optimization of system performance using openmp m5151117 yumiko kimezawa may 25, 20111rps
TRANSCRIPT
Optimization of System Performance using OpenMP
m5151117Yumiko Kimezawa
May 25, 2011 1RPS
Outline
• Problems with Previous Research• What is OpenMP and MPI ?
- A brief example of OpenMP
• Research Goal• Future Work
May 25, 2011 RPS 2
Problems with Previous Research•The scale of systems is large•Accuracy of PPD Algorithm•A/D converters are not implemented
-Raw ECG data is analog data
May 25, 2011 RPS 3
PPD Algorithm will be optimized by using OpenMP or MPI
What are OpenMP and MPI ?
• OpenMP- A shared-memory application programming interface (API)- Facilitator of shared-memory parallel programming- Is used on a board range of SMP architecture- Is usually used to parallelize loops
• MPI- Message Passing Interface- Programming for distributed-memory architecture- Multiple processes operate independently
May 25, 2011 RPS 4
A brief example of OpenMP
May 25, 2011 RPS 5
Fork-join programming model
Team of threads
Initial threads
Initial threads
Fork
Join
Thread 0: for(i=0; I < 25; i++) sum += i;Thread 1: for(i=25; I < 50; i++) sum += i;Thread 2: for(i=50; I < 75; i++) sum += i;Thread 3: for(i=75; I < 100;I ++) sum += i;
+ Sum
Research Goal
• Optimization of - the HW
• Implementation of the A/D converter- the SW
• Parallelization of algorithm• Improvement of accuracy of algorithm
• Real-time evaluation- Conclusive results are sent to the web
May 25, 2011 RPS 6
Future Work
• Continue to study OpenMP• Research of other algorithms
May 25, 2011 RPS 7
May 25, 2011 RPS 8
Period detection
Peaks processing
Data reading
Derivation
Autocorrelation
Finding interval
Extraction
Store of results
Discrimination
Period-Peaks Detection (PPD) Algorithm (2/2)
May 25, 2011 9RPS
Period detection
Peaks processing
Data reading
Derivation
Autocorrelation
Finding interval
Extraction
Store of results
Discrimination
Period-Peaks Detection (PPD) Algorithm (2/2)
nn
nynyt
t
y
)1(
][]1[)(
][ny
nt,
: Current sampling data (filtered ECG signals)
: Current time (step)
][]1[ nyny
EquationEquation
May 25, 2011 10RPS
Period detection
Peaks processing
Data reading
Derivation
Autocorrelation
Finding interval
Extraction
Store of results
Discrimination
Period-Peaks Detection (PPD) Algorithm (2/2)
Autocorrelation functionAutocorrelation function
][][][0
LnynyLRN
ny
yR
][ny
N
: The autocorrelation function
: The number of times needed for
the calculations to get the period (256 in our program)
: The filtered ECG signal
May 25, 2011 11RPS
Period detection
Peaks processing
Data reading
Derivation
Autocorrelation
Finding interval
Extraction
Store of results
Discrimination
Period-Peaks Detection (PPD) Algorithm (2/2)
May 25, 2011 12RPS
System architecture
Graphic LCD Controller
Graphic LCD Controller
Master CPUMemory
Master CPUMemory
MasterCPUTimerTimer
GraphicLCD
GraphicLCDLEDLED JTAG
UARTJTAGUART
PPD Module Master Module
LEDController
LEDController
Avalon Bus
FIR FilterFIR FilterTimerTimer
Slave CPU Memory
Slave CPU Memory
SlaveCPU
ExternalMemoryExternalMemory
SharedMemoryShared
Memory
ECG Data Rom
ECG Data Rom
May 25, 2011 13RPS
: Data flow : Control signal : Data flow : Control signal
System architecture
Graphic LCD Controller
Graphic LCD Controller
Master CPUMemory
Master CPUMemory
MasterCPUTimerTimer
GraphicLCD
GraphicLCDLEDLED JTAG
UARTJTAGUART
PPD Module Master Module
LEDController
LEDController
Avalon Bus
FIR FilterFIR FilterTimerTimer
Slave CPU Memory
Slave CPU Memory
SlaveCPU
ExternalMemoryExternalMemory
SharedMemoryShared
Memory
ECG Data Rom
ECG Data Rom
May 25, 2011 14RPS
1: Signal Reading : Data flow
: Control signal : Data flow : Control signal
System architecture
Graphic LCD Controller
Graphic LCD Controller
Master CPUMemory
Master CPUMemory
MasterCPUTimerTimer
GraphicLCD
GraphicLCDLEDLED JTAG
UARTJTAGUART
PPD Module Master Module
LEDController
LEDController
Avalon Bus
FIR FilterFIR FilterTimerTimer
Slave CPU Memory
Slave CPU Memory
SlaveCPU
ExternalMemoryExternalMemory
SharedMemoryShared
Memory
ECG Data Rom
ECG Data Rom
May 25, 2011 15RPS
2: Filtering : Data flow : Control signal : Data flow : Control signal
System architecture
Graphic LCD Controller
Graphic LCD Controller
Master CPUMemory
Master CPUMemory
MasterCPUTimerTimer
GraphicLCD
GraphicLCDLEDLED JTAG
UARTJTAGUART
PPD Module Master Module
LEDController
LEDController
Avalon Bus
FIR FilterFIR FilterTimerTimer
Slave CPU Memory
Slave CPU Memory
SlaveCPU
ExternalMemoryExternalMemory
SharedMemoryShared
Memory
ECG Data Rom
ECG Data Rom
May 25, 2011 16RPS
3: Analysis : Data flow : Control signal : Data flow : Control signal
System architecture
Graphic LCD Controller
Graphic LCD Controller
Master CPUMemory
Master CPUMemory
MasterCPUTimerTimer
GraphicLCD
GraphicLCDLEDLED JTAG
UARTJTAGUART
PPD Module Master Module
: Data flow : Control signal : Data flow : Control signal
LEDController
LEDController
Avalon Bus
FIR FilterFIR FilterTimerTimer
Slave CPU Memory
Slave CPU Memory
SlaveCPU
ExternalMemoryExternalMemory
SharedMemoryShared
Memory
ECG Data Rom
ECG Data Rom
May 25, 2011 17RPS
4: Display