Computer Organization and

Architecture (AT70.01)
Comp. Sc. and Inf. Mgmt.
Asian Institute of Technology
Instructor: Dr. Sumanta Guha
Slide Sources: Patterson &
Hennessy COD book website
(copyright Morgan Kaufmann)
adapted and supplemented
COD Ch. 2
The Role of Performance
Performance
 Performance is the key to understanding underlying
motivation for the hardware and its organization
 Measure, report, and summarize performance to enable users
to
 make intelligent choices
 see through the marketing hype!

 Why is some hardware better than others for different programs?

 What factors of system performance are hardware related?
(e.g., do we need a new machine, or a new operating system?)
 How does the machine's instruction set affect performance?
 What do we measure?
Define performance….
Airplane Passengers Range (mi) Speed (mph)

Boeing 737-100 101 630 598

Boeing 747 470 4150 610
BAC/Sud Concorde 132 4000 1350
Douglas DC-8-50 146 8720 544

 How much faster is the Concorde compared to the 747?

 How much bigger is the Boeing 747 than the Douglas DC-8?

 So which of these airplanes has the best performance?!

Computer Performance:
TIME, TIME, TIME!!!
 Response Time (elapsed time, latency):
 how long does it take for my job to run?
 how long does it take to execute (start to Individual user
finish) my job? concerns…
 how long must I wait for the database query?
 Throughput:
 how many jobs can the machine run at once?
 what is the average execution rate?
 how much work is getting done? Systems manager
concerns…
 If we upgrade a machine with a new processor what do we increase?
 If we add a new machine to the lab what do we increase?
Execution Time
 Elapsed Time
 counts everything (disk and memory accesses, waiting for I/O, running other
programs, etc.) from start to finish
 a useful number, but often not good for comparison purposes
elapsed time = CPU time + wait time (I/O, other programs, etc.)

 CPU time
 doesn't count waiting for I/O or time spent running other programs
 can be divided into user CPU time and system CPU time (OS calls)
CPU time = user CPU time + system CPU time
 elapsed time = user CPU time + system CPU time + wait time

 Our focus: user CPU time (CPU execution time or, simply, execution time)
 time spent executing the lines of code that are in our program
Definition of Performance
 For some program running on machine X:

PerformanceX = 1 / Execution timeX

 X is n times faster than Y means:

PerformanceX / PerformanceY = n
Clock Cycles
 Instead of reporting execution time in seconds, we often use cycles. In
modern computers hardware events progress cycle by cycle: in other
words, each event, e.g., multiplication, addition, etc., is a sequence of
cycles
seconds cycles seconds
 
program program cycle
 Clock ticks indicate start and end of cycles:

cycle time
tick
tick

 cycle time = time between ticks = seconds per cycle

 clock rate (frequency) = cycles per second (1 Hz. = 1 cycle/sec, 1 MHz.
= 106 cycles/sec)
 Example: A 200 Mhz. clock has a
cycle time 1
 10 9  5 nanoseconds
200  10 6
Performance Equation I
seconds cycles seconds
 
program program cycle

equivalently

CPU execution time CPU clock cycles  Clock cycle time

=
for a program for a program

 So, to improve performance one can either:

 reduce the number of cycles for a program, or
 reduce the clock cycle time, or, equivalently,
 increase the clock rate
How many cycles are required
for a program?
 Could assume that # of cycles = # of instructions
2nd instruction
3rd instruction
1st instruction

4th
5th
6th
...
time

 This assumption is incorrect! Because:

 Different instructions take different amounts of time (cycles)
 Why…?
 How many cycles are required
for a program?
time

 Multiplication takes more time than addition

 Floating point operations take longer than integer ones
 Accessing memory takes more time than accessing registers
 Important point: changing the cycle time often changes the
number of cycles required for various instructions because it
means changing the hardware design. More later…
Example
 Our favorite program runs in 10 seconds on computer A,
which has a 400Mhz. clock.
 We are trying to help a computer designer build a new
machine B, that will run this program in 6 seconds. The
designer can use new (or perhaps more expensive)
technology to substantially increase the clock rate, but has
informed us that this increase will affect the rest of the CPU
design, causing machine B to require 1.2 times as many clock
cycles as machine A for the same program.

 What clock rate should we tell the designer to target?

Terminology
 A given program will require:
 some number of instructions (machine instructions)
 some number of cycles
 some number of seconds
 We have a vocabulary that relates these quantities:
 cycle time (seconds per cycle)
 clock rate (cycles per second)
 (average) CPI (cycles per instruction)
 a floating point intensive application might have a higher average CPI
 MIPS (millions of instructions per second)
 this would be higher for a program using simple instructions
Performance Measure
 Performance is determined by execution time

 Do any of these other variables equal performance?

 # of cycles to execute program?
 # of instructions in program?
 # of cycles per second?
 average # of cycles per instruction?
 average # of instructions per second?

 Common pitfall : thinking one of the variables is indicative of

performance when it really isn’t
Performance Equation II
CPU execution time Instruction count  average CPI  Clock cycle time
=
for a program for a program

 Derive the above equation from Performance Equation I

CPI Example I
 Suppose we have two implementations of the same
instruction set architecture (ISA). For some program:
 machine A has a clock cycle time of 10 ns. and a CPI of 2.0
 machine B has a clock cycle time of 20 ns. and a CPI of 1.2

 Which machine is faster for this program, and by how much?

 If two machines have the same ISA, which of our quantities (e.g.,
clock rate, CPI, execution time, # of instructions, MIPS) will always
be identical?
CPI Example II
 A compiler designer is trying to decide between two code
sequences for a particular machine.
 Based on the hardware implementation, there are three different
classes of instructions: Class A, Class B, and Class C, and they
require 1, 2 and 3 cycles (respectively).
 The first code sequence has 5 instructions:
2 of A, 1 of B, and 2 of C
The second sequence has 6 instructions:
4 of A, 1 of B, and 1 of C.

 Which sequence will be faster? How much? What is the CPI for each
sequence?
MIPS Example
 Two different compilers are being tested for a 500 MHz. machine
with three different classes of instructions: Class A, Class B, and
Class C, which require 1, 2 and 3 cycles (respectively). Both
compilers are used to produce code for a large piece of software.
 Compiler 1 generates code with 5 billion Class A instructions, 1
billion Class B instructions, and 1 billion Class C instructions.
 Compiler 2 generates code with 10 billion Class A instructions, 1
billion Class B instructions, and 1 billion Class C instructions.

 Which sequence will be faster according to MIPS?

 Which sequence will be faster according to execution time?
Benchmarks
 Performance best determined by running a real application
 use programs typical of expected workload
 or, typical of expected class of applications
e.g., compilers/editors, scientific applications, graphics, etc.

 Small benchmarks
 nice for architects and designers
 easy to standardize
 can be abused!

 Benchmark suites
 Perfect Club: set of application codes
 Livermore Loops: 24 loop kernels
 Linpack: linear algebra package
 SPEC: mix of code from industry organization
SPEC (System Performance
Evaluation Corporation)
 Sponsored by industry but independent and self-managed –
trusted by code developers and machine vendors
 Clear guides for testing, see www.spec.org
 Regular updates (benchmarks are dropped and new ones added
periodically according to relevance)
 Specialized benchmarks for particular classes of applications
 Can still be abused…, by selective optimization!
SPEC History
 First Round: SPEC CPU89
 10 programs yielding a single number
 Second Round: SPEC CPU92
 SPEC CINT92 (6 integer programs) and SPEC CFP92 (14 floating point
programs)
 compiler flags can be set differently for different programs
 Third Round: SPEC CPU95
 new set of programs: SPEC CINT95 (8 integer programs) and SPEC CFP95
(10 floating point)
 single flag setting for all programs
 Fourth Round: SPEC CPU2000
 new set of programs: SPEC CINT2000 (12 integer programs) and SPEC
CFP2000 (14 floating point)
 single flag setting for all programs
 programs in C, C++, Fortran 77, and Fortran 90
CINT2000 (Integer component
of SPEC CPU2000)
Program Language What It Is
164.gzip C Compression
175.vpr C FPGA Circuit Placement and Routing
176.gcc C C Programming Language Compiler
181.mcf C Combinatorial Optimization
186.crafty C Game Playing: Chess
197.parser C Word Processing
252.eon C++ Computer Visualization
253.perlbmk C PERL Programming Language
254.gap C Group Theory, Interpreter
255.vortex C Object-oriented Database
256.bzip2 C Compression
300.twolf C Place and Route Simulator
 CFP2000 (Floating point
component of SPEC CPU2000)
Program Language What It Is
168.wupwise Fortran 77 Physics / Quantum Chromodynamics
171.swim Fortran 77 Shallow Water Modeling
172.mgrid Fortran 77 Multi-grid Solver: 3D Potential Field
173.applu Fortran 77 Parabolic / Elliptic Differential Equations
177.mesa C 3-D Graphics Library
178.galgel Fortran 90 Computational Fluid Dynamics
179.art C Image Recognition / Neural Networks
183.equake C Seismic Wave Propagation Simulation
187.facerec Fortran 90 Image Processing: Face Recognition
188.ammp C Computational Chemistry
189.lucas Fortran 90 Number Theory / Primality Testing
191.fma3d Fortran 90 Finite-element Crash Simulation
200.sixtrack Fortran 77 High Energy Physics Accelerator Design
301.apsi Fortran 77 Meteorology: Pollutant Distribution
SPEC CPU2000 reporting
 Refer SPEC website www.spec.org for documentation
 Single number result – geometric mean of normalized ratios for
each code in the suite
 Report precise description of machine
 Report compiler flag setting
Specialized SPEC Benchmarks
 I/O
 Network
 Graphics
 Java
 Web server
 Transaction processing (databases)
Amdahl's Law
 Execution Time After Improvement =
Execution Time Unaffected + ( Execution Time Affected / Rate of
Improvement )

 Example: Improved part of code

 Suppose a program runs in 100 seconds on a machine, with
multiply responsible for 80 seconds of this time.
 How much do we have to improve the speed of multiplication if we want the program
to run 4 times faster?
 How about making it 5 times faster?

 Design Principle: Make the common case fast

Examples
 Suppose we enhance a machine making all floating-point instructions
run five times faster. The execution time of some benchmark before
the floating-point enhancement is 10 seconds.

 What will the speedup be if half of the 10 seconds is spent executing floating-
point instructions?

 We are looking for a benchmark to show off the new floating-point

unit described above, and want the overall benchmark to show a
speedup of 3. One benchmark we are considering runs for 100
seconds with the old floating-point hardware.

 How much of the execution time would floating-point instructions have to account
for in this program in order to yield our desired speedup on this benchmark?
Summary
 Performance is specific to a particular program
 total execution time is a consistent summary of performance
 For a given architecture performance increases come from:
 increases in clock rate (without adverse CPI affects)
 improvements in processor organization that lower CPI
 compiler enhancements that lower CPI and/or instruction count
 Pitfall: expecting improvement in one aspect of a machine’s
performance to affect the total performance
 You should not always believe everything you read! Read
carefully! See newspaper articles, e.g., Exercise 2.37!!