similar to: Is it possible to reproduce the result of opt -O3 manually?

I think this has to do with how the pass manager is populated when we give -O3 vs when we give particular pass names. Some passes have multiple createXYZPass() methods that accept arguments too. These methods call non-default pass constructors, which in turn cause the passes to behave in a different manner. eg: Pass *llvm::createLICMPass() { return new LegacyLICMPass(); } Pass

Hi, Below you can see the updated benchmark results for the new SLP-vectorizer. As you can see, there is a small number of compile time regressions, a single major runtime *regression, and many performance gains. There is a tiny increase in code size: 30k for the whole test-suite. Based on the numbers below I would like to enable the SLP-vectorizer by default for -O3. Please let me know if you

Hello, I am trying to replicate the output from opt -O3 foo.bc -o foo.opt.bc by specifying the individual passes instead of the -O3 flag. Looking at the passes from opt -O3 foo.bc -o foo.bc -debug-pass=Executions it seems there are two passes being run. When I run the flags indicated for the two passes specified in the 'Pass Arguments:' as two sequential opt processes or a single opt

The first one failed on a diff: ******************** TEST (simple) 'sse.expandfft' FAILED! ******************** Execution Context Diff: /home/rkotler/llvmpb3/build/projects/test-suite/tools/fpcmp: Compared: 1.139094e-07 and 1.159249e-07 abs. diff = 2.015500e-09 rel.diff = 1.738626e-02 Out of tolerance: rel/abs: 1.600000e-02/0.000000e+00 ******************** TEST (simple)

Dear R Developers, I should have had the Design library running in R by now but have kept putting off changing some calls to LINPACK routines to use those builtin to R. Specifically I call dqrsl1 and dqr. Would it be an easy task to put those in the next release of R? If not I'll finally bite the bullet and get back into reading LINPACK documentation (which I have but haven't examined

Ordinary functions should not alter their inputs but in R-2.15.2 svd(LINPACK=TRUE,X) does. (It worked in 2.15.0 but not in 2.15.1 or 2.15.2 and became deprecated in 2.15.2.) > X <- matrix(c(1,2,3, 5,7,11, 13,17,19), 3, 3) > X [,1] [,2] [,3] [1,] 1 5 13 [2,] 2 7 17 [3,] 3 11 19 > svd(X, LINPACK=TRUE)$d [1] 31.9718214 2.3882717 0.3143114 Warning message:

when I create the report, there are no failures in it. so maybe these are being filtered for known failures. On 12/12/2012 05:03 PM, reed kotler wrote: > The first one failed on a diff: > ******************** TEST (simple) 'sse.expandfft' FAILED! > ******************** > Execution Context Diff: > /home/rkotler/llvmpb3/build/projects/test-suite/tools/fpcmp: Compared: >

I use the 'make TEST=simple' as a pre-commit test. I think that everybody should run these tests before committing to LLVM. On Dec 12, 2012, at 5:06 PM, reed kotler <rkotler at mips.com> wrote: > when I create the report, there are no failures in it. so maybe these are being filtered for known failures. > > On 12/12/2012 05:03 PM, reed kotler wrote: >> The first

I'm getting three failures. TEST-FAIL: exec /home/rkotler/llvmpb3/build/projects/test-suite/SingleSource/UnitTests/Vector/SSE/sse.expandfft TEST-RESULT-exec-time: user 0.3200 TEST-RESULT-exec-real-time: real 0.3172 TEST-FAIL: exec /home/rkotler/llvmpb3/build/projects/test-suite/SingleSource/UnitTests/Vector/SSE/sse.stepfft TEST-RESULT-exec-time: user 0.4000

I'm probably misunderstanding something in "Writing R Extensions" version 1.0.0. In the chapter on the R API, section 4.7, it is stated that the functions listed in R_ext/Linpack.h are available to users' Fortran code. I am developing a developing a library of ode solvers, based on lsoda and ddassl, and which in turn call some routines from linpack and double precision blas. I

I forgot to mention that you can also run "make TEST=simple report" which will generate a nice report. Do you know why these tests fail ? You can step into the test directory and run 'make TEST=simple' from there. It will save you some time. On Dec 12, 2012, at 4:04 PM, reed kotler <rkotler at mips.com> wrote: > I'm getting three failures. > > TEST-FAIL:

Hello, I am trying to simplify the CFG of a given code and eliminate the conditionals, even though I will obtain codes that are not semantically equivalent. For example, given a simple loop: for(i=0; i<N; i++){    a[i] = i;     if (i%2==0)       a[i] += 12; } I would keep only the loop, without the if statement: for(i=0; i<N; i++){    a[i] = i; } I can eliminate such conditionals on

Hi Everyone, I have modified my version of R-1.4.1 to include choleski with pivoting (like in Splus). I thought R-core might consider including this in the next version of R, so I give below the steps required to facilitate this. 1. Copied Linpack routine "dchdc.f" into src/appl 2. Inserted line F77_SUBROUTINE(dchdc) in src/appl/ROUTINES 3. Inserted "dchdc.f" into

Hello, I'm trying to minimize the processing time for llc -O3 by using a three step compilation process of 1. llc input.bc -stopafter=targetlibinfo -o input.mir 2. llc -run-pass={....min passes...} input.mir -o opt.mir 3. llc -startafter=machine-opt-remark-emitter -filetype=obj opt.mir -o final.obj Examining the passes produced by llc for O1,O2,O3 I compared (with XXX = {1,2,3}): llc

Hi, Jim: Could you please look at svd2.Rd and see what it says? It may give an example, where it gave a better answer than svd -- i.e., a marginal case, where svd2 honestly gave a better answer than svd. If we find -- either in svd2.Rd or in one of the revdepchecks -- an example where svd2 gives a demonstrably different answer, we need to consider what to do about that. 1.

man for `qr` says that the function uses LINPACK's DQRDC, while it in fact uses DQRDC2. ``` The QR decomposition of the matrix as computed by LINPACK or LAPACK. The components in the returned value correspond directly to the values returned by DQRDC/DGEQP3/ZGEQP3 ```

Hello. I am trying to invert a matrix, and I am finding that I can get different answers depending on whether I set LAPACK true or false using "qr". I had understood that LAPACK is, in general more robust and faster than LINPACK, so I am confused as to why I am getting what seems to be invalid answers. The matrix is ostensibly the Hessian for a function I am optimizing. I want to get

I want to speed up computations (involving matrices) by writing some C-code to be loaded. In the C-code, I need to invert matrices etc. As I've understood the "writing R extensions" doc, I can use use #include <R_ext/Linpack.h> in my .c-file and get access to linpack's facilities within my C-code. Is that correct? If so, can anyone point me to examples on how this is

Hey folks, Some progress has been made since the first thread: http://lists.llvm.org/pipermail/llvm-dev/2017-March/111129.html And also I think the consensus is to enable "-ffp-contract=on" by default (instead of "fast"), which seems to be working on some preliminary tests I made. I just ran the test-suite on x86_64 and AArch64. The former is ok, the latter still has some