8.2.0 is a small release, mostly of small bugfixes and improvements.
Changes:
`len(AsyncMapResult)` and progress ports now use the number of items in the map,
not the number of messages.
Enhancements:
- Show output prior to errors in `%%px`
Bugs fixed:
- Fix cases where engine id could be `-1` in tracebacks
- Add missing `pbs` to engine launcher entrypoints
IPython Parallel (ipyparallel) is a Python package and collection of CLI
scripts for controlling clusters of IPython processes, built on the Jupyter
protocol.
Tested on NetBSD-current and 9.1/amd64.
Also update packages tightly coupled to llvm, if new versions are
available.
Mark creduce, include-what-you-use, and zig broken (waiting for a
new release).
Changes:
Non-comprehensive list of changes in this release
Windows Control-flow Enforcement Technology: the -ehcontguard
option now emits valid unwind entrypoints which are validated
when the context is being set during exception handling.
Flang is now included in the binary packages released by LLVM.
The debuginfo-test project has been renamed cross-project-tests
and is now intended for testing components from multiple
projects, not just debug information. The new “cross-project-tests”
name replaces “debuginfo-test” in LLVM_ENABLE_PROJECTS, and a
new check-cross-project-tests target has been added for running
all tests in the project. The pre-existing check-debuginfo-
test target remains for running just the debug information
tests. (D95339 and D96513)
Changes to the LLVM IR
The inalloca attribute now has a mandatory type field, similar
to byval and sret.
The opaque pointer type ptr has been introduced. It is still
in the process of being worked on and should not be used yet.
Using the legacy pass manager for the optimization pipeline is
deprecated and will be removed after LLVM 14. In the meantime,
only minimal effort will be made to maintain the legacy pass
manager for the optimization pipeline.
Changes to building LLVM
The build system now supports building multiple distributions,
so that you can e.g. have one distribution containing just
tools and another for libraries (to enable development). See
Multi-distribution configurations for details.
Changes to the AArch64 Backend
Introduced assembly support for Armv9-A’s Realm Management
Extension (RME) and Scalable Matrix Extension (SME).
Produce proper cross-section relative relocations on COFF
Fixed the calling convention on Windows for variadic functions
involving floats in the fixed arguments
Changes to the ARM Backend
Produce proper cross-section relative relocations on COFF
Changes to the Hexagon Target
The Hexagon target now supports V68/HVX ISA.
Changes to the C API
The C API functions LLVMGetAlignment and LLVMSetAlignment now
allow changing alignment on atomicrmw and cmpxchg instructions
A new entry LLVMDIArgListMetadataKind was added to the
LLVMMetadataKind enum, representing DIArgLists (D88175)
Type attributes have been added to LLVM-C, introducing
LLVMCreateTypeAttribute, LLVMGetTypeAttributeValue and
LLVMIsTypeAttribute. (D977763’)
The LTO_API_VERSION was bumped to 28, introducing a new function
lto_set_debug_options for parsing LTO debug options (D92611)
LLVMJITTargetSymbolFlags was renamed to LLVMJITSymbolTargetFlags
(rG8d718a0bff73af066675a6258c01307937c33cf9)
The C API received support for creating custom ORCv2
MaterializationUnits and APIs to retrieve an LLJIT instance’s
linking layers. An ABI breaking change for LLVMOrcAbsoluteSymbols
was introduced to make it consistent with
LLVMOrcCreateCustomMaterializationUnit.
(rGc8fc5e3ba942057d6c4cdcd1faeae69a28e7b671)
The C API received support for adding ORCv2 object buffers
directly to an object layer.
(rG7b73cd684a8d5fb44d34064200f10e2723085c33)
A breaking change to LLVMGetInlineAsm was introduced, adding
a ninth argument LLVMBool CanThrow (D95745)
The C API received support for calling into the new pass manager.
(D102136)
The C API function LLVMIntrinsicCopyOverloadedName has been
deprecated. Please migrate to LLVMIntrinsicCopyOverloadedName2
which takes an extra module argument and which also handles
unnamed types. (D99173)
The C API received support for dumping objects from ORCv2
(rGcec8e69f01c3374cb38c6683058381b96fab8f89)
A breaking change to LLVMOrcObjectTransformLayerTransformFunction
was introduced, changing the order of the function pointer’s
arguments. (rG8962c68ad007a525f9daa987c99eda57e0d0069a)
The C API received support for accessing utilities from the
LLJIT IRTransformLayer and ThreadSafeModule classes. (D103855)
The C API received support for creating lazy-export
MaterializationUnits (D104672)
The C API function LLVMPassBuilderOptionsSetCoroutines was
removed because couroutine passes have been enabled by default.
(D105877)
comdat noduplicates was renamed to comdat nodeduplicate and as
a result, LLVMNoDuplicatesComdatSelectionKind was renamed to
LLVMNoDeduplicateComdatSelectionKind. (D106319)
Changes to the FastISel infrastructure
FastISel no longer tracks killed registers, and instead leaves
this to the register allocator. This means that hasTrivialKill()
is removed, as well as the OpNIsKill parameters to the fastEmit_*()
family of functions.
Changes to the LLVM tools
The options --build-id-link-{dir,input,output} have been deleted.
(D96310)
Support for in-order processors has been added to llvm-mca.
(D94928)
llvm-objdump supports -M {att,intel} now. --x86-asm-syntax is
a deprecated internal option which will be removed in LLVM
14.0.0. (D101695)
The llvm-readobj short aliases -s (previously --sections) and
-t (previously --syms) have been changed to --syms and
--section-details respectively, to match llvm-readelf. (D105055)
The llvm-nm short aliases -M (--print-armap), -U (--defined-only),
and -W (--no-weak) are now deprecated. Use the long form versions
instead. The alias --just-symbol-name is now deprecated in
favor of --format=just-symbols and -j. (D105330)
In lli the default JIT engine switched from MCJIT (-jit-kind=mcjit)
to ORC (-jit-kind=orc). (D98931)
llvm-rc got support for invoking Clang to preprocess its input.
(D100755)
llvm-rc got a GNU windres compatible frontend, llvm-windres.
(D100756)
llvm-ml has improved compatibility with MS ml.exe, managing to
assemble more asm files.
Changes to LLDB
LLDB executable is now included in pre-built LLVM binaries.
LLDB now includes full featured support for AArch64 SVE register
access.
LLDB now supports AArch64 Pointer Authentication, allowing
stack unwind with signed return address.
LLDB now supports debugging programs on AArch64 Linux that use
memory tagging (MTE).
Added memory tag read and memory tag write commands.
The memory region command will note when a region has memory
tagging enabled.
Synchronous and asynchronous tag faults are recognised.
Synchronous tag faults have memory tag annotations in addition
to the usual fault address.
This package contains machine-readable files for the
SPIR-V Registry. This includes:
* Header files for various languages.
* JSON files describing the grammar for the SPIR-V core instruction
set and the extended instruction sets.
* The XML registry file.
* A tool to build the headers from the JSON grammar.
Notable User Facing Changes
---------------------------
- support for LLVM 13
- CMake: Inter-Procedural Optimization is enabled on code of runtime library
(libpocl.so is compiled with -flto on systems that support it).
- LTTng tracing improved - more command types are traced, and also
some synchronous API calls (like clCreateBuffer) are traced.
- poclcc, tests and examples can be disabled with CMake options
- Valgrind support improved by making Valgrind aware of pocl's
reference counting of cl_* objects
- kernels which are called by kernels are now force-inlined
- Support for NetBSD.
- Support for Unix systems without libdl.
- PoCL can now (optionally) respond to SIGUSR2 by printing
some live debug information.
- improved SPIR support for CUDA devices
Notable Bug Fixes
-----------------
- Fixed a potential crash on Unix systems without sysfs mounted.
- Fixed compilation errors when building on macOS.
- Fixed POCL_FAST_INIT macro; POCL_INIT_LOCK must be invoked with only one argument.
- Fix bin/poclcc to not depend on OpenCL 2.0 symbols
- Fixed miscompilation in kernel loops with multiple conditionals with barriers in them.
Other
-----
- Add cmake options PARALLEL_COMPILE_JOBS, PARALLEL_LINK_JOBS to
use ninja's seperate compile and link job pools.
- Improve memory architecture, buffer migration and allocation.
Buffers are now allocated on a device when first used
(previously each buffer was allocated on every device in context).
- the single global LLVMContext was replaced with
multiple LLVMContexts, one per OpenCL cl_context.
OpenCL code can now be compiled in parallel
when using separate cl_contexts. This feature
is disabled by default since it significantly slowed
down PyOpenCL. This should be resolved by separating
LLVM compilation in their own threads in the future.
- a new OpenCL extension was added to PoCL: cl_pocl_content_size.
The extension allows the user to give optimization hint to PoCL,
which will be used internally by PoCL to optimize buffer transfers
between multiple devices.
New in this release:
* --results no longer prints the result to standard output (stdout) as voted in https://lists.gnu.org/archive/html/parallel/2020-12/msg00003.html
* parset supports associative arrays in bash, ksh, zsh.
* Online HTML is now generated by Sphinx.
* Bug fixes and man page updates.
New in 2021.3 Release
New Features
Since C++20 multiple interfaces are constrained to enforce requirements on argument types.
Preview Features
Extended the high-level task API to simplify migration from TBB to oneTBB.
Added mutex and rw_mutex that are suitable for long critical sections and resistant to high contention.
Added ability to customize the concurrent_hash_map mutex type.
Added heterogeneous lookup, erase, and insert operations to concurrent_hash_map.
Known Limitations
An application using Parallel STL algorithms in libstdc++ versions 9 and 10 may fail to compile due to incompatible interface changes between earlier versions of Threading Building Blocks (TBB) and oneAPI Threading Building Blocks (oneTBB). Disable support for Parallel STL algorithms by defining PSTL_USE_PARALLEL_POLICIES (in libstdc++ 9) or _GLIBCXX_USE_TBB_PAR_BACKEND (in libstdc++ 10) macro to zero before inclusion of the first standard header file in each translation unit.
On Linux, if oneAPI Threading Building Blocks (oneTBB) or Threading Building Blocks (TBB) is installed in a system folder like /usr/lib64, the application may fail to link due to the linker search order for libraries. Use the -L linker option to specify the correct location of oneTBB libraries. This issue does not affect program execution.
The oneapi::tbb::info namespace interfaces might unexpectedly change the process affinity mask on Windows systems (see https://github.com/open-mpi/hwloc/issues/366 for details).
Bugs Fixed
Fixed several issues in parallel algorithms and concurrent containers on ARM.
Improved ThreadSanitizer support for the oneTBB Scalable Memory Allocator.
Improved scheduler to reduce the impact of thread priority inversion.
Fixed the behavior of affinity_partitioner and static_partitioner to resolve possible performance issues.
Fixed crash when finalize is called with null scheduler_handler.
Fixed the oneTBB assertion mechanism to avoid possible leaking of user directory paths.
Fixed crash in resumable tasks on ARM-based macOS.
Fixed compilation failures
OpenCL (Open Computing Language) is an open, royalty-free standard for
cross-platform, parallel programming of diverse accelerators found in
supercomputers, cloud servers, personal computers, mobile devices and embedded
platforms.
PoCL is a portable open source (MIT-licensed) implementation of the OpenCL
standard (1.2 with some 2.0 features supported). In addition to being an easily
portable multi-device (truely heterogeneous) open-source OpenCL implementation,
a major goal of this project is improving interoperability of diversity of
OpenCL-capable devices by integrating them to a single centrally orchestrated
platform.
This package aims at creating an Open Source alternative to vendor specific
OpenCL ICD loaders.
The main difficulties to create such software is that the order of function
pointers in a structure is not publicy available.
This software maintains a YAML database of all known and guessed entries.
OpenCL (Open Computing Language) is an open, royalty-free standard for
cross-platform, parallel programming of diverse accelerators found in
supercomputers, cloud servers, personal computers, mobile devices and
embedded platforms.
This package contains the OpenCL C++ header files.
OpenCL (Open Computing Language) is an open, royalty-free standard for
cross-platform, parallel programming of diverse accelerators found in
supercomputers, cloud servers, personal computers, mobile devices and
embedded platforms.
This package contains C language headers for the OpenCL API.
opencl-clang is a thin wrapper library around clang.
The library has a OpenCL-oriented API and is capable to compile OpenCL C
kernels to SPIR-V modules.
The LLVM/SPIR-V Bi-Directional Translator is a library and tool for translation
between LLVM IR and SPIR-V.
SPIR-V is a binary intermediate language for representing graphical-shader
stages and compute kernels for multiple Khronos APIs, including OpenCL,
OpenGL, and Vulkan. SPIR-V defines a new language and is a successor to the
original Khronos SPIR, which supported only OpenCL device programs.
promises 1.2.0.1
==============
* Added `future_promise()` which returns a `promise` that executes the
expression using `future::future()`. `future_promise()` should
(typically) be a drop-in replacement for any `future::future()`
function call. `future_promise()` will not execute `future` work
faster than `future::future()`, but `future_promise()` will only
submit `future` jobs if a worker is available. If no workers are
available, `future_promise()` will hold the expression information
in a `promise` until a worker does become available to better take
advantage of computing resources available to the main R
session. For more information, please see the [`future_promise()`
article](https://rstudio.github.io/promises/articles/future_promise.html). (#62)
* Added visibility support for `Promise$then(onFulfilled)`. (#59)
promises 1.1.1
==============
* Fix handling of FutureErrors during `future::resolved()` and
`future::value()` by discarding the corrupt future. (#37)
promises 1.1.0
==============
* Fixed#49: `promise_all()` previously did not handle `NULL` values correctly. (#50))
* `new_promise_domain` now takes a `wrapOnFinally` argument, which can
be used to intercept registration of `finally()`. Previous versions
treated `finally` as passing the same callback to
`then(onFulfilled=..., onRejected=...)`, and ignoring the result;
for backward compatibility, promise domains will still treat
`finally` that way by default (i.e. if `wrapOnFinally` is `NULL`,
then `finally` will result in `wrapOnFulfilled` and `wrapOnRejected`
being called, but if `wrapOnFinally` is provided then only
`wrapOnFinally` will be called). (#43)
most of these simply extend matching from "aarch64" to "aarch64eb"
in various forms of code. most remaining uses in pkgsrc of
"MACHINE_ARCH == aarch64" are because of missing aarch64eb support,
such as most of the binary-bootstrap requiring languages like rust,
go, and java.
no pkg-bump because this shouldn't change packages on systems that
could already build all of these.
