In　this　paper,　the　matrix　decomposition　of　existing　DCT　algorithms　is　presented.　Using　a　Very　Long　Instruction　Word　(VLIW)　architectural　view,　existing　state　of　the　art　algorithms　are　compared　with　an　algorithm　that　is　tailored　for　DCT.　VLIW　operation　is　further　utilized　to　design　a　modified　IDCT　algorithm.　Compared　with　existing　algorithms,　the　new　algorithm　reduces　the　number　of　instruction　cycles　needed.　In　addition,　we　use　the　register　character　of　VLIW　for　merging　motion　compensation　(MC)　with　IDCT　(DCT)　in　the　video　codec.　This　reduces　the　number　of　cycles　required　for　MC　to　about　50%　of　the　number　required　for　the　unimproved　one.　The　approach　is　applicable　to　MPEG1/2/4,　H.263　and　H.264/AVC.