#include <assert.h>
#include <math.h>
-//#include "ryg_rans/rans64.h"
-#include "ryg_rans/rans_byte.h"
+#include "ryg_rans/rans64.h"
+//#include "ryg_rans/rans_byte.h"
#include "ryg_rans/renormalize.h"
#include <algorithm>
#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)
+#define BLOCKS_PER_STREAM 320
// If you set this to 1, the program will try to optimize the placement
// of coefficients to rANS probability distributions. This is randomized,
void write_varint(int x, FILE *fp)
{
- while (x >= 128) {
- putc((x & 0x7f) | 0x80, fp);
- x >>= 7;
- }
- putc(x, fp);
+ fwrite(&x, sizeof(x), 1, fp);
}
class RansEncoder {
void init_prob(SymbolStats &s)
{
for (int i = 0; i < NUM_SYMS; i++) {
- 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);
+ //printf("%d: cumfreqs=%d freqs=%d prob_bits=%d\n", i, s.cum_freqs[i], s.freqs[i], prob_bits + 1);
+ Rans64EncSymbolInit(&esyms[i], s.cum_freqs[i], s.freqs[i], prob_bits + 1);
}
sign_bias = s.cum_freqs[NUM_SYMS];
}
{
out_end = out_buf.get() + out_max_size;
ptr = out_end; // *end* of output buffer
- RansEncInit(&rans);
+ Rans64EncInit(&rans);
}
uint32_t save_block(FILE *codedfp) // Returns number of bytes.
{
- RansEncFlush(&rans, &ptr);
+ Rans64EncFlush(&rans, (uint32_t **)&ptr);
//printf("post-flush = %08x\n", rans);
uint32_t num_rans_bytes = out_end - ptr;
clear();
- printf("Saving block: %d rANS bytes\n", num_rans_bytes);
+ //printf("Saving block: %d rANS bytes\n", num_rans_bytes);
return num_rans_bytes;
//return num_rans_bytes;
}
void encode_coeff(short signed_k)
{
- //printf("encoding coeff %d (sym %d), rans before encoding = %08x\n", signed_k, ((abs(signed_k) - 1) & 255), rans);
+ //printf("encoding coeff %d (sym %d), rans before encoding = %016lx\n", signed_k, ((abs(signed_k) - 1) & 255), rans);
unsigned short k = abs(signed_k);
if (k >= ESCAPE_LIMIT) {
// Put the coefficient as a 1/(2^12) symbol _before_
// the 255 coefficient, since the decoder will read the
// 255 coefficient first.
- RansEncPut(&rans, &ptr, k, 1, prob_bits);
+ Rans64EncPut(&rans, (uint32_t **)&ptr, k, 1, prob_bits);
k = ESCAPE_LIMIT;
}
- RansEncPutSymbol(&rans, &ptr, &esyms[(k - 1) & (NUM_SYMS - 1)]);
+ Rans64EncPutSymbol(&rans, (uint32_t **)&ptr, &esyms[(k - 1) & (NUM_SYMS - 1)], prob_bits + 1);
if (signed_k < 0) {
rans += sign_bias;
}
unique_ptr<uint8_t[]> out_buf;
uint8_t *out_end;
uint8_t *ptr;
- RansState rans;
- RansEncSymbol esyms[NUM_SYMS];
+ Rans64State rans;
+ Rans64EncSymbol esyms[NUM_SYMS];
uint32_t sign_bias;
uint32_t last_block = 0; // Not a valid 4-byte rANS block (?)
chroma_energy / (WIDTH * HEIGHT), chroma_energy_pred / (WIDTH * HEIGHT));
#endif
- // DC coefficient pred from the right to left
- for (unsigned yb = 0; yb < HEIGHT; yb += 8) {
- for (unsigned xb = 0; xb < WIDTH - 8; xb += 8) {
- coeff_y[yb * WIDTH + xb] -= coeff_y[yb * WIDTH + (xb + 8)];
+ // DC coefficient pred from the right to left (within each slice)
+ for (unsigned block_idx = 0; block_idx < NUM_BLOCKS; block_idx += BLOCKS_PER_STREAM) {
+ int prev_k = 128;
+
+ for (unsigned subblock_idx = BLOCKS_PER_STREAM; subblock_idx --> 0; ) {
+ unsigned yb = (block_idx + subblock_idx) / WIDTH_BLOCKS;
+ unsigned xb = (block_idx + subblock_idx) % WIDTH_BLOCKS;
+ int k = coeff_y[(yb * 8) * WIDTH + (xb * 8)];
+
+ coeff_y[(yb * 8) * WIDTH + (xb * 8)] = k - prev_k;
+
+ prev_k = k;
}
}
- 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)];
+ for (unsigned block_idx = 0; block_idx < NUM_BLOCKS_CHROMA; block_idx += BLOCKS_PER_STREAM) {
+ int prev_k_cb = 0;
+ int prev_k_cr = 0;
+
+ for (unsigned subblock_idx = BLOCKS_PER_STREAM; subblock_idx --> 0; ) {
+ unsigned yb = (block_idx + subblock_idx) / WIDTH_BLOCKS_CHROMA;
+ unsigned xb = (block_idx + subblock_idx) % WIDTH_BLOCKS_CHROMA;
+ int k_cb = coeff_cb[(yb * 8) * WIDTH/2 + (xb * 8)];
+ int k_cr = coeff_cr[(yb * 8) * WIDTH/2 + (xb * 8)];
+
+ coeff_cb[(yb * 8) * WIDTH/2 + (xb * 8)] = k_cb - prev_k_cb;
+ coeff_cr[(yb * 8) * WIDTH/2 + (xb * 8)] = k_cr - prev_k_cr;
+
+ prev_k_cb = k_cb;
+ prev_k_cr = k_cr;
}
}
// 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 % BLOCKS_PER_STREAM == (BLOCKS_PER_STREAM - 1) || 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);
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 % BLOCKS_PER_STREAM == (BLOCKS_PER_STREAM - 1) || 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);
}
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 % BLOCKS_PER_STREAM == (BLOCKS_PER_STREAM - 1) || 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);
}