Hardware: Broadwell or Skylake
Mesa commit: f940b1665a7f17ad2ae7ae2e951d90d151482875
vk-gl-cts commit: dfcb8e870438f6f2bfe71d4bb63d43120debb3a3
The following ARB_shader_draw_parameters tests fail:
- Expected outcome:
The tests expect gl_BaseVertex to be zero when the draw command does not accept a <basevertex> parameter.
From the ARB_shader_draw_parameters specification:
"<gl_BaseVertexARB> holds the integer value passed to the <baseVertex>
parameter to the command that resulted in the current shader invocation.
In the case where the command has no <baseVertex> parameter, the value
of <gl_BaseVertexARB> is zero."
- Actual outcome:
Although non-indexed draw commands like gl*DrawArrays* do not accept <basevertex> parameters, Mesa sets the value of gl_BaseVertex to the value passed to the <first> parameter of the gl*DrawArrays* command instead of setting it to zero.
Indexed draw calls like glDrawElements which do not accept a <basevertex> parameter work as expected, gl_BaseVertex is zero in that case.
I believe Antia is working on a fix.
(In reply to Kenneth Graunke from comment #1)
> I believe Antia is working on a fix.
I have a patch that fixes the tests. I am currently testing for regressions, if there isn't any I will send it shortly.
The patch also fixes:
which was failing for the same reason. This failure was unnoticed when I filled the bug because at the time we were not supporting ARB_indirect_parameters.
(In reply to Antia Puentes from comment #2)
> (In reply to Kenneth Graunke from comment #1)
> > I believe Antia is working on a fix.
> I have a patch that fixes the tests. I am currently testing for regressions,
> if there isn't any I will send it shortly.
The patch produced regressions related to gl_VertexID. Specifically in the tests which involve gl*DrawArrays*, pass a value different from zero as argument of the <first> parameter of the draw command, and check if gl_VertexID is correct.
According to the spec, for non-indexed draw calls: "The vertex ID of the ith element transferred is <first> + i."
Mesa's compiler lowers the gl_VertexID as gl_VertexID(base zero) + gl_BaseVertex, which is the way expected for indexed draw calls. For non-indexed draw calls, this currently works because we use the <first> argument as gl_BaseVertex. If we give gl_BaseVertex a zero value for gl*DrawArrays* as the spec says, the gl_VertexID will be incorrect.
Not sure how to fix the gl_VertexID issue, it looks like the i965 does not have a way to add a base value during the vertex-id generation.
Ian, any thoughts? This just feels wrong at this point...I thought the intention was that gl_VertexID - gl_BaseVertex would be the zero-based vertex ID.
But if gl_VertexID is <first + i>, and gl_BaseVertex is required to be 0 for DrawArrays, then...there is no GLSL variable to get <first>, so there is no way to get the zero-based index.
It seems like gl_BaseVertex really ought to be <first> for DrawArrays.
glDrawArrays behaves as if you did
for (i = 0; i < count; i++)
indices[i] = first + count;
glDrawElements(mode, count, GL_UNSIGNED_INT, indices);
The language in the spec is a bit more complicated than that, IIRC, but it should be equivalent. I think gl_BaseVertex=0 and gl_VertexID=first+i is consistent with that.
Author: Antia Puentes <email@example.com>
Date: Sat Apr 28 14:09:22 2018 +0200
intel: activate the gl_BaseVertex lowering
Surplus code related to the basevertex is removed.
The Vertex Elements contain now:
* VE 1: <firstvertex, BaseInstance, VertexID, InstanceID>
* VE 2: <DrawID, is_indexed_draw, 0, 0>
Also fixes unreachable message.
Fixes OpenGL CTS tests:
Fixes Piglit tests:
* arb_shader_draw_parameters-drawid-indirect baseinstance
Reviewed-by: Jason Ekstrand <firstname.lastname@example.org>