Hello all,<br>
<br>
With the BBv1 testing infrastructure, I'd like to be able to build a
test executable and then run it several times with different sets of
parameters. Each different run should be considered as a separate test
(some could fail while others do not). I have a kludge to do it, which
basically creates a bunch of targets dependent on the same sources,
which is quite trivial:<br>
<br>
rule mpi-runtest (<br>
&nbsp; target : sources + : requirements * : local-build * : args * : schedule * )<br>
{<br>
&nbsp; schedule ?= 1 2 17 ;<br>
<br>
&nbsp; for local procs in $(schedule)<br>
&nbsp; {<br>
&nbsp;&nbsp;&nbsp; local gRUN_TEST_ARGS = -np $(procs) $(args) ;<br>
&nbsp;&nbsp;&nbsp; local gRUN_TEST_ARGS2 = $(MPI_TEST_ARGS) ;<br>
<br>
&nbsp;&nbsp;&nbsp; # Munge the target to include the number of processors.&nbsp;&nbsp; <br>
&nbsp;&nbsp;&nbsp; local target = &quot;$(target)-$(procs)&quot; ;<br>
<br>
&nbsp;&nbsp;&nbsp; # Build the MPI executable test<br>
&nbsp;&nbsp;&nbsp; local result = [ <br>
&nbsp;&nbsp;&nbsp;&nbsp;&nbsp; boost-test $(sources)<br>
&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp; &lt;template&gt;mpi<br>
&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp; : RUN_MPI<br>
&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp; : $(requirements)<br>
&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp; : $(target)<br>
&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp; : $(local-build)<br>
&nbsp;&nbsp;&nbsp; ] ;<br>
&nbsp; }<br>
}<br>
<br>
Unfortunately, this kludge means that everything $(sources) in sources
is compiled once for each target that we create, e.g., we'll get:<br>
&nbsp; all-gather-1/.../all-gather-1.exe<br>
&nbsp; all-gather-1/.../all-gather-1.o<br>
&nbsp; all-gather-2/.../all-gather-2.exe<br>

&nbsp; all-gather-2/.../all-gather-2.o<br>
&nbsp; all-gather-17/.../all-gather-17.exe<br>

&nbsp; all-gather-17/.../all-gather-17.o<br>
&nbsp; // ...<br>
<br>
I've taken various stabs at sharing at least some of the intermediate
files among all of these tests, but without luck, yet. It seems that we
should either (1) generate a single all-gather/.../all-gather.exe and
then stage/clone it over into the other test locations (e.g.,
all-gather-1/.../all-gather-1.exe) or (2) build all of the .o files in
all-gather/.../ and then link separate
all-gather-1/.../all-gather-1.exe, all-gather-2/.../all-gather-2.exe,
etc. executables. <br>
<br>
Does anyone know how I can achieve this functionality in BBv1's testing environment?<br>
<br>
&nbsp; Doug<br>
<br>
P.S. You can safely ignore the &lt;template&gt;mpi and RUN_MPI stuff; it
just makes it so that we actually run the executables through an
auxiliarly executable called &quot;mpirun&quot;. The morbidly curious can check out boost-sandbox/libs/parallel/test.<br>

<br>