<div dir="ltr">Sorry, I didn't mean to make that much of a stink. Either's fine with me as long as we're not doing 32 for half-float. Reviewed-by: Jason Ekstrand <<a href="mailto:jason.ekstrand@intel.com">jason.ekstrand@intel.com</a>> </div><div class="gmail_extra"> <div class="gmail_quote">On Tue, Aug 12, 2014 at 8:09 AM, Neil Roberts <<a href="mailto:neil@linux.intel.com" target="_blank">neil@linux.intel.com</a>> wrote: <blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">I'm not sure if we came to a decision about what to do with the channel bits. I guess it's not really that important because probably nothing uses it for compressed formats except to check for the existence of the channels. Here is a v2 patch which picks 8/8/8/8 for the RGBA formats and 16/16/16 for the half-float formats. Would that be ok? I guess we should pick a policy and make it consistent for the other formats too but perhaps we can leave that to a later patch. It looks like the NVidia approach might be to return a number of bits that you would need to create a texture to represent the possible generated values from the decompressor. However that doesn't quite add up for the half-float formats because in that case you could accurately represent the values with a half-float texture and only 16 bits. If we wanted to pick that policy we would also have to change what we return for the S3TC formats because they can also return 8 bits of variance. The ETC2 formats already return 8. - Neil ------- >8 --------------- (use git am --scissors to automatically chop here) <div class=""> Adds the ‘bptc’ layout to get_channel_bits. The channel bits for BPTC depend </div>on the mode but as it only has to be an approximation this sets it to 8 for the two UNORM formats and 16 for the two half-float formats. These represent the minimum number of bits of variation that can be generated by the interpolation of the two formats. This doesn't quite match what we do for S3TC which only returns 4 even though it can similarly generate 8 bits from the interpolation. However it does match what we return for ETC2. For reference, NVidia seems to return 8 bits for the UNORM formats and 32 bits for the half-float formats. v2: Change the number of bits to 8/8/8/8 for the UNORM formats and 16/16/16 for the half-float formats. --- src/mesa/main/format_info.py | 3 +++ 1 file changed, 3 insertions(+) diff --git a/src/mesa/main/format_info.py b/src/mesa/main/format_info.py index a0eecd3..7424fe0 100644 --- a/src/mesa/main/format_info.py +++ b/src/mesa/main/format_info.py @@ -127,6 +127,9 @@ def get_channel_bits(fmat, chan_name): bits = 11 if fmat.name.endswith('11_EAC') else 8 return bits if fmat.has_channel(chan_name) else 0 + elif fmat.layout == 'bptc': + bits = 16 if fmat.name.endswith('_FLOAT') else 8 + return bits if fmat.has_channel(chan_name) else 0 else: assert False else: -- 1.9.3 </blockquote></div> </div>