Blame mpn/s390_32/logops_n.asm

Packit 5c3484
dnl  S/390-32 logops.
Packit 5c3484
Packit 5c3484
dnl  Copyright 2011 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     variant 1           variant 2       variant 3
Packit 5c3484
C	        rp!=up  rp=up
Packit 5c3484
C z900		 ?	 ?		 ?		 ?
Packit 5c3484
C z990		 2.5	 1		 2.75		 2.75
Packit 5c3484
C z9		 ?			 ?		 ?
Packit 5c3484
C z10		 ?			 ?		 ?
Packit 5c3484
C z196		 ?			 ?		 ?
Packit 5c3484
Packit 5c3484
C INPUT PARAMETERS
Packit 5c3484
define(`rp',	`%r2')
Packit 5c3484
define(`up',	`%r3')
Packit 5c3484
define(`vp',	`%r4')
Packit 5c3484
define(`nn',	`%r5')
Packit 5c3484
Packit 5c3484
ifdef(`OPERATION_and_n',`
Packit 5c3484
  define(`func',`mpn_and_n')
Packit 5c3484
  define(`VARIANT_1')
Packit 5c3484
  define(`LOGOPC',`nc')
Packit 5c3484
  define(`LOGOP',`n')')
Packit 5c3484
ifdef(`OPERATION_andn_n',`
Packit 5c3484
  define(`func',`mpn_andn_n')
Packit 5c3484
  define(`VARIANT_2')
Packit 5c3484
  define(`LOGOP',`n')')
Packit 5c3484
ifdef(`OPERATION_nand_n',`
Packit 5c3484
  define(`func',`mpn_nand_n')
Packit 5c3484
  define(`VARIANT_3')
Packit 5c3484
  define(`LOGOP',`n')')
Packit 5c3484
ifdef(`OPERATION_ior_n',`
Packit 5c3484
  define(`func',`mpn_ior_n')
Packit 5c3484
  define(`VARIANT_1')
Packit 5c3484
  define(`LOGOPC',`oc')
Packit 5c3484
  define(`LOGOP',`o')')
Packit 5c3484
ifdef(`OPERATION_iorn_n',`
Packit 5c3484
  define(`func',`mpn_iorn_n')
Packit 5c3484
  define(`VARIANT_2')
Packit 5c3484
  define(`LOGOP',`o')')
Packit 5c3484
ifdef(`OPERATION_nior_n',`
Packit 5c3484
  define(`func',`mpn_nior_n')
Packit 5c3484
  define(`VARIANT_3')
Packit 5c3484
  define(`LOGOP',`o')')
Packit 5c3484
ifdef(`OPERATION_xor_n',`
Packit 5c3484
  define(`func',`mpn_xor_n')
Packit 5c3484
  define(`VARIANT_1')
Packit 5c3484
  define(`LOGOPC',`xc')
Packit 5c3484
  define(`LOGOP',`x')')
Packit 5c3484
ifdef(`OPERATION_xnor_n',`
Packit 5c3484
  define(`func',`mpn_xnor_n')
Packit 5c3484
  define(`VARIANT_2')
Packit 5c3484
  define(`LOGOP',`x')')
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
ifdef(`VARIANT_1',`
Packit 5c3484
	cr	rp, up
Packit 5c3484
	jne	L(normal)
Packit 5c3484
Packit 5c3484
	sll	nn, 2
Packit 5c3484
	ahi	nn, -1
Packit 5c3484
	lr	%r1, nn
Packit 5c3484
	srl	%r1, 8
Packit 5c3484
	ltr	%r1, %r1		C < 256 bytes to copy?
Packit 5c3484
	je	L(1)
Packit 5c3484
Packit 5c3484
L(tp):	LOGOPC	0(256, rp), 0(vp)
Packit 5c3484
	la	rp, 256(rp)
Packit 5c3484
	la	vp, 256(vp)
Packit 5c3484
	brct	%r1, L(tp)
Packit 5c3484
Packit 5c3484
L(1):	bras	%r1, L(2)		C make r1 point to mvc insn
Packit 5c3484
	LOGOPC	0(1, rp), 0(vp)
Packit 5c3484
L(2):	ex	nn, 0(%r1)		C execute mvc with length ((nn-1) mod 256)+1
Packit 5c3484
L(rtn):	br	%r14
Packit 5c3484
Packit 5c3484
Packit 5c3484
L(normal):
Packit 5c3484
	stm	%r6, %r8, 12(%r15)
Packit 5c3484
	ahi	nn, 3
Packit 5c3484
	lhi	%r7, 3
Packit 5c3484
	lr	%r0, nn
Packit 5c3484
	srl	%r0, 2
Packit 5c3484
	nr	%r7, nn			C nn mod 4
Packit 5c3484
	je	L(b1)
Packit 5c3484
	chi	%r7, 2
Packit 5c3484
	jl	L(b2)
Packit 5c3484
	jne	L(top)
Packit 5c3484
Packit 5c3484
L(b3):	lm	%r5, %r7, 0(up)
Packit 5c3484
	la	up, 12(up)
Packit 5c3484
	LOGOP	%r5, 0(vp)
Packit 5c3484
	LOGOP	%r6, 4(vp)
Packit 5c3484
	LOGOP	%r7, 8(vp)
Packit 5c3484
	stm	%r5, %r7, 0(rp)
Packit 5c3484
	la	rp, 12(rp)
Packit 5c3484
	la	vp, 12(vp)
Packit 5c3484
	j	L(mid)
Packit 5c3484
Packit 5c3484
L(b1):	l	%r5, 0(up)
Packit 5c3484
	la	up, 4(up)
Packit 5c3484
	LOGOP	%r5, 0(vp)
Packit 5c3484
	st	%r5, 0(rp)
Packit 5c3484
	la	rp, 4(rp)
Packit 5c3484
	la	vp, 4(vp)
Packit 5c3484
	j	L(mid)
Packit 5c3484
Packit 5c3484
L(b2):	lm	%r5, %r6, 0(up)
Packit 5c3484
	la	up, 8(up)
Packit 5c3484
	LOGOP	%r5, 0(vp)
Packit 5c3484
	LOGOP	%r6, 4(vp)
Packit 5c3484
	stm	%r5, %r6, 0(rp)
Packit 5c3484
	la	rp, 8(rp)
Packit 5c3484
	la	vp, 8(vp)
Packit 5c3484
	j	L(mid)
Packit 5c3484
Packit 5c3484
L(top):	lm	%r5, %r8, 0(up)
Packit 5c3484
	la	up, 16(up)
Packit 5c3484
	LOGOP	%r5, 0(vp)
Packit 5c3484
	LOGOP	%r6, 4(vp)
Packit 5c3484
	LOGOP	%r7, 8(vp)
Packit 5c3484
	LOGOP	%r8, 12(vp)
Packit 5c3484
	stm	%r5, %r8, 0(rp)
Packit 5c3484
	la	rp, 16(rp)
Packit 5c3484
	la	vp, 16(vp)
Packit 5c3484
L(mid):	brct	%r0, L(top)
Packit 5c3484
Packit 5c3484
	lm	%r6, %r8, 12(%r15)
Packit 5c3484
	br	%r14
Packit 5c3484
')
Packit 5c3484
Packit 5c3484
ifdef(`VARIANT_2',`
Packit 5c3484
	stm	%r6, %r8, 12(%r15)
Packit 5c3484
	lhi	%r1, -1
Packit 5c3484
Packit 5c3484
	ahi	nn, 3
Packit 5c3484
	lhi	%r7, 3
Packit 5c3484
	lr	%r0, nn
Packit 5c3484
	srl	%r0, 2
Packit 5c3484
	nr	%r7, nn			C nn mod 4
Packit 5c3484
	je	L(b1)
Packit 5c3484
	chi	%r7, 2
Packit 5c3484
	jl	L(b2)
Packit 5c3484
	jne	L(top)
Packit 5c3484
Packit 5c3484
L(b3):	lm	%r5, %r7, 0(vp)
Packit 5c3484
	la	vp, 12(vp)
Packit 5c3484
	xr	%r5, %r1
Packit 5c3484
	xr	%r6, %r1
Packit 5c3484
	xr	%r7, %r1
Packit 5c3484
	LOGOP	%r5, 0(up)
Packit 5c3484
	LOGOP	%r6, 4(up)
Packit 5c3484
	LOGOP	%r7, 8(up)
Packit 5c3484
	stm	%r5, %r7, 0(rp)
Packit 5c3484
	la	rp, 12(rp)
Packit 5c3484
	la	up, 12(up)
Packit 5c3484
	j	L(mid)
Packit 5c3484
Packit 5c3484
L(b1):	l	%r5, 0(vp)
Packit 5c3484
	la	vp, 4(vp)
Packit 5c3484
	xr	%r5, %r1
Packit 5c3484
	LOGOP	%r5, 0(up)
Packit 5c3484
	st	%r5, 0(rp)
Packit 5c3484
	la	rp, 4(rp)
Packit 5c3484
	la	up, 4(up)
Packit 5c3484
	j	L(mid)
Packit 5c3484
Packit 5c3484
L(b2):	lm	%r5, %r6, 0(vp)
Packit 5c3484
	la	vp, 8(vp)
Packit 5c3484
	xr	%r5, %r1
Packit 5c3484
	xr	%r6, %r1
Packit 5c3484
	LOGOP	%r5, 0(up)
Packit 5c3484
	LOGOP	%r6, 4(up)
Packit 5c3484
	stm	%r5, %r6, 0(rp)
Packit 5c3484
	la	rp, 8(rp)
Packit 5c3484
	la	up, 8(up)
Packit 5c3484
	j	L(mid)
Packit 5c3484
Packit 5c3484
L(top):	lm	%r5, %r8, 0(vp)
Packit 5c3484
	la	vp, 16(vp)
Packit 5c3484
	xr	%r5, %r1
Packit 5c3484
	xr	%r6, %r1
Packit 5c3484
	xr	%r7, %r1
Packit 5c3484
	xr	%r8, %r1
Packit 5c3484
	LOGOP	%r5, 0(up)
Packit 5c3484
	LOGOP	%r6, 4(up)
Packit 5c3484
	LOGOP	%r7, 8(up)
Packit 5c3484
	LOGOP	%r8, 12(up)
Packit 5c3484
	la	up, 16(up)
Packit 5c3484
	stm	%r5, %r8, 0(rp)
Packit 5c3484
	la	rp, 16(rp)
Packit 5c3484
L(mid):	brct	%r0, L(top)
Packit 5c3484
Packit 5c3484
	lm	%r6, %r8, 12(%r15)
Packit 5c3484
	br	%r14
Packit 5c3484
')
Packit 5c3484
Packit 5c3484
ifdef(`VARIANT_3',`
Packit 5c3484
	stm	%r6, %r8, 12(%r15)
Packit 5c3484
	lhi	%r1, -1
Packit 5c3484
Packit 5c3484
	ahi	nn, 3
Packit 5c3484
	lhi	%r7, 3
Packit 5c3484
	lr	%r0, nn
Packit 5c3484
	srl	%r0, 2
Packit 5c3484
	nr	%r7, nn			C nn mod 4
Packit 5c3484
	je	L(b1)
Packit 5c3484
	chi	%r7, 2
Packit 5c3484
	jl	L(b2)
Packit 5c3484
	jne	L(top)
Packit 5c3484
Packit 5c3484
L(b3):	lm	%r5, %r7, 0(vp)
Packit 5c3484
	la	vp, 12(vp)
Packit 5c3484
	LOGOP	%r5, 0(up)
Packit 5c3484
	LOGOP	%r6, 4(up)
Packit 5c3484
	xr	%r5, %r1
Packit 5c3484
	xr	%r6, %r1
Packit 5c3484
	LOGOP	%r7, 8(up)
Packit 5c3484
	xr	%r7, %r1
Packit 5c3484
	stm	%r5, %r7, 0(rp)
Packit 5c3484
	la	rp, 12(rp)
Packit 5c3484
	la	up, 12(up)
Packit 5c3484
	j	L(mid)
Packit 5c3484
Packit 5c3484
L(b1):	l	%r5, 0(vp)
Packit 5c3484
	la	vp, 4(vp)
Packit 5c3484
	LOGOP	%r5, 0(up)
Packit 5c3484
	xr	%r5, %r1
Packit 5c3484
	st	%r5, 0(rp)
Packit 5c3484
	la	rp, 4(rp)
Packit 5c3484
	la	up, 4(up)
Packit 5c3484
	j	L(mid)
Packit 5c3484
Packit 5c3484
L(b2):	lm	%r5, %r6, 0(vp)
Packit 5c3484
	la	vp, 8(vp)
Packit 5c3484
	LOGOP	%r5, 0(up)
Packit 5c3484
	LOGOP	%r6, 4(up)
Packit 5c3484
	xr	%r5, %r1
Packit 5c3484
	xr	%r6, %r1
Packit 5c3484
	stm	%r5, %r6, 0(rp)
Packit 5c3484
	la	rp, 8(rp)
Packit 5c3484
	la	up, 8(up)
Packit 5c3484
	j	L(mid)
Packit 5c3484
Packit 5c3484
L(top):	lm	%r5, %r8, 0(vp)
Packit 5c3484
	la	vp, 16(vp)
Packit 5c3484
	LOGOP	%r5, 0(up)
Packit 5c3484
	LOGOP	%r6, 4(up)
Packit 5c3484
	xr	%r5, %r1
Packit 5c3484
	xr	%r6, %r1
Packit 5c3484
	LOGOP	%r7, 8(up)
Packit 5c3484
	LOGOP	%r8, 12(up)
Packit 5c3484
	xr	%r7, %r1
Packit 5c3484
	xr	%r8, %r1
Packit 5c3484
	stm	%r5, %r8, 0(rp)
Packit 5c3484
	la	up, 16(up)
Packit 5c3484
	la	rp, 16(rp)
Packit 5c3484
L(mid):	brct	%r0, L(top)
Packit 5c3484
Packit 5c3484
	lm	%r6, %r8, 12(%r15)
Packit 5c3484
	br	%r14
Packit 5c3484
')
Packit 5c3484
Packit 5c3484
EPILOGUE()