RE: [PATCH v4 1/4] fpga: dfl: change data type of feature id to u16

[David Laight <David.Laight@ACULAB.COM>]

From: Xu Yilun <yilun.xu@intel.com>
> Sent: 13 August 2020 10:04
> 
> On Thu, Aug 13, 2020 at 08:28:05AM +0000, David Laight wrote:
> > From: Xu Yilun
> > > Sent: 13 August 2020 08:59
> > > On Wed, Aug 12, 2020 at 08:52:39AM +0000, David Laight wrote:
> > > > From: Moritz Fischer
> > > > > Sent: 12 August 2020 04:56
> > > > >
> > > > > On Mon, Aug 10, 2020 at 10:41:10AM +0800, Xu Yilun wrote:
> > > > > > The feature id is stored in a 12 bit field in DFH. So a u16 variable is
> > > > > > enough for feature id.
> > > > > >
> > > > > > This patch changes all feature id related places to fit u16.
> > > >
> > > > How much bigger does it make the kernel?
> > >
> > > The patch changes the definition of feature id from u64 to u16, and will
> > > make the kernel slightly smaller.
> >
> > Unlikely.
> > Most of the structures will gain a 'pad' field.
> > Using u16 for function parameters and results almost certainly
> > requires instructions to mask the value.
> > Any arithmetic on u16 will require masking instructions on
> > (probably) all architectures except x86.
> >
> > Using 'unsigned int' is probably best.
> >
> > u16 is never a good idea unless you are defining enough
> > of them in a structure (eg as an array) to reduce the
> > structure size below some threshold.
> > (Or are matching some hardware layout.)
> 
> I got it. Thanks for your detailed explanation. I think we may change them to
> u32. Is it the same case for u8? Think we may also change the dfl_device_id.type.

Loosely 'yes' but it isn't worth the churn of 'random' changes.
And they aren't often passed to/from functions - which I'm 98% sure
requires masking.

I commented because the compiler was going to add pad fields after
your u16 values - so you'd get do space saving and probably more code.

	David

-
Registered Address Lakeside, Bramley Road, Mount Farm, Milton Keynes, MK1 1PT, UK
Registration No: 1397386 (Wales)