|
Packit |
5c3484 |
dnl Intel Pentium mpn_and_n,...,mpn_xnor_n -- bitwise logical operations.
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
dnl Copyright 2001, 2002 Free Software Foundation, Inc.
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
dnl This file is part of the GNU MP Library.
|
|
Packit |
5c3484 |
dnl
|
|
Packit |
5c3484 |
dnl The GNU MP Library is free software; you can redistribute it and/or modify
|
|
Packit |
5c3484 |
dnl it under the terms of either:
|
|
Packit |
5c3484 |
dnl
|
|
Packit |
5c3484 |
dnl * the GNU Lesser General Public License as published by the Free
|
|
Packit |
5c3484 |
dnl Software Foundation; either version 3 of the License, or (at your
|
|
Packit |
5c3484 |
dnl option) any later version.
|
|
Packit |
5c3484 |
dnl
|
|
Packit |
5c3484 |
dnl or
|
|
Packit |
5c3484 |
dnl
|
|
Packit |
5c3484 |
dnl * the GNU General Public License as published by the Free Software
|
|
Packit |
5c3484 |
dnl Foundation; either version 2 of the License, or (at your option) any
|
|
Packit |
5c3484 |
dnl later version.
|
|
Packit |
5c3484 |
dnl
|
|
Packit |
5c3484 |
dnl or both in parallel, as here.
|
|
Packit |
5c3484 |
dnl
|
|
Packit |
5c3484 |
dnl The GNU MP Library is distributed in the hope that it will be useful, but
|
|
Packit |
5c3484 |
dnl WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
|
|
Packit |
5c3484 |
dnl or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
|
|
Packit |
5c3484 |
dnl for more details.
|
|
Packit |
5c3484 |
dnl
|
|
Packit |
5c3484 |
dnl You should have received copies of the GNU General Public License and the
|
|
Packit |
5c3484 |
dnl GNU Lesser General Public License along with the GNU MP Library. If not,
|
|
Packit |
5c3484 |
dnl see https://www.gnu.org/licenses/.
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
include(`../config.m4')
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
C P5: 3.0 c/l and, ior, xor
|
|
Packit |
5c3484 |
C 3.5 c/l andn, iorn, nand, nior, xnor
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
define(M4_choose_op,
|
|
Packit |
5c3484 |
`ifdef(`OPERATION_$1',`
|
|
Packit |
5c3484 |
define(`M4_function', `mpn_$1')
|
|
Packit |
5c3484 |
define(`M4_want_pre', `$4')
|
|
Packit |
5c3484 |
define(`M4op', `$3')
|
|
Packit |
5c3484 |
define(`M4_want_post',`$2')
|
|
Packit |
5c3484 |
')')
|
|
Packit |
5c3484 |
define(M4pre, `ifelse(M4_want_pre, yes,`$1')')
|
|
Packit |
5c3484 |
define(M4post,`ifelse(M4_want_post,yes,`$1')')
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
M4_choose_op( and_n, , andl, )
|
|
Packit |
5c3484 |
M4_choose_op( andn_n, , andl, yes)
|
|
Packit |
5c3484 |
M4_choose_op( nand_n, yes, andl, )
|
|
Packit |
5c3484 |
M4_choose_op( ior_n, , orl, )
|
|
Packit |
5c3484 |
M4_choose_op( iorn_n, , orl, yes)
|
|
Packit |
5c3484 |
M4_choose_op( nior_n, yes, orl, )
|
|
Packit |
5c3484 |
M4_choose_op( xor_n, , xorl, )
|
|
Packit |
5c3484 |
M4_choose_op( xnor_n, yes, xorl, )
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
ifdef(`M4_function',,
|
|
Packit |
5c3484 |
`m4_error(`Unrecognised or undefined OPERATION symbol
|
|
Packit |
5c3484 |
')')
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
MULFUNC_PROLOGUE(mpn_and_n mpn_andn_n mpn_nand_n mpn_ior_n mpn_iorn_n mpn_nior_n mpn_xor_n mpn_xnor_n)
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
NAILS_SUPPORT(0-31)
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
C void M4_function (mp_ptr wp, mp_srcptr xp, mp_srcptr yp, mp_size_t size);
|
|
Packit |
5c3484 |
C
|
|
Packit |
5c3484 |
C Nothing complicated here, just some care to avoid data cache bank clashes
|
|
Packit |
5c3484 |
C and AGIs.
|
|
Packit |
5c3484 |
C
|
|
Packit |
5c3484 |
C We're one register short of being able to do a simple 4 loads, 2 ops, 2
|
|
Packit |
5c3484 |
C stores. Instead %ebp is juggled a bit and nops are introduced to keep the
|
|
Packit |
5c3484 |
C pairings as intended. An in-place operation would free up a register, for
|
|
Packit |
5c3484 |
C an 0.5 c/l speedup, if that's worth bothering with.
|
|
Packit |
5c3484 |
C
|
|
Packit |
5c3484 |
C This code seems best for P55 too. Data alignment is a big problem for MMX
|
|
Packit |
5c3484 |
C and the pairing restrictions on movq and integer instructions make life
|
|
Packit |
5c3484 |
C difficult.
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
defframe(PARAM_SIZE,16)
|
|
Packit |
5c3484 |
defframe(PARAM_YP, 12)
|
|
Packit |
5c3484 |
defframe(PARAM_XP, 8)
|
|
Packit |
5c3484 |
defframe(PARAM_WP, 4)
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
TEXT
|
|
Packit |
5c3484 |
ALIGN(8)
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
PROLOGUE(M4_function)
|
|
Packit |
5c3484 |
deflit(`FRAME',0)
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
pushl %ebx FRAME_pushl()
|
|
Packit |
5c3484 |
pushl %esi FRAME_pushl()
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
pushl %edi FRAME_pushl()
|
|
Packit |
5c3484 |
pushl %ebp FRAME_pushl()
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
movl PARAM_SIZE, %ecx
|
|
Packit |
5c3484 |
movl PARAM_XP, %ebx
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
movl PARAM_YP, %esi
|
|
Packit |
5c3484 |
movl PARAM_WP, %edi
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
shrl %ecx
|
|
Packit |
5c3484 |
jnc L(entry)
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
movl (%ebx,%ecx,8), %eax C risk of data cache bank clash here
|
|
Packit |
5c3484 |
movl (%esi,%ecx,8), %edx
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
M4pre(` notl_or_xorl_GMP_NUMB_MASK(%edx)')
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
M4op %edx, %eax
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
M4post(`xorl $GMP_NUMB_MASK, %eax')
|
|
Packit |
5c3484 |
orl %ecx, %ecx
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
movl %eax, (%edi,%ecx,8)
|
|
Packit |
5c3484 |
jz L(done)
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
jmp L(entry)
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
L(top):
|
|
Packit |
5c3484 |
C eax
|
|
Packit |
5c3484 |
C ebx xp
|
|
Packit |
5c3484 |
C ecx counter, limb pairs, decrementing
|
|
Packit |
5c3484 |
C edx
|
|
Packit |
5c3484 |
C esi yp
|
|
Packit |
5c3484 |
C edi wp
|
|
Packit |
5c3484 |
C ebp
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
M4op %ebp, %edx
|
|
Packit |
5c3484 |
nop
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
M4post(`xorl $GMP_NUMB_MASK, %eax')
|
|
Packit |
5c3484 |
M4post(`xorl $GMP_NUMB_MASK, %edx')
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
movl %eax, 4(%edi,%ecx,8)
|
|
Packit |
5c3484 |
movl %edx, (%edi,%ecx,8)
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
L(entry):
|
|
Packit |
5c3484 |
movl -4(%ebx,%ecx,8), %ebp
|
|
Packit |
5c3484 |
nop
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
movl -4(%esi,%ecx,8), %eax
|
|
Packit |
5c3484 |
movl -8(%esi,%ecx,8), %edx
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
M4pre(` xorl $GMP_NUMB_MASK, %eax')
|
|
Packit |
5c3484 |
M4pre(` xorl $GMP_NUMB_MASK, %edx')
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
M4op %ebp, %eax
|
|
Packit |
5c3484 |
movl -8(%ebx,%ecx,8), %ebp
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
decl %ecx
|
|
Packit |
5c3484 |
jnz L(top)
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
M4op %ebp, %edx
|
|
Packit |
5c3484 |
nop
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
M4post(`xorl $GMP_NUMB_MASK, %eax')
|
|
Packit |
5c3484 |
M4post(`xorl $GMP_NUMB_MASK, %edx')
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
movl %eax, 4(%edi,%ecx,8)
|
|
Packit |
5c3484 |
movl %edx, (%edi,%ecx,8)
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
L(done):
|
|
Packit |
5c3484 |
popl %ebp
|
|
Packit |
5c3484 |
popl %edi
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
popl %esi
|
|
Packit |
5c3484 |
popl %ebx
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
ret
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
EPILOGUE()
|