OK, guys, bad news, we will have to build Altlas on Lion! Or wait for Apple to fix their blas... I checked the attached C code on sqrt5, both with gcc and with cc (i.e. clang), and it's all confirmed.
The correct output is: nash:tmp dima$ ./test_blas y = [ 0.00 1.00 2.00 3.00 4.00 5.00 6.00 7.00 ] y = [-0.50 0.50 1.50 2.50 3.50 4.50 5.50 6.50 ] y = [-1.00 0.00 1.00 2.00 3.00 4.00 5.00 ] nash:tmp dima$ ./test_cblas y = [ 0.00 1.00 2.00 3.00 4.00 5.00 6.00 7.00 ] y = [-0.50 0.50 1.50 2.50 3.50 4.50 5.50 6.50 ] y = [-1.00 0.00 1.00 2.00 3.00 4.00 5.00 ] And on sqrt5 one gets: sqrt5:tmp dima$ ./test_blas y = [ 0.00 1.00 2.00 3.00 4.00 5.00 6.00 7.00 ] y = [ 0.00 1.00 2.00 3.00 4.00 5.00 6.00 7.00 ] y = [-0.50 0.50 1.50 2.50 3.50 4.50 5.50 ] sqrt5:tmp dima$ ./test_cblas y = [ 0.00 1.00 2.00 3.00 4.00 5.00 6.00 7.00 ] y = [ 0.00 1.00 2.00 3.00 4.00 5.00 6.00 7.00 ] y = [-0.50 0.50 1.50 2.50 3.50 4.50 5.50 ] ---------- Forwarded message ---------- From: Martin S. Andersen <martin.ander...@ucla.edu> Date: 6 March 2012 17:54 Subject: Re: cvxopt on MacOSX 10.7 To: dimp...@gmail.com Cc: Joachim Dahl <dahl.joac...@gmail.com>, Lieven Vandenberghe < lieven.vandenber...@gmail.com> Dear Dima, Thank you for your example. The example helped me pinpoint the problem and to create a simple stand-alone C program that fails to produce the correct result when linking against the native BLAS library. In your example, the result is correct when b is a vector of length 7 or less, and otherwise it is incorrect. The attached C examples replicate this problem without CVXOPT. The problem occurs both with the Fortran interface (test_blas.c) and with the CBLAS interface (test_cblas.c). I filed a bug report with Apple in November which I have now updated. I don't know if there is a faster way to get Apple's attention? In the meantime, I am avoiding the native BLAS library in Lion. Best, Martin On Mar 5, 2012, at 5:52 PM, Lieven Vandenberghe wrote: For your information. It is a very strange bug. -------- Original Message -------- Subject: Re: cvxopt on MacOSX 10.7 Date: Tue, 6 Mar 2012 02:12:13 +1300 From: Dima Pasechnik <dimp...@gmail.com><dimp...@gmail.com> To: Lieven Vandenberghe <lieven.vandenber...@gmail.com><lieven.vandenber...@gmail.com> Dear Lieven, here is a bizarre bug we get on OSX 10.7 (compiled with native lapack/blas) from cvxopt import matrix b=matrix([ 1.00e+00, 1.63e-01, 2.83e-01, 9.47e-01, 2.32e-01, 4.85e-01, 9.57e-01, 7.44e-01]) c=1.-b print "b :\n", b print "1-b:\n", c b : [ 1.00e+00] [ 1.63e-01] [ 2.83e-01] [ 9.47e-01] [ 2.32e-01] [ 4.85e-01] [ 9.57e-01] [ 7.44e-01] 1-b: [-1.00e+00] [-1.63e-01] [-2.83e-01] [-9.47e-01] [-2.32e-01] [-4.85e-01] [-9.57e-01] [-7.44e-01] Interestingly, shorter vector b works OK... Do you have any idea what goes wrong here? (this is actually the bug responsible for failure of "acent" test in here:http://sage.math.washington.edu/home/palmieri/misc/cvxopt-1.1.3.log) Best, Dima On 28 February 2012 07:25, Lieven Vandenberghe<lieven.vandenber...@gmail.com> <lieven.vandenber...@gmail.com> wrote: > > Dear Dima, > > I am not sure what the problem is. I will check with Martin Andersen who > provided the MacOSX binaries and has looked into this. > > Best regards, > > Lieven > > > > > On 02/26/2012 04:17 AM, Dima Pasechnik wrote: >> >> Dear Lieven, >> I wonder if were able to determine what caused problems with running >> CVXOPT on MacOSX 10.7. We're going to release Sage on this platform >> soon, so it's become >> an issue... >> >> I see you're distributing pre-compiled CVXOPTs for this platform, >> using the "non-native" BLAS/LAPACKs. >> Do you have convincing evidence that Apple's BLAS/LAPACK is broken on OSX >> 10.7? >> Somehow, Apple's BLAS/LAPACK is used in other parts of Sage, and there >> we do not seem to see problems (although perhaps we do not try hard >> enough). >> >> Many thanks in advance, >> Dima -- To post to this group, send an email to sage-devel@googlegroups.com To unsubscribe from this group, send an email to sage-devel+unsubscr...@googlegroups.com For more options, visit this group at http://groups.google.com/group/sage-devel URL: http://www.sagemath.org
Makefile
Description: Binary data
test_blas.c
Description: Binary data
test_cblas.c
Description: Binary data
