LCOV - code coverage report
Current view: top level - third_party/aom/av1/encoder - mathutils.h (source / functions) Hit Total Coverage
Test: output.info Lines: 0 227 0.0 %
Date: 2017-07-14 16:53:18 Functions: 0 7 0.0 %
Legend: Lines: hit not hit

          Line data    Source code
       1             : /*
       2             :  * Copyright (c) 2017, Alliance for Open Media. All rights reserved
       3             :  *
       4             :  * This source code is subject to the terms of the BSD 2 Clause License and
       5             :  * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
       6             :  * was not distributed with this source code in the LICENSE file, you can
       7             :  * obtain it at www.aomedia.org/license/software. If the Alliance for Open
       8             :  * Media Patent License 1.0 was not distributed with this source code in the
       9             :  * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
      10             :  */
      11             : 
      12             : #include <memory.h>
      13             : #include <math.h>
      14             : #include <stdio.h>
      15             : #include <stdlib.h>
      16             : #include <assert.h>
      17             : 
      18             : static const double TINY_NEAR_ZERO = 1.0E-16;
      19             : 
      20             : // Solves Ax = b, where x and b are column vectors of size nx1 and A is nxn
      21           0 : static INLINE int linsolve(int n, double *A, int stride, double *b, double *x) {
      22             :   int i, j, k;
      23             :   double c;
      24             :   // Forward elimination
      25           0 :   for (k = 0; k < n - 1; k++) {
      26             :     // Bring the largest magitude to the diagonal position
      27           0 :     for (i = n - 1; i > k; i--) {
      28           0 :       if (fabs(A[(i - 1) * stride + k]) < fabs(A[i * stride + k])) {
      29           0 :         for (j = 0; j < n; j++) {
      30           0 :           c = A[i * stride + j];
      31           0 :           A[i * stride + j] = A[(i - 1) * stride + j];
      32           0 :           A[(i - 1) * stride + j] = c;
      33             :         }
      34           0 :         c = b[i];
      35           0 :         b[i] = b[i - 1];
      36           0 :         b[i - 1] = c;
      37             :       }
      38             :     }
      39           0 :     for (i = k; i < n - 1; i++) {
      40           0 :       if (fabs(A[k * stride + k]) < TINY_NEAR_ZERO) return 0;
      41           0 :       c = A[(i + 1) * stride + k] / A[k * stride + k];
      42           0 :       for (j = 0; j < n; j++) A[(i + 1) * stride + j] -= c * A[k * stride + j];
      43           0 :       b[i + 1] -= c * b[k];
      44             :     }
      45             :   }
      46             :   // Backward substitution
      47           0 :   for (i = n - 1; i >= 0; i--) {
      48           0 :     if (fabs(A[i * stride + i]) < TINY_NEAR_ZERO) return 0;
      49           0 :     c = 0;
      50           0 :     for (j = i + 1; j <= n - 1; j++) c += A[i * stride + j] * x[j];
      51           0 :     x[i] = (b[i] - c) / A[i * stride + i];
      52             :   }
      53             : 
      54           0 :   return 1;
      55             : }
      56             : 
      57             : ////////////////////////////////////////////////////////////////////////////////
      58             : // Least-squares
      59             : // Solves for n-dim x in a least squares sense to minimize |Ax - b|^2
      60             : // The solution is simply x = (A'A)^-1 A'b or simply the solution for
      61             : // the system: A'A x = A'b
      62           0 : static INLINE int least_squares(int n, double *A, int rows, int stride,
      63             :                                 double *b, double *scratch, double *x) {
      64             :   int i, j, k;
      65           0 :   double *scratch_ = NULL;
      66             :   double *AtA, *Atb;
      67           0 :   if (!scratch) {
      68           0 :     scratch_ = (double *)aom_malloc(sizeof(*scratch) * n * (n + 1));
      69           0 :     scratch = scratch_;
      70             :   }
      71           0 :   AtA = scratch;
      72           0 :   Atb = scratch + n * n;
      73             : 
      74           0 :   for (i = 0; i < n; ++i) {
      75           0 :     for (j = i; j < n; ++j) {
      76           0 :       AtA[i * n + j] = 0.0;
      77           0 :       for (k = 0; k < rows; ++k)
      78           0 :         AtA[i * n + j] += A[k * stride + i] * A[k * stride + j];
      79           0 :       AtA[j * n + i] = AtA[i * n + j];
      80             :     }
      81           0 :     Atb[i] = 0;
      82           0 :     for (k = 0; k < rows; ++k) Atb[i] += A[k * stride + i] * b[k];
      83             :   }
      84           0 :   int ret = linsolve(n, AtA, n, Atb, x);
      85           0 :   if (scratch_) aom_free(scratch_);
      86           0 :   return ret;
      87             : }
      88             : 
      89             : // Matrix multiply
      90           0 : static INLINE void multiply_mat(const double *m1, const double *m2, double *res,
      91             :                                 const int m1_rows, const int inner_dim,
      92             :                                 const int m2_cols) {
      93             :   double sum;
      94             : 
      95             :   int row, col, inner;
      96           0 :   for (row = 0; row < m1_rows; ++row) {
      97           0 :     for (col = 0; col < m2_cols; ++col) {
      98           0 :       sum = 0;
      99           0 :       for (inner = 0; inner < inner_dim; ++inner)
     100           0 :         sum += m1[row * inner_dim + inner] * m2[inner * m2_cols + col];
     101           0 :       *(res++) = sum;
     102             :     }
     103             :   }
     104           0 : }
     105             : 
     106             : //
     107             : // The functions below are needed only for homography computation
     108             : // Remove if the homography models are not used.
     109             : //
     110             : ///////////////////////////////////////////////////////////////////////////////
     111             : // svdcmp
     112             : // Adopted from Numerical Recipes in C
     113             : 
     114           0 : static INLINE double sign(double a, double b) {
     115           0 :   return ((b) >= 0 ? fabs(a) : -fabs(a));
     116             : }
     117             : 
     118           0 : static INLINE double pythag(double a, double b) {
     119             :   double ct;
     120           0 :   const double absa = fabs(a);
     121           0 :   const double absb = fabs(b);
     122             : 
     123           0 :   if (absa > absb) {
     124           0 :     ct = absb / absa;
     125           0 :     return absa * sqrt(1.0 + ct * ct);
     126             :   } else {
     127           0 :     ct = absa / absb;
     128           0 :     return (absb == 0) ? 0 : absb * sqrt(1.0 + ct * ct);
     129             :   }
     130             : }
     131             : 
     132           0 : static INLINE int svdcmp(double **u, int m, int n, double w[], double **v) {
     133           0 :   const int max_its = 30;
     134             :   int flag, i, its, j, jj, k, l, nm;
     135             :   double anorm, c, f, g, h, s, scale, x, y, z;
     136           0 :   double *rv1 = (double *)aom_malloc(sizeof(*rv1) * (n + 1));
     137           0 :   g = scale = anorm = 0.0;
     138           0 :   for (i = 0; i < n; i++) {
     139           0 :     l = i + 1;
     140           0 :     rv1[i] = scale * g;
     141           0 :     g = s = scale = 0.0;
     142           0 :     if (i < m) {
     143           0 :       for (k = i; k < m; k++) scale += fabs(u[k][i]);
     144           0 :       if (scale != 0.) {
     145           0 :         for (k = i; k < m; k++) {
     146           0 :           u[k][i] /= scale;
     147           0 :           s += u[k][i] * u[k][i];
     148             :         }
     149           0 :         f = u[i][i];
     150           0 :         g = -sign(sqrt(s), f);
     151           0 :         h = f * g - s;
     152           0 :         u[i][i] = f - g;
     153           0 :         for (j = l; j < n; j++) {
     154           0 :           for (s = 0.0, k = i; k < m; k++) s += u[k][i] * u[k][j];
     155           0 :           f = s / h;
     156           0 :           for (k = i; k < m; k++) u[k][j] += f * u[k][i];
     157             :         }
     158           0 :         for (k = i; k < m; k++) u[k][i] *= scale;
     159             :       }
     160             :     }
     161           0 :     w[i] = scale * g;
     162           0 :     g = s = scale = 0.0;
     163           0 :     if (i < m && i != n - 1) {
     164           0 :       for (k = l; k < n; k++) scale += fabs(u[i][k]);
     165           0 :       if (scale != 0.) {
     166           0 :         for (k = l; k < n; k++) {
     167           0 :           u[i][k] /= scale;
     168           0 :           s += u[i][k] * u[i][k];
     169             :         }
     170           0 :         f = u[i][l];
     171           0 :         g = -sign(sqrt(s), f);
     172           0 :         h = f * g - s;
     173           0 :         u[i][l] = f - g;
     174           0 :         for (k = l; k < n; k++) rv1[k] = u[i][k] / h;
     175           0 :         for (j = l; j < m; j++) {
     176           0 :           for (s = 0.0, k = l; k < n; k++) s += u[j][k] * u[i][k];
     177           0 :           for (k = l; k < n; k++) u[j][k] += s * rv1[k];
     178             :         }
     179           0 :         for (k = l; k < n; k++) u[i][k] *= scale;
     180             :       }
     181             :     }
     182           0 :     anorm = fmax(anorm, (fabs(w[i]) + fabs(rv1[i])));
     183             :   }
     184             : 
     185           0 :   for (i = n - 1; i >= 0; i--) {
     186           0 :     if (i < n - 1) {
     187           0 :       if (g != 0.) {
     188           0 :         for (j = l; j < n; j++) v[j][i] = (u[i][j] / u[i][l]) / g;
     189           0 :         for (j = l; j < n; j++) {
     190           0 :           for (s = 0.0, k = l; k < n; k++) s += u[i][k] * v[k][j];
     191           0 :           for (k = l; k < n; k++) v[k][j] += s * v[k][i];
     192             :         }
     193             :       }
     194           0 :       for (j = l; j < n; j++) v[i][j] = v[j][i] = 0.0;
     195             :     }
     196           0 :     v[i][i] = 1.0;
     197           0 :     g = rv1[i];
     198           0 :     l = i;
     199             :   }
     200           0 :   for (i = AOMMIN(m, n) - 1; i >= 0; i--) {
     201           0 :     l = i + 1;
     202           0 :     g = w[i];
     203           0 :     for (j = l; j < n; j++) u[i][j] = 0.0;
     204           0 :     if (g != 0.) {
     205           0 :       g = 1.0 / g;
     206           0 :       for (j = l; j < n; j++) {
     207           0 :         for (s = 0.0, k = l; k < m; k++) s += u[k][i] * u[k][j];
     208           0 :         f = (s / u[i][i]) * g;
     209           0 :         for (k = i; k < m; k++) u[k][j] += f * u[k][i];
     210             :       }
     211           0 :       for (j = i; j < m; j++) u[j][i] *= g;
     212             :     } else {
     213           0 :       for (j = i; j < m; j++) u[j][i] = 0.0;
     214             :     }
     215           0 :     ++u[i][i];
     216             :   }
     217           0 :   for (k = n - 1; k >= 0; k--) {
     218           0 :     for (its = 0; its < max_its; its++) {
     219           0 :       flag = 1;
     220           0 :       for (l = k; l >= 0; l--) {
     221           0 :         nm = l - 1;
     222           0 :         if ((double)(fabs(rv1[l]) + anorm) == anorm || nm < 0) {
     223           0 :           flag = 0;
     224           0 :           break;
     225             :         }
     226           0 :         if ((double)(fabs(w[nm]) + anorm) == anorm) break;
     227             :       }
     228           0 :       if (flag) {
     229           0 :         c = 0.0;
     230           0 :         s = 1.0;
     231           0 :         for (i = l; i <= k; i++) {
     232           0 :           f = s * rv1[i];
     233           0 :           rv1[i] = c * rv1[i];
     234           0 :           if ((double)(fabs(f) + anorm) == anorm) break;
     235           0 :           g = w[i];
     236           0 :           h = pythag(f, g);
     237           0 :           w[i] = h;
     238           0 :           h = 1.0 / h;
     239           0 :           c = g * h;
     240           0 :           s = -f * h;
     241           0 :           for (j = 0; j < m; j++) {
     242           0 :             y = u[j][nm];
     243           0 :             z = u[j][i];
     244           0 :             u[j][nm] = y * c + z * s;
     245           0 :             u[j][i] = z * c - y * s;
     246             :           }
     247             :         }
     248             :       }
     249           0 :       z = w[k];
     250           0 :       if (l == k) {
     251           0 :         if (z < 0.0) {
     252           0 :           w[k] = -z;
     253           0 :           for (j = 0; j < n; j++) v[j][k] = -v[j][k];
     254             :         }
     255           0 :         break;
     256             :       }
     257           0 :       if (its == max_its - 1) {
     258           0 :         aom_free(rv1);
     259           0 :         return 1;
     260             :       }
     261           0 :       assert(k > 0);
     262           0 :       x = w[l];
     263           0 :       nm = k - 1;
     264           0 :       y = w[nm];
     265           0 :       g = rv1[nm];
     266           0 :       h = rv1[k];
     267           0 :       f = ((y - z) * (y + z) + (g - h) * (g + h)) / (2.0 * h * y);
     268           0 :       g = pythag(f, 1.0);
     269           0 :       f = ((x - z) * (x + z) + h * ((y / (f + sign(g, f))) - h)) / x;
     270           0 :       c = s = 1.0;
     271           0 :       for (j = l; j <= nm; j++) {
     272           0 :         i = j + 1;
     273           0 :         g = rv1[i];
     274           0 :         y = w[i];
     275           0 :         h = s * g;
     276           0 :         g = c * g;
     277           0 :         z = pythag(f, h);
     278           0 :         rv1[j] = z;
     279           0 :         c = f / z;
     280           0 :         s = h / z;
     281           0 :         f = x * c + g * s;
     282           0 :         g = g * c - x * s;
     283           0 :         h = y * s;
     284           0 :         y *= c;
     285           0 :         for (jj = 0; jj < n; jj++) {
     286           0 :           x = v[jj][j];
     287           0 :           z = v[jj][i];
     288           0 :           v[jj][j] = x * c + z * s;
     289           0 :           v[jj][i] = z * c - x * s;
     290             :         }
     291           0 :         z = pythag(f, h);
     292           0 :         w[j] = z;
     293           0 :         if (z != 0.) {
     294           0 :           z = 1.0 / z;
     295           0 :           c = f * z;
     296           0 :           s = h * z;
     297             :         }
     298           0 :         f = c * g + s * y;
     299           0 :         x = c * y - s * g;
     300           0 :         for (jj = 0; jj < m; jj++) {
     301           0 :           y = u[jj][j];
     302           0 :           z = u[jj][i];
     303           0 :           u[jj][j] = y * c + z * s;
     304           0 :           u[jj][i] = z * c - y * s;
     305             :         }
     306             :       }
     307           0 :       rv1[l] = 0.0;
     308           0 :       rv1[k] = f;
     309           0 :       w[k] = x;
     310             :     }
     311             :   }
     312           0 :   aom_free(rv1);
     313           0 :   return 0;
     314             : }
     315             : 
     316           0 : static INLINE int SVD(double *U, double *W, double *V, double *matx, int M,
     317             :                       int N) {
     318             :   // Assumes allocation for U is MxN
     319           0 :   double **nrU = (double **)aom_malloc((M) * sizeof(*nrU));
     320           0 :   double **nrV = (double **)aom_malloc((N) * sizeof(*nrV));
     321             :   int problem, i;
     322             : 
     323           0 :   problem = !(nrU && nrV);
     324           0 :   if (!problem) {
     325           0 :     for (i = 0; i < M; i++) {
     326           0 :       nrU[i] = &U[i * N];
     327             :     }
     328           0 :     for (i = 0; i < N; i++) {
     329           0 :       nrV[i] = &V[i * N];
     330             :     }
     331             :   } else {
     332           0 :     if (nrU) aom_free(nrU);
     333           0 :     if (nrV) aom_free(nrV);
     334           0 :     return 1;
     335             :   }
     336             : 
     337             :   /* copy from given matx into nrU */
     338           0 :   for (i = 0; i < M; i++) {
     339           0 :     memcpy(&(nrU[i][0]), matx + N * i, N * sizeof(*matx));
     340             :   }
     341             : 
     342             :   /* HERE IT IS: do SVD */
     343           0 :   if (svdcmp(nrU, M, N, W, nrV)) {
     344           0 :     aom_free(nrU);
     345           0 :     aom_free(nrV);
     346           0 :     return 1;
     347             :   }
     348             : 
     349             :   /* aom_free Numerical Recipes arrays */
     350           0 :   aom_free(nrU);
     351           0 :   aom_free(nrV);
     352             : 
     353           0 :   return 0;
     354             : }

Generated by: LCOV version 1.13