Original: 2ae5d47dc295bebcf75a85f9d3f93f935054721f

Original: b58e80212d9520568d44a96189ee3551d279a7d7

This speeds up BAYESTAR by about 50%.
Original: 61ef915ad390a9f8a51f7600561ce2fe714f9078

For large numbers of terms, logaddexp() performs about twice as many
transcendental function evaluations as is necessary.
This patch results in a speedup of about 25%.
Original: 41634d246522aefd772d5f6b19d13277476b9c07

Original: aedec99769c16311da156fba50bf0cc53b2ab59c

These small changes result in a speedup of about 30%.
Original: 26d753eb650828ec61faae72afb28680f6850d00

Original: fb499b27424c7d05f43c66771de41958b0318877

Original: 639abf85f15ee6a203b07a0e805f51805ae044bd

Original: f0551ce4104f392a36075c9c6e3f59f4fa894ed2

This adds transparent support for reading and writing multi-order HEALPix
maps. See also #4776.
Original: 2835aa2422cb8a3062ccfda6f1b194dac4821e1d

This will improve speed and memory efficiency for super-accurate
sky localizations arising from next-generation detector networks.
Based on mailing list discussion:
https://sourceforge.net/p/healpix/mailman/message/35598107/
Original: 57524be42c69609f21a23bbb709c3ff796dfb391

Original: c394fd2c5a930a9208d31a2e341b78e09f30713e

Original: c14d5bcebfdf6799c698ca08a9fc8ea78adb4c59

 - LAL: Addition of a linked list LALDict structure to store
        multiple variables (in src/tools/).
 - LALSimulation: Addition of a waveformparams file containing
                  standard methods to access/write variables
                  in a LALDict.
                  Critical changes in the ChooseWaveform interface,
                  affecting waveform generation calls.
                  Addition of a function transforming choosewaveform
                  input parameters into the ones used in lalinference.
                  Test codes updated accordingly (src/ and /test)
                  General LALSimInspiral doxygen documentation
                  changed.
 - LALInspiral: Change in waveform calls to comply with new interface
                (src/LALInspiralWave.c and
                python/lalinspiral/sbank/waveforms.py).
 - LALInference: Change in waveform calls to comply with new interface
                 (src/ and python/lalinference/bayestar/filter.py).
 - LALApps: Change in waveform calls to comply with new interface
            (src/inspiral/inspiralutils.c).
Original: def92e2f7fabb41fe1d4332d5effebfb9973523b

It was only used for making a plot in the BAYESTAR paper. Otherwise, it's
only used internally.
Original: 91154430358acd262220b3f6c4c1e67e37fa482d

Original: 21eea52a603d66512a92c2f7c77e77fe2a01131c

That's what gstlal produces.
Original: 94ad9b064c398fe69438d3f5887a452ab568e2e6

This allows pipelines to provide sub-threshold observations when the
signals in some detectors are too faint to produce triggers.
If the SNR time series are missing, then BAYESTAR will print a warning
and synthesize fake SNR time series from the autocorrelation sequences
of the whitened templates.
Original: 3a135d1cf4bfee4bb566e9972d0e7009a7a7b161

Original: 226f8fc703a9963a5e988d1bd43ce7f42eaa544e

The GSL error handler fundamentally cannot be set reentrantly, so it
is a potential source of bugs in multi-threaded programs.
Original: da4087e1c1b623dd1c6234d59cf08521aadf25ad

The issue was setting the static function pointer `old_handler` to
the function `ignore_underflow`, which itself calls `old_handler.
This patch is a hacky workaround. A better solution would be to
maintain a thread-local stack of GSL error handlers, perhaps using
the C11 `thread_local` access specifier.
See, for example, <http://en.cppreference.com/w/c/thread/thread_local>.
Original: c1a050cdb05f9f0d3bb749e1d012e14be87ee676

Original: 8ff48527858f5e689fc2fcc7494841fde9040ecc

Original: c6821276cb07017ec035ac454a8eee717ba8ff17

Original: fba409019a63b312b31e17370d3545808908b104

Original: 33354394d18ca89c4c1d82b9de0709e4a2a56e7f

Original: ab07ecf072cc6be7a698726716f9387e8c87e017

Original: bfab918395b737a73d9d95241f67315a71f1c37b

The unit test for the signal amplitude model is now much more elaborate,
anchoring the phase convention on injecting a signal with
XLALSimInspiralFD / XLALComputeDetAMResponse. This demonstrated that there
was a mismatch in the signal amplitude model between the code and the
BAYESTAR paper, which this patch corrects.
This patch also corrects a mismatch in the autocorrelation likelihood
between the code and the BAYESTAR paper.
The *default* phase convention for the high-level interface is still
anti-FINDCHIRP (conforming with gstlal and MBTA).
Fixes #2354.
Original: 120830869b6bc8b9f2afc9684ef08b6af2cc6bfc

Original: 0d87a92fc1ebb9afcced7b0d1e38c012034cf245

Original: 3f0a7b24d48952d6dca1c763184ce16170cdbab4

I had inadvertently used the complex-valued Catmull-Rom interpolation
function for real-valued interpolation. The real-valued version is about
twice as fast and includes some special cases for infinite or NaN data
points.
I am also reinstating the Mathematica unit tests.
Original: 6ea910f0b7c187d4672955acf09f66ab6b61ceb6

Compared to the old method of evaluating the distance integral on the
fly using fixed-order Gaussian quadrature, this is orders of magnitude
faster and has better numerial accuracy.
Original: a4aa0637ea0f2acdcc7f90771d35c33f473c5978

- For ssize_t on Debian squeeze
- Refs #2097
Original: 52b01d021ae24d06e8657c872180f021ca359a52

Original: b31fe518ae45d29ffe18f4488ed2808b544970b1

Original: 4d8569e69a74949ff19acee1a88a6e662d1b9be5

Original: f19697c71b388bf0f5cae041c3eaf908bd5eddeb

This reverts commit 51e5c5e30565537fbc4eafa2d94cd0810f1134bc.
Original: cadc3c382457135b98cd52937bfeb073e4462ad0

This reverts commit 1ca1503a08528482708ca864c171212a0dd46f96.
Original: 0a2422507c4f0ea80561970858b9db76609acf21

Original: aea2325f7c8768a985b14d94dd5b136329002ab4

Original: 7cd98f84ca7e4c3e11fbb19a3ff0a323d1fa895d