alex/vello
1
0
Fork 0
mirror of https://github.com/italicsjenga/vello.git synced 2025-01-09 12:21:31 +11:00
Code Issues Packages Projects Releases Wiki Activity

Fix performance issues

Browse source 
Use larger workgroup for tile initialization (utilization was poor).
Provide correct element count to coarse rasterizer.
This commit is contained in:
Raph Levien 2020-06-03 15:32:58 -07:00
parent ff8cee059c
commit 63ba45c774
3 changed files with 3 additions and 3 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

2
piet-gpu/shader/tile_alloc.comp
View file

@ -5,7 +5,7 @@
#include "setup.h"
#define LG_TILE_ALLOC_WG 5
#define LG_TILE_ALLOC_WG 8
#define TILE_ALLOC_WG (1 << LG_TILE_ALLOC_WG)
layout(local_size_x = TILE_ALLOC_WG, local_size_y = 1) in;

BIN
piet-gpu/shader/tile_alloc.spv
View file

Binary file not shown.

4
piet-gpu/src/lib.rs
View file

@ -247,7 +247,7 @@ impl<D: Device> Renderer<D> {
let coarse_alloc_start = WIDTH_IN_TILES * HEIGHT_IN_TILES * PTCL_INITIAL_ALLOC;
device.write_buffer(
&coarse_alloc_buf_host,
&[n_elements as u32, coarse_alloc_start as u32],
&[n_paths as u32, coarse_alloc_start as u32],
)?;
let coarse_code = include_bytes!("../shader/coarse.spv");
let coarse_pipeline = device.create_simple_compute_pipeline(coarse_code, 5, 0)?;
@ -323,7 +323,7 @@ impl<D: Device> Renderer<D> {
cmd_buf.dispatch(
&self.tile_pipeline,
&self.tile_ds,
(((self.n_paths + 31) / 32) as u32, 1, 1),
(((self.n_paths + 255) / 256) as u32, 1, 1),
);
cmd_buf.write_timestamp(&query_pool, 2);
cmd_buf.memory_barrier();