CIS 273 Computer Organization

1
CIS 273Computer Organization Design

• Instructor Dr. Michael Geiger
• Spring 2007
• Exam 2 Review

2
Lecture outline

• Tomorrows office hours 130-400, Group
  II-221C
• General exam information
• Exam review what weve covered so far
• Performance metrics
• Execution time
• Comparing performance
• Speedup
• Processor datapath design
• Processor control
• Single-cycle datapath
• Multi-cycle datapath

3
General exam information

• Very similar (in length, anyway), to Exam 1
• Closed book, closed notes exam
• No electronic devices allowed
• Calculators, cell phones, PDAs, etc.
• Exam will start by 805 (at the latest) and end
  at 915
• 6 or 7 questions, which have multiple parts
• 1 multi-part short answer question
• Some computation
• Take the math as far as you can
• Some will be simple fill-in-the-blank

4
Review Performance metrics

• Measure of how well a computer performs a users
  applications
• Measuring performance requires a basis for
  comparison and metric for evaluation
• Two standard metrics
• Latency How long does it take to complete a
  single task?
• Throughput How many tasks can be completed in a
  unit of time?
• Other metrics
• CPI Cycles per instruction (usually an average)
• MIPS Millions of instructions per second
• Both somewhat inaccurate

5
Review Execution time

• In general
• Execution time typically divided into
• User time time spent executing user program
• System time time spent on system tasks to
  support user program
• Other time system is idle, executing other code,
  etc.
• Can measure execution time in clock cycles
• Cycle time time between ticks sec / cycle
• Frequency clock rate cycles / sec

6
Review Performance comparison

• If machine X is n times faster than machine Y
• Performance of X is n times greater than
  performance of Y
• Times faster doesnt make sense for throughput
• Parameters affecting comparisons
• Changing machines, programs, compilers, etc., has
  effect on each of the three factors governing
  execution time
• Instruction count, average CPI, cycle time
• Multi-programmed workloads
• Use some sort of average
• Arithmetic mean
• Weighted mean
• Benchmarks
• Programs used to standardize performance
  measurements

7
Review Useful performance formulas

• Execution time
• (inst. count) x (average CPI) x (cycle time)
• Average CPI
• MIPS
• Speedup (Amdahls Law)

8
Review Processor implementation

• Can be broken into
• Datapath logic that operates on or stores data
• Control logic and signals determining operation
  / multiplexer selection
• Mix of combinational and sequential logic
• Registers multi-bit state elements
• Weve discussed two possible designs
• Single-cycle implementation
• Every instruction takes one cycle
• Cycle time determined by longest instruction
• Multi-cycle implementation
• Different instructions take different numbers of
  cycles
• Cycle time determined by longest phase of
  execution

9
Review Basic datapath

• Instruction supply (PC, instruction memory)
• Execution core (registers, ALU)
• Data supply (data memory)

10
Review Datapath and control

• Adding jump support additional multiplexer
  feeding PC

11
Review Multi-cycle datapath

• Break instruction into 5 stages
• 1 stage per cycle
• Registers separate stages to store results
• See lecture notes from 4/12 for complete list of
  control signals