RGBCube/serenity
1
Fork 0
mirror of https://github.com/RGBCube/serenity synced 2025-07-26 06:37:35 +00:00
Code Issues Activity Actions

LibSoftGPU: Optimize GPU::Vertex lists

Browse source 
On my machine, benchmarking 3DFileViewer revealed ~2.5% of CPU time
spent in `Vector<GPU::Vertex>::try_append`. By carefully managing list
capacities, we can remove this method from profiles altogether.
This commit is contained in:
Jelle Raaijmakers 2023-10-05 23:02:53 +02:00
parent c978891dda
commit 403d3bbdaf
2 changed files with 12 additions and 5 deletions
Download patch file Download diff file Expand all files Collapse all files

8
Userland/Libraries/LibSoftGPU/Clipper.cpp
View file

@ -76,6 +76,10 @@ FLATTEN static void clip_plane(Vector<GPU::Vertex>& input_list, Vector<GPU::Vert
if (input_list_size == 0)
return;
// Ensure we can perform unchecked appends in the loop below
if (input_list_size * 2 > output_list.capacity())
output_list.ensure_capacity(input_list_size * 2);
auto const* prev_vec = &input_list.data()[0];
auto is_prev_point_within_plane = point_within_plane<plane>(*prev_vec, clip_plane);
@ -84,10 +88,10 @@ FLATTEN static void clip_plane(Vector<GPU::Vertex>& input_list, Vector<GPU::Vert
auto const is_curr_point_within_plane = point_within_plane<plane>(curr_vec, clip_plane);
if (is_curr_point_within_plane != is_prev_point_within_plane)
output_list.append(clip_intersection_point<plane>(*prev_vec, curr_vec, clip_plane));
output_list.unchecked_append(clip_intersection_point<plane>(*prev_vec, curr_vec, clip_plane));
if (is_curr_point_within_plane)
output_list.append(curr_vec);
output_list.unchecked_append(curr_vec);
prev_vec = &curr_vec;
is_prev_point_within_plane = is_curr_point_within_plane;