> Running out of disk space or permissions problems with temp directories can > cause super slow compiles. We ran into something (vaugely) similar as we had > multiple people running compiles at once causing havoc in temp directories. > > You might try explicitly setting these in your simulink startup script, e.g. > export DSP_CACHE_DIR=/tmp/dan > export TEMP=/tmp/dan2 > export TMP=/tmp/dan3
Bingo -- we have a winner! For some reason this worked. I was able to compile other designs before making this change, so I don't think it was a permissions error unless the designs that compiled successfully didn't need a DSP_CACHE_DIR. The disk usage in the newly-defined directories isn't very much (~10's of MB), so I can't have been running out of space wherever it was dumping stuff before. > And see if that helps? Another option would be to delete the contents of > ~/.Xilinx (I admit that I don't know what these do, but it is at the very > least cathartic to obliterate them and it hasn't caused me any issues). Fortunately, I did not have to resort to the nuclear option. Thanks for your help! Oliver > > Cheers > Dan
<<inline: compose-unknown-contact.jpg>>
>> Oliver King 20 November 2013 13:32 >> I thought it might be that and increased the heap memory to 512 MB, but to >> avail. I'll try increasing it even more. >> >> >> ------ >> Oliver King >> Senior Scientist >> Astronomy Department, Caltech >> +1-626-395-4197 >> >>
<<inline: compose-unknown-contact.jpg>>
>> G Jones 20 November 2013 13:30 >> Hi Oliver, >> This may be related to the java heap memory issue I ran into recently. >> Check the mailing list archive for details on increasing the heap and >> see if that helps. >> >> Glenn >>
<<inline: compose-unknown-contact.jpg>>
>> Oliver King 20 November 2013 13:29 >> Hi, >> >> I'm trying to blackbox a PFB / FFT combination (following the instructions >> at https://casper.berkeley.edu/wiki/images/a/a4/Black_box_memo.pdf), and am >> running into a problem. >> >> The compilation gets to the "Compiling: Propagating sample times" step and >> stays there forever. There are two short bursts of CPU activity when the >> "Propagating sample times" step starts, but the CPU activity then drops away >> to zero and stays there. After a day or so I lose patience and kill the >> process. >> >> No error messages are printed to the screen other than a warning about the >> model not having continuous states, use VariableStepDiscrete instead, etc. >> Explicitly specifying a discrete solver, as this warning suggests, makes no >> difference. >> >> I am able to black box simple models that use only Xilinx basic elements, so >> I know that the toolflow is working. However, any model that includes either >> the pfb_fir_real or the fft_wideband_real blocks from the CASPER DSP >> Blockset encounters the problem described above. >> >> The particulars of my setup: >> ska-sa mlib_devel, commit a8c43a8f90 >> 64 bit linux (redhat) >> Matlab 2012b >> Xilinx 14.3 >> >> Has anyone come across this problem before? >> >> Cheers, >> Oliver >> >> ------ >> Oliver King >> Senior Scientist >> Astronomy Department, Caltech >> +1-626-395-4197 >> >> ------ Oliver King Senior Scientist Astronomy Department, Caltech +1-626-395-4197
