imagescript 3.3.0 → 3.4.0

package/dist/codecs/vp8-enc-dsp.js

@@ -0,0 +1,932 @@
+import { littleEndian } from "../util/alpha.js";
+export const BPS = 32;
+export const Y_OFF = 0, U_OFF = 16, V_OFF = 24;
+export const YUV_SIZE = BPS * 16, PRED_SIZE = 56 * BPS;
+export const I16DC16 = 0, I16TM16 = 16, I16VE16 = 16 * BPS, I16HE16 = 16 * BPS + 16;
+export const C8DC8 = 32 * BPS, C8TM8 = C8DC8 + 16, C8VE8 = 40 * BPS, C8HE8 = C8VE8 + 16;
+export const I4DC4 = 48 * BPS, I4HD4 = I4DC4 + 4 * BPS, I4TMP = I4HD4 + 8;
+export const I16_MODE_OFFSETS = [I16DC16, I16TM16, I16VE16, I16HE16];
+export const UV_MODE_OFFSETS = [C8DC8, C8TM8, C8VE8, C8HE8];
+export const I4_MODE_OFFSETS = [I4DC4, I4DC4 + 4, I4DC4 + 8, I4DC4 + 12, I4DC4 + 16, I4DC4 + 20, I4DC4 + 24, I4DC4 + 28, I4HD4, I4HD4 + 4];
+export const SCAN = [
+    0, 4, 8, 12, 4 * BPS, 4 * BPS + 4, 4 * BPS + 8, 4 * BPS + 12,
+    8 * BPS, 8 * BPS + 4, 8 * BPS + 8, 8 * BPS + 12, 12 * BPS, 12 * BPS + 4, 12 * BPS + 8, 12 * BPS + 12
+];
+export const SCAN_UV = [0, 4, 4 * BPS, 4 * BPS + 4, 8, 12, 4 * BPS + 8, 4 * BPS + 12];
+export const ZIGZAG = [0, 1, 4, 8, 5, 2, 3, 6, 9, 12, 13, 10, 7, 11, 14, 15];
+const QFIX = 17;
+export const MAX_LEVEL = 2047;
+const clip8 = (v) => v < 0 ? 0 : v > 255 ? 255 : v;
+const GAMMA_TO_LINEAR = new Uint16Array(256), LINEAR_TO_GAMMA = new Int32Array(33);
+for (let v = 0; v <= 255; v++)
+    GAMMA_TO_LINEAR[v] = Math.trunc(Math.pow((1 / 255) * v, 0.8) * 4095 + 0.5);
+for (let v = 0; v <= 32; v++)
+    LINEAR_TO_GAMMA[v] = Math.trunc(255 * Math.pow((128 / 4095) * v, 1 / 0.8) + 0.5);
+const INV_ALPHA = new Uint32Array(4 * 255 + 1);
+for (let a = 1; a <= 4 * 255; a++)
+    INV_ALPHA[a] = Math.floor((1 << 19) / a);
+function linearToGamma(base, shift) {
+    const v = base << shift;
+    const pos = v >> 9, x = v & 511;
+    return (LINEAR_TO_GAMMA[pos + 1] * x + LINEAR_TO_GAMMA[pos] * (512 - x) + 64) >> 7;
+}
+const SUM_TO_GAMMA = new Uint16Array(4 * 4095 + 1);
+for (let v = 0; v < SUM_TO_GAMMA.length; v++)
+    SUM_TO_GAMMA[v] = linearToGamma(v, 0);
+const rgbToY = (r, g, b) => (16839 * r + 33059 * g + 6420 * b + 0x8000 + 0x100000) >> 16;
+const clipUV = (uv) => {
+    uv = (uv + (1 << 17) + (128 << 18)) >> 18;
+    return uv < 0 ? 0 : uv > 255 ? 255 : uv;
+};
+function weighted(rgba, o, step, next, total) {
+    const g = GAMMA_TO_LINEAR, a = o + 3 - (o & 3);
+    const sum = rgba[a] * g[rgba[o]] + rgba[a + step] * g[rgba[o + step]] + rgba[a + next] * g[rgba[o + next]] + rgba[a + next + step] * g[rgba[o + next + step]];
+    return linearToGamma((sum * INV_ALPHA[total]) >>> 17, 0);
+}
+export function rgbaToYUV(rgba, width, height) {
+    const uvWidth = (width + 1) >> 1, uvHeight = (height + 1) >> 1;
+    const y = new Uint8Array(width * height), u = new Uint8Array(uvWidth * uvHeight), v = new Uint8Array(uvWidth * uvHeight);
+    let hasAlpha = false;
+    for (let i = 3; i < width * height * 4; i += 4)
+        if (rgba[i] !== 255) {
+            hasAlpha = true;
+            break;
+        }
+    const a = hasAlpha ? new Uint8Array(width * height) : null;
+    const stride = width * 4;
+    if (a)
+        for (let j = 0; j < height; j++)
+            alphaRow(rgba, j * stride, a, j * width, width);
+    const words = littleEndian && rgba.byteOffset % 4 === 0 ? new Uint32Array(rgba.buffer, rgba.byteOffset, width * height) : null;
+    for (let j = 0; j < uvHeight; j++) {
+        const row = 2 * j * stride;
+        const next = 2 * j + 1 < height ? stride : 0;
+        let alphaRows = false;
+        if (a)
+            for (let i = row + 3; i < row + next + stride; i += 4)
+                if (rgba[i] !== 255) {
+                    alphaRows = true;
+                    break;
+                }
+        if (words && !alphaRows) {
+            pairRow(words, 2 * j * width, next >> 2, y, u, v, j * uvWidth, width);
+            continue;
+        }
+        lumaRow(rgba, row, y, 2 * j * width, width);
+        if (next)
+            lumaRow(rgba, row + next, y, (2 * j + 1) * width, width);
+        if (alphaRows)
+            chromaRowAlpha(rgba, row, next, u, v, j * uvWidth, width);
+        else
+            chromaRow(rgba, row, next, u, v, j * uvWidth, width);
+    }
+    const picture = { width, height, y, u, v, a, uvStride: uvWidth };
+    if (a)
+        cleanupTransparentArea(picture);
+    return picture;
+}
+function lumaRow(rgba, s, y, d, width) {
+    for (let x = 0; x < width; x++, s += 4)
+        y[d + x] = rgbToY(rgba[s], rgba[s + 1], rgba[s + 2]);
+}
+function alphaRow(rgba, s, a, d, width) {
+    for (let x = 0; x < width; x++, s += 4)
+        a[d + x] = rgba[s + 3];
+}
+function pairRow(words, p, next, y, u, v, d, width) {
+    const g = GAMMA_TO_LINEAR, pairs = width >> 1;
+    for (let i = 0; i < pairs; i++, p += 2) {
+        const p0 = words[p], p1 = words[p + 1], p2 = words[p + next], p3 = words[p + next + 1];
+        const r0 = p0 & 255, g0 = p0 >> 8 & 255, b0 = p0 >> 16 & 255, r1 = p1 & 255, g1 = p1 >> 8 & 255, b1 = p1 >> 16 & 255;
+        const r2 = p2 & 255, g2 = p2 >> 8 & 255, b2 = p2 >> 16 & 255, r3 = p3 & 255, g3 = p3 >> 8 & 255, b3 = p3 >> 16 & 255;
+        y[p] = rgbToY(r0, g0, b0);
+        y[p + 1] = rgbToY(r1, g1, b1);
+        if (next) {
+            y[p + next] = rgbToY(r2, g2, b2);
+            y[p + next + 1] = rgbToY(r3, g3, b3);
+        }
+        const r = SUM_TO_GAMMA[g[r0] + g[r1] + g[r2] + g[r3]];
+        const gr = SUM_TO_GAMMA[g[g0] + g[g1] + g[g2] + g[g3]];
+        const b = SUM_TO_GAMMA[g[b0] + g[b1] + g[b2] + g[b3]];
+        u[d + i] = clipUV(-9719 * r - 19081 * gr + 28800 * b);
+        v[d + i] = clipUV(28800 * r - 24116 * gr - 4684 * b);
+    }
+    if (width & 1) {
+        const p0 = words[p], p2 = words[p + next];
+        const r0 = p0 & 255, g0 = p0 >> 8 & 255, b0 = p0 >> 16 & 255, r2 = p2 & 255, g2 = p2 >> 8 & 255, b2 = p2 >> 16 & 255;
+        y[p] = rgbToY(r0, g0, b0);
+        if (next)
+            y[p + next] = rgbToY(r2, g2, b2);
+        const r = linearToGamma(g[r0] + g[r2], 1), gr = linearToGamma(g[g0] + g[g2], 1), b = linearToGamma(g[b0] + g[b2], 1);
+        u[d + pairs] = clipUV(-9719 * r - 19081 * gr + 28800 * b);
+        v[d + pairs] = clipUV(28800 * r - 24116 * gr - 4684 * b);
+    }
+}
+function chromaRow(rgba, row, next, u, v, d, width) {
+    const g = GAMMA_TO_LINEAR, pairs = width >> 1;
+    let o = row;
+    for (let i = 0; i < pairs; i++, o += 8) {
+        const r = linearToGamma(g[rgba[o]] + g[rgba[o + 4]] + g[rgba[o + next]] + g[rgba[o + next + 4]], 0);
+        const gr = linearToGamma(g[rgba[o + 1]] + g[rgba[o + 5]] + g[rgba[o + next + 1]] + g[rgba[o + next + 5]], 0);
+        const b = linearToGamma(g[rgba[o + 2]] + g[rgba[o + 6]] + g[rgba[o + next + 2]] + g[rgba[o + next + 6]], 0);
+        u[d + i] = clipUV(-9719 * r - 19081 * gr + 28800 * b);
+        v[d + i] = clipUV(28800 * r - 24116 * gr - 4684 * b);
+    }
+    if (width & 1) {
+        const r = linearToGamma(g[rgba[o]] + g[rgba[o + next]], 1);
+        const gr = linearToGamma(g[rgba[o + 1]] + g[rgba[o + next + 1]], 1);
+        const b = linearToGamma(g[rgba[o + 2]] + g[rgba[o + next + 2]], 1);
+        u[d + pairs] = clipUV(-9719 * r - 19081 * gr + 28800 * b);
+        v[d + pairs] = clipUV(28800 * r - 24116 * gr - 4684 * b);
+    }
+}
+function chromaRowAlpha(rgba, row, next, u, v, d, width) {
+    const g = GAMMA_TO_LINEAR, uvWidth = (width + 1) >> 1;
+    for (let i = 0; i < uvWidth; i++) {
+        const o = row + 8 * i, pair = 2 * i + 1 < width;
+        const step = pair ? 4 : 0;
+        let r, gr, b;
+        const total = pair ? rgba[o + 3] + rgba[o + 7] + rgba[o + next + 3] + rgba[o + next + 7] : 2 * (rgba[o + 3] + rgba[o + next + 3]);
+        if (total === 4 * 255 || total === 0) {
+            if (pair) {
+                r = linearToGamma(g[rgba[o]] + g[rgba[o + 4]] + g[rgba[o + next]] + g[rgba[o + next + 4]], 0);
+                gr = linearToGamma(g[rgba[o + 1]] + g[rgba[o + 5]] + g[rgba[o + next + 1]] + g[rgba[o + next + 5]], 0);
+                b = linearToGamma(g[rgba[o + 2]] + g[rgba[o + 6]] + g[rgba[o + next + 2]] + g[rgba[o + next + 6]], 0);
+            }
+            else {
+                r = linearToGamma(g[rgba[o]] + g[rgba[o + next]], 1);
+                gr = linearToGamma(g[rgba[o + 1]] + g[rgba[o + next + 1]], 1);
+                b = linearToGamma(g[rgba[o + 2]] + g[rgba[o + next + 2]], 1);
+            }
+        }
+        else {
+            r = weighted(rgba, o, step, next, total);
+            gr = weighted(rgba, o + 1, step, next, total);
+            b = weighted(rgba, o + 2, step, next, total);
+        }
+        u[d + i] = clipUV(-9719 * r - 19081 * gr + 28800 * b);
+        v[d + i] = clipUV(28800 * r - 24116 * gr - 4684 * b);
+    }
+}
+function smoothenBlock(p, x0, y0, w, h) {
+    const a = p.a;
+    let sum = 0, count = 0;
+    for (let y = y0; y < y0 + h; y++) {
+        for (let x = x0; x < x0 + w; x++) {
+            if (a[y * p.width + x]) {
+                count++;
+                sum += p.y[y * p.width + x];
+            }
+        }
+    }
+    if (count > 0 && count < w * h) {
+        const average = Math.floor(sum / count);
+        for (let y = y0; y < y0 + h; y++)
+            for (let x = x0; x < x0 + w; x++)
+                if (!a[y * p.width + x])
+                    p.y[y * p.width + x] = average;
+    }
+    return count === 0;
+}
+function cleanupTransparentArea(p) {
+    const { width, height, uvStride } = p;
+    const flatten = (plane, stride, x0, y0, size, value) => {
+        for (let y = y0; y < y0 + size; y++)
+            plane.fill(value, y * stride + x0, y * stride + x0 + size);
+    };
+    let y = 0;
+    for (; y + 8 <= height; y += 8) {
+        let reset = true, vy = 0, vu = 0, vv = 0;
+        let x = 0;
+        for (; x + 8 <= width; x += 8) {
+            if (smoothenBlock(p, x, y, 8, 8)) {
+                if (reset) {
+                    vy = p.y[y * width + x];
+                    vu = p.u[(y >> 1) * uvStride + (x >> 1)];
+                    vv = p.v[(y >> 1) * uvStride + (x >> 1)];
+                    reset = false;
+                }
+                flatten(p.y, width, x, y, 8, vy);
+                flatten(p.u, uvStride, x >> 1, y >> 1, 4, vu);
+                flatten(p.v, uvStride, x >> 1, y >> 1, 4, vv);
+            }
+            else
+                reset = true;
+        }
+        if (x < width)
+            smoothenBlock(p, x, y, width - x, 8);
+    }
+    if (y < height) {
+        let x = 0;
+        for (; x + 8 <= width; x += 8)
+            smoothenBlock(p, x, y, 8, height - y);
+        if (x < width)
+            smoothenBlock(p, x, y, width - x, height - y);
+    }
+}
+function fill(dst, d, value, size) {
+    for (let j = 0; j < size; j++, d += BPS)
+        for (let x = 0; x < size; x++)
+            dst[d + x] = value;
+}
+function verticalPred(dst, d, top, t, size) {
+    if (top)
+        for (let j = 0; j < size; j++, d += BPS)
+            for (let x = 0; x < size; x++)
+                dst[d + x] = top[t + x];
+    else
+        fill(dst, d, 127, size);
+}
+function horizontalPred(dst, d, left, l, size) {
+    if (left) {
+        for (let j = 0; j < size; j++, d += BPS) {
+            const v = left[l + j];
+            for (let x = 0; x < size; x++)
+                dst[d + x] = v;
+        }
+    }
+    else
+        fill(dst, d, 129, size);
+}
+function trueMotion(dst, d, left, l, top, t, size) {
+    if (left) {
+        if (top) {
+            const corner = left[l - 1];
+            for (let y = 0; y < size; y++) {
+                const base = left[l + y] - corner;
+                for (let x = 0; x < size; x++)
+                    dst[d + y * BPS + x] = clip8(top[t + x] + base);
+            }
+        }
+        else
+            horizontalPred(dst, d, left, l, size);
+    }
+    else if (top)
+        verticalPred(dst, d, top, t, size);
+    else
+        fill(dst, d, 129, size);
+}
+function dcMode(dst, d, left, l, top, t, size, round, shift) {
+    let dc = 0;
+    if (top) {
+        for (let j = 0; j < size; j++)
+            dc += top[t + j];
+        if (left)
+            for (let j = 0; j < size; j++)
+                dc += left[l + j];
+        else
+            dc += dc;
+        dc = (dc + round) >> shift;
+    }
+    else if (left) {
+        for (let j = 0; j < size; j++)
+            dc += left[l + j];
+        dc += dc;
+        dc = (dc + round) >> shift;
+    }
+    else
+        dc = 0x80;
+    fill(dst, d, dc, size);
+}
+export function predLuma16(dst, left, l, top, t, modes = 15) {
+    if (modes & 1)
+        dcMode(dst, I16DC16, left, l, top, t, 16, 16, 5);
+    if (modes & 2)
+        trueMotion(dst, I16TM16, left, l, top, t, 16);
+    if (modes & 4)
+        verticalPred(dst, I16VE16, top, t, 16);
+    if (modes & 8)
+        horizontalPred(dst, I16HE16, left, l, 16);
+}
+export function predChroma8(dst, uLeft, ul, vLeft, vl, top, t, modes = 15) {
+    for (let c = 0; c < 2; c++) {
+        const left = c ? vLeft : uLeft, l = c ? vl : ul, tt = t + 8 * c, o = 8 * c;
+        if (modes & 1)
+            dcMode(dst, C8DC8 + o, left, l, top, tt, 8, 8, 4);
+        if (modes & 2)
+            trueMotion(dst, C8TM8 + o, left, l, top, tt, 8);
+        if (modes & 4)
+            verticalPred(dst, C8VE8 + o, top, tt, 8);
+        if (modes & 8)
+            horizontalPred(dst, C8HE8 + o, left, l, 8);
+    }
+}
+const avg3 = (a, b, c) => (a + 2 * b + c + 2) >> 2;
+const avg2 = (a, b) => (a + b + 1) >> 1;
+const rowSums = new Int32Array(16), columnSums = new Int32Array(16);
+let sourceSquares = 0, sourceSum = 0;
+function sourceSums16(src, s) {
+    let c0 = 0, c1 = 0, c2 = 0, c3 = 0, c4 = 0, c5 = 0, c6 = 0, c7 = 0, c8 = 0, c9 = 0, c10 = 0, c11 = 0, c12 = 0, c13 = 0, c14 = 0, c15 = 0, squares = 0, sum = 0;
+    for (let y = 0, o = s; y < 16; y++, o += BPS) {
+        const v0 = src[o], v1 = src[o + 1], v2 = src[o + 2], v3 = src[o + 3], v4 = src[o + 4], v5 = src[o + 5], v6 = src[o + 6], v7 = src[o + 7], v8 = src[o + 8], v9 = src[o + 9], v10 = src[o + 10], v11 = src[o + 11], v12 = src[o + 12], v13 = src[o + 13], v14 = src[o + 14], v15 = src[o + 15];
+        const row = v0 + v1 + v2 + v3 + v4 + v5 + v6 + v7 + v8 + v9 + v10 + v11 + v12 + v13 + v14 + v15;
+        rowSums[y] = row;
+        sum += row;
+        squares += v0 * v0 + v1 * v1 + v2 * v2 + v3 * v3 + v4 * v4 + v5 * v5 + v6 * v6 + v7 * v7 + v8 * v8 + v9 * v9 + v10 * v10 + v11 * v11 + v12 * v12 + v13 * v13 + v14 * v14 + v15 * v15;
+        c0 += v0;
+        c1 += v1;
+        c2 += v2;
+        c3 += v3;
+        c4 += v4;
+        c5 += v5;
+        c6 += v6;
+        c7 += v7;
+        c8 += v8;
+        c9 += v9;
+        c10 += v10;
+        c11 += v11;
+        c12 += v12;
+        c13 += v13;
+        c14 += v14;
+        c15 += v15;
+    }
+    columnSums[0] = c0;
+    columnSums[1] = c1;
+    columnSums[2] = c2;
+    columnSums[3] = c3;
+    columnSums[4] = c4;
+    columnSums[5] = c5;
+    columnSums[6] = c6;
+    columnSums[7] = c7;
+    columnSums[8] = c8;
+    columnSums[9] = c9;
+    columnSums[10] = c10;
+    columnSums[11] = c11;
+    columnSums[12] = c12;
+    columnSums[13] = c13;
+    columnSums[14] = c14;
+    columnSums[15] = c15;
+    sourceSquares = squares;
+    sourceSum = sum;
+}
+function sourceSums8(src, s) {
+    let c0 = 0, c1 = 0, c2 = 0, c3 = 0, c4 = 0, c5 = 0, c6 = 0, c7 = 0, squares = 0, sum = 0;
+    for (let y = 0, o = s; y < 8; y++, o += BPS) {
+        const v0 = src[o], v1 = src[o + 1], v2 = src[o + 2], v3 = src[o + 3], v4 = src[o + 4], v5 = src[o + 5], v6 = src[o + 6], v7 = src[o + 7];
+        const row = v0 + v1 + v2 + v3 + v4 + v5 + v6 + v7;
+        rowSums[y] = row;
+        sum += row;
+        squares += v0 * v0 + v1 * v1 + v2 * v2 + v3 * v3 + v4 * v4 + v5 * v5 + v6 * v6 + v7 * v7;
+        c0 += v0;
+        c1 += v1;
+        c2 += v2;
+        c3 += v3;
+        c4 += v4;
+        c5 += v5;
+        c6 += v6;
+        c7 += v7;
+    }
+    columnSums[0] = c0;
+    columnSums[1] = c1;
+    columnSums[2] = c2;
+    columnSums[3] = c3;
+    columnSums[4] = c4;
+    columnSums[5] = c5;
+    columnSums[6] = c6;
+    columnSums[7] = c7;
+    sourceSquares = squares;
+    sourceSum = sum;
+}
+export function predictionSSEs(src, s, size, planes, left0, left1, l, top, t, out) {
+    out.fill(0);
+    const n = size * size, shift = size === 16 ? 5 : 4;
+    for (let c = 0; c < planes; c++, s += 8, t += 8) {
+        const left = c ? left1 : left0;
+        if (size === 16)
+            sourceSums16(src, s);
+        else
+            sourceSums8(src, s);
+        const squares = sourceSquares, sum = sourceSum;
+        const fill = (v) => squares - 2 * v * sum + n * v * v;
+        let topSum = 0, topSquares = 0, topCross = 0, leftSum = 0, leftSquares = 0, leftCross = 0;
+        if (top) {
+            for (let x = 0; x < size; x++) {
+                const v = top[t + x];
+                topSum += v;
+                topSquares += v * v;
+                topCross += v * columnSums[x];
+            }
+        }
+        if (left) {
+            for (let y = 0; y < size; y++) {
+                const v = left[l + y];
+                leftSum += v;
+                leftSquares += v * v;
+                leftCross += v * rowSums[y];
+            }
+        }
+        const vertical = top ? squares - 2 * topCross + size * topSquares : fill(127);
+        const horizontal = left ? squares - 2 * leftCross + size * leftSquares : fill(129);
+        let dc;
+        if (top)
+            dc = (left ? topSum + leftSum : 2 * topSum) + size >> shift;
+        else if (left)
+            dc = 2 * leftSum + size >> shift;
+        else
+            dc = 0x80;
+        let tm;
+        if (left && top) {
+            const corner = left[l - 1];
+            let tMin = 255, tMax = 0, bMin = 255, bMax = -255, bSum = 0, bSquares = 0, bCross = 0;
+            for (let x = 0; x < size; x++) {
+                const v = top[t + x];
+                if (v < tMin)
+                    tMin = v;
+                if (v > tMax)
+                    tMax = v;
+            }
+            for (let y = 0; y < size; y++) {
+                const b = left[l + y] - corner;
+                if (b < bMin)
+                    bMin = b;
+                if (b > bMax)
+                    bMax = b;
+                bSum += b;
+                bSquares += b * b;
+                bCross += b * rowSums[y];
+            }
+            if (tMin + bMin >= 0 && tMax + bMax <= 255) {
+                tm = squares - 2 * topCross - 2 * bCross + size * topSquares + 2 * topSum * bSum + size * bSquares;
+            }
+            else {
+                tm = 0;
+                for (let y = 0, o = s; y < size; y++, o += BPS) {
+                    const b = left[l + y] - corner;
+                    for (let x = 0; x < size; x++) {
+                        const d = src[o + x] - clip8(top[t + x] + b);
+                        tm += d * d;
+                    }
+                }
+            }
+        }
+        else
+            tm = left ? horizontal : top ? vertical : fill(129);
+        out[0] += fill(dc);
+        out[1] += tm;
+        out[2] += vertical;
+        out[3] += horizontal;
+    }
+}
+export function predLuma4(dst, top, t) {
+    const X = top[t - 1], I = top[t - 2], J = top[t - 3], K = top[t - 4], L = top[t - 5];
+    const A = top[t], B = top[t + 1], C = top[t + 2], D = top[t + 3], E = top[t + 4], F = top[t + 5], G = top[t + 6], H = top[t + 7];
+    const R0 = 0, R1 = BPS, R2 = 2 * BPS, R3 = 3 * BPS;
+    const dc = (4 + A + B + C + D + I + J + K + L) >> 3;
+    for (let y = 0, o = I4DC4; y < 4; y++, o += BPS)
+        dst[o] = dst[o + 1] = dst[o + 2] = dst[o + 3] = dc;
+    for (let y = 0; y < 4; y++) {
+        const base = top[t - 2 - y] - X, o = I4DC4 + 4 + y * BPS;
+        dst[o] = clip8(A + base);
+        dst[o + 1] = clip8(B + base);
+        dst[o + 2] = clip8(C + base);
+        dst[o + 3] = clip8(D + base);
+    }
+    let o = I4DC4 + 8;
+    const v0 = avg3(X, A, B), v1 = avg3(A, B, C), v2 = avg3(B, C, D), v3 = avg3(C, D, E);
+    for (let y = 0; y < 4; y++, o += BPS) {
+        dst[o] = v0;
+        dst[o + 1] = v1;
+        dst[o + 2] = v2;
+        dst[o + 3] = v3;
+    }
+    o = I4DC4 + 12;
+    const he0 = avg3(X, I, J), he1 = avg3(I, J, K), he2 = avg3(J, K, L), he3 = avg3(K, L, L);
+    dst[o + R0] = dst[o + R0 + 1] = dst[o + R0 + 2] = dst[o + R0 + 3] = he0;
+    dst[o + R1] = dst[o + R1 + 1] = dst[o + R1 + 2] = dst[o + R1 + 3] = he1;
+    dst[o + R2] = dst[o + R2 + 1] = dst[o + R2 + 2] = dst[o + R2 + 3] = he2;
+    dst[o + R3] = dst[o + R3 + 1] = dst[o + R3 + 2] = dst[o + R3 + 3] = he3;
+    o = I4DC4 + 16;
+    const xij = avg3(X, I, J), ijk = avg3(I, J, K), axi = avg3(A, X, I), bax = avg3(B, A, X);
+    dst[o + R3] = avg3(J, K, L);
+    dst[o + R2] = dst[o + R3 + 1] = ijk;
+    dst[o + R1] = dst[o + R2 + 1] = dst[o + R3 + 2] = xij;
+    dst[o] = dst[o + R1 + 1] = dst[o + R2 + 2] = dst[o + R3 + 3] = axi;
+    dst[o + 1] = dst[o + R1 + 2] = dst[o + R2 + 3] = bax;
+    dst[o + 2] = dst[o + R1 + 3] = avg3(C, B, A);
+    dst[o + 3] = avg3(D, C, B);
+    o = I4DC4 + 20;
+    dst[o] = dst[o + R2 + 1] = avg2(X, A);
+    dst[o + 1] = dst[o + R2 + 2] = avg2(A, B);
+    dst[o + 2] = dst[o + R2 + 3] = avg2(B, C);
+    dst[o + 3] = avg2(C, D);
+    dst[o + R3] = avg3(K, J, I);
+    dst[o + R2] = avg3(J, I, X);
+    dst[o + R1] = dst[o + R3 + 1] = axi;
+    dst[o + R1 + 1] = dst[o + R3 + 2] = avg3(X, A, B);
+    dst[o + R1 + 2] = dst[o + R3 + 3] = avg3(A, B, C);
+    dst[o + R1 + 3] = avg3(B, C, D);
+    o = I4DC4 + 24;
+    const bcd = avg3(B, C, D), cde = avg3(C, D, E), def = avg3(D, E, F), efg = avg3(E, F, G);
+    dst[o] = avg3(A, B, C);
+    dst[o + 1] = dst[o + R1] = bcd;
+    dst[o + 2] = dst[o + R1 + 1] = dst[o + R2] = cde;
+    dst[o + 3] = dst[o + R1 + 2] = dst[o + R2 + 1] = dst[o + R3] = def;
+    dst[o + R1 + 3] = dst[o + R2 + 2] = dst[o + R3 + 1] = efg;
+    dst[o + R2 + 3] = dst[o + R3 + 2] = avg3(F, G, H);
+    dst[o + R3 + 3] = avg3(G, H, H);
+    o = I4DC4 + 28;
+    dst[o] = avg2(A, B);
+    dst[o + 1] = dst[o + R2] = avg2(B, C);
+    dst[o + 2] = dst[o + R2 + 1] = avg2(C, D);
+    dst[o + 3] = dst[o + R2 + 2] = avg2(D, E);
+    dst[o + R1] = avg3(A, B, C);
+    dst[o + R1 + 1] = dst[o + R3] = bcd;
+    dst[o + R1 + 2] = dst[o + R3 + 1] = cde;
+    dst[o + R1 + 3] = dst[o + R3 + 2] = def;
+    dst[o + R2 + 3] = efg;
+    dst[o + R3 + 3] = avg3(F, G, H);
+    o = I4HD4;
+    const jix = avg3(J, I, X), kji = avg3(K, J, I);
+    dst[o] = dst[o + R1 + 2] = avg2(I, X);
+    dst[o + R1] = dst[o + R2 + 2] = avg2(J, I);
+    dst[o + R2] = dst[o + R3 + 2] = avg2(K, J);
+    dst[o + R3] = avg2(L, K);
+    dst[o + 3] = avg3(A, B, C);
+    dst[o + 2] = avg3(X, A, B);
+    dst[o + 1] = dst[o + R1 + 3] = axi;
+    dst[o + R1 + 1] = dst[o + R2 + 3] = jix;
+    dst[o + R2 + 1] = dst[o + R3 + 3] = kji;
+    dst[o + R3 + 1] = avg3(L, K, J);
+    o = I4HD4 + 4;
+    dst[o] = avg2(I, J);
+    dst[o + 2] = dst[o + R1] = avg2(J, K);
+    dst[o + R1 + 2] = dst[o + R2] = avg2(K, L);
+    dst[o + 1] = ijk;
+    dst[o + 3] = dst[o + R1 + 1] = avg3(J, K, L);
+    dst[o + R1 + 3] = dst[o + R2 + 1] = avg3(K, L, L);
+    dst[o + R2 + 3] = dst[o + R2 + 2] = dst[o + R3] = dst[o + R3 + 1] = dst[o + R3 + 2] = dst[o + R3 + 3] = L;
+}
+export function iTransform(ref, r, input, i, dst, d, two) {
+    for (let n = 0; n < (two ? 2 : 1); n++, r += 4, i += 16, d += 4) {
+        let ac = 0;
+        for (let k = 1; k < 16; k++)
+            ac |= input[i + k];
+        if (!ac) {
+            const dc = (input[i] + 4) >> 3;
+            for (let k = 0; k < 4; k++) {
+                const o = k * BPS;
+                dst[d + o] = clip8(ref[r + o] + dc);
+                dst[d + o + 1] = clip8(ref[r + o + 1] + dc);
+                dst[d + o + 2] = clip8(ref[r + o + 2] + dc);
+                dst[d + o + 3] = clip8(ref[r + o + 3] + dc);
+            }
+            continue;
+        }
+        const p0 = input[i + 0], q0 = input[i + 4], u0 = input[i + 8], w0 = input[i + 12];
+        const a0 = p0 + u0, b0 = p0 - u0;
+        const c0 = ((q0 * 35468) >> 16) - (((w0 * 20091) >> 16) + w0), e0 = (((q0 * 20091) >> 16) + q0) + ((w0 * 35468) >> 16);
+        const v00 = a0 + e0, v01 = b0 + c0, v02 = b0 - c0, v03 = a0 - e0;
+        const p1 = input[i + 1], q1 = input[i + 5], u1 = input[i + 9], w1 = input[i + 13];
+        const a1 = p1 + u1, b1 = p1 - u1;
+        const c1 = ((q1 * 35468) >> 16) - (((w1 * 20091) >> 16) + w1), e1 = (((q1 * 20091) >> 16) + q1) + ((w1 * 35468) >> 16);
+        const v10 = a1 + e1, v11 = b1 + c1, v12 = b1 - c1, v13 = a1 - e1;
+        const p2 = input[i + 2], q2 = input[i + 6], u2 = input[i + 10], w2 = input[i + 14];
+        const a2 = p2 + u2, b2 = p2 - u2;
+        const c2 = ((q2 * 35468) >> 16) - (((w2 * 20091) >> 16) + w2), e2 = (((q2 * 20091) >> 16) + q2) + ((w2 * 35468) >> 16);
+        const v20 = a2 + e2, v21 = b2 + c2, v22 = b2 - c2, v23 = a2 - e2;
+        const p3 = input[i + 3], q3 = input[i + 7], u3 = input[i + 11], w3 = input[i + 15];
+        const a3 = p3 + u3, b3 = p3 - u3;
+        const c3 = ((q3 * 35468) >> 16) - (((w3 * 20091) >> 16) + w3), e3 = (((q3 * 20091) >> 16) + q3) + ((w3 * 35468) >> 16);
+        const v30 = a3 + e3, v31 = b3 + c3, v32 = b3 - c3, v33 = a3 - e3;
+        {
+            const dc = v00 + 4;
+            const a = dc + v20, b = dc - v20;
+            const c = ((v10 * 35468) >> 16) - (((v30 * 20091) >> 16) + v30), e = (((v10 * 20091) >> 16) + v10) + ((v30 * 35468) >> 16);
+            dst[d] = clip8(ref[r] + ((a + e) >> 3));
+            dst[d + 1] = clip8(ref[r + 1] + ((b + c) >> 3));
+            dst[d + 2] = clip8(ref[r + 2] + ((b - c) >> 3));
+            dst[d + 3] = clip8(ref[r + 3] + ((a - e) >> 3));
+        }
+        {
+            const dc = v01 + 4;
+            const a = dc + v21, b = dc - v21;
+            const c = ((v11 * 35468) >> 16) - (((v31 * 20091) >> 16) + v31), e = (((v11 * 20091) >> 16) + v11) + ((v31 * 35468) >> 16);
+            dst[d + 1 * BPS] = clip8(ref[r + 1 * BPS] + ((a + e) >> 3));
+            dst[d + 1 * BPS + 1] = clip8(ref[r + 1 * BPS + 1] + ((b + c) >> 3));
+            dst[d + 1 * BPS + 2] = clip8(ref[r + 1 * BPS + 2] + ((b - c) >> 3));
+            dst[d + 1 * BPS + 3] = clip8(ref[r + 1 * BPS + 3] + ((a - e) >> 3));
+        }
+        {
+            const dc = v02 + 4;
+            const a = dc + v22, b = dc - v22;
+            const c = ((v12 * 35468) >> 16) - (((v32 * 20091) >> 16) + v32), e = (((v12 * 20091) >> 16) + v12) + ((v32 * 35468) >> 16);
+            dst[d + 2 * BPS] = clip8(ref[r + 2 * BPS] + ((a + e) >> 3));
+            dst[d + 2 * BPS + 1] = clip8(ref[r + 2 * BPS + 1] + ((b + c) >> 3));
+            dst[d + 2 * BPS + 2] = clip8(ref[r + 2 * BPS + 2] + ((b - c) >> 3));
+            dst[d + 2 * BPS + 3] = clip8(ref[r + 2 * BPS + 3] + ((a - e) >> 3));
+        }
+        {
+            const dc = v03 + 4;
+            const a = dc + v23, b = dc - v23;
+            const c = ((v13 * 35468) >> 16) - (((v33 * 20091) >> 16) + v33), e = (((v13 * 20091) >> 16) + v13) + ((v33 * 35468) >> 16);
+            dst[d + 3 * BPS] = clip8(ref[r + 3 * BPS] + ((a + e) >> 3));
+            dst[d + 3 * BPS + 1] = clip8(ref[r + 3 * BPS + 1] + ((b + c) >> 3));
+            dst[d + 3 * BPS + 2] = clip8(ref[r + 3 * BPS + 2] + ((b - c) >> 3));
+            dst[d + 3 * BPS + 3] = clip8(ref[r + 3 * BPS + 3] + ((a - e) >> 3));
+        }
+    }
+}
+export function fTransform(src, s, ref, r, out, o) {
+    const d00 = src[s] - ref[r], d01 = src[s + 1] - ref[r + 1], d02 = src[s + 2] - ref[r + 2], d03 = src[s + 3] - ref[r + 3];
+    const a00 = d00 + d03, a01 = d01 + d02, a02 = d01 - d02, a03 = d00 - d03;
+    const t00 = (a00 + a01) * 8, t01 = (a02 * 2217 + a03 * 5352 + 1812) >> 9, t02 = (a00 - a01) * 8, t03 = (a03 * 2217 - a02 * 5352 + 937) >> 9;
+    const d10 = src[s + 1 * BPS] - ref[r + 1 * BPS], d11 = src[s + 1 * BPS + 1] - ref[r + 1 * BPS + 1], d12 = src[s + 1 * BPS + 2] - ref[r + 1 * BPS + 2], d13 = src[s + 1 * BPS + 3] - ref[r + 1 * BPS + 3];
+    const a10 = d10 + d13, a11 = d11 + d12, a12 = d11 - d12, a13 = d10 - d13;
+    const t10 = (a10 + a11) * 8, t11 = (a12 * 2217 + a13 * 5352 + 1812) >> 9, t12 = (a10 - a11) * 8, t13 = (a13 * 2217 - a12 * 5352 + 937) >> 9;
+    const d20 = src[s + 2 * BPS] - ref[r + 2 * BPS], d21 = src[s + 2 * BPS + 1] - ref[r + 2 * BPS + 1], d22 = src[s + 2 * BPS + 2] - ref[r + 2 * BPS + 2], d23 = src[s + 2 * BPS + 3] - ref[r + 2 * BPS + 3];
+    const a20 = d20 + d23, a21 = d21 + d22, a22 = d21 - d22, a23 = d20 - d23;
+    const t20 = (a20 + a21) * 8, t21 = (a22 * 2217 + a23 * 5352 + 1812) >> 9, t22 = (a20 - a21) * 8, t23 = (a23 * 2217 - a22 * 5352 + 937) >> 9;
+    const d30 = src[s + 3 * BPS] - ref[r + 3 * BPS], d31 = src[s + 3 * BPS + 1] - ref[r + 3 * BPS + 1], d32 = src[s + 3 * BPS + 2] - ref[r + 3 * BPS + 2], d33 = src[s + 3 * BPS + 3] - ref[r + 3 * BPS + 3];
+    const a30 = d30 + d33, a31 = d31 + d32, a32 = d31 - d32, a33 = d30 - d33;
+    const t30 = (a30 + a31) * 8, t31 = (a32 * 2217 + a33 * 5352 + 1812) >> 9, t32 = (a30 - a31) * 8, t33 = (a33 * 2217 - a32 * 5352 + 937) >> 9;
+    const b00 = t00 + t30, b01 = t10 + t20, b02 = t10 - t20, b03 = t00 - t30;
+    out[o + 0] = (b00 + b01 + 7) >> 4;
+    out[o + 4] = ((b02 * 2217 + b03 * 5352 + 12000) >> 16) + (b03 !== 0 ? 1 : 0);
+    out[o + 8] = (b00 - b01 + 7) >> 4;
+    out[o + 12] = (b03 * 2217 - b02 * 5352 + 51000) >> 16;
+    const b10 = t01 + t31, b11 = t11 + t21, b12 = t11 - t21, b13 = t01 - t31;
+    out[o + 1] = (b10 + b11 + 7) >> 4;
+    out[o + 5] = ((b12 * 2217 + b13 * 5352 + 12000) >> 16) + (b13 !== 0 ? 1 : 0);
+    out[o + 9] = (b10 - b11 + 7) >> 4;
+    out[o + 13] = (b13 * 2217 - b12 * 5352 + 51000) >> 16;
+    const b20 = t02 + t32, b21 = t12 + t22, b22 = t12 - t22, b23 = t02 - t32;
+    out[o + 2] = (b20 + b21 + 7) >> 4;
+    out[o + 6] = ((b22 * 2217 + b23 * 5352 + 12000) >> 16) + (b23 !== 0 ? 1 : 0);
+    out[o + 10] = (b20 - b21 + 7) >> 4;
+    out[o + 14] = (b23 * 2217 - b22 * 5352 + 51000) >> 16;
+    const b30 = t03 + t33, b31 = t13 + t23, b32 = t13 - t23, b33 = t03 - t33;
+    out[o + 3] = (b30 + b31 + 7) >> 4;
+    out[o + 7] = ((b32 * 2217 + b33 * 5352 + 12000) >> 16) + (b33 !== 0 ? 1 : 0);
+    out[o + 11] = (b30 - b31 + 7) >> 4;
+    out[o + 15] = (b33 * 2217 - b32 * 5352 + 51000) >> 16;
+}
+const bin = (c) => {
+    const v = (c < 0 ? -c : c) >> 3;
+    return v < 31 ? v : 31;
+};
+function fTransformHistogram(src, s, ref, r, dist) {
+    const d00 = src[s] - ref[r], d01 = src[s + 1] - ref[r + 1], d02 = src[s + 2] - ref[r + 2], d03 = src[s + 3] - ref[r + 3];
+    const a00 = d00 + d03, a01 = d01 + d02, a02 = d01 - d02, a03 = d00 - d03;
+    const t00 = (a00 + a01) * 8, t01 = (a02 * 2217 + a03 * 5352 + 1812) >> 9, t02 = (a00 - a01) * 8, t03 = (a03 * 2217 - a02 * 5352 + 937) >> 9;
+    const d10 = src[s + 1 * BPS] - ref[r + 1 * BPS], d11 = src[s + 1 * BPS + 1] - ref[r + 1 * BPS + 1], d12 = src[s + 1 * BPS + 2] - ref[r + 1 * BPS + 2], d13 = src[s + 1 * BPS + 3] - ref[r + 1 * BPS + 3];
+    const a10 = d10 + d13, a11 = d11 + d12, a12 = d11 - d12, a13 = d10 - d13;
+    const t10 = (a10 + a11) * 8, t11 = (a12 * 2217 + a13 * 5352 + 1812) >> 9, t12 = (a10 - a11) * 8, t13 = (a13 * 2217 - a12 * 5352 + 937) >> 9;
+    const d20 = src[s + 2 * BPS] - ref[r + 2 * BPS], d21 = src[s + 2 * BPS + 1] - ref[r + 2 * BPS + 1], d22 = src[s + 2 * BPS + 2] - ref[r + 2 * BPS + 2], d23 = src[s + 2 * BPS + 3] - ref[r + 2 * BPS + 3];
+    const a20 = d20 + d23, a21 = d21 + d22, a22 = d21 - d22, a23 = d20 - d23;
+    const t20 = (a20 + a21) * 8, t21 = (a22 * 2217 + a23 * 5352 + 1812) >> 9, t22 = (a20 - a21) * 8, t23 = (a23 * 2217 - a22 * 5352 + 937) >> 9;
+    const d30 = src[s + 3 * BPS] - ref[r + 3 * BPS], d31 = src[s + 3 * BPS + 1] - ref[r + 3 * BPS + 1], d32 = src[s + 3 * BPS + 2] - ref[r + 3 * BPS + 2], d33 = src[s + 3 * BPS + 3] - ref[r + 3 * BPS + 3];
+    const a30 = d30 + d33, a31 = d31 + d32, a32 = d31 - d32, a33 = d30 - d33;
+    const t30 = (a30 + a31) * 8, t31 = (a32 * 2217 + a33 * 5352 + 1812) >> 9, t32 = (a30 - a31) * 8, t33 = (a33 * 2217 - a32 * 5352 + 937) >> 9;
+    const b00 = t00 + t30, b01 = t10 + t20, b02 = t10 - t20, b03 = t00 - t30;
+    const c0 = (b00 + b01 + 7) >> 4;
+    const c4 = ((b02 * 2217 + b03 * 5352 + 12000) >> 16) + (b03 !== 0 ? 1 : 0);
+    const c8 = (b00 - b01 + 7) >> 4;
+    const c12 = (b03 * 2217 - b02 * 5352 + 51000) >> 16;
+    const b10 = t01 + t31, b11 = t11 + t21, b12 = t11 - t21, b13 = t01 - t31;
+    const c1 = (b10 + b11 + 7) >> 4;
+    const c5 = ((b12 * 2217 + b13 * 5352 + 12000) >> 16) + (b13 !== 0 ? 1 : 0);
+    const c9 = (b10 - b11 + 7) >> 4;
+    const c13 = (b13 * 2217 - b12 * 5352 + 51000) >> 16;
+    const b20 = t02 + t32, b21 = t12 + t22, b22 = t12 - t22, b23 = t02 - t32;
+    const c2 = (b20 + b21 + 7) >> 4;
+    const c6 = ((b22 * 2217 + b23 * 5352 + 12000) >> 16) + (b23 !== 0 ? 1 : 0);
+    const c10 = (b20 - b21 + 7) >> 4;
+    const c14 = (b23 * 2217 - b22 * 5352 + 51000) >> 16;
+    const b30 = t03 + t33, b31 = t13 + t23, b32 = t13 - t23, b33 = t03 - t33;
+    const c3 = (b30 + b31 + 7) >> 4;
+    const c7 = ((b32 * 2217 + b33 * 5352 + 12000) >> 16) + (b33 !== 0 ? 1 : 0);
+    const c11 = (b30 - b31 + 7) >> 4;
+    const c15 = (b33 * 2217 - b32 * 5352 + 51000) >> 16;
+    let zeros = 0;
+    if (c0 > -8 && c0 < 8)
+        zeros++;
+    else
+        dist[bin(c0)]++;
+    if (c1 > -8 && c1 < 8)
+        zeros++;
+    else
+        dist[bin(c1)]++;
+    if (c2 > -8 && c2 < 8)
+        zeros++;
+    else
+        dist[bin(c2)]++;
+    if (c3 > -8 && c3 < 8)
+        zeros++;
+    else
+        dist[bin(c3)]++;
+    if (c4 > -8 && c4 < 8)
+        zeros++;
+    else
+        dist[bin(c4)]++;
+    if (c5 > -8 && c5 < 8)
+        zeros++;
+    else
+        dist[bin(c5)]++;
+    if (c6 > -8 && c6 < 8)
+        zeros++;
+    else
+        dist[bin(c6)]++;
+    if (c7 > -8 && c7 < 8)
+        zeros++;
+    else
+        dist[bin(c7)]++;
+    if (c8 > -8 && c8 < 8)
+        zeros++;
+    else
+        dist[bin(c8)]++;
+    if (c9 > -8 && c9 < 8)
+        zeros++;
+    else
+        dist[bin(c9)]++;
+    if (c10 > -8 && c10 < 8)
+        zeros++;
+    else
+        dist[bin(c10)]++;
+    if (c11 > -8 && c11 < 8)
+        zeros++;
+    else
+        dist[bin(c11)]++;
+    if (c12 > -8 && c12 < 8)
+        zeros++;
+    else
+        dist[bin(c12)]++;
+    if (c13 > -8 && c13 < 8)
+        zeros++;
+    else
+        dist[bin(c13)]++;
+    if (c14 > -8 && c14 < 8)
+        zeros++;
+    else
+        dist[bin(c14)]++;
+    if (c15 > -8 && c15 < 8)
+        zeros++;
+    else
+        dist[bin(c15)]++;
+    return zeros;
+}
+const ftmp = new Int32Array(16);
+export function fTransformWHT(input, out) {
+    for (let i = 0, o = 0; i < 4; i++, o += 64) {
+        const a0 = input[o] + input[o + 32], a1 = input[o + 16] + input[o + 48];
+        const a2 = input[o + 16] - input[o + 48], a3 = input[o] - input[o + 32];
+        ftmp[i * 4] = a0 + a1;
+        ftmp[i * 4 + 1] = a3 + a2;
+        ftmp[i * 4 + 2] = a3 - a2;
+        ftmp[i * 4 + 3] = a0 - a1;
+    }
+    for (let i = 0; i < 4; i++) {
+        const a0 = ftmp[i] + ftmp[8 + i], a1 = ftmp[4 + i] + ftmp[12 + i], a2 = ftmp[4 + i] - ftmp[12 + i], a3 = ftmp[i] - ftmp[8 + i];
+        out[i] = (a0 + a1) >> 1;
+        out[4 + i] = (a3 + a2) >> 1;
+        out[8 + i] = (a3 - a2) >> 1;
+        out[12 + i] = (a0 - a1) >> 1;
+    }
+}
+export function iTransformWHT(input, out) {
+    for (let i = 0; i < 4; i++) {
+        const a0 = input[i] + input[12 + i], a1 = input[4 + i] + input[8 + i];
+        const a2 = input[4 + i] - input[8 + i], a3 = input[i] - input[12 + i];
+        ftmp[i] = a0 + a1;
+        ftmp[8 + i] = a0 - a1;
+        ftmp[4 + i] = a3 + a2;
+        ftmp[12 + i] = a3 - a2;
+    }
+    for (let i = 0, o = 0; i < 4; i++, o += 64) {
+        const dc = ftmp[i * 4] + 3;
+        const a0 = dc + ftmp[i * 4 + 3], a1 = ftmp[i * 4 + 1] + ftmp[i * 4 + 2];
+        const a2 = ftmp[i * 4 + 1] - ftmp[i * 4 + 2], a3 = dc - ftmp[i * 4 + 3];
+        out[o] = (a0 + a1) >> 3;
+        out[o + 16] = (a3 + a2) >> 3;
+        out[o + 32] = (a0 - a1) >> 3;
+        out[o + 48] = (a3 - a2) >> 3;
+    }
+}
+export function sse4x4(a, ao, b, bo) {
+    const d00 = a[ao] - b[bo], d01 = a[ao + 1] - b[bo + 1], d02 = a[ao + 2] - b[bo + 2], d03 = a[ao + 3] - b[bo + 3];
+    const d10 = a[ao + 1 * BPS] - b[bo + 1 * BPS], d11 = a[ao + 1 * BPS + 1] - b[bo + 1 * BPS + 1], d12 = a[ao + 1 * BPS + 2] - b[bo + 1 * BPS + 2], d13 = a[ao + 1 * BPS + 3] - b[bo + 1 * BPS + 3];
+    const d20 = a[ao + 2 * BPS] - b[bo + 2 * BPS], d21 = a[ao + 2 * BPS + 1] - b[bo + 2 * BPS + 1], d22 = a[ao + 2 * BPS + 2] - b[bo + 2 * BPS + 2], d23 = a[ao + 2 * BPS + 3] - b[bo + 2 * BPS + 3];
+    const d30 = a[ao + 3 * BPS] - b[bo + 3 * BPS], d31 = a[ao + 3 * BPS + 1] - b[bo + 3 * BPS + 1], d32 = a[ao + 3 * BPS + 2] - b[bo + 3 * BPS + 2], d33 = a[ao + 3 * BPS + 3] - b[bo + 3 * BPS + 3];
+    return d00 * d00 + d01 * d01 + d02 * d02 + d03 * d03 + d10 * d10 + d11 * d11 + d12 * d12 + d13 * d13 + d20 * d20 + d21 * d21 + d22 * d22 + d23 * d23 + d30 * d30 + d31 * d31 + d32 * d32 + d33 * d33;
+}
+export function sse16(a, ao, b, bo, h) {
+    let count = 0;
+    for (let y = 0; y < h; y++, ao += BPS, bo += BPS) {
+        const d0 = a[ao] - b[bo], d1 = a[ao + 1] - b[bo + 1], d2 = a[ao + 2] - b[bo + 2], d3 = a[ao + 3] - b[bo + 3], d4 = a[ao + 4] - b[bo + 4], d5 = a[ao + 5] - b[bo + 5], d6 = a[ao + 6] - b[bo + 6], d7 = a[ao + 7] - b[bo + 7], d8 = a[ao + 8] - b[bo + 8], d9 = a[ao + 9] - b[bo + 9], d10 = a[ao + 10] - b[bo + 10], d11 = a[ao + 11] - b[bo + 11], d12 = a[ao + 12] - b[bo + 12], d13 = a[ao + 13] - b[bo + 13], d14 = a[ao + 14] - b[bo + 14], d15 = a[ao + 15] - b[bo + 15];
+        count += d0 * d0 + d1 * d1 + d2 * d2 + d3 * d3 + d4 * d4 + d5 * d5 + d6 * d6 + d7 * d7 + d8 * d8 + d9 * d9 + d10 * d10 + d11 * d11 + d12 * d12 + d13 * d13 + d14 * d14 + d15 * d15;
+    }
+    return count;
+}
+export function tTransform(input, i, w) {
+    let sum = 0;
+    for (let k = 0; k < 4; k++, i += BPS) {
+        const a0 = input[i] + input[i + 2], a1 = input[i + 1] + input[i + 3], a2 = input[i + 1] - input[i + 3], a3 = input[i] - input[i + 2];
+        ftmp[k * 4] = a0 + a1;
+        ftmp[k * 4 + 1] = a3 + a2;
+        ftmp[k * 4 + 2] = a3 - a2;
+        ftmp[k * 4 + 3] = a0 - a1;
+    }
+    for (let k = 0; k < 4; k++) {
+        const a0 = ftmp[k] + ftmp[8 + k], a1 = ftmp[4 + k] + ftmp[12 + k], a2 = ftmp[4 + k] - ftmp[12 + k], a3 = ftmp[k] - ftmp[8 + k];
+        sum += w[k] * Math.abs(a0 + a1) + w[k + 4] * Math.abs(a3 + a2) + w[k + 8] * Math.abs(a3 - a2) + w[k + 12] * Math.abs(a0 - a1);
+    }
+    return sum;
+}
+export function tDisto4x4(source, b, bo, w) {
+    return Math.abs(tTransform(b, bo, w) - source) >> 5;
+}
+export function tDisto16x16(sources, b, bo, w) {
+    let d = 0;
+    for (let n = 0; n < 16; n++)
+        d += tDisto4x4(sources[n], b, bo + SCAN[n], w);
+    return d;
+}
+export function isFlatSource16(src, s) {
+    const v = src[s];
+    for (let y = 0; y < 16; y++)
+        for (let x = 0; x < 16; x++)
+            if (src[s + y * BPS + x] !== v)
+                return false;
+    return true;
+}
+export function isFlat(levels, l, num, thresh) {
+    let score = 0;
+    for (; num > 0; num--, l += 16) {
+        for (let i = 1; i < 16; i++) {
+            if (levels[l + i])
+                score++;
+            if (score > thresh)
+                return false;
+        }
+    }
+    return true;
+}
+export const newMatrix = () => ({
+    q: new Uint16Array(16), iq: new Uint16Array(16), bias: new Uint32Array(16), zthresh: new Uint32Array(16), sharpen: new Uint16Array(16)
+});
+export function quantizeBlock(input, i, out, o, m) {
+    let nz = 0;
+    const sharpen = m.sharpen;
+    {
+        const c = input[i], sign = c < 0;
+        const coeff = (sign ? -c : c) + sharpen[0];
+        if (coeff > m.zthresh[0]) {
+            let level = (coeff * m.iq[0] + m.bias[0]) >>> QFIX;
+            if (level > MAX_LEVEL)
+                level = MAX_LEVEL;
+            if (sign)
+                level = -level;
+            input[i] = level * m.q[0];
+            out[o] = level;
+            nz |= level;
+        }
+        else
+            out[o] = input[i] = 0;
+    }
+    const q = m.q[1], iq = m.iq[1], bias = m.bias[1], zthresh = m.zthresh[1];
+    for (let n = 1; n < 16; n++) {
+        const j = ZIGZAG[n];
+        const c = input[i + j], sign = c < 0;
+        const coeff = (sign ? -c : c) + sharpen[j];
+        if (coeff > zthresh) {
+            let level = (coeff * iq + bias) >>> QFIX;
+            if (level > MAX_LEVEL)
+                level = MAX_LEVEL;
+            if (sign)
+                level = -level;
+            input[i + j] = level * q;
+            out[o + n] = level;
+            nz |= level;
+        }
+        else
+            out[o + n] = input[i + j] = 0;
+    }
+    return nz !== 0;
+}
+export function mean16x4(src, s, dc, d) {
+    for (let k = 0; k < 4; k++, s += 4) {
+        let sum = 0;
+        for (let y = 0; y < 4; y++)
+            for (let x = 0; x < 4; x++)
+                sum += src[s + x + y * BPS];
+        dc[d + k] = sum;
+    }
+}
+const distribution = new Int32Array(32);
+const DSP_SCAN = [...SCAN, 0, 4, 4 * BPS, 4 * BPS + 4, 8, 12, 4 * BPS + 8, 4 * BPS + 12];
+export function histogramAlpha(ref, r, pred, p, start, end) {
+    distribution.fill(0);
+    let zeros = 0;
+    for (let j = start; j < end; j++) {
+        zeros += fTransformHistogram(ref, r + DSP_SCAN[j], pred, p + DSP_SCAN[j], distribution);
+    }
+    distribution[0] += zeros;
+    let max = 0, last = 1;
+    for (let k = 0; k <= 31; k++) {
+        if (distribution[k] > 0) {
+            if (distribution[k] > max)
+                max = distribution[k];
+            last = k;
+        }
+    }
+    return max > 1 ? Math.trunc(510 * last / max) : 0;
+}