xref: /netbsd-src/external/lgpl3/gmp/dist/mpn/mips64/README (revision 62f324d0121177eaf2e0384f92fd9ca2a751c795)
1Copyright 1996 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 the GNU Lesser General Public License as published by
7the Free Software Foundation; either version 3 of the License, or (at your
8option) any later version.
9
10The GNU MP Library is distributed in the hope that it will be useful, but
11WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
12or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU Lesser General Public
13License for more details.
14
15You should have received a copy of the GNU Lesser General Public License
16along with the GNU MP Library.  If not, see http://www.gnu.org/licenses/.
17
18
19
20
21
22This directory contains mpn functions optimized for MIPS3.  Example of
23processors that implement MIPS3 are R4000, R4400, R4600, R4700, and R8000.
24
25RELEVANT OPTIMIZATION ISSUES
26
271. On the R4000 and R4400, branches, both the plain and the "likely" ones,
28   take 3 cycles to execute.  (The fastest possible loop will take 4 cycles,
29   because of the delay insn.)
30
31   On the R4600, branches takes a single cycle
32
33   On the R8000, branches often take no noticable cycles, as they are
34   executed in a separate function unit..
35
362. The R4000 and R4400 have a load latency of 4 cycles.
37
383. On the R4000 and R4400, multiplies take a data-dependent number of
39   cycles, contrary to the SGI documentation.  There seem to be 3 or 4
40   possible latencies.
41
424. The R1x000 processors can issue one floating-point operation, two integer
43   operations, and one memory operation per cycle.  The FPU has very short
44   latencies, while the integer multiply unit is non-pipelined.  We should
45   therefore write fp based mpn_Xmul_1.
46
47STATUS
48
49Good...
50