OpenHarmony/third_party_mesa3d
2
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

panfrost: Clear with a quad to avoid flushing

Browse Source 
Flushing the batch midframe (splitting a renderpass) is expensive on a tiler, as
it requires the GPU to flush the framebuffer contents to main memory and read
them back. Clearing the framebuffer should not trigger a flush. Apps expect
clears to be (almost) free, flushing for a clear is at the very least unexpected
behaviour.

The only reason we previously flushed is to ensure we could always use a "fast"
clear. But a slow clear is a heck of a lot faster than a flush ;-) Instead of
flushing, we should clear with a draw (via u_blitter) in case a fast clear isn't
possible.

This fixes pathological performance for applications that rely on partial clears
within a frame. This issue was identified with Inochi2D, which repeatedly clears
the stencil buffer midframe, in order to implement masking efficiently with the
stencil buffer. In total, the all-important workload of rendering Asahi Lina is
improved from 17fps to 29fps on a panfrost device.

Fixes: c138ca80d2 ("panfrost: Make sure a clear does not re-use a pre-existing batch")
Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17112>
This commit is contained in:
Alyssa Rosenzweig
2022-06-17 16:40:32 -04:00
committed by Marge Bot
parent 35a80418a1
commit 638b22354e
3 changed files with 23 additions and 14 deletions
Download Patch File Download Diff File Expand all files Collapse all files

10
src/gallium/drivers/panfrost/pan_blit.c
View File

@@ -31,12 +31,10 @@
#include "pan_util.h"
#include "util/format/u_format.h"
static void
panfrost_blitter_save(
struct panfrost_context *ctx,
struct blitter_context *blitter,
bool render_cond)
void
panfrost_blitter_save(struct panfrost_context *ctx, bool render_cond)
{
struct blitter_context *blitter = ctx->blitter;
util_blitter_save_vertex_buffer_slot(blitter, ctx->vertex_buffers);
util_blitter_save_vertex_elements(blitter, ctx->vertex);
@@ -82,6 +80,6 @@ panfrost_blit(struct pipe_context *pipe,
if (!util_blitter_is_blit_supported(ctx->blitter, info))
unreachable("Unsupported blit\n");
panfrost_blitter_save(ctx, ctx->blitter, info->render_condition_enable);
panfrost_blitter_save(ctx, info->render_condition_enable);
util_blitter_blit(ctx->blitter, info);
}

24
src/gallium/drivers/panfrost/pan_context.c
View File

@@ -62,18 +62,26 @@ panfrost_clear(
double depth, unsigned stencil)
{
struct panfrost_context *ctx = pan_context(pipe);
struct panfrost_batch *batch = panfrost_get_batch_for_fbo(ctx);
if (!panfrost_render_condition_check(ctx))
return;
/* TODO: panfrost_get_fresh_batch_for_fbo() instantiates a new batch if
* the existing batch targeting this FBO has draws. We could probably
* avoid that by replacing plain clears by quad-draws with a specific
* color/depth/stencil value, thus avoiding the generation of extra
* fragment jobs.
*/
struct panfrost_batch *batch = panfrost_get_fresh_batch_for_fbo(ctx, "Slow clear");
panfrost_batch_clear(batch, buffers, color, depth, stencil);
/* At the start of the batch, we can clear for free */
if (!batch->scoreboard.first_job) {
panfrost_batch_clear(batch, buffers, color, depth, stencil);
return;
}
/* Once there is content, clear with a fullscreen quad */
panfrost_blitter_save(ctx, false /* render condition */);
util_blitter_clear(ctx->blitter,
ctx->pipe_framebuffer.width,
ctx->pipe_framebuffer.height,
util_framebuffer_get_num_layers(&ctx->pipe_framebuffer),
buffers, color, depth, stencil,
util_framebuffer_get_num_samples(&ctx->pipe_framebuffer) > 1);
}
bool

3
src/gallium/drivers/panfrost/pan_resource.h
View File

@@ -124,6 +124,9 @@ void panfrost_resource_context_init(struct pipe_context *pctx);
/* Blitting */
void
panfrost_blitter_save(struct panfrost_context *ctx, bool render_cond);
void
panfrost_blit(struct pipe_context *pipe,
const struct pipe_blit_info *info);