[Mesa-dev] [PATCH 3/4] radeonsi: don't use the SPI barrier management bug workaround

[Samuel Pitoiset]

On 03/27/2018 02:39 AM, Marek Olšák wrote:
> From: Marek Olšák <marek.olsak at amd.com>
> 
> ---
>   src/gallium/drivers/radeonsi/si_shader.c | 5 +++++
>   1 file changed, 5 insertions(+)
> 
> diff --git a/src/gallium/drivers/radeonsi/si_shader.c b/src/gallium/drivers/radeonsi/si_shader.c
> index 8ae742c93f6..00ebbb9b0f2 100644
> --- a/src/gallium/drivers/radeonsi/si_shader.c
> +++ b/src/gallium/drivers/radeonsi/si_shader.c
> @@ -7964,20 +7964,25 @@ static bool si_shader_select_ps_parts(struct si_screen *sscreen,
>   	if (!shader->key.part.ps.epilog.poly_line_smoothing &&
>   	    !shader->selector->info.reads_samplemask)
>   		shader->config.spi_ps_input_ena &= C_0286CC_SAMPLE_COVERAGE_ENA;
>   
>   	return true;
>   }
>   
>   void si_multiwave_lds_size_workaround(struct si_screen *sscreen,
>   				      unsigned *lds_size)
>   {
> +	/* If tessellation is all offchip and on-chip GS isn't used, this
> +	 * workaround is not needed.
> +	 */
> +	return;

Is this expected to always return here? Why you don't remove the entire 
workaround then?

> +
>   	/* SPI barrier management bug:
>   	 *   Make sure we have at least 4k of LDS in use to avoid the bug.
>   	 *   It applies to workgroup sizes of more than one wavefront.
>   	 */
>   	if (sscreen->info.family == CHIP_BONAIRE ||
>   	    sscreen->info.family == CHIP_KABINI ||
>   	    sscreen->info.family == CHIP_MULLINS)
>   		*lds_size = MAX2(*lds_size, 8);
>   }
>   
>