-
Notifications
You must be signed in to change notification settings - Fork 0
/
f2d_V5_5.cu
1515 lines (1126 loc) · 44.3 KB
/
f2d_V5_5.cu
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
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
//P2P transfers tested and pinned memory tranfers tested for coupled overlapping Halo Exchanges. P2P works better in this case
#include "cuda_runtime.h"
#include "device_launch_parameters.h"
#include "testMultiGPU_Jacobi2D_Decom.cuh"
#include <iostream>
#include <chrono>
#include <memory>
#include <vector>
#include <fstream>
using namespace std;
using namespace std::chrono;
#define IMUL(a,b) __mul24(a,b)
//cudaError_t performMultiGPUJacobi();
//Support for below c++14 on *nix
template<typename T, typename ...Args>
std::unique_ptr<T> make_unique(Args&& ...args)
{
return std::unique_ptr<T>(new T(std::forward<Args>(args)...));
}
struct create_Device
{
int deviceID;
//In a GPU topology set the GPU position
int devicePosition_X;
int devicePosition_Y;
int devicePosition_Z;
vector<float> eHalo;
vector<float> wHalo;
vector<float> nHalo;
vector<float> sHalo;
//Flags check the halos needed by the device
int eHalo_flag = 0;
int wHalo_flag = 0;
int nHalo_flag = 0;
int sHalo_flag = 0;
};
//Simple Jacobi iteration
__global__ void jacobi_Simple(const float *A0, const float *A1, const float *A2, const float *A3, const float *A4, float *x_in, float *x_out, const float *rhs, const int ehalo_flag, const int whalo_flag, const int nhalo_flag, const int shalo_flag, float *ehalo, float *whalo, float *nhalo, float *shalo, const int deviceID, const int numDevices, const int domain_Decom)
{
int index = threadIdx.x + blockDim.x * blockIdx.x;
float result = rhs[index];
int dim_x = blockDim.x;// dim across x
int dim_y = gridDim.x;
int x_pos = blockIdx.x;
int y_pos = threadIdx.x;
//result = nhalo[y_pos];
//x_out[index] = result;
//Get the boundaries
int leftBoundaryElem = x_pos * (dim_x);
int rightBoundaryElem = (x_pos * dim_x) + (dim_x - 1);
int topBoundaryElem = y_pos + ((dim_y - 1) * (dim_x));
int bottomBoundaryElem = y_pos;
//Halo computation for 1D Decompostion: For the First and Last GPU Halo computation on both the sides(nhalo and shalo wont be needed)
if (domain_Decom == 1)
{
if (numDevices > 1)
{
//First GPU
if (deviceID == 0) {
//We need to use nhalos
//Carry out computations for boundary elements
if (index != leftBoundaryElem)
//Left
result -= A1[index] * x_in[index - 1];
if (index != rightBoundaryElem)
//Right
result -= A3[index] * x_in[index + 1];
if (index != bottomBoundaryElem)
//Bottom
result -= A0[index] * x_in[index - dim_x];
if (index != topBoundaryElem)
//Top
result -= A4[index] * x_in[index + dim_x];
//The top boundary needs element from nhalo
if (index == topBoundaryElem)
//nHalos
result -= A4[index] * nhalo[y_pos];
result /= A2[index];
x_out[index] = result;
//Update Halo at the end of computation
if (index == topBoundaryElem)
//nHalos updated
nhalo[y_pos] = result;
return;
}
//Last GPU
else if (deviceID == (numDevices - 1)) {
//We need to use shalos
//Carry out computations for boundary elements
if (index != leftBoundaryElem)
//Left
result -= A1[index] * x_in[index - 1];
if (index != rightBoundaryElem)
//Right
result -= A3[index] * x_in[index + 1];
if (index != bottomBoundaryElem)
//Bottom
result -= A0[index] * x_in[index - dim_x];
//The Bottom boundary needs elements from shalo
if (index == bottomBoundaryElem)
//nHalos
result -= A0[index] * shalo[y_pos];
if (index != topBoundaryElem)
//Top
result -= A4[index] * x_in[index + dim_x];
result /= A2[index];
x_out[index] = result;
//Update Halo at the end of computation
if (index == bottomBoundaryElem)
//sHalos updated
shalo[y_pos] = result;
return;
}
//For all the middle GPUs
else
{
//We need to use both shalos and nhalos
//Carry out computations for boundary elements
if (index != leftBoundaryElem)
//Left
result -= A1[index] * x_in[index - 1];
if (index != rightBoundaryElem)
//Right
result -= A3[index] * x_in[index + 1];
if (index != bottomBoundaryElem)
//Bottom
result -= A0[index] * x_in[index - dim_x];
//The Bottom boundary needs elements from shalo
if (index == bottomBoundaryElem)
//nHalos
result -= A0[index] * shalo[y_pos];
if (index != topBoundaryElem)
//Top
result -= A4[index] * x_in[index + dim_x];
//The top boundary needs element from nhalo
if (index == topBoundaryElem)
//nHalos
result -= A4[index] * nhalo[y_pos];
result /= A2[index];
x_out[index] = result;
//Update Halo at the end of computation
if (index == bottomBoundaryElem)
//sHalos updated
shalo[y_pos] = result;
//Update Halo at the end of computation
if (index == topBoundaryElem)
//nHalos updated
nhalo[y_pos] = result;
return;
}
}
}
else if (domain_Decom == 2) {
//======Left Bounday Elem
if (index != leftBoundaryElem)
//Left
result -= A1[index] * x_in[index - 1];
//Computation using the Halos
if (index == leftBoundaryElem) {
if (whalo_flag == 1) {
result -= A1[index] * whalo[x_pos];
}
}
//======Right Bounday Elem
if (index != rightBoundaryElem)
//Right
result -= A3[index] * x_in[index + 1];
if (index == rightBoundaryElem) {
if (ehalo_flag == 1) {
result -= A3[index] * ehalo[x_pos];
}
}
//======Bottom Bounday Elem
if (index != bottomBoundaryElem)
//Bottom
result -= A0[index] * x_in[index - dim_x];
if (index == bottomBoundaryElem) {
if (shalo_flag == 1) {
result -= A0[index] * shalo[y_pos];
}
}
//======Top Bounday Elem
if (index != topBoundaryElem)
//Top
result -= A4[index] * x_in[index + dim_x];
if (index == topBoundaryElem) {
if (nhalo_flag == 1) {
result -= A4[index] * nhalo[y_pos];
}
}
result /= A2[index];
x_out[index] = result;
//Updating Halos at the End of the computation
if (index == topBoundaryElem) {
if (nhalo_flag == 1) {
nhalo[y_pos] = result;
}
}
if (index == bottomBoundaryElem) {
if (shalo_flag == 1) {
shalo[y_pos] = result;
}
}
if (index == leftBoundaryElem) {
if (whalo_flag == 1) {
whalo[x_pos] = result;
}
}
if (index == rightBoundaryElem) {
if (ehalo_flag == 1) {
ehalo[x_pos] = result;
}
}
return;
}
//For computations on a Machine with a single GPU
else
{
{//For some reason order of computation (left,right,top and bottom) gives a different result
//Carry out computations for boundary elements
if (index != leftBoundaryElem)
//Left
result -= A1[index] * x_in[index - 1];
if (index != rightBoundaryElem)
//Right
result -= A3[index] * x_in[index + 1];
if (index != bottomBoundaryElem)
//Bottom
result -= A0[index] * x_in[index - dim_x];
if (index != topBoundaryElem)
//Top
result -= A4[index] * x_in[index + dim_x];
result /= A2[index];
x_out[index] = result;
return;
}
}
}
//========================MultiGPU utility functions============================================================================
void checkP2Paccess(int numGPUs)
{
for (int i = 0; i<numGPUs; i++)
{
cudaSetDevice(i);
for (int j = 0; j<numGPUs; j++)
{
int access;
if (i != j)
{
cudaDeviceCanAccessPeer(&access, i, j);
if (auto err = cudaGetLastError())
{
cout << "P2P Operations failed : " << cudaGetErrorString(err) << endl;
return;
}
}
}
}
cout << "\n***NOTE: In case a device doesn't have P2P access to other one, it falls back to normal memcopy procedure.\nSo you can see lesser Bandwidth (GB/s) in those cases.\n\n";
}
bool enableP2P(int numGPUs)
{
for (int i = 0; i<numGPUs; i++)
{
cudaSetDevice(i);
for (int j = 0; j<numGPUs; j++)
{
int access;
cudaDeviceCanAccessPeer(&access, i, j);
if (auto err = cudaGetLastError())
{
cout << "P2P Operations failed while enabling: " << cudaGetErrorString(err) << endl;
return false;
}
if (access)
{
cudaDeviceEnablePeerAccess(j, 0);
if (auto err = cudaGetLastError())
{
cout << "P2P Operations failed while enabling: " << cudaGetErrorString(err) << endl;
return false;
}
}
}
}
return true;
}
void disableP2P(int numGPUs)
{
for (int i = 0; i<numGPUs; i++)
{
cudaSetDevice(i);
for (int j = 0; j<numGPUs; j++)
{
int access;
cudaDeviceCanAccessPeer(&access, i, j);
if (auto err = cudaGetLastError())
{
cout << "P2P Operations failed while disabling : " << cudaGetErrorString(err) << endl;
return;
}
if (access)
{
cudaDeviceDisablePeerAccess(j);
if (auto err = cudaGetLastError())
{
cout << "P2P Operations failed while disabling: " << cudaGetErrorString(err) << endl;
return;
}
}
}
}
}
//===============================================================================================================================
//====================================Creating Topology with the number of Devices available====================================
void generateGPUGRID(int numDevices, int &numberOfDevicesAlong_X, int &numberOfDevicesAlong_Y)
{
//Finding GPU topology along x and y
//Assumuing total number of devices is a perfect square(To be changed later)
numberOfDevicesAlong_X = (int)sqrt(numDevices);
numberOfDevicesAlong_Y = (int)numberOfDevicesAlong_X;
}
/* Creates a topology for a number of devices in a system
for ex. The devices are aware of left, right, top and bottom neigbours in 2D
1. It also decides the chunk per devices by determining x-dimension and y-dimensions for per chunk of data per device.
2. It also initializes halos for each devices which can be exchanged with the neighbours
*/
void createTopology(int numDevices, vector<create_Device> &deviceArray, int numberOfDevicesAlong_X, int numberOfDevicesAlong_Y)
{
deviceArray.resize(numDevices);
unsigned int deviceCount = 0;
for (int gridCount_X = 0; gridCount_X < numberOfDevicesAlong_X; gridCount_X++) {
for (int gridCount_Y = 0; gridCount_Y < numberOfDevicesAlong_Y; gridCount_Y++) {
deviceArray[deviceCount].deviceID = deviceCount;
deviceArray[deviceCount].devicePosition_X = gridCount_X;
deviceArray[deviceCount].devicePosition_Y = gridCount_Y;
//devicePosition_Z to be changed later
deviceArray[deviceCount].devicePosition_Z = 1;
deviceCount++;
}
}
}
//==============================================================================================================================
//Init Halos: In 1D decomposition only North and South Halos are used. In 2D decomposition North, South, East and West Halo need to be initialized and computed
//TODO:Create a Halo Exchange Mechanism for 2D Multi GPU topology
void initHalos2D(create_Device &device, int chunk_X, int chunk_Y, float *vec_in, int maxdevicesAlong_X, int maxDevicesAlong_Y, int rowStartPos, int rowEndPos, int dim) {
/*cout << endl << "Inside Halo Computation 2D. printing Details";
cout << endl << "Device ID " << device.deviceID;
cout << endl << "Device position X " << device.devicePosition_X;
cout << endl << "Device position Y " << device.devicePosition_Y;
cout << endl << "Row Start " << rowStartPos;
cout << endl << "Row End " << rowEndPos;*/
//Assigning counter for each individual Halos. To prevent update of the same counter
//int rowStartPosEast = rowStartPos;
int rowStartPosWest = rowStartPos;
int rowStartPosNorth = rowStartPos;
int rowStartPosSouth = rowStartPos;
int rowEndPosEast = rowEndPos;
//int rowEndPosWest = rowEndPos;
//int rowEndPosNorth = rowEndPos;
//int rowEndPosSouth = rowEndPos;
//Checks provided for Boundary devices in GPU topology
if ((device.devicePosition_Y - 1) >= 0) {
//cout << "West Halo needed ";
device.wHalo_flag = 1;
device.wHalo.resize(chunk_Y);
for (int rowNum = 0; rowNum < chunk_Y; rowNum++)
{
device.wHalo[rowNum] = vec_in[rowStartPosWest];
//cout << rowStartPosWest << " ";
rowStartPosWest += dim;
}
}
if ((device.devicePosition_Y + 1) < maxdevicesAlong_X) {
//cout << "East Halo needed ";
device.eHalo_flag = 1;
device.eHalo.resize(chunk_Y);
for (int rowNum = 0; rowNum < chunk_Y; rowNum++)
{
device.eHalo[rowNum] = vec_in[rowEndPosEast];
//cout << rowEndPosEast << " ";
rowEndPosEast += dim;
}
}
if ((device.devicePosition_X - 1) >= 0) {
//cout << "South Halo needed ";
device.sHalo_flag = 1;
device.sHalo.resize(chunk_X);
for (int rowNum = 0; rowNum < chunk_X; rowNum++)
{
device.sHalo[rowNum] = vec_in[rowStartPosSouth];
//cout << rowStartPosSouth << " ";
rowStartPosSouth++;
}
}
if ((device.devicePosition_X + 1) < maxDevicesAlong_Y) {
//cout << "North Halo needed ";
device.nHalo_flag = 1;
device.nHalo.resize(chunk_X);
rowStartPosNorth = rowStartPosNorth + (dim * (chunk_Y - 1));
for (int rowNum = 0; rowNum < chunk_X; rowNum++)
{
device.nHalo[rowNum] = vec_in[rowStartPosNorth];
//cout << rowStartPosNorth << " ";
rowStartPosNorth++;
}
}
}
//======================================Exchange Halos: on Host==============================================
int getDeviceIDfromCoord(int devCoord_x, int devCoord_y, int numberofDevicesAlong_X) {
int devID = (devCoord_x * numberofDevicesAlong_X) + devCoord_y;
return devID;
}
void exchangehalos_onHost(int numDevices, vector<create_Device> &deviceArray, int numberofDevicesAlong_X)
{
//Halos exist in pairs so:
//Important: A device exchanges North-to-South Pairs and East-to-West Pairs only. Not South-to-North pairs and West-to-East pairs
//That way the number of exchanges are kept to minimum
for (int dev = 0;dev < numDevices;dev++)
{
int getDevCoord_X = deviceArray[dev].devicePosition_X;
int getDevCoord_Y = deviceArray[dev].devicePosition_Y;
//Check if device is having a north Halo buffer
if (deviceArray[dev].nHalo_flag == 1) {
int devIDtoNorth = getDeviceIDfromCoord(getDevCoord_X + 1, getDevCoord_Y, numberofDevicesAlong_X);
//Exchange Halos
(deviceArray[dev].nHalo).swap(deviceArray[devIDtoNorth].sHalo);
}
//Check if device is having a east Halo buffer
if (deviceArray[dev].eHalo_flag == 1) {
int devIDtoEast = getDeviceIDfromCoord(getDevCoord_X, getDevCoord_Y + 1, numberofDevicesAlong_X);
//Exchange Halos
(deviceArray[dev].eHalo).swap(deviceArray[devIDtoEast].wHalo);
}
}
}
bool exchangehalos_onHostPinned(int numDevices, vector<create_Device> &deviceArray, int numberofDevicesAlong_X, vector<float*> &nHalosPinned, vector<float*> &sHalosPinned, vector<float*> &eHalosPinned, vector<float*> &wHalosPinned)
{
//Halos exist in pairs so:
//Important: A device exchanges North-to-South Pairs and East-to-West Pairs only. Not South-to-North pairs and West-to-East pairs
//That way the number of exchanges are kept to minimum
for (int dev = 0;dev < numDevices;dev++)
{
int getDevCoord_X = deviceArray[dev].devicePosition_X;
int getDevCoord_Y = deviceArray[dev].devicePosition_Y;
//Check if device is having a north Halo buffer
if (deviceArray[dev].nHalo_flag == 1) {
int devIDtoNorth = getDeviceIDfromCoord(getDevCoord_X + 1, getDevCoord_Y, numberofDevicesAlong_X);
//Exchange Halos
//(deviceArray[dev].nHalo).swap(deviceArray[devIDtoNorth].sHalo);
swap(nHalosPinned[dev], sHalosPinned[devIDtoNorth]);
}
//Check if device is having a east Halo buffer
if (deviceArray[dev].eHalo_flag == 1) {
int devIDtoEast = getDeviceIDfromCoord(getDevCoord_X, getDevCoord_Y + 1, numberofDevicesAlong_X);
//Exchange Halos
//(deviceArray[dev].eHalo).swap(deviceArray[devIDtoEast].wHalo);
swap(eHalosPinned[dev], wHalosPinned[devIDtoEast]);
}
}
return true;
}
//===========================Exchange Halos: on Host Ends=====================================================
//Init matrix Diagonals A0, A1, A2, A3, A4
void copyValues(float *A0, float *A1, float *A2, float *A3, float *A4, float *rhs, float *vec_in, float *vec_out, int dim, float *val_A0, float *val_A1, float *val_A2, float *val_A3, float *val_A4, float *val_rhs, float *val_x_in)
{
unsigned int size = dim * dim;
for (unsigned int i = 0; i < size; i++)
{
A0[i] = val_A0[i];
A1[i] = val_A1[i];
A2[i] = val_A2[i];
A3[i] = val_A3[i];
A4[i] = val_A4[i];
rhs[i] = val_rhs[i];
vec_in[i] = val_x_in[i];
vec_out[i] = 0.0f;
}
}
void getAllDeviceProperties() {
int nDevices;
cudaGetDeviceCount(&nDevices);
for (int i = 0; i < nDevices; i++) {
cudaDeviceProp prop;
cudaGetDeviceProperties(&prop, i);
cout << " Device Number: " << i << endl;
cout << " Device name: " << prop.name << endl;
cout << " Memory Clock Rate (KHz): " << prop.memoryClockRate << endl;
cout << " Memory Bus Width (bits): " << prop.memoryBusWidth << endl;;
cout << " Peak Memory Bandwidth (GB/s): " << 2.0*prop.memoryClockRate*(prop.memoryBusWidth / 8) / 1.0e6 << endl << endl << endl;
}
}
/* Prints an output file for checking results */
void sendToPrint(float *partial_result, int devicePosition_X, int devicePosition_Y, int numberOfDevicesAlong_X, int chunk_X, int chunk_Y, int dim, int totalSize, vector<float> &result, int numDevices, int currentIteration, int numberOfTotalIterations) {
int devicePosX = devicePosition_X;
int devicePosY = devicePosition_Y;
//Calculating data position based on device coords
//numberOfDevicesAlong_X * Chunk_X * Chunk_Y : finds out the total data per row of GPUs allocated
int dataStartPos_X = (devicePosX * numberOfDevicesAlong_X * chunk_X * chunk_Y) + (devicePosY * chunk_X);
int dataEndPos_X = dataStartPos_X + chunk_X;
//One complete row across all GPU is dim in order to get the next element above an element we add (currentPosition + dim )
int rowStartPos = dataStartPos_X;
int rowEndPos = dataEndPos_X;
int indexCounter = 0;
//cout << endl;
for (int rowNum = 0; rowNum < chunk_Y; rowNum++)
{
//Get one complete row for the GPU
for (int pos = rowStartPos; pos < rowEndPos; pos++)
{
result[pos] = partial_result[indexCounter];
indexCounter++;
}
//cout << endl;
rowStartPos += dim;
rowEndPos += dim;
}
//Printing when the last device computation is done: Remove the check to check computation for each device
int deviceID = getDeviceIDfromCoord(devicePosition_X, devicePosition_Y, numberOfDevicesAlong_X);
if ((deviceID == (numDevices - 1)) && (currentIteration == (numberOfTotalIterations - 1)))
{
ofstream myfile;
myfile.open("data2.txt");
//Printing the values here
for (int i = totalSize;i > 0; i--) {
if (i%dim == 0) {
myfile << endl;
}
myfile << result[i - 1] << " ";
}
myfile.close();
}
}
cudaError_t performMultiGPUJacobi(unsigned int val_dim, unsigned int numJacobiIt, float* val_A0, float* val_A1, float* val_A2, float* val_A3, float* val_A4, float* val_rhs, float* val_x_in)
{
//Fixed value changed later
int dim = 8;
if (val_dim != 0) {
dim = val_dim;
}
//TODO: write a 2D domain decomposition method for more than 2 GPUs
int size = dim * dim;
//auto result = make_unique<float[]>(size);
//Create Diagonal Vectors
std::vector<float> a0(size);
std::vector<float> a1(size);
std::vector<float> a2(size);
std::vector<float> a3(size);
std::vector<float> a4(size);
std::vector<float> vec_in(size);
std::vector<float> vec_out(size);
std::vector<float> rhs(size);
std::vector<float> result(size);
//Get the total number of devices
int numDevices;
cudaGetDeviceCount(&numDevices);
cout << endl << "Total number of Devices in the System are : " << numDevices << endl;
getAllDeviceProperties();
//Enable Peer-to-Peer access across all GPUs : Done on phase 2 of development
bool p2penabled = false;
p2penabled = enableP2P(numDevices);
//Configuring the number of GPU's manually
//numDevices=2;
copyValues(&a0[0], &a1[0], &a2[0], &a3[0], &a4[0], &rhs[0], &vec_in[0], &vec_out[0], dim, &val_A0[0], &val_A1[0], &val_A2[0], &val_A3[0], &val_A4[0], &val_rhs[0], &val_x_in[0]);
vector<create_Device> deviceArray;
/* Distributed Compuation using Halos: Algorithm
1. Init Halos.
1.a) In 1D decomposition nhalo and shalo intialized from vector x_in
1.b) In 2D decompsition nhalo,shalo, ehalo and whalo initialozed from vector x_in
2. Pass the halos to Jacobi_kernal.
3. Store the result computed at the boundary into the halo boundary positions.
4. Swap nhalo and shalo pairs in 1D decompostion. Swap (nhalo,shalo) and (ehalo,whalo) in 2D.
*/
//=================================Domain Decomposition Logic Starts=================================================================
/*Generating a GPU Grid with multiple GPUs and creating a Topology*/
int numberOfDevicesAlong_X = 1;
int numberOfDevicesAlong_Y = 1;
generateGPUGRID(numDevices, numberOfDevicesAlong_X, numberOfDevicesAlong_Y);
cout << "GPU grid structure is : " << numberOfDevicesAlong_X << " X " << numberOfDevicesAlong_Y << endl;
//Set Decomposition dimension 1D or 2D: when decomposition is 0. Computation happens on a single GPU
int decom_Dim = 2;
//Total elements along each dim in 2D
int chunk_X = dim / numberOfDevicesAlong_X;
int chunk_Y = dim / numberOfDevicesAlong_Y;
/* Creating a GPU topology with multiple devices*/
createTopology(numDevices, deviceArray, numberOfDevicesAlong_X, numberOfDevicesAlong_Y);
//Let the total number of GPU be 2 : has to be changed later
//Computation divided into (size/2) on first and size-(size/2) on second
std::vector<int> domainDivision(numDevices);
//Logic for total chunk per device (Domain distribution)
for (int i = 0; i < numDevices; i++) {
//Chunk per GPU will be same irrepective of 1D or 2D decomposition
domainDivision[i] = size / numDevices;
}
//For use on Device
std::vector<float*>d_A0(numDevices);
std::vector<float*>d_A1(numDevices);
std::vector<float*>d_A2(numDevices);
std::vector<float*>d_A3(numDevices);
std::vector<float*>d_A4(numDevices);
std::vector<float*>d_Vec_In(numDevices);
std::vector<float*>d_Vec_Out(numDevices);
std::vector<float*>d_nhalos(numDevices);
std::vector<float*>d_shalos(numDevices);
std::vector<float*>d_ehalos(numDevices);
std::vector<float*>d_whalos(numDevices);
std::vector<float*>d_Rhs(numDevices);
std::vector<float*>x_buffer(numDevices);
std::vector<float*>y_buffer(numDevices);
//Note: Using Pinned memory on Host for Halos -> Performance Approach 1
vector<float*>nHalo_pinned(numDevices);
vector<float*>sHalo_pinned(numDevices);
vector<float*>wHalo_pinned(numDevices);
vector<float*>eHalo_pinned(numDevices);
for (int dev = 0;dev < numDevices;dev++)
{
cudaSetDevice(dev);
cudaMallocHost((void**)&nHalo_pinned[dev], (chunk_X) * sizeof(float));
cudaMallocHost((void**)&sHalo_pinned[dev], (chunk_X) * sizeof(float));
cudaMallocHost((void**)&wHalo_pinned[dev], (chunk_Y) * sizeof(float));
cudaMallocHost((void**)&eHalo_pinned[dev], (chunk_Y) * sizeof(float));
}
for (int dev = 0; dev < numDevices; dev++)
{
//Setting the device before allocation
cudaSetDevice(dev);
//cudamalloc the Diagonals
cudaMalloc((void**)&d_A0[dev], domainDivision[dev] * sizeof(float));
cudaMalloc((void**)&d_A1[dev], domainDivision[dev] * sizeof(float));
cudaMalloc((void**)&d_A2[dev], domainDivision[dev] * sizeof(float));
cudaMalloc((void**)&d_A3[dev], domainDivision[dev] * sizeof(float));
cudaMalloc((void**)&d_A4[dev], domainDivision[dev] * sizeof(float));
//Using pinned memory as part of performance upgrade- Phase 2 of development
//cudamalloc the Input Vector and Result vector
cudaMalloc((void**)&d_Vec_In[dev], domainDivision[dev] * sizeof(float));
cudaMalloc((void**)&d_Vec_Out[dev], domainDivision[dev] * sizeof(float));
cudaMalloc((void**)&d_Rhs[dev], domainDivision[dev] * sizeof(float));
//cudaMalloc Halos: North and South--1D. TODO: East and West for 2D
cudaMalloc((void**)&d_nhalos[dev], chunk_X * sizeof(float));
cudaMalloc((void**)&d_shalos[dev], chunk_X * sizeof(float));
cudaMalloc((void**)&d_ehalos[dev], chunk_Y * sizeof(float));
cudaMalloc((void**)&d_whalos[dev], chunk_Y * sizeof(float));
//Buffer memory used for p2p exchange
cudaMalloc((void**)&x_buffer[dev], chunk_X * sizeof(float));
cudaMalloc((void**)&y_buffer[dev], chunk_Y * sizeof(float));
}
/* The transfer of Data from Host to Device : Domain Decomposition in 2D*/
if (decom_Dim == 2) {
//Create Partial Diagonal Vectors
//Size per GPU will be
int chunkSize = chunk_X * chunk_Y;
std::vector<float> partial_a0(chunkSize);
std::vector<float> partial_a1(chunkSize);
std::vector<float> partial_a2(chunkSize);
std::vector<float> partial_a3(chunkSize);
std::vector<float> partial_a4(chunkSize);
std::vector<float> partial_vec_in(chunkSize);
std::vector<float> partial_vec_out(chunkSize);
std::vector<float> partial_rhs(chunkSize);
std::vector<float> partial_result(chunkSize);
for (int dev = 0; dev < numDevices; dev++)
{
//Test the properties of the device assigned
//cout << endl << "New Logical Device created " << deviceArray[dev].deviceID;
//cout << endl << "New Logical Device (X,Y) coord (" << deviceArray[dev].devicePosition_X << "," << deviceArray[dev].devicePosition_Y << ")";
//==========Important: Logic for creation of Chunks to be allocated to GPUs==========================================
//Important : Mention about the correlation between the topology and data position in the thesis
int devicePosX = deviceArray[dev].devicePosition_X;
int devicePosY = deviceArray[dev].devicePosition_Y;
//Calculating data position based on device coords
//numberOfDevicesAlong_X * Chunk_X * Chunk_Y : finds out the total data per row of GPUs allocated
int dataStartPos_X = (devicePosX * numberOfDevicesAlong_X * chunk_X * chunk_Y) + (devicePosY * chunk_X);
int dataEndPos_X = dataStartPos_X + chunk_X;
//One complete row across all GPU is dim in order to get the next element above an element we add (currentPosition + dim )
int rowStartPos = dataStartPos_X;
int rowEndPos = dataEndPos_X;
int indexCounter = 0;
//Initialize Halos
initHalos2D(deviceArray[dev], chunk_X, chunk_Y, &vec_in[0], numberOfDevicesAlong_X, numberOfDevicesAlong_Y, rowStartPos, rowEndPos - 1, dim);
for (int rowNum = 0; rowNum < chunk_Y; rowNum++)
{
//Get one complete row for the GPU
for (int pos = rowStartPos; pos < rowEndPos; pos++)
{
partial_a0[indexCounter] = a0[pos];
partial_a1[indexCounter] = a1[pos];
partial_a2[indexCounter] = a2[pos];
partial_a3[indexCounter] = a3[pos];
partial_a4[indexCounter] = a4[pos];
partial_vec_in[indexCounter] = vec_in[pos];
partial_vec_out[indexCounter] = vec_out[pos];
partial_rhs[indexCounter] = rhs[pos];
partial_result[indexCounter] = result[pos];
indexCounter++;
}
rowStartPos += dim;
rowEndPos += dim;
}
//==========Important: Logic for creation of Chunks to be allocated to GPUs Ends ==========================================
//Setting Cuda device
cudaSetDevice(dev);
//Copy the diagonals from host to device : calling all at once instead of putting inside the for loop
cudaMemcpy(d_A0[dev], &partial_a0[0], domainDivision[dev] * sizeof(float), cudaMemcpyHostToDevice);
cudaMemcpy(d_A1[dev], &partial_a1[0], domainDivision[dev] * sizeof(float), cudaMemcpyHostToDevice);
cudaMemcpy(d_A2[dev], &partial_a2[0], domainDivision[dev] * sizeof(float), cudaMemcpyHostToDevice);
cudaMemcpy(d_A3[dev], &partial_a3[0], domainDivision[dev] * sizeof(float), cudaMemcpyHostToDevice);
cudaMemcpy(d_A4[dev], &partial_a4[0], domainDivision[dev] * sizeof(float), cudaMemcpyHostToDevice);
//Copy in and out vectors and RHS
cudaMemcpy(d_Vec_In[dev], &partial_vec_in[0], domainDivision[dev] * sizeof(float), cudaMemcpyHostToDevice);
cudaMemcpy(d_Vec_Out[dev], &partial_vec_out[0], domainDivision[dev] * sizeof(float), cudaMemcpyHostToDevice);
cudaMemcpy(d_Rhs[dev], &partial_rhs[0], domainDivision[dev] * sizeof(float), cudaMemcpyHostToDevice);
}
if (auto err = cudaGetLastError())
{
cout << "Data copy failed 1: " << cudaGetErrorString(err) << endl;
return err;
}
//Copy intial Halos in 2D
//Initial Exchange Halos: Then do intial cudaMemcopies
exchangehalos_onHost(numDevices, deviceArray, numberOfDevicesAlong_X);
for (int dev = 0; dev < numDevices; dev++)
{
cudaSetDevice(dev);
//Copying Halos to the device
if (deviceArray[dev].nHalo_flag == 1)
{
cudaMemcpy(d_nhalos[dev], &deviceArray[dev].nHalo[0], chunk_X * sizeof(float), cudaMemcpyHostToDevice);
}
if (deviceArray[dev].sHalo_flag == 1)
{
cudaMemcpy(d_shalos[dev], &deviceArray[dev].sHalo[0], chunk_X * sizeof(float), cudaMemcpyHostToDevice);
}
if (deviceArray[dev].eHalo_flag == 1)
{
cudaMemcpy(d_ehalos[dev], &deviceArray[dev].eHalo[0], chunk_Y * sizeof(float), cudaMemcpyHostToDevice);
}
if (deviceArray[dev].wHalo_flag == 1)
{
cudaMemcpy(d_whalos[dev], &deviceArray[dev].wHalo[0], chunk_Y * sizeof(float), cudaMemcpyHostToDevice);
}
}
//Development phase 2 changes : For p2p operation communication initialize buffers
for (int dev = 0; dev < numDevices; dev++)
{
cudaSetDevice(dev);
//Copying Halos to the device
if (deviceArray[dev].nHalo_flag == 1)
{
cudaMemcpy(x_buffer[dev], &deviceArray[dev].nHalo[0], chunk_X * sizeof(float), cudaMemcpyHostToDevice);
}
if (deviceArray[dev].wHalo_flag == 1)
{
cudaMemcpy(y_buffer[dev], &deviceArray[dev].eHalo[0], chunk_Y * sizeof(float), cudaMemcpyHostToDevice);
}
}