Handling odd-sized integers

I have a horrible binary file format, which does have a defined format, and which can contain several odd types of data.

I have loaded the entire file to memory as uint8 data, as the code needs to work on remotely-networked servers, and this is faster than repeated disk accesses.

I have 4 bit, 3 bit, 2 bit and 1 bit integer data fields which may be signed. A single byte can contain two or more fields.

e.g. Field 1: 2 bits, Field 2: 2 bits, Field 3: 4 bits

Currently, for each field, I read the entire byte, right shift it (bitsra) to make the relevant bits the LSB's, use bitand to eliminate the upper, unused bits. If the field is signed, I check the uppermost bit in my valid area, and then set all higher bits equal to the same value using bitor, then I typecast it into int8.

I have a similar process using 2 and 4 bytes for the 24 bit and 12 bit integer fields.

Are there any faster bit operation functions than bitand/bitor/bitsra?

[dpb]

AFAIK, no.

If this is a continuing saga, sounds like mex-file time encapsulated in a class, perhaps.

Obvious application for C struct w/ bitfield and union.

[Walter Roberson]

If you are doing large numbers of these in the same way, then using vectorized arithmetic operations would probably be faster than the bit* operations.

[Iain]

bitsra is massively slower than the straightforward division I thought it would be faster than. bitand/bitor do seem to be quick.