utils/calc+bind: add --no-smt to keep a single PU per core

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

NEWS

@@ -93,6 +93,8 @@ Version 2.1.0
       See dynamic_SVG_example.html for an example.
   + Add --nodeset options to hwloc-calc for converting between cpusets and
     nodesets.
+  + Add --no-smt to hwloc-bind and hwloc-calc to ignore multiple
+    PU in SMT cores.
   + hwloc-annotate may annotate multiple locations at once.
   + Add a HTML/JS version of hwloc-ps. See contrib/hwloc-ps.www/README.
   + Add bash completions.

utils/hwloc/hwloc-bind.1in

@@ -1,5 +1,5 @@
 .\" -*- nroff -*-
-.\" Copyright © 2009-2018 Inria.  All rights reserved.
+.\" Copyright © 2009-2019 Inria.  All rights reserved.
 .\" Copyright © 2010 Université of Bordeaux
 .\" Copyright © 2009-2010 Cisco Systems, Inc.  All rights reserved.
 .\" See COPYING in top-level directory.
@@ -97,6 +97,9 @@ No location may be given since no binding is performed.
 \fB\-\-single\fR
 Bind on a single CPU to prevent migration.
 .TP
+\fB\-\-no\-smt\fR
+Only keep a single PU per core before binding.
+.TP
 \fB\-\-strict\fR
 Require strict binding.
 .TP

utils/hwloc/hwloc-bind.c

@@ -50,6 +50,7 @@ void usage(const char *name, FILE *where)
 #endif
   fprintf(where, "  --taskset      Use taskset-specific format when displaying cpuset strings\n");
   fprintf(where, "Input topology options:\n");
+  fprintf(where, "  --no-smt       Only keep a single PU per core\n");
   fprintf(where, "  --restrict <set>   Restrict the topology to processors listed in <set>\n");
   fprintf(where, "  --disallowed   Include objects disallowed by administrative limitations\n");
   fprintf(where, "  --hbm          Only consider high bandwidth memory nodes\n");
@@ -76,6 +77,7 @@ int main(int argc, char *argv[])
   int force = 0;
   int single = 0;
   int verbose = 0;
+  int no_smt = 0;
   int only_hbm = -1;
   int logical = 1;
   int taskset = 0;
@@ -142,6 +144,10 @@ int main(int argc, char *argv[])
 	single = 1;
 	goto next;
       }
+      if (!strcmp(argv[0], "--no-smt")) {
+	no_smt = 1;
+	goto next;
+      }
       if (!strcmp(argv[0], "-f") || !strcmp(argv[0], "--force")) {
 	force = 1;
 	goto next;
@@ -448,6 +454,11 @@ int main(int argc, char *argv[])
       fprintf(stderr, "--mempolicy ignored unless memory binding is also requested with --membind.\n");
   }
 
+  if (!got_cpubind && no_smt) {
+    hwloc_bitmap_copy(cpubind_set, hwloc_topology_get_topology_cpuset(topology));
+    got_cpubind = 1;
+  }
+
   if (got_cpubind) {
     if (hwloc_bitmap_iszero(cpubind_set)) {
       if (verbose >= 0)
@@ -466,6 +477,23 @@ int main(int argc, char *argv[])
 	fprintf(stderr, "Conflicting CPU and memory binding requested, adding HWLOC_CPUBIND_NOMEMBIND flag.\n");
       cpubind_flags |= HWLOC_CPUBIND_NOMEMBIND;
     }
+    if (no_smt) {
+      if (hwloc_get_type_depth(topology, HWLOC_OBJ_CORE) == HWLOC_TYPE_DEPTH_UNKNOWN) {
+	fprintf(stderr, "Topology has no Core object, ignoring --no-smt\n");
+      } else {
+	hwloc_obj_t core = NULL;
+	while ((core = hwloc_get_next_obj_covering_cpuset_by_type(topology, cpubind_set, HWLOC_OBJ_CORE, core)) != NULL) {
+	  int firstpu = hwloc_bitmap_first(core->cpuset);
+	  int hadpu = hwloc_bitmap_isset(cpubind_set, firstpu);
+	  assert(firstpu >= 0);
+	  /* remove the entire core */
+	  hwloc_bitmap_andnot(cpubind_set, cpubind_set, core->cpuset);
+	  /* put back its first PU if it was there */
+	  if (hadpu)
+	    hwloc_bitmap_set(cpubind_set, firstpu);
+	}
+      }
+    }
     if (single)
       hwloc_bitmap_singlify(cpubind_set);
     if (pid_number > 0)

utils/hwloc/hwloc-calc.1in

@@ -1,5 +1,5 @@
 .\" -*- nroff -*-
-.\" Copyright © 2010-2018 Inria.  All rights reserved.
+.\" Copyright © 2010-2019 Inria.  All rights reserved.
 .\" Copyright © 2009 Cisco Systems, Inc.  All rights reserved.
 .\" See COPYING in top-level directory.
 .TH HWLOC-CALC "1" "%HWLOC_DATE%" "%PACKAGE_VERSION%" "%PACKAGE_NAME%"
@@ -93,6 +93,9 @@ while a comma is used to separate indexes
 \fB\-\-single\fR
 Singlify the output to a single CPU.
 .TP
+\fB\-\-no\-smt\fR
+Only keep a single PU per core in the input locations.
+.TP
 \fB\-\-taskset\fR
 Display CPU set strings in the format recognized by the taskset command-line
 program instead of hwloc-specific CPU set string format.

utils/hwloc/hwloc-calc.c

@@ -46,6 +46,7 @@ void usage(const char *callname __hwloc_attribute_unused, FILE *where)
   fprintf(where, "  --taskset                 Use taskset-specific format when displaying cpuset strings\n");
   fprintf(where, "  --single                  Singlify the output to a single CPU\n");
   fprintf(where, "Input topology options:\n");
+  fprintf(where, "  --no-smt                  Only keep a single PU per core\n");
   fprintf(where, "  --restrict <cpuset>       Restrict the topology to processors listed in <cpuset>\n");
   fprintf(where, "  --disallowed              Include objects disallowed by administrative limitations\n");
   hwloc_utils_input_format_usage(where, 10);
@@ -65,6 +66,7 @@ static int intersectdepth = -1;
 static int hiernblevels = 0;
 static int *hierdepth = NULL;
 static int showobjs = 0;
+static int no_smt = 0;
 static int singlify = 0;
 static int taskset = 0;
 
@@ -101,6 +103,24 @@ hwloc_calc_hierarch_output(hwloc_topology_t topology, const char *prefix, const
 static int
 hwloc_calc_output(hwloc_topology_t topology, const char *sep, hwloc_bitmap_t set)
 {
+  if (no_smt) {
+    if (hwloc_get_type_depth(topology, HWLOC_OBJ_CORE) == HWLOC_TYPE_DEPTH_UNKNOWN) {
+      fprintf(stderr, "Topology has no Core object, ignoring --no-smt\n");
+    } else {
+      hwloc_obj_t core = NULL;
+      while ((core = hwloc_get_next_obj_covering_cpuset_by_type(topology, set, HWLOC_OBJ_CORE, core)) != NULL) {
+	int firstpu = hwloc_bitmap_first(core->cpuset);
+	int hadpu = hwloc_bitmap_isset(set, firstpu);
+	assert(firstpu >= 0);
+	/* remove the entire core */
+	hwloc_bitmap_andnot(set, set, core->cpuset);
+	/* put back its first PU if it was there */
+	if (hadpu)
+	  hwloc_bitmap_set(set, firstpu);
+      }
+    }
+  }
+
   if (singlify)
     hwloc_bitmap_singlify(set);
 
@@ -267,6 +287,10 @@ int main(int argc, char *argv[])
 	usage(callname, stdout);
 	return EXIT_SUCCESS;
       }
+      if (!strcmp (argv[0], "--no-smt")) {
+	no_smt = 1;
+	goto next;
+      }
       if (!strcmp (argv[0], "--restrict")) {
 	hwloc_bitmap_t restrictset;
 	if (argc < 2) {

utils/hwloc/test-hwloc-calc.output

@@ -2,6 +2,7 @@
 0xffffffffffffffff
 
 0x0000000f
+0x11111111,0x0
 0x00000f0c
 0x0000800a
 

utils/hwloc/test-hwloc-calc.sh.in

@@ -3,7 +3,7 @@
 
 #
 # Copyright © 2009 CNRS
-# Copyright © 2009-2018 Inria.  All rights reserved.
+# Copyright © 2009-2019 Inria.  All rights reserved.
 # Copyright © 2009, 2011 Université Bordeaux
 # Copyright © 2014 Cisco Systems, Inc.  All rights reserved.
 # See COPYING in top-level directory.
@@ -40,6 +40,7 @@ set -e
   $calc --if synthetic --input "node:4 core:4 pu:4" all --taskset
   echo
   $calc --if synthetic --input "node:4 core:4 pu:4" 0xf
+  $calc --if synthetic --input "node:4 core:4 pu:4" --no-smt node:2-3
   $calc --if synthetic --input "node:4 core:4 pu:4" 0xf ~0x3 0xff0 '^0xf0'
   $calc --if synthetic --input "node:4 core:4 pu:4" core:0 pu:15 ~pu:0 '^pu:2'
   echo