similar to: [LLVMdev] Callgraph inaccuracy

I'm working on this issue, thank you for reporting it. The short answer is that the callgraph reported by TD is the same as discovered/computed during BU (and BU can't determine the indirect call can only target C2). I've added this example, and a few other related tests, to the DSA's test suite[1] to document this behavior while I'm working out what the best solution is.

Hello, I am trying to apply DSA (from the poolalloc project - I'm on LLVM 3.2) on the following C program and found that the generated callgraph over-approximates the callees for the simple indirect call. #include <stdio.h> __attribute__((noinline)) static int f1(int arg1, int arg2) { return arg1 + arg2; } __attribute__((noinline)) static int run_func(int (*fptr)(int, int), int

Hi all, I would like to create a Pass that given an IR instruction walks starting from that instruction up to the main function to identify all the functions call that have been made to call that instruction. Is it possible? What kind of Pass should I create? Thanks Best, Simone Simone Atzeni simone.at at gmail.com +1 (801) 696-8373

Thanks John. I guess I will use a ModulePass, so when I am implementing the “runOnModule” function, do I have to loop through all the functions, for each functions all the BasicBlocks and for each BasicBlock all the instructions or given the Module I have to call the CallGraph directly? Is there an example out there? I can’t find anything. Thanks. Simone > On Feb 24, 2015, at 13:29, John

On 2/25/15 10:51 AM, Simone Atzeni wrote: > Thanks John. > > I guess I will use a ModulePass, so when I am implementing the “runOnModule” function, > do I have to loop through all the functions, for each functions all the BasicBlocks and for each BasicBlock all the instructions If you know the Instruction, you can get it's basic block using Instruction::getParent(), and then get

Hi Simon, > From: Simone Atzeni <simone.at at gmail.com> > To: John Criswell <jtcriswel at gmail.com> > Cc: llvmdev at cs.uiuc.edu > Subject: Re: [LLVMdev] Walking thru CallGraph bottom up > Message-ID: <318EBA41-2040-4EFE-B330-5813C817C2A2 at gmail.com> > Content-Type: text/plain; charset="windows-1252" > > I think I got it and the example is

Dear Simon, Kevin is correct; as far as I can tell, there is no method of getting the functions calling a given function. Instead, you have to start at the main() function and search for the function using a depth-first or breadth-first search. What may make sense is to build a new data structure that has nodes that point from callees to callers once and then use that for your queries.

Dear John, I intend to implement the improvements on DSA. After running DSA on SPEC, I found DSA gives low precision for mcf and bzip2. I have checked the most imprecise c files in mcf an found that the code seems to be a mixture of "PHI" and "GEP" instructions. Could you please give me some hints about what the big picture of the improvement should be and how to start? Thank

Hi all, I've been using LLVM in my software analysis projects for quite a few years now, and several years back I relied on results of DSA analysis in my SMACK tool for checking C programs. At some point that part of SMACK got deprecated, but now I would like to revisit it since it was working quite well. Therefore, I would like to learn what's the status of the poolalloc project, and

Using a line width > 1 results in not only a thicker line but also some fuzz in the other direction, as shown in this example program. You will see that the thick vertical black lines extend below the gray scale horizontal lines. Does anyone know whether this is intended or is it a bug? The application is for displaying a correlation matrix (here just some random U(0,1)s). Thanks -Frank

Hi, Cosmetic. Starting R with e.g. --max-ppsize=-10 produces the following warning: WARNING: '-max-ppsize' value is negative: ignored The name of the option displayed in the warning is incorrect. Could that be fixed? See src/main/CommandLineArgs.c (there are 3 places in that file where the name of this option needs to be adjusted). This is with current R-alpha. Thanks! H. --

Hi, In man page for duplicated: Value: ?duplicated()?: For a vector input, a logical vector of the same length as ?x?. For a data frame, a logical vector with one element for each row. For a matrix or array, a logical array with the same dimensions and dimnames. When 'x' is a matrix or array, the returned value is NOT a logical array: > m <-

Hello, I have been working with various probability distributions in R, and it seems the gamma distribution is inaccurate for some inputs. For example, qgamma(1e-100, 5e-101, lower.tail=FALSE) gives: 1.0. However, it seems this is incorrect; I think the correct answer should be 0.082372029620717283. When I check these numbers using pgamma, I get: pgamma(1,5e-101, lower.tail=FALSE) =

I came across this: shouldn't the last value be a more exact zero? It did not do that with 1 - sum( rep(0.1, 10) ) > 1 - cumsum( rep(0.1, 10) ) [1] 9.000000e-01 8.000000e-01 7.000000e-01 6.000000e-01 5.000000e-01 4.000000e-01 3.000000e-01 2.000000e-01 [9] 1.000000e-01 1.110223e-16 > version _ platform

Hello, I need to use the confluent function of second kind, also known as Tricomi function. It is implemented as kummerU() function in fAsianOptions package, but I've found very inaccurate values, comparing with those provided by Mathematica. I think Mathematica values are OK because kummerU values leads to negative probabilities. For example, if you try the kummerU() function example, you

Hi, I found the following strange effect with qbinom & partial argument matching p0 <- pbinom(0, size = 3, prob = 0.25) qbinom(p0, size = 3, prob = 0.25) ## 0 o.k. qbinom(p0-0.05, size = 3, prob = 0.25) ## 0 o.k. ## positional matching: qbinom(p0, 3, 0.25) ## 0 o.k. ## partial argument matching: qbinom(p0 , s = 3, p = 0.25) ## 1 ??? qbinom(p0-0.05,

Hi all, I'm having some problems with inaccurate sampling rates between sending and receiving party. (Using a simple raw UDP protocol sending packets from speex to speex) Both PC's have their soundcard set to 8 khz, but the samplerate seems to differ slightly between machines. The problem I have is that the sending machine sometimes samples with a slightly lower rate than the player

On 1/26/11 3:06 PM, songlh at cs.wisc.edu wrote: >> [snip] >> -- John T. > Sorry, I am asking the second question. > " > how to determine, within an LLVM pass, what the > possible target(s) of an indirect function call might be? There are at least two solutions. The first is to use the CallGraph analysis pass. It constructs a conservative call graph, meaning that

> On 1/26/11 3:06 PM, songlh at cs.wisc.edu wrote: >>> [snip] >>> -- John T. >> Sorry, I am asking the second question. >> " >> how to determine, within an LLVM pass, what the >> possible target(s) of an indirect function call might be? > > There are at least two solutions. > > The first is to use the CallGraph analysis pass. It

Hi, Man page for 'convolve' says: conj: logical; if 'TRUE', take the complex _conjugate_ before back-transforming (default, and used for usual convolution). The complex conjugate of 'x', of 'y', of both? In fact it seems that it takes the complex conjugate of 'y' only which is OK but might be worth mentioning because (1) conj=TRUE is the