#define BLOCKS_PER_STREAM 320
#define STREAM_BUF_SIZE 1024 // In bytes.
+#define NUM_STREAMS ((NUM_BLOCKS + BLOCKS_PER_STREAM - 1) / BLOCKS_PER_STREAM)
+
static constexpr uint32_t prob_bits = 12;
static constexpr uint32_t prob_scale = 1 << prob_bits;
GLuint output_ssbo;
glGenBuffers(1, &output_ssbo);
glBindBuffer(GL_SHADER_STORAGE_BUFFER, output_ssbo);
- glNamedBufferStorage(output_ssbo, HEIGHT_BLOCKS * WIDTH_BLOCKS * STREAM_BUF_SIZE, nullptr, GL_MAP_READ_BIT | GL_MAP_PERSISTENT_BIT);
+ glNamedBufferStorage(output_ssbo, 64 * NUM_STREAMS * STREAM_BUF_SIZE, nullptr, GL_MAP_READ_BIT | GL_MAP_PERSISTENT_BIT);
check_error();
GLuint bytes_written_ssbo;
glGenBuffers(1, &bytes_written_ssbo);
glBindBuffer(GL_SHADER_STORAGE_BUFFER, bytes_written_ssbo);
- glNamedBufferStorage(bytes_written_ssbo, HEIGHT_BLOCKS * WIDTH_BLOCKS * sizeof(uint32_t), nullptr, GL_MAP_READ_BIT | GL_MAP_PERSISTENT_BIT);
+ glNamedBufferStorage(bytes_written_ssbo, 64 * NUM_STREAMS * sizeof(uint32_t), nullptr, GL_MAP_READ_BIT | GL_MAP_PERSISTENT_BIT);
check_error();
// Bind SSBOs.
glMemoryBarrier(GL_UNIFORM_BARRIER_BIT);
glUseProgram(glsl_rans_program_num);
- glDispatchCompute(NUM_BLOCKS / BLOCKS_PER_STREAM, 8, 5);
+ glDispatchCompute(NUM_STREAMS, 8, 5);
}
check_error();
glFinish();
// Write out the actual data.
- const uint32_t *bytes_written = (const uint32_t *)glMapNamedBufferRange(bytes_written_ssbo, 0, HEIGHT_BLOCKS * WIDTH_BLOCKS * sizeof(uint32_t), GL_MAP_READ_BIT | GL_MAP_PERSISTENT_BIT);
+ const uint32_t *bytes_written = (const uint32_t *)glMapNamedBufferRange(bytes_written_ssbo, 0, 64 * NUM_STREAMS * sizeof(uint32_t), GL_MAP_READ_BIT | GL_MAP_PERSISTENT_BIT);
#if 0
for (int i = 0; i < HEIGHT_BLOCKS*64; ++i) {
printf("%d,%d,%d: %u\n", i / 64, (i / 8) % 8, i % 8, 1024 * (i + 1) - offsets[i]);
}
#endif
- const uint8_t *data = (const uint8_t *)glMapNamedBufferRange(output_ssbo, 0, HEIGHT_BLOCKS * WIDTH_BLOCKS * STREAM_BUF_SIZE, GL_MAP_READ_BIT | GL_MAP_PERSISTENT_BIT);
+ const uint8_t *data = (const uint8_t *)glMapNamedBufferRange(output_ssbo, 0, 64 * NUM_STREAMS * STREAM_BUF_SIZE, GL_MAP_READ_BIT | GL_MAP_PERSISTENT_BIT);
string last_block;
for (unsigned y = 0; y < 8; ++y) {
for (unsigned x = 0; x < 8; ++x) {
- for (unsigned int stream_idx = 0; stream_idx < HEIGHT_BLOCKS; ++stream_idx) {
+ for (unsigned int stream_idx = 0; stream_idx < NUM_STREAMS; ++stream_idx) {
const uint8_t *out_end = data + (stream_idx * 64 + y * 8 + x + 1) * STREAM_BUF_SIZE;
uint32_t num_rans_bytes = bytes_written[stream_idx * 64 + y * 8 + x];
const uint8_t *ptr = out_end - num_rans_bytes;
const uint STREAM_BUF_SIZE = 256; // In uint32s. 1 kB per stream ought to be enough for everyone :-)
const uint NUM_SYMS = 256;
const uint ESCAPE_LIMIT = NUM_SYMS - 1;
+const uint WIDTH_BLOCKS = 160; // TODO: send in as a uniform.
#define MAPPING(s0, s1, s2, s3, s4, s5, s6, s7) ((s0) | (s1 << 2) | (s2 << 4) | (s3 << 6) | (s4 << 8) | (s5 << 10) | (s6 << 12) | (s7 << 14))
for (uint subblock_idx = 0; subblock_idx < BLOCKS_PER_STREAM; ++subblock_idx) {
// TODO: Use SSBOs instead of a texture?
- uint x = (streamgroup_num * BLOCKS_PER_STREAM + subblock_idx) % 160;
- uint y = (streamgroup_num * BLOCKS_PER_STREAM + subblock_idx) / 160;
+ uint x = (streamgroup_num * BLOCKS_PER_STREAM + subblock_idx) % WIDTH_BLOCKS;
+ uint y = (streamgroup_num * BLOCKS_PER_STREAM + subblock_idx) / WIDTH_BLOCKS;
uint f = imageLoad(tex, ivec2(x, y * 8 + coeff_row)).x;
encode_coeff(sign_extend(f & 0x1ffu, 9), enc1);
for (uint subblock_idx = 0; subblock_idx < BLOCKS_PER_STREAM; ++subblock_idx) {
// TODO: Use SSBOs instead of a texture?
- uint x = (streamgroup_num * BLOCKS_PER_STREAM + subblock_idx) % 160;
- uint y = (streamgroup_num * BLOCKS_PER_STREAM + subblock_idx) / 160;
+ uint x = (streamgroup_num * BLOCKS_PER_STREAM + subblock_idx) % WIDTH_BLOCKS;
+ uint y = (streamgroup_num * BLOCKS_PER_STREAM + subblock_idx) / WIDTH_BLOCKS;
int f = imageLoad(tex, ivec2(x, y * 8 + coeff_row)).x;
encode_coeff(f, enc);