Optimization For Embedded System

Peformance

Performance Indices

Memory Usage Power

80% of time is spent in standby mode 80% of operating time is a subset of peripherals and functions used

80/20 20% of time is spent in operating mode

20% of operating time is spent with all functions in use

Von Neumann

Architecture

Harvard Saving power at boot time

Software Techniques

Scaling voltage and frequency Using sleep modes and idling clock domain Coordinating sleep and scaling

Power

Reduce power supply voltage

Embedded processor architecture

Run at low clock frequency

Chip level

Disable function units with control signals when not in use Disconnect part from power supply when not in use

Optimal performance key algorithm Low power for key algorithm

Pick Criteria

Efficient management of memory (cache) Register Memory hierarchy

Different levels of cache Main memory Disc space

Loosely coupled to OS kernel Set of library coutines; execute in client's context Power Manager OS layer

Configurable as necessary Use platform-specific adaptation library for V/F scaling Application, drivers, CLK register for notifications Application trigger actions

Algorithm Prevent obsessive optimization Inline functions Table lookups

Optimization for Embedded System

Developing efficient code

Hand coded assembly Register variables Polling Fixed point arithmetic

Layers

Avoid standard library routine Decreasing code size

Native word size C++ Subtopic Architecture SW to have natural "idle" points Use interrupt-driven programming Code and data placement close to processor to minimize off-chip accesses Smart placement to allow frequencyly accessed code/data close to CPU Size optimizations to reduce footprint, memory, and corresponding leakage

Top 10 SW power optimization

Optimize for speed for more CPU idle mode or reduce CPU frequency Don't over calculate Use DMA for efficient transfer Use co-processors to efficiently handle/accerlate frequent/specialized processing

Code

Use more buffering and batch processing to allow more computation at once and more time in low power modes Simplifies control flow Allocates variables to registers -o0 (register)

Eliminates unused code Simplifies expressions and statements Expands calls to inline functions Performs local copy/constant propagation

-o1 (local)

Removes unused assignments Eliminates local common expressions Performs local loop optimizations

Optimization levels -o2 (global)

Eliminates global common sub-expressions Eliminates global unused assignments Perform loop unrolling Removes functions that are never called Simplifies functions with return value that are never used

-o3 (file)

Reorder functions so that attributes of called function are known when caller is optimized Identifies file-level variable characteristics

Optimization for Embedded System.mmap - 2007/8/26 -