Vello CPU: Implement optional gamma correction for solid colour paths

sparse_strips/vello_bench/src/coarse.rs

@@ -33,6 +33,7 @@ pub fn coarse(c: &mut Criterion) {
                     0,
                     None,
                     &[],
+                    false,
                 );
                 std::hint::black_box(&wide);
             });
@@ -93,6 +94,7 @@ pub fn coarse_with_layer(c: &mut Criterion) {
                     0,
                     None,
                     &[],
+                    false,
                 );
 
                 // Pop all layers
@@ -167,6 +169,7 @@ pub fn coarse_with_layer_large_viewport(c: &mut Criterion) {
                     0,
                     None,
                     &[],
+                    false,
                 );
 
                 // Pop all layers

sparse_strips/vello_bench/src/fine/fill.rs

@@ -68,6 +68,7 @@ pub(crate) fn fill_single<S: Simd, N: FineKernel<S>>(
             &NoOpImageResolver,
             None,
             None,
+            false,
         );
 
         std::hint::black_box(&fine);

sparse_strips/vello_bench/src/fine/strip.rs

@@ -60,6 +60,7 @@ fn strip_single<S: Simd, N: FineKernel<S>>(
             &NoOpImageResolver,
             Some(&alphas),
             None,
+            false,
         );
 
         std::hint::black_box(&fine);

sparse_strips/vello_common/src/coarse.rs

@@ -502,6 +502,7 @@ impl<const MODE: u8> Wide<MODE> {
         thread_idx: u8,
         mask: Option<Mask>,
         encoded_paints: &[EncodedPaint],
+        gamma_correction: bool,
     ) {
         if strip_buf.is_empty() {
             return;
@@ -516,6 +517,7 @@ impl<const MODE: u8> Wide<MODE> {
             thread_idx,
             paint,
             blend_mode,
+            gamma_correction,
             mask,
             alpha_base_idx,
         });
@@ -2044,6 +2046,8 @@ pub struct FillAttrs {
     pub paint: Paint,
     /// The blend mode to apply before drawing the contents.
     pub blend_mode: BlendMode,
+    /// Whether to use gamma-corrected compositing for solid color fills.
+    pub gamma_correction: bool,
     /// A mask to apply to the command.
     pub mask: Option<Mask>,
     /// Base index into the alpha buffer for this path's commands.
@@ -2348,6 +2352,7 @@ mod tests {
             0,
             None,
             &[],
+            false,
         );
         assert!(wide.tiles_dirty);
 
@@ -2384,6 +2389,7 @@ mod tests {
             0,
             None,
             &[],
+            false,
         );
         assert!(!wide.tiles_dirty);
     }

sparse_strips/vello_cpu/src/dispatch/mod.rs

@@ -37,6 +37,7 @@ pub(crate) trait Dispatcher: Debug + Send + Sync {
         aliasing_threshold: Option<u8>,
         mask: Option<Mask>,
         encoded_paints: &[EncodedPaint],
+        gamma_correction: bool,
     );
     fn stroke_path(
         &mut self,
@@ -48,6 +49,7 @@ pub(crate) trait Dispatcher: Debug + Send + Sync {
         aliasing_threshold: Option<u8>,
         mask: Option<Mask>,
         encoded_paints: &[EncodedPaint],
+        gamma_correction: bool,
     );
     /// Fill a pixel-aligned rectangle with the current paint.
     fn fill_rect_fast(
@@ -57,6 +59,7 @@ pub(crate) trait Dispatcher: Debug + Send + Sync {
         blend_mode: BlendMode,
         mask: Option<Mask>,
         encoded_paints: &[EncodedPaint],
+        gamma_correction: bool,
     );
     fn push_clip_path(
         &mut self,

sparse_strips/vello_cpu/src/dispatch/multi_threaded.rs

@@ -308,6 +308,7 @@ impl MultiThreadedDispatcher {
                                 blend_mode,
                                 thread_id,
                                 mask,
+                                gamma_correction,
                             } => self.wide.generate(
                                 &task.allocation_group.strips
                                     [strip_range.start as usize..strip_range.end as usize],
@@ -316,6 +317,7 @@ impl MultiThreadedDispatcher {
                                 thread_id,
                                 mask,
                                 encoded_paints,
+                                gamma_correction,
                             ),
                             CoarseTaskType::RenderWideCommand {
                                 strips,
@@ -330,6 +332,7 @@ impl MultiThreadedDispatcher {
                                 thread_id,
                                 mask,
                                 encoded_paints,
+                                false,
                             ),
                             CoarseTaskType::PushLayer {
                                 thread_id,
@@ -441,6 +444,7 @@ impl Dispatcher for MultiThreadedDispatcher {
         aliasing_threshold: Option<u8>,
         mask: Option<Mask>,
         _encoded_paints: &[EncodedPaint],
+        gamma_correction: bool,
     ) {
         let start = self.allocation_group.path.len() as u32;
         self.allocation_group.path.extend(path);
@@ -453,6 +457,7 @@ impl Dispatcher for MultiThreadedDispatcher {
             blend_mode,
             aliasing_threshold,
             mask,
+            gamma_correction,
         });
     }
 
@@ -466,6 +471,7 @@ impl Dispatcher for MultiThreadedDispatcher {
         aliasing_threshold: Option<u8>,
         mask: Option<Mask>,
         _encoded_paints: &[EncodedPaint],
+        gamma_correction: bool,
     ) {
         let start = self.allocation_group.path.len() as u32;
         self.allocation_group.path.extend(path);
@@ -478,6 +484,7 @@ impl Dispatcher for MultiThreadedDispatcher {
             blend_mode,
             aliasing_threshold,
             mask,
+            gamma_correction,
         });
     }
 
@@ -488,6 +495,7 @@ impl Dispatcher for MultiThreadedDispatcher {
         blend_mode: BlendMode,
         mask: Option<Mask>,
         _encoded_paints: &[EncodedPaint],
+        gamma_correction: bool,
     ) {
         // For multi-threaded, fall back to path-based rendering.
         // TODO: Implement optimized rect strip generation in worker threads.
@@ -508,6 +516,7 @@ impl Dispatcher for MultiThreadedDispatcher {
             blend_mode,
             aliasing_threshold: None,
             mask,
+            gamma_correction,
         });
     }
 
@@ -834,6 +843,7 @@ pub(crate) enum RenderTaskType {
         blend_mode: BlendMode,
         aliasing_threshold: Option<u8>,
         mask: Option<Mask>,
+        gamma_correction: bool,
     },
     WideCommand {
         strip_buf: Box<[Strip]>,
@@ -849,6 +859,7 @@ pub(crate) enum RenderTaskType {
         blend_mode: BlendMode,
         aliasing_threshold: Option<u8>,
         mask: Option<Mask>,
+        gamma_correction: bool,
     },
     PushLayer {
         clip_path: Option<(Range<u32>, Affine)>,
@@ -873,6 +884,7 @@ pub(crate) enum CoarseTaskType {
         blend_mode: BlendMode,
         paint: Paint,
         mask: Option<Mask>,
+        gamma_correction: bool,
     },
     RenderWideCommand {
         thread_id: u8,
@@ -959,6 +971,7 @@ mod tests {
                 None,
                 None,
                 &[],
+                false,
             );
             dispatcher.flush(&[]);
         }

sparse_strips/vello_cpu/src/dispatch/multi_threaded/worker.rs

@@ -70,6 +70,7 @@ impl Worker {
                     blend_mode,
                     aliasing_threshold,
                     mask,
+                    gamma_correction,
                 } => {
                     let start = self.strip_storage.strips.len() as u32;
                     let path = &render_task.allocation_group.path
@@ -91,6 +92,7 @@ impl Worker {
                         blend_mode,
                         paint,
                         mask,
+                        gamma_correction,
                     };
 
                     render_task
@@ -106,6 +108,7 @@ impl Worker {
                     stroke,
                     aliasing_threshold,
                     mask,
+                    gamma_correction,
                 } => {
                     let start = self.strip_storage.strips.len() as u32;
                     let path = &render_task.allocation_group.path
@@ -127,6 +130,7 @@ impl Worker {
                         blend_mode,
                         paint,
                         mask,
+                        gamma_correction,
                     };
 
                     render_task

sparse_strips/vello_cpu/src/dispatch/single_threaded.rs

@@ -539,6 +539,7 @@ impl Dispatcher for SingleThreadedDispatcher {
         aliasing_threshold: Option<u8>,
         mask: Option<Mask>,
         encoded_paints: &[EncodedPaint],
+        gamma_correction: bool,
     ) {
         let wide = &mut self.wide;
 
@@ -560,6 +561,7 @@ impl Dispatcher for SingleThreadedDispatcher {
             0,
             mask,
             encoded_paints,
+            gamma_correction,
         );
     }
 
@@ -573,6 +575,7 @@ impl Dispatcher for SingleThreadedDispatcher {
         aliasing_threshold: Option<u8>,
         mask: Option<Mask>,
         encoded_paints: &[EncodedPaint],
+        gamma_correction: bool,
     ) {
         let wide = &mut self.wide;
 
@@ -594,6 +597,7 @@ impl Dispatcher for SingleThreadedDispatcher {
             0,
             mask,
             encoded_paints,
+            gamma_correction,
         );
     }
 
@@ -604,6 +608,7 @@ impl Dispatcher for SingleThreadedDispatcher {
         blend_mode: BlendMode,
         mask: Option<Mask>,
         encoded_paints: &[EncodedPaint],
+        gamma_correction: bool,
     ) {
         let wide = &mut self.wide;
 
@@ -622,6 +627,7 @@ impl Dispatcher for SingleThreadedDispatcher {
             0,
             mask,
             encoded_paints,
+            gamma_correction,
         );
     }
 
@@ -738,14 +744,7 @@ impl Dispatcher for SingleThreadedDispatcher {
         {
             // This case never gets hit because there is a compile_error in the root.
             // But have this code disables some warnings and makes the compile error easier to read
-            let _ = (
-                buffer,
-                render_mode,
-                width,
-                height,
-                encoded_paints,
-                image_resolver,
-            );
+            let _ = (buffer, render_mode, width, height, encoded_paints, image_resolver);
         }
     }
 
@@ -827,16 +826,8 @@ impl Dispatcher for SingleThreadedDispatcher {
         #[cfg(all(not(feature = "u8_pipeline"), not(feature = "f32_pipeline")))]
         {
             let _ = (
-                buffer,
-                width,
-                height,
-                dst_x,
-                dst_y,
-                dst_buffer_width,
-                dst_buffer_height,
-                render_mode,
-                encoded_paints,
-                image_resolver,
+                buffer, width, height, dst_x, dst_y, dst_buffer_width, dst_buffer_height,
+                render_mode, encoded_paints, image_resolver,
             );
         }
     }
@@ -850,7 +841,7 @@ impl Dispatcher for SingleThreadedDispatcher {
     ) {
         // Generate coarse-level commands from pre-computed strips (thread_idx 0 for single-threaded).
         self.wide
-            .generate(strip_buf, paint, blend_mode, 0, None, encoded_paints);
+            .generate(strip_buf, paint, blend_mode, 0, None, encoded_paints, false);
     }
 
     fn strip_storage_mut(&mut self) -> &mut StripStorage {
@@ -927,6 +918,7 @@ mod tests {
             None,
             None,
             &[],
+            false,
         );
 
         // Ensure there is data to clear.

sparse_strips/vello_cpu/src/fine/highp/blend.rs

@@ -6,19 +6,27 @@ use crate::util::Premultiply;
 use vello_common::fearless_simd::*;
 
 #[derive(Copy, Clone)]
-struct Channels<S: Simd> {
-    r: f32x4<S>,
-    g: f32x4<S>,
-    b: f32x4<S>,
+pub(crate) struct Channels<S: Simd> {
+    pub(crate) r: f32x4<S>,
+    pub(crate) g: f32x4<S>,
+    pub(crate) b: f32x4<S>,
 }
 
 impl<S: Simd> Channels<S> {
     #[inline(always)]
-    fn unpremultiply(mut self, a: f32x4<S>) -> Self {
+    pub(crate) fn unpremultiply(mut self, a: f32x4<S>) -> Self {
         self.r = self.r.unpremultiply(a);
         self.g = self.g.unpremultiply(a);
         self.b = self.b.unpremultiply(a);
 
+        self
+    }
+    #[inline(always)]
+    pub(crate) fn square(mut self) -> Self {
+        self.r *= self.r;
+        self.g *= self.g;
+        self.b *= self.b;
+
         self
     }
 }