From: David Wells <drwells@email.unc.edu>
Date: Mon, 3 Sep 2018 21:22:01 +0000 (-0400)
Subject: Execute quick tests in parallel.
X-Git-Tag: v9.1.0-rc1~717^2
X-Git-Url: https://gitweb.dealii.org/cgi-bin/gitweb.cgi?a=commitdiff_plain;h=refs%2Fpull%2F7143%2Fhead;p=dealii.git

Execute quick tests in parallel.

The quick tests are currently set up as a custom target which immediately
launches ctest. The difficulty with this is that its not possible (AFAICT) to
get the number of jobs available from make at that point, so there is no way to
run these in parallel without giving an explicit number of jobs to ctest.

This PR sets the number of jobs to the number of CPUs on the machine: while
hardcoding this is not ideal, this is the right usage pattern for a build script
like

    make -j20
    make test

where all processors on the machine should be available.
---

diff --git a/doc/news/changes/minor/20180913DavidWells b/doc/news/changes/minor/20180913DavidWells
new file mode 100644
index 0000000000..882500f487
--- /dev/null
+++ b/doc/news/changes/minor/20180913DavidWells
@@ -0,0 +1,8 @@
+Fixed: <code>make test</code> now runs in parallel automatically (i.e., it uses
+a CMake command to determine the number of cores available). Since <code>make
+test</code> just immediately invokes <code>ctest</code> as a subprocess, it is
+currently not possible to link the number of globally available
+<code>make</code> processes with the number of concurrently run tests; i.e.,
+the number of cores <code>make test</code> uses is not changeable by users.
+<br>
+(David Wells, 2018/09/13)
diff --git a/doc/readme.html b/doc/readme.html
index 6751ee266b..de4c9b9747 100644
--- a/doc/readme.html
+++ b/doc/readme.html
@@ -183,8 +183,8 @@
         </li>
 
         <li> If your machine has multiple processors, use <code>make
-	  -jN</code> in the last step, where <code>N</code> is the number of simultaneous build processes you want <code>make</code> to use at any given time. Allowing <code>make</code> to use more simultaneous build processes (assuming you have that many processor
-            cores) will greatly lower the build time.
+	  -jN install</code> in the second to last step instead, where <code>N</code> is the number of simultaneous build processes you want <code>make</code> to use at any given time. Allowing <code>make</code> to use more simultaneous build processes (assuming you have that many processor
+            cores) will greatly lower the build time. <code>make test</code> automatically runs in parallel and no <code>-jN</code> flag should be supplied to it.
         </li>
 
         <li> If you do not intend to modify the <acronym>deal.II</acronym> sources and recompile things, then you can remove the <code>build/</code> directory after the last step.
diff --git a/tests/quick_tests/run.cmake b/tests/quick_tests/run.cmake
index 4b4567bce4..4e4c212998 100644
--- a/tests/quick_tests/run.cmake
+++ b/tests/quick_tests/run.cmake
@@ -13,13 +13,22 @@
 ##
 ## ---------------------------------------------------------------------
 
-# This file is run when "make test" is executed by the user and is 
+# This file is run when "make test" is executed by the user and is
 # responsible for running the tests and printing some helpful
 # error messages.
+include(ProcessorCount)
+PROCESSORCOUNT(_n_processors)
+IF(_n_processors EQUAL 0)
+  SET(_n_processors "1")
+ENDIF()
 
 SEPARATE_ARGUMENTS(ALL_TESTS)
 
-EXECUTE_PROCESS(COMMAND ${CMAKE_CTEST_COMMAND} --force-new-ctest-process --output-on-failure -O quicktests.log RESULT_VARIABLE res_var)
+EXECUTE_PROCESS(COMMAND ${CMAKE_CTEST_COMMAND} -j${_n_processors}
+  --force-new-ctest-process
+  --output-on-failure
+  -O quicktests.log
+  RESULT_VARIABLE res_var)
 
 IF(NOT "${res_var}" STREQUAL "0")
   MESSAGE("