14.2: RAW import rounds to 8 bits with fine noise, in the kernel; the develop sheet shows Importing… until the layer lands

kernels/DitherPixels.h

@@ -51,4 +51,8 @@ int dither_apply(uint8_t *rgba, size_t width, size_t height, size_t stride, cons
 void dither_dots(uint8_t *rgba, size_t width, size_t height, size_t stride, int block, const uint8_t *gap);
 // Adds `glow` (premultiplied RGBA, same layout) over the pixels at `amount`, never past their own alpha.
 void dither_glow(uint8_t *rgba, const uint8_t *glow, size_t width, size_t height, size_t stride, float amount);
+// Rounds 16-bit premultiplied RGBA (`width` × `height`, tightly packed) to 8 bits into `rgba` (`stride` bytes per
+// row), adding noise of about one 8-bit step first so smooth gradients don't come out in steps. The noise is fixed per
+// pixel, so the same input always gives the same output.
+void dither_quantize16(const uint16_t *wide, uint8_t *rgba, size_t width, size_t height, size_t stride);
 #endif