On 10/18/2016 05:50 PM, Kenneth Graunke wrote:
On Tuesday, October 18, 2016 4:38:17 PM PDT Brian Paul wrote:
I found that this patch causes a regression. There's a Windows medical
app which fails to link some shaders since this change.
Basically, when the gl_Position VS input is declared as invariant the
linker fails with:
error: declarations for uniform `gl_ModelViewProjectionMatrix' have
mismatching invariant qualifiers
I haven't investigated how to fix this. I'm hoping you can see a simple
The attached piglit shader_runner script demonstrates the issue. Passes
Oh, sorry about that! Here are two possible fixes:
They're both kind of hacks...but the whole invariant propagation pass
is kind of a hack, and we've got some other hacks in place already.
So...maybe best to pile another one on. Not sure which though.
Maybe Jason or Curro will have an opinion...
Thanks! Either patch is OK with me (though, I'd suggest putting a
comment on the first one to explain what's happening).
It'd be great if we can commit one or the other in the next day or so.
mesa-dev mailing list