source/libs/gmp/gmp-src/mpn/powerpc32/mul_1.asm

   1 dnl  PowerPC-32 mpn_mul_1 -- Multiply a limb vector with a limb and store the
   2 dnl  result in a second limb vector.
   3
   4 dnl  Copyright 1995, 1997, 2000, 2002, 2003, 2005 Free Software Foundation,
   5 dnl  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 603e:            ?
  37 C 604e:            4.0
  38 C 75x (G3):        4.5-11
  39 C 7400,7410 (G4):  4.5-11
  40 C 744x,745x (G4+): 6.0
  41 C power4/ppc970:   6.0
  42 C power5:          5.63
  43
  44 C INPUT PARAMETERS
  45 C rp    r3
  46 C up    r4
  47 C n     r5
  48 C vl    r6
  49
  50 ASM_START()
  51 PROLOGUE(mpn_mul_1)
  52         mtctr   r5
  53         addi    r3,r3,-4        C adjust res_ptr, it's offset before it's used
  54         li      r12,0           C clear upper product reg
  55         addic   r0,r0,0         C clear cy
  56 C Start software pipeline
  57         lwz     r8,0(r4)
  58         bdz     L(end3)
  59         lwzu    r9,4(r4)
  60         mullw   r11,r8,r6
  61         mulhwu  r0,r8,r6
  62         bdz     L(end1)
  63 C Software pipelined main loop
  64 L(loop):
  65         lwz     r8,4(r4)
  66         mullw   r10,r9,r6
  67         adde    r5,r11,r12
  68         mulhwu  r12,r9,r6
  69         stw     r5,4(r3)
  70         bdz     L(end2)
  71         lwzu    r9,8(r4)
  72         mullw   r11,r8,r6
  73         adde    r7,r10,r0
  74         mulhwu  r0,r8,r6
  75         stwu    r7,8(r3)
  76         bdnz    L(loop)
  77 C Finish software pipeline
  78 L(end1):
  79         mullw   r10,r9,r6
  80         adde    r5,r11,r12
  81         mulhwu  r12,r9,r6
  82         stw     r5,4(r3)
  83         adde    r7,r10,r0
  84         stwu    r7,8(r3)
  85         addze   r3,r12
  86         blr
  87 L(end2):
  88         mullw   r11,r8,r6
  89         adde    r7,r10,r0
  90         mulhwu  r0,r8,r6
  91         stwu    r7,8(r3)
  92         adde    r5,r11,r12
  93         stw     r5,4(r3)
  94         addze   r3,r0
  95         blr
  96 L(end3):
  97         mullw   r11,r8,r6
  98         stw     r11,4(r3)
  99         mulhwu  r3,r8,r6
 100         blr
 101 EPILOGUE(mpn_mul_1)