|
145
|
1 #!/bin/bash
|
|
|
2
|
|
|
3 # Script to measure memset and memcpy for different sizes and strategies.
|
|
|
4 #
|
|
|
5 # Contributed by Jan Hubicka <jh@suse.cz>
|
|
|
6 #
|
|
|
7 # Copyright (C) 2019 Free Software Foundation, Inc.
|
|
|
8 #
|
|
|
9 # This file is part of GCC.
|
|
|
10 #
|
|
|
11 # GCC is free software; you can redistribute it and/or modify
|
|
|
12 # it under the terms of the GNU General Public License as published by
|
|
|
13 # the Free Software Foundation; either version 3, or (at your option)
|
|
|
14 # any later version.
|
|
|
15 #
|
|
|
16 # GCC is distributed in the hope that it will be useful,
|
|
|
17 # but WITHOUT ANY WARRANTY; without even the implied warranty of
|
|
|
18 # MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
|
|
|
19 # GNU General Public License for more details.
|
|
|
20 #
|
|
|
21 # You should have received a copy of the GNU General Public License
|
|
|
22 # along with GCC; see the file COPYING. If not, write to
|
|
|
23 # the Free Software Foundation, 51 Franklin Street, Fifth Floor,
|
|
|
24 # Boston, MA 02110-1301, USA.
|
|
|
25
|
|
|
26 # This script will search a line starting with 'spawn' that includes the
|
|
|
27 # pattern you are looking for (typically a source file name).
|
|
|
28 #
|
|
|
29 # Once it finds that pattern, it re-executes the whole command
|
|
|
30 # in the spawn line. If the pattern matches more than one spawn
|
|
|
31 # command, it asks which one you want.
|
|
|
32
|
|
|
33 test()
|
|
|
34 {
|
|
|
35 rm -f a.out
|
|
|
36 cat <<END | $1 -x c -O3 $3 -DAVG_SIZE=$2 $STRINGOP -DMEMORY_COPIES=$memsize -
|
|
|
37 #define BUFFER_SIZE (16*1024*1024 + AVG_SIZE*2)
|
|
|
38 /*#define MEMORY_COPIES (1024*1024*64*(long long)10)*/
|
|
|
39 $type t[BUFFER_SIZE];
|
|
|
40 int main()
|
|
|
41 {
|
|
|
42 unsigned int i;
|
|
|
43 for (i=0;i<((long long)MEMORY_COPIES + AVG_SIZE * 2 - 1)/AVG_SIZE*2;i++)
|
|
|
44 #ifdef test_memset
|
|
|
45 __builtin_memset (t+(i*1024*1024+i*1)%(BUFFER_SIZE - AVG_SIZE*2), i, (AVG_SIZE + i) % (AVG_SIZE * 2 + 0));
|
|
|
46 #else
|
|
|
47 __builtin_memcpy (t+(i*1024*1024+i*1)%(BUFFER_SIZE - AVG_SIZE*2), t+((i+1)*1024*1024*4+i*1)%(BUFFER_SIZE - AVG_SIZE *2), (AVG_SIZE + i) % (AVG_SIZE * 2 + 0));
|
|
|
48 #endif
|
|
|
49 return 0;
|
|
|
50 }
|
|
|
51 END
|
|
|
52 TIME=`/usr/bin/time -f "%E" ./a.out 2>&1`
|
|
|
53 echo -n " "$TIME
|
|
|
54 echo $TIME $4 >>/tmp/accum
|
|
|
55 }
|
|
|
56
|
|
|
57 test2()
|
|
|
58 {
|
|
|
59 rm -f a.out
|
|
|
60 cat <<END | clang -x c -O3 $3 -DAVG_SIZE=$2 $STRINGOP -DMEMORY_COPIES=$memsize 2>/dev/null -
|
|
|
61 #define BUFFER_SIZE (16*1024*1024 + AVG_SIZE*2)
|
|
|
62 /*#define MEMORY_COPIES (1024*1024*64*(long long)10)*/
|
|
|
63 $type t[BUFFER_SIZE];
|
|
|
64 int main()
|
|
|
65 {
|
|
|
66 unsigned int i;
|
|
|
67 for (i=0;i<((long long)MEMORY_COPIES + AVG_SIZE * 2 - 1)/AVG_SIZE*2;i++)
|
|
|
68 #ifdef test_memset
|
|
|
69 __builtin_memset (t+(i*1024*1024+i*1)%(BUFFER_SIZE - AVG_SIZE*2), i, (AVG_SIZE + i) % (AVG_SIZE * 2 + 0));
|
|
|
70 #else
|
|
|
71 __builtin_memcpy (t+(i*1024*1024+i*1)%(BUFFER_SIZE - AVG_SIZE*2), t+((i+1)*1024*1024*4+i*1)%(BUFFER_SIZE - AVG_SIZE *2), (AVG_SIZE + i) % (AVG_SIZE * 2 + 0));
|
|
|
72 #endif
|
|
|
73 return 0;
|
|
|
74 }
|
|
|
75 END
|
|
|
76 TIME=`/usr/bin/time -f "%E" ./a.out 2>&1`
|
|
|
77 echo -n " "$TIME
|
|
|
78 echo $TIME $4 >>/tmp/accum
|
|
|
79 }
|
|
|
80
|
|
|
81 testrow()
|
|
|
82 {
|
|
|
83 echo -n "" >/tmp/accum
|
|
|
84 printf "%12i " $3
|
|
|
85 test "$2" "$3" "-mstringop-strategy=libcall" libcall
|
|
|
86 test "$2" "$3" "-mstringop-strategy=rep_byte -malign-stringops" rep1
|
|
|
87 test "$2" "$3" "-mstringop-strategy=rep_byte -mno-align-stringops" rep1noalign
|
|
|
88 test "$2" "$3" "-mstringop-strategy=rep_4byte -malign-stringops" rep4
|
|
|
89 test "$2" "$3" "-mstringop-strategy=rep_4byte -mno-align-stringops" rep4noalign
|
|
|
90 if [ "$mode" == 64 ]
|
|
|
91 then
|
|
|
92 test "$2" "$3" "-mstringop-strategy=rep_8byte -malign-stringops" rep8
|
|
|
93 test "$2" "$3" "-mstringop-strategy=rep_8byte -mno-align-stringops" rep8noalign
|
|
|
94 fi
|
|
|
95 test "$2" "$3" "-mstringop-strategy=loop -malign-stringops" loop
|
|
|
96 test "$2" "$3" "-mstringop-strategy=loop -mno-align-stringops" loopnoalign
|
|
|
97 test "$2" "$3" "-mstringop-strategy=unrolled_loop -malign-stringops" unrl
|
|
|
98 test "$2" "$3" "-mstringop-strategy=unrolled_loop -mno-align-stringops" unrlnoalign
|
|
|
99 test "$2" "$3" "-mstringop-strategy=vector_loop -malign-stringops" sse
|
|
|
100 test "$2" "$3" "-mstringop-strategy=vector_loop -mno-align-stringops -msse2" ssenoalign
|
|
|
101 #test2 "$2" "$3" ""
|
|
|
102 test "$2" "$3" "-mstringop-strategy=byte_loop" byte
|
|
|
103 best=`cat /tmp/accum | sort | head -1`
|
|
|
104 test "$2" "$3" " -fprofile-generate" >/dev/null 2>&1
|
|
|
105 test "$2" "$3" " -fprofile-use"
|
|
|
106 test "$2" "$3" " -minline-stringops-dynamically"
|
|
|
107 echo " $best"
|
|
|
108 }
|
|
|
109
|
|
|
110 test_all_sizes()
|
|
|
111 {
|
|
|
112 if [ "$mode" == 64 ]
|
|
|
113 then
|
|
|
114 echo " block size libcall rep1 noalg rep4 noalg rep8 noalg loop noalg unrl noalg sse noalg byte PGO dynamic BEST"
|
|
|
115 else
|
|
|
116 echo " block size libcall rep1 noalg rep4 noalg loop noalg unrl noalg sse noalg byte PGO dynamic BEST"
|
|
|
117 fi
|
|
|
118 #for size in 1 2 3 4 6 8 10 12 14 16 24 32 48 64 128 256 512 1024 4096 8192 81920 819200 8192000
|
|
|
119 #for size in 8192000 819200 81920 8192 4096 2048 1024 512 256 128 64 48 32 24 16 14 12 10 8 6 5 4 3 2 1
|
|
|
120 for size in 8192000 819200 81920 20480 8192 4096 2048 1024 512 256 128 64 48 32 24 16 14 12 10 8 6 4 1
|
|
|
121 #for size in 128 256 1024 4096 8192 81920 819200
|
|
|
122 do
|
|
|
123 testrow "$1" "$2" $size
|
|
|
124 done
|
|
|
125 }
|
|
|
126
|
|
|
127 mode=$1
|
|
|
128 shift
|
|
|
129 export memsize=$1
|
|
|
130 shift
|
|
|
131 cmdline=$*
|
|
|
132 if [ "$mode" != 32 ]
|
|
|
133 then
|
|
|
134 if [ "$mode" != 64 ]
|
|
|
135 then
|
|
|
136 echo "Usage:"
|
|
|
137 echo "test_stringop mode size cmdline"
|
|
|
138 echo "mode is either 32 or 64"
|
|
|
139 echo "size is amount of memory copied in each test. Should be chosed small enough so runtime is less than minute for each test and sorting works"
|
|
|
140 echo "Example: test_stringop 32 640000000 ./xgcc -B ./ -march=pentium3"
|
|
|
141 exit
|
|
|
142 fi
|
|
|
143 fi
|
|
|
144
|
|
|
145 echo "memcpy"
|
|
|
146 export STRINGOP=""
|
|
|
147 type=char
|
|
|
148 test_all_sizes $mode "$cmdline -m$mode"
|
|
|
149 echo "Aligned"
|
|
|
150 type=long
|
|
|
151 test_all_sizes $mode "$cmdline -m$mode"
|
|
|
152 echo "memset"
|
|
|
153 export STRINGOP="-Dtest_memset=1"
|
|
|
154 type=char
|
|
|
155 test_all_sizes $mode "$cmdline -m$mode"
|
|
|
156 echo "Aligned"
|
|
|
157 type=long
|
|
|
158 test_all_sizes $mode "$cmdline -m$mode"
|
