Profiling, Performance Tuning, and Design Issues

1
Profiling, Performance Tuning, and Design Issues

• Basic Efficiency Guidelines
• Select best algorithm.
• How to know? Scalable? Portable?
• Use efficient libraries when possible
• Compiler optimizations.
• Code Optimization

2
Compiler Options for producing the Fastest
Executable

• Using optimization flags when compiling can
  greatly reduce the runtime of an executable.
• Each compiler has a different set of options for
  creating the fastest executable .
• Often the best compiler options can only be
  arrived at by empirical testing and timing of
  your code.
• A good reference for compiler flags that can be
  used with various architectures is the SPEC web
  site www.spec.org.
• Read the Compiler manpages.
• GNU -O3 ffast-math funroll-loops

3
Optimizing Memory Access

• Memory access more of performance bottleneck than
  processor speed
• Largest potential for performance improvement
• Access data to minimize out-of-cache memory use

4
Memory Latencies

• CPU register 0 cycles
• L1 cache hit 2-3 cycles
• L1 cache miss satisfied by L2 cache hit 8-12
  cycles
• L2 cache miss satisfied from main memory, no TLB
  miss 75-250 cycles
• TLB miss requiring only reload of the TLB 2000
  cycles
• TLB miss requiring reload of virtual page page
  fault hundreds of millions of cycles

5
Other Code Optimizations

• Copy Propagation
• Constant Folding
• Dead Code Removal
• Induction Variable Simplification
• Function Inlining
• Loop Invariant Conditionals
• Variable RenamingLoop Invariant Code Motion
• Loop Fusion

• Pushing Loops inside Subroutines
• Loop Index Dependent Conditionals
• Loop Unrolling
• Loop Stride Size
• Floating Point Optimizations
• Faster Algorithms
• External Libraries
• Assembly Code
• Lookup Tables

6
Code Optimization References

• Software Optimizations for High Performance
  Computing by Crawford and Wadleigh
• High Performance Computing by Kevin Dowd et al
• Performance Optimization for Numerically
  Intensive Codes by Goedecker and Hoisie

7
Timing and Profiling Codes

• Need to know where to focus attention
• Premature Optimization is the root of all evil
  
• Donald Knuth
• The 80-20 rule codes generally spend 80 of
  their time executing 20 of their instructions
• flat profile shows how much time your program
  spent in each function, and how many times that
  function was called.
• call graph shows, for each function, which
  functions called it, which other functions it
  called, and how many times.
• annotated source listing is a copy of the
  program's source code, labeled with the number of
  times each line of the program was executed.

8
GNU gprof

• The first step in generating profile information
  for your program is to compile and link it with
  profiling enabled use the -pg' option when you
  run the compiler. (This is in addition to the
  options you normally use.)
• The -pg' option also works with a command that
  both compiles and links
• cc -o myprog myprog.c utils.c -g -pg
• Execute code in normal manner
• ./myprog
• Create profile with gprof
• gprof myprog gt myprog.prof

9
Profiling on the Beowulf Cluster

• Compile
• pgf77 -Mproffunc program.f
• pgcc -Mproffunc program.c
• Run the code
• To produce a profile data file called pgprof.out.
  
• View the execution profile
• pgprof pgprof.out

10
Pgprof (without x windows)

• Loading....
• Datafile pgprof.out
• Processes 1
• pgprofgt print
• Time/ Function
• Calls Call() Time() Cost() Name
• --------------------------------------------------
  ----------------------
• 4100500 0.00 23.43 23 lxi (cdnz3d.f1632)
• 4100500 0.00 21.90 22 damping (cdnz3d.f2319)
• 4100500 0.00 21.87 22 leta (cdnz3d.f1790)
• 4100500 0.00 11.68 12 lzeta (cdnz3d.f1947)
• 4100500 0.00 11.24 33 sum (cdnz3d.f2107)
• 250 0.02 5.99 97 page (cdnz3d.f1527)
• 0.01 2.79 3 tmstep (cdnz3d.f678)
• pgprofgt quit

11
Overview of PAPI

• Performance Application Programming Interface
• The purpose of the PAPI project is to design,
  standardize and implement a portable and
  efficient API to access the hardware performance
  monitor counters found on most modern
  microprocessors.
• Parallel Tools Consortium project
• http//www.ptools.org/

12
PAPI Counter Interfaces

• PAPI provides three interfaces to the underlying
  counter hardware
• The low level interface manages hardware events
  in user defined groups called EventSets.
• The high level interface simply provides the
  ability to start, stop and read the counters for
  a specified list of events.
• Graphical tools to visualize information.

13
(No Transcript)
14
Parallel Communication Profiling

• A significant factor that affects the performance
  of a parallel application is the balance between
  communication and workload.
• The challenge of the message passing model is in
  reducing message traffic over the interconnection
  network. Performance analysis tools are needed.
  
• Two such tools
• VAMPIR http//www.pallas.com
• uses the profile extensions to MPI and permits
  analysis of the message events where data is
  transmitted between processors during execution
  of a parallel program. It has user-interface with
  zooming and filtering.
• PARAVER http//www.cepba.upc.es/
• was developed to respond to the basic need to
  have a qualitative perception of the
• application behavior by visual inspection and
  then to be able to focus on the detailed
  quantitative analysis of the problems.

15
(No Transcript)
16
(No Transcript)