fd4c16a86c
25 Sep 15: - Fixed bug in Suave.tm (oversight of 'nmin' argument). 7 May 15: - Fixed sign error in Cuhre's Rule.c (copy-&-paste error from higher-precision weights in degree-9 rule). 31 Mar 15: - Version 4.2: Added flag for resetting the state even if a state file is available in Vegas (Bit 5 of 'flags'). Together with Bit 4 (retain state file) the grid adapted for one integrand can be re-used on another integrand. - Added --with-real=10 and --with-real=16 configuration options for higher-precision versions of Cuba. 26 Nov 14: - Divonne now applies Flatten to the Given list, hence either a list of coordinates (as before) or a list of points can be given. - Tweaked MathLink code for compatibility with older Mathematica versions. 25 Nov 14: - Version 4.1: Added 'ninsuf' argument to Suave. When computing the integral value of a subregion, samples from a former pass contribute only of their number is larger than ninsuf. Increasing ninsuf may reduce jumps in the chi^2 value. 25 Nov 14: - Fixed subtle bug in Cuhre which could lead to longer run times than in versions 1 and 2. - Fixed miscounting of degrees of freedom in rare cases in Suave. 9 Oct 14: - Divonne now passes the userdata argument to the peakfinder function, too. 27 Aug 14: - Fixed problem due to which CUBACORESMAX was effectively ignored. - Fixed issues with MathLink compile for Mathematica 10. 15 Aug 14: - Version 4.0: distinguishes Accelerators (GPUs) and Cores. New distribution strategy (see manual), e.g. can now prescribe maximum number of points per core, necessary for accelerators, allows for load-levelling on cores. - No. of cores/accelerators and max. points per core/accel can be set with environment variables CUBACORES (as before) and CUBACORESMAX (or API call cubacores), CUBAACCEL and CUBAACCELMAX (or API call cubaaccel). API calls generally supersede the environment. - Can choose minimum verbosity level via environment variable CUBAVERBOSE = 0..3 (or API call cubaverbose). - Can start cores/accels explicitly via cubafork/cubawait calls, or have the integrator keep them running at exit. 18 Apr 14: - Fixed uninitialized variable in update of Apr 11 that could lead to a potential segfault in Divonne. 11 Apr 14: - Added invocation of worker init/exit functions also for the master whenever the number of samples is too small to be distributed. 21 Mar 14: - Fixed glitch in Sobol quasi-random numbers for the long long int versions introduced in the last update (forgotten typecast). 18 Mar 14: - Fixed infinite loop in Sobol quasi-random numbers for maxeval values >~ 2^30 (32-bit versions) or 2^62 (64-bit 'll' versions). 22 Jan 14: - Removed 'Chop' from the Mathematica sampling procedure, which would introduce an absolute lower scale for integrals quite close to zero. 8 Jan 14: - Restricted detection of Qt to version 4. 18 Dec 13: - Minor fixes in the MacOS build of partview (thanks to Hong Xu). 11 Dec 13: - Version 3.3: Added new argument 'nvec' to let the integrand function have more than one point at a time, e.g. to take advantage of vectorization. 5 Nov 13: - Fixed MathLink compilation issues with MacOS 10.9/clang++. - Fixed spurious warning in partview.
0 lines
Plaintext
0 lines
Plaintext