similar to: calling optimize/fmin from C code

… actually, now that I’m able double-check this, I’m quite surprised to find that we didn’t define fmax(+0,–0) in IEEE–754, which says [paraphrased]: minNum(x,y) is x if x < y, y if y < x, and the number if one is a number and the other is NaN. Otherwise, it is either x or y (this means results might differ among implementations). So I think your proposed semantics are perfectly

Hi, I’d like to re-propose adding intrinsics for fmin / fmax. These can be used to implement the equivalent libm functions as defined in C99 and OpenCL, which R600 and AArch64 at least have instructions with the same semantics. This is not equivalent to a simple fcmp + select due to its handling of NaNs. This has been proposed before, but never delivered

This is a problem with all floating point folding, not just with these operations. What Matt is proposing is consistent with how we fold other libm intrinsics. —Owen > On Aug 18, 2014, at 1:22 AM, Mueller-Roemer, Johannes Sebastian <Johannes.Sebastian.Mueller-Roemer at igd.fraunhofer.de> wrote: > > Wouldn’t it be better to use the target’s implementation (if there is one)

Hi Carter, I would strongly advise you against this direction. I’m aware of two directions that existing languages go in defining min/max operations: - IEEE 754, C, Fortran, Matlab, OpenCL, and HLSL all define it not to propagate NaNs - C++ (std::min/std::max) and OpenGL define it in the trinary operator manner: (a < b) ? a : b What you’re proposing does not match any existing languages

Hi, I have read R's Writing Extensions manual and am still confused about how to use some of the routines there when I call from C. Specifically, I am writing a little test function which I will optimize using the nmmin function which underlies R's optim() with Nelder-Mead. I guess I wonder what library/header files I should be using. I was using R_ext/Applic.h and linking with libR but I

On Sep 15, 2014, at 4:17 PM, Owen Anderson <resistor at mac.com> wrote: > I’d be fine with that proposal. I could even be convinced if we wanted to add a pair of NaN-propagating intrinsics as well, for targets and languages that want those semantics, even if I disagree with them. I do think that, if we are using the minnum/maxnum names, we should explicitly note that they are

Dear list, I have tried to setup my Xemacs for use with Sweave, which I indend to learn. I have followed the instructions in the Sweave FAQ, that is to say, I put (defun Rnw-mode () (require 'ess-noweb) (noweb-mode) (if (fboundp 'R-mode) (setq noweb-default-code-mode 'R-mode))) (add-to-list 'auto-mode-alist '("\\.Rnw\\'" . Rnw-mode)) (add-to-list

I need to find a computationally simple process for the movement of interest rates. In this simplified model, an interest rate can have 3--5 possible values, and its movement is characterized by a matrix of transition probabilities (ie, it is a Markov process). I would like to estimate this process from a given set of data. For example, let the interest rate time series be: 7 3 8 2 5 9 6

Is there a way in R to test if a given number is an integer, ie a whole number? I am not referring to the data type of a number, but to its value. That is to say, is.whole(pi-pi+2) would be TRUE, whereas is.whole(4/3) would be false. At the moment I am using is.whole <- function(a) { floor(a)==a } which is OK for real numbers, but not for complex ones (a+bi would be a whole number if both a

Hi, I am plotting a policy function (result from a dynamic stochastic optimization problem, discretized approximation). The policy function maps from an 2 x 2 x 2 x 3 x B x F state space to a B x F state space (B and F are usually between 4-6, and represent domestic and foreign savings. The other variables are income (Y), inflation (Pi), domestic and foreign interest rates (R and Z)). I

Given IEEE-754's sway, and its saying what it does on this point, but given also the popularity of NaN-propagating min and max, how about a compromise? We add intrinsics following the IEEE-754 semantics, but we also follow IEEE-754 (and ARMv8) in renaming them to minnum and maxnum, to clarify which interpretation these intrinsics are using. -------------- next part -------------- An HTML

I was trying to find source for optimize and I ran across function (f, interval, ..., lower = min(interval), upper = max(interval), maximum = FALSE, tol = .Machine$double.eps^0.25) { if (maximum) { val <- .Internal(fmin(function(arg) -f(arg, ...), lower, upper, tol)) list(maximum = val, objective = f(val, ...)) } else { val <-

> On Sep 12, 2014, at 10:27 AM, Dan Gohman <dan433584 at gmail.com> wrote: > > > More generally, I don’t see a compelling reason for LLVM to add intrinsic support for the version you’re proposing. Your choice can easily be expanded into IR, and does not have the wide hardware support (particularly in GPUs) that the IEEE version does. > > The IEEE version can also be

Hi, The specification for the llvm.minnum/llvm.maxnum intrinsics is too unclear right now to usefully optimize. There are two problems. First the expected behavior for signaling NaNs needs to be clarified. Second, whether the returned value is expected to be canonicalized (as if by llvm.canonicalize). Currently according to the LangRef: Follows the IEEE-754 semantics for minNum, which also

Hi, I need to parse a data file (output of a measuring device) of the following format: BEGIN RECORD [first record data] RECORD [second record data] RECORD [third record data] END Line breaks can (and do ;-() occur anywhere. White space behaves very much like TeX, eg it is not important whether there are one or more spaces or linebreaks as long as there is one of them. It is a text file, not

Dear list, I'd like to use the function cut() on matrices, ie that when I apply it to a matrix, it would return a matrix of the same dimensions instead of a vector. I wonder if there is a better (more elegant) solution than matrix(cut(a, ...), ncol=ncol(a), nrow=nrow(a)) because I would like to use cut on both vectors and matrices and avoid testing whether a is a matrix. Thanks, Tamas

Hi, I have sparse (tridiagonal) matrices, and I use the Matrix package for handling them. For certain computations, I need to either set the first row to zero, or double the last row. I find that neither operation preserves sparsity, eg > m <- Diagonal(5) > m 5 x 5 diagonal matrix of class "ddiMatrix" [,1] [,2] [,3] [,4] [,5] [1,] 1 . . . . [2,] . 1

Hi, I am about to but a laptop, and have narrowed the choices down to a Dell Latitude 600 and an Apple Powerbook G4 Aluminium (Princeton provides these models at a discount for grad students). I am biased towards the Powerbook, and would like to run Debian on it. I have only used debian on i386 platforms so far. I use R quite frequently, so I would be interested in your experience of running R

Hi, I wonder if it would make sense to make uniroot detect zeros at the endpoints, eg if f(lower)==0, return lower as the root, else if f(upper)==0, return upper as the root, else stop if f(upper)*f(lower) > 0 (currently it stops if >=), else proceed to the algorithm proper. Currently I am using a wrapper function to implement this, and I found it useful. But I didn't want to send a

Dear List, I have a couple of (~200) 3x3 transition probability matrices (ie each defines a Markov chain). They are all estimated from the same underlying process, so it ie meaningful to take their elemetwise mean and standard deviation. [1] First question: supposing that they are given in a list l, how do I get their elementwise mean and standard deviation? Fortunately, the mean of trans. prob.