[Mesa-dev] [PATCH] anv: don't skip the VUE header if we are reading gl_Layer in a fragment shader

[Iago Toral]

On Thu, 2017-01-05 at 14:08 +0100, Iago Toral Quiroga wrote:
> This is the same we do in the GL driver: the hardware provides
> gl_Layer
> in the VUE header, so when the fragment shader reads it we can't skip
> it.

Forgot to add that this fixes the following Vulkan CTS tests:

dEQP-VK.geometry.layered.1d_array.fragment_layer
dEQP-VK.geometry.layered.2d_array.fragment_layer
dEQP-VK.geometry.layered.cube.fragment_layer

> ---
> 
> With this patch we now successfully read gl_Layer in fragment
> shaders. Layered
> rendering still does not work though, probably because we still need
> to hook up
> the layer_id stuff that Jason added some time ago. I'll look into
> that next.

And by this I mean that the dEQP-VK.geometry.layered.cube_array.* tests
still fail.

>  src/intel/vulkan/genX_pipeline.c | 20 ++++++++++++++++----
>  1 file changed, 16 insertions(+), 4 deletions(-)
> 
> diff --git a/src/intel/vulkan/genX_pipeline.c
> b/src/intel/vulkan/genX_pipeline.c
> index 845d020..c1d8ae6 100644
> --- a/src/intel/vulkan/genX_pipeline.c
> +++ b/src/intel/vulkan/genX_pipeline.c
> @@ -291,6 +291,8 @@ emit_3dstate_sbe(struct anv_pipeline *pipeline)
>  #  define swiz sbe
>  #endif
>  
> +   /* Skip the VUE header and position slots by default */
> +   unsigned urb_entry_read_offset = 1;
>     int max_source_attr = 0;
>     for (int attr = 0; attr < VARYING_SLOT_MAX; attr++) {
>        int input_index = wm_prog_data->urb_setup[attr];
> @@ -298,6 +300,12 @@ emit_3dstate_sbe(struct anv_pipeline *pipeline)
>        if (input_index < 0)
>           continue;
>  
> +      /* gl_Layer is stored in the VUE header */
> +      if (attr == VARYING_SLOT_LAYER) {
> +         urb_entry_read_offset = 0;
> +         continue;
> +      }
> +
>        if (attr == VARYING_SLOT_PNTC) {
>           sbe.PointSpriteTextureCoordinateEnable = 1 << input_index;
>           continue;
> @@ -322,18 +330,22 @@ emit_3dstate_sbe(struct anv_pipeline *pipeline)
>           swiz.Attribute[input_index].ComponentOverrideZ = true;
>           swiz.Attribute[input_index].ComponentOverrideW = true;
>        } else {
> -         assert(slot >= 2);
> -         const int source_attr = slot - 2;
> -         max_source_attr = MAX2(max_source_attr, source_attr);
>           /* We have to subtract two slots to accout for the URB
> entry output
>            * read offset in the VS and GS stages.
>            */
> +         assert(slot >= 2);
> +         const int source_attr = slot - 2 * urb_entry_read_offset;
> +         max_source_attr = MAX2(max_source_attr, source_attr);
>           swiz.Attribute[input_index].SourceAttribute = source_attr;
>        }
>     }
>  
> -   sbe.VertexURBEntryReadOffset = 1; /* Skip the VUE header and
> position slots */
> +   sbe.VertexURBEntryReadOffset = urb_entry_read_offset;
>     sbe.VertexURBEntryReadLength = DIV_ROUND_UP(max_source_attr + 1,
> 2);
> +#if GEN_GEN >= 8
> +   sbe.ForceVertexURBEntryReadOffset = true;
> +   sbe.ForceVertexURBEntryReadLength = true;
> +#endif
>  
>     uint32_t *dw = anv_batch_emit_dwords(&pipeline->batch,
>                                          GENX(3DSTATE_SBE_length));