Merge pull request #182 from intel/develop

Develop

setup.py

@@ -19,11 +19,9 @@ def run(self):
 
 setup(
     name='yask',
-    version='v2-alpha',
-    description='YASK--Yet Another Stencil Kernel: '
-    'A framework to facilitate exploration of the HPC '
-    'stencil-performance design space',
-    url='https://01.org/yask',
+    version='v2.15.11',
+    description='YASK--Yet Another Stencil Kernel',
+    url='https://github.com/intel/yask',
     author='Intel Corporation',
     license='MIT',
     packages = ['yask'],

src/common/common_utils.cpp

@@ -46,7 +46,7 @@ namespace yask {
     // for numbers above 9 (at least up to 99).
 
     // Format: "major.minor.patch".
-    const string version = "2.15.10";
+    const string version = "2.15.11";
 
     string yask_get_version_string() {
         return version;

src/kernel/lib/soln_apis.cpp

@@ -113,20 +113,23 @@ namespace yask {
 
         // Report ranks.
         os << endl;
-        os << "Num MPI ranks: " << _env->get_num_ranks() << endl;
-        os << "This MPI rank index: " << _env->get_rank_index() << endl;
+        os << "Num MPI ranks:            " << _env->get_num_ranks() << endl;
+        os << "This MPI rank index:      " << _env->get_rank_index() << endl;
 
         // report threads.
-        os << "Num OpenMP procs: " << omp_get_num_procs() << endl;
-        set_all_threads();
-        os << "Num OpenMP threads: " << omp_get_max_threads() << endl;
-        set_region_threads(); // Temporary; just for reporting.
-        os << "  Num threads per region: " << omp_get_max_threads() << endl;
-        set_block_threads(); // Temporary; just for reporting.
-        os << "  Num threads per block: " << omp_get_max_threads() << endl;
-
-        // Set the number of threads for a region. It should stay this
-        // way for top-level OpenMP parallel sections.
+        {
+            os << "Num OpenMP procs:         " << omp_get_num_procs() << endl;
+            int rt, bt;
+            int at = get_num_comp_threads(rt, bt);
+            os << "Num OpenMP threads avail: " << opts->max_threads <<
+                "\nNum OpenMP threads used:  " << at <<
+                "\n  Num threads per region: " << rt <<
+                "\n  Num threads per block:  " << bt << endl;
+        }
+
+        // Set the number of threads for a region. The number of threads
+        // used in top-level OpenMP parallel sections should not change
+        // during execution.
         int rthreads = set_region_threads();
 
         // Run a dummy nested OMP loop to make sure nested threading is

src/kernel/lib/utils.hpp

@@ -25,18 +25,18 @@ IN THE SOFTWARE.
 
 #pragma once
 
+// Provide the needed definitions for NUMA support.
+// This is fairly convoluted because of the inconsistency of
+// support on various OS releases.
+// The USE_NUMA* vars are set in the Makefile.
 #ifdef USE_NUMA
 
 // Use numa policy library?
 #ifdef USE_NUMA_POLICY_LIB
 #include <numa.h>
 
-// Use mmap and mbind directly?
-#else
-#include <sys/mman.h>
-
 // Use <numaif.h> if available.
-#ifdef USE_NUMAIF_H
+#elif defined(USE_NUMAIF_H)
 #include <numaif.h>
 
 // This is a hack, but some systems are missing <numaif.h>.
@@ -54,7 +54,6 @@ extern "C" {
 #define MPOL_BIND        2
 #define MPOL_INTERLEAVE  3
 
-#endif
 #endif
 #endif
 

src/kernel/lib/yask.hpp

@@ -82,6 +82,7 @@ typedef int MPI_Request;
 #include <unistd.h>
 #include <stdint.h>
 #include <immintrin.h>
+#include <sys/mman.h>
 #ifdef USE_PMEM
 #include <memkind.h>
 #include <sys/syscall.h>

utils/lib/YaskUtils.pm

@@ -57,7 +57,7 @@ our @log_keys =
    'binary invocation',
    'num MPI ranks',
    'num ranks',
-   'num OpenMP threads',
+   'num OpenMP threads', # also matches 'Num OpenMP threads used'.
    'num threads per region',
    'num threads per block',
    'total overall allocation',
@@ -193,7 +193,10 @@ sub getResultsFromLine($$) {
     }
   }
 
-  # special cases for manual parsing.
+  # special cases for manual parsing...
+  # TODO: catch output of auto-tuner and update relevant results.
+
+  # Output of 'uname -a'
   if ($line =~ /^\s*Linux\s+\w+\s+(\S+)/) {
     $results->{$linux_key} = $1;
   }
@@ -216,13 +219,14 @@ sub getResultsFromLine($$) {
       # short key.
       my $sk = substr $key,0,$klen;
 
-      # quick match?
+      # match to short key?
       return if !exists $proc_keys{$sk};
 
       # look for exact key.
       for my $m (keys %{$proc_keys{$sk}}) {
 
         # match?
+        # only check that beginning of key matches.
         if ($key =~ /^$m/) {
           $val =~ s/^\s+//;
           $val =~ s/\s+$//;