Transcript Multicore, parallelism, and multithreading

MULTICORE, PARALLELISM,
AND MULTITHREADING
By: Eric Boren, Charles Noneman, and Kristen Janick
MULTICORE PROCESSING
Why we care
What is it?


A processor with more than one core on a single
chip
Core: An independent system capable of processing
instructions and modifying registers and memory
Motivation


Advancements in component technology and
optimization are limited in contribution to processor
speed
Many CPU applications attempt to do multiple
things at once:
 Video
editing
 Multi-agent simulation

So, use multiple cores to get it done faster
Hurdles

Instruction assignment (who does what?)
 Mostly
delegated to the operating system
 Can be done to a small degree through dependency
analysis on the chip

Cores must still communicate at times – how?
 Shared-memory
 Message
passing
Advantages

Multiple Programs:
Can be separated between cores
 Other programs don’t suffer when one hogs CPU


Multi-threaded Applications:


Independent threads don’t have to wait as long for each
other – results in faster overall execution
VS Multiple Processors
Less distance between chips - faster communication results in
higher maximum clock rate
 Less expensive due to smaller overall chip area, shared
components (caches, etc.)

Disadvantages




OS and programs must be optimized for multiple cores,
or no gain will be seen
In a singly-threaded application, little to no
improvement
Overhead in assigning tasks to cores
Real bottleneck is typically memory and disk access
time – independent of number of cores
Amdahl’s Law

Potential performance increase on a parallel
computing platform is given by Amdahl’s law.
 Large
problems are made up of several parallelizable
parts and non-parallelizable parts.



S = 1/(1-P)
S = speed-up of program
P = fraction of program that is parallizable
Current State of the Art

Commercial processors:
Most have at least 2 cores
 Quad-core are highly popular for desktop applications
 6-core processors have recently appeared on the market
(Intel’s i7 980X)
 8-core exist but are less common


Academic and research:
MIT: RAW 16-core
 Intel Polaris – 80-core
 UC Davis: AsAP – 36 and 167-core, individually-clocked

PARALLELISM
What is Parallel Computing?

Form of computation in which many calculations are
carried out simultaneously.
 Operating
on the principle that large problems can
often be divided into smaller ones, which are solved
concurrently.
Types of Parallelism

Bit level parallelism


Instruction level parallelism


Instructions combined into groups
Data parallelism


Increase processor word size
Distribute data over different computing environments
Task parallelism

Distribute threads across different computing environments
Flynn’s Taxonomy
Single Instruction, Single Data (SISD)



Provides no parallelism in
hardware
1 data stream processed by
the CPU in 1 clock cycle
Instructions executed in serial
fashion
Multiple Instruction, Single Data (MISD)


Process single data stream using multiple instruction
streams simultaneously
More theoretical model than practical model
Single Instruction, Multiple Data (SIMD)



Single instruction steam has ability to process multiple
data streams in 1 clock cycle
Takes operation specified in one instruction and
applies it to more than 1 set of data elements at 1
time
Suitable for
graphics
and image
processing
Multiple Instruction, Multiple Data (MIMD)


Different processors can execute different
instructions on different pieces of data
Each processor can run independent task
Automatic parallelization



The goal is to relieve programmers from the tedious
and error-prone manual parallelization process.
Parallelizing compiler tries to split up a loops so
that its iterations can be executed on separate
processors concurrently
Identify dependences between references -independent actions can operate in parallel
Parallel Programming languages


Concurrent programming languages, libraries, API’s,
and parallel programming models have been
created for programming parallel computers.
Parallel languages make it easier to write parallel
algorithms
 Resulting
code will run more efficiently because the
compiler will have more information to work with
 Easier to identify data dependencies so that the runtime
system can implicitly schedule independent work
MULTITHREADING
TECHNIQUES
fork()



Make a (nearly) exact duplicate of the process
Good when there is no or almost no need to
communicate between processes
Often used for servers
fork()
Parent
Globals
Heap
Stack
Child
Globals
Heap
Stack
Child
Globals
Heap
Stack
Child
Globals
Heap
Stack
Child
Globals
Heap
Stack
fork()
pid_t pID = fork();
if (pID == 0) {
//child
} else {
//parent
}
POSIX Threads

C library for threading

Available in Linux, OS X

Shared Memory

Threads are created and destroyed manually

Has mechanisms for locking memory
POSIX Threads
Process
Globals
Heap
Thread
Thread
Thread
Thread
Stack
Stack
Stack
Stack
POSIX Threads
pthread_t thread;
pthread_create( &thread, NULL, function_to_call,
(void*) data);
//Do stuff
pthread_join(thread, NULL);
POSIX Threads
int total = 0;
void do_work() {
//Do stuff to create “result”
total = total + result;
}
Thread 1 reads total (0)
Thread 2 reads total (0)
Thread 1 does add and saves total (1)
Thread 2 does add and saves total (2)
POSIX Threads
int total = 0;
pthread_mutex_t mutex = PTHREAD_MUTEX_INITIALIZER;
void do_work() {
//Do stuff to create “result”
pthread_mutex_lock( &mutex );
total = total + result;
pthread_mutex_unlock( &mutex );
}
OpenMP

Library and compiler directives for multi-threading

Support in Visual C++, gcc

Code compiles even if compiler doesn't support OpenMP

Popular in high performance communities

Easy to add parallelism to existing code
OpenMP
Initialize an Array
const int array_size = 100000;
int i, a[array_size];
#pragma omp parallel for
for (i = 0; i < array_size; i++) {
a[i] = 2 * i;
}
OpenMP
Reduction
#pragma omp parallel for reduction(+:total)
for(i = 0; i < array_size; i++)
{
total = total + a[i];
}
Grand Central Dispatch
Apple Technology for Multi-Threading
Programmer puts work into queues
A system central process determines the number threads to
give to each queue
Add code to queues using a closure
Right now Mac only, but open source
Easy to add parallelism to existing code

Grand Central Dispatch
Initialize an Array
dispatch_apply(array_size,
dispatch_get_global_queue(0, 0), ^(int i) {
a[i] = 2*i;
});
Grand Central Dispatch
GUI Example
void analyzeDocument(doc) {
do_analysis(doc); //May take a very long time
update_display();
}
Grand Central Dispatch
GUI Example
void analyzeDocument(doc) {
dispatch_async(dispatch_get_global_queue(0, 0), ^{
do_analysis(doc);
update_display();
});
}
Other Technologies

Threading in Java, Python, etc.

MPI – for clusters
QUESTIONS?
Supplemental Reading

Introduction to Parallel Computing
 https://computing.llnl.gov/tutorials/parallel_comp/#A
bstract

Introduction to Multi-Core Architecture
 http://www.intel.com/intelpress/samples/mcp_samplec
h01.pdf

CPU History: A timeline of microprocessors
 http://everything2.com/title/CPU+history%253A+A+t
imeline+of+microprocessors