hph (10/28/82)
I do not doubt that the 8086 running
Pascal beat the 68000 running Pascal.
It is pretty widely known that Motorala's Pascal is a very poor compiler whereas
Intel's is much better. Benchmarks on
machines, when done in higher level
languages should take the compilers
into account.
Paul H.pascal (10/28/82)
Benchmarks including UCSD rarely turn off range checking. The impact can be rather significant. Softech's new native code generator does an impressive job. We are finding 15 to 1 reductions in run time. So far it is only available for the 8086/8088, Z80, 8080. 68K versions coming soon. No word on a 6502 version yet. 8086/68000 comparisons are caught up in a number of issues. Linear address spaces are great for single task systems. A segment system makes multi-tasking a whole lot easier. Second item worth noting about the benchmarks floating around is that they avoid floating point and can be manipulated around the existence or performance of one or two operators. For example a 32x16 divide may take an order of magnitude longer on one processor compared to another when most of the other instructions are similar. The 8/16 issue is frequently obscured in that few assemblers provide an even or align instruction on the 8086 and on average tables are on odd byte boundaries in many of the programs. In bringup UCSD on the 8086 we ran into far more questions of performance in the disk bios than anywhere else. All the stuff we have run to date suffers far more from i/o time than processor speedJim T.