a unified VLD/IQ stage having a variable-length decoder (VLD) and a inverse-quantizer (IQ), that outputs from the inverse-quantizer non-predicted discrete cosine transform (DCT) coefficients, the non-predicted DCT coefficients not adjusted for AC prediction of coefficients, the unified VLD/IQ stage not performing AC prediction of coefficients;

an inverse DCT transformer, receiving DCT coefficients, for performing an inverse discrete cosine transform on the DCT coefficients to generate pixels for a current block;

a mux supplying the DCT coefficients to the inverse DCT transformer, the mux selecting the non-predicted DCT coefficients from the unified VLD/IQ stage when AC prediction is not performed, but selecting predicted DCT coefficients for a first row or a first column in the current block when AC prediction is preformed;

a coefficient store, coupled to receive the DCT coefficients from the mux, for storing DCT coefficients for prior blocks as stored DCT coefficients; and

a post-IQ calculator, receiving the stored DCT coefficients from the coefficient store, and receiving the non-predicted DCT coefficients from the inverse-quantizer in the unified VLD/IQ stage, for generating the predicted DCT coefficients to the mux by performing post-inverse-quantizer AC prediction;

wherein the unified VLD/IQ stage sends quantized DCT coefficients from the variable-length decoder directly to the inverse-quantizer without outputting the quantized DCT coefficients;

wherein the stored DCT coefficients in the coefficient store are non-quantized DCT coefficients;

wherein the current block is block N, wherein N is a positive integer;

wherein a prior block is an above block N-L when DCT coefficients for the first row of the current block are AC predicted, wherein L is a number of blocks in a image-row of a video frame;

wherein the prior block is an immediately prior block N-1 when DCT coefficients for the first column of the current block are AC predicted;

wherein the post-IQ calculator comprises:

a current Q-subtractor that receives the current quantization parameter sent to the inverse-quantizer for the current block and receives the non-predicted DCT coefficients from the inverse-quantizer, for generating current corrected coefficients;

a prior Q-subtractor that receives the stored quantization parameter and the stored DCT coefficients from the coefficient store for a prior block, for generating first prior corrected coefficients;

a divider/multiplier, receiving the first prior corrected coefficients from the prior Q-subtractor, for dividing and multiplying by the current quantization parameter to generate adjusted prior coefficients;

an adder, coupled to the current Q-subtractor, for adding the current corrected coefficients with the adjusted prior coefficients to generate combined coefficients; and

a final combiner, coupled to the adder, for combining the current quantization parameter with the combined coefficients to generate the predicted DCT coefficients to the mux;

whereby the quantized DCT coefficients are not used for AC prediction but the non-quantized DCT coefficients to the inverse DCT transformer are used for AC prediction and whereby prior blocks for AC prediction are above blocks and immediately prior blocks, and

whereby AC prediction is performed on DCT coefficients input to the inverse DCT transformer after the inverse-quantizer.