source/libs/gmp/gmp-src/mpn/powerpc32/bdiv_dbm1c.asm

   1 dnl  PPC32 mpn_bdiv_dbm1c.
   2
   3 dnl  Copyright 2008 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 C                cycles/limb
  34 C 603e:            ?
  35 C 604e:            ?
  36 C 75x (G3):        ?
  37 C 7400,7410 (G4):  9.43
  38 C 744x,745x (G4+): 6.28
  39 C power4/ppc970:   ?
  40 C power5:          ?
  41
  42 C TODO
  43 C  * Nothing to do...
  44
  45 C INPUT PARAMETERS
  46 define(`rp', `r3')
  47 define(`up', `r4')
  48 define(`n',  `r5')
  49 define(`bd', `r6')
  50 define(`cy', `r7')
  51
  52 ASM_START()
  53 PROLOGUE(mpn_bdiv_dbm1c)
  54         lwz     r0, 0(r4)
  55
  56         rlwinm. r12, r5, 0,30,31
  57         cmplwi  cr6, r12, 2
  58         cmplwi  cr7, r5, 4
  59         addi    r5, r5, 1
  60         srwi    r5, r5, 2
  61         mtctr   r5
  62         beq     cr0, L(b00)
  63         blt     cr6, L(b01)
  64         beq     cr6, L(b10)
  65
  66 L(b11): mullw   r5, r0, r6
  67         mulhwu  r12, r0, r6
  68         lwz     r0, 4(r4)
  69         addi    r4, r4, -12
  70         addi    r3, r3, -12
  71         b       L(3)
  72
  73 L(b00): mullw   r9, r0, r6
  74         mulhwu  r8, r0, r6
  75         lwz     r0, 4(r4)
  76         addi    r4, r4, -8
  77         addi    r3, r3, -8
  78         b       L(0)
  79
  80 L(b01): mullw   r5, r0, r6
  81         mulhwu  r12, r0, r6
  82         addi    r3, r3, -4
  83         ble     cr7, L(e1)
  84         lwz     r0, 4(r4)
  85         addi    r4, r4, -4
  86         b       L(1)
  87
  88 L(b10): mullw   r9, r0, r6
  89         mulhwu  r8, r0, r6
  90         lwz     r0, 4(r4)
  91         ble     cr7, L(e2)
  92
  93         ALIGN(16)
  94 L(top): mullw   r5, r0, r6
  95         mulhwu  r12, r0, r6
  96         subfc   r11, r9, r7
  97         lwz     r0, 8(r4)
  98         subfe   r7, r8, r11
  99         stw     r11, 0(r3)
 100 L(1):   mullw   r9, r0, r6
 101         mulhwu  r8, r0, r6
 102         subfc   r11, r5, r7
 103         lwz     r0, 12(r4)
 104         subfe   r7, r12, r11
 105         stw     r11, 4(r3)
 106 L(0):   mullw   r5, r0, r6
 107         mulhwu  r12, r0, r6
 108         subfc   r11, r9, r7
 109         lwz     r0, 16(r4)
 110         subfe   r7, r8, r11
 111         stw     r11, 8(r3)
 112 L(3):   mullw   r9, r0, r6
 113         mulhwu  r8, r0, r6
 114         subfc   r11, r5, r7
 115         lwz     r0, 20(r4)
 116         subfe   r7, r12, r11
 117         stw     r11, 12(r3)
 118         addi    r4, r4, 16
 119         addi    r3, r3, 16
 120         bdnz    L(top)
 121
 122 L(e2):  mullw   r5, r0, r6
 123         mulhwu  r12, r0, r6
 124         subfc   r11, r9, r7
 125         subfe   r7, r8, r11
 126         stw     r11, 0(r3)
 127 L(e1):  subfc   r11, r5, r7
 128         stw     r11, 4(r3)
 129         subfe   r3, r12, r11
 130         blr
 131 EPILOGUE()