1 Copyright 2001 Free Software Foundation, Inc.
3 This file is part of the GNU MP Library.
5 The GNU MP Library is free software; you can redistribute it and/or modify
6 it under the terms of either:
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.
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
18 or both in parallel, as here.
20 The GNU MP Library is distributed in the hope that it will be useful, but
21 WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
22 or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
25 You should have received copies of the GNU General Public License and the
26 GNU Lesser General Public License along with the GNU MP Library. If not,
27 see https://www.gnu.org/licenses/.
32 INTEL PENTIUM-4 MPN SUBROUTINES
35 This directory contains mpn functions optimized for Intel Pentium-4.
37 The mmx subdirectory has routines using MMX instructions, the sse2
38 subdirectory has routines using SSE2 instructions. All P4s have these, the
39 separate directories are just so configure can omit that code if the
40 assembler doesn't support it.
47 mpn_add_n/sub_n 4 normal, 6 in-place
49 mpn_mul_1 4 normal, 6 in-place
53 mpn_mul_basecase 6 cycles/crossproduct (approx)
55 mpn_sqr_basecase 3.5 cycles/crossproduct (approx)
56 or 7.0 cycles/triangleproduct (approx)
62 The shifts ought to be able to go at 1.5 c/l, but not much effort has been
65 In-place operations, and all addmul, submul, mul_basecase and sqr_basecase
66 calls, suffer from pipeline anomalies associated with write combining and
67 movd reads and writes to the same or nearby locations. The movq
68 instructions do not trigger the same hardware problems. Unfortunately,
69 using movq and splitting/combining seems to require too many extra
70 instructions to help. Perhaps future chip steppings will be better.
76 The Pentium-4 pipeline "Netburst", provides for quite a number of surprises.
77 Many traditional x86 instructions run very slowly, requiring use of
78 alterative instructions for acceptable performance.
80 adcl and sbbl are quite slow at 8 cycles for reg->reg. paddq of 32-bits
81 within a 64-bit mmx register seems better, though the combination
82 paddq/psrlq when propagating a carry is still a 4 cycle latency.
84 incl and decl should be avoided, instead use add $1 and sub $1. Apparently
85 the carry flag is not separately renamed, so incl and decl depend on all
86 previous flags-setting instructions.
88 shll and shrl have a 4 cycle latency, or 8 times the latency of the fastest
89 integer instructions (addl, subl, orl, andl, and some more). shldl and
90 shrdl seem to have 13 and 15 cycles latency, respectively. Bizarre.
92 movq mmx -> mmx does have 6 cycle latency, as noted in the documentation.
93 pxor/por or similar combination at 2 cycles latency can be used instead.
94 The movq however executes in the float unit, thereby saving MMX execution
95 resources. With the right juggling, data moves shouldn't be on a dependent
98 L1 is write-through, but the write-combining sounds like it does enough to
99 not require explicit destination prefetching.
101 xmm registers so far haven't found a use, but not much effort has been
102 expended. A configure test for whether the operating system knows
103 fxsave/fxrestor will be needed if they're used.
109 Intel Pentium-4 processor manuals,
111 http://developer.intel.com/design/pentium4/manuals
113 "Intel Pentium 4 Processor Optimization Reference Manual", Intel, 2001,
114 order number 248966. Available on-line:
116 http://developer.intel.com/design/pentium4/manuals/248966.htm