refactor(graphics): use an ExactSizeIterator for iter_to_ycbcr

Unfortunately, that doesn't seem to help unrolling & vectorizing: no
perf improvements.

Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>

Author: elmarco

crates/ironrdp-graphics/src/color_conversion.rs

@@ -15,11 +15,12 @@ pub fn ycbcr_to_bgra(input: YCbCrBuffer<'_>, mut output: &mut [u8]) -> io::Resul
     Ok(())
 }
 
-fn iter_to_ycbcr<'a, I, C>(input: I, y: &mut [i16], cb: &mut [i16], cr: &mut [i16], conv: C)
+fn iter_to_ycbcr<'a, const HINT: usize, I, C>(input: I, y: &mut [i16], cb: &mut [i16], cr: &mut [i16], conv: C)
 where
-    I: IntoIterator<Item = &'a [u8]>,
+    I: ExactSizeIterator<Item = &'a [u8]>,
     C: Fn(&[u8]) -> Rgb,
 {
+    assert!(input.len() >= HINT);
     for (i, pixel) in input.into_iter().enumerate() {
         let pixel = YCbCr::from(conv(pixel));
 
@@ -85,7 +86,7 @@ pub fn to_ycbcr(
     let bpp = format.bytes_per_pixel() as usize;
 
     for _ in 0..height {
-        iter_to_ycbcr(input[..width * bpp].chunks_exact(bpp), y, cb, cr, to_rgb);
+        iter_to_ycbcr::<1, _, _>(input[..width * bpp].chunks_exact(bpp), y, cb, cr, to_rgb);
         input = &input[stride..];
         y = &mut y[width..];
         cb = &mut cb[width..];
@@ -144,6 +145,12 @@ impl<'a> Iterator for TileIterator<'a> {
     }
 }
 
+impl ExactSizeIterator for TileIterator<'_> {
+    fn len(&self) -> usize {
+        64 * 64
+    }
+}
+
 #[allow(clippy::too_many_arguments)]
 pub fn to_64x64_ycbcr_tile(
     input: &[u8],
@@ -162,7 +169,7 @@ pub fn to_64x64_ycbcr_tile(
     let bpp = format.bytes_per_pixel() as usize;
 
     let input = TileIterator::new(input, width, height, stride, bpp);
-    iter_to_ycbcr(input, y, cb, cr, to_rgb);
+    iter_to_ycbcr::<4096, _, _>(input, y, cb, cr, to_rgb);
 }
 
 /// Convert a 16-bit RDP color to RGB representation. Input value should be represented in