Test, improvements on SSE, AVX

awxkee · Sep 25, 2024 · 841ab4e · 841ab4e
1 parent 2ff92da
commit 841ab4e
Show file tree

Hide file tree

Showing 38 changed files with 1,418 additions and 185 deletions.
diff --git a/Cargo.lock b/Cargo.lock
diff --git a/README.md b/README.md
@@ -50,17 +50,17 @@ M3 Pro, NEON dilation RGBA image 2731x4096 with specified kernel size
 
 SSE dilation RGB image 2731x4096 with specified kernel size
 
-| SE     | 9x9     | 15x15    | 21x21    | 41x41    | 61x61 |
-|--------|---------|----------|----------|----------|-------|
-| FM     | 84.19ms | 186.53ms | 254.70ms | 673.45ms | 1.37s |
-| OpenCV | 28.61ms | 62.43ms  | 114.80ms | 428.87ms | 1.16s |
+| SE     | 9x9     | 15x15   | 21x21    | 41x41    | 61x61    |
+|--------|---------|---------|----------|----------|----------|
+| FM     | 30.71ms | 34.87ms | 39.93ms  | 81.56ms  | 149.37ms |
+| OpenCV | 27.36ms | 63.05ms | 112.54ms | 419.40ms | 1.08s    |
 
 SSE dilation RGBA image 2731x4096 with specified kernel size
 
-| SE     | 9x9      | 15x15    | 21x21    | 41x41    | 61x61 |
-|--------|----------|----------|----------|----------|-------|
-| FM     | 109.37ms | 229.11ms | 329.31ms | 981.48ms | 2.05s |
-| OpenCV | 39.20ms  | 76.09ms  | 149.12ms | 569.36ms | 1.33s |
+| SE     | 9x9     | 15x15   | 21x21    | 41x41    | 61x61    |
+|--------|---------|---------|----------|----------|----------|
+| FM     | 45.03ms | 49.03ms | 56.40ms  | 114.72ms | 206.05ms |
+| OpenCV | 35.50ms | 79.60ms | 147.32ms | 556.56ms | 1.33s    |
 
 This project is licensed under either of
 

diff --git a/app/Cargo.toml b/app/Cargo.toml
@@ -7,7 +7,7 @@ edition = "2021"
 image = "0.25.2"
 fast_morphology = {path = "../"}
 imageproc = "0.25.0"
-opencv = {version = "0.93.0", features = ["imgproc", "clang-runtime"]}
+opencv = {version = "0.93.0", features = ["imgproc"]}
 
 [dev-dependencies]
 criterion = {version = "0.5.1", features = ["html_reports"]}

diff --git a/app/benches/dilation/main.rs b/app/benches/dilation/main.rs
@@ -316,19 +316,19 @@ pub fn criterion_benchmark(c: &mut Criterion) {
     opencv::core::set_use_ipp(false).expect("Failed to disable IPP");
     opencv::core::set_use_optimized(false).expect("Failed to disable opts");
 
-    // exec_bench_rgb(c, 4);
-    // exec_bench_rgb(c, 7);
-    // exec_bench_rgb(c, 10);
-    // exec_bench_rgb(c, 20);
-    // exec_bench_rgb(c, 30);
+    exec_bench_rgb(c, 4);
+    exec_bench_rgb(c, 7);
+    exec_bench_rgb(c, 10);
+    exec_bench_rgb(c, 20);
+    exec_bench_rgb(c, 30);
     //
-    exec_bench_rgba(c, 4);
-    exec_bench_rgba(c, 7);
-    exec_bench_rgba(c, 10);
-    exec_bench_rgba(c, 20);
-    exec_bench_rgba(c, 30);
+    // exec_bench_rgba(c, 4);
+    // exec_bench_rgba(c, 7);
+    // exec_bench_rgba(c, 10);
+    // exec_bench_rgba(c, 20);
+    // exec_bench_rgba(c, 30);
 
-    exec_bench_gray(c, 4);
+    // exec_bench_gray(c, 4);
     // exec_bench_gray(c, 7);
     // exec_bench_gray(c, 10);
     // exec_bench_gray(c, 20);

diff --git a/app/src/main.rs b/app/src/main.rs
@@ -1,10 +1,9 @@
 use fast_morphology::{
-    dilate, dilate_rgb, BorderMode, ImageSize, KernelShape, MorphologyThreadingPolicy,
+    dilate, dilate_rgb, dilate_rgba, BorderMode, ImageSize, KernelShape, MorphologyThreadingPolicy,
 };
 use image::{EncodableLayout, GenericImageView, ImageReader};
 use opencv::core::{
-    Mat, MatTrait, MatTraitConstManual, Point, Scalar,
-    BORDER_REPLICATE, CV_8U, CV_8UC3,
+    Mat, MatTrait, MatTraitConstManual, Point, Scalar, BORDER_REPLICATE, CV_8U, CV_8UC3,
 };
 use opencv::imgproc;
 use std::time::Instant;
@@ -164,11 +163,11 @@ fn main() {
     }
 
     let rgba_image = transient_rgba.as_bytes();
-    let mut dst = vec![0u8; saved_origin.len()];
+    let mut dst = vec![0u8; rgba_image.len()];
 
     let exec_time = Instant::now();
-    dilate_rgb(
-        &saved_origin,
+    dilate_rgba(
+        &rgba_image,
         &mut dst,
         image_size,
         &structuring_element,
@@ -237,7 +236,7 @@ fn main() {
         &dst,
         dimensions.0,
         dimensions.1,
-        image::ColorType::Rgb8,
+        image::ColorType::Rgba8,
     )
     .unwrap();
 

diff --git a/src/filter.rs b/src/filter.rs
@@ -28,15 +28,17 @@
  */
 use crate::filter_op_declare::{Arena, MorthOpFilterFlat2DRow};
 use crate::flat_se::AnalyzedSe;
+use crate::morph_base::MorphNativeOp;
 use crate::op_type::MorphOp;
+#[cfg(any(target_arch = "x86", target_arch = "x86_64"))]
+use crate::ops::avx::MorphOpFilterAvx2DRow;
 #[cfg(all(target_arch = "aarch64", target_feature = "neon"))]
 use crate::ops::neon::MorphOpFilterNeon2DRow;
 #[cfg(any(target_arch = "x86", target_arch = "x86_64"))]
-use crate::ops::sse::{MorphOpFilterSse2DRow};
+use crate::ops::sse::{MorphOpFilterSse2DRow, MorphOpFilterSse2DRowF32, MorphOpFilterSse2DRowU16};
 use crate::ops::MorphFilterFlat2DRow;
 use crate::unsafe_slice::UnsafeSlice;
 use crate::ImageSize;
-use crate::morph_base::MorphNativeOp;
 
 pub struct MorthFilterFlat2DRow<T>
 where
@@ -78,9 +80,34 @@ where
                         }
                         #[cfg(any(target_arch = "x86", target_arch = "x86_64"))]
                         {
-                            _result = Box::new(
-                                MorphOpFilterSse2DRow::<{ MorphOp::Dilate as u8 }>::default(),
-                            );
+                            if std::arch::is_x86_feature_detected!("sse4.1") {
+                                _result = Box::new(MorphOpFilterSse2DRow::<
+                                    { MorphOp::Dilate as u8 },
+                                >::default());
+                            }
+                            if std::arch::is_x86_feature_detected!("avx2") {
+                                _result = Box::new(MorphOpFilterAvx2DRow::<
+                                    { MorphOp::Dilate as u8 },
+                                >::default());
+                            }
+                        }
+                    } else if std::any::type_name::<T>() == "u16" {
+                        #[cfg(any(target_arch = "x86", target_arch = "x86_64"))]
+                        {
+                            if std::arch::is_x86_feature_detected!("sse4.1") {
+                                _result = Box::new(MorphOpFilterSse2DRowU16::<
+                                    { MorphOp::Dilate as u8 },
+                                >::default());
+                            }
+                        }
+                    } else if std::any::type_name::<T>() == "f32" {
+                        #[cfg(any(target_arch = "x86", target_arch = "x86_64"))]
+                        {
+                            if std::arch::is_x86_feature_detected!("sse4.1") {
+                                _result = Box::new(MorphOpFilterSse2DRowF32::<
+                                    { MorphOp::Dilate as u8 },
+                                >::default());
+                            }
                         }
                     }
                     _result
@@ -97,9 +124,34 @@ where
                         }
                         #[cfg(any(target_arch = "x86", target_arch = "x86_64"))]
                         {
-                            _result = Box::new(
-                                MorphOpFilterSse2DRow::<{ MorphOp::Erode as u8 }>::default(),
-                            );
+                            if std::arch::is_x86_feature_detected!("sse4.1") {
+                                _result = Box::new(
+                                    MorphOpFilterSse2DRow::<{ MorphOp::Erode as u8 }>::default(),
+                                );
+                            }
+                            if std::arch::is_x86_feature_detected!("avx2") {
+                                _result = Box::new(
+                                    MorphOpFilterAvx2DRow::<{ MorphOp::Erode as u8 }>::default(),
+                                );
+                            }
+                        }
+                    } else if std::any::type_name::<T>() == "u16" {
+                        #[cfg(any(target_arch = "x86", target_arch = "x86_64"))]
+                        {
+                            if std::arch::is_x86_feature_detected!("sse4.1") {
+                                _result = Box::new(MorphOpFilterSse2DRowU16::<
+                                    { MorphOp::Erode as u8 },
+                                >::default());
+                            }
+                        }
+                    } else if std::any::type_name::<T>() == "f32" {
+                        #[cfg(any(target_arch = "x86", target_arch = "x86_64"))]
+                        {
+                            if std::arch::is_x86_feature_detected!("sse4.1") {
+                                _result = Box::new(MorphOpFilterSse2DRowF32::<
+                                    { MorphOp::Erode as u8 },
+                                >::default());
+                            }
                         }
                     }
                     _result

diff --git a/src/morph_base.rs b/src/morph_base.rs
@@ -60,4 +60,4 @@ impl MorphNativeOp<f32> for f32 {
             MorphOp::Erode => (*self).min(other),
         }
     }
-}
+}
diff --git a/src/morph_rgb.rs b/src/morph_rgb.rs
@@ -41,14 +41,7 @@ pub(crate) unsafe fn make_morphology_rgb<T, const OP_TYPE: u8>(
     threading_policy: MorphologyThreadingPolicy,
 ) -> Result<(), String>
 where
-    T: RgbPackable<T>
-        + Copy
-        + 'static
-        + Sync
-        + Send
-        + Clone
-        + Default
-        + MorphNativeOp<T>,
+    T: RgbPackable<T> + Copy + 'static + Sync + Send + Clone + Default + MorphNativeOp<T>,
 {
     let unpacked = T::unpack(src, image_size);
     let mut dst_unpacked = UnpackedRgbImage::alloc(image_size);

diff --git a/src/morph_rgba.rs b/src/morph_rgba.rs
@@ -41,14 +41,7 @@ pub(crate) unsafe fn make_morphology_rgba<T, const OP_TYPE: u8>(
     threading_policy: MorphologyThreadingPolicy,
 ) -> Result<(), String>
 where
-    T: RgbaPackable<T>
-        + Default
-        + Copy
-        + Clone
-        + Send
-        + Sync
-        + 'static
-        + MorphNativeOp<T>,
+    T: RgbaPackable<T> + Default + Copy + Clone + Send + Sync + 'static + MorphNativeOp<T>,
 {
     let unpacked = T::unpack(src, image_size);
     let mut dst_unpacked = UnpackedRgbaImage::alloc(image_size);

diff --git a/src/op.rs b/src/op.rs
@@ -27,13 +27,13 @@
  * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  */
 use crate::border_mode::BorderMode;
+use crate::morph_gray_alpha::make_morphology_gray_alpha;
 use crate::morph_rgb::make_morphology_rgb;
 use crate::morph_rgba::make_morphology_rgba;
 use crate::op_impl::make_morphology;
 use crate::op_type::MorphOp;
 use crate::structuring_element::KernelShape;
 use crate::{ImageSize, MorphologyThreadingPolicy};
-use crate::morph_gray_alpha::make_morphology_gray_alpha;
 
 /// Dilate a gray (planar) image
 ///

diff --git a/src/op_f32.rs b/src/op_f32.rs
@@ -27,13 +27,13 @@
  * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  */
 use crate::border_mode::BorderMode;
+use crate::morph_gray_alpha::make_morphology_gray_alpha;
 use crate::morph_rgb::make_morphology_rgb;
 use crate::morph_rgba::make_morphology_rgba;
 use crate::op_impl::make_morphology;
 use crate::op_type::MorphOp;
 use crate::structuring_element::KernelShape;
 use crate::{ImageSize, MorphologyThreadingPolicy};
-use crate::morph_gray_alpha::make_morphology_gray_alpha;
 
 /// Dilate a gray (planar) stored in u16 image
 ///

diff --git a/src/op_impl.rs b/src/op_impl.rs
@@ -30,13 +30,13 @@ use crate::arena::make_arena;
 use crate::border_mode::BorderMode;
 use crate::filter::MorthFilterFlat2DRow;
 use crate::filter_op_declare::MorthOpFilterFlat2DRow;
+use crate::morph_base::MorphNativeOp;
 use crate::op_type::MorphOp;
 use crate::se_scan::scan_se;
 use crate::structuring_element::KernelShape;
 use crate::unsafe_slice::UnsafeSlice;
 use crate::{ImageSize, MorphologyThreadingPolicy};
 use std::sync::Arc;
-use crate::morph_base::MorphNativeOp;
 
 pub(crate) unsafe fn make_morphology<T, const OP_TYPE: u8>(
     src: &[T],

diff --git a/src/ops/avx/mod.rs b/src/ops/avx/mod.rs
@@ -0,0 +1,3 @@
+mod morph_op;
+
+pub use morph_op::MorphOpFilterAvx2DRow;
-Original file line number
+Diff line change
@@ Expand Up / @@ -60,4 +60,4 @@ impl MorphNativeOp<f32> for f32 { @@
                 MorphOp::Erode => (*self).min(other),
             }
         }
-    }
+    }
Original file line number	Diff line number	Diff line change
		@@ -0,0 +1,3 @@
		mod morph_op;

		pub use morph_op::MorphOpFilterAvx2DRow;