Greetings! Please do me a favor and type 'disassemble' in gdb at your breakpoint -- let's see if the address reported (0xb670b656) corresponds to:
141d66: 0f 18 86 3c 02 00 00 prefetchnta 0x23c(%esi) You are using the vanilla atlas which should be fine everywhere -- what is the CPU? On my box, this routine is indeed free of vector arithmetic operations, but I don't think the prefetch should be here. This (if memory serves), is one of the new hand assembler l1 routines written by Clint Whaley relatively recently. One other item though -- Debian atlas has threads turned off. I don't understand the origin of the gdb tread switching message. Have you built a custom package? Take care, Douglas Bates <[EMAIL PROTECTED]> writes: > It's dying in an Atlas call to a relatively simple routine (idamax). > Because it gives an illegal instruction error my guess is that it is > picking up the wrong version of the Atlas libraries. The Atlas > libraries are specific to the particular processor being used. You > could try installing one of > > atlas3-3dnow > atlas3-sse2 > atlas3-sse > > according to the type of processor you are running. I believe these > can all coexist on the same system and the installation process sets > up a local map to the optimal version but I'm not sure about that. > Camm? > > On 12/22/05, Joshua N Pritikin <[EMAIL PROTECTED]> wrote: > > On Thu, Dec 22, 2005 at 09:11:44AM -0600, Douglas Bates wrote: > > > On 12/21/05, Dirk Eddelbuettel <[EMAIL PROTECTED]> wrote: > > > > | I tried to octave2.1 also, det(eye(9)) gives a panic: Illegal > > > > | instruction. det(eye(4)), 5, 6, 7, 8 works though. It's a mystery. > > > > > > > > I very vaguely recall a bug report with respect to Lapack that came up > > > > in R > > > > that was triggered by one matrix size only. If memory serves, it was 9 > > > > as > > > > well. This is the reason we build R with its own copy of Lapack. > > > > > > > > I cannot recall what the exact circumstances were, but maybe Doug or > > > > Brian > > > > (CC'ed) can remember ? > > > > > > I believe that bug was in an eigenvalue-eigenvector calculation. I > > > imagine this calculation is done using an LU decomposition or, if the > > > matrix is recognized as being symmetric, a Bunch-Kaufman > > > decomposition. > > > > > > Can you create a traceback to show where it is generating the illegal > > > instruction? > > > > Does the follow provide enough detail or should I recompile with -g? > > > > Program received signal SIGILL, Illegal instruction. > > [Switching to Thread -1240947008 (LWP 9259)] > > 0xb670b656 in ATL_diamax_xp1yp0aXbX () from /usr/lib/atlas/libblas.so.3 > > (gdb) where > > #0 0xb670b656 in ATL_diamax_xp1yp0aXbX () from > > /usr/lib/atlas/libblas.so.3 > > #1 0xb68f33a8 in ?? () from /usr/lib/atlas/libblas.so.3 > > #2 0x00000009 in ?? () > > #3 0x08f508c8 in ?? () > > #4 0xb670b4a5 in ATL_idamax () from /usr/lib/atlas/libblas.so.3 > > #5 0x00000009 in ?? () > > #6 0x08f508c8 in ?? () > > #7 0x00000001 in ?? () > > #8 0x00000009 in ?? () > > #9 0xb670b440 in ATL_dgpr1U_a1_x1_yX () from > > /usr/lib/atlas/libblas.so.3 > > #10 0xb6e3d578 in ?? () from /usr/lib/atlas/liblapack.so.3 > > #11 0x08f508c8 in ?? () > > #12 0xb6932c8c in cblas_idamax () from /usr/lib/atlas/liblapack.so.3 > > #13 0x00000009 in ?? () > > #14 0x08f508c8 in ?? () > > #15 0x00000001 in ?? () > > #16 0x00000009 in ?? () > > #17 0xb6932c50 in cblas_icamax () from /usr/lib/atlas/liblapack.so.3 > > #18 0xb6e3d578 in ?? () from /usr/lib/atlas/liblapack.so.3 > > #19 0x08f50910 in ?? () > > #20 0xb6913d43 in ATL_dgetrfC () from /usr/lib/atlas/liblapack.so.3 > > #21 0x08f50258 in ?? () > > #22 0x08f4fe3c in ?? () > > ---Type <return> to continue, or q <return> to quit--- > > #23 0x08f50058 in ?? () > > #24 0x08f50058 in ?? () > > #25 0x00000000 in ?? () > > #26 0x00000000 in ?? () > > #27 0x08f50058 in ?? () > > #28 0x08f50058 in ?? () > > #29 0x08f50258 in ?? () > > #30 0x08f4fe3c in ?? () > > #31 0x00000000 in ?? () > > #32 0x00000009 in ?? () > > #33 0x08f50258 in ?? () > > #34 0x08f4fe3c in ?? () > > #35 0x08f50058 in ?? () > > #36 0xb6e3d578 in ?? () from /usr/lib/atlas/liblapack.so.3 > > #37 0x00000009 in ?? () > > #38 0x08f4f898 in ?? () > > #39 0x00000002 in ?? () > > #40 0xb6913c27 in ATL_dgetrfC () from /usr/lib/atlas/liblapack.so.3 > > #41 0x08048430 in ?? () > > #42 0xb7880b9c in ?? () from /usr/lib/octave-2.9.4/liboctinterp.so > > #43 0x08f50022 in ?? () > > #44 0x08f50058 in ?? () > > #45 0x08f50258 in ?? () > > ---Type <return> to continue, or q <return> to quit--- > > #46 0x08f4fe3c in ?? () > > #47 0xb783775c in ?? () from /usr/lib/octave-2.9.4/liboctinterp.so > > #48 0x0f3a8b99 in ?? () > > #49 0xb7880b9c in ?? () from /usr/lib/octave-2.9.4/liboctinterp.so > > #50 0x08f4fe3c in ?? () > > #51 0xb612cc34 in free () from /lib/tls/i686/cmov/libc.so.6 > > > > > > > > -----BEGIN PGP SIGNATURE----- > > Version: GnuPG v1.2.4 (GNU/Linux) > > > > iD8DBQFDquX+qcqnlKSmC70RAlEfAKC9gfWXnpGa2fLfyzu6l8oGRDV6owCfX/e5 > > 2nifiShhn4tq4pNx4sxFjIs= > > =R3qI > > -----END PGP SIGNATURE----- > > > > > > > > > -- Camm Maguire [EMAIL PROTECTED] ========================================================================== "The earth is but one country, and mankind its citizens." -- Baha'u'llah -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]