Hi, All I want to implement DSWP Which is used for parallelization of loops. For this purpose, the loop was replaced with a new basic block in main function. And new functions were created and basic blocks of Loop assigned to them.I have checked blocks and branches for Succ and Pred relation and I have not found any problems. However I get the following error: * **opt: /home/llvm/src/include/llvm/**Support/CFG.h:105: llvm::SuccIterator<Term_, BB_>::SuccIterator(Term_) [with Term_ llvm::TerminatorInst*, BB_ = llvm::BasicBlock]: Assertion `T && "getTerminator returned null!"' failed. 0 opt 0x0848e569 Stack dump: 0. Program arguments: opt -dot-cfg -dot-postdom -dot-dom -load /home/llvm/src/Release/lib/**parallel.so -parallel obj.o -o out.o 1. Running pass 'Function Pass Manager' on module 'obj.o'. 2. Running pass 'Loop Pass Manager' on function '@main' Aborted (core dumped)* I do not know why this error occurred.Who knows why this error occurs and what should I do? Best regards neda -------------- next part -------------- An HTML attachment was scrubbed... URL: <http://lists.llvm.org/pipermail/llvm-dev/attachments/20111013/01b35da6/attachment.html>
On 10/13/11 1:20 AM, neda 8664 wrote:> Hi, All > > I want to implement DSWP Which is used for parallelization of loops. > For this purpose, the loop was replaced with a new basic block in main > function. > And new functions were created and basic blocks of Loop assigned to > them.I have checked blocks and branches for Succ and Pred relation and > I have not found any problems. > However I get the following error: > / > //opt: /home/llvm/src/include/llvm///Support/CFG.h:105: > llvm::SuccIterator<Term_, BB_>::SuccIterator(Term_) [with Term_ = > llvm::TerminatorInst*, BB_ = llvm::BasicBlock]: Assertion `T && > "getTerminator returned null!"' failed. > 0 opt 0x0848e569 > Stack dump: > 0. Program arguments: opt -dot-cfg -dot-postdom -dot-dom -load > /home/llvm/src/Release/lib///parallel.so -parallel obj.o -o out.o > 1. Running pass 'Function Pass Manager' on module 'obj.o'. > 2. Running pass 'Loop Pass Manager' on function '@main' > Aborted (core dumped)/ > > > I do not know why this error occurred.Who knows why this error occurs > and what should I do?Every basic block in LLVM is required to have a Terminator instruction (http://llvm.org/docs/LangRef.html#terminators). I suspect that you've created a basic block without a Terminator instruction. Call the dump() method on all the basic blocks you added and see if they have a Terminator instruction. -- John T.> > > Best regards > neda > > > _______________________________________________ > LLVM Developers mailing list > LLVMdev at cs.uiuc.edu http://llvm.cs.uiuc.edu > http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev-------------- next part -------------- An HTML attachment was scrubbed... URL: <http://lists.llvm.org/pipermail/llvm-dev/attachments/20111013/3d797946/attachment.html>
This may also be a starting point for you. I think it was someone's class project: http://code.google.com/p/15745-project-dswp/ On Thu, Oct 13, 2011 at 12:20 AM, neda 8664 <neda8664 at gmail.com> wrote:> Hi, All > > I want to implement DSWP Which is used for parallelization of loops. For > this purpose, the loop was replaced with a new basic block in main function. > And new functions were created and basic blocks of Loop assigned to them.I > have checked blocks and branches for Succ and Pred relation and I have not > found any problems. > However I get the following error: > > opt: /home/llvm/src/include/llvm/Support/CFG.h:105: > llvm::SuccIterator<Term_, BB_>::SuccIterator(Term_) [with Term_ > llvm::TerminatorInst*, BB_ = llvm::BasicBlock]: Assertion `T && > "getTerminator returned null!"' failed. > 0 opt 0x0848e569 > Stack dump: > 0. Program arguments: opt -dot-cfg -dot-postdom -dot-dom -load > /home/llvm/src/Release/lib/parallel.so -parallel obj.o -o out.o > 1. Running pass 'Function Pass Manager' on module 'obj.o'. > 2. Running pass 'Loop Pass Manager' on function '@main' > Aborted (core dumped) > > > I do not know why this error occurred.Who knows why this error occurs and > what should I do? > > > Best regards > neda > > _______________________________________________ > LLVM Developers mailing list > LLVMdev at cs.uiuc.edu http://llvm.cs.uiuc.edu > http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev > >
thank john, i have used dump() , and each blocks had pred except entery block, i get the following output: fun main entry entry: %retval = alloca i32 %0 = alloca i32 %i = alloca i32 %j = alloca i32 %a = alloca [2 x i32] %n = alloca i32 %t = alloca i32 %"alloca point" = bitcast i32 0 to i32 %1 = getelementptr inbounds [2 x i32]* %a, i32 0, i32 0 %2 = load i32* getelementptr inbounds ([2 x i32]* @C.0.2045, i32 0, i32 0), align 4 store i32 %2, i32* %1, align 4 %3 = getelementptr inbounds [2 x i32]* %a, i32 0, i32 1 %4 = load i32* getelementptr inbounds ([2 x i32]* @C.0.2045, i32 0, i32 1), align 4 store i32 %4, i32* %3, align 4 store i32 2, i32* %n, align 4 store i32 0, i32* %j, align 4 br label %loop-replace fun main loop-replace loop-replace: ; preds = %entry %5 = alloca [3 x i64] %j_arg = ptrtoint i32* %j to i64 %6 = getelementptr [3 x i64]* %5, i64 0, i64 0 store i64 %j_arg, i64* %6 %t_arg = ptrtoint i32* %t to i64 %7 = getelementptr [3 x i64]* %5, i64 0, i64 1 store i64 %t_arg, i64* %7 %a_arg = ptrtoint [2 x i32]* %a to i64 %8 = getelementptr [3 x i64]* %5, i64 0, i64 2 store i64 %a_arg, i64* %8 call void @sync_init() %9 = bitcast [3 x i64]* %5 to i64* call void @sync_delegate(i32 0, i8* (i8*)* @"1_subloop_0", i64* %9) %10 = bitcast [3 x i64]* %5 to i64* call void @sync_delegate(i32 1, i8* (i8*)* @"1_subloop_1", i64* %10) call void @sync_join() br label %bb5 fun main bb5 bb5: ; preds = %loop-replace %11 = load i32* @sum, align 4 store i32 %11, i32* %0, align 4 %12 = load i32* %0, align 4 store i32 %12, i32* %retval, align 4 br label %return fun main return return: ; preds = %bb5 %retval6 = load i32* %retval ret i32 %retval6 fun 1_subloop_0 new-entry new-entry: call void @showPlace() %1 = bitcast i8* %0 to i64* %2 = getelementptr i64* %1, i64 0 %j_val = load i64* %2, align 8 call void @showValue(i64 %j_val) %3 = trunc i64 %j_val to i32 %4 = inttoptr i64 %j_val to i32* %5 = getelementptr i64* %1, i64 1 %t_val = load i64* %5, align 8 call void @showValue(i64 %t_val) %6 = trunc i64 %t_val to i32 %7 = inttoptr i64 %t_val to i32* %8 = getelementptr i64* %1, i64 2 %a_val = load i64* %8, align 8 call void @showValue(i64 %a_val) %9 = trunc i64 %a_val to i32 %10 = inttoptr i64 %a_val to [2 x i32]* br label %bb4_0 fun 1_subloop_0 bb1_0 bb1_0: ; No predecessors! %c0 = call i64 @sync_consume(i32 0) %c0_val = trunc i64 %c0 to i32 %t18_0 = load i32* @p, align 4 br label %bb3_0 fun 1_subloop_0 bb2_0 bb2_0: ; No predecessors! br label %bb3_0 fun 1_subloop_0 bb3_0 bb3_0: ; preds = %bb4_0, %bb2_0, %bb1_0 %t3_0 = load i32* %7, align 4 %t4_0 = getelementptr inbounds [2 x i32]* %10, i32 0, i32 %t3_0 %t5_0 = load i32* %t4_0, align 4 %t6_0 = load i32* @sum, align 4 %t7_0 = add nsw i32 %t5_0, %t6_0 store i32 %t7_0, i32* @sum, align 4 %t9_0 = load i32* @sum, align 4 %t10_0 = call i32 (i8*, ...)* @printf(i8* noalias getelementptr inbounds ([4 x i8]* @.str, i32 0, i32 0), i32 %t9_0) nounwind %t11_0 = load i32* %4, align 4 %t12_0 = add nsw i32 %t11_0, 1 store i32 %t12_0, i32* %4, align 4 br label %bb4_0 fun 1_subloop_0 bb4_0 bb4_0: ; preds = %bb3_0, %new-entry %t0_0 = load i32* %4, align 4 %t1_0 = icmp sle i32 %t0_0, 9 br i1 %t1_0, label %bb3_0, label %new-exit fun 1_subloop_0 new-exit new-exit: ; preds = %bb4_0 ret i8* null fun 1_subloop_1 new-entry new-entry: call void @showPlace() %1 = bitcast i8* %0 to i64* %2 = getelementptr i64* %1, i64 0 %j_val = load i64* %2, align 8 call void @showValue(i64 %j_val) %3 = trunc i64 %j_val to i32 %4 = inttoptr i64 %j_val to i32* %5 = getelementptr i64* %1, i64 1 %t_val = load i64* %5, align 8 call void @showValue(i64 %t_val) %6 = trunc i64 %t_val to i32 %7 = inttoptr i64 %t_val to i32* %8 = getelementptr i64* %1, i64 2 %a_val = load i64* %8, align 8 call void @showValue(i64 %a_val) %9 = trunc i64 %a_val to i32 %10 = inttoptr i64 %a_val to [2 x i32]* br label %bb4_1 fun 1_subloop_1 bb_1 bb_1: ; preds = %bb4_1 br i1 undef, label %bb1_1, label %bb2_1 fun 1_subloop_1 bb1_1 bb1_1: ; preds = %bb_1 %t15_1 = load i32* %7, align 4 %t16_1 = getelementptr inbounds [2 x i32]* %10, i32 0, i32 %t15_1 %t17_1 = load i32* %t16_1, align 4 %t19_1 = mul nsw i32 %t17_1, undef %t19_1_64 = sext i32 %t19_1 to i64 call void @sync_produce(i64 %t19_1_64, i32 0) store i32 %t19_1, i32* @p, align 4 br label %bb3_1 fun 1_subloop_1 bb2_1 bb2_1: ; preds = %bb_1 store i32 1, i32* @p, align 4 br label %bb3_1 fun 1_subloop_1 bb3_1 bb3_1: ; preds = %bb2_1, %bb1_1 br label %bb4_1 fun 1_subloop_1 bb4_1 bb4_1: ; preds = %bb3_1, %new-entry br i1 undef, label %bb_1, label %new-exit fun 1_subloop_1 new-exit new-exit: ; preds = %bb4_1 ret i8* null -------------- next part -------------- An HTML attachment was scrubbed... URL: <http://lists.llvm.org/pipermail/llvm-dev/attachments/20111013/a681ee22/attachment.html>
Hi neda,> i have used dump() , and each blocks had pred except entery block, i get the > following output:maybe there are also some basic blocks that you created but forgot to add to any function? Ciao, Duncan.
Hi
I have checked all blocks, each block have a Terminator instruction and each
blocks belongs to a function.
I'm really confused. I guess the problem is caused by the removal of the
Loop,The code is as follows:
* //every block to header (except the ones in the loop), will now
redirect to newblock
for (pred_iterator PI = pred_begin(header); PI != pred_end(header);
++PI) {
BasicBlock *pred = *PI;
if (L->contains(pred)) {
continue;
}
TerminatorInst *termInst = pred->getTerminator();
for (unsigned i = 0; i < termInst->getNumOperands(); i++) {
BasicBlock *bb = dyn_cast<BasicBlock>
(termInst->getOperand(i));
if (bb == header) {
termInst->setOperand(i,newBlock);
}
}
}*
* cout << "begin to delete loop" << endl;
for (Loop::block_iterator bi = L->block_begin(), bi2; bi
!L->block_end(); bi = bi2)
{
bi2 = bi;
bi2++;
BasicBlock * BB = *bi;
for (BasicBlock::iterator ii = BB->begin(), ii2; ii != BB->end();
ii= ii2)
{
ii2 = ii;
ii2++;
Instruction *inst = ii;
inst->replaceAllUsesWith(UndefValue::get(inst->getType()));
inst->eraseFromParent();
}
for (pred_iterator PI = pred_begin(BB); PI != pred_end(BB); ++PI)
{
BasicBlock *pred = *PI;
if (!L->contains(pred))
continue;
pred->getTerminator()->eraseFromParent();
}
BB->eraseFromParent();
}
*
May this code be causing the problem?
Best regards
neda
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20111014/9b755118/attachment.html>
Hi neda, general advice: (1) build LLVM with assertions enabled; (2) run your pass under valgrind. Ciao, Duncan.> I have checked all blocks, each block have a Terminator instruction and each > blocks belongs to a function. > I'mreallyconfused. I guess theproblemiscaused bytheremoval ofthe Loop,Thecodeis > as follows: > > > / //every block to header (except the ones in the loop), will now redirect to > newblock > for (pred_iterator PI = pred_begin(header); PI != pred_end(header); ++PI) { > BasicBlock *pred = *PI; > if (L->contains(pred)) { > continue; > } > TerminatorInst *termInst = pred->getTerminator(); > > for (unsigned i = 0; i < termInst->getNumOperands(); i++) { > BasicBlock *bb = dyn_cast<BasicBlock> (termInst->getOperand(i)); > if (bb == header) { > termInst->setOperand(i,newBlock); > } > } > }/ > > / cout << "begin to delete loop" << endl; > for (Loop::block_iterator bi = L->block_begin(), bi2; bi != L->block_end(); > bi = bi2) > { > bi2 = bi; > bi2++; > BasicBlock * BB = *bi; > for (BasicBlock::iterator ii = BB->begin(), ii2; ii != BB->end(); ii= ii2) > { > ii2 = ii; > ii2++; > Instruction *inst = ii; > inst->replaceAllUsesWith(UndefValue::get(inst->getType())); > inst->eraseFromParent(); > } > for (pred_iterator PI = pred_begin(BB); PI != pred_end(BB); ++PI) > { > BasicBlock *pred = *PI; > if (!L->contains(pred)) > continue; > pred->getTerminator()->eraseFromParent(); > } > BB->eraseFromParent(); > } > / > May this code be causing the problem? > > Best regards > neda >
Possibly Parallel Threads
- [LLVMdev] BasicBlock succ iterator
- [LLVMdev] BasicBlock succ iterator
- [LLVMdev] RFC: Better alternative to llvm.frameallocate for use in Windows EH
- [RFC] [Windows SEH][-EHa] Support Hardware Exception Handling
- [LLVMdev] RFC: Better alternative to llvm.frameallocate for use in Windows EH