Blame mpn/arm/logops_n.asm

Packit 5c3484
dnl  ARM mpn_and_n, mpn_andn_n. mpn_nand_n, etc.
Packit 5c3484
Packit 5c3484
dnl  Contributed to the GNU project by Torbjörn Granlund.
Packit 5c3484
Packit 5c3484
dnl  Copyright 1997, 2000, 2001, 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             cycles/limb
Packit 5c3484
C          and andn ior xor         nand iorn nior xnor
Packit 5c3484
C StrongARM	 ?			 ?
Packit 5c3484
C XScale	 ?			 ?
Packit 5c3484
C Cortex-A7	 ?			 ?
Packit 5c3484
C Cortex-A8	 ?			 ?
Packit 5c3484
C Cortex-A9	2.5-2.72		2.75-3
Packit 5c3484
C Cortex-A15	2.25			2.75
Packit 5c3484
Packit 5c3484
C TODO
Packit 5c3484
C  * It seems that 2.25 c/l and 2.75 c/l is possible for A9.
Packit 5c3484
C  * Debug popping issue, see comment below.
Packit 5c3484
Packit 5c3484
define(`rp', `r0')
Packit 5c3484
define(`up', `r1')
Packit 5c3484
define(`vp', `r2')
Packit 5c3484
define(`n',  `r3')
Packit 5c3484
Packit 5c3484
define(`POSTOP')
Packit 5c3484
Packit 5c3484
ifdef(`OPERATION_and_n',`
Packit 5c3484
  define(`func',    `mpn_and_n')
Packit 5c3484
  define(`LOGOP',   `and	$1, $2, $3')')
Packit 5c3484
ifdef(`OPERATION_andn_n',`
Packit 5c3484
  define(`func',    `mpn_andn_n')
Packit 5c3484
  define(`LOGOP',   `bic	$1, $2, $3')')
Packit 5c3484
ifdef(`OPERATION_nand_n',`
Packit 5c3484
  define(`func',    `mpn_nand_n')
Packit 5c3484
  define(`POSTOP',  `mvn	$1, $1')
Packit 5c3484
  define(`LOGOP',   `and	$1, $2, $3')')
Packit 5c3484
ifdef(`OPERATION_ior_n',`
Packit 5c3484
  define(`func',    `mpn_ior_n')
Packit 5c3484
  define(`LOGOP',   `orr	$1, $2, $3')')
Packit 5c3484
ifdef(`OPERATION_iorn_n',`
Packit 5c3484
  define(`func',    `mpn_iorn_n')
Packit 5c3484
  define(`POSTOP',  `mvn	$1, $1')
Packit 5c3484
  define(`LOGOP',   `bic	$1, $3, $2')')
Packit 5c3484
ifdef(`OPERATION_nior_n',`
Packit 5c3484
  define(`func',    `mpn_nior_n')
Packit 5c3484
  define(`POSTOP',  `mvn	$1, $1')
Packit 5c3484
  define(`LOGOP',   `orr	$1, $2, $3')')
Packit 5c3484
ifdef(`OPERATION_xor_n',`
Packit 5c3484
  define(`func',    `mpn_xor_n')
Packit 5c3484
  define(`LOGOP',   `eor	$1, $2, $3')')
Packit 5c3484
ifdef(`OPERATION_xnor_n',`
Packit 5c3484
  define(`func',    `mpn_xnor_n')
Packit 5c3484
  define(`POSTOP',  `mvn	$1, $1')
Packit 5c3484
  define(`LOGOP',   `eor	$1, $2, $3')')
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
	push	{ r8, r9, r10 }
Packit 5c3484
	tst	n, #1
Packit 5c3484
	beq	L(skip1)
Packit 5c3484
	ldr	r10, [vp], #4
Packit 5c3484
	ldr	r12, [up], #4
Packit 5c3484
	LOGOP(	r12, r12, r10)
Packit 5c3484
	POSTOP(	r12)
Packit 5c3484
	str	r12, [rp], #4
Packit 5c3484
L(skip1):
Packit 5c3484
	tst	n, #2
Packit 5c3484
	beq	L(skip2)
Packit 5c3484
	ldmia	vp!, { r10, r12 }
Packit 5c3484
	ldmia	up!, { r8, r9 }
Packit 5c3484
	LOGOP(	r8, r8, r10)
Packit 5c3484
	LOGOP(	r9, r9, r12)
Packit 5c3484
	POSTOP(	r8)
Packit 5c3484
	POSTOP(	r9)
Packit 5c3484
	stmia	rp!, { r8, r9 }
Packit 5c3484
L(skip2):
Packit 5c3484
	bics	n, n, #3
Packit 5c3484
	beq	L(rtn)
Packit 5c3484
	push	{ r4, r5, r6, r7 }
Packit 5c3484
Packit 5c3484
	ldmia	vp!, { r8, r9, r10, r12 }
Packit 5c3484
	b	L(mid)
Packit 5c3484
Packit 5c3484
L(top):	ldmia	vp!, { r8, r9, r10, r12 }
Packit 5c3484
	POSTOP(	r4)
Packit 5c3484
	POSTOP(	r5)
Packit 5c3484
	POSTOP(	r6)
Packit 5c3484
	POSTOP(	r7)
Packit 5c3484
	stmia	rp!, { r4, r5, r6, r7 }
Packit 5c3484
L(mid):	sub	n, n, #4
Packit 5c3484
	ldmia	up!, { r4, r5, r6, r7 }
Packit 5c3484
	teq	n, #0
Packit 5c3484
	LOGOP(	r4, r4, r8)
Packit 5c3484
	LOGOP(	r5, r5, r9)
Packit 5c3484
	LOGOP(	r6, r6, r10)
Packit 5c3484
	LOGOP(	r7, r7, r12)
Packit 5c3484
	bne	L(top)
Packit 5c3484
Packit 5c3484
	POSTOP(	r4)
Packit 5c3484
	POSTOP(	r5)
Packit 5c3484
	POSTOP(	r6)
Packit 5c3484
	POSTOP(	r7)
Packit 5c3484
	stmia	rp!, { r4, r5, r6, r7 }
Packit 5c3484
Packit 5c3484
	pop	{ r4, r5, r6, r7 }	C popping r8-r10 here strangely fails
Packit 5c3484
Packit 5c3484
L(rtn):	pop	{ r8, r9, r10 }
Packit 5c3484
	ret	r14
Packit 5c3484
EPILOGUE()