Remove a skippable write in the clean_and_write ISPC functions

src/ispc/downsample_ispc.rs

@@ -11,6 +11,8 @@ pub struct WeightVariables {
 }
 #[test]
 fn bindgen_test_layout_WeightVariables() {
+    const UNINIT: ::std::mem::MaybeUninit<WeightVariables> = ::std::mem::MaybeUninit::uninit();
+    let ptr = UNINIT.as_ptr();
     assert_eq!(
         ::std::mem::size_of::<WeightVariables>(),
         12usize,
@@ -22,7 +24,7 @@ fn bindgen_test_layout_WeightVariables() {
         concat!("Alignment of ", stringify!(WeightVariables))
     );
     assert_eq!(
-        unsafe { &(*(::std::ptr::null::<WeightVariables>())).src_center as *const _ as usize },
+        unsafe { ::std::ptr::addr_of!((*ptr).src_center) as usize - ptr as usize },
         0usize,
         concat!(
             "Offset of field: ",
@@ -32,7 +34,7 @@ fn bindgen_test_layout_WeightVariables() {
         )
     );
     assert_eq!(
-        unsafe { &(*(::std::ptr::null::<WeightVariables>())).src_start as *const _ as usize },
+        unsafe { ::std::ptr::addr_of!((*ptr).src_start) as usize - ptr as usize },
         4usize,
         concat!(
             "Offset of field: ",
@@ -42,7 +44,7 @@ fn bindgen_test_layout_WeightVariables() {
         )
     );
     assert_eq!(
-        unsafe { &(*(::std::ptr::null::<WeightVariables>())).src_end as *const _ as usize },
+        unsafe { ::std::ptr::addr_of!((*ptr).src_end) as usize - ptr as usize },
         8usize,
         concat!(
             "Offset of field: ",
@@ -61,6 +63,8 @@ pub struct WeightCollection {
 }
 #[test]
 fn bindgen_test_layout_WeightCollection() {
+    const UNINIT: ::std::mem::MaybeUninit<WeightCollection> = ::std::mem::MaybeUninit::uninit();
+    let ptr = UNINIT.as_ptr();
     assert_eq!(
         ::std::mem::size_of::<WeightCollection>(),
         24usize,
@@ -72,7 +76,7 @@ fn bindgen_test_layout_WeightCollection() {
         concat!("Alignment of ", stringify!(WeightCollection))
     );
     assert_eq!(
-        unsafe { &(*(::std::ptr::null::<WeightCollection>())).starts as *const _ as usize },
+        unsafe { ::std::ptr::addr_of!((*ptr).starts) as usize - ptr as usize },
         0usize,
         concat!(
             "Offset of field: ",
@@ -82,7 +86,7 @@ fn bindgen_test_layout_WeightCollection() {
         )
     );
     assert_eq!(
-        unsafe { &(*(::std::ptr::null::<WeightCollection>())).weight_counts as *const _ as usize },
+        unsafe { ::std::ptr::addr_of!((*ptr).weight_counts) as usize - ptr as usize },
         8usize,
         concat!(
             "Offset of field: ",
@@ -92,7 +96,7 @@ fn bindgen_test_layout_WeightCollection() {
         )
     );
     assert_eq!(
-        unsafe { &(*(::std::ptr::null::<WeightCollection>())).values as *const _ as usize },
+        unsafe { ::std::ptr::addr_of!((*ptr).values) as usize - ptr as usize },
         16usize,
         concat!(
             "Offset of field: ",
@@ -110,6 +114,8 @@ pub struct Cache {
 }
 #[test]
 fn bindgen_test_layout_Cache() {
+    const UNINIT: ::std::mem::MaybeUninit<Cache> = ::std::mem::MaybeUninit::uninit();
+    let ptr = UNINIT.as_ptr();
     assert_eq!(
         ::std::mem::size_of::<Cache>(),
         48usize,
@@ -121,7 +127,7 @@ fn bindgen_test_layout_Cache() {
         concat!("Alignment of ", stringify!(Cache))
     );
     assert_eq!(
-        unsafe { &(*(::std::ptr::null::<Cache>())).vertical_weights as *const _ as usize },
+        unsafe { ::std::ptr::addr_of!((*ptr).vertical_weights) as usize - ptr as usize },
         0usize,
         concat!(
             "Offset of field: ",
@@ -131,7 +137,7 @@ fn bindgen_test_layout_Cache() {
         )
     );
     assert_eq!(
-        unsafe { &(*(::std::ptr::null::<Cache>())).horizontal_weights as *const _ as usize },
+        unsafe { ::std::ptr::addr_of!((*ptr).horizontal_weights) as usize - ptr as usize },
         24usize,
         concat!(
             "Offset of field: ",

src/ispc/kernels/lanczos3.ispc

@@ -90,7 +90,7 @@ struct Cache {
 };
 
 uint8<3> sample_3_channels(varying uint64 read_address, const uniform uint8* varying src) {
-    // Pointer hacks to read all channels at once rather than one-by-one, despite technically reading uint8<3> to a uint8<4>.
+    // Memory reinterpretation to read all channels at once rather than one-by-one.
     // While testing, this proved more performant than reading one-by-one.
     const uniform uint8* pixel_ptr = src + read_address;
     const uniform uint8<3>* pixel_ptr3 = (const uniform uint8<3>*)(pixel_ptr);
@@ -101,17 +101,12 @@ uint8<3> sample_3_channels(varying uint64 read_address, const uniform uint8* var
 
 void clean_and_write_3_channels(varying float<3> color, varying uint64 write_address, uniform uint8* varying dst) {
     // The final color is a sum of numbers that are multiplied by the weights of their respective pixels.
-    // Because of their numbers, floating point precision leads to the final color being potentially outside of the 0-255 range by a slighty margin.
+    // Because of their numbers, floating point precision leads to the final color being potentially outside of the 0-255 range by a slight margin.
     // This would cause an underflow/overflow, which we avoid with the clamps.
-    color.x = clamp(color.x, 0.0f, 255.0f);
-    color.y = clamp(color.y, 0.0f, 255.0f);
-    color.z = clamp(color.z, 0.0f, 255.0f);
-
-    // Pointer hacks to write all channels at once rather than one-by-one, despite technically writing uint8<4> to a uint8<3>.
-    // While testing, this proved more performant than writing one-by-one.
     uniform uint8* pixel_ptr = dst + write_address;
-    uniform uint8<3>* pixel_ptr3 = (uniform uint8<3>*)(pixel_ptr);
-    *pixel_ptr3 = color;
+    pixel_ptr[0] = clamp(color[0], 0.0f, 255.0f);
+    pixel_ptr[1] = clamp(color[1], 0.0f, 255.0f);
+    pixel_ptr[2] = clamp(color[2], 0.0f, 255.0f);
 }
 
 uint8<4> sample_4_channels(varying uint64 read_address, const uniform uint8* varying src) {
@@ -123,14 +118,11 @@ uint8<4> sample_4_channels(varying uint64 read_address, const uniform uint8* var
 }
 
 void clean_and_write_4_channels(varying float<4> color, varying uint64 write_address, uniform uint8* varying dst) {
-    color.x = clamp(color.x, 0.0f, 255.0f);
-    color.y = clamp(color.y, 0.0f, 255.0f);
-    color.z = clamp(color.z, 0.0f, 255.0f);
-    color.w = clamp(color.w, 0.0f, 255.0f);
-
     uniform uint8* pixel_ptr = dst + write_address;
-    uniform uint8<4>* pixel_ptr4 = (uniform uint8<4>*)(pixel_ptr);
-    *pixel_ptr4 = color;
+    pixel_ptr[0] = clamp(color.x, 0.0f, 255.0f);
+    pixel_ptr[1] = clamp(color.y, 0.0f, 255.0f);
+    pixel_ptr[2] = clamp(color.z, 0.0f, 255.0f);
+    pixel_ptr[3] = clamp(color.w, 0.0f, 255.0f);
 }