-
Notifications
You must be signed in to change notification settings - Fork 6
/
loop_cuda_rose.cc
3004 lines (2488 loc) · 108 KB
/
loop_cuda_rose.cc
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
/*****************************************************************************
Copyright (C) 2009 University of Utah
All Rights Reserved.
Purpose:
Cudaize methods
Notes:
History:
1/7/10 Created by Gabe Rudy by migrating code from loop.cc
31/1/11 Modified by Protonu Basu
*****************************************************************************/
#ifdef FRONTEND_ROSE
#include <malloc.h>
//#define TRANSFORMATION_FILE_INFO Sg_File_Info::generateDefaultFileInfoForTransformationNode()
// these are very similar (the same?)
#include "loop_cuda_chill.hh"
#include <code_gen/CG_stringBuilder.h>
#include <codegen.h>
#include <code_gen/CG_utils.h>
#include "loop.hh"
#include <math.h>
//#include <useful.h>
#include "omegatools.hh"
#include "ir_cudarose.hh"
#include "ir_rose.hh"
#include "ir_rose_utils.hh"
#include "chill_error.hh"
#include <vector>
#include "Outliner.hh"
//#define DEBUG
using namespace omega;
using namespace SageBuilder;
using namespace SageInterface;
namespace omega {
extern char* k_cuda_texture_memory; //protonu--added to track texture memory type
}
extern char *k_ocg_comment;
extern bool cudaDebug;
std::string& upcase(std::string& s) {
for (int i = 0; i < s.size(); i++)
s[i] = toupper(s[i]);
return s;
}
void printVs(const std::vector<std::string>& curOrder) {
if (!cudaDebug) return;
for (int i = 0; i < curOrder.size(); i++) {
if (i > 0)
printf(",");
printf("%s", curOrder[i].c_str());
}
printf("\n");
fflush(stdout);
}
void printVS(const std::vector<std::string>& curOrder) {
if(!cudaDebug) return;
for (int i = 0; i < curOrder.size(); i++) {
if (i > 0)
printf(",");
printf("%s", curOrder[i].c_str());
}
printf("\n");
fflush(stdout);
}
LoopCuda::~LoopCuda() {
const int m = stmt.size();
for (int i = 0; i < m; i++)
stmt[i].code->clear();
}
bool LoopCuda::symbolExists(std::string s) {
debug_fprintf(stderr, "LoopCuda::symbolExists( %s ) TODO loop_cuda_rose.cc L89\nDIE\n", s.c_str());
exit(-1); // DFL
/*
if (body_symtab->find_variable(SgName(s.c_str())) commented OUT
|| parameter_symtab->find_variable(SgName(s.c_str())))
return true;
if (globals->lookup_variable_symbol(SgName(s.c_str())))
return true;
for (int i = 0; i < idxNames.size(); i++)
for (int j = 0; j < idxNames[i].size(); j++)
if (strcmp(idxNames[i][j].c_str(), s.c_str()) == 0)
return true;
*/
return false;
}
void LoopCuda::addSync(int stmt_num, std::string idxName) {
debug_fprintf(stderr, "addsync\n");
//we store these and code-gen inserts sync to omega comments where stmt
//in loop that has idxName being generated
syncs.push_back(make_pair(stmt_num, idxName));
}
void LoopCuda::renameIndex(int stmt_num, std::string idx, std::string newName) {
int level = findCurLevel(stmt_num, idx);
if (idxNames.size() <= stmt_num || idxNames[stmt_num].size() < level)
throw std::runtime_error("Invalid statment number of index");
idxNames[stmt_num][level - 1] = newName.c_str();
}
enum Type {
Int
};
int wrapInIfFromMinBound(chillAST_node* then_part,
chillAST_ForStmt* loop,
chillAST_node *symtab,
chillAST_node* bound_sym) {
debug_fprintf(stderr, "wrapInIfFromMinBound()\n"); exit(-1); // DFL
return 0;
/*
SgNode* wrapInIfFromMinBound(SgNode* then_part, SgForStatement* loop,
SgScopeStatement* symtab, SgVariableSymbol* bound_sym) {
// CG_roseBuilder *ocg = new CG_roseBuilder(
SgBinaryOp* test_expr = isSgBinaryOp(loop->get_test_expr());
SgExpression* upperBound;
SgExpression* conditional;
upperBound = test_expr->get_rhs_operand();
CG_outputRepr *ifstmt;
SgCallExpression *call;
if (call = isSgCallExpression(upperBound))
if (isSgVarRefExp(call->get_function())->get_symbol()->get_name().getString()
== "__rose_lt") {
SgExprListExp* arg_list = call->get_args();
SgExpression *if_bound = *(arg_list->get_expressions().begin());
//This relies on the minimum expression being the rhs operand of
// the min instruction.
//
SgIfStmt *ifstmt = buildIfStmt(
buildLessOrEqualOp(buildVarRefExp(bound_sym), if_bound),
isSgStatement(then_part), NULL);
return isSgNode(ifstmt);
}
/* if (isSgConditionalExp(upperBound)) {
conditional = isSgConditionalExp(upperBound)->get_conditional_exp();
if (isSgBinaryOp(conditional)) {
SgBinaryOp* binop = isSgBinaryOp(conditional);
if (isSgLessThanOp(binop) || isSgLessOrEqualOp(binop)) {
SgIfStmt *ifstmt = buildIfStmt(
buildLessOrEqualOp(buildVarRefExp(bound_sym),
test_expr), isSgStatement(then_part), NULL);
return isSgNode(ifstmt);
}
}
}
*/
//return then_part;
}
/**
* This would be better if it was done by a CHiLL xformation instead of at codegen
*
* state:
* for(...)
* for(...)
* cur_body
* stmt1
*
* stm1 is in-between two loops that are going to be reduced. The
* solution is to put stmt1 at the end of cur_body but conditionally run
* in on the last step of the for loop.
*
* A CHiLL command that would work better:
*
* for(...)
* stmt0
* for(for i=0; i<n; i++)
* cur_body
* stmt1
* =>
* for(...)
* for(for i=0; i<n; i++)
* if(i==0) stmt0
* cur_body
* if(i==n-1) stmt1
*/
void findReplacePreferedIdxs(chillAST_node *newkernelcode,
chillAST_FunctionDecl *kernel ) {
debug_fprintf(stderr, "\nrecursiveFindReplacePreferedIdxs( sync 0 ) perhaps adding syncthreads\n");
std::vector<chillAST_VarDecl *>* symtab = kernel->getSymbolTable();
newkernelcode->findLoopIndexesToReplace( symtab, false );
}
void recursiveFindRefs(SgNode* code, std::set<const SgVariableSymbol *>& syms,
SgFunctionDefinition* def) {
SgStatement* s = isSgStatement(code);
// L = {symbols defined within 's'}, local variables declared within 's'
ASTtools::VarSymSet_t L;
ASTtools::collectDefdVarSyms(s, L);
//dump (L, "L = ");
// U = {symbols used within 's'}
ASTtools::VarSymSet_t U;
ASTtools::collectRefdVarSyms(s, U);
//dump (U, "U = ");
// U - L = {symbols used within 's' but not defined in 's'}
// variable references to non-local-declared variables
ASTtools::VarSymSet_t diff_U_L;
set_difference(U.begin(), U.end(), L.begin(), L.end(),
inserter(diff_U_L, diff_U_L.begin()));
//dump (diff_U_L, "U - L = ");
// Q = {symbols defined within the function surrounding 's' that are
// visible at 's'}, including function parameters
ASTtools::VarSymSet_t Q;
ASTtools::collectLocalVisibleVarSyms(def->get_declaration(), s, Q);
// dump (Q, "Q = ");
// (U - L) \cap Q = {variables that need to be passed as parameters
// to the outlined function}
// a sub set of variables that are not globally visible (no need to pass at all)
// It excludes the variables with a scope between global and the enclosing function
set_intersection(diff_U_L.begin(), diff_U_L.end(), Q.begin(), Q.end(),
inserter(syms, syms.begin()));
}
SgNode* recursiveFindReplacePreferedIdxs(SgNode* code, SgSymbolTable* body_syms,
SgSymbolTable* param_syms, SgScopeStatement* body,
std::map<std::string, SgVariableSymbol*>& loop_idxs,
SgGlobal* globalscope, bool sync = false) {
//tree_node_list* tnl = new tree_node_list;
//tree_node_list_iter tnli(code);
SgVariableSymbol* idxSym = 0;
std::vector<SgStatement*> r1;
std::vector<SgNode*> r2;
SgNode* tnli;
SgNode* tnli1;
SgNode* tnli2;
SgBasicBlock * clone;
if (isSgForStatement(code)) {
AstTextAttribute* att =
(AstTextAttribute*) (isSgNode(code)->getAttribute(
"omega_comment"));
std::string comment;
if (att != NULL)
comment = att->toString();
if (comment.find("~cuda~") != std::string::npos
&& comment.find("preferredIdx: ") != std::string::npos) {
std::string idx = comment.substr(
comment.find("preferredIdx: ") + 14, std::string::npos);
if (idx.find(" ") != std::string::npos)
idx = idx.substr(0, idx.find(" "));
if (loop_idxs.find(idx) != loop_idxs.end())
idxSym = loop_idxs.find(idx)->second;
//Get the proc variable sybol for this preferred index
if (idxSym == 0) {
idxSym = body_syms->find_variable(idx.c_str());
if (!idxSym)
idxSym = param_syms->find_variable(idx.c_str());
//printf("idx not found: lookup %p\n", idxSym);
if (!idxSym) {
SgVariableDeclaration* defn = buildVariableDeclaration(
SgName((char*) idx.c_str()), buildIntType());
//idxSym = new var_sym(type_s32, (char*)idx.c_str());
SgInitializedNamePtrList& variables = defn->get_variables();
SgInitializedNamePtrList::const_iterator i =
variables.begin();
SgInitializedName* initializedName = *i;
SgVariableSymbol* vs = new SgVariableSymbol(
initializedName);
prependStatement(defn, body);
vs->set_parent(body_syms);
body_syms->insert(SgName((char*) idx.c_str()), vs);
idxSym = vs;
//printf("idx created and inserted\n");
}
//Now insert into our map for future
if (cudaDebug) std::cout << idx << "\n\n";
loop_idxs.insert(make_pair(idx, idxSym));
}
//See if we have a sync as well
if (comment.find("sync") != std::string::npos) {
//printf("Inserting sync after current block\n");
sync = true;
}
}
if (idxSym) {
SgForInitStatement* list =
isSgForStatement(code)->get_for_init_stmt();
SgStatementPtrList& initStatements = list->get_init_stmt();
SgStatementPtrList::const_iterator j = initStatements.begin();
const SgVariableSymbol* index;
if (SgExprStatement *expr = isSgExprStatement(*j))
if (SgAssignOp* op = isSgAssignOp(expr->get_expression()))
if (SgVarRefExp* var_ref = isSgVarRefExp(
op->get_lhs_operand()))
index = var_ref->get_symbol();
std::vector<SgVarRefExp *> array = substitute(code, index, NULL,
isSgNode(body_syms));
for (int j = 0; j < array.size(); j++)
array[j]->set_symbol(idxSym);
}
SgStatement* body_ = isSgStatement(
recursiveFindReplacePreferedIdxs(
isSgNode((isSgForStatement(code)->get_loop_body())),
body_syms, param_syms, body, loop_idxs, globalscope));
omega::CG_roseRepr * tnl = new omega::CG_roseRepr(code);
omega::CG_outputRepr* block = tnl->clone();
tnli = static_cast<const omega::CG_roseRepr *>(block)->GetCode();
isSgForStatement(tnli)->set_loop_body(body_);
body_->set_parent(tnli);
if (idxSym) {
SgForInitStatement* list =
isSgForStatement(tnli)->get_for_init_stmt();
SgStatementPtrList& initStatements = list->get_init_stmt();
SgStatementPtrList::const_iterator j = initStatements.begin();
const SgVariableSymbol* index;
if (SgExprStatement *expr = isSgExprStatement(*j))
if (SgAssignOp* op = isSgAssignOp(expr->get_expression()))
if (SgVarRefExp* var_ref = isSgVarRefExp(
op->get_lhs_operand()))
index = var_ref->get_symbol();
std::vector<SgVarRefExp *> array = substitute(tnli, index, NULL,
isSgNode(body_syms));
for (int j = 0; j < array.size(); j++)
array[j]->set_symbol(idxSym);
}
// std::cout << isSgNode(body_)->unparseToString() << "\n\n";
if (att != NULL)
tnli->setAttribute("omega_comment", att);
if (sync) {
SgName name_syncthreads("__syncthreads");
SgFunctionSymbol * syncthreads_symbol =
globalscope->lookup_function_symbol(name_syncthreads);
// Create a call to __syncthreads():
SgFunctionCallExp * syncthreads_call = buildFunctionCallExp(
syncthreads_symbol, buildExprListExp());
SgExprStatement* stmt = buildExprStatement(syncthreads_call);
/* if (SgBasicBlock* bb = isSgBasicBlock(
isSgForStatement(code)->get_loop_body()))
appendStatement(isSgStatement(stmt), bb);
else if (SgStatement* ss = isSgStatement(
isSgForStatement(code)->get_loop_body())) {
SgBasicBlock* bb2 = buildBasicBlock();
isSgNode(ss)->set_parent(bb2);
appendStatement(ss, bb2);
appendStatement(isSgStatement(stmt), bb2);
isSgNode(stmt)->set_parent(bb2);
isSgForStatement(code)->set_loop_body(bb2);
isSgNode(bb2)->set_parent(code);
}
*/
SgBasicBlock* bb2 = buildBasicBlock();
bb2->append_statement(isSgStatement(tnli));
bb2->append_statement(stmt);
/* SgNode* parent = code->get_parent();
if(!isSgStatement(parent))
throw loop_error("Parent not a statement");
if(isSgForStatement(parent)){
if(SgStatement *ss = isSgForStatement(isSgForStatement(parent)->get_loop_body())){
omega::CG_roseRepr * tnl = new omega::CG_roseRepr(ss);
omega::CG_outputRepr* block= tnl->clone();
SgNode *new_ss = static_cast<const omega::CG_roseRepr *>(block)->GetCode();
SgBasicBlock* bb2 = buildBasicBlock();
isSgNode(new_ss)->set_parent(bb2);
appendStatement(isSgStatement(new_ss), bb2);
appendStatement(isSgStatement(stmt), bb2);
isSgNode(stmt)->set_parent(bb2);
isSgStatement(parent)->replace_statement_from_basicBlock(ss, isSgStatement(bb2));
}else if(isSgBasicBlock(isSgForStatement(parent)->get_loop_body()))
isSgStatement(isSgForStatement(parent)->get_loop_body())->insert_statement(isSgStatement(code), stmt, false);
else
throw loop_error("parent statement type undefined!!");
}
else if(isSgBasicBlock(parent))
isSgStatement(parent)->insert_statement(isSgStatement(code), stmt, false);
else
throw loop_error("parent statement type undefined!!");
//tnl->print();
*
*
*/
sync = true;
return isSgNode(bb2);
} else
return tnli;
} else if (isSgIfStmt(code)) {
SgStatement* body_ = isSgStatement(
recursiveFindReplacePreferedIdxs(
isSgNode((isSgIfStmt(code)->get_true_body())),
body_syms, param_syms, body, loop_idxs, globalscope));
omega::CG_roseRepr * tnl = new omega::CG_roseRepr(code);
omega::CG_outputRepr* block = tnl->clone();
tnli = static_cast<const omega::CG_roseRepr *>(block)->GetCode();
isSgIfStmt(tnli)->set_true_body(body_);
if ((isSgIfStmt(code)->get_false_body()))
isSgIfStmt(tnli)->set_false_body(
isSgStatement(
recursiveFindReplacePreferedIdxs(
isSgNode(
(isSgIfStmt(code)->get_false_body())),
body_syms, param_syms, body, loop_idxs,
globalscope)));
return tnli;
} else if (isSgStatement(code) && !isSgBasicBlock(code)) {
omega::CG_roseRepr * tnl = new omega::CG_roseRepr(code);
omega::CG_outputRepr* block = tnl->clone();
tnli = static_cast<const omega::CG_roseRepr *>(block)->GetCode();
return tnli;
} else if (isSgBasicBlock(code)) {
SgStatementPtrList& tnl = isSgBasicBlock(code)->get_statements();
SgStatementPtrList::iterator temp;
clone = buildBasicBlock();
bool sync_found = false;
for (SgStatementPtrList::const_iterator it = tnl.begin();
it != tnl.end(); it++) {
if (isSgForStatement(*it)) {
AstTextAttribute* att =
(AstTextAttribute*) (isSgNode(*it)->getAttribute(
"omega_comment"));
std::string comment;
if (att != NULL)
comment = att->toString();
if (comment.find("~cuda~") != std::string::npos
&& comment.find("preferredIdx: ")
!= std::string::npos) {
std::string idx = comment.substr(
comment.find("preferredIdx: ") + 14,
std::string::npos);
if (idx.find(" ") != std::string::npos)
idx = idx.substr(0, idx.find(" "));
//printf("sym_tab preferred index: %s\n", idx.c_str());
if (loop_idxs.find(idx) != loop_idxs.end())
idxSym = loop_idxs.find(idx)->second;
//Get the proc variable sybol for this preferred index
if (idxSym == 0) {
idxSym = body_syms->find_variable(idx.c_str());
if (!idxSym)
idxSym = param_syms->find_variable(idx.c_str());
//printf("idx not found: lookup %p\n", idxSym);
if (!idxSym) {
SgVariableDeclaration* defn =
buildVariableDeclaration(
SgName((char*) idx.c_str()),
buildIntType());
//idxSym = new var_sym(type_s32, (char*)idx.c_str());
SgInitializedNamePtrList& variables =
defn->get_variables();
SgInitializedNamePtrList::const_iterator i =
variables.begin();
SgInitializedName* initializedName = *i;
SgVariableSymbol* vs = new SgVariableSymbol(
initializedName);
prependStatement(defn, body);
vs->set_parent(body_syms);
body_syms->insert(SgName((char*) idx.c_str()), vs);
//printf("idx created and inserted\n");
idxSym = vs;
}
//Now insert into our map for future
if (cudaDebug) std::cout << idx << "\n\n";
loop_idxs.insert(make_pair(idx, idxSym));
}
//See if we have a sync as well
if (comment.find("sync") != std::string::npos) {
//printf("Inserting sync after current block\n");
sync = true;
}
}
if (idxSym) {
SgForInitStatement* list =
isSgForStatement(*it)->get_for_init_stmt();
SgStatementPtrList& initStatements = list->get_init_stmt();
SgStatementPtrList::const_iterator j =
initStatements.begin();
const SgVariableSymbol* index;
if (SgExprStatement *expr = isSgExprStatement(*j))
if (SgAssignOp* op = isSgAssignOp(
expr->get_expression()))
if (SgVarRefExp* var_ref = isSgVarRefExp(
op->get_lhs_operand()))
index = var_ref->get_symbol();
std::vector<SgVarRefExp *> array = substitute(*it, index,
NULL, isSgNode(body_syms));
for (int j = 0; j < array.size(); j++)
array[j]->set_symbol(idxSym);
}
SgStatement* body_ =
isSgStatement(
recursiveFindReplacePreferedIdxs(
isSgNode(
(isSgForStatement(*it)->get_loop_body())),
body_syms, param_syms, body, loop_idxs,
globalscope));
omega::CG_roseRepr * tnl = new omega::CG_roseRepr(*it);
omega::CG_outputRepr* block = tnl->clone();
tnli =
static_cast<const omega::CG_roseRepr *>(block)->GetCode();
isSgForStatement(tnli)->set_loop_body(body_);
body_->set_parent(tnli);
if (idxSym) {
SgForInitStatement* list =
isSgForStatement(tnli)->get_for_init_stmt();
SgStatementPtrList& initStatements = list->get_init_stmt();
SgStatementPtrList::const_iterator j =
initStatements.begin();
const SgVariableSymbol* index;
if (SgExprStatement *expr = isSgExprStatement(*j))
if (SgAssignOp* op = isSgAssignOp(
expr->get_expression()))
if (SgVarRefExp* var_ref = isSgVarRefExp(
op->get_lhs_operand()))
index = var_ref->get_symbol();
std::vector<SgVarRefExp *> array = substitute(tnli, index,
NULL, isSgNode(body_syms));
for (int j = 0; j < array.size(); j++)
array[j]->set_symbol(idxSym);
}
idxSym = 0;
// std::cout << isSgNode(body_)->unparseToString() << "\n\n";
if (att != NULL)
tnli->setAttribute("omega_comment", att);
clone->append_statement(isSgStatement(tnli));
if (sync) {
SgName name_syncthreads("__syncthreads");
SgFunctionSymbol * syncthreads_symbol =
globalscope->lookup_function_symbol(
name_syncthreads);
// Create a call to __syncthreads():
SgFunctionCallExp * syncthreads_call = buildFunctionCallExp(
syncthreads_symbol, buildExprListExp());
SgExprStatement* stmt = buildExprStatement(
syncthreads_call);
/* if (SgBasicBlock* bb = isSgBasicBlock(
isSgForStatement(code)->get_loop_body()))
appendStatement(isSgStatement(stmt), bb);
else if (SgStatement* ss = isSgStatement(
isSgForStatement(code)->get_loop_body())) {
SgBasicBlock* bb2 = buildBasicBlock();
isSgNode(ss)->set_parent(bb2);
appendStatement(ss, bb2);
appendStatement(isSgStatement(stmt), bb2);
isSgNode(stmt)->set_parent(bb2);
isSgForStatement(code)->set_loop_body(bb2);
isSgNode(bb2)->set_parent(code);
}
*/
//SgBasicBlock* bb2 = buildBasicBlock();
clone->append_statement(stmt);
/* SgNode* parent = code->get_parent();
if(!isSgStatement(parent))
throw loop_error("Parent not a statement");
if(isSgForStatement(parent)){
if(SgStatement *ss = isSgForStatement(isSgForStatement(parent)->get_loop_body())){
omega::CG_roseRepr * tnl = new omega::CG_roseRepr(ss);
omega::CG_outputRepr* block= tnl->clone();
SgNode *new_ss = static_cast<const omega::CG_roseRepr *>(block)->GetCode();
SgBasicBlock* bb2 = buildBasicBlock();
isSgNode(new_ss)->set_parent(bb2);
appendStatement(isSgStatement(new_ss), bb2);
appendStatement(isSgStatement(stmt), bb2);
isSgNode(stmt)->set_parent(bb2);
isSgStatement(parent)->replace_statement_from_basicBlock(ss, isSgStatement(bb2));
}else if(isSgBasicBlock(isSgForStatement(parent)->get_loop_body()))
isSgStatement(isSgForStatement(parent)->get_loop_body())->insert_statement(isSgStatement(code), stmt, false);
else
throw loop_error("parent statement type undefined!!");
}
else if(isSgBasicBlock(parent))
isSgStatement(parent)->insert_statement(isSgStatement(code), stmt, false);
else
throw loop_error("parent statement type undefined!!");
//tnl->print();
*
*
*/
sync = true;
// return isSgNode(bb2);
}
// return tnli;
} else if (isSgIfStmt(*it)) {
SgStatement* body_ = isSgStatement(
recursiveFindReplacePreferedIdxs(
isSgNode((isSgIfStmt(*it)->get_true_body())),
body_syms, param_syms, body, loop_idxs,
globalscope));
omega::CG_roseRepr * tnl = new omega::CG_roseRepr(*it);
omega::CG_outputRepr* block = tnl->clone();
tnli1 =
static_cast<const omega::CG_roseRepr *>(block)->GetCode();
isSgIfStmt(tnli1)->set_true_body(body_);
body_->set_parent(tnli1);
if ((isSgIfStmt(*it)->get_false_body()))
isSgIfStmt(tnli1)->set_false_body(
isSgStatement(
recursiveFindReplacePreferedIdxs(
isSgNode(
(isSgIfStmt(*it)->get_false_body())),
body_syms, param_syms, body,
loop_idxs, globalscope)));
clone->append_statement(isSgStatement(tnli1));
//return tnli;
} else if (isSgStatement(*it)) {
omega::CG_roseRepr * tnl = new omega::CG_roseRepr(*it);
omega::CG_outputRepr* block = tnl->clone();
tnli2 =
static_cast<const omega::CG_roseRepr *>(block)->GetCode();
clone->append_statement(isSgStatement(tnli2));
//return tnli;
}
}
return isSgNode(clone);
}
}
// loop_vars -> array references
// loop_idxs -> <idx_name,idx_sym> map for when we encounter a loop with a different preferredIndex
// dim_vars -> out param, fills with <old,new> var_sym pair for 2D array dimensions (messy stuff)
void swapVarReferences( chillAST_node *newkernelcode,
chillAST_FunctionDecl *kernel ) {
// find what variables are referenced in the new kernel code
vector<chillAST_VarDecl*> newdecls;
newkernelcode->gatherVarDecls( newdecls );
debug_fprintf(stderr, "%d variables in kernel\n", newdecls.size());
for (int i=0; i<newdecls.size(); i++) {
debug_fprintf(stderr, "variable name %s ", newdecls[i]->varname);
chillAST_VarDecl *isParam = kernel->hasParameterNamed( newdecls[i]->varname );
chillAST_VarDecl *isLocalVar = kernel->hasVariableNamed( newdecls[i]->varname );
if (isParam) { debug_fprintf(stderr, "is a parameter\n"); }
if (isLocalVar) { debug_fprintf(stderr, "is already defined in the kernel\n"); }
if (!isParam && (!isLocalVar)) {
debug_fprintf(stderr, "needed to be added to kernel symbol table\n");
kernel->addDecl( newdecls[i] );
kernel->addChild( newdecls[i] ); // adds to body!
}
}
}
bool LoopCuda::validIndexes(int stmt, const std::vector<std::string>& idxs) {
for (int i = 0; i < idxs.size(); i++) {
bool found = false;
for (int j = 0; j < idxNames[stmt].size(); j++) {
if (strcmp(idxNames[stmt][j].c_str(), idxs[i].c_str()) == 0) {
found = true;
}
}
if (!found) {
return false;
}
}
return true;
}
bool LoopCuda::cudaize_v2(std::string kernel_name,
std::map<std::string, int> array_dims,
std::vector<std::string> blockIdxs,
std::vector<std::string> threadIdxs) {
debug_fprintf(stderr, " LoopCuda::cudaize_v2() ROSE\n");
exit(-1);
for(std::map<std::string, int>::iterator it = array_dims->begin(); it != array_dims->end(); it++) {
debug_fprintf(stderr, "array_dims '%s' %d\n", it->first.c_str(), it->second)
}
this->array_dims = array_dims;
debug_fprintf(stderr, "\nBEFORE ir->builder()\n");
CG_outputBuilder *ocg = ir->builder();
debug_fprintf(stderr, "AFTER ir->builder()\n");
int stmt_num = 0;
if (cudaDebug) {
printf("cudaize_v2(%s, {", kernel_name.c_str());
//for(
printf("}, blocks={");
printVs(blockIdxs);
printf("}, thread={");
printVs(threadIdxs);
printf("})\n");
fflush(stdout);
}
this->array_dims = array_dims;
if (!validIndexes(stmt_num, blockIdxs)) {
throw std::runtime_error("One of the indexes in the block list was not "
"found in the current set of indexes.");
}
if (!validIndexes(stmt_num, threadIdxs)) {
throw std::runtime_error(
"One of the indexes in the thread list was not "
"found in the current set of indexes.");
}
if (blockIdxs.size() == 0)
throw std::runtime_error("Cudaize: Need at least one block dimension");
int block_level = 0;
//Now, we will determine the actual size (if possible, otherwise
//complain) for the block dimensions and thread dimensions based on our
//indexes and the relations for our stmt;
for (int i = 0; i < blockIdxs.size(); i++) {
int level = findCurLevel(stmt_num, blockIdxs[i]);
int ub, lb;
CG_outputRepr* ubrepr = extractCudaUB(stmt_num, level, ub, lb);
fflush(stdout);
debug_fprintf(stderr, "lb %ldL\n", lb);
if (lb != 0) {
debug_fprintf(stderr, "lb != 0?\n");
//attempt to "normalize" the loop with an in-place tile and then re-check our bounds
if (cudaDebug) {
printf(
"Cudaize1: doing tile at level %d to try and normalize lower bounds\n",
level);
fflush(stdout);
}
debug_fprintf(stderr, "calling tile()\n");
tile(stmt_num, level, 1, level, CountedTile);
idxNames[stmt_num].insert(idxNames[stmt_num].begin() + (level), ""); //TODO: possibly handle this for all sibling stmts
ubrepr = extractCudaUB(stmt_num, level, ub, lb);
}
else debug_fprintf(stderr, "lb == 0?\n");
fflush(stdout);
debug_fprintf(stderr, "lb2 %ldL\n", lb);
if (lb != 0) {
debug_fprintf(stderr, "lb2 != 0?\n");
char buf[1024];
sprintf(buf,
"Cudaize: Loop at level %d does not have 0 as its lower bound",
level);
throw std::runtime_error(buf);
}
else debug_fprintf(stderr, "lb2 == 0?\n");
if (ub < 0) {
char buf[1024];
sprintf(buf,
"Cudaize: Loop at level %d does not have a hard upper bound",
level);
//Anand: Commenting out error indication for lack of constant upper bound
//throw std::runtime_error(buf);
}
if (cudaDebug) {
printf("block idx %s level %d lb: %d ub %d\n", blockIdxs[i].c_str(),
level, lb, ub);
fflush(stdout);
}
else debug_fprintf(stderr, "NO CUDADEBUG\n");
if (i == 0) {
block_level = level;
if (ubrepr == NULL) {
cu_bx = ub + 1;
cu_bx_repr = NULL;
} else {
cu_bx = 0;
cu_bx_repr = (omega::CG_chillRepr*)ocg->CreatePlus(ubrepr, ocg->CreateInt(1));
}
idxNames[stmt_num][level - 1] = "bx";
} else if (i == 1) {
if (ubrepr == NULL) {
cu_by = ub + 1; debug_fprintf(stderr, "cu_by = %d\n", cu_by);
cu_by_repr = NULL;
} else {
cu_by = 0; debug_fprintf(stderr, "cu_by = %d\n", cu_by);
cu_by_repr = (omega::CG_chillRepr*)ocg->CreatePlus(ubrepr, ocg->CreateInt(1));
}
idxNames[stmt_num][level - 1] = "by";
}
}
if (!cu_by && !cu_by_repr)
block_level = 0;
int thread_level1 = 0;
int thread_level2 = 0;
for (int i = 0; i < threadIdxs.size(); i++) {
int level = findCurLevel(stmt_num, threadIdxs[i]);
int ub, lb;
CG_outputRepr* ubrepr = extractCudaUB(stmt_num, level, ub, lb);
fflush(stdout);
debug_fprintf(stderr, "lb3 %ldL\n", lb);
if (lb != 0) {
debug_fprintf(stderr, "lb3 != 0?\n");
//attempt to "normalize" the loop with an in-place tile and then re-check our bounds
if (cudaDebug) {
printf(
"Cudaize2: doing tile at level %d to try and normalize lower bounds\n",
level);
fflush(stdout);
}
tile(stmt_num, level, 1, level, CountedTile);
idxNames[stmt_num].insert(idxNames[stmt_num].begin() + (level), "");
ubrepr = extractCudaUB(stmt_num, level, ub, lb);
}
else debug_fprintf(stderr, "lb3 == 0?\n");
if (lb != 0) {
char buf[1024];
sprintf(buf,
"Cudaize: Loop at level %d does not have 0 as it's lower bound",
level);
throw std::runtime_error(buf);
}
if (ub < 0) {
char buf[1024];
sprintf(buf,
"Cudaize: Loop at level %d does not have a hard upper bound",
level);
//Anand: Commenting out error indication for lack of constant upper bound
//throw std::runtime_error(buf);
}
if (cudaDebug)
printf("thread idx %s level %d lb: %d ub %d\n",
threadIdxs[i].c_str(), level, lb, ub);
if (i == 0) {
thread_level1 = level;
if (ubrepr == NULL) {
cu_tx = ub + 1;
cu_tx_repr = NULL;
} else {
cu_tx = 0;
cu_tx_repr = (omega::CG_chillRepr*)ocg->CreatePlus(ubrepr, ocg->CreateInt(1));
}
idxNames[stmt_num][level - 1] = "tx";
} else if (i == 1) {
thread_level2 = level;
if (ubrepr == NULL) {
cu_ty = ub + 1;
cu_ty_repr = NULL;
} else {
cu_ty = 0;
cu_ty_repr = (omega::CG_chillRepr*)ocg->CreatePlus(ubrepr, ocg->CreateInt(1));
}
idxNames[stmt_num][level - 1] = "ty";
} else if (i == 2) {
if (ubrepr == NULL) {
cu_tz = ub + 1;
cu_tz_repr = NULL;
} else {
cu_tz = 0;
cu_tz_repr = (omega::CG_chillRepr*)ocg->CreatePlus(ubrepr, ocg->CreateInt(1));
}
idxNames[stmt_num][level - 1] = "tz";
}
}
if (!cu_ty && !cu_ty_repr)
thread_level1 = 0;
if (!cu_tz && !cu_tz_repr)
thread_level2 = 0;
//Make changes to nonsplitlevels
const int m = stmt.size();
for (int i = 0; i < m; i++) {
if (block_level) {
//stmt[i].nonSplitLevels.append((block_level)*2);
stmt_nonSplitLevels[i].push_back((block_level) * 2);
}
if (thread_level1) {
//stmt[i].nonSplitLevels.append((thread_level1)*2);
stmt_nonSplitLevels[i].push_back((thread_level1) * 2);
}
if (thread_level2) {
//stmt[i].nonSplitLevels.append((thread_level1)*2);
stmt_nonSplitLevels[i].push_back((thread_level1) * 2);
}
}
if (cudaDebug) {
printf("cudaize_vman svn 2: current names: ");
printVS(idxNames[stmt_num]);
}
//Set codegen flag
code_gen_flags |= GenCudaizeV2;
//Save array dimension sizes
this->array_dims = array_dims;
cu_kernel_name = kernel_name.c_str();
}
chillAST_VarDecl *addBuiltin( char *nameofbuiltin, char *typeOfBuiltin, chillAST_node *somecode) {
// create a builtin, like "blockIdx.x", so we can use it in assignment statements
chillAST_VarDecl *bi = new chillAST_VarDecl( typeOfBuiltin, nameofbuiltin, "", NULL); // somecode ); // ->getSourceFile() );
bi->isABuiltin = true;
return bi;
}
/*
* setupConstantVar comment
* handles CUDA constant variable declaration