit |
5c3484 |
mov %rax, (rp)
|
|
Packit |
5c3484 |
lea 8(rp), rp
|
|
Packit |
5c3484 |
dec n
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
sub $14, n
|
|
Packit |
5c3484 |
jc L(sma)
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
ALIGN(16)
|
|
Packit |
5c3484 |
L(top): movdqu (up), %xmm0
|
|
Packit |
5c3484 |
movdqu 16(up), %xmm1
|
|
Packit |
5c3484 |
movdqu 32(up), %xmm2
|
|
Packit |
5c3484 |
movdqu 48(up), %xmm3
|
|
Packit |
5c3484 |
movdqu 64(up), %xmm4
|
|
Packit |
5c3484 |
movdqu 80(up), %xmm5
|
|
Packit |
5c3484 |
movdqu 96(up), %xmm6
|
|
Packit |
5c3484 |
lea 112(up), up
|
|
Packit |
5c3484 |
pxor %xmm7, %xmm0
|
|
Packit |
5c3484 |
pxor %xmm7, %xmm1
|
|
Packit |
5c3484 |
pxor %xmm7, %xmm2
|
|
Packit |
5c3484 |
pxor %xmm7, %xmm3
|
|
Packit |
5c3484 |
pxor %xmm7, %xmm4
|
|
Packit |
5c3484 |
pxor %xmm7, %xmm5
|
|
Packit |
5c3484 |
pxor %xmm7, %xmm6
|
|
Packit |
5c3484 |
movdqa %xmm0, (rp)
|
|
Packit |
5c3484 |
movdqa %xmm1, 16(rp)
|
|
Packit |
5c3484 |
movdqa %xmm2, 32(rp)
|
|
Packit |
5c3484 |
movdqa %xmm3, 48(rp)
|
|
Packit |
5c3484 |
movdqa %xmm4, 64(rp)
|
|
Packit |
5c3484 |
movdqa %xmm5, 80(rp)
|
|
Packit |
5c3484 |
movdqa %xmm6, 96(rp)
|
|
Packit |
5c3484 |
lea 112(rp), rp
|
|
Packit |
5c3484 |
L(ali): sub $14, n
|
|
Packit |
5c3484 |
jnc L(top)
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
L(sma): add $14, n
|
|
Packit |
5c3484 |
test $8, R8(n)
|
|
Packit |
5c3484 |
jz 1f
|
|
Packit |
5c3484 |
movdqu (up), %xmm0
|
|
Packit |
5c3484 |
movdqu 16(up), %xmm1
|
|
Packit |
5c3484 |
movdqu 32(up), %xmm2
|
|
Packit |
5c3484 |
movdqu 48(up), %xmm3
|
|
Packit |
5c3484 |
lea 64(up), up
|
|
Packit |
5c3484 |
pxor %xmm7, %xmm0
|
|
Packit |
5c3484 |
pxor %xmm7, %xmm1
|
|
Packit |
5c3484 |
pxor %xmm7, %xmm2
|
|
Packit |
5c3484 |
pxor %xmm7, %xmm3
|
|
Packit |
5c3484 |
movdqa %xmm0, (rp)
|
|
Packit |
5c3484 |
movdqa %xmm1, 16(rp)
|
|
Packit |
5c3484 |
movdqa %xmm2, 32(rp)
|
|
Packit |
5c3484 |
movdqa %xmm3, 48(rp)
|
|
Packit |
5c3484 |
lea 64(rp), rp
|
|
Packit |
5c3484 |
1:
|
|
Packit |
5c3484 |
test $4, R8(n)
|
|
Packit |
5c3484 |
jz 1f
|
|
Packit |
5c3484 |
movdqu (up), %xmm0
|
|
Packit |
5c3484 |
movdqu 16(up), %xmm1
|
|
Packit |
5c3484 |
lea 32(up), up
|
|
Packit |
5c3484 |
pxor %xmm7, %xmm0
|
|
Packit |
5c3484 |
pxor %xmm7, %xmm1
|
|
Packit |
5c3484 |
movdqa %xmm0, (rp)
|
|
Packit |
5c3484 |
movdqa %xmm1, 16(rp)
|
|
Packit |
5c3484 |
lea 32(rp), rp
|
|
Packit |
5c3484 |
1:
|
|
Packit |
5c3484 |
test $2, R8(n)
|
|
Packit |
5c3484 |
jz 1f
|
|
Packit |
5c3484 |
movdqu (up), %xmm0
|
|
Packit |
5c3484 |
lea 16(up), up
|
|
Packit |
5c3484 |
pxor %xmm7, %xmm0
|
|
Packit |
5c3484 |
movdqa %xmm0, (rp)
|
|
Packit |
5c3484 |
lea 16(rp), rp
|
|
Packit |
5c3484 |
1:
|
|
Packit |
5c3484 |
test $1, R8(n)
|
|
Packit |
5c3484 |
jz 1f
|
|
Packit |
5c3484 |
mov (up), %rax
|
|
Packit |
5c3484 |
not %rax
|
|
Packit |
5c3484 |
mov %rax, (rp)
|
|
Packit |
5c3484 |
1:
|
|
Packit |
5c3484 |
L(don): FUNC_EXIT()
|
|
Packit |
5c3484 |
ret
|
|
Packit |
5c3484 |
EPILOGUE()
|
|
rpm-build |
c3cd4f |
CF_PROT
|