1dnl  Intel P6 mpn_modexact_1_odd -- exact division style remainder.
2
3dnl  Copyright 2000-2002, 2007 Free Software Foundation, Inc.
4
5dnl  This file is part of the GNU MP Library.
6dnl
7dnl  The GNU MP Library is free software; you can redistribute it and/or modify
8dnl  it under the terms of either:
9dnl
10dnl    * the GNU Lesser General Public License as published by the Free
11dnl      Software Foundation; either version 3 of the License, or (at your
12dnl      option) any later version.
13dnl
14dnl  or
15dnl
16dnl    * the GNU General Public License as published by the Free Software
17dnl      Foundation; either version 2 of the License, or (at your option) any
18dnl      later version.
19dnl
20dnl  or both in parallel, as here.
21dnl
22dnl  The GNU MP Library is distributed in the hope that it will be useful, but
23dnl  WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
24dnl  or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
25dnl  for more details.
26dnl
27dnl  You should have received copies of the GNU General Public License and the
28dnl  GNU Lesser General Public License along with the GNU MP Library.  If not,
29dnl  see https://www.gnu.org/licenses/.
30
31include(`../config.m4')
32
33
34C P6: 10.0 cycles/limb
35
36
37C mp_limb_t mpn_modexact_1_odd (mp_srcptr src, mp_size_t size,
38C                               mp_limb_t divisor);
39C mp_limb_t mpn_modexact_1c_odd (mp_srcptr src, mp_size_t size,
40C                                mp_limb_t divisor, mp_limb_t carry);
41C
42C It's not worth skipping a step at the end when high<divisor since the main
43C loop is only 10 cycles.
44
45defframe(PARAM_CARRY,  16)
46defframe(PARAM_DIVISOR,12)
47defframe(PARAM_SIZE,   8)
48defframe(PARAM_SRC,    4)
49
50dnl  Not enough room under modexact_1 to make these re-use the parameter
51dnl  space, unfortunately.
52defframe(SAVE_EBX,     -4)
53defframe(SAVE_ESI,     -8)
54defframe(SAVE_EDI,    -12)
55deflit(STACK_SPACE, 12)
56
57          TEXT
58
59          ALIGN(16)
60PROLOGUE(mpn_modexact_1c_odd)
61deflit(`FRAME',0)
62
63          movl      PARAM_CARRY, %ecx
64          jmp       L(start_1c)
65
66EPILOGUE()
67
68          ALIGN(16)
69PROLOGUE(mpn_modexact_1_odd)
70deflit(`FRAME',0)
71
72          xorl      %ecx, %ecx
73L(start_1c):
74          movl      PARAM_DIVISOR, %eax
75
76          subl      $STACK_SPACE, %esp  FRAME_subl_esp(STACK_SPACE)
77
78          movl      %esi, SAVE_ESI
79          movl      PARAM_SRC, %esi
80
81          shrl      %eax                          C d/2
82          movl      %edi, SAVE_EDI
83
84          andl      $127, %eax
85
86ifdef(`PIC',`
87          LEA(      binvert_limb_table, %edi)
88          movzbl    (%eax,%edi), %edi             C inv 8 bits
89',`
90          movzbl    binvert_limb_table(%eax), %edi          C inv 8 bits
91')
92
93          xorl      %edx, %edx                    C initial extra carry
94          leal      (%edi,%edi), %eax   C 2*inv
95
96          imull     %edi, %edi                    C inv*inv
97
98          movl      %ebx, SAVE_EBX
99          movl      PARAM_SIZE, %ebx
100
101          imull     PARAM_DIVISOR, %edi C inv*inv*d
102
103          subl      %edi, %eax                    C inv = 2*inv - inv*inv*d
104          leal      (%eax,%eax), %edi   C 2*inv
105
106          imull     %eax, %eax                    C inv*inv
107
108          imull     PARAM_DIVISOR, %eax C inv*inv*d
109
110          leal      (%esi,%ebx,4), %esi C src end
111          negl      %ebx                          C -size
112
113          subl      %eax, %edi                    C inv = 2*inv - inv*inv*d
114
115          ASSERT(e,`          C d*inv == 1 mod 2^GMP_LIMB_BITS
116          movl      PARAM_DIVISOR, %eax
117          imull     %edi, %eax
118          cmpl      $1, %eax')
119
120
121C The dependent chain here is
122C
123C         subl      %edx, %eax       1
124C         imull     %edi, %eax       4
125C         mull      PARAM_DIVISOR    5
126C                                    ----
127C         total                         10
128C
129C and this is the measured speed.  No special scheduling is necessary, out
130C of order execution hides the load latency.
131
132L(top):
133          C eax     scratch (src limb)
134          C ebx     counter, limbs, negative
135          C ecx     carry bit, 0 or 1
136          C edx     carry limb, high of last product
137          C esi     &src[size]
138          C edi     inverse
139          C ebp
140
141          movl      (%esi,%ebx,4), %eax
142          subl      %ecx, %eax
143
144          sbbl      %ecx, %ecx
145          subl      %edx, %eax
146
147          sbbl      $0, %ecx
148
149          imull     %edi, %eax
150
151          negl      %ecx
152
153          mull      PARAM_DIVISOR
154
155          incl      %ebx
156          jnz       L(top)
157
158
159          movl      SAVE_ESI, %esi
160          leal      (%ecx,%edx), %eax
161
162          movl      SAVE_EDI, %edi
163
164          movl      SAVE_EBX, %ebx
165          addl      $STACK_SPACE, %esp
166
167          ret
168
169EPILOGUE()
170ASM_END()
171