README
1Copyright 1996, 1999, 2001, 2002, 2004 Free Software Foundation, Inc.
2
3This file is part of the GNU MP Library.
4
5The GNU MP Library is free software; you can redistribute it and/or modify
6it under the terms of either:
7
8 * the GNU Lesser General Public License as published by the Free
9 Software Foundation; either version 3 of the License, or (at your
10 option) any later version.
11
12or
13
14 * the GNU General Public License as published by the Free Software
15 Foundation; either version 2 of the License, or (at your option) any
16 later version.
17
18or both in parallel, as here.
19
20The GNU MP Library is distributed in the hope that it will be useful, but
21WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
22or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
23for more details.
24
25You should have received copies of the GNU General Public License and the
26GNU Lesser General Public License along with the GNU MP Library. If not,
27see https://www.gnu.org/licenses/.
28
29
30
31
32
33
34This directory contains mpn functions for various HP PA-RISC chips. Code
35that runs faster on the PA7100 and later implementations, is in the pa7100
36directory.
37
38RELEVANT OPTIMIZATION ISSUES
39
40 Load and Store timing
41
42On the PA7000 no memory instructions can issue the two cycles after a store.
43For the PA7100, this is reduced to one cycle.
44
45The PA7100 has a lookup-free cache, so it helps to schedule loads and the
46dependent instruction really far from each other.
47
48STATUS
49
501. mpn_mul_1 could be improved to 6.5 cycles/limb on the PA7100, using the
51 instructions below (but some sw pipelining is needed to avoid the
52 xmpyu-fstds delay):
53
54 fldds s1_ptr
55
56 xmpyu
57 fstds N(%r30)
58 xmpyu
59 fstds N(%r30)
60
61 ldws N(%r30)
62 ldws N(%r30)
63 ldws N(%r30)
64 ldws N(%r30)
65
66 addc
67 stws res_ptr
68 addc
69 stws res_ptr
70
71 addib Loop
72
732. mpn_addmul_1 could be improved from the current 10 to 7.5 cycles/limb
74 (asymptotically) on the PA7100, using the instructions below. With proper
75 sw pipelining and the unrolling level below, the speed becomes 8
76 cycles/limb.
77
78 fldds s1_ptr
79 fldds s1_ptr
80
81 xmpyu
82 fstds N(%r30)
83 xmpyu
84 fstds N(%r30)
85 xmpyu
86 fstds N(%r30)
87 xmpyu
88 fstds N(%r30)
89
90 ldws N(%r30)
91 ldws N(%r30)
92 ldws N(%r30)
93 ldws N(%r30)
94 ldws N(%r30)
95 ldws N(%r30)
96 ldws N(%r30)
97 ldws N(%r30)
98 addc
99 addc
100 addc
101 addc
102 addc %r0,%r0,cy-limb
103
104 ldws res_ptr
105 ldws res_ptr
106 ldws res_ptr
107 ldws res_ptr
108 add
109 stws res_ptr
110 addc
111 stws res_ptr
112 addc
113 stws res_ptr
114 addc
115 stws res_ptr
116
117 addib
118
1193. For the PA8000 we have to stick to using 32-bit limbs before compiler
120 support emerges. But we want to use 64-bit operations whenever possible,
121 in particular for loads and stores. It is possible to handle mpn_add_n
122 efficiently by rotating (when s1/s2 are aligned), masking+bit field
123 inserting when (they are not). The speed should double compared to the
124 code used today.
125
126
127
128
129LABEL SYNTAX
130
131The HP-UX assembler takes labels starting in column 0 with no colon,
132
133 L$loop ldws,mb -4(0,%r25),%r22
134
135Gas on hppa GNU/Linux however requires a colon,
136
137 L$loop: ldws,mb -4(0,%r25),%r22
138
139This is covered by using LDEF() from asm-defs.m4. An alternative would be
140to use ".label" which is accepted by both,
141
142 .label L$loop
143 ldws,mb -4(0,%r25),%r22
144
145but that's not as nice to look at, not if you're used to assembler code
146having labels in column 0.
147
148
149
150
151REFERENCES
152
153Hewlett Packard, "HP Assembler Reference Manual", 9th edition, June 1998,
154part number 92432-90012.
155
156
157
158----------------
159Local variables:
160mode: text
161fill-column: 76
162End:
163