Stanford EE 392C - Study of Multimedia Applications

Unformatted text preview:

Study of Multimedia ApplicationsHonggo WijayaArjun SinghVarun Sagar MalhotraNavneet AronCharacteristics• Intensive computation for highly regular operations• Data locality (Spatial and Temporal)• Frequent encounter of small integer operands• Usually demand real-time processing capabilitiesCustom Hardware Support• Application specific processors (DSPs)• General Purpose Processors with Media coprocessors (Imagine).• Adding Multimedia Instructions to the ISA (Intel MMX, Sun VIS, MIPS MDMX)Benchmarks• MediaBench+• Video : MPEG-2, MPEG-4, H.263• Audio : ADPCM coder• Graphics : Mesa• Image : JPEG, EPIC, Ghostscript• Security : PGP, Pegwit• Speech : GSM, G.271, Rasta– Also includes MPEG-4 and H.263 as a representative of emerging video applicationsInstruction Mix Comparison0102030405060Spec Int Spec Fp Multimedia%age of Instruction mixMem ory Integer Alu FP ops Branch OtherBranches• The branches mainly due to loops.• Static branch prediction performs nearly as well as dynamic branch prediction (Fritts et al.)Memory Access Patterns– Load small amount of data at a time that can fit inside a cache, processes it, then throws it away• Most of the data is used before ejected from the cache– Good spatial as well as reasonably good temporal locality• Prefetching techniques will work well since branches are also predictable• Stream buffers beneficialMemory Access Patterns(contd)• Multimedia applications generate fewer or equal data memory references per instruct as compared to SPEC Int95 applications• They have a slightly lower or equal cache miss rate• They do not place more stringent memory requirement than other applicationsParallelism• Thread Level: Video (De)compression – coarse grained-suitable for CMPs• Instruction Level: Image Processing– Fine grained –Superscalar/Wide Issue– Average Basic Block Size: 8• Data Level: Graphics Applications– Producer Consumer locality – suitable for stream architectures.– Compiler support in identifying data parallelism can boost performance (Vector processors)Future Trends• MPEG-4: Less processing regularity and high data rate• Graphics Application : – Lots of floating point ops– High memory bandwidth– Distributed frame buffers is one proposed solution (one processor per frame buffer)• Communication overhead, high network


View Full Document

Stanford EE 392C - Study of Multimedia Applications

Download Study of Multimedia Applications
Our administrator received your request to download this document. We will send you the file to your email shortly.
Loading Unlocking...
Login

Join to view Study of Multimedia Applications and access 3M+ class-specific study document.

or
We will never post anything without your permission.
Don't have an account?
Sign Up

Join to view Study of Multimedia Applications 2 2 and access 3M+ class-specific study document.

or

By creating an account you agree to our Privacy Policy and Terms Of Use

Already a member?