Some unscientific benchmarks on RISC OS 3.71 under RPCEmu

Soruk
Soruk
edited January 2022 in Brandy BASIC including Matrix Brandy
riscos-benchmark.png

I also ran some comparisons with the Sphere program included with PiTubeDirect (slightly edited, only to run in MODE 1 instead of MODE 5)...

First, under ARM BBC BASIC V:

sphere.armbasic5.png

Then, under ARM BBC BASIC VI - this is surprisingly slow compared to BASIC V:

sphere.armbasic6.png

Matrix Brandy compiled against the standard SharedCLibrary is also quite slow:

sphere.brandyclib.png

But the big surprise here is the UnixLib build, which compares quite favourably against ARM BBC BASIC VI - indeed in this program it is actually slightly faster, not a bad feat for compiled code against hand-written ARM assembly!

sphere.brandyulib.png


Comments

  • Soruk
    Soruk

    Of course, the slowness of the BASIC VI build using FPA is down to the RiscPC not having a floating point processor so the FPEmulator module has to not only go through a hardware trap but then emulate the instructions - as does the SharedCLibrary build of Matrix Brandy. By comparison, the UnixLib build just assumes floating point (but can me made to compile with FPA or even VFP (for the RasPi)) so the hardware trap isn't tripped.

Categories