[Mesa-dev] [PATCH 2/2] radeonsi: use u_blitter for mipmap generation

[Nicolai Hähnle]

On 13.06.2016 19:30, Marek Olšák wrote:
> From: Marek Olšák <marek.olsak at amd.com>
>
> This reduces time spend in glGenerateMipmap by a half.
> ---
>   src/gallium/drivers/radeonsi/si_blit.c | 37 +++++++++++++++++++++++++++++-----
>   src/gallium/drivers/radeonsi/si_pipe.c |  2 +-
>   2 files changed, 33 insertions(+), 6 deletions(-)
>
> diff --git a/src/gallium/drivers/radeonsi/si_blit.c b/src/gallium/drivers/radeonsi/si_blit.c
> index f2f1ef5..25d531b 100644
> --- a/src/gallium/drivers/radeonsi/si_blit.c
> +++ b/src/gallium/drivers/radeonsi/si_blit.c
> @@ -690,7 +690,8 @@ static void si_clear_depth_stencil(struct pipe_context *ctx,
>    * rendering. */
>   static void si_decompress_subresource(struct pipe_context *ctx,
>   				      struct pipe_resource *tex,
> -				      unsigned planes, unsigned level,
> +				      unsigned planes,
> +				      unsigned first_level, unsigned last_level,
>   				      unsigned first_layer, unsigned last_layer)
>   {
>   	struct si_context *sctx = (struct si_context *)ctx;
> @@ -703,10 +704,10 @@ static void si_decompress_subresource(struct pipe_context *ctx,
>   			planes &= ~PIPE_MASK_S;
>
>   		si_blit_decompress_zs_in_place(sctx, rtex, planes,
> -					       level, level,
> +					       first_level, last_level,
>   					       first_layer, last_layer);
>   	} else if (rtex->fmask.size || rtex->cmask.size || rtex->dcc_offset) {
> -		si_blit_decompress_color(ctx, rtex, level, level,
> +		si_blit_decompress_color(ctx, rtex, first_level, last_level,
>   					 first_layer, last_layer, false);
>   	}
>   }
> @@ -746,7 +747,7 @@ void si_resource_copy_region(struct pipe_context *ctx,
>
>   	/* The driver doesn't decompress resources automatically while
>   	 * u_blitter is rendering. */
> -	si_decompress_subresource(ctx, src, PIPE_MASK_RGBAZS, src_level,
> +	si_decompress_subresource(ctx, src, PIPE_MASK_RGBAZS, src_level, src_level,
>   				  src_box->z, src_box->z + src_box->depth - 1);
>
>   	dst_width = u_minify(dst->width0, dst_level);
> @@ -1005,7 +1006,7 @@ static void si_blit(struct pipe_context *ctx,
>   	/* The driver doesn't decompress resources automatically while
>   	 * u_blitter is rendering. */
>   	si_decompress_subresource(ctx, info->src.resource, info->mask,
> -				  info->src.level,
> +				  info->src.level, info->src.level,
>   				  info->src.box.z,
>   				  info->src.box.z + info->src.box.depth - 1);
>
> @@ -1019,6 +1020,31 @@ static void si_blit(struct pipe_context *ctx,
>   	si_blitter_end(ctx);
>   }
>
> +static boolean si_generate_mipmap(struct pipe_context *ctx,
> +				  struct pipe_resource *tex,
> +				  enum pipe_format format,
> +				  unsigned base_level, unsigned last_level,
> +				  unsigned first_layer, unsigned last_layer)
> +{
> +	struct si_context *sctx = (struct si_context*)ctx;
> +
> +	if (!util_blitter_is_copy_supported(sctx->blitter, tex, tex))
> +		return false;
> +
> +	/* The driver doesn't decompress resources automatically while
> +	 * u_blitter is rendering. */
> +	si_decompress_subresource(ctx, tex, PIPE_MASK_RGBAZS,
> +				  base_level, last_level,
> +				  first_layer, last_layer);

Since the lower levels are overwritten anyway, is decompressing all 
levels really needed? Seems like it should be enough to reset cmask and 
htile to an uncompressed state.

Nicolai

> +	si_blitter_begin(ctx, SI_BLIT | SI_DISABLE_RENDER_COND);
> +	util_blitter_generate_mipmap(sctx->blitter, tex, format,
> +				     base_level, last_level,
> +				     first_layer, last_layer);
> +	si_blitter_end(ctx);
> +	return true;
> +}
> +
>   static void si_flush_resource(struct pipe_context *ctx,
>   			      struct pipe_resource *res)
>   {
> @@ -1113,6 +1139,7 @@ void si_init_blit_functions(struct si_context *sctx)
>   	sctx->b.b.resource_copy_region = si_resource_copy_region;
>   	sctx->b.b.blit = si_blit;
>   	sctx->b.b.flush_resource = si_flush_resource;
> +	sctx->b.b.generate_mipmap = si_generate_mipmap;
>   	sctx->b.blit_decompress_depth = si_blit_decompress_depth;
>   	sctx->b.decompress_dcc = si_decompress_dcc;
>   }
> diff --git a/src/gallium/drivers/radeonsi/si_pipe.c b/src/gallium/drivers/radeonsi/si_pipe.c
> index 0c601da..1ee1877 100644
> --- a/src/gallium/drivers/radeonsi/si_pipe.c
> +++ b/src/gallium/drivers/radeonsi/si_pipe.c
> @@ -357,6 +357,7 @@ static int si_get_param(struct pipe_screen* pscreen, enum pipe_cap param)
>   	case PIPE_CAP_TGSI_PACK_HALF_FLOAT:
>   	case PIPE_CAP_FRAMEBUFFER_NO_ATTACHMENT:
>   	case PIPE_CAP_ROBUST_BUFFER_ACCESS_BEHAVIOR:
> +	case PIPE_CAP_GENERATE_MIPMAP:
>   		return 1;
>
>   	case PIPE_CAP_RESOURCE_FROM_USER_MEMORY:
> @@ -409,7 +410,6 @@ static int si_get_param(struct pipe_screen* pscreen, enum pipe_cap param)
>   	case PIPE_CAP_DRAW_PARAMETERS:
>   	case PIPE_CAP_MULTI_DRAW_INDIRECT:
>   	case PIPE_CAP_MULTI_DRAW_INDIRECT_PARAMS:
> -	case PIPE_CAP_GENERATE_MIPMAP:
>   	case PIPE_CAP_STRING_MARKER:
>   	case PIPE_CAP_QUERY_BUFFER_OBJECT:
>   	case PIPE_CAP_CULL_DISTANCE:
>