Re:Bitcoin Core Development Status Report #2
by eduardordm on Friday December 21, 2012 @07:53PM
(#4955328)
Even if RVDS could do that like gcc or vcc, it would make any operations at least 60x slower than just using the add instruction directly. Unlike IA32, 32 bits ARM processors have only 32 bits addr space and registers (against 36 bit in IA32 which allows some 64 bit support) and have no support for 64 bit operations. You are confusing compiler wizardry and the actual program.
That said, NO you should never do that.
Re:Bitcoin Core Development Status Report #2
by gizmo686 on Saturday December 22, 2012 @01:56AM
(#4956009)
Where did you get the 60x number from? Using the proccessor's 32-bit primitive operations, you can treat 64-bit arithmatic as 2-digit arithmetic. If you take advantage of the carry registers, it seems like you should have no more than about 3x slow down.
Re:Bitcoin Core Development Status Report #2
by eduardordm on Saturday December 22, 2012 @09:27AM
(#4956621)
The sheer number of instructions generated by vcc.
Err, you can use 64-bit integers on 32-bit systems. Or 16-bit systems, or 8-bit systems. Just because it's not the native size for integer math doesn't mean you can't use it.