similar to: [LLVMdev] Preserve RegionInfo after regionsimplify

Hi fellows, I am writing to ask what is the algorithm implemented in LLVM's RegionInfo.h/cpp. In the header file "RegionInfo.h", it says "Calculates a program structure tree built out of single entry single exit regions (defined in a 1994 paper called "The Program Structure Tree"). ... ... The algorithm to calculate these data structures however is COMPLETELY

Tobias, Thanks for taking the time to summarize all this. It's a great writeup. I'm moving the thread to llvm-dev. My responses below. On May 21, 2012, at 5:06 AM, Tobias Grosser <tobias at grosser.es> wrote: > First of all some information about the RegionInfo pass: > > ======================================================================= > The very first paper I

Hi Tobias, Thanks a lot for the detailed reply. I am working on several new optimizations for OpenCL kernels for a DSP architecture. The project itself has an NDA associated with it, so I cannot go into more details, but the source will be open to public after completion. One of the first steps is to serialize the work-items in a work-group (e.g., insert nested loops around REGIONs in

Thanks very much for the quick response. I have read the text many times, but it was not very clear to me why checking the two conditions involving dominance frontiers is equivalent to proving the pair {entry, exit} defines a refined region. I was asking for an mathematical proof really. It sounds to me like there should be a theorem or two in the graph theory endorsing it. Or do you mean, the

Hi Tobi, I have one additional question about the RegionInfo::isRegion function. In the second case (i.e. Entry dominates Exit), why is checking the following two conditions are equivalent to checking it's a refined region: For any BB in DF(exit), 1) BB should be in DF(entry) 2) BB reachable only from entry through a path passing exit.

Hi Fellows, I notice an unexpected behavior in RegionInfo's block_iterator. Consider the following situation: a user creates her own region containing just a single basic block TheBB, then tries to have the block_iterator to print a DFS traversal of the region. The expected behavior should be that only the single basic block TheBB will be printed, but the real behavior is that

Hi all, I ran across an assertion failure while using region analysis in my passes. I get the same thing when doing: opt -regions -analyze [...]   [1] entry => if.end End region tree Printing analysis 'Detect single entry single exit regions' for function 'njDecodeSOF': Region tree: [0] entry => <Function Return>   [1] entry => return     [2] if.end => return

Hi Tobias, Thanks so much for the quick response. Your approach fixes the issue. On a bigger context, would it make more sense to make the region exit part of the region? For example, a while loop gets lowered down to LLVM IR contains while.cond, while.body and while.end. If one tries to use RegionInfo as a substitute for structural analysis, she might think while.end should belong to

Hi , It will be good ,If you get us the sample that assert. Thanks ~Umesh On Tue, Jul 31, 2012 at 11:40 AM, 田中 太郎 <dontforgettobringatowel at live.jp>wrote: > > Hi all, > > I ran across an assertion failure while using region analysis in my passes. > I get the same thing when doing: opt -regions -analyze > > [...] > [1] entry => if.end > End region tree

On Fri, May 2, 2014 at 5:30 PM, Tobias Grosser <tobias at grosser.es> wrote: > On 03/05/2014 00:15, Paul Vario wrote: > >> Hi Tobias, >> >> Thanks so much for the quick response. Your approach fixes the >> issue. >> On a bigger context, would it make more sense to make the region >> exit >> part of the region? For example, a while

Last time I asked here about superblocks (which we were using in the first version of ThreadSanitizer instrumentation) Tobias Grosser has suggested to use RegionInfo: did you have a look at the RegionInfo pass? It currently detects some kind of refinded SESE regions. I use them in Polly and as far as I know the Intel OpenCL SDK also uses them in some way. It is not SEME, but it may

On 04/26/2013 05:08 AM, tanmx_star wrote: > Hi all, Hi, thanks for the update and sorry for the delay in reviewing. I just had a look at your proposal. > I have updated my GSoS proposal: "FastPolly: Reducing LLVM-Polly Compiling overhead" (https://gist.github.com/tanstar/5441808). I think the pass ordering problem you discussed early can be also investigated in this project!

On 10/22/2011 08:41 AM, Marcello Maggioni wrote: > I was trying the new feature you introduce about printing out the > graphs, so I updated my version of llvm/clang/polly synchronizing them > to the last version, but I get this error launching clang (also , I > recently switched to MacOS X for development): > > $ clang not_so_simple_loop.c -O3 -Xclang -load -Xclang >

On 01/26/2011 07:50 PM, Vu Le wrote: > I guess I didn't have a clear question. > > Suppose we have BB1 and BB2 both point to BB3. > BB1 has variable x. BB2 also as variable x. > BB3 will have PHINode for x with 2 value from BB1 and BB2. > BB1 BB2 > \ / > BB3 > > Now if BB1 and BB2 is extracted into a function > (using ExtractCodeRegion), they

Hi all, I have updated my GSoS proposal: "FastPolly: Reducing LLVM-Polly Compiling overhead" (https://gist.github.com/tanstar/5441808). I think the pass ordering problem you discussed early can be also investigated in this project! Is there any comment or advice about my proposal? I appreciate all your help and advice. Thanks, Star Tan Proposal:

Why aren't polly.merge and entry in the same color region in the attached file? What I'm looking to do is match up a control flow to it's end merge point. This is simple to do based on preds providing each block is in an independent path; however, this is not always the case. -------------- next part -------------- An HTML attachment was scrubbed... URL:

Btw, here is another interesting paper about post-dominators and control dependence: https://pdfs.semanticscholar.org/cbb2/9a0e4895025bd9df24f9263217df12f1ed1e.pdf I think a great outcome of your internship would be some precise documentation regarding the guarantees the LLVM dominators give -- possibly also considering classic and weak control dependence and the difference between

hi all, On 2010-1-7 0:11, John Mosby wrote: > In LLVM we could add support for generalized CFG regions and > RegionPasses. A region is a part of the CFG. The only information we > have is, that it has one entry and one exit, this it can be optimized > separately. > I think this is the best way to add region analysis. I must admit > this approach > helps me on another,

Hi, 2011/4/8 Vlad Krylov <krvladislav at gmail.com>: > Hi. > > I see that to detect scops firstly we search for regions in CFG ( by > RegionInfo ) and then select regions that answer some requirements ( > in ScopDetection ). Because only affine expressions in conditions and > bounds are permissible, we trying to get scalar expressions into > affine form by

Hi. I see that to detect scops firstly we search for regions in CFG ( by RegionInfo ) and then select regions that answer some requirements ( in ScopDetection ). Because only affine expressions in conditions and bounds are permissible, we trying to get scalar expressions into affine form by AffineSCEVIterator. At present there plugs for scev types Truncate, ZeroExtend, SignExtend, UDivExpr,