Merge branch 'control_replication' into shardrefine

StanfordLegion · Oct 5, 2023 · 522ef64 · 522ef64
2 parents 8247c2a + 71d152d
commit 522ef64
Show file tree

Hide file tree

Showing 28 changed files with 698 additions and 206 deletions.
diff --git a/CHANGES.txt b/CHANGES.txt
@@ -6,11 +6,14 @@ Version 23.09.0 (September ??, 2023)
     - Elide future maps in index launches
     - Improvements to Pygion interop
   * Realm
-    - Machine configuration API that allows applications to configure the machine model without using the command line
-    - Expose Realm managed CUDA/HIP stream to applications to launch GPU tasks without device-wise synchronization when hijack is disabled
-    - Update the timer to use of rdtsc for different processors
-    - Performance improvement for getting highest priority task available from any task queue
-    - Implementation of framebuffer memory with cuMemMap
+    - Add a machine configuration API that allows applications to configure
+       the machine model without using the command line
+    - Expose Realm managed CUDA/HIP stream to applications to launch GPU tasks
+       without device-wise synchronization when hijack is disabled
+    - Change timers to use rdtsc
+    - Improve performance for getting highest priority task available in any
+       task queue
+    - Implement framebuffer memory with `cuMemMap`
     - Initial work for moving STL dependencies to header only
 
 Version 23.06.0 (June 28, 2023)

diff --git a/CMakeLists.txt b/CMakeLists.txt
@@ -743,6 +743,17 @@ if(NOT WIN32)
   option(REALM_LIMIT_SYMBOL_VISIBILITY "hide private symbols in realm" ON)
 endif()
 
+# Only enable shm on linux and if GASNET1 is not available.
+# OSX is not well tested yet, so disable by default for the time being
+# GASNET1 does not allow communication before network attach, which is needed
+# in order to coordinate a job identifier required to set up named system
+# resources
+if (UNIX AND NOT APPLE AND NOT REALM_USE_GASNET1)
+  option(REALM_USE_SHM "Enable shared memory usage" ON)  
+else()
+  option(REALM_USE_SHM "Enable shared memory usage" OFF)
+endif()
+
 if (UNIX)
   # Look for memfd_create to see if we can use anonymous memory sharing
   check_function_exists(memfd_create REALM_HAS_MEMFD)

diff --git a/runtime/legion/legion_context.cc b/runtime/legion/legion_context.cc
@@ -18367,15 +18367,6 @@ namespace Legion {
                                                    bool from_application)
     //--------------------------------------------------------------------------
     {
-      for (int i = 0; runtime->safe_control_replication && from_application &&
-        (i < 2) && ((current_trace == NULL) || !current_trace->is_fixed()); i++)
-      {
-        Murmur3Hasher hasher(this, runtime->safe_control_replication > 1,i > 0);
-        hasher.hash(REPLICATE_DESTROY_FIELD_ALLOCATOR, __func__);
-        hasher.hash(node->handle, "handle");
-        if (hasher.verify(__func__))
-          break;
-      }
       if (from_application)
       {
         AutoRuntimeCall call(this);

diff --git a/runtime/legion/legion_context.h b/runtime/legion/legion_context.h
@@ -2353,7 +2353,6 @@ namespace Legion {
         REPLICATE_DESTROY_LOGICAL_REGION,
         REPLICATE_RESET_EQUIVALENCE_SETS,
         REPLICATE_CREATE_FIELD_ALLOCATOR,
-        REPLICATE_DESTROY_FIELD_ALLOCATOR,
         REPLICATE_EXECUTE_TASK,
         REPLICATE_EXECUTE_INDEX_SPACE,
         REPLICATE_REDUCE_FUTURE_MAP,

diff --git a/runtime/legion/legion_profiling.cc b/runtime/legion/legion_profiling.cc
@@ -1736,6 +1736,11 @@ namespace Legion {
 
       serializer->serialize(machine_desc);
 
+      LegionProfDesc::ZeroTime zero_time;
+      zero_time.zero_time = Legion::Runtime::get_zero_time();
+
+      serializer->serialize(zero_time);
+
       for (unsigned idx = 0; idx < num_meta_tasks; idx++)
       {
         LegionProfDesc::MetaDesc meta_desc;

diff --git a/runtime/legion/legion_profiling.h b/runtime/legion/legion_profiling.h
@@ -127,6 +127,11 @@ namespace Legion {
         unsigned node_id;
 	unsigned num_nodes;
       };
+      struct ZeroTime {
+      public:
+        long long zero_time;
+      };
+
     };
 
     class LegionProfInstance {

diff --git a/runtime/legion/legion_profiling_serializer.cc b/runtime/legion/legion_profiling_serializer.cc
@@ -91,6 +91,11 @@ namespace Legion {
          << "num_nodes:unsigned:" << sizeof(unsigned)
          << "}" << std::endl;
 
+      ss << "ZeroTime {"
+         << "id:" << ZERO_TIME_ID  << delim
+         << "zero_time:long long:" << sizeof(long long)
+         << "}" << std::endl;
+
       ss << "MemDesc {" 
          << "id:" << MEM_DESC_ID                               << delim
          << "mem_id:MemID:"                << sizeof(MemID)    << delim
@@ -510,6 +515,18 @@ namespace Legion {
 		sizeof(machine_desc.num_nodes));
     }
 
+    //--------------------------------------------------------------------------
+    void LegionProfBinarySerializer::serialize(
+                                      const LegionProfDesc::ZeroTime
+				      &zero_time)
+    //--------------------------------------------------------------------------
+    {
+      int ID = ZERO_TIME_ID;
+      lp_fwrite(f, (char*)&ID, sizeof(ID));
+      lp_fwrite(f, (char*)&(zero_time.zero_time),
+		sizeof(zero_time.zero_time));
+    }
+
     //--------------------------------------------------------------------------
     void LegionProfBinarySerializer::serialize(
                const LegionProfInstance::IndexSpacePointDesc &ispace_point_desc)
@@ -1593,6 +1610,15 @@ namespace Legion {
                      machine_desc.node_id, machine_desc.num_nodes);
     }
 
+    //--------------------------------------------------------------------------
+    void LegionProfASCIISerializer::serialize(
+                                      const LegionProfDesc::ZeroTime
+				      &zero_time)
+    //--------------------------------------------------------------------------
+    {
+      log_prof.print("Zero Time %lld", zero_time.zero_time);
+    }
+
     //--------------------------------------------------------------------------
     void LegionProfASCIISerializer::serialize(
                                   const LegionProfInstance::TaskKind &task_kind)

diff --git a/runtime/legion/legion_profiling_serializer.h b/runtime/legion/legion_profiling_serializer.h
@@ -49,6 +49,7 @@ namespace Legion {
       virtual void serialize(const LegionProfDesc::OpDesc&) = 0;
       virtual void serialize(const LegionProfDesc::MaxDimDesc&) = 0;
       virtual void serialize(const LegionProfDesc::MachineDesc&) = 0;
+      virtual void serialize(const LegionProfDesc::ZeroTime&) = 0;
       virtual void serialize(const LegionProfInstance::IndexSpacePointDesc&) = 0;
       virtual void serialize(const LegionProfInstance::IndexSpaceRectDesc&) = 0;
       virtual void serialize(const LegionProfInstance::IndexSpaceEmptyDesc&) = 0;
@@ -117,6 +118,7 @@ namespace Legion {
       void serialize(const LegionProfDesc::OpDesc&);
       void serialize(const LegionProfDesc::MaxDimDesc&);
       void serialize(const LegionProfDesc::MachineDesc&);
+      void serialize(const LegionProfDesc::ZeroTime&);
       void serialize(const LegionProfInstance::IndexSpacePointDesc&);
       void serialize(const LegionProfInstance::IndexSpaceRectDesc&);
       void serialize(const LegionProfInstance::IndexSpaceEmptyDesc&);
@@ -215,8 +217,9 @@ namespace Legion {
         COPY_INST_INFO_ID,
         FILL_INST_INFO_ID,
 #ifdef LEGION_PROF_SELF_PROFILE
-        PROFTASK_INFO_ID
+        PROFTASK_INFO_ID,
 #endif
+        ZERO_TIME_ID,
       };
     };
 
@@ -234,6 +237,7 @@ namespace Legion {
       void serialize(const LegionProfDesc::OpDesc&);
       void serialize(const LegionProfDesc::MaxDimDesc&);
       void serialize(const LegionProfDesc::MachineDesc&);
+      void serialize(const LegionProfDesc::ZeroTime&);
       void serialize(const LegionProfInstance::IndexSpacePointDesc&);
       void serialize(const LegionProfInstance::IndexSpaceRectDesc&);
       void serialize(const LegionProfInstance::IndexSpaceEmptyDesc&);