Messages in this thread | | | Date | Wed, 1 Feb 2006 14:09:03 -0800 | From | Grant Grundler <> | Subject | Re: [PATCH 1/12] generic *_bit() |
| |
On Wed, Feb 01, 2006 at 01:41:03PM -0800, Chen, Kenneth W wrote: > > Well, if it doesn't matter, why is unsigned int better? > > I was coming from the angle of having bitop operate on unsigned > int *, so people don't have to type cast or change bit flag variable > to unsigned long for various structures. With unsigned int type for > bit flag, some of them are not even close to fully utilized. for example: > > thread_info->flags uses 18 bits > thread_struct->flags uses 7 bits > > It's a waste of memory to define a variable that kernel will *never* > touch the 4 MSB in that field.
Agreed. Good point. But this can be mitigated if the code using "unsigned int" (or unsigned byte) first loads the value into a local unsigned long variable. That typically translates into a tmp register anyway. Compiler will help you find places where that needs to happen.
Counter point is bit arrays (e.g. bit maps) like cpumask_t are typically much larger than 32-bits (typically distro's ship with NR_CPUS set to 256 or so). File system code also likes bit arrays for block allocation tables. Searching a bit array using unsigned long is 2x faster on 64-bit architectures. I don't want to give that up and I'm pretty sure Tony Luck, Paul Mckerras and a few others would object unless you can give a better reason.
Obviously neither memory footprint nor speed of walking memory is an the issue for 32-bit arches (where unsigned long == unsigned int).
> > unsigned long is typically the native register size, right? > > I'd expect that to be more efficient on most arches. > > The only difference that I can think of on Itanium processor is the > memory operation, you either load/store 4 or 8 bytes. Once the data > is in the CPU register, it doesn't make any difference whether it is > operating on 32bit or entire 64 bit. I don't know about others RISC > arch though whether it is more efficient with native register size.
agreed. I was thinking mostly of the bit map search - not searching within a single unsigned int.
grant - To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
| |