Using　a　very　long　instruction　word　(VLIW)　architecture　view,　the　existing　state　of　the　art　algorithms　were　compared　with　an　algorithm　that　is　tailored　for　VLIW　architecture　in　the　form　of　matrix.　The　architecture＇s　character　is　further　exploited　to　design　a　new　fast　inverse　discrete　cosine　transform　(IDCT)　algorithm.　Compared　with　existing　algorithms,　the　new　algorithm　reduces　the　number　of　instruction　cycles　needed.　Further,　the　register　character　of　VLIW　is　used　for　merging　motion　compensation　with　IDCT　(DCT)　in　the　video　codec　process.　This　achieves　50%　speed　improvement.　The　concept　is　applicable　for　MPEG1/2/4　and　H.264/AVC.