1. 09 Apr, 2015 9 commits
  2. 08 Apr, 2015 8 commits
  3. 06 Apr, 2015 2 commits
  4. 05 Apr, 2015 3 commits
    • Evan Goetz's avatar
      Fix fabsf in TwoSpecSpecFunc · ebdad711
      Evan Goetz authored
      Original: 2cda1e2539dc2ffe9d0f44187e60da4955f84437
      ebdad711
    • Evan Goetz's avatar
      Fix automake file for TwoSpect · 6210fd52
      Evan Goetz authored
      Original: 74c550632e69b3e493d58fc8aa8635aa095bc3ce
      6210fd52
    • Evan Goetz's avatar
      Patch updating TwoSpect for template based search, coherent SFT analysis,... · 1802f912
      Evan Goetz authored
      Patch updating TwoSpect for template based search, coherent SFT analysis, speed improvements, other minor modifications
        - Fixes #1759
        Added computeSignalDetector program to test detector-signal relationship for coherent analysis
        Fixed some coherent SFT additions
        Fixed a few minor bugs in scripts
        Moved FAR and FAP functions to separate source file
        Moved SFT functions to separate source file
        Rename fastchisqinv to TwoSpectSpecFunc
        CDF functions now in cdfdist
        Rename templateStruct to TwoSpectTemplate
        New TwoSpectParamSpaceSearchVals struct to hold the range of parameter space values
        Generic template generation for pixel locations indexed from the non-Doppler-modulated frequency bin
        Fully embrace the UserInput_t struct and get rid of inputParamsStruct
        Change user input Tcoh to Tsft
        Need to pass detectors/rng to functions since we use the UserInput_t struct
        Added TwoSpectTemplateBank program to save a vector of templates to a file
        User can pass tempalate bank file after saving a vector from the TwoSpectTemplateBank program
        Alpha/Delta sky locations removed from some functions in favor of SkyPosition struct
        Added skyposTest to test scripts
        Truncate calculations of Gaussian template weights where the weight will be zero
        Move loop counters to be defined in loop
        Added template speed calculation program to test improvements. This could also be leveraged for unit tests in the future
        New mem alignment of REAL4Vector from VectorMath
        VectorMath exp and sin/cos functions
        Alignment for REAL8Vector and  alignedREAL4VectorArray
        Fix up getMultiTimestampsFromSegmentsFile
        Improve speed of Davies algorithm
        Add coherentSFT test features
        testVectorMath update
        Update doxygen details
        Coherent SFT sum only includes sensitive detectors
        Fix start time issues by automatically shifting t0 and printing warning
        Change user options from useSSE and useAVX booleans to vectorMath integer argument
        Improve speed of coherent SFT addition function
        Fixes number of SFTs loaded into catalog and possible divide by zero in coherent analysis when first detector is badly oriented to source
        Add templateVectorIndex ID for use when doing template bank searches
        Remove old KSKtests code
        Implement these alignedREAL4VectorArray in IHS functions to speed up computations
        Remove unused vectormath functions
        Remove unused statistics functions
        Function to avoid zeros in denominator of Dirichlet kernel ratio
        Added some additional vector math functions
        For any of the coherent SFT additions, don't add other detectors that are less sensitive
      Original: 902af6deffe7ef735609aa8e2778a09666f0c9aa
      1802f912
  5. 04 Apr, 2015 3 commits
  6. 03 Apr, 2015 6 commits
  7. 02 Apr, 2015 9 commits
    • Karl Wette's avatar
      LALSIMD.h: namespace enum and macros with LAL_... prefixes · e9b6ee68
      Karl Wette authored
      - Refs #1950
      Original: 5c24f0c7d84f8dd8c692ba7f49b9b20a530c2bbf
      e9b6ee68
    • Karl Wette's avatar
      LALSimulation: disable broken test · e9ea791b
      Karl Wette authored
      Original: 48e8d5182b02b800790726904a5e85a682555666
      e9ea791b
    • Karl Wette's avatar
      Implement runtime SIMD detection and dispatch for VectorMath functions · d896f77d
      Karl Wette authored
      - Autoconf macro LALSUITE_WITH_SIMD now just checks for compiler support,
        no runtime checks (so removed unneeded AX_... macros). These checks are
        always performed so that specialised SIMD code can always be compiled.
      - Remove ./configure option --with-simd, no longer needed.
      - Public LALSIMD.h headers exposes SIMD runtime detection and some macros.
        Guru users can set LAL_SIMD_ISET to down-select SIMD instruction set.
      - Internal simd_dispatch.h contains useful macros for runtime dispatching.
      - lal-simd-detect is a utility which prints SIMD compiler/machine support.
      - VectorMath uses runtime dispatch to select SSE|SSE2|AVX|AVX2 versions of
        XLALVectorMath...() functions, using an internal function pointer and
        dispatch function. Internal details of implementation are hidden behind
        API, and everything is macro-generated for ease of adding new functions.
      - VectorMathTests.h test script runs VectorMathTest with different SIMD
        instructions, try to test at least SSE and AVX if supported.
      - Remove LAL_WITH_SIMD from LALPulsar, no longer needed.
      - Refs #1950
      Original: 62849868479af666596defcf6ee255bd1bd1de76
      d896f77d
    • Chris Pankow's avatar
      LALSimInspiral.{c,h}: Add another convenience mode adder · 78726aa5
      Chris Pankow authored
      As a complement to XLALSimAddModeFroModes, XLALSimNewTimeSeriesFromModes will
      allocate the memory to the passed pointers before calling
      XLALSimAddModeFromModes.
      Original: 071f98d2c5b5b791afcedf694f3b1e17a2c56d00
      78726aa5
    • Chris Pankow's avatar
      LALSimInspiral.{c,h}: convenience mode adder func · 4e247718
      Chris Pankow authored
      XLALAddModeFromModes is a convenience function which takes
      a SphHarmTimeSeries as input and calls AddMode for each valid
      member of the SphHarmTimeSeries. This simply allows for a h+, hx
      from a SphHarmTimeSeries within one function.
      Original: 2b485d00e827be2cd087ed7f15aa4530d8778bc1
      4e247718
    • Chris Pankow's avatar
      rejigger scheme from de-LALification of time series · 9ef5ca8a
      Chris Pankow authored
      The assumption made about the mode time series in caf68c8196c45510b5538615cdba3f457a14e242 was incorrect. The pared down mode time series did not have the same epoch between detectors. This replaces some of that patch with the old version and opts, instead, to hand the factored likelihood function another instrument-keyed dictionary with the epochs, while retaining the numpy (instead of LAL) based time series. This also fixed some broken logic in the conversion process.
      Original: 48a8d3dc4941890fd0da2f84416486c0091802a4
      9ef5ca8a
    • Chris Pankow's avatar
      factored_likelihood.py: Move indexing calculation outside of loop / aesthetics · f1592457
      Chris Pankow authored
      Original: 7956f8d3ae39c7ccb0def4c4791c2e9bd1226e14
      f1592457
    • Chris Pankow's avatar
      factored_likelihood.py: Remove redundant rounding · a46947ab
      Chris Pankow authored
      Original: 7ef07811d1acb9543f48512f76e548dcd706093e
      a46947ab
    • Chris Pankow's avatar
      rapidpe: Don't use LAL time series in rho_lm loop · b304ae03
      Chris Pankow authored
      The mode time series handed to the network log likelihood function was using LAL timeseries. This invoked a copy of the sliced numpy array (ts.data.data member) instead of a view since (we suppose) the 'array' object did not actually own the data and thus could not provide a copy. This is ameliorated by eschwing the LAL time series in favor of numpy arrays well before the loop, and moving the epoch of the time series (now ensure to be the same for all detector) to the time grid points. The construction of the time grid points is also moved outside the likelihood definition so as not to be redundnantly recreated.
      Original: f93043326cdbe067ec492e3c42434b4cb4e4a37a
      b304ae03