Lines Matching defs:pixels
57 // Decode up to 8 output pixels' x-coordinates and weights.
81 // Gather the 32 32-bit pixels that we'll bilerp into our 8 output pixels.
119 // Treat 32-bit pixels as 4 8-bit values, and expand to 16-bit for room to multiply.
124 // Sum up weighted sample pixels. The naive, redundant math would be,
161 pixels;
163 bilerp(skvx::bit_pun<skvx::Vec<8,uint32_t>>(coords)).store(&pixels);
164 _mm256_maskstore_epi32((int*)colors, active, pixels);
182 // interpolating in X for up to two output pixels (A and B) using _mm_maddubs_epi16().
207 // Returns two pixels, with each color channel in a 16-bit lane of the __m128i.
244 // We can really get going, loading 4 X-pairs at a time to produce 4 output pixels.
255 // Splat each x weight 4x (for each color channel) as wr for pixels on the right at x1,
256 // and sixteen minus that as wl for pixels on the left at x0.
266 // interpolate_in_x_and_y() can produce two output pixels (A and B) at a time
267 // from eight input pixels {A0..A3} and {B0..B3}, arranged in a 2x2 grid for each.
274 // Once more with the other half of the x-weights for two more pixels C,D.
281 // Scale by alpha, pack back together to 8-bit lanes, and write out four pixels!
335 // Load the 4 pixels we're interpolating, in this grid:
341 // We want to calculate a sum of 4 pixels weighted in two directions:
393 // filtering step after we've loaded all four pixels we want to bilerp.