IBM MVS mainframes, (ref.time = DAT50) :
- With only single precision values, the 3032 should be equal to the
370/168-3. In fact, it has been known that the difference could go as high
as 15% if the code is 100% floating point & double precision (eg. 64 bits)
values; here we have: 3032/168-3 = 1167/1033 = 113% .
- IBM 3090-180S:
- AIX/370, O/O1/O2/O3 = 74.4/63.6/48.4/48.4
- Metaware cc, noopt/opt = 73.7/70.6
| optimization | none | O1 | O2 | O3 |
| VM/CMS | 79.46 | | | 50.60 |
| AIX/370, fvs -XA | 81.52 | | | 52.90 |
| UTS, Fujitsu-f77 | | 59.25 | 55.02 | 54.33 |
- IBM 3090-200, VSFortran 1.1.0 (Sept.85) = opt=0: 97.26 & opt=3: 68.50
- IBM 3090-200, Fujitsu f77 2.10 v10l20 = opt=2: 71.08 & opt=3: 70.58
- IBM 9000/720, noopt = 75.7, opt(3) novector = 50.2, opt(3) vector(report) = 54.4
- Siemens 7890, Fujitsu f77 2.10 v10l20 = opt=2: 63.48 & opt=3: 63.98
[Return]