Committing Intel(R) TBB 4.4 Update 4 source code

uxlfoundation · Sep 15, 2016 · 3d8b180 · 3d8b180
1 parent 97d449b
commit 3d8b180
Show file tree

Hide file tree

Showing 744 changed files with 22,692 additions and 31,465 deletions.
diff --git a/CHANGES b/CHANGES
@@ -2,6 +2,31 @@
 The list of most significant changes made over time in
 Intel(R) Threading Building Blocks (Intel(R) TBB).
 
+Intel TBB 4.4 Update 4
+TBB_INTERFACE_VERSION == 9004
+
+Changes (w.r.t. Intel TBB 4.4 Update 3):
+
+- Removed a few cases of excessive user data copying in the flow graph.
+- Improved robustness of concurrent_bounded_queue::abort() in case of
+    simultaneous push and pop operations.
+
+Preview Features:
+
+- Added tbb::flow::async_msg, a special message type to support
+    communications between the flow graph and external asynchronous
+    activities.
+- async_node modified to support use with C++03 compilers.
+
+Bugs fixed:
+
+- Fixed a bug in dynamic memory allocation replacement for Windows* OS.
+- Fixed excessive memory consumption on Linux* OS caused by enabling
+    zero-copy realloc.
+- Fixed performance regression on Intel(R) Xeon Phi(tm) coprocessor with
+    auto_partitioner.
+
+------------------------------------------------------------------------
 Intel TBB 4.4 Update 3
 TBB_INTERFACE_VERSION == 9003
 

diff --git a/README.md b/README.md
@@ -1,4 +1,4 @@
-# Intel(R) Threading Building Blocks 4.4 Update 3
+# Intel(R) Threading Building Blocks 4.4 Update 4
 
 Intel(R) Threading Building Blocks (Intel(R) TBB) lets you easily write parallel C++ programs that take
 full advantage of multicore performance, that are portable, composable and have future-proof scalability.
@@ -8,7 +8,7 @@ Here are the latest [Changes] (CHANGES) and [Release Notes]
 (doc/Release_Notes.txt) (contains system requirements and known issues).
 
 ## Licensing
-Intel(R) TBB 4.4 Update 3 is licensed under [GPLv2] (COPYING) with the runtime exception.
+Intel(R) TBB 4.4 Update 4 is licensed under [GPLv2] (COPYING) with the runtime exception.
 
 ## Documentation
 * Intel(R) TBB [tutorial] (https://software.intel.com/en-us/tbb-tutorial)

diff --git a/build/FreeBSD.clang.inc b/build/FreeBSD.clang.inc
@@ -50,12 +50,6 @@ ifeq (libc++,$(stdlib))
     LIB_LINK_FLAGS += -stdlib=libc++
 endif
 
-CPP11_FLAGS = -std=c++11 -D_TBB_CPP0X
-
-ifneq (00,$(lambdas)$(cpp0x))
-    CXX_ONLY_FLAGS += $(CPP11_FLAGS)
-endif
-
 TBB_ASM.OBJ=
 MALLOC_ASM.OBJ=
 

diff --git a/build/Makefile.test b/build/Makefile.test
@@ -78,15 +78,6 @@ $(SCHEDULER_DIRECTLY_INCLUDED): LINK_TBB.LIB =
 $(SCHEDULER_DIRECTLY_INCLUDED): LINK_FILES += $(SCHEDULER_DEPENDENCIES)
 $(SCHEDULER_DIRECTLY_INCLUDED): $(SCHEDULER_DEPENDENCIES)
 
-# Tests that use some features of C++11
-TEST_TBB_CPP11 = test_lambda.$(TEST_EXT) test_cache_aligned_allocator_STL.$(TEST_EXT)
-
-ifneq (0,$(cpp0x))
-# Made CPP11 tests use NOSTRICT flags because -strict-ansi combined with
-# -std=c++0x on ICC 13.0 results in a compile error when stdlib is included
-$(TEST_TBB_CPP11):  CPLUS_FLAGS := $(CPLUS_FLAGS_NOSTRICT)
-endif
-
 # test_tbb_header detects "multiple definition" linker error using the test that covers the whole library
 TWICE_LINKED_TESTS = test_tbb_header.$(TEST_EXT) \
                      test_concurrent_unordered_set.$(TEST_EXT)
@@ -134,6 +125,7 @@ TEST_TBB_PLAIN.EXE = test_assembly.$(TEST_EXT)   \
 	test_mutex_native_threads.$(TEST_EXT)        \
 	test_rwm_upgrade_downgrade.$(TEST_EXT)       \
 	test_cache_aligned_allocator.$(TEST_EXT)     \
+        test_cache_aligned_allocator_STL.$(TEST_EXT) \
 	test_parallel_for.$(TEST_EXT)                \
 	test_parallel_reduce.$(TEST_EXT)             \
 	test_parallel_sort.$(TEST_EXT)               \
@@ -143,6 +135,7 @@ TEST_TBB_PLAIN.EXE = test_assembly.$(TEST_EXT)   \
 	test_pipeline.$(TEST_EXT)                    \
 	test_pipeline_with_tbf.$(TEST_EXT)           \
 	test_parallel_pipeline.$(TEST_EXT)           \
+        test_lambda.$(TEST_EXT)                      \
 	test_task_scheduler_init.$(TEST_EXT)         \
 	test_task_scheduler_observer.$(TEST_EXT)     \
 	test_task.$(TEST_EXT)                        \
@@ -202,10 +195,9 @@ TEST_TBB_PLAIN.EXE = test_assembly.$(TEST_EXT)   \
 	test_flow_graph_whitebox.$(TEST_EXT)         \
 	test_composite_node.$(TEST_EXT)              \
 	test_async_node.$(TEST_EXT)                  \
+	test_async_msg.$(TEST_EXT)                   \
 	test_tbb_version.$(TEST_EXT)                 # insert new files right above
 
-TEST_TBB_PLAIN.EXE += $(TEST_TBB_CPP11)
-
 # skip mode_plugin for now
 skip_tests += test_model_plugin
 

diff --git a/build/android.gcc.inc b/build/android.gcc.inc
@@ -48,10 +48,6 @@ endif
 LINK_FLAGS = -Wl,-rpath-link=. -rdynamic 
 C_FLAGS = $(CPLUS_FLAGS)
 
-# Supported g++ versions support C++11
-# On Android/gcc 4.4.3, -std=c++0x causes ::int64_t and ::uint64_t to be undefined.
-CPP11_FLAGS = -std=c++11 $(DEFINE_KEY)_TBB_CPP0X
-
 ifeq ($(cfg), release)
     CPLUS_FLAGS = -O2
 endif
@@ -75,12 +71,6 @@ CPLUS_FLAGS    += --sysroot=$(SYSROOT)
 LIB_LINK_FLAGS += --sysroot=$(SYSROOT)
 LIBS           = -L$(CPLUS_LIB_PATH) -lgnustl_shared
 
-# This causes CPP11_FLAGS to be issued twice for test_lambda.cpp
-# TODO: Fix this in general for all platforms once the correct strategy is determined.
-ifneq (00,$(lambdas)$(cpp0x))
-    CXX_ONLY_FLAGS += $(CPP11_FLAGS)
-endif
-
 ifeq (ia32,$(arch))
     # TODO: Determine best setting of -march and add to CPLUS_FLAGS
     CPLUS_FLAGS += -m32 

diff --git a/build/android.icc.inc b/build/android.icc.inc
@@ -47,8 +47,6 @@ endif
 
 LINK_FLAGS = -Wl,-rpath-link=. -rdynamic 
 C_FLAGS = $(CPLUS_FLAGS)
-
-CPP11_FLAGS = -std=c++11 $(DEFINE_KEY)_TBB_CPP0X
 
 ifeq ($(cfg), release)
     CPLUS_FLAGS = -O2
@@ -74,12 +72,6 @@ LIB_LINK_FLAGS += --sysroot=$(SYSROOT)
 # the -static-intel flag is to remove the need to copy Intel-specific libs to the device.
 LIBS           = -L$(CPLUS_LIB_PATH) -lgnustl_shared -static-intel
 
-# This causes CPP11_FLAGS to be issued twice for test_lambda.cpp
-# TODO: Fix this in general for all platforms once the correct strategy is determined.
-ifneq (00,$(lambdas)$(cpp0x))
-    CXX_ONLY_FLAGS += $(CPP11_FLAGS)
-endif
-
 ifeq (ia32,$(arch))
     # TODO: Determine best setting of -march and add to CPLUS_FLAGS
     CPLUS_FLAGS += -m32 -march=pentium4 -falign-stack=maintain-16-byte

diff --git a/build/common.inc b/build/common.inc
@@ -43,6 +43,25 @@ ifeq ($(tbb_cpf),1)
   export CPF_SUFFIX ?=_preview
 endif
 
+ifdef cpp0x
+    $(warning "Warning: deprecated cpp0x=$(cpp0x) is used, stdver must be used instead. Building in stdver=c++0x mode.")
+    export stdver ?= c++0x
+    override cpp0x=
+endif
+
+ifneq (,$(stdver))
+    ifeq (,$(findstring ++, $(stdver)))
+        $(warning "Warning: unexpected stdver=$(stdver) is used.")
+    endif
+
+    CXX_STD_FLAGS = -std=$(stdver) -D_TBB_CPP0X
+endif
+
+# The requested option is added unconditionally.
+# If it is not supported, a compiler warning or error is expected.
+# Note that CXX_STD_FLAGS can be changed in <os>.<compiler>.inc.
+CXX_ONLY_FLAGS += $(CXX_STD_FLAGS)
+
 ifeq (,$(wildcard $(tbb_root)/build/$(tbb_os).inc))
   $(error "$(tbb_os)" is not supported. Add build/$(tbb_os).inc file with os-specific settings )
 endif
@@ -82,10 +101,6 @@ cross_cfg = $(if $(crosstest),$(call flip_cfg,$(1)),$(1))
 cfg?=release
 
 ifdef BUILDING_PHASE
- # No lambdas or other C++0x extensions by default for compilers that implement them as experimental features
- # TODO: it should become unnecessary when all relevant tests are "moved" to the TEST_TBB_CPP11 set
- lambdas ?= 0
- cpp0x ?= 0
 
  ifndef target
   target:=$(tbb_os)

diff --git a/build/common_rules.inc b/build/common_rules.inc
@@ -51,7 +51,6 @@ CPLUS_FLAGS += $(DEFINE_KEY)__TBB_CPF_BUILD=1
 endif
 LINK_FLAGS += $(LDFLAGS)
 LIB_LINK_FLAGS += $(LDFLAGS)
-CPLUS_FLAGS_NOSTRICT = $(subst -strict-ansi,-ansi,$(CPLUS_FLAGS))
 
 LIB_LINK_CMD ?= $(CPLUS) $(PIC_KEY)
 ifeq ($(origin LIB_OUTPUT_KEY), undefined)
@@ -112,7 +111,8 @@ LINK_FILES+=$(TEST_LIBS)
 	$(ASM) $(ASM_FLAGS) -o $@ $*.tmp
 
 # Rule for generating .E file if needed for visual inspection
-# Note that due to mapping for ICL all uses of PREPROC_ONLY should be immediately followed by a file name
+# Note that ICL treats an argument after PREPROC_ONLY as a file to open,
+# so all uses of PREPROC_ONLY should be immediately followed by a file name
 %.E: %.cpp
 	$(CPLUS) $(CPLUS_FLAGS) $(CXX_ONLY_FLAGS) $(INCLUDES) $(PREPROC_ONLY) $< >$@
 

diff --git a/build/index.html b/build/index.html
@@ -124,6 +124,10 @@ <H3>Software prerequisites:</H3>
     </ol>
     <DT><TT>make compiler=clang stdlib=libc++ <B>[</B>(above options or targets)<B>]</B></TT>
     <DD>Build and run as above, but use <TT>libc++</TT> as a standard c++ library for clang.
+    <DT><TT>make stdver=<B>{</B>c++11, c++14, ...<B>}</B> <B>[</B>(above options or targets)<B>]</B></TT>
+    <DD>Build and run as above, but additionally specify the version of the C++ standard or dialect to be used by
+      the compiler. The specified value of <TT>stdver</TT> will be used as a parameter to the appropriate
+      compiler option (such as <TT>-std</TT>); the behavior in case of unsupported value is compiler-specific.
     <DT><TT>make target_ui=win8ui [target_ui_mode=production] <B>[</B>(above options or targets)<B>]</B></TT>
     <DD>Build and run as above, but use API that is compliant with Windows Store* applications. 
     <TT>target_ui_mode=production</TT> is used to produce binaries that are compliant with Windows Store* application container. In later case they won't with Intel TBB unit tests but work only with Windows Store* applications. 
@@ -170,7 +174,7 @@ <H4>To port the Intel TBB source code:</H4>
 		and the scheduler yield function.  See <A HREF="../include/tbb/machine/mac_ppc.h">include/tbb/machine/mac_ppc.h</A>
 		for an example of a minimal implementation.
 	    <LI>More complex implementation examples can also be found in the
-		<A HREF="../include/tbb/machine>include/tbb/machine"</A> directory
+		<A HREF="../include/tbb/machine">include/tbb/machine</A> directory
 		that implement all the individual variants of synchronization operations that Intel TBB uses.
 		Such implementations are more verbose but may achieve better performance on a given architecture.
 	    <LI>In a given implementation, any synchronization operation that is not defined is implemented, by default,

diff --git a/build/linux.clang.inc b/build/linux.clang.inc
@@ -50,12 +50,6 @@ ifeq (libc++,$(stdlib))
     LIB_LINK_FLAGS += -stdlib=libc++
 endif
 
-CPP11_FLAGS = -std=c++11 -D_TBB_CPP0X
-
-ifneq (00,$(lambdas)$(cpp0x))
-    CXX_ONLY_FLAGS += $(CPP11_FLAGS)
-endif
-
 TBB_ASM.OBJ=
 MALLOC_ASM.OBJ=
 

diff --git a/build/linux.gcc.inc b/build/linux.gcc.inc
@@ -38,10 +38,6 @@ LIB_LINK_FLAGS = $(DYLIB_KEY) -Wl,-soname=$(BUILDING_LIBRARY)
 LIBS += -lpthread -lrt
 LINK_FLAGS = -Wl,-rpath-link=. -rdynamic
 C_FLAGS = $(CPLUS_FLAGS)
-# gcc 4.4 and higher support -std=c++0x
-ifneq (,$(shell gcc -dumpversion | egrep  "^(4\.[4-9]|[5-9])"))
-    CPP11_FLAGS = -std=c++0x -D_TBB_CPP0X
-endif
 
 # gcc 4.2 and higher support OpenMP
 ifneq (,$(shell gcc -dumpversion | egrep  "^(4\.[2-9]|[5-9])"))
@@ -60,10 +56,6 @@ ifeq ($(cfg), debug)
         CPLUS_FLAGS = -DTBB_USE_DEBUG $(ITT_NOTIFY) -g -O0 -DUSE_PTHREAD
 endif
 
-ifneq (00,$(lambdas)$(cpp0x))
-    CXX_ONLY_FLAGS += $(CPP11_FLAGS)
-endif
-
 TBB_ASM.OBJ=
 MALLOC_ASM.OBJ=
 

diff --git a/build/linux.icc.inc b/build/linux.icc.inc
@@ -44,10 +44,6 @@ LIB_LINK_FLAGS = -shared -static-intel -Wl,-soname=$(BUILDING_LIBRARY)
 LIBS += -lpthread -lrt
 LINK_FLAGS = -rdynamic
 C_FLAGS = $(CPLUS_FLAGS)
-# ICC 11.0 and higher support -std=c++0x
-ifneq (,$(shell icc -dumpversion | egrep  "^1[1-9]\."))
-    CPP11_FLAGS = -std=c++0x -D_TBB_CPP0X
-endif
 
 ifneq (,$(shell icc -dumpversion | egrep  "^1[6-9]\."))
 OPENMP_FLAG = -qopenmp
@@ -82,10 +78,6 @@ ifneq (,$(codecov))
     ITT_NOTIFY = -prof-genx
 endif
 
-ifneq (00,$(lambdas)$(cpp0x))
-	CXX_ONLY_FLAGS += $(CPP11_FLAGS)
-endif
-
 #------------------------------------------------------------------------------
 # Setting assembler data.
 #------------------------------------------------------------------------------

diff --git a/build/linux.pathcc.inc b/build/linux.pathcc.inc
@@ -39,7 +39,6 @@ LIBS += -lstl -lpthread -lrt
 LINK_FLAGS = -Wl,-rpath-link=. -rdynamic
 C_FLAGS = $(CPLUS_FLAGS)
 
-CPP11_FLAGS = -std=c++0x -D_TBB_CPP0X
 OPENMP_FLAG = -openmp
 
 ifeq ($(cfg), release)
@@ -49,10 +48,6 @@ ifeq ($(cfg), debug)
         CPLUS_FLAGS = -DTBB_USE_DEBUG $(ITT_NOTIFY) -g -O0 -DUSE_PTHREAD
 endif
 
-ifneq (00,$(lambdas)$(cpp0x))
-    CXX_ONLY_FLAGS += $(CPP11_FLAGS)
-endif
-
 TBB_ASM.OBJ=
 MALLOC_ASM.OBJ=
 

diff --git a/build/macos.clang.inc b/build/macos.clang.inc
@@ -56,12 +56,6 @@ ifneq (,$(stdlib))
     LIB_LINK_FLAGS += -stdlib=$(stdlib)
 endif
 
-CPP11_FLAGS = -std=c++11 -D_TBB_CPP0X
-
-ifneq (00,$(lambdas)$(cpp0x))
-    CXX_ONLY_FLAGS += $(CPP11_FLAGS)
-endif
-
 ifeq (intel64,$(arch))
     CPLUS_FLAGS += -m64 $(RTM_KEY)
     LINK_FLAGS += -m64

diff --git a/build/macos.icc.inc b/build/macos.icc.inc
@@ -35,10 +35,6 @@ LIBS = -lpthread
 LINK_FLAGS =
 LIB_LINK_FLAGS = -dynamiclib -static-intel -install_name @rpath/$(BUILDING_LIBRARY)
 C_FLAGS = $(CPLUS_FLAGS)
-# ICC 11.0 and higher support -std=c++0x
-ifneq (,$(shell icc -dumpversion | egrep  "^1[1-9]\."))
-    CPP11_FLAGS = -std=c++0x -D_TBB_CPP0X
-endif
 
 ifneq (,$(shell icc -dumpversion | egrep  "^1[6-9]\."))
 OPENMP_FLAG = -qopenmp
@@ -63,10 +59,6 @@ ifneq (,$(codecov))
     CPLUS_FLAGS += -prof-genx
 endif
 
-ifneq (00,$(lambdas)$(cpp0x))
-	CXX_ONLY_FLAGS += $(CPP11_FLAGS)
-endif
-
 # ICC 14.0 and higher support usage of libc++, clang standard library
 # Extending minimal deployment version of official binaries to unsupported 10.8
 # per customer requests. TODO: make a dedicated variable for this

diff --git a/build/mic.icc.inc b/build/mic.icc.inc
@@ -57,16 +57,11 @@ endif
 LIB_LINK_FLAGS = -shared -static-intel -Wl,-soname=$(BUILDING_LIBRARY)
 LIBS += -lpthread -lrt
 C_FLAGS = $(CPLUS_FLAGS)
-CPP11_FLAGS = -std=c++0x -D_TBB_CPP0X
 CILK_AVAILABLE = yes
 
 TBB_ASM.OBJ=
 MALLOC_ASM.OBJ=
 
-ifneq (00,$(lambdas)$(cpp0x))
-    CXX_ONLY_FLAGS += $(CPP11_FLAGS)
-endif
-
 CPLUS_FLAGS += -DHARNESS_INCOMPLETE_SOURCES=1 -D__TBB_MIC_NATIVE -DTBB_USE_EXCEPTIONS=0 -opt-streaming-stores never
 CPLUS += -mmic
 CONLY += -mmic