Additional small SIMD-related improvements for Paeth unfilter.

This commit implements 2 changes proposed by @okaneco:

* Changing how `load3` and `load6` work
* Using a manual `if_then_else` instead of `std::simd::Mask::select`

src/filter.rs

@@ -10,9 +10,23 @@ use crate::common::BytesPerPixel;
 #[cfg(feature = "unstable")]
 mod simd {
     use std::simd::{
-        u8x4, u8x8, LaneCount, Simd, SimdInt, SimdOrd, SimdPartialEq, SimdUint, SupportedLaneCount,
+        u8x4, u8x8, LaneCount, Mask, Simd, SimdInt, SimdOrd, SimdPartialEq, SimdUint, SupportedLaneCount,
     };
 
+    /// Equivalent to `condition.select(true_value, false_value)` but generates more compact
+    /// and faster assembly - see https://rust.godbolt.org/z/cbYK5xchf
+    fn if_then_else<const N: usize>(
+        condition: Mask<i16, N>,
+        true_value: Simd<i16, N>,
+        false_value: Simd<i16, N>,
+    ) -> Simd<i16, N>
+    where
+        LaneCount<N>: SupportedLaneCount,
+    {
+        let condition = condition.to_int();
+        (condition & true_value) | (condition & !false_value)
+    }
+
     /// This is an equivalent of the `PaethPredictor` function from
     /// [the spec](http://www.libpng.org/pub/png/spec/1.2/PNG-Filters.html#Filter-type-4-Paeth)
     /// except that it simultaenously calculates the predictor for all SIMD lanes.
@@ -48,9 +62,11 @@ mod simd {
         //     if smalest == pa
         //         then select a
         //         else select (if smallest == pb then select b else select c)
-        smallest
-            .simd_eq(pa)
-            .select(a, smallest.simd_eq(pb).select(b, c))
+        if_then_else(
+            smallest.simd_eq(pa),
+            a,
+            if_then_else(smallest.simd_eq(pb), b, c),
+        )
     }
 
     /// Memory of previous pixels (as needed to unfilter `FilterType::Paeth`).
@@ -88,7 +104,9 @@ mod simd {
     }
 
     fn load3(src: &[u8]) -> u8x4 {
-        u8x4::from_array([src[0], src[1], src[2], 0])
+        let mut temp = [0; 4];
+        temp[..3].copy_from_slice(&src[..3]);
+        u8x4::from_slice(&temp)
     }
 
     fn store3(src: u8x4, dest: &mut [u8]) {
@@ -127,7 +145,9 @@ mod simd {
     }
 
     fn load6(src: &[u8]) -> u8x8 {
-        u8x8::from_array([src[0], src[1], src[2], src[3], src[4], src[5], 0, 0])
+        let mut temp = [0; 8];
+        temp[..6].copy_from_slice(&src[..6]);
+        u8x8::from_slice(&temp)
     }
 
     fn store6(src: u8x8, dest: &mut [u8]) {