The below table shows median performance of Blitz++ on
21 loop kernels, relative to the best native Fortran compiler
with typical optimization switches (-O3, -Ofast):
Platform |
Compiler |
Out-of-cache |
In-cache |
Cray T3E | KAI C++ | 95.7% | 98.1% |
HPC-160 | KAI C++ | 100.2% | 97.5% |
Origin 2000 | KAI C++ | 88.1% | 79.8% |
Origin 2000 | SGI C++ 7.3 | 88.1% | 97.1% |
Pentium II | egcs | 98.4% | 79.6% |
RS 6000 | KAI C++ | 93.5% | 97.3% |
UltraSPARC | KAI C++ | 91.1% | 79.0% |
DEC Alpha | DECcxx | 98.0% | 69.0% |
Most of these benchmark results were measured on a 100 MHz RS/6000 using
KAI C++ at +K3 -O3. Fortran examples were compiled with XL Fortran 77
and XL Fortran 90 using -O3.