Merge pull request #1958 from vbystricky:ocl_filter2D

modules/imgproc/src/filter.cpp

@@ -41,6 +41,7 @@
 //M*/
 
 #include "precomp.hpp"
+#include "opencl_kernels.hpp"
 
 /****************************************************************************************\
                                     Base Image Filter
@@ -3115,6 +3116,206 @@ template<typename ST, class CastOp, class VecOp> struct Filter2D : public BaseFi
 
 }
 
+namespace cv
+{
+
+#define DIVUP(total, grain) (((total) + (grain) - 1) / (grain))
+#define ROUNDUP(sz, n)      ((sz) + (n) - 1 - (((sz) + (n) - 1) % (n)))
+
+// prepare kernel: transpose and make double rows (+align). Returns size of aligned row
+// Samples:
+//        a b c
+// Input: d e f
+//        g h i
+// Output, last two zeros is the alignment:
+// a d g a d g 0 0
+// b e h b e h 0 0
+// c f i c f i 0 0
+template <typename T>
+static int _prepareKernelFilter2D(std::vector<T>& data, const Mat &kernel)
+{
+    Mat _kernel; kernel.convertTo(_kernel, DataDepth<T>::value);
+    int size_y_aligned = ROUNDUP(kernel.rows * 2, 4);
+    data.clear(); data.resize(size_y_aligned * kernel.cols, 0);
+    for (int x = 0; x < kernel.cols; x++)
+    {
+        for (int y = 0; y < kernel.rows; y++)
+        {
+            data[x * size_y_aligned + y] = _kernel.at<T>(y, x);
+            data[x * size_y_aligned + y + kernel.rows] = _kernel.at<T>(y, x);
+        }
+    }
+    return size_y_aligned;
+}
+
+static bool ocl_filter2D( InputArray _src, OutputArray _dst, int ddepth,
+                   InputArray _kernel, Point anchor,
+                   double delta, int borderType )
+{
+    if (abs(delta) > FLT_MIN)
+        return false;
+
+    int type = _src.type();
+    int cn = CV_MAT_CN(type);
+    if ((1 != cn) && (2 != cn) && (4 != cn))
+        return false;//TODO
+
+    int sdepth = CV_MAT_DEPTH(type);
+    Size ksize = _kernel.size();
+    if( anchor.x < 0 )
+        anchor.x = ksize.width / 2;
+    if( anchor.y < 0 )
+        anchor.y = ksize.height / 2;
+    if( ddepth < 0 )
+        ddepth = sdepth;
+    else if (ddepth != sdepth)
+        return false;
+
+    bool isIsolatedBorder = (borderType & BORDER_ISOLATED) != 0;
+    bool useDouble = (CV_64F == sdepth);
+    const cv::ocl::Device &device = cv::ocl::Device::getDefault();
+    int doubleFPConfig = device.doubleFPConfig();
+    if (useDouble && (0 == doubleFPConfig))
+        return false;
+
+    const char* btype = NULL;
+    switch (borderType & ~BORDER_ISOLATED)
+    {
+    case BORDER_CONSTANT:
+        btype = "BORDER_CONSTANT";
+        break;
+    case BORDER_REPLICATE:
+        btype = "BORDER_REPLICATE";
+        break;
+    case BORDER_REFLECT:
+        btype = "BORDER_REFLECT";
+        break;
+    case BORDER_WRAP:
+        return false;
+    case BORDER_REFLECT101:
+        btype = "BORDER_REFLECT_101";
+        break;
+    }
+
+    cv::Mat kernelMat = _kernel.getMat();
+    std::vector<float> kernelMatDataFloat;
+    std::vector<double> kernelMatDataDouble;
+    int kernel_size_y2_aligned = useDouble ?
+            _prepareKernelFilter2D<double>(kernelMatDataDouble, kernelMat)
+            : _prepareKernelFilter2D<float>(kernelMatDataFloat, kernelMat);
+
+
+    cv::Size sz = _src.size();
+    size_t globalsize[2] = {sz.width, sz.height};
+    size_t localsize[2] = {0, 1};
+
+    ocl::Kernel kernel;
+    UMat src; Size wholeSize;
+    if (!isIsolatedBorder)
+    {
+        src = _src.getUMat();
+        Point ofs;
+        src.locateROI(wholeSize, ofs);
+    }
+
+    size_t maxWorkItemSizes[32]; device.maxWorkItemSizes(maxWorkItemSizes);
+    size_t tryWorkItems = maxWorkItemSizes[0];
+    for (;;)
+    {
+        size_t BLOCK_SIZE = tryWorkItems;
+        while (BLOCK_SIZE > 32 && BLOCK_SIZE >= (size_t)ksize.width * 2 && BLOCK_SIZE > (size_t)sz.width * 2)
+            BLOCK_SIZE /= 2;
+#if 1 // TODO Mode with several blocks requires a much more VGPRs, so this optimization is not actual for the current devices
+        size_t BLOCK_SIZE_Y = 1;
+#else
+        size_t BLOCK_SIZE_Y = 8; // TODO Check heuristic value on devices
+        while (BLOCK_SIZE_Y < BLOCK_SIZE / 8 && BLOCK_SIZE_Y * src.clCxt->getDeviceInfo().maxComputeUnits * 32 < (size_t)src.rows)
+            BLOCK_SIZE_Y *= 2;
+#endif
+
+        if ((size_t)ksize.width > BLOCK_SIZE)
+            return false;
+
+        int requiredTop = anchor.y;
+        int requiredLeft = (int)BLOCK_SIZE; // not this: anchor.x;
+        int requiredBottom = ksize.height - 1 - anchor.y;
+        int requiredRight = (int)BLOCK_SIZE; // not this: ksize.width - 1 - anchor.x;
+        int h = isIsolatedBorder ? sz.height : wholeSize.height;
+        int w = isIsolatedBorder ? sz.width : wholeSize.width;
+        bool extra_extrapolation = h < requiredTop || h < requiredBottom || w < requiredLeft || w < requiredRight;
+
+        if ((w < ksize.width) || (h < ksize.height))
+            return false;
+
+        char build_options[1024];
+        sprintf(build_options, "-D LOCAL_SIZE=%d -D BLOCK_SIZE_Y=%d -D DATA_DEPTH=%d -D DATA_CHAN=%d -D USE_DOUBLE=%d "
+                "-D ANCHOR_X=%d -D ANCHOR_Y=%d -D KERNEL_SIZE_X=%d -D KERNEL_SIZE_Y=%d -D KERNEL_SIZE_Y2_ALIGNED=%d "
+                "-D %s -D %s -D %s",
+                (int)BLOCK_SIZE, (int)BLOCK_SIZE_Y,
+                sdepth, cn, useDouble ? 1 : 0,
+                anchor.x, anchor.y, ksize.width, ksize.height, kernel_size_y2_aligned,
+                btype,
+                extra_extrapolation ? "EXTRA_EXTRAPOLATION" : "NO_EXTRA_EXTRAPOLATION",
+                isIsolatedBorder ? "BORDER_ISOLATED" : "NO_BORDER_ISOLATED");
+
+        localsize[0] = BLOCK_SIZE;
+        globalsize[0] = DIVUP(sz.width, BLOCK_SIZE - (ksize.width - 1)) * BLOCK_SIZE;
+        globalsize[1] = DIVUP(sz.height, BLOCK_SIZE_Y);
+
+        cv::String errmsg;
+        if (!kernel.create("filter2D", cv::ocl::imgproc::filter2D_oclsrc, build_options))
+            return false;
+        size_t kernelWorkGroupSize = kernel.workGroupSize();
+        if (localsize[0] <= kernelWorkGroupSize)
+            break;
+        if (BLOCK_SIZE < kernelWorkGroupSize)
+            return false;
+        tryWorkItems = kernelWorkGroupSize;
+    }
+
+    _dst.create(sz, CV_MAKETYPE(ddepth, cn));
+    UMat dst = _dst.getUMat();
+    if (src.empty())
+        src = _src.getUMat();
+
+    int idxArg = 0;
+    idxArg = kernel.set(idxArg, ocl::KernelArg::PtrReadOnly(src));
+    idxArg = kernel.set(idxArg, (int)src.step);
+
+    int srcOffsetX = (int)((src.offset % src.step) / src.elemSize());
+    int srcOffsetY = (int)(src.offset / src.step);
+    int srcEndX = (isIsolatedBorder ? (srcOffsetX + sz.width) : wholeSize.width);
+    int srcEndY = (isIsolatedBorder ? (srcOffsetY + sz.height) : wholeSize.height);
+    idxArg = kernel.set(idxArg, srcOffsetX);
+    idxArg = kernel.set(idxArg, srcOffsetY);
+    idxArg = kernel.set(idxArg, srcEndX);
+    idxArg = kernel.set(idxArg, srcEndY);
+
+    idxArg = kernel.set(idxArg, ocl::KernelArg::WriteOnly(dst));
+    float borderValue[4] = {0, 0, 0, 0};
+    double borderValueDouble[4] = {0, 0, 0, 0};
+    if ((borderType & ~BORDER_ISOLATED) == BORDER_CONSTANT)
+    {
+        int cnocl = (3 == cn) ? 4 : cn;
+        if (useDouble)
+            idxArg = kernel.set(idxArg, (void *)&borderValueDouble[0], sizeof(double) * cnocl);
+        else
+            idxArg = kernel.set(idxArg, (void *)&borderValue[0], sizeof(float) * cnocl);
+    }
+    if (useDouble)
+    {
+        UMat kernalDataUMat(kernelMatDataDouble, true);
+        idxArg = kernel.set(idxArg, ocl::KernelArg::PtrReadOnly(kernalDataUMat));
+    }
+    else
+    {
+        UMat kernalDataUMat(kernelMatDataFloat, true);
+        idxArg = kernel.set(idxArg, ocl::KernelArg::PtrReadOnly(kernalDataUMat));
+    }
+    return kernel.run(2, globalsize, localsize, true);
+}
+}
+
 cv::Ptr<cv::BaseFilter> cv::getLinearFilter(int srcType, int dstType,
                                 InputArray filter_kernel, Point anchor,
                                 double delta, int bits)
@@ -3230,6 +3431,10 @@ void cv::filter2D( InputArray _src, OutputArray _dst, int ddepth,
                    InputArray _kernel, Point anchor,
                    double delta, int borderType )
 {
+    bool use_opencl = ocl::useOpenCL() && _dst.isUMat();
+    if( use_opencl && ocl_filter2D(_src, _dst, ddepth, _kernel, anchor, delta, borderType))
+        return;
+
     Mat src = _src.getMat(), kernel = _kernel.getMat();
 
     if( ddepth < 0 )

modules/imgproc/test/ocl/test_filter2d.cpp

+/*M///////////////////////////////////////////////////////////////////////////////////////
+//
+//  IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
+//
+//  By downloading, copying, installing or using the software you agree to this license.
+//  If you do not agree to this license, do not download, install,
+//  copy or use the software.
+//
+//
+//                           License Agreement
+//                For Open Source Computer Vision Library
+//
+// Copyright (C) 2010-2012, Institute Of Software Chinese Academy Of Science, all rights reserved.
+// Copyright (C) 2010-2012, Advanced Micro Devices, Inc., all rights reserved.
+// Copyright (C) 2010-2012, Multicoreware, Inc., all rights reserved.
+// Third party copyrights are property of their respective owners.
+//
+// Redistribution and use in source and binary forms, with or without modification,
+// are permitted provided that the following conditions are met:
+//
+//   * Redistribution's of source code must retain the above copyright notice,
+//     this list of conditions and the following disclaimer.
+//
+//   * Redistribution's in binary form must reproduce the above copyright notice,
+//     this list of conditions and the following disclaimer in the documentation
+//     and/or other materials provided with the distribution.
+//
+//   * The name of the copyright holders may not be used to endorse or promote products
+//     derived from this software without specific prior written permission.
+//
+// This software is provided by the copyright holders and contributors "as is" and
+// any express or implied warranties, including, but not limited to, the implied
+// warranties of merchantability and fitness for a particular purpose are disclaimed.
+// In no event shall the Intel Corporation or contributors be liable for any direct,
+// indirect, incidental, special, exemplary, or consequential damages
+// (including, but not limited to, procurement of substitute goods or services;
+// loss of use, data, or profits; or business interruption) however caused
+// and on any theory of liability, whether in contract, strict liability,
+// or tort (including negligence or otherwise) arising in any way out of
+// the use of this software, even if advised of the possibility of such damage.
+//
+//M*/
+
+#include "test_precomp.hpp"
+#include "opencv2/ts/ocl_test.hpp"
+
+#ifdef HAVE_OPENCL
+
+namespace cvtest {
+namespace ocl {
+
+/////////////////////////////////////////////////////////////////////////////////////////////////
+// Filter2D
+PARAM_TEST_CASE(Filter2D, MatDepth, Channels, BorderType, bool, bool)
+{
+    static const int kernelMinSize = 2;
+    static const int kernelMaxSize = 10;
+
+    int type;
+    Size dsize;
+    Point anchor;
+    int borderType;
+    bool useRoi;
+    Mat kernel;
+
+    TEST_DECLARE_INPUT_PARAMETER(src)
+    TEST_DECLARE_OUTPUT_PARAMETER(dst)
+
+    virtual void SetUp()
+    {
+        type = CV_MAKE_TYPE(GET_PARAM(0), GET_PARAM(1));
+        borderType = GET_PARAM(2) | (GET_PARAM(3) ? BORDER_ISOLATED : 0);
+        useRoi = GET_PARAM(4);
+    }
+
+    void random_roi()
+    {
+        dsize = randomSize(1, MAX_VALUE);
+
+        Size ksize = randomSize(kernelMinSize, kernelMaxSize);
+        Mat temp = randomMat(ksize, CV_MAKE_TYPE(((CV_64F == CV_MAT_DEPTH(type)) ? CV_64F : CV_32F), 1), -MAX_VALUE, MAX_VALUE);
+        cv::normalize(temp, kernel, 1.0, 0.0, NORM_L1);
+
+        Size roiSize = randomSize(ksize.width, MAX_VALUE, ksize.height, MAX_VALUE);
+        Border srcBorder = randomBorder(0, useRoi ? MAX_VALUE : 0);
+        randomSubMat(src, src_roi, roiSize, srcBorder, type, -MAX_VALUE, MAX_VALUE);
+
+        Border dstBorder = randomBorder(0, useRoi ? MAX_VALUE : 0);
+        randomSubMat(dst, dst_roi, dsize, dstBorder, type, -MAX_VALUE, MAX_VALUE);
+
+        anchor.x = randomInt(-1, ksize.width);
+        anchor.y = randomInt(-1, ksize.height);
+
+        UMAT_UPLOAD_INPUT_PARAMETER(src)
+        UMAT_UPLOAD_OUTPUT_PARAMETER(dst)
+    }
+
+    void Near(double threshold = 0.0)
+    {
+        EXPECT_MAT_NEAR(dst, udst, threshold);
+        EXPECT_MAT_NEAR(dst_roi, udst_roi, threshold);
+    }
+};
+
+OCL_TEST_P(Filter2D, Mat)
+{
+    for (int j = 0; j < test_loop_times; j++)
+    {
+        random_roi();
+
+        OCL_OFF(cv::filter2D(src_roi, dst_roi, -1, kernel, anchor, 0.0, borderType));
+        OCL_ON(cv::filter2D(usrc_roi, udst_roi, -1, kernel, anchor, 0.0, borderType));
+
+        Near(1.0);
+    }
+}
+
+
+OCL_INSTANTIATE_TEST_CASE_P(ImageProc, Filter2D,
+                            Combine(
+                                Values(CV_8U, CV_16U, CV_16S, CV_32F, CV_64F),
+                                Values(1, 2, 4),
+                                Values((BorderType)BORDER_CONSTANT,
+                                       (BorderType)BORDER_REPLICATE,
+                                       (BorderType)BORDER_REFLECT,
+                                       (BorderType)BORDER_REFLECT_101),
+                                Bool(), // BORDER_ISOLATED
+                                Bool()  // ROI
+                                )
+                           );
+
+
+} } // namespace cvtest::ocl
+
+#endif // HAVE_OPENCL