On Fri, Nov 18, 2022 at 01:23:36AM +0100, Andrew Lunn wrote:
> > struct vchiq_mmal_component {
> > - u32 in_use:1;
> > + bool in_use:1;
> > bool enabled:1;
>
> The patch you referenced says:
>
> +If a structure has many true/false values, consider consolidating them into a
> +bitfield with 1 bit members, or using an appropriate fixed width type, such as
> +u8.
>
> The code did exactly this, using two bits fields, in one u32. A bool
> probably takes up 4 bytes, maybe 8 bytes, so this change probably
> doubles the storage size for these two fields.
In GCC and Clang bools take a byte, but the C language is vague and
other compilers are free to do it differently.
> Are these fields on the
> hot path, where an extra AND instruction would make a difference?
This patch takes the first u32 for "in_use" and squeezes it into the
same byte as "enabled" so it makes the struct four bytes smaller. There
is still a 3 byte struct hole between "enabled" and "handle" so we could
add more 62 bool bitfields if we wanted.
In the v2 patch these become:
bool in_use;
bool enabled;
One byte each and there is a two byte gap before "handle".
regards,
dan carpenter
