1 | #include "rq_bench.hfa" |
---|
2 | |
---|
3 | struct Result { |
---|
4 | uint64_t count; |
---|
5 | uint64_t dmigs; |
---|
6 | uint64_t gmigs; |
---|
7 | }; |
---|
8 | |
---|
9 | // ================================================== |
---|
10 | thread __attribute__((aligned(128))) MyThread { |
---|
11 | struct MyData * volatile data; |
---|
12 | |
---|
13 | struct { |
---|
14 | struct MySpot ** ptr; |
---|
15 | size_t len; |
---|
16 | } spots; |
---|
17 | |
---|
18 | bench_sem sem; |
---|
19 | |
---|
20 | Result result; |
---|
21 | |
---|
22 | bool share; |
---|
23 | size_t cnt; |
---|
24 | processor * ttid; |
---|
25 | size_t id; |
---|
26 | }; |
---|
27 | |
---|
28 | uint64_t moved(MyThread & this, processor * ttid) { |
---|
29 | if(this.ttid == ttid) { |
---|
30 | return 0; |
---|
31 | } |
---|
32 | this.ttid = ttid; |
---|
33 | return 1; |
---|
34 | } |
---|
35 | |
---|
36 | // ================================================== |
---|
37 | struct __attribute__((aligned(128))) MyData { |
---|
38 | uint64_t _p1[16]; // padding |
---|
39 | uint64_t * data; |
---|
40 | size_t len; |
---|
41 | processor * ttid; |
---|
42 | size_t id; |
---|
43 | uint64_t _p2[16]; // padding |
---|
44 | }; |
---|
45 | |
---|
46 | void ?{}(MyData & this, size_t id, size_t size) { |
---|
47 | this.len = size; |
---|
48 | this.data = alloc(this.len, 128`align); |
---|
49 | this.ttid = active_processor(); |
---|
50 | this.id = id; |
---|
51 | |
---|
52 | for(i; this.len) { |
---|
53 | this.data[i] = 0; |
---|
54 | } |
---|
55 | } |
---|
56 | |
---|
57 | uint64_t moved(MyData & this, processor * ttid) { |
---|
58 | if(this.ttid == ttid) { |
---|
59 | return 0; |
---|
60 | } |
---|
61 | this.ttid = ttid; |
---|
62 | return 1; |
---|
63 | } |
---|
64 | |
---|
65 | __attribute__((noinline)) void access(MyData & this, size_t idx) { |
---|
66 | size_t l = this.len; |
---|
67 | this.data[idx % l] += 1; |
---|
68 | } |
---|
69 | |
---|
70 | // ================================================== |
---|
71 | // Atomic object where a single thread can wait |
---|
72 | // May exchanges data |
---|
73 | struct __attribute__((aligned(128))) MySpot { |
---|
74 | MyThread * volatile ptr; |
---|
75 | size_t id; |
---|
76 | uint64_t _p1[16]; // padding |
---|
77 | }; |
---|
78 | |
---|
79 | void ?{}(MySpot & this, size_t id) { |
---|
80 | this.ptr = 0p; |
---|
81 | this.id = id; |
---|
82 | } |
---|
83 | |
---|
84 | // Main handshake of the code |
---|
85 | // Single seat, first thread arriving waits |
---|
86 | // Next threads unblocks current one and blocks in its place |
---|
87 | // if share == true, exchange data in the process |
---|
88 | bool put( MySpot & this, MyThread & ctx, MyData * data, bool share) { |
---|
89 | // Attempt to CAS our context into the seat |
---|
90 | for() { |
---|
91 | MyThread * expected = this.ptr; |
---|
92 | if (expected == 1p) { // Seat is closed, return |
---|
93 | return true; |
---|
94 | } |
---|
95 | |
---|
96 | if (__atomic_compare_exchange_n(&this.ptr, &expected, &ctx, false, __ATOMIC_SEQ_CST, __ATOMIC_SEQ_CST)) { |
---|
97 | if(expected) { |
---|
98 | if(share) { |
---|
99 | expected->data = data; |
---|
100 | } |
---|
101 | post( expected->sem ); |
---|
102 | } |
---|
103 | break; // We got the seat |
---|
104 | } |
---|
105 | } |
---|
106 | |
---|
107 | // Block once on the seat |
---|
108 | wait(ctx.sem); |
---|
109 | |
---|
110 | // Someone woke us up, get the new data |
---|
111 | return false; |
---|
112 | } |
---|
113 | |
---|
114 | // Shutdown the spot |
---|
115 | // Wake current thread and mark seat as closed |
---|
116 | void release( MySpot & this ) { |
---|
117 | MyThread * val = __atomic_exchange_n(&this.ptr, 1p, __ATOMIC_SEQ_CST); |
---|
118 | if (!val) { |
---|
119 | return; |
---|
120 | } |
---|
121 | |
---|
122 | // Someone was there, release them |
---|
123 | post( val->sem ); |
---|
124 | } |
---|
125 | |
---|
126 | // ================================================== |
---|
127 | // Do some work by accessing 'cnt' cells in the array |
---|
128 | __attribute__((noinline)) void work(MyData & data, size_t cnt_, uint64_t & state) { |
---|
129 | for (cnt_) { |
---|
130 | access(data, __xorshift64(state)); |
---|
131 | } |
---|
132 | } |
---|
133 | |
---|
134 | void main(MyThread & this) { |
---|
135 | uint64_t state = thread_rand(); |
---|
136 | |
---|
137 | // Wait for start |
---|
138 | wait(this.sem); |
---|
139 | |
---|
140 | // Main loop |
---|
141 | for() { |
---|
142 | // Touch our current data, write to invalidate remote cache lines |
---|
143 | work(*this.data, this.cnt, state); |
---|
144 | |
---|
145 | // Wait on a random spot |
---|
146 | uint64_t idx = __xorshift64(state) % this.spots.len; |
---|
147 | bool closed = put(*this.spots.ptr[idx], this, this.data, this.share); |
---|
148 | |
---|
149 | // Check if the experiment is over |
---|
150 | if (closed) break; |
---|
151 | if ( clock_mode && stop) break; |
---|
152 | if (!clock_mode && this.result.count >= stop_count) break; |
---|
153 | |
---|
154 | // Check everything is consistent |
---|
155 | verify(this.data); |
---|
156 | |
---|
157 | // write down progress and check migrations |
---|
158 | processor * ttid = active_processor(); |
---|
159 | this.result.count += 1; |
---|
160 | this.result.gmigs += moved(this, ttid); |
---|
161 | this.result.dmigs += moved(*this.data, ttid); |
---|
162 | } |
---|
163 | |
---|
164 | __atomic_fetch_add(&threads_left, -1, __ATOMIC_SEQ_CST); |
---|
165 | } |
---|
166 | |
---|
167 | void ?{}( MyThread & this, MyData * data, MySpot ** spots, size_t spot_len, size_t cnt, bool share, size_t id) { |
---|
168 | ((thread&)this){ bench_cluster }; |
---|
169 | this.data = data; |
---|
170 | this.spots.ptr = spots; |
---|
171 | this.spots.len = spot_len; |
---|
172 | (this.sem){}; |
---|
173 | this.result.count = 0; |
---|
174 | this.result.gmigs = 0; |
---|
175 | this.result.dmigs = 0; |
---|
176 | this.share = share; |
---|
177 | this.cnt = cnt; |
---|
178 | this.ttid = active_processor(); |
---|
179 | this.id = id; |
---|
180 | } |
---|
181 | |
---|
182 | // ================================================== |
---|
183 | int main(int argc, char * argv[]) { |
---|
184 | unsigned wsize = 2; |
---|
185 | unsigned wcnt = 2; |
---|
186 | unsigned nspots = 0; |
---|
187 | bool share = false; |
---|
188 | cfa_option opt[] = { |
---|
189 | BENCH_OPT, |
---|
190 | { 'n', "nspots", "Number of spots where threads sleep (nthreads - nspots are active at the same time)", nspots}, |
---|
191 | { 'w', "worksize", "Size of the array for each threads, in words (64bit)", wsize}, |
---|
192 | { 'c', "workcnt" , "Number of words to touch when working (random pick, cells can be picked more than once)", wcnt }, |
---|
193 | { 's', "share" , "Pass the work data to the next thread when blocking", share, parse_truefalse } |
---|
194 | }; |
---|
195 | BENCH_OPT_PARSE("cforall cycle benchmark"); |
---|
196 | |
---|
197 | unsigned long long global_count = 0; |
---|
198 | unsigned long long global_gmigs = 0; |
---|
199 | unsigned long long global_dmigs = 0; |
---|
200 | |
---|
201 | if( nspots == 0 ) { nspots = nthreads - nprocs; } |
---|
202 | if( nspots == 0 ) { |
---|
203 | fprintf(stderr, "--nspots must be set or --nthreads set to something bigger than --nprocs\n"); |
---|
204 | exit(EXIT_FAILURE); |
---|
205 | } |
---|
206 | |
---|
207 | Time start, end; |
---|
208 | { |
---|
209 | MyData * data_arrays[nthreads]; |
---|
210 | for(i; nthreads) { |
---|
211 | data_arrays[i] = malloc(); |
---|
212 | (*data_arrays[i]){ i, wsize }; |
---|
213 | } |
---|
214 | |
---|
215 | MySpot * spots[nspots]; |
---|
216 | for(i; nspots) { |
---|
217 | spots[i] = malloc(); |
---|
218 | (*spots[i]){ i }; |
---|
219 | } |
---|
220 | |
---|
221 | BenchCluster bc = { nprocs }; |
---|
222 | threads_left = nprocs; |
---|
223 | { |
---|
224 | MyThread * threads[nthreads]; |
---|
225 | for(i; nthreads) { |
---|
226 | threads[i] = malloc(); |
---|
227 | (*threads[i]){ |
---|
228 | data_arrays[i], |
---|
229 | spots, |
---|
230 | nspots, |
---|
231 | wcnt, |
---|
232 | share, |
---|
233 | i |
---|
234 | }; |
---|
235 | } |
---|
236 | |
---|
237 | bool is_tty = isatty(STDOUT_FILENO); |
---|
238 | start = timeHiRes(); |
---|
239 | |
---|
240 | for(i; nthreads) { |
---|
241 | post( threads[i]->sem ); |
---|
242 | } |
---|
243 | wait(start, is_tty); |
---|
244 | |
---|
245 | stop = true; |
---|
246 | end = timeHiRes(); |
---|
247 | printf("\nDone\n"); |
---|
248 | |
---|
249 | for(i; nthreads) { |
---|
250 | post( threads[i]->sem ); |
---|
251 | MyThread & thrd = join( *threads[i] ); |
---|
252 | global_count += thrd.result.count; |
---|
253 | global_gmigs += thrd.result.gmigs; |
---|
254 | global_dmigs += thrd.result.dmigs; |
---|
255 | } |
---|
256 | |
---|
257 | for(i; nthreads) { |
---|
258 | ^( *threads[i] ){}; |
---|
259 | free( threads[i] ); |
---|
260 | } |
---|
261 | } |
---|
262 | |
---|
263 | for(i; nthreads) { |
---|
264 | ^( *data_arrays[i] ){}; |
---|
265 | free( data_arrays[i] ); |
---|
266 | } |
---|
267 | |
---|
268 | for(i; nspots) { |
---|
269 | ^( *spots[i] ){}; |
---|
270 | free( spots[i] ); |
---|
271 | } |
---|
272 | } |
---|
273 | |
---|
274 | printf("Duration (ms) : %'lf\n", (end - start)`dms); |
---|
275 | printf("Number of processors : %'d\n", nprocs); |
---|
276 | printf("Number of threads : %'d\n", nthreads); |
---|
277 | printf("Total Operations(ops) : %'15llu\n", global_count); |
---|
278 | printf("Work size (64bit words): %'15u\n", wsize); |
---|
279 | printf("Total Operations(ops) : %'15llu\n", global_count); |
---|
280 | printf("Total G Migrations : %'15llu\n", global_gmigs); |
---|
281 | printf("Total D Migrations : %'15llu\n", global_dmigs); |
---|
282 | printf("Ops per second : %'18.2lf\n", ((double)global_count) / (end - start)`ds); |
---|
283 | printf("ns per ops : %'18.2lf\n", (end - start)`dns / global_count); |
---|
284 | printf("Ops per threads : %'15llu\n", global_count / nthreads); |
---|
285 | printf("Ops per procs : %'15llu\n", global_count / nprocs); |
---|
286 | printf("Ops/sec/procs : %'18.2lf\n", (((double)global_count) / nprocs) / (end - start)`ds); |
---|
287 | printf("ns per ops/procs : %'18.2lf\n", (end - start)`dns / (global_count / nprocs)); |
---|
288 | fflush(stdout); |
---|
289 | } |
---|