lexain12 / AlphaBlending

Task about optimisation of AlphaBlending algorithm with intrinsic instrucitons

Alpha blending optimization

General information

This repository is the next step of "SIMD optimizations" lab. It is dedicated to alpha blending algorithm (and to cats), which I have optimized by using SSE instruction set.

Result of Alpha Blending

Simple algorithm

The main idea of alpha blending is to blend two RGBA pictures using the formula below:

Front_CLR = Front_CLR * Front_Alpha + Back_CLR * (1 - Front_CLR)

In simple algorithm I am iterating by every pixel of front image, and use the alpha blending formula.

Optimization

To optimize the algorithm, I use the SIMD instructions and calculate four pixels in one iteration.

Measurements

I measure time, using sf::Clock.

Compilation flags	Optimization	FPS	Performance
-O0	None	400	0.27.
-Ofast	None	1500	1
-Ofast	SSE	5500	3.67

The -Ofast and -O2 has the same FPS in this task, so I put -Ofast results in table.

The performance boost is $5500/1500 = 3.67$ times.

Conclusion

Optimization is a very hard task. The compiler does it, but in some cases we need to do optimization by our own and SIMD instruction is useful instrument for this. By this method I increased performance in $3.67$ times and the result worth it.