Skip to content

Commit

Permalink
ENH: Update to TBB 2017 (2016-09-08 release)
Browse files Browse the repository at this point in the history 
The new release TBB is now under a new more
open license.
                                 Apache License
                           Version 2.0, January 2004
                        http://www.apache.org/licenses/

The list of most significant changes made over time in
Intel(R) Threading Building Blocks (Intel(R) TBB).

Intel TBB 2017
TBB_INTERFACE_VERSION == 9100

Changes (w.r.t. Intel TBB 4.4 Update 5):

- static_partitioner class is now a fully supported feature.
- async_node class is now a fully supported feature.
- Improved dynamic memory allocation replacement on Windows* OS to skip
    DLLs for which replacement cannot be done, instead of aborting.
- Intel TBB no longer performs dynamic memory allocation replacement
    for Microsoft* Visual Studio* 2008.
- For 64-bit platforms, quadrupled the worst-case limit on the amount
    of memory the Intel TBB allocator can handle.
- Added TBB_USE_GLIBCXX_VERSION macro to specify the version of GNU
    libstdc++ when it cannot be properly recognized, e.g. when used
    with Clang on Linux* OS. Inspired by a contribution from David A.
- Added graph/stereo example to demostrate tbb::flow::async_msg.
- Removed a few cases of excessive user data copying in the flow graph.
- Reworked split_node to eliminate unnecessary overheads.
- Added support for C++11 move semantics to the argument of
    tbb::parallel_do_feeder::add() method.
- Added C++11 move constructor and assignment operator to
    tbb::combinable template class.
- Added tbb::this_task_arena::max_concurrency() function and
    max_concurrency() method of class task_arena returning the maximal
    number of threads that can work inside an arena.
- Deprecated tbb::task_arena::current_thread_index() static method;
    use tbb::this_task_arena::current_thread_index() function instead.
- All examples for commercial version of library moved online:
    https://software.intel.com/en-us/product-code-samples. Examples are
    available as a standalone package or as a part of Intel(R) Parallel
    Studio XE or Intel(R) System Studio Online Samples packages.

Changes affecting backward compatibility:

- Renamed following methods and types in async_node class:
    Old                   New
    async_gateway_type => gateway_type
    async_gateway()    => gateway()
    async_try_put()    => try_put()
    async_reserve()    => reserve_wait()
    async_commit()     => release_wait()
- Internal layout of some flow graph nodes has changed; recompilation
    is recommended for all binaries that use the flow graph.

Preview Features:

- Added template class streaming_node to the flow graph API. It allows
    a flow graph to offload computations to other devices through
    streaming or offloading APIs.
- Template class opencl_node reimplemented as a specialization of
    streaming_node that works with OpenCL*.
- Added tbb::this_task_arena::isolate() function to isolate execution
    of a group of tasks or an algorithm from other tasks submitted
    to the scheduler.

Bugs fixed:

- Added a workaround for GCC bug #62258 in std::rethrow_exception()
    to prevent possible problems in case of exception propagation.
- Fixed parallel_scan to provide correct result if the initial value
    of an accumulator is not the operation identity value.
- Fixed a memory corruption in the memory allocator when it meets
    internal limits.
- Fixed the memory allocator on 64-bit platforms to align memory
    to 16 bytes by default for all allocations bigger than 8 bytes.
- As a workaround for crashes in the Intel TBB library compiled with
    GCC 6, added -flifetime-dse=1 to compilation options on Linux* OS.
- Fixed a race in the flow graph implementation.

Open-source contributions integrated:

- Enabling use of C++11 'override' keyword by Raf Schietekat.

------------------------------------------------------------------------
  • Loading branch information
@hjmjohnson
hjmjohnson committed Sep 11, 2016
1 parent f9befdd commit c0ee940
Show file tree
Hide file tree
Showing 2,177 changed files with 101,795 additions and 53,212 deletions.
105 changes: 90 additions & 15 deletions CHANGES
View file
Original file line number Diff line number Diff line change
Expand Up @@ -2,6 +2,81 @@
The list of most significant changes made over time in
Intel(R) Threading Building Blocks (Intel(R) TBB).

Intel TBB 2017
TBB_INTERFACE_VERSION == 9100

Changes (w.r.t. Intel TBB 4.4 Update 5):

- static_partitioner class is now a fully supported feature.
- async_node class is now a fully supported feature.
- Improved dynamic memory allocation replacement on Windows* OS to skip
DLLs for which replacement cannot be done, instead of aborting.
- Intel TBB no longer performs dynamic memory allocation replacement
for Microsoft* Visual Studio* 2008.
- For 64-bit platforms, quadrupled the worst-case limit on the amount
of memory the Intel TBB allocator can handle.
- Added TBB_USE_GLIBCXX_VERSION macro to specify the version of GNU
libstdc++ when it cannot be properly recognized, e.g. when used
with Clang on Linux* OS. Inspired by a contribution from David A.
- Added graph/stereo example to demostrate tbb::flow::async_msg.
- Removed a few cases of excessive user data copying in the flow graph.
- Reworked split_node to eliminate unnecessary overheads.
- Added support for C++11 move semantics to the argument of
tbb::parallel_do_feeder::add() method.
- Added C++11 move constructor and assignment operator to
tbb::combinable template class.
- Added tbb::this_task_arena::max_concurrency() function and
max_concurrency() method of class task_arena returning the maximal
number of threads that can work inside an arena.
- Deprecated tbb::task_arena::current_thread_index() static method;
use tbb::this_task_arena::current_thread_index() function instead.
- All examples for commercial version of library moved online:
https://software.intel.com/en-us/product-code-samples. Examples are
available as a standalone package or as a part of Intel(R) Parallel
Studio XE or Intel(R) System Studio Online Samples packages.

Changes affecting backward compatibility:

- Renamed following methods and types in async_node class:
Old New
async_gateway_type => gateway_type
async_gateway() => gateway()
async_try_put() => try_put()
async_reserve() => reserve_wait()
async_commit() => release_wait()
- Internal layout of some flow graph nodes has changed; recompilation
is recommended for all binaries that use the flow graph.

Preview Features:

- Added template class streaming_node to the flow graph API. It allows
a flow graph to offload computations to other devices through
streaming or offloading APIs.
- Template class opencl_node reimplemented as a specialization of
streaming_node that works with OpenCL*.
- Added tbb::this_task_arena::isolate() function to isolate execution
of a group of tasks or an algorithm from other tasks submitted
to the scheduler.

Bugs fixed:

- Added a workaround for GCC bug #62258 in std::rethrow_exception()
to prevent possible problems in case of exception propagation.
- Fixed parallel_scan to provide correct result if the initial value
of an accumulator is not the operation identity value.
- Fixed a memory corruption in the memory allocator when it meets
internal limits.
- Fixed the memory allocator on 64-bit platforms to align memory
to 16 bytes by default for all allocations bigger than 8 bytes.
- As a workaround for crashes in the Intel TBB library compiled with
GCC 6, added -flifetime-dse=1 to compilation options on Linux* OS.
- Fixed a race in the flow graph implementation.

Open-source contributions integrated:

- Enabling use of C++11 'override' keyword by Raf Schietekat.

------------------------------------------------------------------------
Intel TBB 4.4 Update 5
TBB_INTERFACE_VERSION == 9005

Expand All @@ -11,7 +86,7 @@ Changes (w.r.t. Intel TBB 4.4 Update 4):

Preview Features:

- Added a Python* module which is able to replace Python's thread pool
- Added a Python* module which is able to replace Python's thread pool
class with the implementation based on Intel TBB task scheduler.

Bugs fixed:
Expand Down Expand Up @@ -520,7 +595,7 @@ Bugs fixed:

- Fixed data races in preview extensions of task_scheduler_observer.
- Added noexcept(false) for destructor of task_group_base to avoid
crash on cancelation of structured task group in C++11.
crash on cancellation of structured task group in C++11.

Open-source contributions integrated:

Expand All @@ -536,7 +611,7 @@ TBB_INTERFACE_VERSION == 7003
Changes (w.r.t. Intel TBB 4.2 Update 2):

- Added support for Microsoft* Visual Studio* 2013.
- Improved Microsoft* PPL-compatible form of parallel_for for better
- Improved Microsoft* PPL-compatible form of parallel_for for better
support of auto-vectorization.
- Added a new example for cancellation and reset in the flow graph:
Kohonen self-organizing map (examples/graph/som).
Expand Down Expand Up @@ -602,7 +677,7 @@ Preview Features:
- Class task_arena no longer requires linking with a preview library,
though still remains a community preview feature.
- The method task_arena::wait_until_empty() is removed.
- The method task_arena::current_slot() now returns -1 if
- The method task_arena::current_slot() now returns -1 if
the task scheduler is not initialized in the thread.

Changes affecting backward compatibility:
Expand Down Expand Up @@ -783,7 +858,7 @@ Changes (w.r.t. Intel TBB 4.1):
any platform supported by compiler version 12.1 and newer.
- Using GetNativeSystemInfo() instead of GetSystemInfo() to support
more than 32 processors for 32-bit applications under WOW64.
- The following form of parallel_for:
- The following form of parallel_for:
parallel_for(first, last, [step,] f[, context]) now accepts an
optional partitioner parameter after the function f.

Expand Down Expand Up @@ -854,7 +929,7 @@ Changes (w.r.t. Intel TBB 4.0 Update 4):

Bugs fixed:

- Fixed a tv_nsec overflow bug in condition_variable::wait_for.
- Fixed a tv_nsec overflow bug in condition_variable::wait_for.
- Fixed execution order of enqueued tasks with different priorities.
- Fixed a bug with task priority changes causing lack of progress
for fire-and-forget tasks when TBB was initialized to use 1 thread.
Expand Down Expand Up @@ -923,7 +998,7 @@ Changes (w.r.t. Intel TBB 4.0 Update 2):
Backward-incompatible API changes:

- a graph reference parameter is now required to be passed to the
constructors of the following flow graph nodes: overwrite_node,
constructors of the following flow graph nodes: overwrite_node,
write_once_node, broadcast_node, and the CPF or_node.
- the following tbb::flow node methods and typedefs have been renamed:
Old New
Expand All @@ -944,10 +1019,10 @@ TBB_INTERFACE_VERSION == 6002

Changes (w.r.t. Intel TBB 4.0 Update 1 commercial-aligned release):

- concurrent_bounded_queue now has an abort() operation that releases
threads involved in pending push or pop operations. The released
- concurrent_bounded_queue now has an abort() operation that releases
threads involved in pending push or pop operations. The released
threads will receive a tbb::user_abort exception.
- Added Community Preview Feature: concurrent_lru_cache container,
- Added Community Preview Feature: concurrent_lru_cache container,
a concurrent implementation of LRU (least-recently-used) cache.

Bugs fixed:
Expand Down Expand Up @@ -1038,8 +1113,8 @@ TBB_INTERFACE_VERSION == 5006 (forgotten to increment)

Changes (w.r.t. Intel TBB 3.0 Update 6 commercial-aligned release):

- Added implementation of the platform isolation layer based on
GCC atomic built-ins; it is supposed to work on any platform
- Added implementation of the platform isolation layer based on
GCC atomic built-ins; it is supposed to work on any platform
where GCC has these built-ins.

Community Preview Features:
Expand Down Expand Up @@ -1116,9 +1191,9 @@ Changes (w.r.t. Intel TBB 3.0 Update 3 commercial-aligned release):
- Fixed library loading to avoid possibility for remote code execution,
see http://www.microsoft.com/technet/security/advisory/2269637.mspx.
- Added support of more than 64 cores for appropriate Microsoft*
Windows* versions. For more details, see
Windows* versions. For more details, see
http://msdn.microsoft.com/en-us/library/dd405503.aspx.
- Default number of worker threads is adjusted in accordance with
- Default number of worker threads is adjusted in accordance with
process affinity mask.

Bugs fixed:
Expand Down Expand Up @@ -1203,7 +1278,7 @@ Bugs fixed:
was a temporary object.
- Incorrect usage of memory fences on PowerPC and XBOX360 platforms.
- A subtle issue in task group context binding that could result
in cancelation signal being missed by nested task groups.
in cancellation signal being missed by nested task groups.
- Incorrect construction of concurrent_unordered_map if specified
number of buckets is not power of two.
- Broken count() and equal_range() of concurrent_unordered_map.
Expand Down
Loading

0 comments on commit c0ee940

Please sign in to comment.