Skip to content

Latest commit

 

History

History
383 lines (288 loc) · 16.4 KB

CHANGELOG.md

File metadata and controls

383 lines (288 loc) · 16.4 KB
Raw

Nanos6 Release Notes

All notable changes to this project will be documented in this file.

Version 4.1, Wed May 15, 2024

The 4.1 version corresponds to the OmpSs-2 2024.05 release. It introduces the directory/cache (D/C) for Host and CUDA devices. It also adds several fixes for the instrumentation.

General

  • Add directory/cache (D/C) for Host and CUDA devices
  • Add device memory allocation API for D/C-managed memory
  • Require ovni 1.5.0 or greater
  • Fix NUMA tests to accept executions with no NUMA_ALL_ACTIVE

Instrumentation

  • Fix thread instrumentation order when blocking task
  • Call ovni_thread_requires for the Nanos6 model
  • Call ovni_thread_free when threads end
  • Always mark task as paused when entering taskwait in ovni

Version 4.0, Fri Nov 17, 2023

The 4.0 version corresponds to the OmpSs-2 2023.11 release. It provides support for the ALPI tasking interface, reduces the external software requirements, and removes obsolete features.

General

  • Implement the ALPI tasking interface to support external task-aware libraries
  • Allow embedding jemalloc allocator
  • Embed hwloc and jemalloc by default
  • Add devices.cuda.prefetch option to control CUDA prefetching of data dependencies (enabled by default)
  • Install the nanos6.toml configuration file in $prefix/share
  • Remove obsolete instrument.h public interface
  • Remove obsolete stats and graph instrumentations
  • Remove software dependency with libunwind and elfutils
  • Improve ovni library discovery
  • Fix execution when enabling extrae instrumentation
  • Fix memory leaks
  • Fix several tests

Version 3.0, Wed May 24, 2023

The 3.0 release corresponds to the OmpSs-2 2023.05 release. It introduces several performance improvements, important bug fixes, and improved usability and programmability. It also improves the support for the ovni instrumentation.

General

  • Leverage C++17 standard, which may require newer GCC (such as GCC 7 or later)
  • Fix visualization of task labels for programmatically spawned tasks (e.g., polling tasks from task-aware libraries)
  • Deprecate CTF instrumentation; use ovni instrumentation instead
  • Remove support for the task for clause

Performance

  • Decrease the default immediate successor probability to 0.75 instead of 1.0. Always applying the immediate successor policy can degrade the performance of some applications. Instead, if your program considerability relies on it, set it back to 1.0
  • Remove several dynamic memory allocations that were on the critical path of Nanos6 code
  • The turbo.warmup indicates whether the runtime should perform a warmup of Jemalloc arenas (enabled by default)
  • Add the config list option cpumanager.sponge_cpus to indicate which CPUs should not be used by the runtime. A sponge CPU is a CPU that the runtime system has available but it does not execute any task (or runtime code) on it. Such CPUs are useful to reduce the system noise. The runtime leaves these CPUs free (without consuming CPU time) so that the system can schedule other threads and interruptions on them

Building and Usability

  • Add the autogen.sh script to prepare autotools, instead of autoreconf
  • Allow embedding a hwloc library into Nanos6 to avoid conflicts with other external hwloc libraries
  • Add the configure option --with-hwloc to specify whether hwloc should be external or embedded
  • Attach the hwloc 2.9.1 tarball inside the deps folder for the embedded default hwloc. See autogen.sh --help for more information

ovni Instrumentation

  • Add support for the ovni's Idle view that can be displayed with Paraver
  • Add support for the ovni's Breakdown view that can be displayed with Paraver
  • Support ovni 1.2.0 version and higher compatible versions
  • Perform a run-time version check to verify if the loaded ovni library is compatible
  • Link Nanos6 with ovni library using runpath instead of rpath to allow changing the ovni library through LD_LIBRARY_PATH

Version 2.8, Tue Nov 15, 2022

The 2.8 release corresponds to the OmpSs-2 2022.11 release. It introduces LLVM support for CUDA tasks and runtime loading. It also adds OVNI instrumentation support and several bug fixes and features that improve the overall performance and programmability.

General

  • Add a probabilistic attribute to the Immediate Successor feature (enabled by default)
  • Improve locking, reduce allocations, and fix alignment issues
  • Improve task creation performance

ovni Instrumentation

  • Add support for ovni instrumentation through the configuration: version.instrument=ovni
  • Choose detail level with the option instrument.ovni.level

LLVM Support for CUDA

  • Support device(cuda) tasks in OmpSs-2 programs built with the LLVM compiler
  • Drop support for device(cuda) in Mercurium
  • Support both building kernels separately with NVCC and linking them to the final binary or building directly with LLVM
  • Support building PTX binaries and CUDA kernels at runtime when placed in a specific folder (by default nanos6-cuda-kernels)
  • Add a new configuration option for the default CUDA kernels folder devices.cuda.kernels_folder

Version 2.7.1, Fri May 27, 2022

The 2.7.1 release corresponds to the OmpSs-2 2021.11.1 release. It introduces some bug and code fixes, and some minor improvements

General

  • Adapt taskfor to avoid overwriting task args in compiler-generated code
  • Improve support for custom CXXFLAGS at configure time
  • Add --disable-all-instrumentations configure option
  • Modify API to allow setting task labels
  • Provide nanos6-info with new options to show compile/link runtime flags
  • Provide nanos6-info with new options to show current and default config files

Version 2.7, Wed Nov 17, 2021

The 2.7 release corresponds to the OmpSs-2 2021.11 release. It introduces some performance and code fixes, and several fixes for the CTF tracing tools.

General

  • Set hybrid CPU manager policy as default
  • Fix the setting of a floating-point optimization bit in the CSR register (x86) when enabling turbo mode
  • Add several fixes to CTF tracing tools
  • Add support for if(0) and taskwaits with dependencies in fast CTF converter (nanos6-ctf2prv-fast)
  • Remove unnecessary warning at run-time in the NUMA-aware code

Version 2.6, Wed Jun 30, 2021

The 2.6 release corresponds to the OmpSs-2 2021.06 release. It introduces several features and fixes that improve the general performance and programmability. It introduces a NUMA-aware API to allocate structures in taskified applications. It also adds a new and faster CTF trace converter and support for multi-process tracing.

General

  • Add NUMA-aware API to allocate structures in taskified programs
  • Add new hybrid CPU manager policy that combines both idle and busy policies
  • Make idle CPUs block inside the scheduler while there are no ready tasks when enabling the busy policy
  • Add support for the onready task clause
  • Consolidate the new polling mechanism using the nanos6_wait_for function
  • Remove polling services API
  • Remove OmpSs-2@Cluster features and code

Instrumentation

  • Add fast CTF trace converter enabled through the config option instrument.ctf.converter.fast
  • Add support for multi-process tracing enabled by the Task-Aware MPI and Task-Aware GASPI libraries
  • Add merger tool for multi-process traces named nanos6-mergeprv

Version 2.5.1, Tue Dec 22, 2020

The 2.5.1 release corresponds to the OmpSs-2 2020.11.1 release. It introduces bug fixes and code improvements.

General

  • Unify instrumentation, monitoring and hwcounter points
  • Efficient support for taskloop dependencies
  • Fix reductions in taskloops and taskfors
  • Centralize configuration variables
  • Fully implement assert directive
  • Abort execution when an invalid config variable is enabled
  • Fix CTF instrumentation bugs
  • Bugfixes, performance and code improvements

Version 2.5, Wed Nov 18, 2020

The 2.5 release corresponds to the OmpSs-2 2020.11 release. It introduces several features and fixes that improve general performance. It replaces the configuration environment variables with a configuration file, improving the usability of the runtime system. It also makes the discrete dependency system the default implementation.

General

  • Replace all environment variables with a configuration file
  • Add NANOS6_CONFIG environment variable to specify the configuration file
  • Add NANOS6_CONFIG_OVERRIDE to override options from the configuration file
  • Enhance performance in architectures with hyperthreading
  • Improve locking performance
  • Allocate critical C++ containers with the custom memory allocator
  • Support the assert directive to check the loaded dependency system

Dependency System

  • Make discrete the default dependency system
  • Improve allocations in discrete dependencies
  • Add support for CUDA task reductions in discrete dependencies
  • Use address translation tables for specifying task reductions' storage

Instrumentation

  • Add support for kernel events in CTF instrumentation
  • Add new Paraver views for CTF traces

Devices

  • Add fixes for OpenACC and CUDA devices

Version 2.4.1, Tue Sep 22, 2020

The 2.4.1 release corresponds to the OmpSs-2 2020.06.1 release. It introduces bug fixes and performance improvements.

General

  • Improve the interface and performance of the scheduler's lock
  • Fix CTF instrumentation bugs and limitations
  • Fix PAPI hardware counters backend
  • Support newer versions of GCC, Clang and GLIBC
  • Fix task external events API
  • Remove preemption mechanism from critical sections
  • Fix initialization of locks
  • Add test suite built with the OmpSs-2 compiler based on LLVM
  • Add new tests

Version 2.4, Mon Jun 22, 2020

The 2.4 release corresponds to the OmpSs-2 2020.06 release. It introduces several features that improve the general performance of OmpSs-2 applications. It adds a new variant to extract execution traces with a lightweight internal tracer. It also improves the support for CUDA and provides support for OpenACC tasks.

General

  • Use jemalloc as a scalable multi-threading memory allocator
  • Add turbo variant enabling floating-point optimizations and the discrete dependency system
  • Refactor of CPU Manager and DLB support improvements
  • Bugfixes, performance and code improvements

Scheduling

  • Improve taskfor distribution policy
  • Improve scheduling performance and code
  • Add the nanos6_wait_for function to efficiently pause a task for a given time

Dependency System

  • Implement the discrete dependency system with lock-free techniques
  • Add support for weak dependencies in discrete
  • Add support for commutative and concurrent dependencies in discrete

Instrumentation

  • Refactor the hardware counters infrastructure and support both PAPI and PQoS counters
  • Add ctf variant to extract execution traces in CTF format using a lightweight internal tracer
  • Provide the ctf2prv tool to convert CTF traces to Paraver traces
  • Avoid Extrae trace desynchronizations in hybrid MPI+OmpSs-2 executions
  • Remove the stats-papi instrumentation variant

Devices

  • Refactor of the devices' infrastructure
  • Perform transparent CUDA Unified Memory prefetching
  • Add support for cuBLAS and similar CUDA APIs
  • Add support for OpenACC tasks

Version 2.3.2, Wed Jan 8, 2020

The 2.3.2 release corresponds to the OmpSs-2 2019.11.2 release. It mainly introduces bug fixes.

General

  • Fix important error at the runtime initialization
  • Fix in discrete dependency system
  • Several fixes for OmpSs-2@Cluster

Version 2.3.1, Tue Dec 10, 2019

The 2.3.1 release corresponds to the OmpSs-2 2019.11.1 release. It introduces bug fixes and performance improvements.

General

  • Fix execution of CUDA tasks
  • Fix dmalloc in OmpSs-2@Cluster
  • Add missing calls to CPU Manager
  • Improve taskfor performance
  • Improve general performance by using a reasonable cache line size padding
  • Add tests checking the execution of CUDA tasks

Version 2.3, Mon Nov 18, 2019

The 2.3 release corresponds to the OmpSs-2 2019.11 release. It introduces a new optimized data dependency implementation. It improves the usability, performance and code of the scheduling infrastructure and the task for feature. It also adds support for DLB and OmpSs-2@Linter.

General

  • Data dependency implementation can be decided at run-time through NANOS6_DEPENDENCIES variable
  • Performance and code improvements on the task for feature
  • Add support for Dynamic Load Balancing (DLB) tool
  • Add support for OmpSs-2@Linter
  • Important bugfix in memory allocator (used by OmpSs-2@Cluster)
  • Bugfixes, performance and code improvements

Dependency System

  • Add new optimized discrete dependency system implementation; enabled by NANOS6_DEPENDENCIES=discrete

Scheduling

  • Usability, performance and code improvements on the scheduling infrastructure

Instrumentation

  • Remove profile instrumentation variant
  • Remove interception mechanism of memory allocation functions

Version 2.2.2, Mon Oct 7, 2019

The 2.2.2 release corresponds to the OmpSs-2 2019.06.2 release. It introduces bug fixes.

General

  • Compile extrae variant with high optimization flags
  • Remove backtrace sampling from the extrae variant

Version 2.2.1, Fri Sep 27, 2019

The 2.2.1 release corresponds to the OmpSs-2 2019.06.1 release. It mainly introduces bug fixes and code improvements.

General

  • Rename loop directive to task for
  • Tasks can leverage reductions and external events at the same time (over distinct data regions)
  • OmpSs-2@Cluster bugfixes
  • Fix binding information reported by nanos6-info binary
  • Support for the TAGASPI library
  • Other bugfixes and code improvements

Version 2.2, Mon Jun 17, 2019

The 2.2 release corresponds to the OmpSs-2 2019.06 release. It mainly introduces the new support for OmpSs-2@Cluster. It also includes some improvements and optimizations for array task reductions and general bugfixes.

General

  • Support for OmpSs-2@Cluster
  • Bugfixes and performance improvements

Dependency System

  • Bugfixes and optimization for array reductions
  • Delete obsolete task data dependency implementations

Scheduling

  • Delete obsolete schedulers

Version 2.1, Fri Nov 9, 2018

The 2.1 release corresponds to the OmpSs-2 2018.11 release. It provides full support for the TAMPI library. It also includes general bugfixes and performance improvements.

General

  • Full support for TAMPI
  • Bugfixes and performance improvements

Others

  • Bugfixes in task external events API

Version 2.0.2, Mon Jun 25, 2018

The 2.0.2 release corresponds to the OmpSs-2 2018.06.2 release.

General

  • Bugfixes in HWLOC support

Version 2.0.1, Mon Jun 25, 2018

The 2.0.1 release corresponds to the OmpSs-2 2018.06.1 release.

General

  • Bugfixes in task reductions

Version 2.0, Mon Jun 25, 2018

The 2.0 release corresponds to the OmpSs-2 2018.06 release. It introduces support for OmpSs-2@CUDA in Unified Memory NVIDIA devices. It also supports array task reductions in C/C++ and task priorities. Additionally, it provides two new APIs used by the TAMPI library.

General

  • Support for OmpSs-2@CUDA Unified Memory
  • Bugfixes and performance improvements

Dependency System

  • Support for array task reductions in C/C++

Scheduling

  • Support for task priorities
  • Add priority scheduler

Others

  • Add polling services API
  • Add task external events API
  • Rename taskloop construct to loop

Version 1.0.1, Thu Nov 23, 2017

The 1.0.1 release corresponds to the OmpSs-2 2017.11.1 release.

General

  • Fixes for the building system
  • Fixes for the loading system

Version 1.0, Mon Nov 13, 2017

The 1.0 release corresponds to the OmpSs-2 2017.11 release. It is the first release of the Nanos6 runtime system. It implements the basic infrastructure to manage the parallelism of user tasks (task creation, task scheduling, etc) and their data dependencies. The task dependency system supports the nested dependency domain connection, and both early release and weak dependency models.

General

  • General infrastructure of the runtime system
  • Support for user tasks and nesting of tasks

Scheduling

  • Implement different schedulers: FIFO, LIFO, etc

Dependency System

  • Implementation of a task data dependency system
  • Support for nested dependency domain connection
  • Support for early release of task dependencies
  • Support for weak task dependencies
  • Support for reductions

Others

  • Taskloop construct with dependencies
  • Task pause/resume API