fix CUDA LUT implementation

In CUDA 6.0 there was a bug in NPP LUT implementation (invalid results when
src == 255). In CUDA 6.5 the bug was fixed.

Replaced NPP LUT call with own implementation (ported from master branch)
to be independant from CUDA Toolkit version.
(cherry picked from commit eaaa2d27)

modules/gpu/src/arithm.cpp

@@ -317,6 +317,11 @@ void cv::gpu::flip(const GpuMat& src, GpuMat& dst, int flipCode, Stream& stream)
 ////////////////////////////////////////////////////////////////////////
 // LUT
 
+namespace arithm
+{
+    void lut(PtrStepSzb src, uchar* lut, int lut_cn, PtrStepSzb dst, bool cc30, cudaStream_t stream);
+}
+
 void cv::gpu::LUT(const GpuMat& src, const Mat& lut, GpuMat& dst, Stream& s)
 {
     const int cn = src.channels();
@@ -328,82 +333,21 @@ void cv::gpu::LUT(const GpuMat& src, const Mat& lut, GpuMat& dst, Stream& s)
 
     dst.create(src.size(), CV_MAKE_TYPE(lut.depth(), cn));
 
-    NppiSize sz;
-    sz.height = src.rows;
-    sz.width = src.cols;
-
-    Mat nppLut;
-    lut.convertTo(nppLut, CV_32S);
-
-    int nValues3[] = {256, 256, 256};
-
-    Npp32s pLevels[256];
-    for (int i = 0; i < 256; ++i)
-        pLevels[i] = i;
-
-    const Npp32s* pLevels3[3];
-
-#if (CUDA_VERSION <= 4020)
-    pLevels3[0] = pLevels3[1] = pLevels3[2] = pLevels;
-#else
-    GpuMat d_pLevels;
-    d_pLevels.upload(Mat(1, 256, CV_32S, pLevels));
-    pLevels3[0] = pLevels3[1] = pLevels3[2] = d_pLevels.ptr<Npp32s>();
-#endif
+    GpuMat d_lut;
+    d_lut.upload(Mat(1, 256, lut.type(), lut.data));
 
+    int lut_cn = d_lut.channels();
+    bool cc30 = deviceSupports(FEATURE_SET_COMPUTE_30);
     cudaStream_t stream = StreamAccessor::getStream(s);
-    NppStreamHandler h(stream);
 
-    if (src.type() == CV_8UC1)
+    if (lut_cn == 1)
     {
-#if (CUDA_VERSION <= 4020)
-        nppSafeCall( nppiLUT_Linear_8u_C1R(src.ptr<Npp8u>(), static_cast<int>(src.step),
-            dst.ptr<Npp8u>(), static_cast<int>(dst.step), sz, nppLut.ptr<Npp32s>(), pLevels, 256) );
-#else
-        GpuMat d_nppLut(Mat(1, 256, CV_32S, nppLut.data));
-        nppSafeCall( nppiLUT_Linear_8u_C1R(src.ptr<Npp8u>(), static_cast<int>(src.step),
-            dst.ptr<Npp8u>(), static_cast<int>(dst.step), sz, d_nppLut.ptr<Npp32s>(), d_pLevels.ptr<Npp32s>(), 256) );
-#endif
+        arithm::lut(src.reshape(1), d_lut.data, lut_cn, dst.reshape(1), cc30, stream);
     }
-    else
+    else if (lut_cn == 3)
     {
-        const Npp32s* pValues3[3];
-
-        Mat nppLut3[3];
-        if (nppLut.channels() == 1)
-        {
-#if (CUDA_VERSION <= 4020)
-            pValues3[0] = pValues3[1] = pValues3[2] = nppLut.ptr<Npp32s>();
-#else
-            GpuMat d_nppLut(Mat(1, 256, CV_32S, nppLut.data));
-            pValues3[0] = pValues3[1] = pValues3[2] = d_nppLut.ptr<Npp32s>();
-#endif
-        }
-        else
-        {
-            cv::split(nppLut, nppLut3);
-
-#if (CUDA_VERSION <= 4020)
-            pValues3[0] = nppLut3[0].ptr<Npp32s>();
-            pValues3[1] = nppLut3[1].ptr<Npp32s>();
-            pValues3[2] = nppLut3[2].ptr<Npp32s>();
-#else
-            GpuMat d_nppLut0(Mat(1, 256, CV_32S, nppLut3[0].data));
-            GpuMat d_nppLut1(Mat(1, 256, CV_32S, nppLut3[1].data));
-            GpuMat d_nppLut2(Mat(1, 256, CV_32S, nppLut3[2].data));
-
-            pValues3[0] = d_nppLut0.ptr<Npp32s>();
-            pValues3[1] = d_nppLut1.ptr<Npp32s>();
-            pValues3[2] = d_nppLut2.ptr<Npp32s>();
-#endif
-        }
-
-        nppSafeCall( nppiLUT_Linear_8u_C3R(src.ptr<Npp8u>(), static_cast<int>(src.step),
-            dst.ptr<Npp8u>(), static_cast<int>(dst.step), sz, pValues3, pLevels3, nValues3) );
+        arithm::lut(src, d_lut.data, lut_cn, dst, cc30, stream);
     }
-
-    if (stream == 0)
-        cudaSafeCall( cudaDeviceSynchronize() );
 }
 
 ////////////////////////////////////////////////////////////////////////

modules/gpu/src/cuda/lut.cu

+/*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) 2000-2008, Intel Corporation, all rights reserved.
+// Copyright (C) 2009, Willow Garage 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*/
+
+#if !defined CUDA_DISABLER
+
+#include <cstring>
+#include "opencv2/gpu/device/common.hpp"
+#include "opencv2/gpu/device/transform.hpp"
+#include "opencv2/gpu/device/functional.hpp"
+
+using namespace cv::gpu;
+using namespace cv::gpu::device;
+
+namespace
+{
+    texture<uchar, cudaTextureType1D, cudaReadModeElementType> texLutTable;
+
+    struct LutC1 : public unary_function<uchar, uchar>
+    {
+        typedef uchar value_type;
+        typedef uchar index_type;
+
+        cudaTextureObject_t texLutTableObj;
+
+        __device__ __forceinline__ uchar operator ()(uchar x) const
+        {
+        #if defined(__CUDA_ARCH__) && (__CUDA_ARCH__ < 300)
+            // Use the texture reference
+            return tex1Dfetch(texLutTable, x);
+        #else
+            // Use the texture object
+            return tex1Dfetch<uchar>(texLutTableObj, x);
+        #endif
+        }
+    };
+    struct LutC3 : public unary_function<uchar3, uchar3>
+    {
+        typedef uchar3 value_type;
+        typedef uchar3 index_type;
+
+        cudaTextureObject_t texLutTableObj;
+
+        __device__ __forceinline__ uchar3 operator ()(const uchar3& x) const
+        {
+        #if defined(__CUDA_ARCH__) && (__CUDA_ARCH__ < 300)
+            // Use the texture reference
+            return make_uchar3(tex1Dfetch(texLutTable, x.x * 3), tex1Dfetch(texLutTable, x.y * 3 + 1), tex1Dfetch(texLutTable, x.z * 3 + 2));
+        #else
+            // Use the texture object
+            return make_uchar3(tex1Dfetch<uchar>(texLutTableObj, x.x * 3), tex1Dfetch<uchar>(texLutTableObj, x.y * 3 + 1), tex1Dfetch<uchar>(texLutTableObj, x.z * 3 + 2));
+        #endif
+        }
+    };
+}
+
+namespace arithm
+{
+    void lut(PtrStepSzb src, uchar* lut, int lut_cn, PtrStepSzb dst, bool cc30, cudaStream_t stream)
+    {
+        cudaTextureObject_t texLutTableObj;
+
+        if (cc30)
+        {
+            // Use the texture object
+            cudaResourceDesc texRes;
+            std::memset(&texRes, 0, sizeof(texRes));
+            texRes.resType = cudaResourceTypeLinear;
+            texRes.res.linear.devPtr = lut;
+            texRes.res.linear.desc = cudaCreateChannelDesc<uchar>();
+            texRes.res.linear.sizeInBytes = 256 * lut_cn * sizeof(uchar);
+
+            cudaTextureDesc texDescr;
+            std::memset(&texDescr, 0, sizeof(texDescr));
+
+            cudaSafeCall( cudaCreateTextureObject(&texLutTableObj, &texRes, &texDescr, 0) );
+        }
+        else
+        {
+            // Use the texture reference
+            cudaChannelFormatDesc desc = cudaCreateChannelDesc<uchar>();
+            cudaSafeCall( cudaBindTexture(0, &texLutTable, lut, &desc) );
+        }
+
+        if (lut_cn == 1)
+        {
+            LutC1 op;
+            op.texLutTableObj = texLutTableObj;
+
+            transform((PtrStepSz<uchar>) src, (PtrStepSz<uchar>) dst, op, WithOutMask(), stream);
+        }
+        else if (lut_cn == 3)
+        {
+            LutC3 op;
+            op.texLutTableObj = texLutTableObj;
+
+            transform((PtrStepSz<uchar3>) src, (PtrStepSz<uchar3>) dst, op, WithOutMask(), stream);
+        }
+
+        if (cc30)
+        {
+            // Use the texture object
+            cudaSafeCall( cudaDestroyTextureObject(texLutTableObj) );
+        }
+        else
+        {
+            // Use the texture reference
+            cudaSafeCall( cudaUnbindTexture(texLutTable) );
+        }
+    }
+}
+
+#endif