Improvements, bugfixes

awxkee · Jul 2, 2024 · 89931dc · 89931dc
1 parent 88a7f82
commit 89931dc
Show file tree

Hide file tree

Showing 14 changed files with 329 additions and 421 deletions.
diff --git a/Cargo.lock b/Cargo.lock
diff --git a/README.md b/README.md
@@ -7,7 +7,7 @@ You may receive gaussian blur in 100 FPS for 4K photo.
 
 Much faster than `image` default blur.
 
-When 4-channels mode is in use that always considered thad alpha channel is the last.
+When 4-channels mode is in use that always considered that alpha channel is the last.
 
 Also there are some available options to perform blurring in linear colorspace, or if methods do not fit you `f32`
 options also available

diff --git a/src/lib/Cargo.toml b/src/lib/Cargo.toml
@@ -1,6 +1,6 @@
 [package]
 name = "libblur"
-version = "0.12.1"
+version = "0.12.2"
 edition = "2021"
 description = "Fast image blurring in pure Rust"
 readme = "../../README.md"

diff --git a/src/lib/fast_gaussian.rs b/src/lib/fast_gaussian.rs
diff --git a/src/lib/gaussian/gaussian.rs b/src/lib/gaussian/gaussian.rs
@@ -31,7 +31,6 @@ use rayon::ThreadPool;
 
 use crate::channels_configuration::FastBlurChannels;
 use crate::edge_mode::EdgeMode;
-use crate::gaussian::gaussian_f16::gaussian_f16::gaussian_blur_impl_f16;
 use crate::gaussian::gaussian_filter::create_filter;
 use crate::gaussian::gaussian_horizontal::gaussian_blur_horizontal_pass_impl;
 use crate::gaussian::gaussian_kernel::get_gaussian_kernel_1d;
@@ -54,7 +53,7 @@ use crate::unsafe_slice::UnsafeSlice;
 use crate::ThreadingPolicy;
 
 fn gaussian_blur_horizontal_pass<
-    T: FromPrimitive + Default + Into<f32> + Send + Sync,
+    T: FromPrimitive + Default + Send + Sync,
     const CHANNEL_CONFIGURATION: usize,
     const EDGE_MODE: usize,
 >(
@@ -69,7 +68,7 @@ fn gaussian_blur_horizontal_pass<
     thread_pool: &ThreadPool,
     thread_count: u32,
 ) where
-    T: std::ops::AddAssign + std::ops::SubAssign + Copy + 'static,
+    T: std::ops::AddAssign + std::ops::SubAssign + Copy + 'static + AsPrimitive<f32>,
     f32: AsPrimitive<T> + ToStorage<T>,
 {
     let mut _dispatcher: fn(
@@ -126,7 +125,7 @@ fn gaussian_blur_horizontal_pass<
 }
 
 fn gaussian_blur_vertical_pass_impl<
-    T: FromPrimitive + Default + Into<f32> + Send + Sync,
+    T: FromPrimitive + Default + Send + Sync,
     const CHANNEL_CONFIGURATION: usize,
     const EDGE_MODE: usize,
 >(
@@ -141,7 +140,7 @@ fn gaussian_blur_vertical_pass_impl<
     start_y: u32,
     end_y: u32,
 ) where
-    T: std::ops::AddAssign + std::ops::SubAssign + Copy + 'static,
+    T: std::ops::AddAssign + std::ops::SubAssign + Copy + 'static + AsPrimitive<f32>,
     f32: AsPrimitive<T> + ToStorage<T>,
 {
     gaussian_blur_vertical_pass_c_impl::<T, CHANNEL_CONFIGURATION, EDGE_MODE>(
@@ -159,7 +158,7 @@ fn gaussian_blur_vertical_pass_impl<
 }
 
 fn gaussian_blur_vertical_pass<
-    T: FromPrimitive + Default + Into<f32> + Send + Sync,
+    T: FromPrimitive + Default + Send + Sync,
     const CHANNEL_CONFIGURATION: usize,
     const EDGE_MODE: usize,
 >(
@@ -174,7 +173,7 @@ fn gaussian_blur_vertical_pass<
     thread_pool: &ThreadPool,
     thread_count: u32,
 ) where
-    T: std::ops::AddAssign + std::ops::SubAssign + Copy + 'static,
+    T: std::ops::AddAssign + std::ops::SubAssign + Copy + 'static + AsPrimitive<f32>,
     f32: AsPrimitive<T> + ToStorage<T>,
 {
     let mut _dispatcher: fn(
@@ -234,7 +233,7 @@ fn gaussian_blur_vertical_pass<
 }
 
 fn gaussian_blur_impl<
-    T: FromPrimitive + Default + Into<f32> + Send + Sync,
+    T: FromPrimitive + Default + Send + Sync,
     const CHANNEL_CONFIGURATION: usize,
 >(
     src: &[T],
@@ -248,7 +247,7 @@ fn gaussian_blur_impl<
     threading_policy: ThreadingPolicy,
     edge_mode: EdgeMode,
 ) where
-    T: std::ops::AddAssign + std::ops::SubAssign + Copy + 'static,
+    T: std::ops::AddAssign + std::ops::SubAssign + Copy + 'static + AsPrimitive<f32>,
     f32: AsPrimitive<T> + ToStorage<T>,
 {
     if kernel_size % 2 == 0 {
@@ -608,6 +607,7 @@ pub fn gaussian_blur_f32(
 /// * `kernel_size` - Length of gaussian kernel. Panic if kernel size is not odd, even kernels with unbalanced center is not accepted.
 /// * `sigma` - Sigma for a gaussian kernel, corresponds to kernel flattening level. Default - kernel_size / 6
 /// * `channels` - Count of channels in the image
+/// * `edge_mode` - Rule to handle edge mode
 /// * `threading_policy` - Threading policy according to *ThreadingPolicy*
 ///
 /// # Panics
@@ -620,18 +620,37 @@ pub fn gaussian_blur_f16(
     kernel_size: u32,
     sigma: f32,
     channels: FastBlurChannels,
+    edge_mode: EdgeMode,
     threading_policy: ThreadingPolicy,
 ) {
-    gaussian_blur_impl_f16(
-        src,
-        width * channels.get_channels() as u32,
-        dst,
-        width * channels.get_channels() as u32,
-        width,
-        height,
-        kernel_size,
-        sigma,
-        channels,
-        threading_policy,
-    );
+    match channels {
+        FastBlurChannels::Channels3 => {
+            gaussian_blur_impl::<half::f16, 3>(
+                unsafe { std::mem::transmute(src) },
+                width * channels.get_channels() as u32,
+                unsafe { std::mem::transmute(dst) },
+                width * channels.get_channels() as u32,
+                width,
+                height,
+                kernel_size,
+                sigma,
+                threading_policy,
+                edge_mode,
+            );
+        }
+        FastBlurChannels::Channels4 => {
+            gaussian_blur_impl::<half::f16, 4>(
+                unsafe { std::mem::transmute(src) },
+                width * channels.get_channels() as u32,
+                unsafe { std::mem::transmute(dst) },
+                width * channels.get_channels() as u32,
+                width,
+                height,
+                kernel_size,
+                sigma,
+                threading_policy,
+                edge_mode,
+            );
+        }
+    }
 }