1. © 2004 Wayne Wolf Memory system optimizations Strictly software:  Effectively using the cache and partitioned memory. Hardware + software:  Scratch-pad memories.  Custom memory hierarchies.

2. © 2004 Wayne Wolf Taxonomy of memory optimizations (Wolf/Kandemir) Data vs. code. Array/buffer vs. non-array. Cache/scratch pad vs. main memory. Code size vs. data size. Program vs. process. Languages.

3. © 2004 Wayne Wolf Software performance analysis Worst-case execution time (WCET) analysis (Li/Malik):  Find longest path through CDFG.  Can use annotations of branch probabilities.  Can be mapped onto cache lines.  Difficult in practice---must analyze optimized code. Trace-driven analysis:  Well understood.  Requires code, input vectors.

4. © 2004 Wayne Wolf Software energy/power analysis Analytical models of cache (Su/Despain, Kamble/Ghose, etc.):  Decoding, memory core, I/O path, etc. System-level models (Li/Henkel). Power simulators (Vijaykrishnan et al, Brooks et al).

5. © 2004 Wayne Wolf Power-optimizing transformations Kandemir et al:  Most energy is consumed by the memory system, not the CPU core.  Performance-oriented optimizations reduce memory system energy but increase datapath energy consumption.  Larger caches increase cache energy consumption but reduce overall memory system energy.

6. © 2004 Wayne Wolf Cacheing for real-time systems Kirk and Strosnider---SMART  Strategic Memory Allocation for Real-Time cache design  cache is divided into segments  critical processes get their own cache segments  hardware flag selects private cache segment or pooled cache  heuristic algorithm groups tasks into cache segments Wolfe---software cache partitioning  map routines at link time to addresses which remove conflicts for critical routines

7. © 2004 Wayne Wolf Scratch pad memories Explicitly managed local memory. Panda et al used a static management scheme.  Data structures assigned to off-chip memory or scratch pad at compile time.  Put scalars in scratch pad, arrays in main. May want to manage scratch pad at run time.

8. © 2004 Wayne Wolf Reconfigurable caches Use compiler to determine best cache configuration for various program regions.  Must be able to quickly reconfigure the cache.  Must be able to identify where program behavior changes.

9. © 2004 Wayne Wolf Software methods for cache placement McFarling analyzed inter-function dependencies. Tomiyama and Yasuura used ILP. Li and Wolf used a process-level model. Kirovski et al use profiling information plus graph model. Dwyer/Fernando use bit vectors to construct boudns in instruction caches. Parmeswaran and Henkel use heuristics.

10. © 2004 Wayne Wolf Addressing optimizations Addressing can be expensive:  55% of DSP56000 instructions performed addressing operations in MediaBench. Utilize specialized addressing registers, pre/post-incr/decrement, etc.  Place variables in proper order in memory so that simpler operations can be used to calculate next address from previous address.