用于收集任何spec cpu(2000/2006/2017)的数据
基于python的脚本，替代spec官方的runspec脚本，可以运行在spec的命令前加上各种前缀命令, 与config中的submit和use_submit_for_speed参数比较相似。
git clone https://github.com/rrwhx/runspec
编译好的spec cpu目录
保证 run 目录下生成三个文件夹分别对应test train ref
cd <spec_root>
cp run.sh .
cp setup_spec.sh .
./run.sh <config_file> test 1 all
# cpu 2017
# ./run.sh <config_file> test 1 intrate fprate -T base
./setup_spec.sh <config_file>- spec cpu2000
保证 run 目录下生成
00000001 00000002 00000003分别对应test train ref
# build
./run.sh <config_file> test 1 all
# setup run
rm -rf benchspec/*/*/run/
./run.sh <config_file> test 1 all
./run.sh <config_file> train 1 all
./run.sh <config_file> ref 1 all- spec cpu2006
保证 run 目录下生成
run_base_test_<config_file>.0000 run_base_train_<config_file>.0000 run_base_ref_<config_file>.0000分别对应test train ref
cd <spec_root>
# build
./run.sh <config_file> test 1 all
# setup run
rm -rf benchspec/CPU2006/*/run/
./run.sh <config_file> test 1 all -a setup
./run.sh <config_file> train 1 all -a setup
./run.sh <config_file> ref 1 all -a setup- spec cpu2017
保证 run 目录下生成
run_base_test_<config_file>.0000 run_base_train_<config_file>.0000 run_base_refrate_<config_file>.0000分别对应test train ref
source shrc
# build
./run.sh <config_file> test 1 intrate fprate -T base
# setup run
rm -rf benchspec/CPU/*/run/
./run.sh <config_file> test 1 -T base -a setup specrate
./run.sh <config_file> train 1 -T base -a setup specrate
./run.sh <config_file> ref 1 -T base -a setup specrateusage: runspec.py [-h] [-i {test,train,ref,refrate}] [-b BENCHMARK] [-T {base,peak}] [--ext EXT] [--suffix SUFFIX] [-t THREADS] [-l] [-n] [-v] [--intfp] [-c CMD_PREFIX] [--title TITLE]
[--stamp STAMP] [--result_dir RESULT_DIR] [--dir DIR] [--slimit SLIMIT]
Run spec cpu with prefix(none, qemu, perf, pin, dynamorio, strace, time), get log or performance,
Spec run directory should be prepared carefully,
Run test, train and ref in spec directory(00), or just -a setup(06,17),
options:
-h, --help show this help message and exit
-i {test,train,ref,refrate}, --size {test,train,ref,refrate}
-b BENCHMARK, --benchmark BENCHMARK
benchmark selection, all/int/fp, comma separated items
-T {base,peak}, --tune {base,peak}
--ext EXT auto probe, need not set
--suffix SUFFIX suffix of output file
-t THREADS, --threads THREADS
Allow N jobs at once;
-l, --loose ignore errors
-n, --dry_run Don't actually run any cmd; just print them.
-v, --verbose Print cmd before exec cmd
--intfp intfp prefix
-c CMD_PREFIX, --cmd_prefix CMD_PREFIX
cmd prefix before real cmd, %s for output, eg: -c "perf stat -o %s "
--title TITLE
--stamp STAMP
--result_dir RESULT_DIR
location of cmd_prefix logs, defaults to ~/runspec_result
--dir DIR
--slimit SLIMIT The limit of the stack size, 0 ulimited, or a number(MB), default: not modified./runspec.py -i ref -b all -t 4 \
--dir /2t/SPEC/SPEC2006/lxy/spec2006_x64_avx2/ \
--cmd_prefix "/usr/bin/time -f %%M -o %s " \
--title physical_memory_usagecd ~/runspec_result/physical_memory_usage_2006_ref_2023_04_24_09_36_48/
grep -r . . | sort
./400.perlbench_1:242852
./400.perlbench_2:309456
./400.perlbench_3:679740
./401.bzip2_1:870216
./401.bzip2_2:102740
./401.bzip2_3:102760
./401.bzip2_4:870216
./401.bzip2_5:870208
./401.bzip2_6:625140
./403.gcc_1:237160
./403.gcc_2:257476
./403.gcc_3:454100
..../runspec.py -i ref -b int \
--dir /2t/SPEC/SPEC2006/lxy/spec2006_x64_avx2/ \
--cmd_prefix "qemu-x86_64 -- " \
--title qemu_log./runspec.py -i ref -b int \
--dir /2t/SPEC/SPEC2006/lxy/spec2006_x64_avx2/./runspec.py -i ref -b int \
--dir /2t/SPEC/SPEC2006/lxy/spec2006_x64_avx2/ \
--cmd_prefix "taskset -c 8 perf record -F 100 -e instructions -o %s -- " \
--title perf_record_instructions命令执行失败，添加-v选项查看运行的目录和命令，手动运行查看原因。