Blame mpn/powerpc64/p7/copyd.asm

Packit 5c3484
dnl  PowerPC-64 mpn_copyd.
Packit 5c3484
Packit 5c3484
dnl  Copyright 2004, 2005, 2013 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          ?
Packit 5c3484
C POWER4/PPC970          ?
Packit 5c3484
C POWER5                 ?
Packit 5c3484
C POWER6                 1.25
Packit 5c3484
C POWER7                 1.09
Packit 5c3484
Packit 5c3484
C INPUT PARAMETERS
Packit 5c3484
define(`rp',	`r3')
Packit 5c3484
define(`up',	`r4')
Packit 5c3484
define(`n',	`r5')
Packit 5c3484
Packit 5c3484
ASM_START()
Packit 5c3484
PROLOGUE(mpn_copyd)
Packit 5c3484
Packit 5c3484
ifdef(`HAVE_ABI_mode32',
Packit 5c3484
`	rldicl	n, n, 0,32')
Packit 5c3484
Packit 5c3484
	sldi	r0, n, 3
Packit 5c3484
	add	up, up, r0		C point at u[] end
Packit 5c3484
	add	rp, rp, r0		C point at r[] end
Packit 5c3484
Packit 5c3484
	cmpdi	cr0, n, 4
Packit 5c3484
	blt	L(sml)
Packit 5c3484
Packit 5c3484
	addi	r10, n, 4
Packit 5c3484
	srdi	r10, r10, 3
Packit 5c3484
	mtctr	r10
Packit 5c3484
Packit 5c3484
	andi.	r0, n, 1
Packit 5c3484
	rlwinm	r11, n, 0,30,30
Packit 5c3484
	rlwinm	r12, n, 0,29,29
Packit 5c3484
	cmpdi	cr6, r11, 0
Packit 5c3484
	cmpdi	cr7, r12, 0
Packit 5c3484
Packit 5c3484
	beq	cr0, L(xx0)
Packit 5c3484
L(xx1):	ld	r6, -8(up)
Packit 5c3484
	addi	up, up, -8
Packit 5c3484
	std	r6, -8(rp)
Packit 5c3484
	addi	rp, rp, -8
Packit 5c3484
Packit 5c3484
L(xx0):	bne	cr6, L(x10)
Packit 5c3484
L(x00):	ld	r6, -8(up)
Packit 5c3484
	ld	r7, -16(up)
Packit 5c3484
	bne	cr7, L(100)
Packit 5c3484
L(000):	addi	rp, rp, 32
Packit 5c3484
	b	L(lo0)
Packit 5c3484
L(100):	addi	up, up, 32
Packit 5c3484
	b	L(lo4)
Packit 5c3484
L(x10):	ld	r8, -8(up)
Packit 5c3484
	ld	r9, -16(up)
Packit 5c3484
	bne	cr7, L(110)
Packit 5c3484
L(010):	addi	up, up, -16
Packit 5c3484
	addi	rp, rp, 16
Packit 5c3484
	b	L(lo2)
Packit 5c3484
L(110):	addi	up, up, 16
Packit 5c3484
	addi	rp, rp, 48
Packit 5c3484
	b	L(lo6)
Packit 5c3484
Packit 5c3484
L(sml):	cmpdi	cr0, n, 0
Packit 5c3484
	beqlr-	cr0
Packit 5c3484
	mtctr	n
Packit 5c3484
L(t):	ld	r6, -8(up)
Packit 5c3484
	addi	up, up, -8
Packit 5c3484
	std	r6, -8(rp)
Packit 5c3484
	addi	rp, rp, -8
Packit 5c3484
	bdnz	L(t)
Packit 5c3484
	blr
Packit 5c3484
Packit 5c3484
	ALIGN(32)
Packit 5c3484
L(top):	std	r6, -8(rp)
Packit 5c3484
	std	r7, -16(rp)
Packit 5c3484
L(lo2):	ld	r6, -8(up)
Packit 5c3484
	ld	r7, -16(up)
Packit 5c3484
	std	r8, -24(rp)
Packit 5c3484
	std	r9, -32(rp)
Packit 5c3484
L(lo0):	ld	r8, -24(up)
Packit 5c3484
	ld	r9, -32(up)
Packit 5c3484
	std	r6, -40(rp)
Packit 5c3484
	std	r7, -48(rp)
Packit 5c3484
L(lo6):	ld	r6, -40(up)
Packit 5c3484
	ld	r7, -48(up)
Packit 5c3484
	std	r8, -56(rp)
Packit 5c3484
	std	r9, -64(rp)
Packit 5c3484
	addi	rp, rp, -64
Packit 5c3484
L(lo4):	ld	r8, -56(up)
Packit 5c3484
	ld	r9, -64(up)
Packit 5c3484
	addi	up, up, -64
Packit 5c3484
	bdnz	L(top)
Packit 5c3484
Packit 5c3484
L(end):	std	r6, -8(rp)
Packit 5c3484
	std	r7, -16(rp)
Packit 5c3484
	std	r8, -24(rp)
Packit 5c3484
	std	r9, -32(rp)
Packit 5c3484
	blr
Packit 5c3484
EPILOGUE()