Avoid branch by adding xfer_align to cf->buff

[AtariDreams]

We can avoid a branch by not comparing with 0.

[jsonn]

Are you sure?

[AtariDreams]

Are you sure?

Yes

[kientzle]

This is more efficient than comparing with 0.

Do you have any benchmark numbers to help support this claim? I'd be surprised if this produced a noticeable difference in overall libarchive performance. When I've done performance profiling in the past, the biggest performance bottlenecks were system calls.

[AtariDreams]

This is more efficient than comparing with 0.

Do you have any benchmark numbers to help support this claim? I'd be surprised if this produced a noticeable difference in overall libarchive performance. When I've done performance profiling in the past, the biggest performance bottlenecks were system calls.

I applied your suggestion. Thanks! Though it may not have a noticeable effect on performance, I do think this code is better written and more straightforward to other programmers who may see this code.

[kientzle]

It's possible to avoid the branch above if you assume xfer_align is a power of two. However, the change you've made here actually wastes xfer_align bytes in the case where the original allocation is already aligned (which should be the common case).

[kientzle]

Oh. I just realized that you can do this correctly, but you need an additional modulo operation:

cf->buff = cf->allocation_ptr + ((xfer_align - s) % xfer_align)