Blame mpn/powerpc64/mode64/bdiv_dbm1c.asm

Packit 5c3484
dnl  PPC64 mpn_bdiv_dbm1c.
Packit 5c3484
Packit 5c3484
dnl  Copyright 2008, 2010 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       6-18
Packit 5c3484
C POWER4/PPC970       8.25
Packit 5c3484
C POWER5              8.5  fluctuating as function of n % 3
Packit 5c3484
C POWER6             15
Packit 5c3484
C POWER7              4.75
Packit 5c3484
Packit 5c3484
C TODO
Packit 5c3484
C  * Nothing to do...
Packit 5c3484
Packit 5c3484
C INPUT PARAMETERS
Packit 5c3484
define(`rp', `r3')
Packit 5c3484
define(`up', `r4')
Packit 5c3484
define(`n',  `r5')
Packit 5c3484
define(`bd', `r6')
Packit 5c3484
define(`cy', `r7')
Packit 5c3484
Packit 5c3484
ASM_START()
Packit 5c3484
PROLOGUE(mpn_bdiv_dbm1c)
Packit 5c3484
	ld	r0, 0(r4)
Packit 5c3484
Packit 5c3484
	rldicl.	r12, r5, 0,62
Packit 5c3484
	cmpldi	cr6, r12, 2
Packit 5c3484
	cmpldi	cr7, r5, 4
Packit 5c3484
	addi	r5, r5, 1
Packit 5c3484
	srwi	r5, r5, 2
Packit 5c3484
	mtctr	r5
Packit 5c3484
	beq	cr0, L(b00)
Packit 5c3484
	blt	cr6, L(b01)
Packit 5c3484
	beq	cr6, L(b10)
Packit 5c3484
Packit 5c3484
	ALIGN(16)
Packit 5c3484
L(b11):	mulld	r5, r0, r6
Packit 5c3484
	mulhdu	r12, r0, r6
Packit 5c3484
	ld	r0, 8(r4)
Packit 5c3484
	addi	r4, r4, -24
Packit 5c3484
	addi	r3, r3, -24
Packit 5c3484
	b	L(3)
Packit 5c3484
Packit 5c3484
	ALIGN(16)
Packit 5c3484
L(b00):	mulld	r9, r0, r6
Packit 5c3484
	mulhdu	r8, r0, r6
Packit 5c3484
	addi	r4, r4, -16
Packit 5c3484
	addi	r3, r3, -16
Packit 5c3484
	b	L(0)
Packit 5c3484
Packit 5c3484
	ALIGN(16)
Packit 5c3484
L(b01):	mulld	r5, r0, r6
Packit 5c3484
	mulhdu	r12, r0, r6
Packit 5c3484
	addi	r3, r3, -8
Packit 5c3484
	ble	cr7, L(e1)
Packit 5c3484
	ld	r0, 8(r4)
Packit 5c3484
	addi	r4, r4, -8
Packit 5c3484
	b	L(1)
Packit 5c3484
Packit 5c3484
	ALIGN(16)
Packit 5c3484
L(b10):	mulld	r9, r0, r6
Packit 5c3484
	mulhdu	r8, r0, r6
Packit 5c3484
	ble	cr7, L(e2)
Packit 5c3484
Packit 5c3484
	ALIGN(16)
Packit 5c3484
L(top):	subfc	r11, r9, r7
Packit 5c3484
	ld	r10, 8(r4)
Packit 5c3484
	ld	r0, 16(r4)
Packit 5c3484
	subfe	r7, r8, r11
Packit 5c3484
	std	r11, 0(r3)
Packit 5c3484
	mulld	r5, r10, r6
Packit 5c3484
	mulhdu	r12, r10, r6
Packit 5c3484
L(1):	mulld	r9, r0, r6
Packit 5c3484
	mulhdu	r8, r0, r6
Packit 5c3484
	subfc	r11, r5, r7
Packit 5c3484
	subfe	r7, r12, r11
Packit 5c3484
	std	r11, 8(r3)
Packit 5c3484
L(0):	subfc	r11, r9, r7
Packit 5c3484
	ld	r10, 24(r4)
Packit 5c3484
	ld	r0, 32(r4)
Packit 5c3484
	subfe	r7, r8, r11
Packit 5c3484
	std	r11, 16(r3)
Packit 5c3484
	mulld	r5, r10, r6
Packit 5c3484
	mulhdu	r12, r10, r6
Packit 5c3484
L(3):	mulld	r9, r0, r6
Packit 5c3484
	mulhdu	r8, r0, r6
Packit 5c3484
	subfc	r11, r5, r7
Packit 5c3484
	subfe	r7, r12, r11
Packit 5c3484
	std	r11, 24(r3)
Packit 5c3484
	addi	r4, r4, 32
Packit 5c3484
	addi	r3, r3, 32
Packit 5c3484
	bdnz	L(top)
Packit 5c3484
Packit 5c3484
L(e2):	ld	r10, 8(r4)
Packit 5c3484
	mulld	r5, r10, r6
Packit 5c3484
	mulhdu	r12, r10, r6
Packit 5c3484
	subfc	r11, r9, r7
Packit 5c3484
	subfe	r7, r8, r11
Packit 5c3484
	std	r11, 0(r3)
Packit 5c3484
L(e1):	subfc	r11, r5, r7
Packit 5c3484
	std	r11, 8(r3)
Packit 5c3484
	subfe	r3, r12, r11
Packit 5c3484
	blr
Packit 5c3484
EPILOGUE()