github.com/aergoio/aergo@v1.3.1/libtool/src/gmp-6.1.2/mpn/pa32/hppa1_1/pa7100/add_n.asm (about)

     1  dnl  HP-PA mpn_add_n -- Add two limb vectors of the same length > 0 and store
     2  dnl  sum in a third limb vector.  Optimized for the PA7100, where is runs at
     3  dnl  4.25 cycles/limb.
     4  
     5  dnl  Copyright 1992, 1994, 2000-2003 Free Software Foundation, Inc.
     6  
     7  dnl  This file is part of the GNU MP Library.
     8  dnl
     9  dnl  The GNU MP Library is free software; you can redistribute it and/or modify
    10  dnl  it under the terms of either:
    11  dnl
    12  dnl    * the GNU Lesser General Public License as published by the Free
    13  dnl      Software Foundation; either version 3 of the License, or (at your
    14  dnl      option) any later version.
    15  dnl
    16  dnl  or
    17  dnl
    18  dnl    * the GNU General Public License as published by the Free Software
    19  dnl      Foundation; either version 2 of the License, or (at your option) any
    20  dnl      later version.
    21  dnl
    22  dnl  or both in parallel, as here.
    23  dnl
    24  dnl  The GNU MP Library is distributed in the hope that it will be useful, but
    25  dnl  WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
    26  dnl  or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
    27  dnl  for more details.
    28  dnl
    29  dnl  You should have received copies of the GNU General Public License and the
    30  dnl  GNU Lesser General Public License along with the GNU MP Library.  If not,
    31  dnl  see https://www.gnu.org/licenses/.
    32  
    33  include(`../config.m4')
    34  
    35  C INPUT PARAMETERS
    36  C res_ptr	r26
    37  C s1_ptr	r25
    38  C s2_ptr	r24
    39  C size		r23
    40  
    41  ASM_START()
    42  PROLOGUE(mpn_add_n)
    43  	ldws,ma		4(0,%r25),%r20
    44  	ldws,ma		4(0,%r24),%r19
    45  
    46  	addib,<=	-5,%r23,L(rest)
    47  	 add		%r20,%r19,%r28	C add first limbs ignoring cy
    48  
    49  LDEF(loop)
    50  	ldws,ma		4(0,%r25),%r20
    51  	ldws,ma		4(0,%r24),%r19
    52  	stws,ma		%r28,4(0,%r26)
    53  	addc		%r20,%r19,%r28
    54  	ldws,ma		4(0,%r25),%r20
    55  	ldws,ma		4(0,%r24),%r19
    56  	stws,ma		%r28,4(0,%r26)
    57  	addc		%r20,%r19,%r28
    58  	ldws,ma		4(0,%r25),%r20
    59  	ldws,ma		4(0,%r24),%r19
    60  	stws,ma		%r28,4(0,%r26)
    61  	addc		%r20,%r19,%r28
    62  	ldws,ma		4(0,%r25),%r20
    63  	ldws,ma		4(0,%r24),%r19
    64  	stws,ma		%r28,4(0,%r26)
    65  	addib,>		-4,%r23,L(loop)
    66  	addc		%r20,%r19,%r28
    67  
    68  LDEF(rest)
    69  	addib,=		4,%r23,L(end)
    70  	nop
    71  
    72  LDEF(eloop)
    73  	ldws,ma		4(0,%r25),%r20
    74  	ldws,ma		4(0,%r24),%r19
    75  	stws,ma		%r28,4(0,%r26)
    76  	addib,>		-1,%r23,L(eloop)
    77  	addc		%r20,%r19,%r28
    78  
    79  LDEF(end)
    80  	stws		%r28,0(0,%r26)
    81  	bv		0(%r2)
    82  	 addc		%r0,%r0,%r28
    83  EPILOGUE()