bearophile wrote:
Lot of time ago I have instead suggest BigInt.sizeof == size_t.sizeof, because 
in most programs you need integers smaller than 63 bit, 31 bits are enough. 
This increases the performance a little on 32 bit systems.

The compiler also has to make sure the tests for out-of-smallint-range are 
always inlined, otherwise you are mostly wasting your time optimizing BigInt 
for small integers.

Bye,
bearophile

The big savings come from avoiding memory allocation.

Reply via email to