|
Packit |
5c3484 |
dnl Intel Atom mpn_and_n,...,mpn_xnor_n -- bitwise logical operations.
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
dnl Copyright 2011 Free Software Foundation, Inc.
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
dnl Contributed to the GNU project by Marco Bodrato.
|
|
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 |
C cycles/limb
|
|
Packit |
5c3484 |
C op nop opn
|
|
Packit |
5c3484 |
C P5
|
|
Packit |
5c3484 |
C P6 model 0-8,10-12
|
|
Packit |
5c3484 |
C P6 model 9 (Banias)
|
|
Packit |
5c3484 |
C P6 model 13 (Dothan)
|
|
Packit |
5c3484 |
C P4 model 0 (Willamette)
|
|
Packit |
5c3484 |
C P4 model 1 (?)
|
|
Packit |
5c3484 |
C P4 model 2 (Northwood)
|
|
Packit |
5c3484 |
C P4 model 3 (Prescott)
|
|
Packit |
5c3484 |
C P4 model 4 (Nocona)
|
|
Packit |
5c3484 |
C Intel Atom 3 3.5 3.5
|
|
Packit |
5c3484 |
C AMD K6
|
|
Packit |
5c3484 |
C AMD K7
|
|
Packit |
5c3484 |
C AMD K8
|
|
Packit |
5c3484 |
C AMD K10
|
|
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(`M4_inst', `$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 |
C void M4_function (mp_ptr dst, mp_srcptr src2, mp_srcptr src1, mp_size_t size);
|
|
Packit |
5c3484 |
C
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
defframe(PARAM_SIZE, 16)
|
|
Packit |
5c3484 |
defframe(PARAM_SRC1, 12)
|
|
Packit |
5c3484 |
defframe(PARAM_SRC2, 8)
|
|
Packit |
5c3484 |
defframe(PARAM_DST, 4)
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
dnl re-use parameter space
|
|
Packit |
5c3484 |
define(SAVE_RP,`PARAM_SIZE')
|
|
Packit |
5c3484 |
define(SAVE_VP,`PARAM_SRC1')
|
|
Packit |
5c3484 |
define(SAVE_UP,`PARAM_DST')
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
define(`rp', `%edi')
|
|
Packit |
5c3484 |
define(`up', `%esi')
|
|
Packit |
5c3484 |
define(`vp', `%ebx')
|
|
Packit |
5c3484 |
define(`cnt', `%eax')
|
|
Packit |
5c3484 |
define(`r1', `%ecx')
|
|
Packit |
5c3484 |
define(`r2', `%edx')
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
ASM_START()
|
|
Packit |
5c3484 |
TEXT
|
|
Packit |
5c3484 |
ALIGN(16)
|
|
Packit |
5c3484 |
deflit(`FRAME',0)
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
PROLOGUE(M4_function)
|
|
Packit |
5c3484 |
mov PARAM_SIZE, cnt C size
|
|
Packit |
5c3484 |
mov rp, SAVE_RP
|
|
Packit |
5c3484 |
mov PARAM_DST, rp
|
|
Packit |
5c3484 |
mov up, SAVE_UP
|
|
Packit |
5c3484 |
mov PARAM_SRC1, up
|
|
Packit |
5c3484 |
shr cnt C size >> 1
|
|
Packit |
5c3484 |
mov vp, SAVE_VP
|
|
Packit |
5c3484 |
mov PARAM_SRC2, vp
|
|
Packit |
5c3484 |
mov (up), r1
|
|
Packit |
5c3484 |
jz L(end) C size == 1
|
|
Packit |
5c3484 |
jnc L(even) C size % 2 == 0
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
ALIGN(16)
|
|
Packit |
5c3484 |
L(oop):
|
|
Packit |
5c3484 |
M4pre(` notl_or_xorl_GMP_NUMB_MASK(r1)')
|
|
Packit |
5c3484 |
M4_inst (vp), r1
|
|
Packit |
5c3484 |
lea 8(up), up
|
|
Packit |
5c3484 |
mov -4(up), r2
|
|
Packit |
5c3484 |
M4post(` notl_or_xorl_GMP_NUMB_MASK(r1)')
|
|
Packit |
5c3484 |
lea 8(vp), vp
|
|
Packit |
5c3484 |
mov r1, (rp)
|
|
Packit |
5c3484 |
L(entry):
|
|
Packit |
5c3484 |
M4pre(` notl_or_xorl_GMP_NUMB_MASK(r2)')
|
|
Packit |
5c3484 |
M4_inst -4(vp), r2
|
|
Packit |
5c3484 |
lea 8(rp), rp
|
|
Packit |
5c3484 |
M4post(` notl_or_xorl_GMP_NUMB_MASK(r2)')
|
|
Packit |
5c3484 |
dec cnt
|
|
Packit |
5c3484 |
mov (up), r1
|
|
Packit |
5c3484 |
mov r2, -4(rp)
|
|
Packit |
5c3484 |
jnz L(oop)
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
L(end):
|
|
Packit |
5c3484 |
M4pre(` notl_or_xorl_GMP_NUMB_MASK(r1)')
|
|
Packit |
5c3484 |
mov SAVE_UP, up
|
|
Packit |
5c3484 |
M4_inst (vp), r1
|
|
Packit |
5c3484 |
M4post(`notl_or_xorl_GMP_NUMB_MASK(r1)')
|
|
Packit |
5c3484 |
mov SAVE_VP, vp
|
|
Packit |
5c3484 |
mov r1, (rp)
|
|
Packit |
5c3484 |
mov SAVE_RP, rp
|
|
Packit |
5c3484 |
ret
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
L(even):
|
|
Packit |
5c3484 |
mov r1, r2
|
|
Packit |
5c3484 |
lea 4(up), up
|
|
Packit |
5c3484 |
lea 4(vp), vp
|
|
Packit |
5c3484 |
lea -4(rp), rp
|
|
Packit |
5c3484 |
jmp L(entry)
|
|
Packit |
5c3484 |
EPILOGUE()
|
|
Packit |
5c3484 |
ASM_END()
|