source: doc/theses/colby_parsons_MMAth/benchmarks/waituntil/run@ b28ce93

Last change on this file since b28ce93 was 14e1053, checked in by caparsons <caparson@…>, 2 years ago

first draft of full waituntil chapter and conclusion chapter. Lots of graph/plotting utilities cleanup. Reran all CFA actor benchmarks after recent changes. Small changes to actor.tex in performance section

  • Property mode set to 100755
File size: 8.3 KB
Line 
1#!/bin/bash -
2
3false=0; true=1
4
5# Usage: arch [ hostname ] returns hostname, cores, startcore
6#
7# Define machine architecture based on starting socket, CPUs (cores) per socket, number of
8# sockets, has hyperthreading.
9
10start=0
11
12arch() {
13 hostname=${1:-`hostname`} # return value
14 hashyper=${true} # assume machine has hyperthreads
15 if [ "${hostname}" = "plg2" ] ; then
16 startsocket=${start}
17 cps=16 # coresPerSocket
18 sockets=2
19 hashyper=${false} # has no hyperthreads
20 elif [ "${hostname}" = "nasus" ] ; then
21 startsocket=${start}
22 cps=64 # coresPerSocket
23 sockets=2
24 elif [ "${hostname}" = "pyke" ] ; then
25 startsocket=${start}
26 cps=24 # coresPerSocket
27 sockets=2
28 elif [ "${hostname}" = "jax" ] ; then
29 startsocket=${start}
30 cps=24 # coresPerSocket
31 sockets=4
32 else
33 echo "unsupported host" ${hostname}
34 exit 1
35 fi
36 cores=$(( ${cps} * ${sockets} ))
37 startcore=$(( ${startsocket} * ${cps} ))
38}
39
40# Usage: affinity (global cps, sockets, startsocket, hashyper, cores, startcore, wrap)
41# returns taskset argument
42#
43# This routine assumes hyperthreading has only 2 hyperthreads per core.
44#
45# If hyperthread scanning is used: processor units are assigned across the low-number hyperthreads
46# of the socket's cores. When the low-number hyperthreads are filled, the high-number hyperhtreads
47# are assigned across the socket's cores. Then the next socket is assigned.
48#
49# If hyperthread wrapping is used: processor units are assigned in low/high-number pairs of
50# hyperthreads across the socket's cores. Then the next socket is assigned.
51
52wrap=${false} # set to control hyperthread assignment across socket cores
53
54affinity() {
55 if [ ${wrap} -eq ${true} -a ${hashyper} -eq ${false} ] ; then
56 echo "architecture does not support hyperthreading for wrapping"
57 exit 1
58 fi
59 taskset="" # return value
60 set -- $(( ${1} - 1 )) # decrement $1
61 if [ ${1} -eq 0 ] ; then taskset="${startcore}-${startcore}"; return; fi
62 if [ ${1} -ge $(( ${cps} * ( ${sockets} - ${startsocket} ) * ( ${hashyper} + 1 ) )) ] ; then # error
63 echo "not enough cores $(( ${cores} * ${sockets} )) for $(( ${1} + 1 )) starting at ${startcore}"
64 exit 1
65 fi
66 if [ ${hashyper} -eq ${false} ] ; then taskset="${startcore}-$(( ${1} + ${startcore} ))"; return; fi # no hyperthreads
67 start2=$(( ${startcore} + ${cores} ))
68 if [ ${wrap} -eq ${true} ] ; then # hyperthread wrapping
69 end1=$(( ${1} / 2 + ${startcore} ))
70 end2=$(( ${end1} + ${cores} ))
71 if [ $(( ${1} % 2 )) -eq 0 ] ; then
72 end2=$(( ${end2} - 1 ))
73 fi
74 taskset="${startcore}-${end1},${start2}-${end2}"
75 else # hyperthread scanning
76 if [ ${1} -lt ${cps} ] ; then taskset="${startcore}-$(( ${1} + ${startcore} ))"; return; fi
77 filled=$(( ${1} / ( ${cps} * 2 ) * ${cps} ))
78 modulus=$(( ${1} % ( ${cps} * 2 ) )) # leftover cores added to saturated sockets
79 if [ ${modulus} -gt ${cps} ] ; then
80 taskset="${startcore}-$(( ${startcore} + ${filled} + ${cps} - 1 )),${start2}-$(( ${start2} + ${filled} + ${modulus} % ${cps} ))"
81 else
82 taskset="${startcore}-$(( ${startcore} + ${filled} + ${modulus} )),${start2}-$(( ${start2} + ${filled} - 1 ))"
83 fi
84 fi
85}
86
87numtimes=5
88# numtimes=1
89
90# num_threads='2 4 8 16 24 32'
91side_chan_threads='6 12 18 24 30' # must be mults of 6
92num_threads='2'
93# side_chan_threads='6'
94
95chan_size='10'
96future_time='10'
97future_flags=('-DOR' '-DAND3' '-DANDOR' '-DORAND')
98future_names=('OR' 'AND' 'ANDOR' 'ORAND')
99
100# toggle benchmarks
101spin=${true}
102contend=${true}
103sidechan=${true}
104future=${true}
105order=${true}
106spin=${false}
107contend=${false}
108sidechan=${false}
109future=${false}
110# order=${false}
111
112runCFA=${true}
113runGO=${true}
114# runUCPP=${true}
115# runCFA=${false}
116# runGO=${false}
117runUCPP=${false}
118
119cfa=~/cfa-cc/driver/cfa
120
121# Helpers to minimize code duplication
122
123# repeats a command ${numtimes}
124preprint=''
125repeat_command() {
126 t=1
127 while [ ${t} -le ${numtimes} ] ; do
128 echo -n -e ${preprint}
129 "${@}"
130 t=`expr ${t} + 1`
131 done
132}
133
134# prints the leading info for a given run of a variant
135print_header() {
136 echo ${1}':'
137 echo -e "cores\tthroughput (entries)"
138}
139
140# runs the current benchmark with provided args
141# only works for standard-run benchmarks (not Akka)
142# must split into pre and post args to be able to supply val of p
143pre_args=''
144post_args=''
145single_run() {
146 affinity ${1}
147 preprint="${1}\t"
148 repeat_command taskset -c ${taskset} ./a.${hostname} ${pre_args} ${1} ${post_args}
149}
150
151# runs the current bench for all processor vals
152# works for standard benchs that dont need to set a config file (not Akka or CAF)
153run_bench() {
154 for p in ${num_threads} ; do
155 single_run ${p}
156 done
157}
158
159run_side_chan() {
160 i=1
161 for p in ${side_chan_threads} ; do
162 affinity ${p}
163 preprint="${p}\t"
164 repeat_command taskset -c ${taskset} ./a.${hostname} ${pre_args} ${i} ${post_args}
165 i=`expr ${i} + 1`
166 done
167}
168
169run_order() {
170 affinity 4
171 preprint="4\t"
172 repeat_command taskset -c ${taskset} ./a.${hostname}
173}
174
175run_future() {
176 affinity 2
177 preprint="2\t"
178 repeat_command taskset -c ${taskset} ./a.${hostname} ${post_args}
179}
180
181arch # get hostname
182
183# set up leading info for python script
184echo $numtimes
185echo $num_threads
186echo $side_chan_threads
187
188if [ ${runCFA} -eq ${true} ]; then
189 echo -n 'CFA '
190fi
191if [ ${runGO} -eq ${true} ]; then
192 echo -n 'Go '
193fi
194echo ""
195
196# done printing header info for output
197
198# cfa flags
199cfa_flags='-quiet -O3 -nodebug -DNDEBUG'
200
201# UCPP flags
202UCPPflags="-quiet -g -Wall -Wextra -O3 -nodebug -DNDEBUG -multi"
203UCPP=~/ucpp/u++-7.0.0/bin/u++
204
205# run the benchmarks
206
207run_contend() {
208 post_args=${1}
209
210 if [ ${runCFA} -eq ${true} ] ; then
211 cd cfa # CFA RUN
212 print_header 'CFA'
213 ${cfa} ${cfa_flags} '-DNUM_CHANS='${3} ${2}.cfa -o a.${hostname} > /dev/null 2>&1
214 run_bench
215 rm a.${hostname}
216 cd - > /dev/null
217 fi # done CFA
218
219 if [ ${runGO} -eq ${true} ] ; then
220 cd go/${2}${3} # Go RUN
221 print_header 'Go'
222 go build -o a.${hostname} > /dev/null 2>&1
223 run_bench
224 rm a.${hostname}
225 cd - > /dev/null
226 fi # done Go
227}
228
229# /usr/bin/time -f "%Uu %Ss %Er %Mkb"
230if [ ${contend} -eq ${true} ] ; then
231 echo "contend2: "
232 run_contend ${chan_size} 'contend' '2'
233 echo "contend4: "
234 run_contend ${chan_size} 'contend' '4'
235 echo "contend8: "
236 run_contend ${chan_size} 'contend' '8'
237fi
238
239if [ ${spin} -eq ${true} ] ; then
240 echo "spin2: "
241 run_contend ${chan_size} 'spin' '2'
242 echo "spin4: "
243 run_contend ${chan_size} 'spin' '4'
244 echo "spin8: "
245 run_contend ${chan_size} 'spin' '8'
246fi
247
248if [ ${sidechan} -eq ${true} ] ; then
249 echo "sidechan: "
250 post_args=${chan_size}
251 if [ ${runCFA} -eq ${true} ] ; then
252 cd cfa # CFA RUN
253 print_header 'CFA'
254 ${cfa} ${cfa_flags} sidechan.cfa -o a.${hostname} > /dev/null 2>&1
255 run_side_chan
256 rm a.${hostname}
257 cd - > /dev/null
258 fi # done CFA
259
260 if [ ${runGO} -eq ${true} ] ; then
261 cd go/sidechan
262 print_header 'Go'
263 go build -o a.${hostname} > /dev/null 2>&1
264 run_side_chan
265 rm a.${hostname}
266 cd - > /dev/null
267 fi # done Go
268fi
269
270if [ ${future} -eq ${true} ] ; then
271 post_args=${future_time}
272 for i in ${!future_flags[@]}; do
273 echo 'future '${future_names[$i]}':'
274 if [ ${runCFA} -eq ${true} ] ; then
275 cd cfa # CFA RUN
276 print_header 'CFA'
277 ${cfa} ${cfa_flags} ${future_flags[$i]} future.cfa -o a.${hostname} > /dev/null 2>&1
278 run_future
279 rm a.${hostname}
280 cd - > /dev/null
281 fi # done CFA
282
283 if [ ${runUCPP} -eq ${true} ] ; then
284 cd ucpp
285 print_header 'uC++'
286 ${UCPP} ${UCPPflags} ${future_flags[$i]} future.cc -o a.${hostname} > /dev/null 2>&1
287 run_future
288 rm a.${hostname}
289 cd - > /dev/null
290 fi # done Go
291 done
292fi
293
294if [ ${order} -eq ${true} ] ; then
295 echo "order: "
296 post_args=${chan_size}
297 if [ ${runCFA} -eq ${true} ] ; then
298 cd cfa # CFA RUN
299 print_header 'CFA'
300 ${cfa} ${cfa_flags} order.cfa -o a.${hostname} > /dev/null 2>&1
301 run_order
302 rm a.${hostname}
303 cd - > /dev/null
304 fi # done CFA
305
306 if [ ${runGO} -eq ${true} ] ; then
307 cd go/order
308 print_header 'Go'
309 go build -o a.${hostname} > /dev/null 2>&1
310 run_order
311 rm a.${hostname}
312 cd - > /dev/null
313 fi # done Go
314fi
Note: See TracBrowser for help on using the repository browser.