Better error message code. Remove commented out code.

This commit was SVN r28793.

ompi/mca/common/cuda/common_cuda.c

@@ -502,9 +502,11 @@ static int mca_common_cuda_load_libcuda(void)
     int retval, i, j;
     int advise_support = 1;
     bool loaded = false;
-    char *errs[NUMLIBS*SEARCHPATHS];
     char *cudalibs[NUMLIBS] = {"libcuda.so.1", "libcuda.so"};
     char *searchpaths[SEARCHPATHS] = {NULL, "/usr/lib64"};
+    char **errmsgs = NULL;
+    char *errmsg = NULL;
+    int errsize;
 
     if (0 != (retval = opal_lt_dlinit())) {
         if (OPAL_ERR_NOT_SUPPORTED == retval) {
@@ -528,10 +530,6 @@ static int mca_common_cuda_load_libcuda(void)
         }
     }
 
-    for (i = 0; i < NUMLIBS*SEARCHPATHS; i++) {
-        errs[i] = NULL; /* just to be safe */
-    }
-
     /* Now walk through all the potential names libcuda and find one
      * that works.  If it does, all is good.  If not, print out all
      * the messages about why things failed.  This code was careful
@@ -564,13 +562,13 @@ static int mca_common_cuda_load_libcuda(void)
                 if (NULL == libcuda_handle) {
                     str = opal_lt_dlerror();
                     if (NULL != str) {
-                        errs[j*NUMLIBS + i] = strdup(str);
+                        opal_argv_append(&errsize, &errmsgs, str);
                     } else {
-                        errs[j*NUMLIBS + i] = strdup("lt_dlerror() returned NULL.");
+                        opal_argv_append(&errsize, &errmsgs, "lt_dlerror() returned NULL.");
                     }
                     opal_output_verbose(10, mca_common_cuda_output,
                                         "CUDA: Library open error: %s",
-                                        errs[j*NUMLIBS + i]);
+                                        errmsgs[errsize-1]);
                 } else {
                     opal_output_verbose(10, mca_common_cuda_output,
                                         "CUDA: Library successfully opened %s",
@@ -594,13 +592,15 @@ static int mca_common_cuda_load_libcuda(void)
                 if (NULL == libcuda_handle) {
                     str = opal_lt_dlerror();
                     if (NULL != str) {
-                        errs[j*NUMLIBS + i] = strdup(str);
+                        opal_argv_append(&errsize, &errmsgs, str);
                     } else {
-                        errs[j*NUMLIBS + i] = strdup("lt_dlerror() returned NULL.");
+                        opal_argv_append(&errsize, &errmsgs, "lt_dlerror() returned NULL.");
                     }
+
                     opal_output_verbose(10, mca_common_cuda_output,
                                         "CUDA: Library open error: %s",
-                                        errs[j*NUMLIBS + i]);
+                                        errmsgs[errsize-1]);
+
                 } else {
                     opal_output_verbose(10, mca_common_cuda_output,
                                         "CUDA: Library successfully opened %s",
@@ -614,17 +614,12 @@ static int mca_common_cuda_load_libcuda(void)
     }
 
     if (loaded != true) {
+        errmsg = opal_argv_join(errmsgs, '\n');
         opal_show_help("help-mpi-common-cuda.txt", "dlopen failed", true,
-                       cudalibs[0], errs[0], cudalibs[1], errs[1], 
-                       cudalibs[2], errs[2], cudalibs[3], errs[3]);
-    }
-    
-    /* Cleanup error messages.  Need to do this after printing them. */
-    for (i = 0; i < NUMLIBS*SEARCHPATHS; i++) {
-        if (NULL != errs[i]) {
-            free(errs[i]);
-        }
+                       errmsg);
     }
+    opal_argv_free(errmsgs);
+    free(errmsg);
         
     if (loaded != true) {
         return 1;

ompi/mca/common/cuda/help-mpi-common-cuda.txt

@@ -142,20 +142,10 @@ cause the program to abort.
 Check the cuda.h file for what the return vale means.
 #
 [dlopen disabled] 
-While trying to load the supporting libcuda.so library, an error was 
-detected.  This error indicates that the Open MPI library was probably
-configured with the --disable-dlopen flag.  When the library is 
-configured in this way, CUDA support is disabled because CUDA support
-depends on the ability to dynamically open libraries.  Reconfigure
-without the --disable-dlopen flag to get around this problem.
-#
-[dladvise disabled]
-While trying to initialize the lt_dladvise structure, an error was 
-detected.  This error indicates that the Open MPI library was
-configured such that there is no support for the lt_dladvise 
-structure.  This is needed for properly opening the libcuda library.
-Look around for the OPAL_HAVE_LTDL_ADVISE macro and ensure that it
-is defined as a 1.  
+Open MPI was compiled without dynamic library support (e.g., with the
+ --disable-dlopen flag), and therefore cannot utilize CUDA support.  
+
+If you need CUDA support, reconfigure Open MPI with dynamic library support enabled.
 #
 [unknown ltdl error]
 While attempting to load the supporting libcuda.so library, an error
@@ -166,15 +156,9 @@ MPI developers.
    Error string: %s
 #
 [dlopen failed]
-The library attempted to open the supporting CUDA libraries but failed.
-  Library attempted: %s
-  Error string:      %s
-  Library attempted: %s
-  Error string:      %s
-  Library attempted: %s
-  Error string:      %s
-  Library attempted: %s
-  Error string:      %s
+The library attempted to open the following supporting CUDA libraries, 
+but each of them failed.
+%s
 #
 [dlsym failed]
 An error occurred while trying to map in the address of a function.

ompi/mca/pml/ob1/Makefile.am

@@ -66,10 +66,6 @@ mcacomponentdir = $(pkglibdir)
 mcacomponent_LTLIBRARIES = $(component_install)
 mca_pml_ob1_la_SOURCES = $(ob1_sources)
 mca_pml_ob1_la_LDFLAGS = -module -avoid-version
-#if MCA_ompi_cuda_support
-#mca_pml_ob1_la_LIBADD = \
-#    $(top_ompi_builddir)/ompi/mca/common/cuda/libmca_common_cuda.la
-#endif
 
 noinst_LTLIBRARIES = $(component_noinst)
 libmca_pml_ob1_la_SOURCES = $(ob1_sources)