similar to: [LLVMdev] idf_iterator and MachineFunctions

On 2007-03-18, at 03:22, Lang Hames wrote: > Is there a recommended way to find the final block (the one with > successors={}) in a machine function? This isn't a property of the CFG in the general case. However, the UnifyFunctionExitNodes transformation/analysis provides it. From getAnalysis<UnifyFunctionExitNodes>().getReturnBlock(), you can visit the basic block

On Sun, 18 Mar 2007, Lang Hames wrote: > I need to do an inverse-depth-first iteration over the basic blocks in > a machine function (i.e. starting from the last block and following > predecessor edges) for a liveness analysis I'm writing. As Gordon mentioned, there isn't a trivial way to do this, as there isn't a single exit node in the MBB CFG. > idf_iterator seems

Hi, Well, yes i did try your suggestion but i keep on running into a compilation problem. The error is: llvm[0]: Compiling Hello.cpp for Release build (PIC) /home/saraswat/llvm/llvm-2.1/include/llvm/ADT/GraphTraits.h: In instantiation of `llvm::GraphTraits<llvm::ilist_iterator<llvm::BasicBlock> >': Hello.cpp:59: instantiated from here

Hi Pabhat, Have you checked out DepthFirstIterator? (include/llvm/ADT/ DepthFirstIterator.h). It provides an iterator abstraction to perform a forward/reverse DFS traversal of a graph. Many of the LLVM datatypes that represent graphs have a template specialization of the GraphTraits<> class which allows separate algorithms to treat them as graphs, walk them, etc. (Both BasicBlock

Hi all, Is there a way to walk through ALL the predecessors of a basic block in a CFG. I tried to iterate over the preds using this method for (pred_iterator PI = pred_begin(BB), E = pred_end(BB); PI != E; ++I) { BasicBlock *PredBB = *PI; } but this only gives the immediate predecessors for a basic block. For example, in this sample control flow graph. entry -> bb1 -> bb2 -> bb4

Dear Trevor, I'm too lazy to convert your .dot file into a graph file, but I'll make some comments anyway. :) First, LLVM does not guarantee that a function has a single exit block. Multiple basic blocks can have a return instruction as their terminator instruction. There is a pass (Unify Function Exit nodes i.e., -mergereturn <http://llvm.org/docs/Passes.html#mergereturn>)

Hi, I'm confused about the entry and exit blocks of an LLVM CFG. I understand that every CFG has one and only one entry block, and this is confirmed by the existence of the getEntryBlock function in the Function class. But what about exit (a.k.a. return) blocks? At first I assumed that LLVM CFGs have one and only one exit block, too, but the following code is a counter-example:

On Mar 30, 2010, at 7:51 PM, John Criswell wrote: > I'm too lazy to convert your .dot file into a graph file What format should I have posted? (I'm not sure what you mean by "graph file".) I had thought that .dot was the preferred format here, since that's what LLVM generates (e.g., "opt -dot-cfg ..."). > First, LLVM does not guarantee that a function

Hi, A number of transforms are actually independent, so here's a partial fix. I updated the dependencies in a cluster of transforms: LowerSwitch, Mem2Reg, LowerSelect, LowerAllocations, UnifyFunctionExitNodes. The patch has been tested, but not extensively. PassManager doesn't complain, and the result of a test pass that requires all these (except for LowerAllocations) together works

Hi all, Why are these two passes conflicting: UnifyFunctionExitNodes and LowerSwitch? AU.addRequired(LowerSwitchID) works, so does AU.addRequired<unifyFunctionExitNodes>(), but the two together don't... opt: PassManagerT.h:387: void llvm::PassManagerT<Trait>::markPassUsed(const llvm::PassInfo*, llvm::Pass*) [with Trait = llvm::MTraits]: Assertion `getAnalysisOrNullUp(P)

On Jul 22, 2010, at 2:05 PM, John Criswell wrote: > If you write your pass as a ModulePass, then you can iterate over the > functions in any order that you want. I had considered that, but my FunctionPass depends on other passes processing the functions first: void MyPass::getAnalysisUsage(AnalysisUsage &AU) const { AU.addRequired<UnifyFunctionExitNodes>();

I have a ModulePass, which we'll call MP, that generates a dependency graph for an entire program. I want MP to require the UnifyFunctionExitNodes pass, which is a FunctionPass. Since its not possible to make a ModulePass depend on a FunctionPass, is my only choice to make MP a FunctionPass in which the runOnFunction() routine does nothing, and the doFinalization routine does all the

Trevor Harmon wrote: > On Jul 22, 2010, at 2:05 PM, John Criswell wrote: > > >> If you write your pass as a ModulePass, then you can iterate over the >> functions in any order that you want. >> > > I had considered that, but my FunctionPass depends on other passes > processing the functions first: > Two things to consider: 1) The PassManager

John Criswell wrote: > Trevor Harmon wrote: >> On Jul 22, 2010, at 2:05 PM, John Criswell wrote: >> >> >>> If you write your pass as a ModulePass, then you can iterate over the >>> functions in any order that you want. >>> >> >> I had considered that, but my FunctionPass depends on other passes >> processing the functions first:

Is there a facility with which we may automagically create a "fake" exit node, one that is the target of all BasicBlocks ending in return. This would be very helpful to IP dataflow analysis... Dave

Hi, I found an existing pass "CalledValuePropagation" that can solve the problem I raised a few days ago regarding the "callees" metadata ( https://groups.google.com/forum/#!topic/llvm-dev/yjtZVMH_aC4). Now I have difficulty in calling this pass in my own pass. In my own pass, I called "getAnalysis<CalledValuePropagationPass>()" and in the

>> All SingleSource and MultiSource tests failed in last run of FreeBSD >> nighttester (and in previous). >> Mostly with error: > > This looks like your crtend.a file did not get built correctly. Try doing > make clean/make install in llvm/runtime/ again. > gmake clean in llvm/runtime/ remove all build tree of nighttest starting with llvm/runtime/../.. And then i

Upon writing my register allocation algorithm, I am concerned that a load may be skipped due to a branch. For instance consider the psudocode int x = 1 ... if( false ) load x y = 3*x else y = 4*x ... The load happens in the false block because this is when the variable is first seen. The load is not preformed in the else block because the register allocator may see it loaded already. Do

Hi All, I've developed a pass to render machine functions as HTML pages with some accompanying information about liveness and register pressure. Current features: Renders machine functions, optionally displaying estimated register pressure for selected register classes, and liveness for selected intervals. The following command line options can be used to enable and customise the

On Wed, Mar 31, 2010 at 2:59 PM, Trevor Harmon <trevor.w.harmon at nasa.gov> wrote: > I'm wondering what would cause a CFG not to have a return block. The > comments in UnifyFunctionExitNodes.cpp say: "If there are no return > stmts in the Function, a null pointer is returned." But this doesn't > make sense; even an empty void function has a return block with a