-
Notifications
You must be signed in to change notification settings - Fork 23
/
prefixsum-packedstore.cpp
542 lines (414 loc) · 14.1 KB
/
prefixsum-packedstore.cpp
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
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
#include <assert.h>
#include <stdio.h>
#include <algorithm>
#include <stdlib.h>
#include <iostream>
#include "hip/hip_runtime.h"
#include <math.h>
#define HIP_ASSERT(x) (assert((x)==hipSuccess))
//Algorithm£º
//Step 0: Prefix sum every 256 mask into 1 GROUP_IDX
//step 1: Prefix sum every 256 group idx
//step 2: prefix sum every 256x256 group idx
//step 3: prefix sum every 256x256 group idx
//step 4: packed store by
__global__ void mask_prefixsum(hipLaunchParm lp, const unsigned char* mask, unsigned int* group_offset, int total_length)
{
int localId = hipThreadIdx_x;
int localSize = hipBlockDim_x;
int globalIdx = hipBlockIdx_x;
__shared__ unsigned int prefix_sum[1024];
globalIdx = globalIdx * localSize + localId;
prefix_sum[localId] = 0;
if (globalIdx < total_length)
{
if (mask[globalIdx] > 0)
{
prefix_sum[localId] = 1;
}
}
//Do Prefix Sum;
//https://en.wikipedia.org/wiki/Prefix_sum
//Circuit representation of a highly parallel 16-input parallel prefix sum
for (int ll = 1; ll <= localSize / 2; ll = ll << 1)
{
__syncthreads();
if (localId >= ll)
{
int pre_val = prefix_sum[localId - ll];
int cur_val = prefix_sum[localId];
cur_val += pre_val;
__syncthreads();
prefix_sum[localId] = cur_val;
}
}
__syncthreads();
//Store max value into hipBlockIdx_X
if (0 == hipThreadIdx_x)
{
//Get Max from Last Valid Shared Memory
unsigned int max_value = prefix_sum[hipBlockDim_x - 1];
group_offset[hipBlockIdx_x] = max_value;
}
}
__global__ void group_prefixsum_failed_function_single_barrier(hipLaunchParm lp, unsigned int* group_offset, int total_length)
{
int localId = hipThreadIdx_x;
int localSize = hipBlockDim_x;
int globalIdx = hipBlockIdx_x;
__shared__ unsigned int prefix_sum[1024];
globalIdx = globalIdx * localSize + localId;
prefix_sum[localId] = 0;
if (globalIdx < total_length)
{
prefix_sum[localId] = group_offset[globalIdx];
}
//Do Prefix Sum;
//Circuit representation of a highly parallel 16-input parallel prefix sum
for (int ll = 1; ll <= localSize / 2; ll = ll << 1)
{
__syncthreads();
if (localId >= ll) {
int pre_val = prefix_sum[localId - ll];
int cur_val = prefix_sum[localId];
cur_val += pre_val;
//__syncthreads();
prefix_sum[localId] = cur_val;
}
}
__syncthreads();
//Store max value into hipBlockIdx_X
if (globalIdx < total_length)
group_offset[globalIdx] = prefix_sum[localId];
}
__global__ void group_prefixsum_Pass1(hipLaunchParm lp, unsigned int* group_offset, unsigned int total_length)
{
int localId = hipThreadIdx_x;
int localSize = hipBlockDim_x;
int globalIdx = hipBlockIdx_x;
__shared__ unsigned int prefix_sum[1024];
globalIdx = globalIdx * localSize + localId;
prefix_sum[localId] = 0;
if (globalIdx < total_length)
{
prefix_sum[localId] = group_offset[globalIdx];
}
//Do Prefix Sum;
//Circuit representation of a highly parallel 16-input parallel prefix sum
for (int ll = 1; ll <= localSize / 2; ll = ll << 1)
{
__syncthreads();
if (localId >= ll) {
int pre_val = prefix_sum[localId - ll];
int cur_val = prefix_sum[localId];
cur_val += pre_val;
__syncthreads();
prefix_sum[localId] = cur_val;
}
}
__syncthreads();
//Store max value into hipBlockIdx_X
if(globalIdx < total_length)
group_offset[globalIdx] = prefix_sum[localId];
}
__global__ void group_prefixsum_Pass2(hipLaunchParm lp, unsigned int* group_offset, unsigned int* group_offset2, unsigned int offset, unsigned int total_length)
{
unsigned int globalIdx = hipBlockIdx_x * hipBlockDim_x + hipThreadIdx_x;
unsigned int localId = hipThreadIdx_x;
__shared__ unsigned int prefix_sum[1024];
unsigned int big_group_size = offset * hipBlockDim_x;
unsigned int big_group_base = (hipBlockIdx_x * hipBlockDim_x) & (~(big_group_size - 1));
//inside its own offset*hipBlockDim_x, so must from offset gtroup 1,
//Otherwise it will read 1 more data from previous group's last data
unsigned int to_prefixsum_idx = big_group_base + (hipThreadIdx_x+1) * offset -1;
if (globalIdx < offset)
{
group_offset2[globalIdx] = group_offset[globalIdx];
return;
}
unsigned int to_prefixsum_val = 0;
//from second offset * hipBlockDim_x: can not read 1
if ((to_prefixsum_idx < (hipBlockIdx_x * hipBlockDim_x)) )
to_prefixsum_val = group_offset[to_prefixsum_idx];
prefix_sum[hipThreadIdx_x] = to_prefixsum_val;
unsigned int cur_val = 0;
if(globalIdx < total_length)
cur_val = group_offset[globalIdx];
__syncthreads();
//DO PrefixSUm of (max hipBlockDim_x )
for (int ll = 1; ll <= hipBlockDim_x / 2; ll = ll << 1)
{
__syncthreads();
if (localId >= ll) {
int pre_val = prefix_sum[localId - ll];
int cur_val = prefix_sum[localId];
cur_val += pre_val;
__syncthreads();
prefix_sum[localId] = cur_val;
}
}
__syncthreads();
if (globalIdx >= total_length)
return;
unsigned int max_val = prefix_sum[hipBlockDim_x - 1];
group_offset2[globalIdx] = max_val + cur_val;
}
__global__ void packed_store_prefixsum(hipLaunchParm lp, const unsigned int* in, const unsigned char* mask, unsigned int* out, unsigned int* group_offset, unsigned int total_length)
{
int localId = hipThreadIdx_x;
int localSize = hipBlockDim_x;
int globalIdx = hipBlockIdx_x * hipBlockDim_x + hipThreadIdx_x;
__shared__ unsigned int prefix_sum[1024]; //
if (globalIdx >= total_length)
return;
int group_prefixsum = 0;
//1st workgroup offset begins from 0
if (hipBlockIdx_x > 0)
{
group_prefixsum = group_offset[hipBlockIdx_x - 1];
}
unsigned int mask_val = 0;
mask_val = mask[globalIdx];
if (mask_val > 0)
mask_val = 1;
prefix_sum[localId] = mask_val;
//Do Prefix Sum locally;
for (int ll = 1; ll <= localSize / 2; ll = ll << 1)
{
__syncthreads();
if (localId >= ll) {
int pre_val = prefix_sum[localId - ll];
int cur_val = prefix_sum[localId];
cur_val += pre_val;
__syncthreads();
prefix_sum[localId] = cur_val;
}
}
__syncthreads();
if (mask_val > 0)
{
//first thread with Mask>1, the prefix_sum result for 1st thread = 1, so must -1s,
out[group_prefixsum + prefix_sum[localId] - 1] = in[globalIdx];
}
}
using namespace std;
void PackedStore(const unsigned int* in, const unsigned char* mask, unsigned int* out, unsigned int* groupoffset, unsigned int* groupoffset2, unsigned int inlength, unsigned int& outlength, unsigned int verify = 0) {
unsigned int localThreads = 256;
unsigned int gloal_blocks = (inlength + localThreads - 1) / localThreads;
//DO Prefix for every 256 threads
hipLaunchKernel(mask_prefixsum,
dim3(gloal_blocks),
dim3(localThreads),
0, 0,
mask, groupoffset, inlength);
#if 0
if (verify > 0) {
unsigned int packedLen = 0;
unsigned int* tempBuf = (unsigned int*)malloc(sizeof(unsigned int) * (gloal_blocks));
HIP_ASSERT(hipMemcpy(tempBuf, groupoffset, sizeof(unsigned int) * (gloal_blocks), hipMemcpyDeviceToHost));
hipStreamSynchronize(0);
for (int i = 0; i < gloal_blocks; i++)
{
if (128 != tempBuf[i])
{
if (i != (gloal_blocks - 1))
printf("must == 128 , [%d]th %9d %9d\n", i, tempBuf[i], tempBuf[i - 1]);
}
}
printf("passed for mask_prefixsum\n");
free(tempBuf);
}
#endif
//Prefix Sum for GroupOffset
unsigned int offset = 1;
unsigned int group_blocks = (gloal_blocks + localThreads - 1) / localThreads;
for (offset = 1; offset < gloal_blocks; offset = offset * localThreads)
{
if (offset == 1) {
hipLaunchKernel(group_prefixsum_Pass1,
dim3(group_blocks),
dim3(localThreads),
0, 0,
groupoffset, gloal_blocks);
}
else if (offset == 256){
hipLaunchKernel(group_prefixsum_Pass2,
dim3(group_blocks),
dim3(localThreads),
0, 0,
groupoffset, groupoffset2, offset, gloal_blocks);
}
else
{
hipLaunchKernel(group_prefixsum_Pass2,
dim3(group_blocks),
dim3(localThreads),
0, 0,
groupoffset2, groupoffset, offset, gloal_blocks);
}
}
#if 0
if (verify > 0) {
if ((gloal_blocks > 256) && (gloal_blocks <= 256 * 256))
{
HIP_ASSERT(hipMemcpyDtoD(groupoffset, groupoffset2, sizeof(unsigned int) * (gloal_blocks)));
}
unsigned int packedLen = 0;
unsigned int* tempBuf = (unsigned int*)malloc(sizeof(unsigned int) * (gloal_blocks));
HIP_ASSERT(hipMemcpy(tempBuf, groupoffset, sizeof(unsigned int) * (gloal_blocks), hipMemcpyDeviceToHost));
hipStreamSynchronize(0);
for (int i = 0; i < gloal_blocks; i++)
{
if ((128 * (i+1)) != tempBuf[i])
{
if(i != (gloal_blocks-1))
printf(" not equal 128x == , [%d]th %9d \n", i, tempBuf[i]);
}
}
free(tempBuf);
}
#endif
if ((gloal_blocks > 256) && (gloal_blocks <= 256 * 256))
{
HIP_ASSERT(hipMemcpy(&outlength, groupoffset2 + (gloal_blocks - 1), sizeof(unsigned int) * 1, hipMemcpyDeviceToHost));
//Finally output
hipLaunchKernel(packed_store_prefixsum,
dim3(gloal_blocks),
dim3(localThreads),
0, 0,
in, mask, out, groupoffset2, inlength);
}
else
{
HIP_ASSERT(hipMemcpy(&outlength, groupoffset + (gloal_blocks - 1), sizeof(unsigned int) * 1, hipMemcpyDeviceToHost));
//Finally output
hipLaunchKernel(packed_store_prefixsum,
dim3(gloal_blocks),
dim3(localThreads),
0, 0,
in, mask, out, groupoffset, inlength);
}
}
void test(unsigned int inlength)
{
unsigned int A_NUM = inlength;
unsigned int B_NUM = inlength;
unsigned int C_NUM = inlength;
unsigned int D_NUM = ((inlength + 256 - 1) / 256);
unsigned int* hostA;
unsigned char* hostB;
unsigned int* hostC;
unsigned int* hostD;
unsigned int* deviceA;
unsigned char* deviceB;
unsigned int* deviceC;
unsigned int* deviceD;
unsigned int* deviceD2;
hipDeviceProp_t devProp;
hipGetDeviceProperties(&devProp, 0);
cout << " System minor " << devProp.minor << endl;
cout << " System major " << devProp.major << endl;
cout << " agent prop name " << devProp.name << endl;
cout << "hip Device prop succeeded " << endl;
hipEvent_t start, stop;
hipEventCreate(&start);
hipEventCreate(&stop);
float eventMs = 1.0f;
int errors;
hostA = (unsigned int*)malloc(A_NUM * sizeof(unsigned int));
hostB = (unsigned char*)malloc(B_NUM * sizeof(unsigned char));
hostC = (unsigned int*)malloc(C_NUM * sizeof(unsigned int));
hostD = (unsigned int*)malloc(D_NUM * sizeof(unsigned int));
for (unsigned int i = 0; i < inlength; i = i + 1) {
hostA[i] = i;
hostB[i] = (i) & (0x1); //Magic Mask
//printf("%d ", hostB[i]);
hostC[i] = 0;
}
//printf("\n");
cout << "host allocated \n";
HIP_ASSERT(hipMalloc((void**)& deviceA, A_NUM * sizeof(unsigned int)));
HIP_ASSERT(hipMalloc((void**)& deviceB, B_NUM * sizeof(unsigned char)));
HIP_ASSERT(hipMalloc((void**)& deviceC, C_NUM * sizeof(unsigned int)));
HIP_ASSERT(hipMalloc((void**)& deviceD, D_NUM * sizeof(unsigned int)));
HIP_ASSERT(hipMalloc((void**)& deviceD2, D_NUM * sizeof(unsigned int)));
cout << "device allocated \n";
HIP_ASSERT(hipMemcpy(deviceA, hostA, A_NUM * sizeof(unsigned int), hipMemcpyHostToDevice));
HIP_ASSERT(hipMemcpy(deviceB, hostB, B_NUM * sizeof(unsigned char), hipMemcpyHostToDevice));
HIP_ASSERT(hipMemcpy(deviceC, hostC, C_NUM * sizeof(unsigned int), hipMemcpyHostToDevice));
cout << "Host to Device Copied\n";
{
unsigned int outLength = 0;
hipStreamSynchronize(0);
unsigned verify = 0;
PackedStore(deviceA, deviceB, deviceC, deviceD, deviceD2, inlength, outLength, verify);
hipStreamSynchronize(0);
printf("inLen: %d, outLength %d\n", inlength, outLength);
if (verify > 0) {
unsigned int targetLen = (inlength / 2) + (inlength & 0x1);
unsigned int targetVal = (inlength - 1) & 0xFFFFFFFF;
unsigned int packedVal = 0;
HIP_ASSERT(hipMemcpy(hostC, deviceC, sizeof(unsigned int) * (targetLen), hipMemcpyDeviceToHost));
hipStreamSynchronize(0);
packedVal = hostC[targetLen - 1];
if (targetVal == packedVal)
{
printf("--->Passed! target [%d] == packed [%d] \n", targetVal, packedVal);
}
else
{
printf("--->Failed! target [%d] == packed [%d] \n", targetVal, packedVal);
}
}
hipStreamSynchronize(0);
hipEventRecord(start, NULL);
int iterations = 10;
for (int i = 0; i < iterations; i++)
{
PackedStore(deviceA, deviceB, deviceC, deviceD, deviceD2, inlength, outLength, 0);
if (0) {
unsigned int* tmpBuffer = (unsigned int*)malloc(outLength * sizeof(unsigned int));
HIP_ASSERT(hipMemcpy(tmpBuffer, deviceC , outLength * sizeof(unsigned int), hipMemcpyDeviceToHost));
free(tmpBuffer);
}
}
hipEventRecord(stop, NULL);
hipEventSynchronize(stop);
hipEventElapsedTime(&eventMs, start, stop);
float bandwidth = float(inlength) * float( sizeof(unsigned int) * 2 + sizeof(unsigned char) * 2) / (float)1024 / (float)1024 / (float)1024;
bandwidth = bandwidth / (eventMs / iterations / 1000.0);
printf("Inlen:[%d] costs %f millseconds , Bandwidth = %f GB/s\n", inlength, eventMs / iterations, bandwidth);
}
HIP_ASSERT(hipMemcpy(hostC, deviceC, sizeof(int) * C_NUM, hipMemcpyDeviceToHost));
//CPU Verifying here
HIP_ASSERT(hipFree(deviceA));
HIP_ASSERT(hipFree(deviceB));
HIP_ASSERT(hipFree(deviceC));
HIP_ASSERT(hipFree(deviceD));
HIP_ASSERT(hipFree(deviceD2));
free(hostA);
free(hostB);
free(hostC);
free(hostD);
//return errors;
}
int main(int argc, char** argv)
{
if (argc != 2)
{
printf("it is to demo packed store:\n");
printf(" <length> total length to do pack store \n");
printf("----Packed Store--- \n");
printf("void func(uint32_t inlen, const uint32_t *in, const uint8_t *mask, uint32_t *out) { \n");
printf(" uint32_t oidx = 0; \n");
printf(" for (uint32_t_t i = 0; i < inlen; i++) { \n");
printf(" if (mask[i]) out[oidx++] = in[i]; \n");
printf("}\n");
return 0;
}
unsigned int packedLength = 0;
packedLength = atoi(argv[1]);
//packedLength = 256 * 256 * 255;
test(packedLength);
return 0;
}