<div dir="ltr">I think an indicator for float can double as indicating the size is multiplied by 8. All float formats I am familiar with take a mulitple of 8 bits, including an 8-bit format that is sometimes used in CG.<div> </div></div> <div class="gmail_quote"><div dir="ltr">On Mon, Oct 29, 2018 at 2:18 AM Maarten Lankhorst <<a href="mailto:maarten.lankhorst@linux.intel.com">maarten.lankhorst@linux.intel.com</a>> wrote: </div><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">Op 29-10-18 om 09:16 schreef Siarhei Siamashka: > On Wed, 03 Oct 2018 10:11:46 +0100 > Chris Wilson <<a href="mailto:chris@chris-wilson.co.uk" target="_blank">chris@chris-wilson.co.uk</a>> wrote: > >> Quoting Maarten Lankhorst (2018-08-01 13:41:33) >>> diff --git a/pixman/pixman.h b/pixman/pixman.h >>> index 509ba5e534a8..c9bf4faa80d4 100644 >>> --- a/pixman/pixman.h >>> +++ b/pixman/pixman.h >>> @@ -647,19 +647,24 @@ struct pixman_indexed >>> * sample implementation allows only packed RGB and GBR >>> * representations for data to simplify software rendering, >>> */ >>> -#define PIXMAN_FORMAT(bpp,type,a,r,g,b) (((bpp) << 24) | \ >>> +#define PIXMAN_FORMAT_PACKC(val) ((val) <= 10 ? (val) : \ >>> + (val) == 32 ? 11 : 0) >>> +#define PIXMAN_FORMAT_UNPACKC(val) ((val) <= 10 ? (val) : \ >>> + (val) == 11 ? 32 : 0) >> We have 4bits, why not reserve 0xf for float32? Certainly you >> want space for >> >> #define PIXMAN_FORMAT_PACKED_FLOAT16 0xb >> #define PIXMAN_FORMAT_PACKED_FLOAT32 0xc >> #define PIXMAN_FORMAT_PACKED_FLOAT64 0xd >> >> I just wonder if we may have 12bpc one day as well, so leaving 0xc clear >> has some some appeal. > We could probably also try to do something like this: > > /* > * While the protocol is generous in format support, the > * sample implementation allows only packed RGB and GBR > * representations for data to simplify software rendering, > * > * Bit 23 selects the granularity of channel color depth > * 0: 1-bit granularity (allows up to 15 bits per channel) > * 1: 1-byte granularity (allows up to 120 bits per channel) > */ > #define PIXMAN_FORMAT(bpp,type,a,r,g,b) \ > (((bpp) << 24) | \ > ((type) << 16) | \ > (((a) < 16) ? ((a) << 12) : (((a / 8) << 12) | (1 << 23))) | \ > (((r) < 16) ? ((r) << 8) : (((r / 8) << 8) | (1 << 23))) | \ > (((g) < 16) ? ((g) << 4) : (((g / 8) << 4) | (1 << 23))) | \ > (((b) < 16) ? ((b)) : (((b / 8)) | (1 << 23)))) > > /* 0 for 1-bit granularity and 3 for 1-byte granularity */ > #define PIXMAN_FORMAT_CHANDEPTH_SHIFT(f) \ > ((((f) >> 23) & 1) | ((((f) >> 23) & 1) << 1)) I would use 2 bits then, 6 is still plenty for the rest of the type. Perhaps make a separate PIXMAN_FORMAT_TYPE_RGBA_FLOAT and PIXMAN_FORMAT_TYPE_RGB_FLOAT? > #define PIXMAN_FORMAT_A(f) \ > ((((f) >> 12) & 0x0f) << PIXMAN_FORMAT_CHANDEPTH_SHIFT(f)) > #define PIXMAN_FORMAT_R(f) \ > ((((f) >> 8) & 0x0f) << PIXMAN_FORMAT_CHANDEPTH_SHIFT(f)) > #define PIXMAN_FORMAT_G(f) \ > ((((f) >> 4) & 0x0f) << PIXMAN_FORMAT_CHANDEPTH_SHIFT(f)) > #define PIXMAN_FORMAT_B(f) \ > ((((f) ) & 0x0f) << PIXMAN_FORMAT_CHANDEPTH_SHIFT(f)) > > #define PIXMAN_FORMAT_BPP(f) (((uint32_t)(f)) >> 24) > #define PIXMAN_FORMAT_TYPE(f) (((f) >> 16) & 0x7f) > #define PIXMAN_FORMAT_RGB(f) (((f) ) & 0xfff) > #define PIXMAN_FORMAT_VIS(f) (((f) ) & 0xffff) > #define PIXMAN_FORMAT_DEPTH(f) (PIXMAN_FORMAT_A(f) + \ > PIXMAN_FORMAT_R(f) + \ > PIXMAN_FORMAT_G(f) + \ > PIXMAN_FORMAT_B(f)) > > > Right now the format type occupies 8 bits. The highest bit could be > repurposed as a flag for storing channel depth as bytes instead > of bits. This way 16-bit and 64-bit per color component will be > also supported. The limitation of this approach is that the depth > of every color component has to be a multiple of 8 bits if any of > them is 16 bits or larger. > > I don't feel very comfortable about the fact that some applications > are using the PIXMAN_FORMAT_A/R/G/B macros and this code gets compiled > as part of these application binaries. > > Are there any other interesting >32bpp formats that we may need > to support in the future? > Doubt it, pixman doesn't have that accuracy currently, but F16 might be interesting at some point.. ~Maarten _______________________________________________ Pixman mailing list <a href="mailto:Pixman@lists.freedesktop.org" target="_blank">Pixman@lists.freedesktop.org</a> <a href="https://lists.freedesktop.org/mailman/listinfo/pixman" rel="noreferrer" target="_blank">https://lists.freedesktop.org/mailman/listinfo/pixman</a> </blockquote></div>