[Mesa-dev] [PATCH] radeonsi/gfx9: use the correct condition for the DPBB + QUANT_MODE workaround

[Samuel Pitoiset]

On 4/18/19 4:17 AM, Marek Olšák wrote:
> From: Marek Olšák <marek.olsak at amd.com>
>
> ---
>   src/gallium/drivers/radeonsi/si_state_viewport.c | 3 ++-
>   1 file changed, 2 insertions(+), 1 deletion(-)
>
> diff --git a/src/gallium/drivers/radeonsi/si_state_viewport.c b/src/gallium/drivers/radeonsi/si_state_viewport.c
> index 1ec69216841..83905d36ee6 100644
> --- a/src/gallium/drivers/radeonsi/si_state_viewport.c
> +++ b/src/gallium/drivers/radeonsi/si_state_viewport.c
> @@ -359,21 +359,22 @@ static void si_set_viewport_states(struct pipe_context *pctx,
>   		max_extent += distance_off_center;
>   
>   		/* Determine the best quantization mode (subpixel precision),
>   		 * but also leave enough space for the guardband.
>   		 *
>   		 * Note that primitive binning requires QUANT_MODE == 16_8 on Vega10
>   		 * and Raven1. What we do depends on the chip:
>   		 * - Vega10: Never use primitive binning.

Update the comment? Primitive binning can be enabled with an envvar.

With that, patch is:

Reviewed-by: Samuel Pitoiset <samuel.pitoiset at gmail.com>

>   		 * - Raven1: Always use QUANT_MODE == 16_8.
>   		 */
> -		if (ctx->family == CHIP_RAVEN)
> +		if ((ctx->family == CHIP_VEGA10 || ctx->family == CHIP_RAVEN) &&
> +		    ctx->screen->dpbb_allowed)
>   			max_extent = 16384; /* Use QUANT_MODE == 16_8. */
>   
>   		/* Another constraint is that all coordinates in the viewport
>   		 * are representable in fixed point with respect to the
>   		 * surface origin.
>   		 *
>   		 * It means that PA_SU_HARDWARE_SCREEN_OFFSET can't be given
>   		 * an offset that would make the upper corner of the viewport
>   		 * greater than the maximum representable number post
>   		 * quantization, ie 2^quant_bits.