Hit performance goals with configurable processors
October 17, 2006 -- dspdesignline.com
With mainstream DSPs, code must be hand-tuned using assembly code in order to meet performance goals. A more productive approach is to tailor the processor to the algorithm. This article explains why, using FFT, Viterbi, and MPEG4 examples.
For more than 30 years, the fixed-ISA (instruction set architecture) microprocessor has largely defined electronic system design. Similarly, single-chip DSPs have dominated DSP system design since they were introduced 20 years ago. Fixed-ISA processors have addressed application performance problems in two ways. The first approach has been to increase clock rate, mirroring the same trend in PC processors. If power dissipation is not important, rapidly escalating clock rates can go a long way towards force-fitting a particular processor to an application. The alternative approach for fixed-ISA processors is to provide more computational resources that operate in parallel so that the processor can perform more work per clock.
Designers of fixed-ISA microprocessors and DSPs attempt to develop architectures that are good at executing a wide range of algorithms, but not tailored to any specific application. This design approach can reduce clock-rate requirements, but it doesn't match the clock-rate or gate-usage efficiencies of a tailored solution. The design of complex portable and battery-powered processor-based systems increasingly calls for taking a different approach, one that more closely matches the processing resources to the application tasks so that very high clock rates (and correspondingly high power and heat dissipations) are not required. FPGA, ASIC and SOC technologies provide an ideal medium for tailoring processors to specific applications, using newly available configurable and extensible processor technology.
To read the full article, click here
Related Semiconductor IP
- RVA23, Multi-cluster, Hypervisor and Android
- 64 bit RISC-V Multicore Processor with 2048-bit VLEN and AMM
- NPU IP Core for Mobile
- RISC-V AI Acceleration Platform - Scalable, standards-aligned soft chiplet IP
- H.264 Decoder
Related White Papers
- Tuning Fork - A Tool For Optimizing Parallel Configurable Processors
- Configurable processors or RTL -- evaluating the tradeoffs
- Configurable Processors: Ready for Prime Time
- Configurable Processors for Video Processing SOCs
Latest White Papers
- QiMeng: Fully Automated Hardware and Software Design for Processor Chip
- RISC-V source class riscv_asm_program_gen, the brain behind assembly instruction generator
- Concealable physical unclonable functions using vertical NAND flash memory
- Ramping Up Open-Source RISC-V Cores: Assessing the Energy Efficiency of Superscalar, Out-of-Order Execution
- Transition Fixes in 3nm Multi-Voltage SoC Design