-
Notifications
You must be signed in to change notification settings - Fork 102
/
index.html
399 lines (395 loc) · 12.5 KB
/
index.html
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
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
<!DOCTYPE html>
<html>
<head>
<meta charset="utf-8">
<link rel="stylesheet" href="../common-revealjs/css/reveal.css">
<link rel="stylesheet" href="../common-revealjs/css/theme/white.css">
<link rel="stylesheet" href="../common-revealjs/css/custom.css">
<script>
// This is needed when printing the slides to pdf
var link = document.createElement( 'link' );
link.rel = 'stylesheet';
link.type = 'text/css';
link.href = window.location.search.match( /print-pdf/gi ) ? '../common-revealjs/css/print/pdf.css' : '../common-revealjs/css/print/paper.css';
document.getElementsByTagName( 'head' )[0].appendChild( link );
</script>
<script>
// This is used to display the static images on each slide,
// See global-images in this html file and custom.css
(function() {
if(window.addEventListener) {
window.addEventListener('load', () => {
let slides = document.getElementsByClassName("slide-background");
if (slides.length === 0) {
slides = document.getElementsByClassName("pdf-page")
}
// Insert global images on each slide
for(let i = 0, max = slides.length; i < max; i++) {
let cln = document.getElementById("global-images").cloneNode(true);
cln.removeAttribute("id");
slides[i].appendChild(cln);
}
// Remove top level global images
let elem = document.getElementById("global-images");
elem.parentElement.removeChild(elem);
}, false);
}
})();
</script>
<style>
/* Custom styling for the highlighted code */
.highlight {
color: red;
font-weight: bold;
}
</style>
</head>
<body>
<div class="reveal">
<div class="slides">
<div id="global-images" class="global-images">
<img src="../common-revealjs/images/sycl_academy.png" />
<img src="../common-revealjs/images/sycl_logo.png" />
<img src="../common-revealjs/images/trademarks.png" />
<img src="../common-revealjs/images/codeplay.png" />
</div>
<!--Slide 1-->
<section class="hbox">
<div class="hbox" data-markdown>
## More SYCL Features
</div>
</section>
<!--Slide 2-->
<section class="hbox" data-markdown>
## Learning Objectives
* Learn about atomic operations and how to use them in SYCL kernels
* Learn about SYCL group algorithms
* Learn about SYCL reductions
</section>
<!--Slide 3-->
<section>
<div class="hbox" data-markdown>
#### Race Condition
</div>
<div class="container">
<div class="col" data-markdown>
* In a multithreaded environment, multiple work items writing indiscriminately
to the same area of memory causes a race condition.
</div>
<div class="col" data-markdown>
```
q.parallel_for([=](sycl::item<1> it) {
// Race condition! Multiple threads
// writing to same area of memory
a[0] = it.get_global_linear_id();
});
```
</div>
</div>
</section>
<!--Slide 4-->
<section>
<div class="hbox" data-markdown>
#### Race Condition
</div>
<div class="container">
<div class="col" data-markdown>
```
q.parallel_for([=](sycl::item<1> it) {
// Race condition! Multiple threads
// writing to same area of memory
a[0] = it.get_global_linear_id();
});
```
</div>
<div class="col" data-markdown>
* SYCL does not guarantee any particular ordering for the execution of work
items.
* When multiple work items concurrently write different values to the same area
of memory, there is no way of knowing which value will be held in memory
once all work items have finished writing.
* This is called a race condition and can be a source of non determinism in
code execution.
</div>
</div>
</section>
<!--Slide 5-->
<section>
<div class="hbox" data-markdown>
#### Atomic Operations
</div>
<div class="container">
<div class="col" data-markdown>
```
q.parallel_for([=](sycl::item<1> it) {
sycl::atomic_ref<T,
sycl::memory_order_relaxed,
sycl::memory_scope_device>
(a[0])
.fetch_add(it.get_global_linear_id());
});
```
</div>
<div class="col" data-markdown>
* Atomic operations are needed in order to deterministically combine values
from different work items.
* Atomic operations enforce a particular ordering of instructions across work
items. Some orderings include `memory_order_relaxed`, `memory_order_acq_rel`,
`memory_order_seq_cst`.
</div>
</div>
</section>
<!--Slide 6 -->
<section>
<div class="hbox" data-markdown>
#### Atomic Operations
</div>
<div class="container">
<div class="col" data-markdown>
```
q.parallel_for([=](sycl::item<1> it) {
sycl::atomic_ref<T,
sycl::memory_order_relaxed,
sycl::memory_scope_device>
(a[0])
.fetch_add(it.get_global_linear_id());
});
```
</div>
<div class="col" data-markdown>
* Atomic operations are needed in order to deterministically combine values
from different work items.
* Atomic operations enforce a particular ordering of instructions across work
items. Some orderings include `memory_order_relaxed`, `memory_order_acq_rel`,
`memory_order_seq_cst`.
</div>
</div>
</section>
<!--Slide 7 -->
<section>
<div class="hbox" data-markdown>
#### Atomic Operations
</div>
<div class="container">
<div class="col" data-markdown>
```
q.parallel_for([=](sycl::item<1> it) {
sycl::atomic_ref<T,
sycl::memory_order_relaxed,
sycl::memory_scope_device>
(a[0])
.fetch_add(it.get_global_linear_id());
});
```
</div>
<div class="col" data-markdown>
* Atomic operations are needed in order to deterministically combine values
from different work items.
* Atomic operations enforce a particular ordering of instructions across work
items. Some orderings include `memory_order_relaxed`, `memory_order_acq_rel`,
`memory_order_seq_cst`.
</div>
</div>
</section>
<!--Slide 8 -->
<section>
<div class="hbox" data-markdown>
#### Atomic Operations
</div>
<div class="container">
<div class="col" data-markdown>
```
q.parallel_for([=](sycl::item<1> it) {
sycl::atomic_ref<T,
sycl::memory_order_relaxed,
sycl::memory_scope_device>
(a[0])
.fetch_add(it.get_global_linear_id());
});
```
</div>
<div class="col" data-markdown>
* Atomic operations are needed in order to deterministically combine values
from different work items.
* Atomic operations enforce a particular ordering of instructions across work
items. Some orderings include `memory_order_relaxed`, `memory_order_acq_rel`,
`memory_order_seq_cst`.
</div>
</div>
</section>
<!--Slide 9 -->
<section>
<div class="hbox" data-markdown>
#### Atomic Operations
</div>
<div class="container">
<div class="col" data-markdown>
```
q.parallel_for([=](sycl::item<1> it) {
sycl::atomic_ref<T,
sycl::memory_order_relaxed,
sycl::memory_scope_device>
(a[0])
.fetch_add(it.get_global_linear_id());
});
```
</div>
<div class="col" data-markdown>
* Using atomics, values can be combined across work items without data races.
* `fetch_add`, `fetch_sub`, `fetch_max`, `fetch_min` are some of the ways we can
combine values atomically.
* `fetch_and` and `fetch_or` can also be used for
integral types
* Please see the SYCL Specification for more details.
</div>
</div>
</section>
<!--Slide 10 -->
<section>
<div class="hbox" data-markdown>
#### Atomic Operations
</div>
<div class="container">
<div class="col">
<pre><code>
q.parallel_for([=](sycl::item<1> it) {
sycl::atomic_ref<T,
sycl::memory_order_relaxed,
sycl::memory_scope_<span class="highlight">work_group</span>,
<span class="highlight">sycl::access::address_space::local_space</span>>
(a[0])
.fetch_add(it.get_global_linear_id());
});
</code></pre>
</div>
<div class="col" data-markdown>
* We can also specify the memory space of `a[0]`.
* If `a[0]` is in local memory, we should expect a speedup for using the local
memory atomic over the default atomic (which uses the generic address space).
</div>
</div>
</section>
<!--Slide 11 -->
<section>
<div class="hbox" data-markdown>
#### Group Algorithms
</div>
<div class="container" data-markdown>
* SYCL provides group algorithms which perform common operations over a single
workgroup.
* `reduce` algorithms perform fast work group reduction operation for some op
such as `plus`, `max`, etc.
* `any_of`, `all_of`, `none_of` as well as the `joint_*` counterparts perform
some predicate checking and return the same value to all items in a work
group.
* `permute_group_by_xor` permutes values among work items in a work group,
according to a provided mask.
* `inclusive_scan` and `exclusive_scan`. For a scan of elements: `[x0,...,xn]`
the ith result of an exclusive scan is the combination of the elements
[x0,...,x{i-1}] and the ith result of an inclusive scan is the combination of
elements [x0,...,xi] using some binary op.
* Please see the SYCL specification for more details.
</div>
</section>
<!--Slide 12 -->
<section>
<div class="hbox" data-markdown>
#### Group Algorithms
</div>
<div class="container" data-markdown>
* Group algorithms can operate on different group scopes, such as `work_group`,
`sub_group`.
* All work items in a given group scope must call the function in convergent
control flow.
* Please see the SYCL specification for more details.
</div>
</section>
</section>
<!--Slide 13 -->
<section>
<div class="hbox" data-markdown>
#### SYCL Reductions
</div>
<div class="container">
<div class="col">
<pre><code>
q.submit([&](sycl::handler &cgh) {
// Output of reduction will be in ptr
auto sumReduction = sycl::reduction(ptr,
sycl::plus<T>());
cgh.parallel_for(myNd, <span class="highlight">sumReduction</span>,
[=](sycl::nd_item<1> item, <span class="highlight">auto &sum</span>) {
sum += devA[item.get_global_linear_id()];
});
});
</code></pre>
</div>
<div class="col" data-markdown>
* SYCL provides reduction operators to perform fast reductions using some binary
operation.
* Reductions can be performed with `sycl::plus`, `sycl::maximum`, `sycl::multiplies` etc.
* Please see the SYCL specification for more details.
</div>
</div>
</section>
<!--Slide 14 -->
<section>
<div class="hbox" data-markdown>
#### SYCL Reductions
</div>
<div class="container">
<div class="col">
<pre><code>
q.submit([&](sycl::handler &cgh) {
// Output of reduction will be in ptr
auto maxReduction = sycl::reduction(ptr,
<span class="highlight">sycl::maximum</span><T>());
cgh.parallel_for(myNd, maxReduction,
[=](sycl::nd_item<1> item, <span class="highlight">auto &myMax</span>) {
myMax.<span class="highlight">combine</span>(devA[item.get_global_linear_id()]);
});
});
</code></pre>
</div>
<div class="col" data-markdown>
* SYCL provides reduction operators to perform fast reductions using some binary
operation.
* Reductions can be performed with `sycl::plus`, `sycl::maximum`,
`sycl::multiplies` etc.
* Please see the SYCL specification for more details.
</div>
</div>
</section>
<!--Slide 15-->
<section>
<div class="hbox" data-markdown>
## Questions
</div>
</section>
<!--Slide 16 -->
<section>
<div class="hbox" data-markdown>
#### Exercise
* See how atomics, group algorithms and `sycl::reductions` are used in
implementing a simple reduction operation.
* Which code runs fastest? Which code is simplest?
</div>
</section>
</div>
</div>
<script src="../common-revealjs/js/reveal.js"></script>
<script src="../common-revealjs/plugin/markdown/marked.js"></script>
<script src="../common-revealjs/plugin/markdown/markdown.js"></script>
<script src="../common-revealjs/plugin/notes/notes.js"></script>
<script>
Reveal.initialize({mouseWheel: true, defaultNotes: true});
Reveal.configure({ slideNumber: true });
</script>
</body>
</html>