Comparison of forward_DCT accuracy
----------------------------------

Test parameters : ieee1180.exe -256 255 1 10000

Test procedure :

           |
  test-data| ---> fdct_reference -> idct_reference -> reference_output
           | ---> fdct_testfunc  -> idct_reference -> test_output
           |

  ERROR_CALCULATION< reference_output vs test_output >

----------------------------------------------------------------------
tabulated results...


                Partial Mean     |                 |
                Square Error     |   MEAN_ERROR    |      PEAK-ERROR
FDCT_TYPE       WORST   OVERALL  | WORST   OVERALL |   WORST (#elements)
---------------------------------+-----------------+---------------
(1)x87d-AAN     0.0229  *0.0164  | 0.0229  0.0049  |    +/- 1
(2)3dnow-AAN    0.0231  *0.015   | 0.0229  0.0050  |    +/- 1
                                 |                 |
(3)fdctam32     0.0595   0.0538  | 0.0245  0.0051  |    +/- 2 (1)
(4)fdctmm32     0.0814   0.0696  | 0.0263  0.0052  |    +/- 2 (1)
(5)int32-AAN    0.4455   0.1194  | 0.4377  0.0074  |    +/- 2 (13)

* With different test parameters, the x87d-AAN PMSE_OVERALL is less
  than 3DNow-AAN PMSE_OVERALL.

For all error categories, lower #s represent better accuracy.
Worst peak-error shows

Legend :

(1) x87d-AAN  {fdctref.c } - (BBMPEG) double-precision AAN
(2) 3dnow-AAN {fdct3dn.c } - 3D-Now single-precision AAN
(3) fdctam32  {fdctam32.c} - Intel AP-922 MMX(3DNow enhanced with PMULHRW)
(4) fdctmm32  {fdctmm32.c} - Intel AP-922 standard MMX
(5) int32-AAN {fdctint.c } - integer32 AAN (JPEG/MPEG codec)
