RenderScript · Parallel Computing on Android, the Easy Way by Marchetti, Alberto -- Read -- Imperial Library of Trantor

Index

RenderScript: parallel computing on Android, the easy way Special thanks Table of contents Introduction

Who this book is for What is parallel computing?

CPU side GPU side Result

What is needed to use this book Author thoughts How this book is structured Source code

What is RenderScript

Experience-based API Note How RenderScript works First example

Setup RenderScript Create an RS Script

Conclusions

RenderScript components

The language Element

Predefined elements Custom struct elements Custom Java elements

Type Allocation Context

Debug note

Script

#pragma rs_fp_relaxed Script components FilterScript

Message queue RenderScript support library

USAGE_IO flags Book note

Conclusions

Performance notes

Memory architecture

Example: convolution GPU or CPU?

Profiler sample project

Profiler: memory access Profiler: pure calculation Profiler: kernels calling Test specifics

Crowdsourced testing

Memory access Pure calculation RGBA to grayscale Kernels calling

Conclusions

Native analysis

RenderScript call workflow

static void nAllocationCopyToBitmap(...) rsAllocationCopyToBitmap(...) void rsi_AllocationCopyToBitmap(...) void Allocation::read(...) void (*data2D)(...) void rsdAllocationData2D(...) Bonus function: static void Update2DTexture(...) RenderScript driver Note: Java pointers

RenderScript and 64-bit RenderScript compile and runtime

slang - Compiler for RenderScript language libbcc - Executor of RenderScript bitcode resources Diagram

Conclusions

RenderScript and NDK

Access RenderScript elements using Java reflection

Allocation pointer access Native kernel calling

RenderScript NDK native support

Setup C++ code Notes

Conclusions

Use cases

Debug Random numbers Blur YUV to RGBA conversion RGBA to grayscale conversion Surface rendering

Usage Workflow

Camera capture

Workflow

Profiling code Resizing images

Custom implementation

Color normalization Custom struct element copy to Java

Padding Offset Reading elements

Dynamic bitcode loading

Workflow Module - rsscriptparser Module - rslibrary Module - app

Porting case - FAST features detection

FAST features Detection process The porting process

Input Porting - Original FAST library Porting - OpenCV implementation Porting - conclusions

The comparison

Input CPU version GPU version

Results Further works - Scoring Conclusions

Conclusions

The framework The book What is missing?

Appendix - API changes

API 19 API 20 API 21 API 22 API 23

Appendix: Enable NDK support

Install NDK Simple implementation External ndk-build implementation

Appendix: Other Android parallel computing frameworks

OpenGL ES OpenCL Vulkan CUDA (Only for Tegra-powered devices)

← Prev
Back
Next →

← Prev
Back
Next →