-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1 Brian Paul wrote: | Brian Paul wrote: |> Ian Romanick wrote: |>> |>> Currently in Gallium no state object is bound for blend state if blend |>> has not been enabled. Or is it just if none of the blend state is |>> modified? Either way, I have a hack in the current Cell driver that |>> sets some default state on the SPE if cell_bind_blend_state is never |>> called. |>> |>> What would it take to change this? I'd prefer it if some blend state |>> object was always bound. I haven't dug around in the Gallium code |>> outside the Cell driver, so I don't know where the right place is to do |>> this. |> |> Ian, can you point me at a particular program to test? |> |> I just tried a few tests w/ gdb and cell_bind_blend_state() is getting |> called with a non-null blend state before anything's drawn. Looks OK |> here.
That seems to be the case here as well. I could swear that it wasn't working correctly a few days ago. | Slightly off topic... I've been off the cell project for a while and | haven't been following your code-gen work. But I took a quick look at | things. Looks nice, but one thing: the intention is that code-gen | should be done at 'create_XXX' time, rather than at 'bind_XXX' time, | whenever possible. | | For example, can cell_generate_alpha_blend() be moved into | cell_create_blend_state()? Done for both blend and depth test. At this point alpha / stencil / depth testing is complete. Alpha blending and color mask are also complete. However, logic op is not complete. Right now blending is performed in floating point. This seemed the better route for when floating-point framebuffer support is added. However, logic op expects to operate on colors in framebuffer format. That makes the logic op code implicitly dependent on the framebuffer configuration. I'm still thinking about the right way to implement this within the driver. After that, I have two things at the top of my to-do list: 1. Implement xshm support. 2. Better connection between the SPE vertex processor and the SPE rasterizer. The extra round-trip through main memory is just silly. I expect this to be a fairly significant piece of work. There are some bugs in the current Cell code that I don't know how to approach. I've noticed that a *LOT* of the demos hit assertions about the vertex size not being a multiple of 16 bytes. -----BEGIN PGP SIGNATURE----- Version: GnuPG v1.4.7 (GNU/Linux) iD8DBQFH5AmFX1gOwKyEAw8RApvwAJ9a5pcd6+ir69S0GmW1bHfHu1C+LgCeNcwz pcFQxIxclim+Hkgd2tP22jY= =Ar2R -----END PGP SIGNATURE----- ------------------------------------------------------------------------- This SF.net email is sponsored by: Microsoft Defy all challenges. Microsoft(R) Visual Studio 2008. http://clk.atdmt.com/MRT/go/vse0120000070mrt/direct/01/ _______________________________________________ Mesa3d-dev mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/mesa3d-dev
