On 24.06.2016 16:40, Brian Paul wrote:
On 06/24/2016 03:56 AM, Nicolai Hähnle wrote:
On 24.06.2016 04:07, Brian Paul wrote:
If the user requests an RGB drawing surface but we actually create an
RGBA surface, we need it to act as if A=1. For blending, this means
adjusting the blending terms to use 1/0 instead of
DST_ALPHA/INV_DST_ALPHA.
Drivers can use this flag to determine when that's needed.
Having the bit is fine, but shouldn't there be a PIPE_CAP for it?
No. What would the state tracker do differently if there was a cap?
If there's a driver currently simulating RGB with RGBA they're probably
already doing blending wrong for the cases in question. The new flags
gives the driver a chance of fixing this. It can be ignored otherwise.
>
And keep in mind that this is a corner case that should never come up in
practice: if the app is blending into an RGB surface using blending
terms that involve destination alpha, that's kind of senseless.
The piglit fbo-blending-formats test hits this, however.
Those are passing with radeonsi.
I guess what happens is that RGB gets mapped to RGBX (as per Roland),
and then we do have code in the driver that sets a DST_ALPHA_1 bit in
that case which specifically tells the hardware to treat alpha as being
always one. (So the bit would be trivial to implement for us, but now
that I think about it again, it does seem a bit redundant with the RGBX
formats.)
Nicolai
-Brian
Nicolai
A previous patch I posted last year did this entirely in the state
tracker
but it involved making blend state dependent on the framebuffer state.
This approach avoids that dependency.
---
src/gallium/include/pipe/p_state.h | 1 +
1 file changed, 1 insertion(+)
diff --git a/src/gallium/include/pipe/p_state.h
b/src/gallium/include/pipe/p_state.h
index 9c69355..8b0c3a2 100644
--- a/src/gallium/include/pipe/p_state.h
+++ b/src/gallium/include/pipe/p_state.h
@@ -387,6 +387,7 @@ struct pipe_surface
unsigned height; /**< logical height in pixels */
unsigned writable:1; /**< writable shader resource */
+ unsigned alpha_one:1; /**< Should an RGBA surface should act
like RGB? */
union {
struct {
_______________________________________________
mesa-dev mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/mesa-dev
