RichardB's notes from the seminar

From eLinux.org
Revision as of 12:10, 23 January 2009 by Richardb (Talk | contribs)

Jump to: navigation, search

Texas Instruments & ARM: OMAP Seminar - 21 Jan. 09 – ARM, Cambridge, UK

Cortex A8 Core– Bryan Lawrence – ARM

  • Cortex A8 Core is the design. OMAP is the physical implementation of this design by TI
  • Cortex A8 is based on V7-A instruction-set architecture and includes:
    • NEON advanced SIMD (multimedia accelerator – integer and floating-point SIMD (single-instruction multiple-data)
    • Jazelle-RCT (Java accelerator)
    • TrustZone security foundation (effectively virtualisation of the core)
  • Particularly aimed at applications (rather than real-time or ‘deeply’ embedded)
  • A8 Processor Core (design) can run up too 1GHz +, c. 2000 DMIPS, depending on silicon
  • MMU for OS virtual memory management
  • Thumb-2 allows 16 & 32-bit instructions. Allows efficient, but small (compressed) code size if required
    • Dependant on the compiler t**produce better ‘code density’
  • Thumb-2, for e.g., gives a 29% reduction in Linux kernel size. E.g.: http://www.arm.com/products/os/linux.html
  • CoreSight; non-invasive real-time trace for debugging
    • JTAG port
    • Debug access port (DAP)
    • Embedded Trace Macrocell (ETM) – captures instruction and data


NEON SIMD – Ashley Stevens – ARM

  • Flexible, generic multimedia acceleration
  • High-power consumption than dedicated hardware but supports emerging standards
  • Hybrid 64/128-bit SIMD architecture
  • Supports up too 64-bit integers, single-precision floating-point
  • Adds additional registers
  • Variety of ways to use: assembler , C Intrinsics, through too OpenMAX DL library (recommended), Vectorizing compilers (generates NEON SIMD instructions)
  • Provides, for e.g., faster FFT’s
  • Armcc vs gcc : armcc produces more compact, faster code.
  • Lots of NEON-optimised codecs available


OMAP35x Processor Overview – Chris Bowers – Snr Field Applications Engineer – TI

  • TI have a range of microcontrollers through t**Application processors & DSP
  • OMAP tends t**be seen in things like digital signage, POS terminals, portable infotainment etc (Lower power, high performance); “Laptop-like performance”
  • Up t**1200 Dhrystone MIPS
  • ARCHOS7 Internet Media Table built on OMAP3
  • TI are “nicely surprised” by things like BeagleBoard
  • Has a DSP (in addition t**NEON) for vide**processing, up t**HD
    • DSP is generic; not limited t**video/audi**processing
  • Peripheral connectivity (USB, MMC, Serial , USB etc.)
  • OMAP35 models:
    • 3503 - ARM Cortex A8, Peripherals
    • 3515 - ARM Cortex A8, Peripherals, PowerVR SGX (OpenGL ES) graphics engine
    • 3525 - ARM Cortex A8, Peripherals, C64x DSP & video accelerator
    • 3530 - ARM Cortex A8, Peripherals, PowerVR SGX (OpenGL ES) graphics engine, DSP & video accelerator
  • Camera interface
    • Auto-focus engine
    • CCD & CMOS imager interface
    • Preview engine etc.
  • Display subsystem
    • (24-bit RGB up to 1024x768 HD, 2 x 10-bit DAC’s; rotation, image resizing)
    • Overlay, scaling, picture-in-picture
  • Also discussed TI DaVinci platform: video-centric, based an ARM9, has some overlap with OMAP
  • OMAP35x has power-management module. (PRCM), active and static (standby) modes of consumption
    • Can reduce core voltage and frequency
    • Various major components can be turned on/off as required – “power domains”
  • Various complete boards available:
    • OMAP35x evaluation module (EVM); OMAP 3530 plus touchscreen, RAM & NAND flash, Ethernet etc.
    • BeagleBoard
    • Gumstix Overo(tiny)
    • LogicPD
    • Analog & Micro


Understanding 2D/3D Graphics Dev using OMAP 35x - Jason Brand – Fields Apps Engineer – TI

  • Lots of uses/major apps;
    • Scalable UI’s , navigation, Games, Visualisations, Automotive
  • OMAP 35x has NEON vector floating-point processor (VFP) +
  • PowerVR SGX (graphics engine):
    • Tile-based architecture
    • Universal Scalable Shader Engine (USSE)
    • Support for: OpenGL ES (Embedded Standard) 1.1 and 2.0, OpenVG 1.0 (t**accelerate Adobe Flash and SVG Tiny (Scalable Vector Graphics) and UI’s built on these)
    • ~10M polygons/second, ~0.9 GFLOPS
  • OpenGL ES is a well-defined subset of desktop OpenGL
  • (lots of details on SGX engine)
  • OpenGL ES support seems powerful
  • Graphics SDK is available from TI; tools, headers, libs, demos etc
  • IVA 2.2 – Image, Video, Audio subsystem- C64x DSP core:
    • 32-bit fiex-point media processor
    • Video & image accelerator
    • TI supply compiler tools to optimize for this hardware


ARM Software Development Tools – Elan Lennard – System Design Division – ARM

  • “Enabling all developers to get the best from their ARM-based system”
  • Quality, high-performance s/w
  • Tools: Compilation, Optimization, Middleware, Device Support, verification & debug, Fast simulation
  • RealiVew Development Suite:
    • Co-developed and validated with ARM processor IP; best code
    • Extensive support for CoreSight (debug tech)
    • Supports all ARM processors
    • Std and Pro editions. ***Pro includes NEON compiler, RealView profiler, fast simulator (RTSM), ICE
    • Automatic optimisation; data from profiler feeds back int**compiler == some perf improvement (c. 6%) and 40% (ish) code side reduction.
  • Loop unrolling (where appropriate)
  • Code reordering
    • Link-time compilation; allows optimizations across source files, 5% size reduction, 5% perf improvement
    • ARM compiler vs. GCC: ARM is 30% faster, 43% smaller. (similar when using Thumb code)
    • NEON Vectorizing compiler; up t**400% (4x) performance improvement on a particular vide**decoder, compared t**regular ARM compiler
  • ARM Workbench IDE – based on Eclipse 3.3
    • ARM Eclipse plugins; ARM profiler, Flash programmer, ARM Linux project wizard etc. etc.
    • Only really useful if RealView is used
  • ARM Profiler:
    • “Get the best out of ARM processors”
    • Performance and code coverage analysis; detailed analysis of performance/usage, call-chain analysis
    • Traces can be logged and replayed
    • Completely non-intrusive; analyse running system/application
    • Good e.g. show one instruction using 27% of application time
  • RealView ICE and Trace,
    • Hardware trace/debugger


Tool Chain Overview – Chris Bower – TI

  • TI Code Composer Toolset
    • DSPBIOS, low level ARM debug, DSP development and debug
  • Montavista (for DaVinci)
    • Linux-based, licensed through TI.
    • Linux app development, Eclipse-based IDE
  • Green Hills
    • Integrity Linux based, MULTI debug environment for DSP and ARM. Application too
  • Code Sourcery
    • Linux (and Windows) – GNU Toolchain. For building Linux apps
    • Eclipse-based for Pers and Pr**editions
  • MPC
    • WinCE
  • Microsoft Platform Builder
    • The choice for Windows CE etc. development
  • Lauterbach TRACE32
    • Low-level debug of ARM & DSP


OMAP3 OS Support – Jason Brand – TI

  • Fundamentally this is Linux or Windows CE.
  • TI issue a Linux 2.6.22 kernel, includes lots of device drivers, EVM drivers, on top of which:
    • There is also DSPBIOS – scheduler, resource manager for DSP.
    • Als**layers on top of these; codec interfaces, algorithm abstraction, Open VG, OpenGL ES, audio/vide**(GStreamer) etc.
  • Windows CE 6.0 can als**function as the ‘underlying’ OS, some of the higher layers are different
  • OMAP353 - SDK Beta SDK:
    • Board boot, test, & flash utils
    • Platform support:
  • U-boot Linux boot-load and flashing
  • Linux kernel with drivers
  • Root fs
    • Dem**apps
  • Image viewer
  • DaVinci I/F dump
    • Code Sourcery tools
    • X-Loader
  • Small user boot-loader t**boot from on-board flash
  • Must be signed before use
    • U-Boot
  • The next-stage boot-loader
  • Flexible open-source utility for boot-loading Linux
  • Capable of reading kernel image from flash, Ethernet TFTP, and ?
    • ITBOK (is the board ok)
  • Based on u-boot
  • Basic H/W functionality tests
  • OMAP35x WinCE Support:
    • TI seeing 40% WinCE vs 60% Linux
    • MS suggest that total cost of development is cheaper , and to-market faster than with Linux
    • BSQUARE’s WinCE 6.0 R2 BSP (board support package) Dem**and Source is free with OMAP EVM
  • Visual Tools plugin
  • WinCE R2 Pr**compiled with Visual Tools
  • Various codecs, DirectShow filters etc. for a/v
  • Production Tested (tw**full QA passes)
  • 100% CETK passed
    • Adobe Flash Lite 3 port available for WinCE R2 BSP – OMAP35x EVM


Power for OMAP35x Processors – Miriam Corder – TI

  • Max power consumption is 360mW
  • With dynamic voltage/freq scaling – averaging <100mW
  • External power-management chips available (“analog companion “)
    • Includes audi**codec, RTC, USB OTG transceiver, battery charger etc.