similar to: [LLVMdev] sincos functions

Hi Suresh, > I was trying to compare the performance of icc, gcc and llvm on the > program almabench.c in Coyote Benchmark suite. Here is a line of code > from the program. > > > da = da + (ca[np][k] * cos(arga) + sa[np][k] * sin(arga)) * 0.0000001; > > gcc and icc are performing way better than llvm as they are using > 'sincos' library function to

Hi, While writing http://reviews.llvm.org/D20916, I stumbled across some code affecting ARM and X86_64 environments that looks like it might be unintentional. I thought I should ask about it here since that patch has a '[mips]' tag and therefore might not be noticed by someone who knows. I've noticed that the GNUX32 and GNUEABI/GNUEABIHF environments don't make use of the sincos

Hi, I'm looking at http://llvm.org/bugs/show_bug.cgi?id=13204 which involves converting calls to sin and cos to sincos (when available) Initially I thought about transforming calls to sinf/cosf to sincosf. However, I don't think this is a legal transformation given that a declaration for a function called sinf is not necessarily the standard library function. Therefore it makes sense to

Hi Sanjay, For sincos calls, I’m currently just going through isTriviallyVectorizable(), which was good enough to get things working so that I could test the translation. I don’t see why this cannot be changed to use addVectorizableFunctionsFromVecLib(). The other functions that I’m working with are already vectorized using the loop pragma. Those include sin, cos, exp, log, and pow. From: Sanjay

RFC: A proposal for vectorizing loops with calls to math functions using SVML (short vector math library). ========= Overview ========= Very simply, SVML (Intel short vector math library) functions are vector variants of scalar math functions that take vector arguments, apply an operation to each element, and store the result in a vector register. These vector variants can be generated by the

I'm cross-posting the message below (from GCC list) because I believe it would (at some point) be very beneficial to build an evolutionary optimization pass into LLVM. The idea would be to discover the perfect set of optimizations for a given program by trying them all and analyzing the execution times of each. This would be somewhat like profile driven optimization except the profile is

On 22/01/13 05:30, Redmond, Paul wrote: [...] > I'm looking at http://llvm.org/bugs/show_bug.cgi?id=13204 which involves converting calls to sin and cos to sincos (when available) > > Initially I thought about transforming calls to sinf/cosf to sincosf. However, I don't think this is a legal transformation given that a declaration for a function called sinf is not necessarily the

Devang, I meant as an end user. -Suresh On Thu, Jun 16, 2011 at 11:00 PM, Devang Patel <dpatel at apple.com> wrote: > Suresh, > > > On Jun 15, 2011, at 9:13 PM, Suresh Purini wrote: > >> Dear all, >> >> What is the default loop-unroll factor in llvm? How can we specify >> our own unroll-factor? > > Here "we" means end user or a

Dear all, What is the default loop-unroll factor in llvm? How can we specify our own unroll-factor? -Suresh

Hello, When we invoke the loop-unroll pass, the compiler is crashing. From the earlier posts in the mailing-list and from the bug reports, it is a known problem. Is there some one working on this bug? -Suresh

Hello, I am trying to use adaptIntegrate function but I need to pass on a few additional parameters to the integrand. However, this function seems not to have the flexibility of passing on such additional parameters. Am I missing something or this is a known limitation. Is there a good alternative to such restrictions, if there at all are? Many thanks for your time. HC -- View this message in

Dear all, I am doing few experiments to do understand optimization phase interactions. Here is a brief description of my experiements. 1. I picked the list of machine independent optimizations acting on llvm IR (those that are enabled at O3). 2. for each optimzation in the optimization-list a) Compiled the program using 'clang -c O0 -flto program.c' b) opt

Suresh, On Jun 15, 2011, at 9:13 PM, Suresh Purini wrote: > Dear all, > > What is the default loop-unroll factor in llvm? How can we specify > our own unroll-factor? Here "we" means end user or a compiler developer ? The threshold is 150, see LoopUnrollPass.cpp - Devang

Hello, I am working on finding good optimization sequences for a given program (phase ordering problem). I have the following setup. 1) The source programs are translated into LLVM IR using -O0 + -scalarrepl. 2) Find an optimization sequence using some strategy which translates the IR generated in the previous step into another IR. 3) Apply llc -O2 and map the IR in to target assembly code.

Hello, How can I convert Fortran Programs to llvm IR? Can I use dragonegg to generate an llvm IR and then use rest of the llvm tool set as it is? -Suresh

This is a hot topic in the compiler research community, but the focus there is on (a) choosing the right optimization sequences internally and transparently, rather than through combinations of options, (b) performance prediction techniques so you don't actually have to run gazillion different choices, and perhaps can even avoid the problem of choosing representative inputs, as you talked

On 19 June 2011 14:44, Suresh Purini <suresh.purini at gmail.com> wrote: >  I am doing few experiments to do understand optimization phase > interactions. Here is a brief description of my experiements. > > 1. I picked the list of machine independent optimizations acting on > llvm IR (those that are enabled at O3). > 2.  for each optimzation in the optimization-list >  

Hello, I tried to do some small experiments on the loop unroll transformation. Following is the Test Program. I compiled it as follows: $ opt -loop-rotate -debug-only=loop-unroll -loop-unroll -unroll-count=2 test1.o -S -o test1.s ------------------ int a[1024]; int main() { int i, sum=0; for(i=0; i<1024; ++i) sum += a[i]; printf("%d",sum); } ------------------- I got

FYI, Yesterday's Slashdot had an article about Linux compiler benchmarks from Coyote Gulch (Scott Ladd). In this update he compares GCC and ICC. You can read the article here: http://www.coyotegulch.com/reviews/linux_compilers/ Of particular note was his use of SciMark 2.0 which is a NIST developed benchmark for scientific computing. Its available in both java and C and computes a MFLOPS

Dear all, Is there a command line argument which prints the order of application of various analysis/transformation passes on a program using clang? -Suresh