[Mesa-dev] [PATCH 1/2] radeonsi: disable ReZ

[Andy Furniss]

Marek Olšák wrote:
> From: Marek Olšák <marek.olsak at amd.com>
>
> This is a serious performance fix. Discovered by luck.

Hmm, maybe I should bump my bugs more aggressively :-)

https://bugs.freedesktop.org/show_bug.cgi?id=94354

>
> Cc: 12.0 <mesa-stable at lists.freedesktop.org>
> ---
>   src/gallium/drivers/radeonsi/si_state_shaders.c | 11 ++++-------
>   1 file changed, 4 insertions(+), 7 deletions(-)
>
> diff --git a/src/gallium/drivers/radeonsi/si_state_shaders.c b/src/gallium/drivers/radeonsi/si_state_shaders.c
> index c41c519..be5c659 100644
> --- a/src/gallium/drivers/radeonsi/si_state_shaders.c
> +++ b/src/gallium/drivers/radeonsi/si_state_shaders.c
> @@ -801,34 +801,31 @@ static void si_shader_ps(struct si_shader *shader)
>   	si_pm4_set_reg(pm4, R_00B028_SPI_SHADER_PGM_RSRC1_PS,
>   		       S_00B028_VGPRS((shader->config.num_vgprs - 1) / 4) |
>   		       S_00B028_SGPRS((shader->config.num_sgprs - 1) / 8) |
>   		       S_00B028_DX10_CLAMP(1) |
>   		       S_00B028_FLOAT_MODE(shader->config.float_mode));
>   	si_pm4_set_reg(pm4, R_00B02C_SPI_SHADER_PGM_RSRC2_PS,
>   		       S_00B02C_EXTRA_LDS_SIZE(shader->config.lds_size) |
>   		       S_00B02C_USER_SGPR(SI_PS_NUM_USER_SGPR) |
>   		       S_00B32C_SCRATCH_EN(shader->config.scratch_bytes_per_wave > 0));
>
> -	/* Prefer RE_Z if the shader is complex enough. The requirement is either:
> -	 * - the shader uses at least 2 VMEM instructions, or
> -	 * - the code size is at least 50 2-dword instructions or 100 1-dword
> -	 *   instructions.
> +	/* DON'T USE EARLY_Z_THEN_RE_Z !!!
> +	 *
> +	 * It decreases performance by 15% in DiRT: Showdown on Ultra settings.
> +	 * And it has pretty complex shaders.
>   	 *
>   	 * Shaders with side effects that must execute independently of the
>   	 * depth test require LATE_Z.
>   	 */
>   	if (info->writes_memory &&
>   	    !info->properties[TGSI_PROPERTY_FS_EARLY_DEPTH_STENCIL])
>   		shader->z_order = V_02880C_LATE_Z;
> -	else if (info->num_memory_instructions >= 2 ||
> -	         shader->binary.code_size > 100*4)
> -		shader->z_order = V_02880C_EARLY_Z_THEN_RE_Z;
>   	else
>   		shader->z_order = V_02880C_EARLY_Z_THEN_LATE_Z;
>   }
>
>   static void si_shader_init_pm4_state(struct si_screen *sscreen,
>                                        struct si_shader *shader)
>   {
>   	switch (shader->selector->type) {
>   	case PIPE_SHADER_VERTEX:
>   		if (shader->key.vs.as_ls)
>