forked from AMDComputeLibraries/Nekbone
-
Notifications
You must be signed in to change notification settings - Fork 0
/
CHANGES
executable file
·93 lines (76 loc) · 3.82 KB
/
CHANGES
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
***********************************************************
* Changes in 3.1 *
***********************************************************
-Fixed small bug that would cause floating point errors for
large element runs
-Allow the code to continue when specific domain decomp. is
not included in the data.rea file. The code will find the
appropriate decomposition, so there is no need to fail
without user provided input
-Fixed logical comparisons for BG platforms.
***********************************************************
* Changes in 3.0 *
***********************************************************
-New Multigrid preconditioner added as an option to the
mini app cases. To run with the new option, add 'MGRID'
to the PPLIST parameter set in makenek.
-Added nx0 and nxN capabilities back into nekbone so that
tests may be ran with multiple polynomial orders in one
test submission, without recompiling the source code.
-Implemented a step increment variable for the number of
elements and polynomial order. Set in the data.rea file,
users can now control the increase of the element counts
and the polynomial order increase. Default is set to 1,
which is the standard case.
-Added user control of the element and processor decomposition.
Parameters are read in through the data.rea file to set the
processor and element x,y,z values. If invalid values are
provided, nekbone will calculate the correct values.
-Added new examples, nek_comm, nek_dlay, nek_mgrid, example2,
and example3.
Nek_comm example only runs
the communication/platform timer. There is no geometry for
this case and is set to test the latency, bandwidth, matrix
matrix products used in nekbone.
Nek_dlay example runs the standard nekbone example, but
includes a simulated processor load imbalance. A delay
is applied on a node basis to all MPI calls.
Nek_mgrid example runs an example with the multigrid
preconditioner turned on.
These tests are triggered by adding "NEKCOMM", "NEKDLAY",
"MGRID"
to the PPLIST in the makenek script.
examples 2 and 3 demonstrate some of the new user control
options included in this update.
More details are in the readme files included in each example.
-Cleaned up the memory and include files to rid nekbone of
unused arrays
-Cleaned up stdout output to be more clear and concise
***********************************************************
* Changes in 2.1 *
***********************************************************
-Fixed nx0 and nxN control of polynomial order. Default is
now to use lx1 until further notice. Variable
nx1 caused memory unstabilities and needs further
development.
-Fixed a memory copy bug in the jl/ array transfer code.
sarray_trasfer, used for the tuple transfer, should
be fixed now.
***********************************************************
* Changes in 2.0 *
***********************************************************
-Subroutine gsync() has changed to nekgsync() to avoid
possible conflict on certain architectures
-Executable is renamed 'nekbone' to replace 'nekproxy' and
other naming changes.
-iel0 and ielN set in data.rea file are now used to control the
range of tests ran. Test range in size from iel0
to ielN elements per process. (prevoiusly tests
were ran from 1 to lelt elements per process) The
maximum value of ielN is lelt.
-nx0 and nxN set in data.rea file are now used to control the
range of polynomial orders. Ranging from nx0 to
nxN, where nxN<=lx1 (which is set in SIZE). Previously
tests only ran with nx1=lx1. The default is set to
reflect this, but nekbone now supports a range of
polynomial orders without recompiling the code.