It seems that TMuC0.7 and TMuC0.7.1 may yield different results with different compilers. For example, for random access high efficiency configuration, when Qp=37, the results I got(compiled with VS2008 or VS2010, in 64bits) are as follows.

Bitrate: 480.2856 PSNR: 30.4399(Y) 35.0552(U) 36.6993(V)

The log file is attached. While the results generated by Frank and Kenneth are as follows.

Bitrate: 480.47 PSNR: 30.44 (Y) 35.05 (U) 36.69 (V)

3/8th and 5/8th position coefficients of 6-tap DCT-IF filter have bugs because of typos.

The 6 taps coefficients in TMuC 0.71 with QC_AMVRES on.

{ 11, -42, 196, 118, -36, 10 }, 3/8 { 10, -36, 196, 118, -42, 11 }, 5/8

Sum of the coefficients should be 256, while that of current coefficients is 257.

The correct coefficients should be

{ 11, -43, 196, 118, -36, 10 }, 3/8 { 10, -36, 196, 118, -43, 11 }, 5/8

I noticed that in TComRdCost.cpp, all the functions for SAD,SSE,HAD where bRound is being passed as parameter, xClip() is being used which affects the encoding time. xClip() is redundant step. (255+255+0)>>1 = 255 and (255+255+1)>>1 = 255.

So xClip should be removed from all these functions.

For e.g. So in TMuC0.7.2, in TComRdCost.cpp, line 471 can be changed from pred = xClip( (piCur[n] + piRef[n] + bRound) >> 1 ); to pred = (piCur[n] + piRef[n] + bRound) >> 1 ;

Results should still be bitexact.