[Mesa-dev] [RFC 1/6] dri: Support 64 bit rgba masks

[Strasser, Kevin]

Adam Jackson wrote:
> > On Fri, 2019-01-11 at 15:01 +0000, Emil Velikov wrote:
> > 
> > > > @@ -460,6 +464,14 @@ driGetConfigAttribIndex(const __DRIconfig
> > *config,
> > > >  	else
> > > >  	    *value = 0;
> > > >  	break;
> > > > +    case __DRI_ATTRIB_RED_MASK_HI:
> > > > +    case __DRI_ATTRIB_GREEN_MASK_HI:
> > > > +    case __DRI_ATTRIB_BLUE_MASK_HI:
> > > > +    case __DRI_ATTRIB_ALPHA_MASK_HI:
> > > > +        /* upper 32 bits of 64 bit fields */
> > > > +        *value = *(unsigned int *)
> > > > +            ((char *) &config->modes + attribMap[index].offset +
> > > > + 4);
> > >
> > > Is the "+ 4" going to work on big endian systems?
> > 
> > No.
> > 
> > I think I'd prefer to just expand config attribute values to 64-bit across the board
> > internally, rather than have paired 32-bit attributes like this.

Emil is right, big endian wouldn't work with my patch as-is, but I suppose I
could incorporate some macros to make it work. I experimented with changing
getConfigAttrib/indexConfigAttrib to allow for 64-bit values, like you suggest,
but doing so without breaking ABI got pretty ugly - required adding new
functions (getConfigAttrib2/indexConfigAttrib2), and checking the driver's
extension version level at every call site. However, given the issue with
endianness, I'm now more inclined to go in that direction.

Thanks,
Kevin