NRC Arch meeting 2/14/2006
- Status reports
- Jae Lee: H.264 progress.
- Steve: 802.11a C implementation almost done, one more block left to do, couple of weeks left to write the input generators and output verifiers. The class work on a Bluespec implementation of this.
- Ken: Looked at ARM simulators. Tools from ARM don't rally provide what we need. Open source tools only implement ARM v5 instruction set. Some discussion, we probably only need armv5.
Anantha: They have a DSP board to do some power measurements. Also, see Jouletrack
- Daniel started looking at H.264 for the DSP
- Gopal: working on ARM processor modeling
- H.264 decoders shown at ISSCC, very low power (hundred microwatts)?
- Action Items:
- Choose cross toolchain for our use (GCC or ARM compiler)
- Need to agree on H.264 profile and benchmark streams
- Get best H.264 code inside Nokia to use as a baseline, performance wise (Gopal and Soracha)
- Look at the H.264 decoders shown at ISSCC
- Arvind: next publication deadline is ICCAD in April, always a good idea to submit a paper
- Need to define data formats so that simulators and executable models can feed into thermal and power models
- For one level of results, can feed event counts into separate power/thermal models, but execution can effect power dynamically, so need to cosimulate it
- Write a description for what is required for the simulator
- Take the instruction set simulator from skyeye, wrap it for SMASH framework, hook it up to magic memory model
- Also need proxy kernel, perhaps from simit or simplescalar kernel?