1 dnl AMD K6 mpn_lshift
-- mpn left shift.
3 dnl Copyright
1999, 2000, 2002 Free Software Foundation
, Inc.
5 dnl
This file is part of the GNU MP Library.
7 dnl The GNU MP Library is free software
; you can redistribute it and/or modify
8 dnl it under the terms of
either:
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.
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
20 dnl
or both
in parallel
, as here.
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
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
/.
31 include(`..
/config.m4
')
37 C mp_limb_t mpn_lshift (mp_ptr dst, mp_srcptr src, mp_size_t size,
40 C The loop runs at 3 cycles/limb, limited by decoding and by having 3 mmx
41 C instructions. This is despite every second fetch being unaligned.
44 defframe(PARAM_SHIFT,16)
45 defframe(PARAM_SIZE, 12)
46 defframe(PARAM_SRC, 8)
47 defframe(PARAM_DST, 4)
55 C The
1 limb case can be done without the
push %ebx, but it
's then
56 C still the same speed. The push is left as a free helping hand for
57 C the two_or_more code.
60 pushl %ebx FRAME_pushl()
65 movl PARAM_SHIFT, %ecx
68 movl (%ebx), %edx C src limb
71 shldl( %cl, %edx, %eax) C return value
75 movl %edx, (%ebx) C dst limb
81 ALIGN(16) C avoid offset 0x1f
82 nop C avoid bad cache line crossing
89 movl (%ebx,%eax,4), %edx C src high limb
92 movd PARAM_SHIFT, %mm6
93 addl $32, %ecx C 32-shift
101 C eax counter, size-1 to 1
110 movq -4(%ebx,%eax,4), %mm0
115 movd %mm0, 4(%ecx,%eax,4)