github.com/aergoio/aergo@v1.3.1/libtool/src/gmp-6.1.2/mpn/x86/atom/sse2/mul_1.asm (about)

     1  dnl  Intel Atom mpn_mul_1.
     2  
     3  dnl  Contributed to the GNU project by Torbjorn Granlund and Marco Bodrato.
     4  
     5  dnl  Copyright 2011 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			    cycles/limb
    36  C			    cycles/limb
    37  C P5				 -
    38  C P6 model 0-8,10-12		 -
    39  C P6 model 9  (Banias)
    40  C P6 model 13 (Dothan)
    41  C P4 model 0  (Willamette)
    42  C P4 model 1  (?)
    43  C P4 model 2  (Northwood)
    44  C P4 model 3  (Prescott)
    45  C P4 model 4  (Nocona)
    46  C Intel Atom			 7.5
    47  C AMD K6			 -
    48  C AMD K7			 -
    49  C AMD K8
    50  C AMD K10
    51  
    52  defframe(PARAM_CARRY,20)
    53  defframe(PARAM_MUL,  16)
    54  defframe(PARAM_SIZE, 12)
    55  defframe(PARAM_SRC,  8)
    56  defframe(PARAM_DST,  4)
    57  
    58  define(`rp', `%edx')
    59  define(`up', `%esi')
    60  define(`n',  `%ecx')
    61  
    62  ASM_START()
    63  	TEXT
    64  	ALIGN(16)
    65  deflit(`FRAME',0)
    66  
    67  PROLOGUE(mpn_mul_1c)
    68  	movd	PARAM_CARRY, %mm6	C carry
    69  	jmp	L(ent)
    70  EPILOGUE()
    71  
    72  	ALIGN(8)			C for compact code
    73  PROLOGUE(mpn_mul_1)
    74  	pxor	%mm6, %mm6
    75  L(ent):	push	%esi			FRAME_pushl()
    76  	mov	PARAM_SRC, up
    77  	mov	PARAM_SIZE, %eax	C size
    78  	movd	PARAM_MUL, %mm7
    79  	movd	(up), %mm0
    80  	mov	%eax, n
    81  	and	$3, %eax
    82  	pmuludq	%mm7, %mm0
    83  	mov	PARAM_DST, rp
    84  	jz	L(lo0)
    85  	cmp	$2, %eax
    86  	lea	-16(up,%eax,4),up
    87  	lea	-16(rp,%eax,4),rp
    88  	jc	L(lo1)
    89  	jz	L(lo2)
    90  	jmp	L(lo3)
    91  
    92  	ALIGN(16)
    93  L(top):	movd	(up), %mm0
    94  	pmuludq	%mm7, %mm0
    95  	psrlq	$32, %mm6
    96  	lea	16(rp), rp
    97  L(lo0):	paddq	%mm0, %mm6
    98  	movd	4(up), %mm0
    99  	pmuludq	%mm7, %mm0
   100  	movd	%mm6, (rp)
   101  	psrlq	$32, %mm6
   102  L(lo3):	paddq	%mm0, %mm6
   103  	movd	8(up), %mm0
   104  	pmuludq	%mm7, %mm0
   105  	movd	%mm6, 4(rp)
   106  	psrlq	$32, %mm6
   107  L(lo2):	paddq	%mm0, %mm6
   108  	movd	12(up), %mm0
   109  	pmuludq	%mm7, %mm0
   110  	movd	%mm6, 8(rp)
   111  	psrlq	$32, %mm6
   112  L(lo1):	paddq	%mm0, %mm6
   113  	sub	$4, n
   114  	movd	%mm6, 12(rp)
   115  	lea	16(up), up
   116  	ja	L(top)
   117  
   118  	psrlq	$32, %mm6
   119  	movd	%mm6, %eax
   120  	emms
   121  	pop	%esi			FRAME_popl()
   122  	ret
   123  EPILOGUE()
   124  ASM_END()