source/libs/gmp/gmp-src/mpn/x86/pentium4/sse2/mod_1_1.asm

   1 dnl  x86-32 mpn_mod_1_1p for Pentium 4 and P6 models with SSE2 (i.e., 9,D,E,F).
   2
   3 dnl  Contributed to the GNU project by Torbjorn Granlund.
   4
   5 dnl  Copyright 2009, 2010 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 TODO:
  36 C  * Optimize.  The present code was written quite straightforwardly.
  37 C  * Optimize post-loop reduction code; it is from mod_1s_4p, thus overkill.
  38 C  * Write a cps function that uses sse2 insns.
  39
  40 C                           cycles/limb
  41 C P6 model 0-8,10-12            -
  42 C P6 model 9   (Banias)         ?
  43 C P6 model 13  (Dothan)         ?
  44 C P4 model 0-1 (Willamette)     ?
  45 C P4 model 2   (Northwood)     16
  46 C P4 model 3-4 (Prescott)      18
  47
  48 C INPUT PARAMETERS
  49 C ap            sp + 4
  50 C n             sp + 8
  51 C b             sp + 12
  52 C cps           sp + 16
  53
  54 define(`B1modb', `%mm1')
  55 define(`B2modb', `%mm2')
  56 define(`ap',     `%edx')
  57 define(`n',      `%eax')
  58
  59         TEXT
  60         ALIGN(16)
  61 PROLOGUE(mpn_mod_1_1p)
  62         push    %ebx
  63         mov     8(%esp), ap
  64         mov     12(%esp), n
  65         mov     20(%esp), %ecx
  66         movd    8(%ecx), B1modb
  67         movd    12(%ecx), B2modb
  68
  69         lea     -4(ap,n,4), ap
  70
  71 C FIXME: See comment in generic/mod_1_1.c.
  72         movd    (ap), %mm7
  73         movd    -4(ap), %mm4
  74         pmuludq B1modb, %mm7
  75         paddq   %mm4, %mm7
  76         add     $-2, n
  77         jz      L(end)
  78
  79         ALIGN(8)
  80 L(top): movq    %mm7, %mm6
  81         psrlq   $32, %mm7               C rh
  82         movd    -8(ap), %mm0
  83         add     $-4, ap
  84         pmuludq B2modb, %mm7
  85         pmuludq B1modb, %mm6
  86         add     $-1, n
  87         paddq   %mm0, %mm7
  88         paddq   %mm6, %mm7
  89         jnz     L(top)
  90
  91 L(end): pcmpeqd %mm4, %mm4
  92         psrlq   $32, %mm4               C 0x00000000FFFFFFFF
  93         pand    %mm7, %mm4              C rl
  94         psrlq   $32, %mm7               C rh
  95         pmuludq B1modb, %mm7            C rh,cl
  96         paddq   %mm4, %mm7              C rh,rl
  97         movd    4(%ecx), %mm4           C cnt
  98         psllq   %mm4, %mm7              C rh,rl normalized
  99         movq    %mm7, %mm2              C rl in low half
 100         psrlq   $32, %mm7               C rh
 101         movd    (%ecx), %mm1            C bi
 102         pmuludq %mm7, %mm1              C qh,ql
 103         paddq   %mm2, %mm1              C qh-1,ql
 104         movd    %mm1, %ecx              C ql
 105         psrlq   $32, %mm1               C qh-1
 106         movd    16(%esp), %mm3          C b
 107         pmuludq %mm1, %mm3              C (qh-1) * b
 108         psubq   %mm3, %mm2              C r in low half (could use psubd)
 109         movd    %mm2, %eax              C r
 110         mov     16(%esp), %ebx
 111         sub     %ebx, %eax              C r
 112         cmp     %eax, %ecx
 113         lea     (%eax,%ebx), %edx
 114         cmovc(  %edx, %eax)
 115         movd    %mm4, %ecx              C cnt
 116         cmp     %ebx, %eax
 117         jae     L(fix)
 118         emms
 119         pop     %ebx
 120         shr     %cl, %eax
 121         ret
 122
 123 L(fix): sub     %ebx, %eax
 124         emms
 125         pop     %ebx
 126         shr     %cl, %eax
 127         ret
 128 EPILOGUE()
 129
 130 PROLOGUE(mpn_mod_1_1p_cps)
 131 C CAUTION: This is the same code as in k7/mod_1_1.asm
 132         push    %ebp
 133         mov     12(%esp), %ebp
 134         push    %esi
 135         bsr     %ebp, %ecx
 136         push    %ebx
 137         xor     $31, %ecx
 138         mov     16(%esp), %esi
 139         sal     %cl, %ebp
 140         mov     %ebp, %edx
 141         not     %edx
 142         mov     $-1, %eax
 143         div     %ebp
 144         mov     %eax, (%esi)            C store bi
 145         mov     %ecx, 4(%esi)           C store cnt
 146         xor     %ebx, %ebx
 147         sub     %ebp, %ebx
 148         mov     $1, %edx
 149         shld    %cl, %eax, %edx
 150         imul    %edx, %ebx
 151         mul     %ebx
 152         add     %ebx, %edx
 153         not     %edx
 154         imul    %ebp, %edx
 155         add     %edx, %ebp
 156         cmp     %edx, %eax
 157         cmovc(  %ebp, %edx)
 158         shr     %cl, %ebx
 159         mov     %ebx, 8(%esi)           C store B1modb
 160         shr     %cl, %edx
 161         mov     %edx, 12(%esi)          C store B2modb
 162         pop     %ebx
 163         pop     %esi
 164         pop     %ebp
 165         ret
 166 EPILOGUE()