Add Magic Kernel Sharp (no relation) to resizing kernels

docs/src/content/docs/api-resize.md

@@ -38,6 +38,8 @@ Possible downsizing kernels are:
 - `mitchell`: Use a [Mitchell-Netravali spline](https://www.cs.utexas.edu/~fussell/courses/cs384g-fall2013/lectures/mitchell/Mitchell.pdf).
 - `lanczos2`: Use a [Lanczos kernel](https://en.wikipedia.org/wiki/Lanczos_resampling#Lanczos_kernel) with `a=2`.
 - `lanczos3`: Use a Lanczos kernel with `a=3` (the default).
+- `mks2013`: Use a [Magic Kernel Sharp](https://johncostella.com/magic/mks.pdf) 2013 kernel, as adopted by Facebook.
+- `mks2021`: Use a Magic Kernel Sharp 2021 kernel, with more accurate (reduced) sharpening than the 2013 version.
 
 When upsampling, these kernels map to `nearest`, `linear` and `cubic` interpolators.
 Downsampling kernels without a matching upsampling interpolator map to `cubic`.

docs/src/content/docs/changelog.md

@@ -10,6 +10,8 @@ Requires libvips v8.17.0
 
 * Upgrade to libvips v8.17.0 for upstream bug fixes.
 
+* Add "Magic Kernel Sharp" (no relation) to resizing kernels.
+
 * Expose JPEG 2000 `oneshot` decoder option.
   [#4262](https://github.com/lovell/sharp/pull/4262)
   [@mbklein](https://github.com/mbklein)

lib/index.d.ts

@@ -1729,6 +1729,8 @@ declare namespace sharp {
         mitchell: 'mitchell';
         lanczos2: 'lanczos2';
         lanczos3: 'lanczos3';
+        mks2013: 'mks2013';
+        mks2021: 'mks2021';
     }
 
     interface PresetEnum {

lib/resize.js

@@ -150,6 +150,8 @@ function isResizeExpected (options) {
  * - `mitchell`: Use a [Mitchell-Netravali spline](https://www.cs.utexas.edu/~fussell/courses/cs384g-fall2013/lectures/mitchell/Mitchell.pdf).
  * - `lanczos2`: Use a [Lanczos kernel](https://en.wikipedia.org/wiki/Lanczos_resampling#Lanczos_kernel) with `a=2`.
  * - `lanczos3`: Use a Lanczos kernel with `a=3` (the default).
+ * - `mks2013`: Use a [Magic Kernel Sharp](https://johncostella.com/magic/mks.pdf) 2013 kernel, as adopted by Facebook.
+ * - `mks2021`: Use a Magic Kernel Sharp 2021 kernel, with more accurate (reduced) sharpening than the 2013 version.
  *
  * When upsampling, these kernels map to `nearest`, `linear` and `cubic` interpolators.
  * Downsampling kernels without a matching upsampling interpolator map to `cubic`.

test/types/sharp.test-d.ts

@@ -188,6 +188,8 @@ sharp(input)
     // of the image data in inputBuffer
   });
 
+sharp(input).resize({ kernel: 'mks2013' });
+
 transformer = sharp()
   .resize(200, 200, {
     fit: 'cover',