Use rayon to speed up BC7 compression

vE5li · Jan 7, 2025 · 04d0eab · 04d0eab
1 parent 3558e9b
commit 04d0eab
Show file tree

Hide file tree

Showing 2 changed files with 35 additions and 12 deletions.
diff --git a/korangar/src/loaders/async/mod.rs b/korangar/src/loaders/async/mod.rs
@@ -75,9 +75,11 @@ impl AsyncLoader {
         sprite_loader: Arc<SpriteLoader>,
         texture_loader: Arc<TextureLoader>,
     ) -> Self {
+        let num_cpu = rayon::current_num_threads();
+
         let thread_pool = ThreadPoolBuilder::new()
-            .num_threads(2)
-            .thread_name(|_| "async loader".to_string())
+            .num_threads(num_cpu)
+            .thread_name(|number| format!("async loader {number}"))
             .build()
             .unwrap();
 

diff --git a/korangar/src/loaders/texture/mod.rs b/korangar/src/loaders/texture/mod.rs
@@ -11,6 +11,8 @@ use korangar_util::color::contains_transparent_pixel;
 use korangar_util::container::SimpleCache;
 use korangar_util::texture_atlas::{AllocationId, AtlasAllocation, OfflineTextureAtlas};
 use korangar_util::FileLoader;
+use rayon::iter::{IndexedParallelIterator, ParallelIterator};
+use rayon::prelude::ParallelSliceMut;
 use wgpu::{
     CommandEncoderDescriptor, Device, Extent3d, Queue, TextureAspect, TextureDescriptor, TextureDimension, TextureFormat, TextureUsages,
     TextureViewDescriptor, TextureViewDimension,
@@ -246,16 +248,35 @@ impl TextureLoader {
                         }
                         TextureCompression::BC7 => {
                             let mip_size = intel_tex_2::bc7::calc_output_size(width, height);
-                            intel_tex_2::bc7::compress_blocks_into(
-                                &intel_tex_2::bc7::alpha_ultra_fast_settings(),
-                                &RgbaSurface {
-                                    data: current_image.as_raw(),
-                                    width,
-                                    height,
-                                    stride: width * 4,
-                                },
-                                &mut final_buffer[offset..offset + mip_size],
-                            );
+                            let settings = intel_tex_2::bc7::alpha_ultra_fast_settings();
+
+                            let strip_height = 16;
+                            let bytes_per_row = width as usize * 4;
+                            let blocks_per_row = width as usize / 4;
+                            let strip_output_size = blocks_per_row * strip_height / 4 * 16;
+
+                            final_buffer[offset..offset + mip_size]
+                                .par_chunks_mut(strip_output_size)
+                                .enumerate()
+                                .for_each(|(strip_idx, output_chunk)| {
+                                    let strip_y = strip_idx * strip_height;
+                                    let strip_height = (height as usize - strip_y).min(strip_height);
+                                    let strip_height = strip_height - (strip_height % 4);
+
+                                    let src_offset = strip_y * bytes_per_row;
+
+                                    intel_tex_2::bc7::compress_blocks_into(
+                                        &settings,
+                                        &RgbaSurface {
+                                            data: &current_image.as_raw()[src_offset..],
+                                            width,
+                                            height: strip_height as u32,
+                                            stride: (width as usize * 4) as u32,
+                                        },
+                                        output_chunk,
+                                    );
+                                });
+
                             offset += mip_size;
                         }
                     }