-
Notifications
You must be signed in to change notification settings - Fork 11
/
Copy pathChangeLog
113 lines (74 loc) · 4.89 KB
/
ChangeLog
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
2019-5-3 Pierre Kestener
* Version 1.5 : implement cmake build system (simpler to use and maintain than autotools); autotools still usable.
2019-4-29 Pierre Kestener
* Version 1.4 : refresh autotools build system; compiling and running again on up-to-date systems
2015-1-31 Pierre Kestener
* Version 1.3 : some refactoring in hydrodynamics schemes
* both CPU and GPU implementation are slightly improved: each of
them have now 3 versions which varies by the amount of memory required.
2014-11-3 Pierre Kestener
* Version 1.2 : bug fixes and minor improvements
* turbulence simulation can now output structure functions computation on-line
* add numpy file format to dump structure function computation results (easy
to reload in python)
* add Fourier power spectrum compution for turbulence simulation analysis
* update CUDA kernel to use read-only data cache (const restrict)
* add new initial condition for stratified shearing box problems
* minor improvements in example parameter files
2013-8-27 Pierre Kestener
* Version 1.1 : bug fixes and features implementation
* major bug fix: in both CPU / GPU version of the MHD numerical
scheme, there was a problem in the computation of the
electromotive force (see file riemann_mhd.h, routine compute_emf)
* new major feature: refactoring numerical scheme to reduce memory
footprint. See C++ class suffixed by "ZSlab"; this new
implementation is the one that must be used for very large
simulation (ex. 2000^3 for MHD turbulence simulation on Keeneland)
* new major feature: output using PNetCDF library; restart can
load a NetCDF file; minor feature: one can output optionnally
output data on external faces for making movies (output at much
high frequency than regular output of the entire 3D domain).
* new feature: new init condition for hydro and MHD turbulence
application; implementation of a simple forcing term
* new feature: implement history routine for computing on-line
analysis quantities (ex: alpha in MRI simulations, Mach number in
turbulence simulation, ...)
* new feature: analysis directory containing simple
post-processing tools adapted to large data files (PNetCDF):
Fourier power spectrum, slice extraction
* new minor feature: simple MPI app to convert large file from
parallel HDF5 to PNetCDF
* new minor feature: small utility to save data in python numpy files (.npy)
* new minor feature: some python script for post-processing analysis
(for data up to 1000^3); for larger data, use tools in the
analysis sub-directory.
* fix typo in slope_mhd.h: routine slope_unsplit_hydro_3d for
slope type = 3
* new feature: add Riemann solver LLF (Local Lax-Friedrich)
* new feature: add 2D magnetic Riemann solvers HLLA, HLLF and LLF
2012-8-27 Pierre Kestener
* Version 1.0.1 : bug fixes and minor improvements
* major bug fix: in the CUDA version of the code, when running the MRI problem, total mass inside box was slightly increasing with time. The bug was fixed in the kernel_mhd_flux_update_hydro_v4_shear CUDA kernel.
* bug fix: this bug only affects MPI/CUDA version for the RMI problem (wrong index in remapping formula in shear border condition). This bug was causing the simulation run to stop after several 1000's of time steps
* bug fix in viscosity tensor: affect both CPU/GPU: only one term was slightly false (almost neutral bug)
* new feature: now GPU memory array can be allocated using pitched or linear memory (before pitched was the default). On new hardware, pitch memory is not anymore a must. Linear memory gives the same performane (thanks to cache improvements). Besides, for a given problem, the amouint of required memory is almost twice smaller when using linear memory, so linear memory is now the default choice.
* fix MRI history method for MPI run: output was wrong
* improvement: much better restart procedure; now one can read an HDF5 file at a twice smaller resolution and perform an upscale. This is usefull for MRI problem to restart a job by increasing resolution with a factor of 2.
2012-5-23 Pierre Kestener
* Version 1.0; package name is now RAMSES-GPU with a dedicated website :
http://www.maisondelasimulation.fr/projects/RAMSES-GPU/html/
* MRI with both viscous/resistive terms working
2012-3-14 Pierre Kestener
* Version 0.4.0 is the first version with 3D MHD on CPU/GPU and
with MHD shearing box capabilities
* fix bug in MHD time step computation routine (could
only be seen on large spatial domain causing memory fault)
* backport unsplit scheme into pure hydro routines
* improve HDF5/VTK output routines
2011-4-20 Pierre Kestener
* ChangeLog file was never filled. Let's start !
* SoftwareName : As 2D MHD solver has been implemented on GPU, let's take the
opportunity to change the name of the software package from
HybridHydro to PANAME (PArallel Numerical Algorithms for MHD).
* Version 0.3.0 is the first version with MHD capabilities (for
now only 2D on both CPU / GPU has been tested).