cpublit 0.4.2

Drawing and image composing library.


To use this package, run the following command in your project's root directory:

Manual usage
Put the following dependency into your project's dependences section:

CPUblit

Drawing and image composing library.

Description

  • Uses SSE2 or NEON where its possible for high performance.
  • Can work without garbage collection.

Use

Add this library to your project's dependency via dub or your chosen IDE.

Currently most functions are very low-level, so experience with pointers is recommended. This is to make vector optimizations easier, and pointers will be hidden in higher-level functions.

Composing function operator legends

  • src: The image that we want to compose onto the destination.
  • dest: The image we want to compose onto.
  • dest0: If present, the result of the image composition will be copied there.
  • length: The amount of pixels to be composed.
  • mask: If not present, mask will be taken from src's alpha channel or value. It controls how the composing is done, e.g. transparency, amount, etc.
  • color/value: Sets a fix amount for a given composition.

Composing function code example

Many composing functions are semi-interchangeable with virtual function calls.

alias cmpFunc = void function(uint*, uint* size_t) @nogc pure nothrow;
cmpFunc f = &alphaBlend;

NOTE: there's currently an issue with how D handles templates with overloads, so it might not really work.

The recommended solution for composing two images with different sizes is a per-line approach:

for (int i ; i < lineNum ; i++) {
    alphaBlend(src + i * srcPitch, dest + i * destPitch, lineLength);
}

General guidelines

  • Memory allocation will cause performance drop. To avoid it, use pre-allocated destinations. This is also the reason why the functions don't return either an array of result, or a pointer to it.
  • As of now, GC allocation have very minimal or no performance impact compared to using manual C-style ones. However GC allocation have the advantage of better memory safety. However it's recommended to use @nogc labels on functions where GC allocation is not needed.
  • Use of LDC2 is recommended over DMD due to it's better performance.

To do list

  • Add optimization for ARM Neon. (partly done)
  • Make a GPGPU based variant called GPUblit with either D-Compute, CUDA, and/or OpenCL.
  • Add functions for RLE compression and decompression.
  • Add higher-level functions and types.
  • Add ability of running it in betterC mode with limitations.
  • More testing.

Known issues

DMD for x86-64 targets treats vector optimization features differently from LDC. This will be fixed in the near future.

Authors:
  • Laszlo Szeremi
Dependencies:
intel-intrinsics, bitleveld
Versions:
0.4.2 2021-Apr-17
0.4.1 2021-Mar-15
0.4.0 2021-Jan-23
0.4.0-beta 2020-Dec-08
0.4.0-alpha.2 2020-Nov-05
Show all 17 versions
Download Stats:
  • 0 downloads today

  • 0 downloads this week

  • 0 downloads this month

  • 50 downloads total

Score:
1.2
Short URL:
cpublit.dub.pm