BOAST 2.0.2 → 2.1.0

data/lib/BOAST/Runtime/OpenCLRuntime.rb

@@ -139,7 +139,7 @@ def self.run(*args)
     @kernel.set_arg(i, params[i])
   }
   gws = opts[:global_work_size]
-  if not gws then
+  unless gws then
     raise ":global_work_size or :block_number are required to run OpenCL kernels!" unless opts[:block_number]
     gws = []
     opts[:block_number].each_index { |i|
@@ -148,7 +148,7 @@ def self.run(*args)
     }
   end
   lws = opts[:local_work_size]
-  if not lws then
+  unless lws then
     lws = opts[:block_size]
   end
   event1 = @queue.enqueue_NDrange_kernel(@kernel, gws, :local_work_size => lws)

data/lib/BOAST/Runtime/Probe.rb

@@ -32,52 +32,109 @@ EOF
       end
     end
 
-    def decl
+    def preamble
+      get_output.print <<EOF
+struct _boast_timer_struct {
+EOF
       if OS.mac? then
-        get_output.print "  uint64_t _mac_boast_start, _mac_boast_stop;\n"
-        get_output.print "  mach_timebase_info_data_t _mac_boast_timebase_info;\n"
+        get_output.print <<EOF
+  uint64_t start, stop;
+  mach_timebase_info_data_t timebase_info;
+EOF
       else
-        get_output.print "  struct timespec _boast_start, _boast_stop;\n"
+        get_output.print <<EOF
+  struct timespec start, stop;
+EOF
       end
-      BOAST::decl RESULT
-    end
+      push_env(:indent_level => 2) {
+        BOAST::decl RESULT
+      }
+      get_output.print <<EOF
+};
 
-    def configure
-    end
+static inline void _boast_timer_start(struct _boast_timer_struct * _boast_timer) {
+EOF
+      if OS.mac? then
+        get_output.print "  _boast_timer->start = mach_absolute_time();\n"
+      else
+        get_output.print "  clock_gettime(CLOCK_REALTIME, &_boast_timer->start);\n"
+      end
+      get_output.print <<EOF
+}
 
-    def start
+static inline void _boast_timer_stop(struct _boast_timer_struct * _boast_timer) {
+EOF
       if OS.mac? then
-        get_output.print "  _mac_boast_start = mach_absolute_time();\n"
+        get_output.print "  _boast_timer->stop = mach_absolute_time();\n"
       else
-        get_output.print "  clock_gettime(CLOCK_REALTIME, &_boast_start);\n"
+        get_output.print "  clock_gettime(CLOCK_REALTIME, &_boast_timer->stop);\n"
       end
-    end
+      get_output.print <<EOF
+}
 
-    def stop
+static inline void _boast_timer_compute(struct _boast_timer_struct * _boast_timer) {
+EOF
       if OS.mac? then
-        get_output.print "  _mac_boast_stop = mach_absolute_time();\n"
+        get_output.print "  mach_timebase_info(&_boast_timer->timebase_info);\n"
+        get_output.print "  _boast_timer->#{RESULT} = (_boast_timer->stop - _boast_timer->start) * _boast_timer->timebase_info.numer / _boast_timer->timebase_info.denom;\n"
       else
-        get_output.print "  clock_gettime(CLOCK_REALTIME, &_boast_stop);\n"
+        get_output.print "  _boast_timer->#{RESULT} = (int64_t)(_boast_timer->stop.tv_sec - _boast_timer->start.tv_sec) * 1000000000ll + _boast_timer->stop.tv_nsec - _boast_timer->start.tv_nsec;\n"
       end
-    end
+      get_output.print <<EOF
+}
 
-    def compute
+#ifdef RUBY
+static inline void _boast_timer_store(struct _boast_timer_struct * _boast_timer, VALUE _boast_stats) {
+EOF
+      get_output.puts "  rb_hash_aset(_boast_stats,ID2SYM(rb_intern(\"duration\")),rb_float_new((double)_boast_timer->#{RESULT}*(double)1e-9));"
       if OS.mac? then
-        get_output.print "  mach_timebase_info(&_mac_boast_timebase_info);\n"
-        get_output.print "  #{RESULT} = (_mac_boast_stop - _mac_boast_start) * _mac_boast_timebase_info.numer / _mac_boast_timebase_info.denom;\n"
+        get_output.puts "  rb_hash_aset(_boast_stats,ID2SYM(rb_intern(\"start\")),rb_int_new((int64_t)(_boast_timer->start * _boast_timer->timebase_info.numer / _boast_timer->timebase_info.denom)*1000000000ll));"
+        get_output.puts "  rb_hash_aset(_boast_stats,ID2SYM(rb_intern(\"end\")),rb_int_new((int64_t)(_boast_timer->stop * _boast_timer->timebase_info.numer / _boast_timer->timebase_info.denom)*1000000000ll));"
       else
-        get_output.print "  #{RESULT} = (int64_t)(_boast_stop.tv_sec - _boast_start.tv_sec) * 1000000000ll + _boast_stop.tv_nsec - _boast_start.tv_nsec;\n"
+        get_output.puts "  rb_hash_aset(_boast_stats,ID2SYM(rb_intern(\"start\")),rb_int_new((int64_t)_boast_timer->start.tv_sec * 1000000000ll+_boast_timer->start.tv_nsec));"
+        get_output.puts "  rb_hash_aset(_boast_stats,ID2SYM(rb_intern(\"end\")),rb_int_new((int64_t)_boast_timer->stop.tv_sec * 1000000000ll+_boast_timer->stop.tv_nsec));"
       end
+      get_output.print <<EOF
+}
+#endif
+
+EOF
+    end
+
+    def decl
+      get_output.print "  struct _boast_timer_struct _boast_timer;\n"
+    end
+
+    def configure
+    end
+
+    def start
+      get_output.puts "  _boast_timer_start(&_boast_params._boast_timer);"
+    end
+
+    def stop
+      get_output.puts "  _boast_timer_stop(&_boast_params._boast_timer);"
+    end
+
+    def compute
+      get_output.puts "  _boast_timer_compute(&_boast_params._boast_timer);"
     end
 
     def store
-      get_output.print "  rb_hash_aset(_boast_stats,ID2SYM(rb_intern(\"duration\")),rb_float_new((double)#{RESULT}*(double)1e-9));\n"
+      get_output.puts "  _boast_timer_store(&_boast_params._boast_timer, _boast_stats);"
     end
 
     def to_yaml
       get_output.print <<EOF
-  printf(":duration: %lf\\n", (double)#{RESULT}*(double)1e-9);
+  printf(":duration: %lf\\n", (double)_boast_params._boast_timer.#{RESULT}*(double)1e-9);
 EOF
+      if OS.mac? then
+        get_output.puts "  printf(\":start: %lld\\n\",(int64_t)(_boast_params._boast_timer.start * _boast_params._boast_timer.timebase_info.numer / _boast_params._boast_timer.timebase_info.denom)*1000000000ll);"
+        get_output.puts "  printf(\":end: %lld\\n\",(int64_t)(_boast_params._boast_timer.stop * _boast_params._boast_timer.timebase_info.numer / _boast_params._boast_timer.timebase_info.denom)*1000000000ll);"
+      else
+        get_output.puts "  printf(\":start: %lld\\n\",(int64_t)_boast_params._boast_timer.start.tv_sec * 1000000000ll+_boast_params._boast_timer.start.tv_nsec);"
+        get_output.puts "  printf(\":end: %lld\\n\",(int64_t)_boast_params._boast_timer.stop.tv_sec * 1000000000ll+_boast_params._boast_timer.stop.tv_nsec);"
+      end
     end
 
   end
@@ -95,41 +152,72 @@ EOF
     def header
     end
 
-    def decl
-      get_output.print "  VALUE _boast_event_set = Qnil;\n"
-      get_output.print "  VALUE _boast_papi_results = Qnil;\n"
-    end
-
-    def configure
+    def preamble
       get_output.print <<EOF
+struct _boast_papi_struct {
+  VALUE event_set;
+  VALUE results;
+};
+
+static void _boast_get_papi_envent_set( VALUE _boast_rb_opts, struct _boast_papi_struct *_boast_papi );
+static void _boast_get_papi_envent_set( VALUE _boast_rb_opts, struct _boast_papi_struct *_boast_papi ) {
+  VALUE _boast_event_set = Qnil;
   if( _boast_rb_opts != Qnil ) {
      VALUE _boast_PAPI_rb_ptr = Qnil;
     _boast_PAPI_rb_ptr = rb_hash_aref(_boast_rb_opts, ID2SYM(rb_intern("PAPI")));
     if( _boast_PAPI_rb_ptr != Qnil ) {
       VALUE _boast_PAPI = Qnil;
       VALUE _boast_EventSet = Qnil;
-      rb_eval_string("require 'PAPI'");
       _boast_PAPI =  rb_const_get(rb_cObject, rb_intern("PAPI"));
       _boast_EventSet =  rb_const_get(_boast_PAPI, rb_intern("EventSet"));
       _boast_event_set = rb_funcall(_boast_EventSet, rb_intern("new"), 0);
       rb_funcall(_boast_event_set, rb_intern("add_named"), 1, _boast_PAPI_rb_ptr);
     }
   }
+  _boast_papi->event_set = _boast_event_set;
+}
+
+static void _boast_store_papi_results( struct _boast_papi_struct *_boast_papi, VALUE _boast_rb_opts, VALUE _boast_stats );
+static void _boast_store_papi_results( struct _boast_papi_struct *_boast_papi, VALUE _boast_rb_opts, VALUE _boast_stats ) {
+  if( _boast_papi->results != Qnil) {
+    VALUE _boast_papi_stats = Qnil;
+    _boast_papi_stats = rb_ary_new3(1,rb_hash_aref(_boast_rb_opts, ID2SYM(rb_intern("PAPI"))));
+    _boast_papi_stats = rb_funcall(_boast_papi_stats, rb_intern("flatten"), 0);
+    _boast_papi_stats = rb_funcall(_boast_papi_stats, rb_intern("zip"), 1, _boast_papi->results);
+    _boast_papi->results = rb_funcall(rb_const_get(rb_cObject, rb_intern("Hash")), rb_intern("send"), 2, ID2SYM(rb_intern("[]")), _boast_papi_stats );
+    rb_hash_aset(_boast_stats, ID2SYM(rb_intern(\"PAPI\")), _boast_papi->results);
+    rb_funcall(_boast_papi->event_set, rb_intern("cleanup"), 0);
+    rb_funcall(_boast_papi->event_set, rb_intern("destroy"), 0);
+  }
+}
+
+EOF
+    end
+
+    def decl
+      get_output.print <<EOF
+  struct _boast_papi_struct _boast_papi = { Qnil, Qnil };
+EOF
+    end
+
+    def configure
+      get_output.print <<EOF
+  _boast_get_papi_envent_set( _boast_rb_opts, &_boast_papi );
 EOF
     end
 
     def start
       get_output.print <<EOF
-  if( _boast_event_set != Qnil) {
-    rb_funcall(_boast_event_set, rb_intern("start"), 0);
+  if( _boast_papi.event_set != Qnil) {
+    rb_funcall(_boast_papi.event_set, rb_intern("start"), 0);
   }
 EOF
     end
 
     def stop
       get_output.print <<EOF
-  if( _boast_event_set != Qnil) {
-    _boast_papi_results = rb_funcall(_boast_event_set, rb_intern("stop"), 0);
+  if( _boast_papi.event_set != Qnil) {
+    _boast_papi.results = rb_funcall(_boast_papi.event_set, rb_intern("stop"), 0);
   }
 EOF
     end
@@ -139,14 +227,7 @@ EOF
 
     def store
       get_output.print <<EOF
-  if( _boast_papi_results != Qnil) {
-    VALUE _boast_papi_stats = Qnil;
-    _boast_papi_stats = rb_ary_new3(1,rb_hash_aref(_boast_rb_opts, ID2SYM(rb_intern("PAPI"))));
-    _boast_papi_stats = rb_funcall(_boast_papi_stats, rb_intern("flatten"), 0);
-    _boast_papi_stats = rb_funcall(_boast_papi_stats, rb_intern("zip"), 1, _boast_papi_results);
-    _boast_papi_results = rb_funcall(rb_const_get(rb_cObject, rb_intern("Hash")), rb_intern("send"), 2, ID2SYM(rb_intern("[]")), _boast_papi_stats );
-    rb_hash_aset(_boast_stats,ID2SYM(rb_intern(\"PAPI\")),_boast_papi_results);
-  }
+  _boast_store_papi_results( &_boast_papi, _boast_rb_opts, _boast_stats );
 EOF
     end
 

metadata

@@ -1,14 +1,14 @@
 --- !ruby/object:Gem::Specification
 name: BOAST
 version: !ruby/object:Gem::Version
-  version: 2.0.2
+  version: 2.1.0
 platform: ruby
 authors:
 - Brice Videau
 autorequire: 
 bindir: bin
 cert_chain: []
-date: 2017-02-18 00:00:00.000000000 Z
+date: 2017-06-19 00:00:00.000000000 Z
 dependencies:
 - !ruby/object:Gem::Dependency
   name: narray
@@ -116,20 +116,40 @@ dependencies:
     requirements:
     - - "~>"
       - !ruby/object:Gem::Version
-        version: '0'
+        version: '1.0'
     - - ">="
       - !ruby/object:Gem::Version
-        version: '0.101'
+        version: 1.0.0
   type: :runtime
   prerelease: false
   version_requirements: !ruby/object:Gem::Requirement
     requirements:
     - - "~>"
       - !ruby/object:Gem::Version
-        version: '0'
+        version: '1.0'
     - - ">="
       - !ruby/object:Gem::Version
-        version: '0.101'
+        version: 1.0.0
+- !ruby/object:Gem::Dependency
+  name: hwloc
+  requirement: !ruby/object:Gem::Requirement
+    requirements:
+    - - "~>"
+      - !ruby/object:Gem::Version
+        version: '0.3'
+    - - ">="
+      - !ruby/object:Gem::Version
+        version: 0.3.0
+  type: :runtime
+  prerelease: false
+  version_requirements: !ruby/object:Gem::Requirement
+    requirements:
+    - - "~>"
+      - !ruby/object:Gem::Version
+        version: '0.3'
+    - - ">="
+      - !ruby/object:Gem::Version
+        version: 0.3.0
 - !ruby/object:Gem::Dependency
   name: ffi
   requirement: !ruby/object:Gem::Requirement
@@ -234,6 +254,7 @@ files:
 - lib/BOAST/Runtime/CKernel.rb
 - lib/BOAST/Runtime/CRuntime.rb
 - lib/BOAST/Runtime/CUDARuntime.rb
+- lib/BOAST/Runtime/CoExecute.rb
 - lib/BOAST/Runtime/CompiledRuntime.rb
 - lib/BOAST/Runtime/Compilers.rb
 - lib/BOAST/Runtime/Config.rb
@@ -260,7 +281,7 @@ required_ruby_version: !ruby/object:Gem::Requirement
   requirements:
   - - ">="
     - !ruby/object:Gem::Version
-      version: 1.9.3
+      version: 2.0.0
 required_rubygems_version: !ruby/object:Gem::Requirement
   requirements:
   - - ">="
@@ -268,7 +289,7 @@ required_rubygems_version: !ruby/object:Gem::Requirement
       version: '0'
 requirements: []
 rubyforge_project: 
-rubygems_version: 2.5.1
+rubygems_version: 2.5.2
 signing_key: 
 specification_version: 4
 summary: BOAST is a computing kernel metaprogramming tool.