X-Git-Url: https://git.sesse.net/?a=blobdiff_plain;f=qdc.cpp;h=556a8a5d541315a54d6bddf332cdd746a5c1769a;hb=34138d5b1a1302a7a1050fd46d2fb95c0186140a;hp=b5594b76856918d9b14bee3f55dd5bcec24fdd7a;hpb=cd50d72afd7a54ec57827049c8bb934cfbb5a42c;p=narabu

diff --git a/qdc.cpp b/qdc.cpp
index b5594b7..556a8a5 100644
--- a/qdc.cpp
+++ b/qdc.cpp
@@ -18,6 +18,11 @@
 
 #define WIDTH 1280
 #define HEIGHT 720
+#define WIDTH_BLOCKS (WIDTH/8)
+#define WIDTH_BLOCKS_CHROMA (WIDTH/16)
+#define HEIGHT_BLOCKS (HEIGHT/8)
+#define NUM_BLOCKS (WIDTH_BLOCKS * HEIGHT_BLOCKS)
+#define NUM_BLOCKS_CHROMA (WIDTH_BLOCKS_CHROMA * HEIGHT_BLOCKS)
 #define NUM_SYMS 256
 #define ESCAPE_LIMIT (NUM_SYMS - 1)
 
@@ -25,7 +30,7 @@
 // of coefficients to rANS probability distributions. This is randomized,
 // so you might want to run it a few times.
 #define FIND_OPTIMAL_STREAM_ASSIGNMENT 0
-#define NUM_CLUSTERS 8
+#define NUM_CLUSTERS 4
 
 static constexpr uint32_t prob_bits = 12;
 static constexpr uint32_t prob_scale = 1 << prob_bits;
@@ -154,12 +159,37 @@ SymbolStats stats[128];
 float kl_dist[64][64];
 #endif
 
+const int luma_mapping[64] = {
+	0, 0, 1, 1, 2, 2, 3, 3,
+	0, 0, 1, 2, 2, 2, 3, 3,
+	1, 1, 2, 2, 2, 3, 3, 3,
+	1, 1, 2, 2, 2, 3, 3, 3,
+	1, 2, 2, 2, 2, 3, 3, 3,
+	2, 2, 2, 2, 3, 3, 3, 3,
+	2, 2, 3, 3, 3, 3, 3, 3,
+	3, 3, 3, 3, 3, 3, 3, 3,
+};
+const int chroma_mapping[64] = {
+	0, 1, 1, 2, 2, 2, 3, 3,
+	1, 1, 2, 2, 2, 3, 3, 3,
+	2, 2, 2, 2, 3, 3, 3, 3,
+	2, 2, 2, 3, 3, 3, 3, 3,
+	2, 3, 3, 3, 3, 3, 3, 3,
+	3, 3, 3, 3, 3, 3, 3, 3,
+	3, 3, 3, 3, 3, 3, 3, 3,
+	3, 3, 3, 3, 3, 3, 3, 3,
+};
+
 int pick_stats_for(int x, int y, bool is_chroma)
 {
 #if FIND_OPTIMAL_STREAM_ASSIGNMENT
 	return y * 8 + x + is_chroma * 64;
 #else
-	return std::min<int>(x + y, 7) + is_chroma * 8;
+	if (is_chroma) {
+		return chroma_mapping[y * 8 + x] + 4;
+	} else {
+		return luma_mapping[y * 8 + x];
+	}
 #endif
 }
 		
@@ -187,7 +217,7 @@ public:
 			printf("%d: cumfreqs=%d freqs=%d prob_bits=%d\n", i, s.cum_freqs[i], s.freqs[i], prob_bits + 1);
 			RansEncSymbolInit(&esyms[i], s.cum_freqs[i], s.freqs[i], prob_bits + 1);
 		}
-		sign_bias = s.cum_freqs[256];
+		sign_bias = s.cum_freqs[NUM_SYMS];
 	}
 
 	void clear()
@@ -245,7 +275,7 @@ public:
 			RansEncPut(&rans, &ptr, k, 1, prob_bits);
 			k = ESCAPE_LIMIT;
 		}
-		RansEncPutSymbol(&rans, &ptr, &esyms[(k - 1) & 255]);
+		RansEncPutSymbol(&rans, &ptr, &esyms[(k - 1) & (NUM_SYMS - 1)]);
 		if (signed_k < 0) {
 			rans += sign_bias;
 		}
@@ -379,7 +409,7 @@ void find_optimal_stream_assignment(int base)
 	double inv_sum[64];
 	for (unsigned i = 0; i < 64; ++i) {
 		double s = 0.0;
-		for (unsigned k = 0; k < 256; ++k) {
+		for (unsigned k = 0; k < NUM_SYMS; ++k) {
 			s += stats[i + base].freqs[k] + 0.5;
 		}
 		inv_sum[i] = 1.0 / s;
@@ -388,7 +418,7 @@ void find_optimal_stream_assignment(int base)
 	for (unsigned i = 0; i < 64; ++i) {
 		for (unsigned j = 0; j < 64; ++j) {
 			double d = 0.0;
-			for (unsigned k = 0; k < 256; ++k) {
+			for (unsigned k = 0; k < NUM_SYMS; ++k) {
 				double p1 = (stats[i + base].freqs[k] + 0.5) * inv_sum[i];
 				double p2 = (stats[j + base].freqs[k] + 0.5) * inv_sum[j];
 
@@ -695,6 +725,12 @@ int main(int argc, char **argv)
 			coeff_y[yb * WIDTH + xb] -= coeff_y[yb * WIDTH + (xb + 8)];
 		}
 	}
+	for (unsigned yb = 0; yb < HEIGHT; yb += 8) {
+		for (unsigned xb = 0; xb < WIDTH/2 - 8; xb += 8) {
+			coeff_cb[yb * WIDTH/2 + xb] -= coeff_cb[yb * WIDTH/2 + (xb + 8)];
+			coeff_cr[yb * WIDTH/2 + xb] -= coeff_cr[yb * WIDTH/2 + (xb + 8)];
+		}
+	}
 
 	FILE *fp = fopen("reconstructed.pgm", "wb");
 	fprintf(fp, "P5\n%d %d\n255\n", WIDTH, HEIGHT);
@@ -746,7 +782,7 @@ int main(int argc, char **argv)
 						k = ESCAPE_LIMIT;
 						extra_bits += 12;  // escape this one
 					}
-					++s_luma.freqs[(k - 1) & 255];
+					++s_luma.freqs[(k - 1) & (NUM_SYMS - 1)];
 				}
 			}
 			// Chroma
@@ -762,8 +798,8 @@ int main(int argc, char **argv)
 						k_cr = ESCAPE_LIMIT;
 						extra_bits += 12;  // escape this one
 					}
-					++s_chroma.freqs[(k_cb - 1) & 255];
-					++s_chroma.freqs[(k_cr - 1) & 255];
+					++s_chroma.freqs[(k_cb - 1) & (NUM_SYMS - 1)];
+					++s_chroma.freqs[(k_cr - 1) & (NUM_SYMS - 1)];
 				}
 			}
 		}
@@ -778,10 +814,10 @@ int main(int argc, char **argv)
 #endif
 
 	for (unsigned i = 0; i < 64; ++i) {
-		stats[i].freqs[255] /= 2;  // zero, has no sign bits (yes, this is trickery)
+		stats[i].freqs[NUM_SYMS - 1] /= 2;  // zero, has no sign bits (yes, this is trickery)
 		stats[i].normalize_freqs(prob_scale);
-		stats[i].cum_freqs[256] += stats[i].freqs[255];
-		stats[i].freqs[255] *= 2;
+		stats[i].cum_freqs[NUM_SYMS] += stats[i].freqs[NUM_SYMS - 1];
+		stats[i].freqs[NUM_SYMS - 1] *= 2;
 	}
 
 	FILE *codedfp = fopen("coded.dat", "wb");
@@ -817,21 +853,20 @@ int main(int argc, char **argv)
 			// need to reverse later
 			rans_encoder.clear();
 			size_t num_bytes = 0;
-			for (unsigned yb = 0; yb < HEIGHT; yb += 8) {
-				for (unsigned xb = 0; xb < WIDTH; xb += 8) {
-					int k = coeff_y[(yb + y) * WIDTH + (xb + x)];
-					//printf("encoding coeff %d xb,yb=%d,%d: %d\n", y*8+x, xb, yb, k);
-					rans_encoder.encode_coeff(k);
-				}
-				if (yb % 16 == 8) {
+			for (unsigned block_idx = 0; block_idx < NUM_BLOCKS; ++block_idx) {
+				unsigned yb = block_idx / WIDTH_BLOCKS;
+				unsigned xb = block_idx % WIDTH_BLOCKS;
+
+				int k = coeff_y[(yb * 8 + y) * WIDTH + (xb * 8 + x)];
+				//printf("encoding coeff %d xb,yb=%d,%d: %d\n", y*8+x, xb, yb, k);
+				rans_encoder.encode_coeff(k);
+
+				if (block_idx % 320 == 319 || block_idx == NUM_BLOCKS - 1) {
 					int l = rans_encoder.save_block(codedfp);
 					num_bytes += l;
 					lens.push_back(l);
 				}
 			}
-			if (HEIGHT % 16 != 0) {
-				num_bytes += rans_encoder.save_block(codedfp);
-			}
 			tot_bytes += num_bytes;
 			printf("coeff %d Y': %ld bytes\n", y * 8 + x, num_bytes);
 
@@ -858,18 +893,18 @@ int main(int argc, char **argv)
 
 			rans_encoder.clear();
 			size_t num_bytes = 0;
-			for (unsigned yb = 0; yb < HEIGHT; yb += 8) {
-				for (unsigned xb = 0; xb < WIDTH/2; xb += 8) {
-					int k = coeff_cb[(yb + y) * WIDTH/2 + (xb + x)];
-					rans_encoder.encode_coeff(k);
-				}
-				if (yb % 16 == 8) {
+			for (unsigned block_idx = 0; block_idx < NUM_BLOCKS_CHROMA; ++block_idx) {
+				unsigned yb = block_idx / WIDTH_BLOCKS_CHROMA;
+				unsigned xb = block_idx % WIDTH_BLOCKS_CHROMA;
+
+				int k = coeff_cb[(yb * 8 + y) * WIDTH/2 + (xb * 8 + x)];
+				//printf("encoding coeff %d xb,yb=%d,%d: %d\n", y*8+x, xb, yb, k);
+				rans_encoder.encode_coeff(k);
+
+				if (block_idx % 320 == 319 || block_idx == NUM_BLOCKS - 1) {
 					num_bytes += rans_encoder.save_block(codedfp);
 				}
 			}
-			if (HEIGHT % 16 != 0) {
-				num_bytes += rans_encoder.save_block(codedfp);
-			}
 			tot_bytes += num_bytes;
 			printf("coeff %d Cb: %ld bytes\n", y * 8 + x, num_bytes);
 		}
@@ -883,18 +918,18 @@ int main(int argc, char **argv)
 
 			rans_encoder.clear();
 			size_t num_bytes = 0;
-			for (unsigned yb = 0; yb < HEIGHT; yb += 8) {
-				for (unsigned xb = 0; xb < WIDTH/2; xb += 8) {
-					int k = coeff_cr[(yb + y) * WIDTH/2 + (xb + x)];
-					rans_encoder.encode_coeff(k);
-				}
-				if (yb % 16 == 8) {
+			for (unsigned block_idx = 0; block_idx < NUM_BLOCKS_CHROMA; ++block_idx) {
+				unsigned yb = block_idx / WIDTH_BLOCKS_CHROMA;
+				unsigned xb = block_idx % WIDTH_BLOCKS_CHROMA;
+
+				int k = coeff_cr[(yb * 8 + y) * WIDTH/2 + (xb * 8 + x)];
+				//printf("encoding coeff %d xb,yb=%d,%d: %d\n", y*8+x, xb, yb, k);
+				rans_encoder.encode_coeff(k);
+
+				if (block_idx % 320 == 319 || block_idx == NUM_BLOCKS - 1) {
 					num_bytes += rans_encoder.save_block(codedfp);
 				}
 			}
-			if (HEIGHT % 16 != 0) {
-				num_bytes += rans_encoder.save_block(codedfp);
-			}
 			tot_bytes += num_bytes;
 			printf("coeff %d Cr: %ld bytes\n", y * 8 + x, num_bytes);
 		}