Civil Engineering Reference
In-Depth Information
New dynamic integer arrays:
jeig
intermediate array
intermediate array
nu
New dynamic real arrays:
alfa
intermediate array
intermediate array
beta
intermediate array
del
diagonal mass matrix
diag_tmp
element consistent mass matrix
ecm
intermediate array
eig
accumulated element mass matrix
emm
distributed
{
U
}
in product
{
U
}+
[
A
]
{
V
}
ua_pp
distributed
{
V
}
in product
{
U
}+
[
A
]
{
V
}
va_pp
vdiag_pp
distributed diagonal vector
v_store_pp
distributed stored Lanczos vectors
w1_pp
intermediate array
local element array
x
intermediate array
z_pp
Loop
elements_1
is a good illustration of the transition from serial to paral-
lel code. Routine
geometry_8bxz
, to create a cuboidal mesh of 8-node hexahedra,
replaces
geom_rect
,and
find_g
replaces
num_to_g
. Distributed arrays
g_num_pp
,
p_g_co_pp
and
g_g_pp
replace their serial counterparts
g_num
,
g_coord
and
g_g
.
Loop
elements_2
with embedded loop
integrating_points_1
clearly carries over
although the parallel version assumes uniform element properties throughout the mesh. In
all other aspects the serial and parallel programs are essentially identical although of course
the serial and parallel solution algorithm libraries are not interchangeable. Data are listed as
Figure 12.35 involving about a quarter of a million equations with results as Figure 12.36
and some performance statistics as Figure 12.37. For larger data sets, Figures 12.38 and
12.39 show performance measured on an SGI Origin 3000 computer.
nels nxe nze nip
78125 25 25 8
aa bb cc rho
0.04 0.04 0.04 1.0
e v
1.0 0.3
nmodes
5
el er
0.0 1.0
lalfa leig lx lz acc
5000 20 100 5000 1.0e-9
Figure 12.35
Data for Program 12.8 example