Summary of results for:

  Benchmark --all --normalise

OOC2 Benchmarks
---------------

Code is compiled with ooc2 SSA/C output near 24/August 2002. This
includes a new opcode "indexed" as a short form for base+index*scale,
where scale is one of 1,2,4, or 8. On the i386, this makes gcc use
scaled addressing mode.

The changes improve the QSORT test significantly, and also the BSORT
on the Athlon to a certain degree.  For the rest, the difference is
minimal.  

The PPC performance figures are not significantly different. FFT2
appears to run slightly faster. The generated code pattern for BSORT
is close to the C version, and the speedup previously seen on the G4
has gone.

Pentium-II / 400Mhz  (mva)

  Test             tests/sec % optimal
  -------------------------- -------
  QSORT                361.6   99.5%
  QSORTC               363.3
  BSORT                131.3   83.6%
  BSORTC               157.1
  FFT                 1485.4   34.4%
  FFTC                4313.0
  FFT2                2893.4   73.8%
  FFT2C               3920.6
  -------------------------- -------

Athlon XP / 1533Mhz (mva)

  Test             tests/sec % optimal
  -------------------------- -------
  QSORT               1409.9   99.3%
  QSORTC              1420.4
  BSORT                604.9   80.2%
  BSORTC               754.3
  FFT                 7526.8   40.8%
  FFTC               18429.7
  FFT2               13514.7   77.8%
  FFT2C              17377.8
  -------------------------- -------

PowerPC G3 / 700Mhz  (sg)
gcc 2.95.2, -O2

  Test             tests/sec % optimal
  -------------------------- -------
  QSORT                376.0   85.7%
  QSORTC               438.6
  BSORT                153.8  100.0%
  BSORTC               153.8
  FFT                 2517.7   35.1%
  FFTC                7164.8
  FFT2                3294.1   60.0%
  FFT2C               5490.6
  -------------------------- -------

PPC G4 / Dual 800Mhz (sg)
gcc 2.95.2, -O2

  Test             tests/sec % optimal
  -------------------------- -------
  QSORT                501.7   99.2%
  QSORTC               505.9
  BSORT                158.7   99.7%
  BSORTC               159.2
  FFT                 2408.3   35.3%
  FFTC                6815.3
  FFT2                3032.7   60.6%
  FFT2C               5006.6
  -------------------------- -------

OOC1 Benchmarks
---------------
Code is compiled with OOC1 1.5.7.1

On the G3, the SORT benchmarks appear to run at roughly the same speed
for OOC1 and OOC2. Code generated by OOC1 is roughly twice as fast as
OOC2 for FFT (OOC1: 81 percent, OOC2: 35 percent). FFT2 is also better
for OOC1 (OOC1: 101 percent, OOC2: 60 percent) and appears to be
optimal.

The G4 performance shows similar trends. The OOC1 BSORT has a slight
advantage over the C version (104 percent), whereas the OOC2 BSORT
appears to be slightly slower but still good (99.7 percent). FFT2
performance exceeds the C version on the G4.

Pentium-II / 400Mhz  (mva)
gcc-2.95.4, -O2

  Test             tests/sec % optimal
  -------------------------- -------
  QSORT                248.5   68.3%
  QSORTC               364.0
  BSORT                 87.8   55.8%
  BSORTC               157.4
  FFT                 3757.2   87.0%
  FFTC                4318.7
  FFT2                2752.5   70.1%
  FFT2C               3929.0
  -------------------------- -------

Athlon XP / 1533Mhz (mva)
gcc-2.95.4, -O2

  Test             tests/sec % optimal
  -------------------------- -------
  QSORT                917.3   64.9%
  QSORTC              1414.4
  BSORT                374.4   49.7%
  BSORTC               753.6
  FFT                14876.8   80.6%
  FFTC               18466.0
  FFT2               12698.3   72.7%
  FFT2C              17470.4
  -------------------------- -------

PowerPC G3 / 700Mhz  (sg)
gcc 2.95.2, -O2

  Test             tests/sec % optimal
  -------------------------- -------
  QSORT                377.6   86.0%
  QSORTC               438.8
  BSORT                153.7   99.9%
  BSORTC               153.8
  FFT                 5838.4   81.5%
  FFTC                7162.0
  FFT2                5543.6  100.9%
  FFT2C               5491.5
  -------------------------- -------

PPC G4 / Dual 800Mhz (sg)
gcc 2.95.2, -O2

  Test             tests/sec % optimal
  -------------------------- -------
  QSORT                502.8   99.3%
  QSORTC               506.1
  BSORT                158.9  104.2%
  BSORTC               152.5
  FFT                 5221.6   76.6%
  FFTC                6819.6
  FFT2                5536.1  110.6%
  FFT2C               5007.3
 -------------------------- -------
  
