1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
//! Integration test for Request::wait_any and Request::wait_some.
//!
//! Each rank posts 3 nonblocking receives and 3 nonblocking sends in a ring
//! pattern, then drives completion with wait_any (Part 1) and wait_some
//! (Part 2). Asserts that the total number of completions equals the number
//! of posted requests.
//!
//! Run with: mpiexec -n 4 ./target/debug/examples/test_waitany
use ferrompi::{Mpi, Request};
use std::time::Duration;
fn main() {
let mpi = Mpi::init().expect("MPI init failed");
let world = mpi.world();
let rank = world.rank();
let size = world.size();
assert!(
size >= 2,
"test_waitany requires at least 2 processes, got {size}"
);
let next = (rank + 1) % size;
let prev = (rank + size - 1) % size;
const N: usize = 3;
// ========================================================================
// Part 1: Post N irecvs + N isends, drive with wait_any
// ========================================================================
{
// Allocate receive buffers — must outlive the requests.
let mut recv_bufs: Vec<Vec<f64>> = (0..N).map(|_| vec![0.0f64; 4]).collect();
let send_bufs: Vec<Vec<f64>> = (0..N)
.map(|i| vec![(rank * 100 + i as i32) as f64; 4])
.collect();
// Post all receives first, then all sends (standard deadlock-free ordering).
let mut requests: Vec<Request> = Vec::with_capacity(N * 2);
for (i, buf) in recv_bufs.iter_mut().enumerate() {
let req = world
.irecv(buf, prev, 100 + i as i32)
.expect("irecv failed");
requests.push(req);
}
for (i, buf) in send_bufs.iter().enumerate() {
let req = world
.isend(buf, next, 100 + i as i32)
.expect("isend failed");
requests.push(req);
}
let total_posted = requests.len();
let mut completions = 0usize;
// Drive completions with wait_any; remove completed entries by swap-remove
// so the vector shrinks naturally. We track original indices by keeping a
// parallel index map so that wait_any indices stay valid after removals.
//
// Simpler approach matching the ticket spec: loop until the vec is empty,
// using swap-remove on each returned index.
while !requests.is_empty() {
let idx = Request::wait_any(&mut requests)
.expect("wait_any failed")
.expect("wait_any returned None on non-empty active request list");
// Remove the completed request (swap-remove preserves compactness).
requests.swap_remove(idx);
completions += 1;
}
assert_eq!(
completions, total_posted,
"rank {rank}: wait_any Part 1: expected {total_posted} completions, got {completions}"
);
// Verify received data from the previous rank.
for (i, buf) in recv_bufs.iter().enumerate() {
let expected_val = (prev * 100 + i as i32) as f64;
for (j, &v) in buf.iter().enumerate() {
assert!(
(v - expected_val).abs() < f64::EPSILON,
"rank {rank}: wait_any Part 1: recv_bufs[{i}][{j}] = {v}, expected {expected_val}"
);
}
}
println!("waitany: rank {rank} completed {completions} requests");
}
world.barrier().expect("barrier after Part 1 failed");
// ========================================================================
// Part 2: Post N isends, drive with wait_some, assert completions count
// ========================================================================
{
// We only post sends here; the previous rank's Part 1 receives are gone, so
// we pair with fresh receives on the next rank. Use a different tag range to
// avoid message matching confusion with Part 1 traffic.
//
// To keep the test self-contained, post N receives AND N sends again.
let mut recv_bufs2: Vec<Vec<f64>> = (0..N).map(|_| vec![0.0f64; 2]).collect();
let send_bufs2: Vec<Vec<f64>> = (0..N)
.map(|i| vec![(rank * 10 + i as i32) as f64; 2])
.collect();
let mut requests: Vec<Request> = Vec::with_capacity(N * 2);
for (i, buf) in recv_bufs2.iter_mut().enumerate() {
let req = world
.irecv(buf, prev, 200 + i as i32)
.expect("irecv Part 2 failed");
requests.push(req);
}
for (i, buf) in send_bufs2.iter().enumerate() {
let req = world
.isend(buf, next, 200 + i as i32)
.expect("isend Part 2 failed");
requests.push(req);
}
let total_posted = requests.len();
let mut all_completed_indices: Vec<usize> = Vec::new();
// Drive with wait_some; accumulate all returned indices, then remove.
while !requests.is_empty() {
let batch = Request::wait_some(&mut requests).expect("wait_some failed");
// wait_some returning empty on a non-empty active list is an error.
assert!(
!batch.is_empty(),
"rank {rank}: wait_some returned empty on non-empty active request list"
);
// Sort descending so swap-removes do not invalidate earlier indices.
let mut sorted = batch.clone();
sorted.sort_unstable_by(|a, b| b.cmp(a));
for idx in sorted {
requests.swap_remove(idx);
}
all_completed_indices.extend(batch);
}
assert_eq!(
all_completed_indices.len(),
total_posted,
"rank {rank}: wait_some Part 2: expected {total_posted} completions, got {}",
all_completed_indices.len()
);
if rank == 0 {
println!(
"waitany: rank {rank} completed {} requests via wait_some",
all_completed_indices.len()
);
}
}
world.barrier().expect("barrier after Part 2 failed");
// ========================================================================
// Part 3: Post N irecvs + N isends, drive with test_any polling loop
// ========================================================================
{
let mut recv_bufs3: Vec<Vec<f64>> = (0..N).map(|_| vec![0.0f64; 4]).collect();
let send_bufs3: Vec<Vec<f64>> = (0..N)
.map(|i| vec![(rank * 1000 + i as i32) as f64; 4])
.collect();
let mut requests: Vec<Request> = Vec::with_capacity(N * 2);
for (i, buf) in recv_bufs3.iter_mut().enumerate() {
let req = world
.irecv(buf, prev, 300 + i as i32)
.expect("irecv Part 3 failed");
requests.push(req);
}
for (i, buf) in send_bufs3.iter().enumerate() {
let req = world
.isend(buf, next, 300 + i as i32)
.expect("isend Part 3 failed");
requests.push(req);
}
let mut completions = 0usize;
while !requests.is_empty() {
match Request::test_any(&mut requests) {
Ok(Some(idx)) => {
requests.swap_remove(idx);
completions += 1;
}
Ok(None) => {
std::thread::sleep(Duration::from_millis(1));
}
Err(e) => panic!("test_any failed: {e}"),
}
}
assert_eq!(
completions,
N * 2,
"rank {rank}: test_any Part 3: expected {} completions, got {completions}",
N * 2
);
// Verify received data from the previous rank.
for (i, buf) in recv_bufs3.iter().enumerate() {
let expected_val = (prev * 1000 + i as i32) as f64;
for (j, &v) in buf.iter().enumerate() {
assert!(
(v - expected_val).abs() < f64::EPSILON,
"rank {rank}: test_any Part 3: recv_bufs3[{i}][{j}] = {v}, expected {expected_val}"
);
}
}
if rank == 0 {
println!("PASS: test_any polling completed {completions} requests");
}
}
world.barrier().expect("barrier after Part 3 failed");
// ========================================================================
// Part 4: Post N irecvs + N isends, drive with test_some polling loop
// ========================================================================
{
let mut recv_bufs4: Vec<Vec<f64>> = (0..N).map(|_| vec![0.0f64; 4]).collect();
let send_bufs4: Vec<Vec<f64>> = (0..N)
.map(|i| vec![(rank * 1000 + i as i32) as f64; 4])
.collect();
let mut requests: Vec<Request> = Vec::with_capacity(N * 2);
for (i, buf) in recv_bufs4.iter_mut().enumerate() {
let req = world
.irecv(buf, prev, 400 + i as i32)
.expect("irecv Part 4 failed");
requests.push(req);
}
for (i, buf) in send_bufs4.iter().enumerate() {
let req = world
.isend(buf, next, 400 + i as i32)
.expect("isend Part 4 failed");
requests.push(req);
}
let mut completions = 0usize;
while !requests.is_empty() {
match Request::test_some(&mut requests) {
Ok(batch) if !batch.is_empty() => {
let mut sorted = batch.clone();
sorted.sort_unstable_by(|a, b| b.cmp(a));
for idx in sorted {
requests.swap_remove(idx);
}
completions += batch.len();
}
Ok(_) => {
std::thread::sleep(Duration::from_millis(1));
}
Err(e) => panic!("test_some failed: {e}"),
}
}
assert_eq!(
completions,
N * 2,
"rank {rank}: test_some Part 4: expected {} completions, got {completions}",
N * 2
);
// Verify received data from the previous rank.
for (i, buf) in recv_bufs4.iter().enumerate() {
let expected_val = (prev * 1000 + i as i32) as f64;
for (j, &v) in buf.iter().enumerate() {
assert!(
(v - expected_val).abs() < f64::EPSILON,
"rank {rank}: test_some Part 4: recv_bufs4[{i}][{j}] = {v}, expected {expected_val}"
);
}
}
if rank == 0 {
println!("PASS: test_some polling completed {completions} requests");
}
}
world.barrier().expect("barrier after Part 4 failed");
if rank == 0 {
println!("\n========================================");
println!("All wait_any / wait_some tests passed!");
println!("========================================");
}
}