Add FastCV DSP Initialization, QcAllocator and FastCV DSP Extension APIs

modules/fastcv/include/opencv2/fastcv/blur_dsp.hpp

@@ -24,7 +24,7 @@ namespace dsp {
  *
  * @sa Filter2D
  */
-CV_EXPORTS_W void filter2D(InputArray _src, OutputArray _dst, int ddepth, InputArray _kernel);
+CV_EXPORTS void filter2D(InputArray _src, OutputArray _dst, int ddepth, InputArray _kernel);
 
 //! @}
 } // dsp::

modules/fastcv/include/opencv2/fastcv/dsp_init.hpp

@@ -27,7 +27,7 @@ namespace dsp {
  * 
  * @return int Returns 0 on success, and a non-zero value on failure.
  */
-CV_EXPORTS_W int fcvdspinit();
+CV_EXPORTS int fcvdspinit();
 
 /**
  * @brief Deinitializes the FastCV DSP environment.
@@ -39,7 +39,7 @@ CV_EXPORTS_W int fcvdspinit();
  * @note This function must be called at the end of the use case or program, after all DSP-related 
  *       operations are complete.
  */
-CV_EXPORTS_W void fcvdspdeinit();
+CV_EXPORTS void fcvdspdeinit();
 //! @}
 
 } // dsp::

modules/fastcv/include/opencv2/fastcv/edges_dsp.hpp

@@ -28,7 +28,7 @@ namespace dsp {
  * @param apertureSize  The Sobel kernel size for calculating gradient. Supported sizes are 3, 5 and 7.
  * @param L2gradient    L2 Gradient or L1 Gradient
 */
-CV_EXPORTS_W void canny(InputArray _src, OutputArray _dst, int lowThreshold, int highThreshold, int apertureSize = 3, bool L2gradient = false);
+CV_EXPORTS void Canny(InputArray _src, OutputArray _dst, int lowThreshold, int highThreshold, int apertureSize = 3, bool L2gradient = false);
 //! @}
 
 } // dsp::

modules/fastcv/include/opencv2/fastcv/fft_dsp.hpp

@@ -25,7 +25,7 @@ namespace dsp {
 * @param dst The computed FFT matrix of type CV_32FC2. The FFT Re and Im coefficients are stored in different channels.
             Hence the dimensions of the dst are (srcWidth, srcHeight)
 */
-CV_EXPORTS_W void FFT(InputArray src, OutputArray dst);
+CV_EXPORTS void FFT(InputArray src, OutputArray dst);
 
 /**
 * @brief Computes the 1D or 2D Inverse Fast Fourier Transform of a complex valued matrix.
@@ -38,7 +38,7 @@ CV_EXPORTS_W void FFT(InputArray src, OutputArray dst);
 * @param dst The computed IFFT matrix of type CV_8U. The matrix is real valued and has no imaginary components.
             Hence the dimensions of the dst are (srcWidth , srcHeight)
 */
-CV_EXPORTS_W void IFFT(InputArray src, OutputArray dst);
+CV_EXPORTS void IFFT(InputArray src, OutputArray dst);
 
 //! @}
 

modules/fastcv/include/opencv2/fastcv/sad_dsp.hpp

@@ -24,7 +24,7 @@ namespace dsp {
  * @param _src The input image data, type CV_8UC1
  * @param _dst The output image data, type CV_16UC1
 */
-CV_EXPORTS_W void sumOfAbsoluteDiffs(cv::InputArray _patch, cv::InputArray _src, cv::OutputArray _dst);
+CV_EXPORTS void sumOfAbsoluteDiffs(cv::InputArray _patch, cv::InputArray _src, cv::OutputArray _dst);
 //! @}
 
 } // dsp::

modules/fastcv/include/opencv2/fastcv/thresh_dsp.hpp

@@ -12,26 +12,26 @@ namespace cv {
 namespace fastcv {
 namespace dsp {
 
-    //! @addtogroup fastcv
-    //! @{
+//! @addtogroup fastcv
+//! @{
 
-    /**
-     * @brief Binarizes a grayscale image using Otsu's method.
-     *        Sets the pixel to max(255) if it's value is greater than the threshold;
-     *        else, set the pixel to min(0). The threshold is searched that minimizes
-     *        the intra-class variance (the variance within the class).
-     * 
-     * @param _src Input 8-bit grayscale image. Size of buffer is srcStride*srcHeight bytes.
-     * @param _dst Output 8-bit binarized image. Size of buffer is dstStride*srcHeight bytes.
-     * @param type Threshold type that can be either 0 or 1.
-     *             NOTE: For threshold type=0, the pixel is set as
-     *             maxValue if it's value is greater than the threshold; else, it is set as zero.
-     *             For threshold type=1, the pixel is set as zero if it's
-     *             value is greater than the threshold; else, it is set as maxValue.
-     */
-    CV_EXPORTS_W void thresholdOtsu(InputArray _src, OutputArray _dst, bool type);
+/**
+ * @brief Binarizes a grayscale image using Otsu's method.
+ *        Sets the pixel to max(255) if it's value is greater than the threshold;
+ *        else, set the pixel to min(0). The threshold is searched that minimizes
+ *        the intra-class variance (the variance within the class).
+ * 
+ * @param _src Input 8-bit grayscale image. Size of buffer is srcStride*srcHeight bytes.
+ * @param _dst Output 8-bit binarized image. Size of buffer is dstStride*srcHeight bytes.
+ * @param type Threshold type that can be either 0 or 1.
+ *             NOTE: For threshold type=0, the pixel is set as
+ *             maxValue if it's value is greater than the threshold; else, it is set as zero.
+ *             For threshold type=1, the pixel is set as zero if it's
+ *             value is greater than the threshold; else, it is set as maxValue.
+ */
+CV_EXPORTS void thresholdOtsu(InputArray _src, OutputArray _dst, bool type);
 
-    //! @}
+//! @}
 } // dsp::
 } // fastcv::
 } // cv::

modules/fastcv/perf/perf_edges_dsp.cpp

@@ -41,7 +41,7 @@ PERF_TEST_P(CannyPerfTest, run,
     while (next())
     {
         startTimer();
-        cv::fastcv::dsp::canny(src, dst, lowThreshold, highThreshold, apertureSize, L2gradient);
+        cv::fastcv::dsp::Canny(src, dst, lowThreshold, highThreshold, apertureSize, L2gradient);
         stopTimer();
     }
 

modules/fastcv/src/allocator.cpp

@@ -61,8 +61,6 @@ cv::UMatData* QcAllocator::allocate(int dims, const int* sizes, int type,
     u->size = total;
     if(data0)
         u->flags |= cv::UMatData::USER_ALLOCATED;
-
-    u->userdata = new std::string("QCOM");
 
     // Add to active allocations
     cv::fastcv::QcResourceManager::getInstance().addAllocation(data);
@@ -94,12 +92,6 @@ void QcAllocator::deallocate(cv::UMatData* u) const
         u->origdata = 0;
     }
 
-    if (u->userdata)
-    {
-        delete static_cast<std::string*>(u->userdata);
-        u->userdata = nullptr;
-    }
-
     delete u;
 }
 

modules/fastcv/src/blur_dsp.cpp

@@ -9,75 +9,6 @@ namespace cv {
 namespace fastcv {
 namespace dsp {
 
-class FcvFilter2DLoop_Invoker : public ParallelLoopBody
-{
-    public:
-
-    FcvFilter2DLoop_Invoker(const Mat& _src, Mat& _dst, const Mat& _kernel) :
-        ParallelLoopBody(), src(_src), dst(_dst), kernel(_kernel)
-    {
-        width     = src.cols;
-        height    = src.rows;
-        ksize     = kernel.size().width;
-        halfKsize = ksize/2;
-    }
-
-    virtual void operator()(const Range& range) const CV_OVERRIDE
-    {
-        int topLines     = 0;
-        int rangeHeight  = range.end-range.start;
-        int paddedHeight = rangeHeight;
-
-        if(range.start >= halfKsize)
-        {
-            topLines    += halfKsize;
-            paddedHeight += halfKsize;
-        }
-
-        if(range.end <= height-halfKsize)
-        {
-            paddedHeight += halfKsize;
-        }
-
-        Mat srcPadded, dstPadded;
-        srcPadded.allocator = cv::fastcv::getQcAllocator();
-        dstPadded.allocator = cv::fastcv::getQcAllocator();
-
-        srcPadded = src(Rect(0, range.start - topLines, width, paddedHeight)); 
-        dstPadded.create(paddedHeight, width, dst.depth());
-
-        CV_Assert(IS_FASTCV_ALLOCATED(srcPadded));
-        CV_Assert(IS_FASTCV_ALLOCATED(dstPadded));
-
-        if (dst.depth() == CV_8U)
-            fcvFilterCorrNxNu8Q((int8_t*)kernel.data, ksize, 0, srcPadded.data, width, paddedHeight, srcPadded.step,
-                dstPadded.data, dstPadded.step);
-        else if (dst.depth() == CV_16S)
-            fcvFilterCorrNxNu8s16Q((int8_t*)kernel.data, ksize, 0, srcPadded.data, width, paddedHeight, srcPadded.step,
-                (int16_t*)dstPadded.data, dstPadded.step);
-        else if (dst.depth() == CV_32F)
-            fcvFilterCorrNxNu8f32Q((float32_t*)kernel.data, ksize, srcPadded.data, width, paddedHeight, srcPadded.step,
-                (float32_t*)dstPadded.data, dstPadded.step);
-
-        // Only copy center part back to output image and ignore the padded lines
-        Mat temp1 = dstPadded(Rect(0, topLines, width, rangeHeight));
-        Mat temp2 = dst(Rect(0, range.start, width, rangeHeight));
-        temp1.copyTo(temp2);
-    }
-
-    private:
-    const Mat&  src;
-    Mat&        dst;
-    const Mat&  kernel;
-    int         width;
-    int         height;
-    int         ksize;
-    int         halfKsize;
-
-    FcvFilter2DLoop_Invoker(const FcvFilter2DLoop_Invoker &);  // = delete;
-    const FcvFilter2DLoop_Invoker& operator= (const FcvFilter2DLoop_Invoker &);  // = delete;
-};
-
 void filter2D(InputArray _src, OutputArray _dst, int ddepth, InputArray _kernel)
 {
     CV_Assert(
@@ -103,25 +34,25 @@ void filter2D(InputArray _src, OutputArray _dst, int ddepth, InputArray _kernel)
     // Check DSP initialization status and initialize if needed
     FASTCV_CHECK_DSP_INIT();
 
-    int nThreads = getNumThreads();
-    int nStripes = (nThreads > 1) ? ((src.rows > 60) ? 3 * nThreads : 1) : 1;
-
-    if (ddepth == CV_8U && ksize.width == 3)
-        fcvFilterCorr3x3s8_v2Q((int8_t*)kernel.data, src.data, src.cols, src.rows, src.step, dst.data, dst.step);
-
     switch (ddepth)
     {
         case CV_8U:
+        {
+            if(ksize.width == 3)
+                fcvFilterCorr3x3s8_v2Q((int8_t*)kernel.data, src.data, src.cols, src.rows, src.step, dst.data, dst.step);
+            else
+                fcvFilterCorrNxNu8Q((int8_t*)kernel.data, ksize.width, 0, src.data, src.cols, src.rows, src.step, dst.data, dst.step);
+
+            break;
+        }
         case CV_16S:
         {
-            CV_Assert(CV_MAT_DEPTH(kernel.type()) == CV_8S);
-            parallel_for_(Range(0, src.rows), FcvFilter2DLoop_Invoker(src, dst, kernel), nStripes);
+            fcvFilterCorrNxNu8s16Q((int8_t*)kernel.data, ksize.width, 0, src.data, src.cols, src.rows, src.step, (int16_t*)dst.data, dst.step);
             break;
         }
         case CV_32F:
         {
-            CV_Assert(CV_MAT_DEPTH(kernel.type()) == CV_32F);
-            parallel_for_(Range(0, src.rows), FcvFilter2DLoop_Invoker(src, dst, kernel), nStripes);
+            fcvFilterCorrNxNu8f32Q((float32_t*)kernel.data, ksize.width, src.data, src.cols, src.rows, src.step, (float32_t*)dst.data, dst.step);
             break;
         }
         default:
@@ -131,6 +62,7 @@ void filter2D(InputArray _src, OutputArray _dst, int ddepth, InputArray _kernel)
             break;
         }
     }
+
 }
 
 } // dsp::

modules/fastcv/src/edges_dsp.cpp

@@ -9,7 +9,7 @@ namespace cv {
 namespace fastcv {
 namespace dsp {
 
-void canny(InputArray _src, OutputArray _dst, int lowThreshold, int highThreshold, int apertureSize, bool L2gradient)
+void Canny(InputArray _src, OutputArray _dst, int lowThreshold, int highThreshold, int apertureSize, bool L2gradient)
 {
     CV_Assert(
         !_src.empty() && 

modules/fastcv/src/precomp.hpp

@@ -79,8 +79,7 @@ namespace dsp {
     struct FastCvDspContext;
 
     #define IS_FASTCV_ALLOCATED(mat) \
-    ((mat.u && mat.u->userdata && \
-        *static_cast<std::string*>(mat.u->userdata) == "QCOM") ? true : \
+    ((mat.allocator == cv::fastcv::getQcAllocator()) ? true : \
         (CV_Error(cv::Error::StsBadArg, cv::format("Matrix '%s' not allocated with FastCV allocator. " \
                                     "Please ensure that the matrix is created using " \
                                     "cv::fastcv::getQcAllocator().", #mat)), false))

modules/fastcv/test/test_blur_dsp.cpp

@@ -65,7 +65,7 @@ TEST_P(Filter2DTest_DSP, accuracy)
 }
 
 INSTANTIATE_TEST_CASE_P(FastCV_Extension, Filter2DTest_DSP, Combine(
-/*image sie*/      Values(perf::szVGA, perf::sz720p, perf::sz1080p),
+/*image sie*/      Values(perf::szVGA, perf::sz720p),
 /*dst depth*/      Values(CV_8U,CV_16S,CV_32F),
 /*kernel size*/    Values(3, 5, 7, 9, 11)
 ));

modules/fastcv/test/test_edges_dsp.cpp

@@ -24,7 +24,7 @@ TEST(DSP_CannyTest, accuracy)
     int lowThreshold = 0;
     int highThreshold = 150;
 
-    cv::fastcv::dsp::canny(src, dst, lowThreshold, highThreshold, 3, true);
+    cv::fastcv::dsp::Canny(src, dst, lowThreshold, highThreshold, 3, true);
 
     //De-Initialize DSP
     cv::fastcv::dsp::fcvdspdeinit();