< previous page page_132 next page >

Page 132
Some observations:
1. For a three-cycle memory, a single register set of about 16 registers plus a data cache seems to be more area-effective than multiple register windows.
2. As the memory access time increases beyond three cycles, data indicate that the single register set approach plus the data cache is even more effective than the register window approach.
3. Since register windows require a significant amount of chip area and increased control complexity compared to the single register set plus cache approach, MRS (windows) may cause a deterioration in the cycle time due to increased register select logic.
4. If a significant amount of area is available for data cache and a data cache is added to an MRS implementation, then ultimately this combination will use fewer cycles than a single register set implementation with data cache using the same area. The MRS approach saves additional time at procedure call.
Basic Optimization Design Rules:
1. Within instruction set (compatibility, etc.) constraints, determine effect of instruction set enhancements on code density and I-cache size and decoder area. For purposes of this optimization assume separate I- and D-caches split about evenly.
2. Within constraints outlined in (1), determine the effects of register set enhancements on memory traffic and memory access delay. Optimize D-cache/register area allocation.
3. Based on any changes made in (1) and (2), reevaluate cycle time and any adverse effect on performance.
2.7 Conclusions
Cycle time is of paramount importance in microprocessor design. It is largely determined by technology, but is significantly influenced by secondary considerations such as clocking philosophy and pipeline segmentation.
Once cycle time has been determined, the designer's next challenge is to optimize the cost-performance of a design by making maximum use of chip areausing chip area to the best possible advantage of performance. A technology-independent measure of area called the register bit equivalentthe rbeprovides the basis for storage hierarchy tradeoffs among a number of important architectural considerations.
The instruction set is the glue that brings together implementation and technology and forms the specification of the ''user-visible logical machine."
The dynamic character of the high-level language source program is represented by HLL measures of the program. The number of HLL instructions

 
< previous page page_132 next page >