Multi-threaded frame calculation for movie clip proxies

This commit implements multi-threaded calculation of frames
when building proxies. Both scaling and undistortion steps
are now threaded.

Frames and proxy resolution are still handled one-by-one,
saving files after every single step. So if HDD is not so
fast, this commit could have not so much benefit.

Internal changes:

- Added IMB_scaleImBuf_threaded which scales given image
  buffer in multiple threads and uses bilinear filtering.

- libmv's camera intrinsics now have SetThreads() method
  which is used to specify how many OpenMP threads to use
  for buffer distortion/undistortion.

  And yeah, this code is using OpenMP for threading.

- Reshuffled a bit libmv-capi calls and added function
  BKE_tracking_distortion_set_threads to specify number
  of threads used by intrinscis.

4 files changed, 51 insertions, 19 deletions

diff --git a/extern/libmv/libmv-capi.cpp b/extern/libmv/libmv-capi.cpp
index a85c3268b16..eccf6fb7cb6 100644
--- a/extern/libmv/libmv-capi.cpp
+++ b/extern/libmv/libmv-capi.cpp
@@ -821,6 +821,13 @@ struct libmv_CameraIntrinsics *libmv_ReconstructionExtractIntrinsics(struct libm
 	return (struct libmv_CameraIntrinsics *)&libmv_Reconstruction->intrinsics;
 }
 
+struct libmv_CameraIntrinsics *libmv_CameraIntrinsicsNewEmpty(void)
+{
+	libmv::CameraIntrinsics *camera_intrinsics = new libmv::CameraIntrinsics();
+
+	return (struct libmv_CameraIntrinsics *) camera_intrinsics;
+}
+
 struct libmv_CameraIntrinsics *libmv_CameraIntrinsicsNew(libmv_cameraIntrinsicsOptions *libmv_camera_intrinsics_options)
 {
 	libmv::CameraIntrinsics *camera_intrinsics = new libmv::CameraIntrinsics();
@@ -884,6 +891,13 @@ void libmv_CameraIntrinsicsUpdate(struct libmv_CameraIntrinsics *libmv_intrinsic
 	}
 }
 
+void libmv_CameraIntrinsicsSetThreads(struct libmv_CameraIntrinsics *libmv_intrinsics, int threads)
+{
+	libmv::CameraIntrinsics *camera_intrinsics = (libmv::CameraIntrinsics *) libmv_intrinsics;
+
+	camera_intrinsics->SetThreads(threads);
+}
+
 void libmv_CameraIntrinsicsExtract(struct libmv_CameraIntrinsics *libmv_intrinsics, double *focal_length,
 			double *principal_x, double *principal_y, double *k1, double *k2, double *k3, int *width, int *height)
 {
diff --git a/extern/libmv/libmv-capi.h b/extern/libmv/libmv-capi.h
index c90322c56c5..c3debe45fd1 100644
--- a/extern/libmv/libmv-capi.h
+++ b/extern/libmv/libmv-capi.h
@@ -132,6 +132,8 @@ void libmv_destroyFeatures(struct libmv_Features *libmv_features);
 /* camera intrinsics */
 struct libmv_CameraIntrinsics *libmv_ReconstructionExtractIntrinsics(struct libmv_Reconstruction *libmv_Reconstruction);
 
+struct libmv_CameraIntrinsics *libmv_CameraIntrinsicsNewEmpty(void);
+
 struct libmv_CameraIntrinsics *libmv_CameraIntrinsicsNew(libmv_cameraIntrinsicsOptions *libmv_camera_intrinsics_options);
 
 struct libmv_CameraIntrinsics *libmv_CameraIntrinsicsCopy(struct libmv_CameraIntrinsics *libmv_intrinsics);
@@ -141,6 +143,8 @@ void libmv_CameraIntrinsicsDestroy(struct libmv_CameraIntrinsics *libmv_intrinsi
 void libmv_CameraIntrinsicsUpdate(struct libmv_CameraIntrinsics *libmv_intrinsics,
                                   libmv_cameraIntrinsicsOptions *libmv_camera_intrinsics_options);
 
+void libmv_CameraIntrinsicsSetThreads(struct libmv_CameraIntrinsics *libmv_intrinsics, int threads);
+
 void libmv_CameraIntrinsicsExtract(struct libmv_CameraIntrinsics *libmv_intrinsics, double *focal_length,
 			double *principal_x, double *principal_y, double *k1, double *k2, double *k3, int *width, int *height);
 
diff --git a/extern/libmv/libmv/simple_pipeline/camera_intrinsics.cc b/extern/libmv/libmv/simple_pipeline/camera_intrinsics.cc
index 658f65c1367..543fddedd0c 100644
--- a/extern/libmv/libmv/simple_pipeline/camera_intrinsics.cc
+++ b/extern/libmv/libmv/simple_pipeline/camera_intrinsics.cc
@@ -62,7 +62,8 @@ CameraIntrinsics::CameraIntrinsics()
       p1_(0),
       p2_(0),
       distort_(0),
-      undistort_(0) {}
+      undistort_(0),
+      threads_(1) {}
 
 CameraIntrinsics::CameraIntrinsics(const CameraIntrinsics &from)
     : K_(from.K_),
@@ -72,7 +73,8 @@ CameraIntrinsics::CameraIntrinsics(const CameraIntrinsics &from)
       k2_(from.k2_),
       k3_(from.k3_),
       p1_(from.p1_),
-      p2_(from.p2_)
+      p2_(from.p2_),
+      threads_(from.threads_)
 {
   distort_ = copyGrid(from.distort_);
   undistort_ = copyGrid(from.undistort_);
@@ -120,6 +122,11 @@ void CameraIntrinsics::SetTangentialDistortion(double p1, double p2) {
   FreeLookupGrid();
 }
 
+void CameraIntrinsics::SetThreads(int threads)
+{
+	threads_ = threads;
+}
+
 void CameraIntrinsics::ApplyIntrinsics(double normalized_x,
                                        double normalized_y,
                                        double *image_x,
@@ -188,6 +195,7 @@ void CameraIntrinsics::ComputeLookupGrid(Grid* grid, int width, int height, doub
   double aspx = (double)w / image_width_;
   double aspy = (double)h / image_height_;
 
+  #pragma omp parallel for schedule(dynamic) num_threads(threads_) if (threads_ > 1 && height > 100)
   for (int y = 0; y < height; y++) {
     for (int x = 0; x < width; x++) {
       double src_x = (x - 0.5 * overscan * w) / aspx, src_y = (y - 0.5 * overscan * h) / aspy;
@@ -214,7 +222,8 @@ void CameraIntrinsics::ComputeLookupGrid(Grid* grid, int width, int height, doub
 // TODO(MatthiasF): cubic B-Spline image sampling, bilinear lookup
 template<typename T,int N>
 static void Warp(const Grid* grid, const T* src, T* dst,
-                 int width, int height) {
+                 int width, int height, int threads) {
+  #pragma omp parallel for schedule(dynamic) num_threads(threads) if (threads > 1 && height > 100)
   for (int y = 0; y < height; y++) {
     for (int x = 0; x < width; x++) {
       Offset offset = grid->offset[y*width+x];
@@ -306,37 +315,37 @@ void CameraIntrinsics::CheckUndistortLookupGrid(int width, int height, double ov
 
 void CameraIntrinsics::Distort(const float* src, float* dst, int width, int height, double overscan, int channels) {
   CheckDistortLookupGrid(width, height, overscan);
-       if(channels==1) Warp<float,1>(distort_,src,dst,width,height);
-  else if(channels==2) Warp<float,2>(distort_,src,dst,width,height);
-  else if(channels==3) Warp<float,3>(distort_,src,dst,width,height);
-  else if(channels==4) Warp<float,4>(distort_,src,dst,width,height);
+       if(channels==1) Warp<float,1>(distort_,src,dst,width,height,threads_);
+  else if(channels==2) Warp<float,2>(distort_,src,dst,width,height,threads_);
+  else if(channels==3) Warp<float,3>(distort_,src,dst,width,height,threads_);
+  else if(channels==4) Warp<float,4>(distort_,src,dst,width,height,threads_);
   //else assert("channels must be between 1 and 4");
 }
 
 void CameraIntrinsics::Distort(const unsigned char* src, unsigned char* dst, int width, int height, double overscan, int channels) {
   CheckDistortLookupGrid(width, height, overscan);
-       if(channels==1) Warp<unsigned char,1>(distort_,src,dst,width,height);
-  else if(channels==2) Warp<unsigned char,2>(distort_,src,dst,width,height);
-  else if(channels==3) Warp<unsigned char,3>(distort_,src,dst,width,height);
-  else if(channels==4) Warp<unsigned char,4>(distort_,src,dst,width,height);
+       if(channels==1) Warp<unsigned char,1>(distort_,src,dst,width,height,threads_);
+  else if(channels==2) Warp<unsigned char,2>(distort_,src,dst,width,height,threads_);
+  else if(channels==3) Warp<unsigned char,3>(distort_,src,dst,width,height,threads_);
+  else if(channels==4) Warp<unsigned char,4>(distort_,src,dst,width,height,threads_);
   //else assert("channels must be between 1 and 4");
 }
 
 void CameraIntrinsics::Undistort(const float* src, float* dst, int width, int height, double overscan, int channels) {
   CheckUndistortLookupGrid(width, height, overscan);
-       if(channels==1) Warp<float,1>(undistort_,src,dst,width,height);
-  else if(channels==2) Warp<float,2>(undistort_,src,dst,width,height);
-  else if(channels==3) Warp<float,3>(undistort_,src,dst,width,height);
-  else if(channels==4) Warp<float,4>(undistort_,src,dst,width,height);
+       if(channels==1) Warp<float,1>(undistort_,src,dst,width,height,threads_);
+  else if(channels==2) Warp<float,2>(undistort_,src,dst,width,height,threads_);
+  else if(channels==3) Warp<float,3>(undistort_,src,dst,width,height,threads_);
+  else if(channels==4) Warp<float,4>(undistort_,src,dst,width,height,threads_);
   //else assert("channels must be between 1 and 4");
 }
 
 void CameraIntrinsics::Undistort(const unsigned char* src, unsigned char* dst, int width, int height, double overscan, int channels) {
   CheckUndistortLookupGrid(width, height, overscan);
-       if(channels==1) Warp<unsigned char,1>(undistort_,src,dst,width,height);
-  else if(channels==2) Warp<unsigned char,2>(undistort_,src,dst,width,height);
-  else if(channels==3) Warp<unsigned char,3>(undistort_,src,dst,width,height);
-  else if(channels==4) Warp<unsigned char,4>(undistort_,src,dst,width,height);
+       if(channels==1) Warp<unsigned char,1>(undistort_,src,dst,width,height,threads_);
+  else if(channels==2) Warp<unsigned char,2>(undistort_,src,dst,width,height,threads_);
+  else if(channels==3) Warp<unsigned char,3>(undistort_,src,dst,width,height,threads_);
+  else if(channels==4) Warp<unsigned char,4>(undistort_,src,dst,width,height,threads_);
   //else assert("channels must be between 1 and 4");
 }
 
diff --git a/extern/libmv/libmv/simple_pipeline/camera_intrinsics.h b/extern/libmv/libmv/simple_pipeline/camera_intrinsics.h
index b51b28a4bfb..632922c38ff 100644
--- a/extern/libmv/libmv/simple_pipeline/camera_intrinsics.h
+++ b/extern/libmv/libmv/simple_pipeline/camera_intrinsics.h
@@ -68,6 +68,9 @@ class CameraIntrinsics {
 
   void SetTangentialDistortion(double p1, double p2);
 
+  /// Set number of threads using for buffer distortion/undistortion
+  void SetThreads(int threads);
+
   /*!
       Apply camera intrinsics to the normalized point to get image coordinates.
 
@@ -153,6 +156,8 @@ class CameraIntrinsics {
 
   struct Grid *distort_;
   struct Grid *undistort_;
+
+  int threads_;
 };
 
 /// A human-readable representation of the camera intrinsic parameters.