Implement gradient filter of tight decoder, fixing issue #1767

This commit is a basic implementation of the gradient filter required by
qemu `lossy` option.

core/decoders/tight.js

@@ -285,7 +285,73 @@ export default class TightDecoder {
  }
 
  _gradientFilter(streamId, x, y, width, height, sock, display, depth) {
- throw new Error("Gradient filter not implemented");
+ // assume the TPIXEL is 3 bytes long
+ const uncompressedSize = width * height * 3;
+ let data;
+
+ if (uncompressedSize === 0) {
+ return true;
+ }
+
+ if (uncompressedSize < 12) {
+ if (sock.rQwait("TIGHT", uncompressedSize)) {
+ return false;
+ }
+
+ data = sock.rQshiftBytes(uncompressedSize);
+ } else {
+ data = this._readData(sock);
+ if (data === null) {
+ return false;
+ }
+
+ this._zlibs[streamId].setInput(data);
+ data = this._zlibs[streamId].inflate(uncompressedSize);
+ this._zlibs[streamId].setInput(null);
+ }
+
+ let rgbx = new Uint8Array(4 * width * height);
+
+ let rgbxIndex = 0, dataIndex = 0;
+ let left = new Uint8Array(3);
+ for (let x = 0; x < width; x++) {
+ for (let c = 0; c < 3; c++) {
+ const prediction = left[c];
+ const value = data[dataIndex++] + prediction;
+ rgbx[rgbxIndex++] = value;
+ left[c] = value;
+ }
+ rgbx[rgbxIndex++] = 255;
+ }
+
+ let upperIndex = 0;
+ let upper = new Uint8Array(3),
+ upperleft = new Uint8Array(3);
+ for (let y = 1; y < height; y++) {
+ left.fill(0);
+ upperleft.fill(0);
+ for (let x = 0; x < width; x++) {
+ for (let c = 0; c < 3; c++) {
+ upper[c] = rgbx[upperIndex++];
+ let prediction = left[c] + upper[c] - upperleft[c];
+ if (prediction < 0) {
+ prediction = 0;
+ } else if (prediction > 255) {
+ prediction = 255;
+ }
+ const value = data[dataIndex++] + prediction;
+ rgbx[rgbxIndex++] = value;
+ upperleft[c] = upper[c];
+ left[c] = value;
+ }
+ rgbx[rgbxIndex++] = 255;
+ upperIndex++;
+ }
+ }
+
+ display.blitImage(x, y, width, height, rgbx, 0, false);
+
+ return true;
  }
 
  _readData(sock) {