Hello!
Following are the results of povray 3.6.1 official benchmark
(benchmark.ini) on:
vendor_id : AuthenticAMD
cpu family : 15
model : 47
model name : AMD Athlon(tm) 64 Processor 3000+
stepping : 2
cpu MHz : 1809.278
cache size : 512 KB
Base compile flags of "gcc version 4.3.0 20061029" were set to:
-O3 -msse3 -ffast-math -march=k8 -mtune=k8 -minline-all-stringops
Different -mpfmath selections were benchmarked:
a) -mfpmath=sse
user 27m33.848s
b) -mfpmath=387
user 27m42.136s
c) -mfpmath=sse,387
user 26m0.312s
These results were obtained with Richard's SSE rounding functions, top
of ChangeLog was:
2006-10-29 Richard Guenther <[EMAIL PROTECTED]>
* config/i386/i386-protos.h (ix86_expand_trunc): Declare.
Nice to see that there is life in -mfpmath=sse,387 ;) It was faster than
a) and b) by some 6%.
Uros.
