rawspec failing on ATA_guppi_59730_78085_225768981_J0332+5434_off_0001-beam0000
texadactyl opened this issue · comments
rawspec -f 1 -t 32 -g 3 -d . /mnt_blpd20/scratch/rawspec_testing/baseline//ATA_guppi_59730_78085_225768981_J0332+5434_off_0001-beam0000
rawspec 3.1.1+11@g188981a using librawspec 3.1.1+11@g188981a and cuFFT 10.2.1.245
using requested GPU: 3
writing output files in SIGPROC Filterbank format
working stem: /mnt_blpd20/scratch/rawspec_testing/baseline//ATA_guppi_59730_78085_225768981_J0332+5434_off_0001-beam0000
opening file: /mnt_blpd20/scratch/rawspec_testing/baseline//ATA_guppi_59730_78085_225768981_J0332+5434_off_0001-beam0000.0000.raw
bad obsnchan/nants: 192 % 20 != 0
output product 0: 0 spectra
It must be one of the 2 commits after June 1st because regression testing was successful that day.
See https://github.com/UCBerkeleySETI/rawspec/commits/master
@radonnachie Is this file no longer valid? I got it from you: UCBerkeleySETI/rawspec_testing@9a36ae2
@wfarah I have tried to assign @radonnachie but he does not appear in the assignee list (?). Basically, it seems like a file that Ross gave me is no longer valid because obsnchan/nants: 192 % 20 != 0. Maybe we can replace this file?
Yeah it is invalid because it required the NBEAMS override of NANTS that I've recently redacted from #58. The NANTS value in this beam file should be 1.
@radonnachie Is a replacement available?
I have just run rsync -av /mnt/buf1/blade/GUPPI/./guppi_59811_38723_118314086_AzEl_0001-beam0001.0000.raw rsync://blpd18.ssl.berkeley.edu/datax/ and manually inspected it to have NANTS=1, NBEAM=2 and BEAM_ID=1.
Prior to that I transferred rsync -av /mnt/buf0/blade/GUPPI/./guppi_59811_38073_118274414_AzEl_0001-beam0000.0001.raw rsync://blpd18.ssl.berkeley.edu/datax/, which has NANTS=1 and NBEAMS=1, so I'd ask you to discard it please (I've yet to actually log in to the Berkeley servers.
@radonnachie
Downloaded:
rsync -v rsync://blpd18.ssl.berkeley.edu/datax/guppi_59811_38723_118314086_AzEl_0001-beam0001.0000.raw .
guppi_59811_38723_118314086_AzEl_0001-beam0001.0000.raw
sent 43 bytes received 8,525,911,930 bytes 215,845,872.73 bytes/sec
total size is 8,523,830,784 speedup is 1.00
Prepended "ATA_" to file name.
Tried to use it:
rawspec -f 1 -t 32 -j -g 3 -d . ATA_guppi_59811_38723_118314086_AzEl_0001-beam0001
rawspec 3.1.1+11@g188981a-dirty using librawspec 3.1.1+11@g188981a-dirty and cuFFT 10.2.1.245
using requested GPU: 3
writing output files in FBH5 format
working stem: ATA_guppi_59811_38723_118314086_AzEl_0001-beam0001
opening file: ATA_guppi_59811_38723_118314086_AzEl_0001-beam0001.0000.raw
Number of spectra per dump (Nd) = 256
Number of fine channels per coarse channel (nfpc) = 1
Chunk dimensions = (256, 1, 1)
Cache nslots = 257, nbytes = 196609
BLOCSIZE not found in header
error getting obs params from ATA_guppi_59811_38723_118314086_AzEl_0001-beam0001.0000.raw [No such file or directory]
opening file: ATA_guppi_59811_38723_118314086_AzEl_0001-beam0001.0001.raw [No such file or directory]
output product 0: 1536 spectra
"BLOCSIZE not found in header"?
watutil -i ATA_guppi_59811_38723_118314086_AzEl_0001-beam0001.rawspec.0000.h5
--- File Info ---
DIMENSION_LABELS : [b'time' b'feed_id' b'frequency']
az_start : 0.0
data_type : 1
fch1 : 6240.0 MHz
foff : 0.5 MHz
ibeam : 1
machine_id : 20
nbeams : 1
nbits : 32
nchans : 192
nfpc : 1
nifs : 1
rawdatafile : ATA_guppi_59811_38723_118314086_AzEl_0001-beam0001.0000.raw
source_name : AzEl
src_dej : 67:15:07.0502
src_raj : 12:33:08.4039
telescope_id : 9
tsamp : 6.4e-05
tstart (ISOT) : 2022-08-20T10:45:23.000
tstart (MJD) : 59811.44818287037
za_start : 0.0
Num ints in file : 1536
File shape : (1536, 1, 192)
--- Selection Info ---
Data selection shape : (1536, 1, 192)
Minimum freq (MHz) : 6240.0
Maximum freq (MHz) : 6335.5
nfpc=1? Number of fine channels per coarse channel = FFT length = 1? This bothered me with the previous file too. How could dedoppler search a coarse channel of width 1? Of course, for this testing, we are not doing dedoppler searching.
How does that Waterfall info display look to you other than the nfpc value?
@radonnachie Was my rawspec -f value wrong? How about the -t value? I was using the same values that you recommended for the previous file (ATA_guppi_59730_78085_225768981_J0332+5434_off_0001-beam0000) which is now discarded.
@radonnachie The new file successfully replaces the old one in rawspec_testing. I will try out the floating_point PR now that this is resolved.
The waterfall header-info displayed looks good to me. The nfpc is surely the result of -f, as you said the FFT length. In reality we'd probably really want to get to 1 Hz wide fine channels, so -f 524288. That would require much more coarse time though, for even a single fine-spectra output. It seems I only provided you with 32*1536=49152 time samples. Thanks @texadactyl!