github.com/aergoio/aergo@v1.3.1/libtool/src/gmp-6.1.2/mpn/x86_64/lshiftc.asm (about)

     1  dnl  AMD64 mpn_lshiftc -- mpn left shift with complement.
     2  
     3  dnl  Copyright 2003, 2005, 2006, 2009, 2011, 2012 Free Software Foundation, Inc.
     4  
     5  dnl  This file is part of the GNU MP Library.
     6  dnl
     7  dnl  The GNU MP Library is free software; you can redistribute it and/or modify
     8  dnl  it under the terms of either:
     9  dnl
    10  dnl    * the GNU Lesser General Public License as published by the Free
    11  dnl      Software Foundation; either version 3 of the License, or (at your
    12  dnl      option) any later version.
    13  dnl
    14  dnl  or
    15  dnl
    16  dnl    * the GNU General Public License as published by the Free Software
    17  dnl      Foundation; either version 2 of the License, or (at your option) any
    18  dnl      later version.
    19  dnl
    20  dnl  or both in parallel, as here.
    21  dnl
    22  dnl  The GNU MP Library is distributed in the hope that it will be useful, but
    23  dnl  WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
    24  dnl  or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
    25  dnl  for more details.
    26  dnl
    27  dnl  You should have received copies of the GNU General Public License and the
    28  dnl  GNU Lesser General Public License along with the GNU MP Library.  If not,
    29  dnl  see https://www.gnu.org/licenses/.
    30  
    31  include(`../config.m4')
    32  
    33  
    34  C	     cycles/limb
    35  C AMD K8,K9	 2.75
    36  C AMD K10	 2.75
    37  C Intel P4	 ?
    38  C Intel core2	 ?
    39  C Intel corei	 ?
    40  C Intel atom	 ?
    41  C VIA nano	 3.75
    42  
    43  
    44  C INPUT PARAMETERS
    45  define(`rp',	`%rdi')
    46  define(`up',	`%rsi')
    47  define(`n',	`%rdx')
    48  define(`cnt',	`%rcx')
    49  
    50  ABI_SUPPORT(DOS64)
    51  ABI_SUPPORT(STD64)
    52  
    53  ASM_START()
    54  	TEXT
    55  	ALIGN(32)
    56  PROLOGUE(mpn_lshiftc)
    57  	FUNC_ENTRY(4)
    58  	neg	R32(%rcx)		C put rsh count in cl
    59  	mov	-8(up,n,8), %rax
    60  	shr	R8(%rcx), %rax		C function return value
    61  
    62  	neg	R32(%rcx)		C put lsh count in cl
    63  	lea	1(n), R32(%r8)
    64  	and	$3, R32(%r8)
    65  	je	L(rlx)			C jump for n = 3, 7, 11, ...
    66  
    67  	dec	R32(%r8)
    68  	jne	L(1)
    69  C	n = 4, 8, 12, ...
    70  	mov	-8(up,n,8), %r10
    71  	shl	R8(%rcx), %r10
    72  	neg	R32(%rcx)		C put rsh count in cl
    73  	mov	-16(up,n,8), %r8
    74  	shr	R8(%rcx), %r8
    75  	or	%r8, %r10
    76  	not	%r10
    77  	mov	%r10, -8(rp,n,8)
    78  	dec	n
    79  	jmp	L(rll)
    80  
    81  L(1):	dec	R32(%r8)
    82  	je	L(1x)			C jump for n = 1, 5, 9, 13, ...
    83  C	n = 2, 6, 10, 16, ...
    84  	mov	-8(up,n,8), %r10
    85  	shl	R8(%rcx), %r10
    86  	neg	R32(%rcx)		C put rsh count in cl
    87  	mov	-16(up,n,8), %r8
    88  	shr	R8(%rcx), %r8
    89  	or	%r8, %r10
    90  	not	%r10
    91  	mov	%r10, -8(rp,n,8)
    92  	dec	n
    93  	neg	R32(%rcx)		C put lsh count in cl
    94  L(1x):
    95  	cmp	$1, n
    96  	je	L(ast)
    97  	mov	-8(up,n,8), %r10
    98  	shl	R8(%rcx), %r10
    99  	mov	-16(up,n,8), %r11
   100  	shl	R8(%rcx), %r11
   101  	neg	R32(%rcx)		C put rsh count in cl
   102  	mov	-16(up,n,8), %r8
   103  	mov	-24(up,n,8), %r9
   104  	shr	R8(%rcx), %r8
   105  	or	%r8, %r10
   106  	shr	R8(%rcx), %r9
   107  	or	%r9, %r11
   108  	not	%r10
   109  	not	%r11
   110  	mov	%r10, -8(rp,n,8)
   111  	mov	%r11, -16(rp,n,8)
   112  	sub	$2, n
   113  
   114  L(rll):	neg	R32(%rcx)		C put lsh count in cl
   115  L(rlx):	mov	-8(up,n,8), %r10
   116  	shl	R8(%rcx), %r10
   117  	mov	-16(up,n,8), %r11
   118  	shl	R8(%rcx), %r11
   119  
   120  	sub	$4, n			C				      4
   121  	jb	L(end)			C				      2
   122  	ALIGN(16)
   123  L(top):
   124  	C finish stuff from lsh block
   125  	neg	R32(%rcx)		C put rsh count in cl
   126  	mov	16(up,n,8), %r8
   127  	mov	8(up,n,8), %r9
   128  	shr	R8(%rcx), %r8
   129  	or	%r8, %r10
   130  	shr	R8(%rcx), %r9
   131  	or	%r9, %r11
   132  	not	%r10
   133  	not	%r11
   134  	mov	%r10, 24(rp,n,8)
   135  	mov	%r11, 16(rp,n,8)
   136  	C start two new rsh
   137  	mov	0(up,n,8), %r8
   138  	mov	-8(up,n,8), %r9
   139  	shr	R8(%rcx), %r8
   140  	shr	R8(%rcx), %r9
   141  
   142  	C finish stuff from rsh block
   143  	neg	R32(%rcx)		C put lsh count in cl
   144  	mov	8(up,n,8), %r10
   145  	mov	0(up,n,8), %r11
   146  	shl	R8(%rcx), %r10
   147  	or	%r10, %r8
   148  	shl	R8(%rcx), %r11
   149  	or	%r11, %r9
   150  	not	%r8
   151  	not	%r9
   152  	mov	%r8, 8(rp,n,8)
   153  	mov	%r9, 0(rp,n,8)
   154  	C start two new lsh
   155  	mov	-8(up,n,8), %r10
   156  	mov	-16(up,n,8), %r11
   157  	shl	R8(%rcx), %r10
   158  	shl	R8(%rcx), %r11
   159  
   160  	sub	$4, n
   161  	jae	L(top)			C				      2
   162  L(end):
   163  	neg	R32(%rcx)		C put rsh count in cl
   164  	mov	8(up), %r8
   165  	shr	R8(%rcx), %r8
   166  	or	%r8, %r10
   167  	mov	(up), %r9
   168  	shr	R8(%rcx), %r9
   169  	or	%r9, %r11
   170  	not	%r10
   171  	not	%r11
   172  	mov	%r10, 16(rp)
   173  	mov	%r11, 8(rp)
   174  
   175  	neg	R32(%rcx)		C put lsh count in cl
   176  L(ast):	mov	(up), %r10
   177  	shl	R8(%rcx), %r10
   178  	not	%r10
   179  	mov	%r10, (rp)
   180  	FUNC_EXIT()
   181  	ret
   182  EPILOGUE()