Multicore group works on task management API
SAN JOSE, Calif — The Multicore Association has launched a new working group to define a Multicore Task Management API to ease parallel programming for embedded systems developers working on heterogeneous processors. The group hopes to release MTAPI in November 2012.
Existing thread management tools require too much memory and lack the flexibility embedded developers need, said Urs Gleim, a program manager in a corporate R&D group at Siemens AG who chairs the new group. Members include representatives from ENEA, Freescale Semiconductor, LSI, Qualcomm, Plurality, PolyCore Software, Texas Instruments, and Wind River.
Gleim is part of a small team at Siemens that has developed parallel-programming APIs for its internal use on some of the proprietary health care, industrial automation, and energy systems it builds.
"This was the next logical step for us and something we think can benefit the industry," Gleim said.
Tools such as the Thread Building Blocks and Cilk libraries from Intel and the open-source OpenMP tools typically focus on homogenous shared-memory multicore processors that use a single operating system and can require megabytes of memory. MTAPI aims to enable processors that may use a mix of different cores, accelerators and OSes in chips that may have only tens of kilobytes of internal memory, Gleim said.
Another goal of MTAPI is to enable software developers to improve task scheduling in flexible ways that can optimize a design for diverse kinds of embedded systems. The API will be defined in C and work with the association's existing communications and resource-management APIs.
MTAPI aims to address a variety of applications including recursive algorithms such as Fourier transforms, data-parallel tasks such as video streaming, and real-time control tasks. It also hopes to form a basis for building higher level software abstractions such as data-flow graphs and pipelines.
The group started in May and has already collected a core set of requirements. It is starting implementation work on the API now, but is still open to new members.
"Our experience has been that if you have more companies involved it increases the chances of more complete coverage in the standard," said Markus Levy, president of the association.
This story was originally posted by EE Times.
Multicore group works on task management API
Existing thread management tools require too much memory and lack the flexibility embedded developers need, said Urs Gleim, a program manager in a corporate R&D group at Siemens AG who chairs the new group. Members include representatives from ENEA, Freescale Semiconductor, LSI, Qualcomm, Plurality, PolyCore Software, Texas Instruments, and Wind River.
Gleim is part of a small team at Siemens that has developed parallel-programming APIs for its internal use on some of the proprietary health care, industrial automation, and energy systems it builds.
"This was the next logical step for us and something we think can benefit the industry," Gleim said.
Tools such as the Thread Building Blocks and Cilk libraries from Intel and the open-source OpenMP tools typically focus on homogenous shared-memory multicore processors that use a single operating system and can require megabytes of memory. MTAPI aims to enable processors that may use a mix of different cores, accelerators and OSes in chips that may have only tens of kilobytes of internal memory, Gleim said.
Another goal of MTAPI is to enable software developers to improve task scheduling in flexible ways that can optimize a design for diverse kinds of embedded systems. The API will be defined in C and work with the association's existing communications and resource-management APIs.
MTAPI aims to address a variety of applications including recursive algorithms such as Fourier transforms, data-parallel tasks such as video streaming, and real-time control tasks. It also hopes to form a basis for building higher level software abstractions such as data-flow graphs and pipelines.
The group started in May and has already collected a core set of requirements. It is starting implementation work on the API now, but is still open to new members.
"Our experience has been that if you have more companies involved it increases the chances of more complete coverage in the standard," said Markus Levy, president of the association.
This story was originally posted by EE Times.
Multicore group works on task management API
Comments
Post a Comment