Return to popcount.asm CVS log

Up to [local] / OpenXM_contrib / gmp / mpn / x86 / pentium

Annotation of OpenXM_contrib/gmp/mpn/x86/pentium/popcount.asm, Revision 1.1.1.1

1.1       ohara       1: dnl  Intel P5 mpn_popcount -- mpn bit population count.
                      2:
                      3: dnl  Copyright 2001, 2002 Free Software Foundation, Inc.
                      4: dnl
                      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
                      8: dnl  modify it under the terms of the GNU Lesser General Public License as
                      9: dnl  published by the Free Software Foundation; either version 2.1 of the
                     10: dnl  License, or (at your option) any later version.
                     11: dnl
                     12: dnl  The GNU MP Library is distributed in the hope that it will be useful,
                     13: dnl  but WITHOUT ANY WARRANTY; without even the implied warranty of
                     14: dnl  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
                     15: dnl  Lesser General Public License for more details.
                     16: dnl
                     17: dnl  You should have received a copy of the GNU Lesser General Public
                     18: dnl  License along with the GNU MP Library; see the file COPYING.LIB.  If
                     19: dnl  not, write to the Free Software Foundation, Inc., 59 Temple Place -
                     20: dnl  Suite 330, Boston, MA 02111-1307, USA.
                     21:
                     22: include(`../config.m4')
                     23:
                     24:
                     25: C P5: 8.0 cycles/limb
                     26:
                     27:
                     28: C unsigned long mpn_popcount (mp_srcptr src, mp_size_t size);
                     29: C
                     30: C An arithmetic approach has been found to be slower than the table lookup,
                     31: C due to needing too many instructions.
                     32:
                     33: C The slightly strange quoting here helps the renaming done by tune/many.pl.
                     34: deflit(TABLE_NAME,
                     35: m4_assert_defined(`GSYM_PREFIX')
                     36: GSYM_PREFIX`'mpn_popcount``'_table')
                     37:
                     38:        RODATA
                     39:        ALIGN(8)
                     40:        GLOBL   TABLE_NAME
                     41: TABLE_NAME:
                     42: forloop(i,0,255,
                     43: `      .byte   m4_popcount(i)
                     44: ')
                     45:
                     46: defframe(PARAM_SIZE,8)
                     47: defframe(PARAM_SRC, 4)
                     48:
                     49:        TEXT
                     50:        ALIGN(8)
                     51:
                     52: PROLOGUE(mpn_popcount)
                     53: deflit(`FRAME',0)
                     54:
                     55:        movl    PARAM_SIZE, %ecx
                     56:        pushl   %esi    FRAME_pushl()
                     57:
                     58: ifdef(`PIC',`
                     59:        pushl   %ebx    FRAME_pushl()
                     60:        pushl   %ebp    FRAME_pushl()
                     61:
                     62:        call    L(here)
                     63: L(here):
                     64:        popl    %ebp
                     65:        shll    %ecx            C size in byte pairs
                     66:
                     67:        addl    $_GLOBAL_OFFSET_TABLE_+[.-L(here)], %ebp
                     68:        movl    PARAM_SRC, %esi
                     69:
                     70:        xorl    %eax, %eax      C total
                     71:        xorl    %ebx, %ebx      C byte
                     72:
                     73:        movl    TABLE_NAME@GOT(%ebp), %ebp
                     74:        xorl    %edx, %edx      C byte
                     75: define(TABLE,`(%ebp,$1)')
                     76: ',`
                     77: dnl non-PIC
                     78:        shll    %ecx            C size in byte pairs
                     79:        movl    PARAM_SRC, %esi
                     80:
                     81:        pushl   %ebx    FRAME_pushl()
                     82:        xorl    %eax, %eax      C total
                     83:
                     84:        xorl    %ebx, %ebx      C byte
                     85:        xorl    %edx, %edx      C byte
                     86:
                     87: define(TABLE,`TABLE_NAME`'($1)')
                     88: ')
                     89:
                     90:
                     91:        ALIGN(8)        C necessary on P55 for claimed speed
                     92: L(top):
                     93:        C eax   total
                     94:        C ebx   byte
                     95:        C ecx   counter, 2*size to 2
                     96:        C edx   byte
                     97:        C esi   src
                     98:        C edi
                     99:        C ebp   [PIC] table
                    100:
                    101:        addl    %ebx, %eax
                    102:        movb    -1(%esi,%ecx,2), %bl
                    103:
                    104:        addl    %edx, %eax
                    105:        movb    -2(%esi,%ecx,2), %dl
                    106:
                    107:        movb    TABLE(%ebx), %bl
                    108:        decl    %ecx
                    109:
                    110:        movb    TABLE(%edx), %dl
                    111:        jnz     L(top)
                    112:
                    113:
                    114: ifdef(`PIC',`
                    115:        popl    %ebp
                    116: ')
                    117:        addl    %ebx, %eax
                    118:        popl    %ebx
                    119:
                    120:        addl    %edx, %eax
                    121:        popl    %esi
                    122:
                    123:        ret
                    124:
                    125: EPILOGUE()