[Mesa-dev] [PATCH 1/2] radeonsi: fix viewport clipping handling.

[Nicolai Hähnle]

On 18.12.2015 21:16, Dave Airlie wrote:
> From: Dave Airlie <airlied at redhat.com>
>
> If oViewport is written, vertex reuse need to be turned off.
> If oViewport is constant, vertex reuse is fine, and VPORT_PROVOKE_DISABLE
> need to be set.
>
> Fixes: arb_viewport_array-render-viewport-2 and some CTS tests.
>
> Signed-off-by: Dave Airlie <airlied at redhat.com>
> ---
>   src/gallium/drivers/radeonsi/si_state.c | 6 ++++++
>   1 file changed, 6 insertions(+)
>
> diff --git a/src/gallium/drivers/radeonsi/si_state.c b/src/gallium/drivers/radeonsi/si_state.c
> index 4086819..e3bce12 100644
> --- a/src/gallium/drivers/radeonsi/si_state.c
> +++ b/src/gallium/drivers/radeonsi/si_state.c
> @@ -586,6 +586,7 @@ static void si_emit_clip_regs(struct si_context *sctx, struct r600_atom *atom)
>   	unsigned clipdist_mask =
>   		info->writes_clipvertex ? SIX_BITS : info->clipdist_writemask;
>
> +	/* vport provoke should be set to 1 if oViewport isn't written */

The comment feels misleading to me. vport provoke should be disabled if 
oViewport isn't written, or perhaps vport provoke _disable_ should be 
set to 1.

>   	radeon_set_context_reg(cs, R_02881C_PA_CL_VS_OUT_CNTL,
>   		S_02881C_USE_VTX_POINT_SIZE(info->writes_psize) |
>   		S_02881C_USE_VTX_EDGE_FLAG(info->writes_edgeflag) |
> @@ -604,7 +605,12 @@ static void si_emit_clip_regs(struct si_context *sctx, struct r600_atom *atom)
>   		sctx->queued.named.rasterizer->pa_cl_clip_cntl |
>   		(clipdist_mask ? 0 :
>   		 sctx->queued.named.rasterizer->clip_plane_enable & SIX_BITS) |
> +		S_028810_VTE_VPORT_PROVOKE_DISABLE(!info->writes_viewport_index) |
>   		S_028810_CLIP_DISABLE(window_space));
> +
> +	/* reuse needs to be set off if we write oViewport */
> +	radeon_set_context_reg(cs, R_028AB4_VGT_REUSE_OFF,
> +			       S_028AB4_REUSE_OFF(info->writes_viewport_index));
>   }
>
>   static void si_set_scissor_states(struct pipe_context *ctx,
>

Now that VGT_REUSE_OFF is also programmed in this atom, it should be 
possible to remove it from si_init_config, right?

Cheers,
Nicolai