Lecture 4: Optimizing for the Memory Hierarchy

Date: Oct. 15

Background Material

• Chapter 6 of Computer Systems: A Programmer's Perspective

Topics

• Motivation for memory heirarchy
  1. Processor memory gap
  2. Principle of locality
  3. Memory performance equation:
     Memory stall cycles = Instruction Count X Memory references per instruction * Miss rate * Miss penalty
• Basics of caches
  1. Where can a block be placed in cache? (block placement)
  2. How can a block be found in cache? (block identification)
  3. Which block should be replaced on a miss? (block replacement)
  4. What happens on a write? (write strategy)
• Three C model
  1. Compulsory misses
  2. Capacity misses
  3. Conflict misses
• Program to detect cache structure and performance (cache.c)
• Cache simulator (cache.map - also see cache.mw for a Maple worksheet containing the program and traces or cache.html for a viewable copy of the worksheet)
• PAPI counters for memory performance
• cachegrind cache simulator.
• Dinero IV cache simulator.

Lecture notes

• Lecture Slides class.ppt and class14.ppt from Chapter 6 of Computer Systems: A Programmer's Perspective

Programs

• cache.c - Program to empirically determine cache parameters.
• dellxpst600.xls - Results of running cache.c on a (Pentium III).

Practice Assignments

• Determine the cache configuration for your machine. (use /usr/local/papi/bin/mem_info)
• Run cache.c and analyze (try to determine cache configuration from access times).
• Run cache.c with PAPI
• Run cache.c with cachegrind
• Run cache.c with dinero and explore different cache configurations.
• Create memory trace for matrix multiplication for the Cache simulator (cache.map and explore miss rate with different cache configurations.