[Mesa-dev] [PATCH 1/2] mesa: fix the format of glEdgeFlagPointer

[Ian Romanick]

On 03/04/2014 01:18 PM, Marek Olšák wrote:
> On Tue, Mar 4, 2014 at 8:05 AM, Ian Romanick <idr at freedesktop.org> wrote:
>> On 03/03/2014 03:25 AM, Marek Olšák wrote:
>>>
>>> From: Marek Olšák <marek.olsak at amd.com>
>>>
>>> Softpipe expects a float in the vertex shader, which is what glEdgeFlag
>>> generates.
>>>
>>> This fixes piglit/gl-2.0-edgeflag.
>>
>>
>> I think this may be a problem elsewhere inside softpipe.  This test passes
>> with or without this patch on i965.  It was also previously passing on i915
>> and r200.  I don't have that hardware with me right now, so I can't test it
>> with the patch.  Does this test fail on Gallium hardware drivers?
>
> As far as I know, the test fails on all Gallium drivers. It couldn't
> have passed on r200, because it doesn't support GL 2.0. The type

Of course. :)  I just looked and didn't see it on the "problems" tab
from a recent piglit run on r200... if it wasn't a problem, it must have
passed, right?

> cannot be integer, because the immediate mode function glEdgeFlag
> emits a float, so if an edge flag array works on i965, glEdgeFlag must
> be broken. I'm just following what glEdgeFlag does. Drivers can emit a

That is possible.  It doesn't look like there are any piglit tests that
use glEdgeFlag with GLSL. :(  It also looks like Eric found where we
were handling glEdgeFlag differently than glEdgeFlagPointer.

http://lists.freedesktop.org/archives/mesa-dev/2014-March/055352.html

So...

Reviewed-by: Ian Romanick <ian.d.romanick at intel.com>

> float-to-int conversion in the shader like r600g and radeonsi already
> do to make glEdgeFlag work.
>
>> The "integer" column was added in OpenGL 3.0 (though the old comment says
>> GL_EXT_gpu_shader4), so maybe this setting should be conditional with the GL
>> version?
>
> A float-to-int conversion in the shader solves this.
>
> Marek