void init_analyzer() {
const int aligned_width = ALIGN_POWER_OF_TWO(info->frame_width, MI_SIZE_LOG2);
const int aligned_height = ALIGN_POWER_OF_TWO(info->frame_height, MI_SIZE_LOG2);
int mi_cols = aligned_width >> MI_SIZE_LOG2;
int mi_rows = aligned_height >> MI_SIZE_LOG2;
int mi_length = mi_cols * mi_rows;
printf("init_analyzer: %d:%d (mi)\n", mi_cols, mi_rows);
analyzer_data.mi_grid.buffer = aom_malloc(sizeof(AnalyzerMI) * mi_length);
analyzer_data.mi_grid.length = mi_length;
size_t size = aligned_width * aligned_height * 2;
AnalyzerImagePlane *planes = analyzer_data.predicted_image.planes;
planes[AOM_PLANE_Y].size = size;
planes[AOM_PLANE_U].size = size;
planes[AOM_PLANE_V].size = size;
planes[AOM_PLANE_Y].stride = info->frame_width;
planes[AOM_PLANE_U].stride = info->frame_width >> 1;
planes[AOM_PLANE_V].stride = info->frame_width >> 1;
planes[AOM_PLANE_Y].buffer = aom_malloc(size);
planes[AOM_PLANE_U].buffer = aom_malloc(size);
planes[AOM_PLANE_V].buffer = aom_malloc(size);
}
void *aom_realloc(void *memblk, size_t size) {
void *addr, *new_addr = NULL;
int align = DEFAULT_ALIGNMENT;
/*
The realloc() function changes the size of the object pointed to by
ptr to the size specified by size, and returns a pointer to the
possibly moved block. The contents are unchanged up to the lesser
of the new and old sizes. If ptr is null, realloc() behaves like
malloc() for the specified size. If size is zero (0) and ptr is
not a null pointer, the object pointed to is freed.
*/
if (!memblk)
new_addr = aom_malloc(size);
else if (!size)
aom_free(memblk);
else {
addr = (void *)(((size_t *)memblk)[-1]);
memblk = NULL;
new_addr = realloc(addr, size + align + ADDRESS_STORAGE_SIZE);
if (new_addr) {
addr = new_addr;
new_addr =
(void *)(((size_t)((unsigned char *)new_addr + ADDRESS_STORAGE_SIZE) +
(align - 1)) &
(size_t)-align);
/* save the actual malloc address */
((size_t *)new_addr)[-1] = (size_t)addr;
}
}
return new_addr;
}
int pseudo_inverse(double *inv, double *matx, const int M, const int N) {
double ans;
int i, j, k;
double *const U = (double *)aom_malloc(M * N * sizeof(*matx));
double *const W = (double *)aom_malloc(N * sizeof(*matx));
double *const V = (double *)aom_malloc(N * N * sizeof(*matx));
if (!(U && W && V)) {
return 1;
}
if (SVD(U, W, V, matx, M, N)) {
aom_free(U);
aom_free(W);
aom_free(V);
return 1;
}
for (i = 0; i < N; i++) {
if (fabs(W[i]) < TINY_NEAR_ZERO) {
aom_free(U);
aom_free(W);
aom_free(V);
return 1;
}
}
for (i = 0; i < N; i++) {
for (j = 0; j < M; j++) {
ans = 0;
for (k = 0; k < N; k++) {
ans += V[k + N * i] * U[k + N * j] / W[k];
}
inv[j + M * i] = ans;
}
}
aom_free(U);
aom_free(W);
aom_free(V);
return 0;
}
static int SVD(double *U, double *W, double *V, double *matx, int M, int N) {
// Assumes allocation for U is MxN
double **nrU = (double **)aom_malloc((M) * sizeof(*nrU));
double **nrV = (double **)aom_malloc((N) * sizeof(*nrV));
int problem, i;
problem = !(nrU && nrV);
if (!problem) {
for (i = 0; i < M; i++) {
nrU[i] = &U[i * N];
}
for (i = 0; i < N; i++) {
nrV[i] = &V[i * N];
}
} else {
if (nrU) aom_free(nrU);
if (nrV) aom_free(nrV);
return 1;
}
/* copy from given matx into nrU */
for (i = 0; i < M; i++) {
memcpy(&(nrU[i][0]), matx + N * i, N * sizeof(*matx));
}
/* HERE IT IS: do SVD */
if (svdcmp(nrU, M, N, W, nrV)) {
aom_free(nrU);
aom_free(nrV);
return 1;
}
/* aom_free Numerical Recipes arrays */
aom_free(nrU);
aom_free(nrV);
return 0;
}
static int find_affine(int np, double *pts1, double *pts2, double *mat) {
const int np2 = np * 2;
double *a = (double *)aom_malloc(sizeof(*a) * np2 * 13);
double *b = a + np2 * 6;
double *temp = b + np2;
int i;
double sx, sy, dx, dy;
double T1[9], T2[9];
normalize_homography(pts1, np, T1);
normalize_homography(pts2, np, T2);
for (i = 0; i < np; ++i) {
dx = *(pts2++);
dy = *(pts2++);
sx = *(pts1++);
sy = *(pts1++);
a[i * 2 * 6 + 0] = sx;
a[i * 2 * 6 + 1] = sy;
a[i * 2 * 6 + 2] = 0;
a[i * 2 * 6 + 3] = 0;
a[i * 2 * 6 + 4] = 1;
a[i * 2 * 6 + 5] = 0;
a[(i * 2 + 1) * 6 + 0] = 0;
a[(i * 2 + 1) * 6 + 1] = 0;
a[(i * 2 + 1) * 6 + 2] = sx;
a[(i * 2 + 1) * 6 + 3] = sy;
a[(i * 2 + 1) * 6 + 4] = 0;
a[(i * 2 + 1) * 6 + 5] = 1;
b[2 * i] = dx;
b[2 * i + 1] = dy;
}
if (pseudo_inverse(temp, a, np2, 6)) {
aom_free(a);
return 1;
}
multiply_mat(temp, b, mat, 6, np2, 1);
denormalize_affine_reorder(mat, T1, T2);
aom_free(a);
return 0;
}
static int find_affine(int np, double *pts1, double *pts2, double *mat) {
const int np2 = np * 2;
double *a = (double *)aom_malloc(sizeof(*a) * (np2 * 7 + 42));
double *b = a + np2 * 6;
double *temp = b + np2;
int i;
double sx, sy, dx, dy;
double T1[9], T2[9];
normalize_homography(pts1, np, T1);
normalize_homography(pts2, np, T2);
for (i = 0; i < np; ++i) {
dx = *(pts2++);
dy = *(pts2++);
sx = *(pts1++);
sy = *(pts1++);
a[i * 2 * 6 + 0] = sx;
a[i * 2 * 6 + 1] = sy;
a[i * 2 * 6 + 2] = 0;
a[i * 2 * 6 + 3] = 0;
a[i * 2 * 6 + 4] = 1;
a[i * 2 * 6 + 5] = 0;
a[(i * 2 + 1) * 6 + 0] = 0;
a[(i * 2 + 1) * 6 + 1] = 0;
a[(i * 2 + 1) * 6 + 2] = sx;
a[(i * 2 + 1) * 6 + 3] = sy;
a[(i * 2 + 1) * 6 + 4] = 0;
a[(i * 2 + 1) * 6 + 5] = 1;
b[2 * i] = dx;
b[2 * i + 1] = dy;
}
if (!least_squares(6, a, np2, 6, b, temp, mat)) {
aom_free(a);
return 1;
}
denormalize_affine_reorder(mat, T1, T2);
aom_free(a);
return 0;
}
struct aom_noise_tx_t *aom_noise_tx_malloc(int block_size) {
struct aom_noise_tx_t *noise_tx =
(struct aom_noise_tx_t *)aom_malloc(sizeof(struct aom_noise_tx_t));
if (!noise_tx) return NULL;
memset(noise_tx, 0, sizeof(*noise_tx));
switch (block_size) {
case 2:
noise_tx->fft = aom_fft2x2_float;
noise_tx->ifft = aom_ifft2x2_float;
break;
case 4:
noise_tx->fft = aom_fft4x4_float;
noise_tx->ifft = aom_ifft4x4_float;
break;
case 8:
noise_tx->fft = aom_fft8x8_float;
noise_tx->ifft = aom_ifft8x8_float;
break;
case 16:
noise_tx->fft = aom_fft16x16_float;
noise_tx->ifft = aom_ifft16x16_float;
break;
case 32:
noise_tx->fft = aom_fft32x32_float;
noise_tx->ifft = aom_ifft32x32_float;
break;
default:
aom_free(noise_tx);
fprintf(stderr, "Unsupported block size %d\n", block_size);
return NULL;
}
noise_tx->block_size = block_size;
noise_tx->tx_block = (float *)aom_memalign(
32, 2 * sizeof(*noise_tx->tx_block) * block_size * block_size);
noise_tx->temp = (float *)aom_memalign(
32, 2 * sizeof(*noise_tx->temp) * block_size * block_size);
if (!noise_tx->tx_block || !noise_tx->temp) {
aom_noise_tx_free(noise_tx);
return NULL;
}
return noise_tx;
}
CYCLIC_REFRESH *av1_cyclic_refresh_alloc(int mi_rows, int mi_cols) {
size_t last_coded_q_map_size;
CYCLIC_REFRESH *const cr = aom_calloc(1, sizeof(*cr));
if (cr == NULL) return NULL;
cr->map = aom_calloc(mi_rows * mi_cols, sizeof(*cr->map));
if (cr->map == NULL) {
aom_free(cr);
return NULL;
}
last_coded_q_map_size = mi_rows * mi_cols * sizeof(*cr->last_coded_q_map);
cr->last_coded_q_map = aom_malloc(last_coded_q_map_size);
if (cr->last_coded_q_map == NULL) {
aom_free(cr);
return NULL;
}
assert(MAXQ <= 255);
memset(cr->last_coded_q_map, MAXQ, last_coded_q_map_size);
return cr;
}
void av1_encode_tiles_mt(AV1_COMP *cpi) {
AV1_COMMON *const cm = &cpi->common;
const int tile_cols = cm->tile_cols;
const AVxWorkerInterface *const winterface = aom_get_worker_interface();
const int num_workers = AOMMIN(cpi->oxcf.max_threads, tile_cols);
int i;
av1_init_tile_data(cpi);
// Only run once to create threads and allocate thread data.
if (cpi->num_workers == 0) {
CHECK_MEM_ERROR(cm, cpi->workers,
aom_malloc(num_workers * sizeof(*cpi->workers)));
CHECK_MEM_ERROR(cm, cpi->tile_thr_data,
aom_calloc(num_workers, sizeof(*cpi->tile_thr_data)));
for (i = 0; i < num_workers; i++) {
AVxWorker *const worker = &cpi->workers[i];
EncWorkerData *const thread_data = &cpi->tile_thr_data[i];
++cpi->num_workers;
winterface->init(worker);
thread_data->cpi = cpi;
if (i < num_workers - 1) {
// Allocate thread data.
CHECK_MEM_ERROR(cm, thread_data->td,
aom_memalign(32, sizeof(*thread_data->td)));
av1_zero(*thread_data->td);
// Set up pc_tree.
thread_data->td->leaf_tree = NULL;
thread_data->td->pc_tree = NULL;
av1_setup_pc_tree(cm, thread_data->td);
// Set up variance tree if needed.
if (cpi->sf.partition_search_type == VAR_BASED_PARTITION)
av1_setup_var_tree(cm, thread_data->td);
// Allocate frame counters in thread data.
CHECK_MEM_ERROR(cm, thread_data->td->counts,
aom_calloc(1, sizeof(*thread_data->td->counts)));
// Create threads
if (!winterface->reset(worker))
aom_internal_error(&cm->error, AOM_CODEC_ERROR,
"Tile encoder thread creation failed");
} else {
// Main thread acts as a worker and uses the thread data in cpi.
thread_data->td = &cpi->td;
}
winterface->sync(worker);
}
}
for (i = 0; i < num_workers; i++) {
AVxWorker *const worker = &cpi->workers[i];
EncWorkerData *thread_data;
worker->hook = (AVxWorkerHook)enc_worker_hook;
worker->data1 = &cpi->tile_thr_data[i];
worker->data2 = NULL;
thread_data = (EncWorkerData *)worker->data1;
// Before encoding a frame, copy the thread data from cpi.
if (thread_data->td != &cpi->td) {
thread_data->td->mb = cpi->td.mb;
thread_data->td->rd_counts = cpi->td.rd_counts;
}
if (thread_data->td->counts != &cpi->common.counts) {
memcpy(thread_data->td->counts, &cpi->common.counts,
sizeof(cpi->common.counts));
}
#if CONFIG_PALETTE
// Allocate buffers used by palette coding mode.
if (cpi->common.allow_screen_content_tools && i < num_workers - 1) {
MACROBLOCK *x = &thread_data->td->mb;
CHECK_MEM_ERROR(cm, x->palette_buffer,
aom_memalign(16, sizeof(*x->palette_buffer)));
}
#endif // CONFIG_PALETTE
}
// Encode a frame
for (i = 0; i < num_workers; i++) {
AVxWorker *const worker = &cpi->workers[i];
EncWorkerData *const thread_data = (EncWorkerData *)worker->data1;
// Set the starting tile for each thread.
thread_data->start = i;
if (i == cpi->num_workers - 1)
winterface->execute(worker);
else
winterface->launch(worker);
}
// Encoding ends.
for (i = 0; i < num_workers; i++) {
AVxWorker *const worker = &cpi->workers[i];
winterface->sync(worker);
}
for (i = 0; i < num_workers; i++) {
AVxWorker *const worker = &cpi->workers[i];
EncWorkerData *const thread_data = (EncWorkerData *)worker->data1;
// Accumulate counters.
if (i < cpi->num_workers - 1) {
av1_accumulate_frame_counts(&cm->counts, thread_data->td->counts);
accumulate_rd_opt(&cpi->td, thread_data->td);
}
}
}
static int ransac(const int *matched_points, int npoints,
int *num_inliers_by_motion, double *params_by_motion,
int num_desired_motions, const int minpts,
IsDegenerateFunc is_degenerate,
FindTransformationFunc find_transformation,
ProjectPointsDoubleFunc projectpoints) {
static const double PROBABILITY_REQUIRED = 0.9;
static const double EPS = 1e-12;
int N = 10000, trial_count = 0;
int i = 0;
int ret_val = 0;
unsigned int seed = (unsigned int)npoints;
int indices[MAX_MINPTS] = { 0 };
double *points1, *points2;
double *corners1, *corners2;
double *image1_coord;
// Store information for the num_desired_motions best transformations found
// and the worst motion among them, as well as the motion currently under
// consideration.
RANSAC_MOTION *motions, *worst_kept_motion = NULL;
RANSAC_MOTION current_motion;
// Store the parameters and the indices of the inlier points for the motion
// currently under consideration.
double params_this_motion[MAX_PARAMDIM];
double *cnp1, *cnp2;
for (i = 0; i < num_desired_motions; ++i) {
num_inliers_by_motion[i] = 0;
}
if (npoints < minpts * MINPTS_MULTIPLIER || npoints == 0) {
return 1;
}
points1 = (double *)aom_malloc(sizeof(*points1) * npoints * 2);
points2 = (double *)aom_malloc(sizeof(*points2) * npoints * 2);
corners1 = (double *)aom_malloc(sizeof(*corners1) * npoints * 2);
corners2 = (double *)aom_malloc(sizeof(*corners2) * npoints * 2);
image1_coord = (double *)aom_malloc(sizeof(*image1_coord) * npoints * 2);
motions =
(RANSAC_MOTION *)aom_malloc(sizeof(RANSAC_MOTION) * num_desired_motions);
for (i = 0; i < num_desired_motions; ++i) {
motions[i].inlier_indices =
(int *)aom_malloc(sizeof(*motions->inlier_indices) * npoints);
clear_motion(motions + i, npoints);
}
current_motion.inlier_indices =
(int *)aom_malloc(sizeof(*current_motion.inlier_indices) * npoints);
clear_motion(¤t_motion, npoints);
worst_kept_motion = motions;
if (!(points1 && points2 && corners1 && corners2 && image1_coord && motions &&
current_motion.inlier_indices)) {
ret_val = 1;
goto finish_ransac;
}
cnp1 = corners1;
cnp2 = corners2;
for (i = 0; i < npoints; ++i) {
*(cnp1++) = *(matched_points++);
*(cnp1++) = *(matched_points++);
*(cnp2++) = *(matched_points++);
*(cnp2++) = *(matched_points++);
}
while (N > trial_count) {
double sum_distance = 0.0;
double sum_distance_squared = 0.0;
clear_motion(¤t_motion, npoints);
int degenerate = 1;
int num_degenerate_iter = 0;
while (degenerate) {
num_degenerate_iter++;
if (!get_rand_indices(npoints, minpts, indices, &seed)) {
ret_val = 1;
goto finish_ransac;
}
copy_points_at_indices(points1, corners1, indices, minpts);
copy_points_at_indices(points2, corners2, indices, minpts);
degenerate = is_degenerate(points1);
if (num_degenerate_iter > MAX_DEGENERATE_ITER) {
ret_val = 1;
goto finish_ransac;
}
}
if (find_transformation(minpts, points1, points2, params_this_motion)) {
trial_count++;
continue;
}
projectpoints(params_this_motion, corners1, image1_coord, npoints, 2, 2);
for (i = 0; i < npoints; ++i) {
double dx = image1_coord[i * 2] - corners2[i * 2];
double dy = image1_coord[i * 2 + 1] - corners2[i * 2 + 1];
double distance = sqrt(dx * dx + dy * dy);
if (distance < INLIER_THRESHOLD) {
current_motion.inlier_indices[current_motion.num_inliers++] = i;
sum_distance += distance;
sum_distance_squared += distance * distance;
}
}
if (current_motion.num_inliers >= worst_kept_motion->num_inliers &&
current_motion.num_inliers > 1) {
int temp;
double fracinliers, pNoOutliers, mean_distance, dtemp;
mean_distance = sum_distance / ((double)current_motion.num_inliers);
current_motion.variance =
sum_distance_squared / ((double)current_motion.num_inliers - 1.0) -
mean_distance * mean_distance * ((double)current_motion.num_inliers) /
((double)current_motion.num_inliers - 1.0);
if (is_better_motion(¤t_motion, worst_kept_motion)) {
// This motion is better than the worst currently kept motion. Remember
// the inlier points and variance. The parameters for each kept motion
// will be recomputed later using only the inliers.
worst_kept_motion->num_inliers = current_motion.num_inliers;
worst_kept_motion->variance = current_motion.variance;
memcpy(worst_kept_motion->inlier_indices, current_motion.inlier_indices,
sizeof(*current_motion.inlier_indices) * npoints);
assert(npoints > 0);
fracinliers = (double)current_motion.num_inliers / (double)npoints;
pNoOutliers = 1 - pow(fracinliers, minpts);
pNoOutliers = fmax(EPS, pNoOutliers);
pNoOutliers = fmin(1 - EPS, pNoOutliers);
dtemp = log(1.0 - PROBABILITY_REQUIRED) / log(pNoOutliers);
temp = (dtemp > (double)INT32_MAX)
? INT32_MAX
: dtemp < (double)INT32_MIN ? INT32_MIN : (int)dtemp;
if (temp > 0 && temp < N) {
N = AOMMAX(temp, MIN_TRIALS);
}
// Determine the new worst kept motion and its num_inliers and variance.
for (i = 0; i < num_desired_motions; ++i) {
if (is_better_motion(worst_kept_motion, &motions[i])) {
worst_kept_motion = &motions[i];
}
}
}
}
trial_count++;
}
// Sort the motions, best first.
qsort(motions, num_desired_motions, sizeof(RANSAC_MOTION), compare_motions);
// Recompute the motions using only the inliers.
for (i = 0; i < num_desired_motions; ++i) {
if (motions[i].num_inliers >= minpts) {
copy_points_at_indices(points1, corners1, motions[i].inlier_indices,
motions[i].num_inliers);
copy_points_at_indices(points2, corners2, motions[i].inlier_indices,
motions[i].num_inliers);
find_transformation(motions[i].num_inliers, points1, points2,
params_by_motion + (MAX_PARAMDIM - 1) * i);
}
num_inliers_by_motion[i] = motions[i].num_inliers;
}
finish_ransac:
aom_free(points1);
aom_free(points2);
aom_free(corners1);
aom_free(corners2);
aom_free(image1_coord);
aom_free(current_motion.inlier_indices);
for (i = 0; i < num_desired_motions; ++i) {
aom_free(motions[i].inlier_indices);
}
aom_free(motions);
return ret_val;
}
static int find_homography(int np, double *pts1, double *pts2, double *mat) {
// Implemented from Peter Kovesi's normalized implementation
const int np3 = np * 3;
double *a = (double *)aom_malloc(sizeof(*a) * np3 * 18);
double *U = a + np3 * 9;
double S[9], V[9 * 9], H[9];
int i, mini;
double sx, sy, dx, dy;
double T1[9], T2[9];
normalize_homography(pts1, np, T1);
normalize_homography(pts2, np, T2);
for (i = 0; i < np; ++i) {
dx = *(pts2++);
dy = *(pts2++);
sx = *(pts1++);
sy = *(pts1++);
a[i * 3 * 9 + 0] = a[i * 3 * 9 + 1] = a[i * 3 * 9 + 2] = 0;
a[i * 3 * 9 + 3] = -sx;
a[i * 3 * 9 + 4] = -sy;
a[i * 3 * 9 + 5] = -1;
a[i * 3 * 9 + 6] = dy * sx;
a[i * 3 * 9 + 7] = dy * sy;
a[i * 3 * 9 + 8] = dy;
a[(i * 3 + 1) * 9 + 0] = sx;
a[(i * 3 + 1) * 9 + 1] = sy;
a[(i * 3 + 1) * 9 + 2] = 1;
a[(i * 3 + 1) * 9 + 3] = a[(i * 3 + 1) * 9 + 4] = a[(i * 3 + 1) * 9 + 5] =
0;
a[(i * 3 + 1) * 9 + 6] = -dx * sx;
a[(i * 3 + 1) * 9 + 7] = -dx * sy;
a[(i * 3 + 1) * 9 + 8] = -dx;
a[(i * 3 + 2) * 9 + 0] = -dy * sx;
a[(i * 3 + 2) * 9 + 1] = -dy * sy;
a[(i * 3 + 2) * 9 + 2] = -dy;
a[(i * 3 + 2) * 9 + 3] = dx * sx;
a[(i * 3 + 2) * 9 + 4] = dx * sy;
a[(i * 3 + 2) * 9 + 5] = dx;
a[(i * 3 + 2) * 9 + 6] = a[(i * 3 + 2) * 9 + 7] = a[(i * 3 + 2) * 9 + 8] =
0;
}
if (SVD(U, S, V, a, np3, 9)) {
aom_free(a);
return 1;
} else {
double minS = 1e12;
mini = -1;
for (i = 0; i < 9; ++i) {
if (S[i] < minS) {
minS = S[i];
mini = i;
}
}
}
for (i = 0; i < 9; i++) H[i] = V[i * 9 + mini];
denormalize_homography_reorder(H, T1, T2);
aom_free(a);
if (H[8] == 0.0) {
return 1;
} else {
// normalize
double f = 1.0 / H[8];
for (i = 0; i < 8; i++) mat[i] = f * H[i];
}
return 0;
}
static int find_hortrapezoid(int np, double *pts1, double *pts2, double *mat) {
const int np3 = np * 3;
double *a = (double *)aom_malloc(sizeof(*a) * np3 * 14);
double *U = a + np3 * 7;
double S[7], V[7 * 7], H[9];
int i, mini;
double sx, sy, dx, dy;
double T1[9], T2[9];
normalize_homography(pts1, np, T1);
normalize_homography(pts2, np, T2);
for (i = 0; i < np; ++i) {
dx = *(pts2++);
dy = *(pts2++);
sx = *(pts1++);
sy = *(pts1++);
a[i * 3 * 7 + 0] = a[i * 3 * 7 + 1] = a[i * 3 * 7 + 2] = 0;
a[i * 3 * 7 + 3] = -sy;
a[i * 3 * 7 + 4] = -1;
a[i * 3 * 7 + 5] = dy * sy;
a[i * 3 * 7 + 6] = dy;
a[(i * 3 + 1) * 7 + 0] = sx;
a[(i * 3 + 1) * 7 + 1] = sy;
a[(i * 3 + 1) * 7 + 2] = 1;
a[(i * 3 + 1) * 7 + 3] = a[(i * 3 + 1) * 7 + 4] = 0;
a[(i * 3 + 1) * 7 + 5] = -dx * sy;
a[(i * 3 + 1) * 7 + 6] = -dx;
a[(i * 3 + 2) * 7 + 0] = -dy * sx;
a[(i * 3 + 2) * 7 + 1] = -dy * sy;
a[(i * 3 + 2) * 7 + 2] = -dy;
a[(i * 3 + 2) * 7 + 3] = dx * sy;
a[(i * 3 + 2) * 7 + 4] = dx;
a[(i * 3 + 2) * 7 + 5] = a[(i * 3 + 2) * 7 + 6] = 0;
}
if (SVD(U, S, V, a, np3, 7)) {
aom_free(a);
return 1;
} else {
double minS = 1e12;
mini = -1;
for (i = 0; i < 7; ++i) {
if (S[i] < minS) {
minS = S[i];
mini = i;
}
}
}
H[3] = H[6] = 0;
for (i = 0; i < 3; i++) H[i] = V[i * 7 + mini];
for (; i < 5; i++) H[i + 1] = V[i * 7 + mini];
for (; i < 7; i++) H[i + 2] = V[i * 7 + mini];
denormalize_homography_reorder(H, T1, T2);
aom_free(a);
if (H[8] == 0.0) {
return 1;
} else {
// normalize
double f = 1.0 / H[8];
for (i = 0; i < 8; i++) mat[i] = f * H[i];
}
return 0;
}
void av1_dering_frame(YV12_BUFFER_CONFIG *frame, AV1_COMMON *cm,
MACROBLOCKD *xd, int global_level) {
int r, c;
int sbr, sbc;
int nhsb, nvsb;
int16_t src[OD_DERING_INBUF_SIZE];
int16_t *linebuf[3];
int16_t colbuf[3][OD_BSIZE_MAX + 2 * OD_FILT_VBORDER][OD_FILT_HBORDER];
dering_list dlist[MAX_MIB_SIZE * MAX_MIB_SIZE];
unsigned char *row_dering, *prev_row_dering, *curr_row_dering;
int dering_count;
int dir[OD_DERING_NBLOCKS][OD_DERING_NBLOCKS] = { { 0 } };
int stride;
int bsize[3];
int dec[3];
int pli;
int dering_left;
int coeff_shift = AOMMAX(cm->bit_depth - 8, 0);
int nplanes;
if (xd->plane[1].subsampling_x == xd->plane[1].subsampling_y &&
xd->plane[2].subsampling_x == xd->plane[2].subsampling_y)
nplanes = 3;
else
nplanes = 1;
nvsb = (cm->mi_rows + MAX_MIB_SIZE - 1) / MAX_MIB_SIZE;
nhsb = (cm->mi_cols + MAX_MIB_SIZE - 1) / MAX_MIB_SIZE;
av1_setup_dst_planes(xd->plane, frame, 0, 0);
row_dering = aom_malloc(sizeof(*row_dering) * nhsb * 2);
memset(row_dering, 1, sizeof(*row_dering) * (nhsb + 2) * 2);
prev_row_dering = row_dering + 1;
curr_row_dering = prev_row_dering + nhsb + 2;
for (pli = 0; pli < nplanes; pli++) {
dec[pli] = xd->plane[pli].subsampling_x;
bsize[pli] = OD_DERING_SIZE_LOG2 - dec[pli];
}
stride = (cm->mi_cols << bsize[0]) + 2 * OD_FILT_HBORDER;
for (pli = 0; pli < nplanes; pli++) {
linebuf[pli] = aom_malloc(sizeof(*linebuf) * OD_FILT_VBORDER * stride);
}
for (sbr = 0; sbr < nvsb; sbr++) {
for (pli = 0; pli < nplanes; pli++) {
for (r = 0; r < (MAX_MIB_SIZE << bsize[pli]) + 2 * OD_FILT_VBORDER; r++) {
for (c = 0; c < OD_FILT_HBORDER; c++) {
colbuf[pli][r][c] = OD_DERING_VERY_LARGE;
}
}
}
dering_left = 1;
for (sbc = 0; sbc < nhsb; sbc++) {
int level;
int nhb, nvb;
int cstart = 0;
if (!dering_left) cstart = -OD_FILT_HBORDER;
nhb = AOMMIN(MAX_MIB_SIZE, cm->mi_cols - MAX_MIB_SIZE * sbc);
nvb = AOMMIN(MAX_MIB_SIZE, cm->mi_rows - MAX_MIB_SIZE * sbr);
level = compute_level_from_index(
global_level, cm->mi_grid_visible[MAX_MIB_SIZE * sbr * cm->mi_stride +
MAX_MIB_SIZE * sbc]
->mbmi.dering_gain);
curr_row_dering[sbc] = 0;
if (level == 0 ||
(dering_count = sb_compute_dering_list(
cm, sbr * MAX_MIB_SIZE, sbc * MAX_MIB_SIZE, dlist)) == 0) {
dering_left = 0;
continue;
}
curr_row_dering[sbc] = 1;
for (pli = 0; pli < nplanes; pli++) {
int16_t dst[OD_BSIZE_MAX * OD_BSIZE_MAX];
int threshold;
int coffset;
int rend, cend;
if (sbc == nhsb - 1)
cend = (nhb << bsize[pli]);
else
cend = (nhb << bsize[pli]) + OD_FILT_HBORDER;
if (sbr == nvsb - 1)
rend = (nvb << bsize[pli]);
else
rend = (nvb << bsize[pli]) + OD_FILT_VBORDER;
coffset = sbc * MAX_MIB_SIZE << bsize[pli];
if (sbc == nhsb - 1) {
/* On the last superblock column, fill in the right border with
OD_DERING_VERY_LARGE to avoid filtering with the outside. */
for (r = 0; r < rend + OD_FILT_VBORDER; r++) {
for (c = cend; c < (nhb << bsize[pli]) + OD_FILT_HBORDER; ++c) {
src[r * OD_FILT_BSTRIDE + c + OD_FILT_HBORDER] =
OD_DERING_VERY_LARGE;
}
}
}
if (sbr == nvsb - 1) {
/* On the last superblock row, fill in the bottom border with
OD_DERING_VERY_LARGE to avoid filtering with the outside. */
for (r = rend; r < rend + OD_FILT_VBORDER; r++) {
for (c = 0; c < (nhb << bsize[pli]) + 2 * OD_FILT_HBORDER; c++) {
src[(r + OD_FILT_VBORDER) * OD_FILT_BSTRIDE + c] =
OD_DERING_VERY_LARGE;
}
}
}
/* Copy in the pixels we need from the current superblock for
deringing.*/
copy_sb8_16(
cm,
&src[OD_FILT_VBORDER * OD_FILT_BSTRIDE + OD_FILT_HBORDER + cstart],
OD_FILT_BSTRIDE, xd->plane[pli].dst.buf,
(MAX_MIB_SIZE << bsize[pli]) * sbr, coffset + cstart,
xd->plane[pli].dst.stride, rend, cend - cstart);
if (!prev_row_dering[sbc]) {
copy_sb8_16(cm, &src[OD_FILT_HBORDER], OD_FILT_BSTRIDE,
xd->plane[pli].dst.buf,
(MAX_MIB_SIZE << bsize[pli]) * sbr - OD_FILT_VBORDER,
coffset, xd->plane[pli].dst.stride, OD_FILT_VBORDER,
nhb << bsize[pli]);
} else if (sbr > 0) {
for (r = 0; r < OD_FILT_VBORDER; r++) {
for (c = 0; c < nhb << bsize[pli]; c++) {
src[r * OD_FILT_BSTRIDE + c + OD_FILT_HBORDER] =
linebuf[pli][r * stride + coffset + c];
}
}
} else {
for (r = 0; r < OD_FILT_VBORDER; r++) {
for (c = 0; c < nhb << bsize[pli]; c++) {
src[r * OD_FILT_BSTRIDE + c + OD_FILT_HBORDER] =
OD_DERING_VERY_LARGE;
}
}
}
if (!prev_row_dering[sbc - 1]) {
copy_sb8_16(cm, src, OD_FILT_BSTRIDE, xd->plane[pli].dst.buf,
(MAX_MIB_SIZE << bsize[pli]) * sbr - OD_FILT_VBORDER,
coffset - OD_FILT_HBORDER, xd->plane[pli].dst.stride,
OD_FILT_VBORDER, OD_FILT_HBORDER);
} else if (sbr > 0 && sbc > 0) {
for (r = 0; r < OD_FILT_VBORDER; r++) {
for (c = -OD_FILT_HBORDER; c < 0; c++) {
src[r * OD_FILT_BSTRIDE + c + OD_FILT_HBORDER] =
linebuf[pli][r * stride + coffset + c];
}
}
} else {
for (r = 0; r < OD_FILT_VBORDER; r++) {
for (c = -OD_FILT_HBORDER; c < 0; c++) {
src[r * OD_FILT_BSTRIDE + c + OD_FILT_HBORDER] =
OD_DERING_VERY_LARGE;
}
}
}
if (!prev_row_dering[sbc + 1]) {
copy_sb8_16(cm, &src[OD_FILT_HBORDER + (nhb << bsize[pli])],
OD_FILT_BSTRIDE, xd->plane[pli].dst.buf,
(MAX_MIB_SIZE << bsize[pli]) * sbr - OD_FILT_VBORDER,
coffset + (nhb << bsize[pli]), xd->plane[pli].dst.stride,
OD_FILT_VBORDER, OD_FILT_HBORDER);
} else if (sbr > 0 && sbc < nhsb - 1) {
for (r = 0; r < OD_FILT_VBORDER; r++) {
for (c = nhb << bsize[pli];
c < (nhb << bsize[pli]) + OD_FILT_HBORDER; c++) {
src[r * OD_FILT_BSTRIDE + c + OD_FILT_HBORDER] =
linebuf[pli][r * stride + coffset + c];
}
}
} else {
for (r = 0; r < OD_FILT_VBORDER; r++) {
for (c = nhb << bsize[pli];
c < (nhb << bsize[pli]) + OD_FILT_HBORDER; c++) {
src[r * OD_FILT_BSTRIDE + c + OD_FILT_HBORDER] =
OD_DERING_VERY_LARGE;
}
}
}
if (dering_left) {
/* If we deringed the superblock on the left then we need to copy in
saved pixels. */
for (r = 0; r < rend + OD_FILT_VBORDER; r++) {
for (c = 0; c < OD_FILT_HBORDER; c++) {
src[r * OD_FILT_BSTRIDE + c] = colbuf[pli][r][c];
}
}
}
for (r = 0; r < rend + OD_FILT_VBORDER; r++) {
for (c = 0; c < OD_FILT_HBORDER; c++) {
/* Saving pixels in case we need to dering the superblock on the
right. */
colbuf[pli][r][c] =
src[r * OD_FILT_BSTRIDE + c + (nhb << bsize[pli])];
}
}
copy_sb8_16(cm, &linebuf[pli][coffset], stride, xd->plane[pli].dst.buf,
(MAX_MIB_SIZE << bsize[pli]) * (sbr + 1) - OD_FILT_VBORDER,
coffset, xd->plane[pli].dst.stride, OD_FILT_VBORDER,
(nhb << bsize[pli]));
/* FIXME: This is a temporary hack that uses more conservative
deringing for chroma. */
if (pli)
threshold = (level * 5 + 4) >> 3 << coeff_shift;
else
threshold = level << coeff_shift;
if (threshold == 0) continue;
od_dering(
dst, &src[OD_FILT_VBORDER * OD_FILT_BSTRIDE + OD_FILT_HBORDER],
dec[pli], dir, pli, dlist, dering_count, threshold, coeff_shift);
#if CONFIG_AOM_HIGHBITDEPTH
if (cm->use_highbitdepth) {
copy_dering_16bit_to_16bit(
(int16_t *)&CONVERT_TO_SHORTPTR(
xd->plane[pli]
.dst.buf)[xd->plane[pli].dst.stride *
(MAX_MIB_SIZE * sbr << bsize[pli]) +
(sbc * MAX_MIB_SIZE << bsize[pli])],
xd->plane[pli].dst.stride, dst, dlist, dering_count,
3 - dec[pli]);
} else {
#endif
copy_dering_16bit_to_8bit(
&xd->plane[pli].dst.buf[xd->plane[pli].dst.stride *
(MAX_MIB_SIZE * sbr << bsize[pli]) +
(sbc * MAX_MIB_SIZE << bsize[pli])],
xd->plane[pli].dst.stride, dst, dlist, dering_count, bsize[pli]);
#if CONFIG_AOM_HIGHBITDEPTH
}
#endif
}
dering_left = 1;
}
{
unsigned char *tmp;
tmp = prev_row_dering;
prev_row_dering = curr_row_dering;
curr_row_dering = tmp;
}
}
static int svdcmp(double **u, int m, int n, double w[], double **v) {
const int max_its = 30;
int flag, i, its, j, jj, k, l, nm;
double anorm, c, f, g, h, s, scale, x, y, z;
double *rv1 = (double *)aom_malloc(sizeof(*rv1) * (n + 1));
g = scale = anorm = 0.0;
for (i = 0; i < n; i++) {
l = i + 1;
rv1[i] = scale * g;
g = s = scale = 0.0;
if (i < m) {
for (k = i; k < m; k++) scale += fabs(u[k][i]);
if (scale != 0.) {
for (k = i; k < m; k++) {
u[k][i] /= scale;
s += u[k][i] * u[k][i];
}
f = u[i][i];
g = -sign(sqrt(s), f);
h = f * g - s;
u[i][i] = f - g;
for (j = l; j < n; j++) {
for (s = 0.0, k = i; k < m; k++) s += u[k][i] * u[k][j];
f = s / h;
for (k = i; k < m; k++) u[k][j] += f * u[k][i];
}
for (k = i; k < m; k++) u[k][i] *= scale;
}
}
w[i] = scale * g;
g = s = scale = 0.0;
if (i < m && i != n - 1) {
for (k = l; k < n; k++) scale += fabs(u[i][k]);
if (scale != 0.) {
for (k = l; k < n; k++) {
u[i][k] /= scale;
s += u[i][k] * u[i][k];
}
f = u[i][l];
g = -sign(sqrt(s), f);
h = f * g - s;
u[i][l] = f - g;
for (k = l; k < n; k++) rv1[k] = u[i][k] / h;
for (j = l; j < m; j++) {
for (s = 0.0, k = l; k < n; k++) s += u[j][k] * u[i][k];
for (k = l; k < n; k++) u[j][k] += s * rv1[k];
}
for (k = l; k < n; k++) u[i][k] *= scale;
}
}
anorm = fmax(anorm, (fabs(w[i]) + fabs(rv1[i])));
}
for (i = n - 1; i >= 0; i--) {
if (i < n - 1) {
if (g != 0.) {
for (j = l; j < n; j++) v[j][i] = (u[i][j] / u[i][l]) / g;
for (j = l; j < n; j++) {
for (s = 0.0, k = l; k < n; k++) s += u[i][k] * v[k][j];
for (k = l; k < n; k++) v[k][j] += s * v[k][i];
}
}
for (j = l; j < n; j++) v[i][j] = v[j][i] = 0.0;
}
v[i][i] = 1.0;
g = rv1[i];
l = i;
}
for (i = AOMMIN(m, n) - 1; i >= 0; i--) {
l = i + 1;
g = w[i];
for (j = l; j < n; j++) u[i][j] = 0.0;
if (g != 0.) {
g = 1.0 / g;
for (j = l; j < n; j++) {
for (s = 0.0, k = l; k < m; k++) s += u[k][i] * u[k][j];
f = (s / u[i][i]) * g;
for (k = i; k < m; k++) u[k][j] += f * u[k][i];
}
for (j = i; j < m; j++) u[j][i] *= g;
} else {
for (j = i; j < m; j++) u[j][i] = 0.0;
}
++u[i][i];
}
for (k = n - 1; k >= 0; k--) {
for (its = 0; its < max_its; its++) {
flag = 1;
for (l = k; l >= 0; l--) {
nm = l - 1;
if ((double)(fabs(rv1[l]) + anorm) == anorm || nm < 0) {
flag = 0;
break;
}
if ((double)(fabs(w[nm]) + anorm) == anorm) break;
}
if (flag) {
c = 0.0;
s = 1.0;
for (i = l; i <= k; i++) {
f = s * rv1[i];
rv1[i] = c * rv1[i];
if ((double)(fabs(f) + anorm) == anorm) break;
g = w[i];
h = pythag(f, g);
w[i] = h;
h = 1.0 / h;
c = g * h;
s = -f * h;
for (j = 0; j < m; j++) {
y = u[j][nm];
z = u[j][i];
u[j][nm] = y * c + z * s;
u[j][i] = z * c - y * s;
}
}
}
z = w[k];
if (l == k) {
if (z < 0.0) {
w[k] = -z;
for (j = 0; j < n; j++) v[j][k] = -v[j][k];
}
break;
}
if (its == max_its - 1) {
aom_free(rv1);
return 1;
}
assert(k > 0);
x = w[l];
nm = k - 1;
y = w[nm];
g = rv1[nm];
h = rv1[k];
f = ((y - z) * (y + z) + (g - h) * (g + h)) / (2.0 * h * y);
g = pythag(f, 1.0);
f = ((x - z) * (x + z) + h * ((y / (f + sign(g, f))) - h)) / x;
c = s = 1.0;
for (j = l; j <= nm; j++) {
i = j + 1;
g = rv1[i];
y = w[i];
h = s * g;
g = c * g;
z = pythag(f, h);
rv1[j] = z;
c = f / z;
s = h / z;
f = x * c + g * s;
g = g * c - x * s;
h = y * s;
y *= c;
for (jj = 0; jj < n; jj++) {
x = v[jj][j];
z = v[jj][i];
v[jj][j] = x * c + z * s;
v[jj][i] = z * c - x * s;
}
z = pythag(f, h);
w[j] = z;
if (z != 0.) {
z = 1.0 / z;
c = f * z;
s = h * z;
}
f = c * g + s * y;
x = c * y - s * g;
for (jj = 0; jj < m; jj++) {
y = u[jj][j];
z = u[jj][i];
u[jj][j] = y * c + z * s;
u[jj][i] = z * c - y * s;
}
}
rv1[l] = 0.0;
rv1[k] = f;
w[k] = x;
}
}
aom_free(rv1);
return 0;
}
int aom_noise_data_validate(const double *data, int w, int h) {
const double kVarianceThreshold = 2;
const double kMeanThreshold = 2;
int x = 0, y = 0;
int ret_value = 1;
double var = 0, mean = 0;
double *mean_x, *mean_y, *var_x, *var_y;
// Check that noise variance is not increasing in x or y
// and that the data is zero mean.
mean_x = (double *)aom_malloc(sizeof(*mean_x) * w);
var_x = (double *)aom_malloc(sizeof(*var_x) * w);
mean_y = (double *)aom_malloc(sizeof(*mean_x) * h);
var_y = (double *)aom_malloc(sizeof(*var_y) * h);
memset(mean_x, 0, sizeof(*mean_x) * w);
memset(var_x, 0, sizeof(*var_x) * w);
memset(mean_y, 0, sizeof(*mean_y) * h);
memset(var_y, 0, sizeof(*var_y) * h);
for (y = 0; y < h; ++y) {
for (x = 0; x < w; ++x) {
const double d = data[y * w + x];
var_x[x] += d * d;
var_y[y] += d * d;
mean_x[x] += d;
mean_y[y] += d;
var += d * d;
mean += d;
}
}
mean /= (w * h);
var = var / (w * h) - mean * mean;
for (y = 0; y < h; ++y) {
mean_y[y] /= h;
var_y[y] = var_y[y] / h - mean_y[y] * mean_y[y];
if (fabs(var_y[y] - var) >= kVarianceThreshold) {
fprintf(stderr, "Variance distance too large %f %f\n", var_y[y], var);
ret_value = 0;
break;
}
if (fabs(mean_y[y] - mean) >= kMeanThreshold) {
fprintf(stderr, "Mean distance too large %f %f\n", mean_y[y], mean);
ret_value = 0;
break;
}
}
for (x = 0; x < w; ++x) {
mean_x[x] /= w;
var_x[x] = var_x[x] / w - mean_x[x] * mean_x[x];
if (fabs(var_x[x] - var) >= kVarianceThreshold) {
fprintf(stderr, "Variance distance too large %f %f\n", var_x[x], var);
ret_value = 0;
break;
}
if (fabs(mean_x[x] - mean) >= kMeanThreshold) {
fprintf(stderr, "Mean distance too large %f %f\n", mean_x[x], mean);
ret_value = 0;
break;
}
}
aom_free(mean_x);
aom_free(mean_y);
aom_free(var_x);
aom_free(var_y);
return ret_value;
}
static void ifd_init_mi_rc(insp_frame_data *fd, int mi_cols, int mi_rows) {
fd->mi_cols = mi_cols;
fd->mi_rows = mi_rows;
fd->mi_grid = (insp_mi_data *)aom_malloc(sizeof(insp_mi_data) * fd->mi_rows *
fd->mi_cols);
}