Make KCF member variables constexpr where possible

[hercules2020/kcf.git] / src / dynmem.hpp

diff --git a/src/dynmem.hpp b/src/dynmem.hpp
index 5a32bd6360148463caca6e34c6873bf2bea5d717..f45609419b4f4883960e3b8496f90a64e037915f 100644 (file)
--- a/src/dynmem.hpp
+++ b/src/dynmem.hpp
@@ -4,11 +4,12 @@
 #include <cstdlib>
 #include <opencv2/opencv.hpp>
 #include <cassert>
+#include <numeric>
 
 #if defined(CUFFT) || defined(CUFFTW)
 #include "cuda_runtime.h"
 #ifdef CUFFT
-#include "cuda/cuda_error_check.cuh"
+#include "cuda_error_check.hpp"
 #endif
 #endif
 
@@ -18,19 +19,25 @@ template <typename T> class DynMem_ {
 #ifdef CUFFT
     T *ptr_d = nullptr;
 #endif
-    size_t size;
   public:
-    typedef T type;
-    DynMem_(size_t size) : size(size)
+    typedef T value_type;
+    const size_t num_elem;
+
+    DynMem_(size_t num_elem) : num_elem(num_elem)
     {
 #ifdef CUFFT
-        CudaSafeCall(cudaHostAlloc(reinterpret_cast<void **>(&ptr_h), size, cudaHostAllocMapped));
+        CudaSafeCall(cudaHostAlloc(reinterpret_cast<void **>(&ptr_h), num_elem * sizeof(T), cudaHostAllocMapped));
         CudaSafeCall(cudaHostGetDevicePointer(reinterpret_cast<void **>(&ptr_d), reinterpret_cast<void *>(ptr_h), 0));
 #else
-        ptr_h = new float[size];
+        ptr_h = new T[num_elem];
 #endif
     }
-    DynMem_(DynMem_&& other) {
+    DynMem_(const DynMem_ &other) : DynMem_(other.num_elem)
+    {
+        memcpy(ptr_h, other.ptr_h, num_elem * sizeof(T));
+    }
+    DynMem_(DynMem_ &&other) : num_elem(other.num_elem)
+    {
         ptr_h = other.ptr_h;
         other.ptr_h = nullptr;
 #ifdef CUFFT
@@ -40,26 +47,37 @@ template <typename T> class DynMem_ {
     }
     ~DynMem_()
     {
-#ifdef CUFFT
-        CudaSafeCall(cudaFreeHost(ptr_h));
-#else
-        delete[] ptr_h;
-#endif
+        release();
     }
     T *hostMem() { return ptr_h; }
+    const T *hostMem() const { return ptr_h; }
 #ifdef CUFFT
     T *deviceMem() { return ptr_d; }
+    const T *deviceMem() const { return ptr_d; }
 #endif
     void operator=(DynMem_ &rhs) {
-        memcpy(ptr_h, rhs.ptr_h, size * sizeof(T));
+        assert(num_elem == rhs.num_elem);
+        memcpy(ptr_h, rhs.ptr_h, num_elem * sizeof(T));
     }
     void operator=(DynMem_ &&rhs)
     {
+        assert(num_elem == rhs.num_elem);
+        release();
         ptr_h = rhs.ptr_h;
         rhs.ptr_h = nullptr;
 #ifdef CUFFT
         ptr_d = rhs.ptr_d;
         rhs.ptr_d = nullptr;
+#endif
+    }
+    T operator[](uint i) const { return ptr_h[i]; }
+private:
+    void release()
+    {
+#ifdef CUFFT
+        CudaSafeCall(cudaFreeHost(ptr_h));
+#else
+        delete[] ptr_h;
 #endif
     }
 };
@@ -70,20 +88,27 @@ typedef DynMem_<float> DynMem;
 class MatDynMem : public DynMem, public cv::Mat {
   public:
     MatDynMem(cv::Size size, int type)
-        : DynMem(size.area() * sizeof(DynMem::type) * CV_MAT_CN(type)), cv::Mat(size, type, hostMem())
+        : DynMem(size.area() * CV_MAT_CN(type)), cv::Mat(size, type, hostMem())
     {
         assert((type & CV_MAT_DEPTH_MASK) == CV_32F);
     }
     MatDynMem(int height, int width, int type)
-        : DynMem(width * height * sizeof(DynMem::type) * CV_MAT_CN(type)), cv::Mat(height, width, type, hostMem())
+        : DynMem(width * height * CV_MAT_CN(type)), cv::Mat(height, width, type, hostMem())
     {
         assert((type & CV_MAT_DEPTH_MASK) == CV_32F);
     }
     MatDynMem(int ndims, const int *sizes, int type)
-        : DynMem(volume(ndims, sizes) * sizeof(DynMem::type) * CV_MAT_CN(type)), cv::Mat(ndims, sizes, type, hostMem())
+        : DynMem(volume(ndims, sizes) * CV_MAT_CN(type)), cv::Mat(ndims, sizes, type, hostMem())
     {
         assert((type & CV_MAT_DEPTH_MASK) == CV_32F);
     }
+    MatDynMem(std::vector<int> size, int type)
+        : DynMem(std::accumulate(size.begin(), size.end(), 1, std::multiplies<int>()))
+        , cv::Mat(size.size(), size.data(), type, hostMem()) {}
+    MatDynMem(MatDynMem &&other) = default;
+    MatDynMem(const cv::Mat &other)
+        : DynMem(other.total()) , cv::Mat(other) {}
+
     void operator=(const cv::MatExpr &expr) {
         static_cast<cv::Mat>(*this) = expr;
     }
@@ -96,6 +121,56 @@ class MatDynMem : public DynMem, public cv::Mat {
             vol *= sizes[i];
         return vol;
     }
+
+    using cv::Mat::create;
+};
+
+class Mat3d : public MatDynMem
+{
+public:
+    Mat3d(uint dim0, cv::Size size) : MatDynMem({{int(dim0), size.height, size.width}}, CV_32F) {}
+
+    cv::Mat plane(uint idx) {
+        assert(dims == 3);
+        assert(int(idx) < size[0]);
+        return cv::Mat(size[1], size[2], cv::Mat::type(), ptr(idx));
+    }
+    const cv::Mat plane(uint idx) const {
+        assert(dims == 3);
+        assert(int(idx) < size[0]);
+        return cv::Mat(size[1], size[2], cv::Mat::type(), const_cast<uchar*>(ptr(idx)));
+    }
+
+};
+
+class MatFeats : public Mat3d
+{
+public:
+    MatFeats(uint num_features, cv::Size size) : Mat3d(num_features, size) {}
+};
+class MatScales : public Mat3d
+{
+public:
+    MatScales(uint num_scales, cv::Size size) : Mat3d(num_scales, size) {}
+};
+
+class MatScaleFeats : public MatDynMem
+{
+public:
+    MatScaleFeats(uint num_scales, uint num_features, cv::Size size)
+        : MatDynMem({{int(num_scales), int(num_features), size.height, size.width}}, CV_32F) {}
+
+    cv::Mat plane(uint scale, uint feature) {
+        assert(dims == 4);
+        assert(int(scale) < size[0]);
+        assert(int(feature) < size[1]);
+        return cv::Mat(size[2], size[3], cv::Mat::type(), ptr(scale, feature));
+    }
+    cv::Mat scale(uint scale) {
+        assert(dims == 4);
+        assert(int(scale) < size[0]);
+        return cv::Mat(3, std::vector<int>({size[1], size[2], size[3]}).data(), cv::Mat::type(), ptr(scale));
+    }
 };
 
 #endif // DYNMEM_HPP