CUDA-supported Real-Time DXT Compression of HD Video: Design and Implementation

CUDA-supported real-time DXT
compression of HD video: design and

implementation
Feb. 23rd, 2011
HDTV WG Session of 31st APAN Hong Kong Meeting

@ Hong Kong, China
Dr. JongWon Kim

{jongwon}@nm.gist.ac.kr
Networked Media Lab., School of Information and Communications,
Gwangju Institute of Science and Technology (GIST), KOREA
Abstraction for Multi-party Visual Sharing
NeTD: Networked Tiled Display
• A networked display system using multiple tiled display

devices to form a large logical display wall
• Provides ultra-high resolutions and large physical display sizes
• SAGE: Scalable Adaptive Graphics Environment
– An “Operation System” for tiled-display environments
– Manages the parallel graphics streams between the rendering nodes
and tiled display nodes
3
SAGE Visualcasting for Multi-party
Collected Collaboration
HD Media Compression
• Many codecs for HD Media Compression

– MPEG2, MPEG4, H.264, etc.,
– Hard to realize real-time support
• DXT
– S/W based light-weight lossy compression algorithm
– Most of VGAs support H/W accelerated DXT-
decompression
– Each pixel block is independent of other blocks
5
HDMI-based HD Media Transport System
DXT DXT
Decompressor Decompressor
Network
DXT DXT
HDMI V Compressor Compressor V HDMI
A Audio Audio A
Transport Transport
Audio Out Audio Out
Pre- 1394 1394 Pre-
Amplifier S/W based S/W based Amplifier
Echo Controller Echo Controller
6
DXT: DirectX Texture Compression
• 4x4 block of pixels (512-bit or 384-bit) to a 64-bit or 128-bit

quantity
FOURCC Description Comp. ratio Texture
DXT1 (BC1) 1-bit Alpha / Opaque 8:1 or 6:1 Simple non-alpha
DXT3 (BC2) Explicit alpha 4:1 Sharp alpha
DXT5 (BC3) Interpolated alpha 4:1 Gradient alpha
• Select DXT1 for HD Media Compression

– HD Media does not include alpha channel
Color 0
32bit (2x16bit)
Color 1
xx xx xx xx
xx xx xx xx
32bit (16x2bit)
xx xx xx xx
xx xx xx xx
< 4x4 RGB pixels > < DXT1 block > 7

FastDXT: Realtime DXT Compression
• Focused on compression speed rather than quality

• Using Multi-Threads (2 or 4) for DXT Compression
• Optimized with SSE2 instruction-set
• Sequential compression processing each 4x4 pixel block
ㆍㆍㆍ Block1 Compression
Block2 Compression
< Image Frame > ㆍ

ㆍ
ㆍ 8
Performance of FastDXT for SAGE
• Using Decklinkcapture (HDMI Transport Application) by

VisLab@UQ
• Performance
CPU
Machine O/S Compression fps B/W(Mbps) MTU
usage
Dell Precision Ubuntu 10.04.1 DXT 12~13 105~110 75% 8900

670 (single core) x64 (64bit) Uncompressed 23~24 730~750 35% 8900
Dell Precision Ubuntu 10.04.1 DXT 30 260~270 85% 1450

T3400 (quad core) x64 (64bit) Uncompressed 13 440 8% 1450
9
GPU acceleration approach
• OpenGL/Cg
– high-level shading language developed by NVIDIA
– suitable for GPU programming and it does not replace a
general programming language
– Cg compiler outputs DirectX or OpenGL shader
programs
• CUDA: Compute Unified Device Architecture
– parallel computing architecture developed by NVIDIA
– gives developers access to the virtual instruction set and
memory of the parallel computational elements in CUDA
GPUs
10
CUDA Programming Model
• Parallel portions of an application are executed on device
(GPU) as kernels
– One kernel is executed at a time
– Many threads execute each kernel
– Kernels are lunched in grids
• Threads and Blocks have IDs

– Each Thread can decide what data to work on
11
DXT Compression using CUDA
• Each pixel block is processed by CUDA Thread Block
– Many pixel block compression simultaneously
– Enables CUDA performance scalability
• CUDA Thread processing
– Unpack color space each pixel
– Compute DXT color index each pixel
• Apply FastDXT algorithm
– Min/Max Color selection
– Emit color indices
12
Reference
1) OpenGL DXT texture compression,
http://www.opengl.org/registry/specs/EXT/texture_compression_s3tc.txt.
2) Libsquish, http://code.google.com/p/libsquish.
3) J.M.P. van Waveren “Real-Time DXT compression” May 20th 2006 ©
2006, Id Software, Inc., http://www.intel.com/cd/ids/developer/asmo-
na/eng/324337.htm.
4) L. Renambot, B. Jeong and J. Leigh, “Real-time compression for high-
resolution content,” Proceedings Access Grid Retreat 2007.
5) W. R. Mark, R. S. Glanville, K. Akeley and M. J. Kilgard, “Cg: A system for
programming graphics hardware in a C-like language,” ACM SIGGRAPH
2003
6) Nvidia-texture-tools, http://code.google.com/p/nvidia-texture-tools.
7) NVIDIA CUDA, http://developer.nvidia.com/object/cuda.html.
8) SAGE, http://www.sagecommons.org.
9) WIKIPEDIA, http://en.wikipedia.org/wiki/S3_Texture_Compression
13
Gwangju Institute of
Science & Technology
Thank you!
Send Inquiry to jongwon@nm.gist.ac.kr
http://nm.gist.ac.kr

CUDA-supported Real-Time DXT Compression of HD Video: Design and Implementation

Uploaded by

Copyright:

Available Formats

CUDA-supported Real-Time DXT Compression of HD Video: Design and Implementation

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

CUDA-supported Real-Time DXT Compression of HD Video: Design and Implementation

Uploaded by

Copyright:

Available Formats

CUDA-supported real-time DXT

compression of HD video: design and

HDTV WG Session of 31st APAN Hong Kong Meeting

Dr. JongWon Kim

• A networked display system using multiple tiled display

• Many codecs for HD Media Compression

• 4x4 block of pixels (512-bit or 384-bit) to a 64-bit or 128-bit

• Select DXT1 for HD Media Compression

< 4x4 RGB pixels > < DXT1 block > 7

• Focused on compression speed rather than quality

ㆍㆍㆍ Block1 Compression

< Image Frame > ㆍ

• Using Decklinkcapture (HDMI Transport Application) by

Dell Precision Ubuntu 10.04.1 DXT 12~13 105~110 75% 8900

Dell Precision Ubuntu 10.04.1 DXT 30 260~270 85% 1450

• Threads and Blocks have IDs

You might also like

Pfad - The Proxy pFad of © 2024 Garber Painting. All rights reserved.