From db25e44a17b1e1b1b756049c71af2334c79d9539 Mon Sep 17 00:00:00 2001
From: Vladislav Vinogradov <vlad.vinogradov@itseez.com>
Date: Wed, 4 Mar 2015 14:44:52 +0300
Subject: [PATCH] reduce brute_force_matcher instantiates for tiny build

---
 modules/gpu/src/brute_force_matcher.cpp | 97 +++++++++++++++++++++++++
 modules/gpu/src/cuda/bf_knnmatch.cu     | 12 +++
 modules/gpu/src/cuda/bf_match.cu        | 12 +++
 modules/gpu/src/cuda/bf_radius_match.cu | 12 +++
 4 files changed, 133 insertions(+)

diff --git a/modules/gpu/src/brute_force_matcher.cpp b/modules/gpu/src/brute_force_matcher.cpp
index 5da22e156b..e39bce2270 100644
--- a/modules/gpu/src/brute_force_matcher.cpp
+++ b/modules/gpu/src/brute_force_matcher.cpp
@@ -204,6 +204,26 @@ void cv::gpu::BruteForceMatcher_GPU_base::matchSingle(const GpuMat& query, const
                              const PtrStepSzi& trainIdx, const PtrStepSzf& distance,
                              cudaStream_t stream);
 
+#ifdef OPENCV_TINY_GPU_MODULE
+    static const caller_t callers[3][6] =
+    {
+        {
+            0/*matchL1_gpu<unsigned char>*/, 0/*matchL1_gpu<signed char>*/,
+            0/*matchL1_gpu<unsigned short>*/, 0/*matchL1_gpu<short>*/,
+            0/*matchL1_gpu<int>*/, 0/*matchL1_gpu<float>*/
+        },
+        {
+            0/*matchL2_gpu<unsigned char>*/, 0/*matchL2_gpu<signed char>*/,
+            0/*matchL2_gpu<unsigned short>*/, 0/*matchL2_gpu<short>*/,
+            0/*matchL2_gpu<int>*/, matchL2_gpu<float>
+        },
+        {
+            matchHamming_gpu<unsigned char>, 0/*matchHamming_gpu<signed char>*/,
+            0/*matchHamming_gpu<unsigned short>*/, 0/*matchHamming_gpu<short>*/,
+            0/*matchHamming_gpu<int>*/, 0/*matchHamming_gpu<float>*/
+        }
+    };
+#else
     static const caller_t callers[3][6] =
     {
         {
@@ -222,6 +242,7 @@ void cv::gpu::BruteForceMatcher_GPU_base::matchSingle(const GpuMat& query, const
             matchHamming_gpu<int>, 0/*matchHamming_gpu<float>*/
         }
     };
+#endif
 
     CV_Assert(query.channels() == 1 && query.depth() < CV_64F);
     CV_Assert(train.cols == query.cols && train.type() == query.type());
@@ -334,6 +355,16 @@ void cv::gpu::BruteForceMatcher_GPU_base::matchCollection(const GpuMat& query, c
     GpuMat& trainIdx, GpuMat& imgIdx, GpuMat& distance,
     const GpuMat& masks, Stream& stream)
 {
+#ifdef OPENCV_TINY_GPU_MODULE
+    (void)query;
+    (void)trainCollection;
+    (void)trainIdx;
+    (void)imgIdx;
+    (void)distance;
+    (void)masks;
+    (void)stream;
+    CV_Error(CV_StsNotImplemented, "not available in tiny build");
+#else
     if (query.empty() || trainCollection.empty())
         return;
 
@@ -374,6 +405,7 @@ void cv::gpu::BruteForceMatcher_GPU_base::matchCollection(const GpuMat& query, c
     CV_Assert(func != 0);
 
     func(query, trainCollection, masks, trainIdx, imgIdx, distance, StreamAccessor::getStream(stream));
+#endif
 }
 
 void cv::gpu::BruteForceMatcher_GPU_base::matchDownload(const GpuMat& trainIdx, const GpuMat& imgIdx, const GpuMat& distance, vector<DMatch>& matches)
@@ -451,6 +483,26 @@ void cv::gpu::BruteForceMatcher_GPU_base::knnMatchSingle(const GpuMat& query, co
                              const PtrStepSzb& trainIdx, const PtrStepSzb& distance, const PtrStepSzf& allDist,
                              cudaStream_t stream);
 
+#ifdef OPENCV_TINY_GPU_MODULE
+    static const caller_t callers[3][6] =
+    {
+        {
+            0/*matchL1_gpu<unsigned char>*/, 0/*matchL1_gpu<signed char>*/,
+            0/*matchL1_gpu<unsigned short>*/, 0/*matchL1_gpu<short>*/,
+            0/*matchL1_gpu<int>*/, 0/*matchL1_gpu<float>*/
+        },
+        {
+            0/*matchL2_gpu<unsigned char>*/, 0/*matchL2_gpu<signed char>*/,
+            0/*matchL2_gpu<unsigned short>*/, 0/*matchL2_gpu<short>*/,
+            0/*matchL2_gpu<int>*/, matchL2_gpu<float>
+        },
+        {
+            matchHamming_gpu<unsigned char>, 0/*matchHamming_gpu<signed char>*/,
+            0/*matchHamming_gpu<unsigned short>*/, 0/*matchHamming_gpu<short>*/,
+            0/*matchHamming_gpu<int>*/, 0/*matchHamming_gpu<float>*/
+        }
+    };
+#else
     static const caller_t callers[3][6] =
     {
         {
@@ -469,6 +521,7 @@ void cv::gpu::BruteForceMatcher_GPU_base::knnMatchSingle(const GpuMat& query, co
             matchHamming_gpu<int>, 0/*matchHamming_gpu<float>*/
         }
     };
+#endif
 
     CV_Assert(query.channels() == 1 && query.depth() < CV_64F);
     CV_Assert(train.type() == query.type() && train.cols == query.cols);
@@ -568,6 +621,16 @@ void cv::gpu::BruteForceMatcher_GPU_base::knnMatch2Collection(const GpuMat& quer
     GpuMat& trainIdx, GpuMat& imgIdx, GpuMat& distance,
     const GpuMat& maskCollection, Stream& stream)
 {
+#ifdef OPENCV_TINY_GPU_MODULE
+    (void)query;
+    (void)trainCollection;
+    (void)trainIdx;
+    (void)imgIdx;
+    (void)distance;
+    (void)maskCollection;
+    (void)stream;
+    CV_Error(CV_StsNotImplemented, "not available in tiny build");
+#else
     if (query.empty() || trainCollection.empty())
         return;
 
@@ -613,6 +676,7 @@ void cv::gpu::BruteForceMatcher_GPU_base::knnMatch2Collection(const GpuMat& quer
     CV_Assert(func != 0);
 
     func(query, trainCollection, maskCollection, trainIdx, imgIdx, distance, StreamAccessor::getStream(stream));
+#endif
 }
 
 void cv::gpu::BruteForceMatcher_GPU_base::knnMatch2Download(const GpuMat& trainIdx, const GpuMat& imgIdx, const GpuMat& distance,
@@ -755,6 +819,26 @@ void cv::gpu::BruteForceMatcher_GPU_base::radiusMatchSingle(const GpuMat& query,
                              const PtrStepSzi& trainIdx, const PtrStepSzf& distance, const PtrStepSz<unsigned int>& nMatches,
                              cudaStream_t stream);
 
+#ifdef OPENCV_TINY_GPU_MODULE
+    static const caller_t callers[3][6] =
+    {
+        {
+            0/*matchL1_gpu<unsigned char>*/, 0/*matchL1_gpu<signed char>*/,
+            0/*matchL1_gpu<unsigned short>*/, 0/*matchL1_gpu<short>*/,
+            0/*matchL1_gpu<int>*/, 0/*matchL1_gpu<float>*/
+        },
+        {
+            0/*matchL2_gpu<unsigned char>*/, 0/*matchL2_gpu<signed char>*/,
+            0/*matchL2_gpu<unsigned short>*/, 0/*matchL2_gpu<short>*/,
+            0/*matchL2_gpu<int>*/, matchL2_gpu<float>
+        },
+        {
+            matchHamming_gpu<unsigned char>, 0/*matchHamming_gpu<signed char>*/,
+            0/*matchHamming_gpu<unsigned short>*/, 0/*matchHamming_gpu<short>*/,
+            0/*matchHamming_gpu<int>*/, 0/*matchHamming_gpu<float>*/
+        }
+    };
+#else
     static const caller_t callers[3][6] =
     {
         {
@@ -773,6 +857,7 @@ void cv::gpu::BruteForceMatcher_GPU_base::radiusMatchSingle(const GpuMat& query,
             matchHamming_gpu<int>, 0/*matchHamming_gpu<float>*/
         }
     };
+#endif
 
     const int nQuery = query.rows;
     const int nTrain = train.rows;
@@ -872,6 +957,17 @@ void cv::gpu::BruteForceMatcher_GPU_base::radiusMatch(const GpuMat& query, const
 void cv::gpu::BruteForceMatcher_GPU_base::radiusMatchCollection(const GpuMat& query, GpuMat& trainIdx, GpuMat& imgIdx, GpuMat& distance, GpuMat& nMatches,
     float maxDistance, const vector<GpuMat>& masks, Stream& stream)
 {
+#ifdef OPENCV_TINY_GPU_MODULE
+    (void)query;
+    (void)trainIdx;
+    (void)imgIdx;
+    (void)distance;
+    (void)nMatches;
+    (void)maxDistance;
+    (void)masks;
+    (void)stream;
+    CV_Error(CV_StsNotImplemented, "not available in tiny build");
+#else
     if (query.empty() || empty())
         return;
 
@@ -926,6 +1022,7 @@ void cv::gpu::BruteForceMatcher_GPU_base::radiusMatchCollection(const GpuMat& qu
 
     func(query, &trains_[0], static_cast<int>(trains_.size()), maxDistance, masks_.size() == 0 ? 0 : &masks_[0],
         trainIdx, imgIdx, distance, nMatches, StreamAccessor::getStream(stream));
+#endif
 }
 
 void cv::gpu::BruteForceMatcher_GPU_base::radiusMatchDownload(const GpuMat& trainIdx, const GpuMat& imgIdx, const GpuMat& distance, const GpuMat& nMatches,
diff --git a/modules/gpu/src/cuda/bf_knnmatch.cu b/modules/gpu/src/cuda/bf_knnmatch.cu
index 3e5bc741ff..640dafb816 100644
--- a/modules/gpu/src/cuda/bf_knnmatch.cu
+++ b/modules/gpu/src/cuda/bf_knnmatch.cu
@@ -1168,12 +1168,14 @@ namespace cv { namespace gpu { namespace device
                 matchDispatcher< L1Dist<T> >(static_cast< PtrStepSz<T> >(query), static_cast< PtrStepSz<T> >(train), k, WithOutMask(), trainIdx, distance, allDist, stream);
         }
 
+#ifndef OPENCV_TINY_GPU_MODULE
         template void matchL1_gpu<uchar >(const PtrStepSzb& queryDescs, const PtrStepSzb& trainDescs, int k, const PtrStepSzb& mask, const PtrStepSzb& trainIdx, const PtrStepSzb& distance, const PtrStepSzf& allDist, cudaStream_t stream);
         //template void matchL1_gpu<schar >(const PtrStepSzb& queryDescs, const PtrStepSzb& trainDescs, int k, const PtrStepSzb& mask, const PtrStepSzb& trainIdx, const PtrStepSzb& distance, const PtrStepSzf& allDist, cudaStream_t stream);
         template void matchL1_gpu<ushort>(const PtrStepSzb& queryDescs, const PtrStepSzb& trainDescs, int k, const PtrStepSzb& mask, const PtrStepSzb& trainIdx, const PtrStepSzb& distance, const PtrStepSzf& allDist, cudaStream_t stream);
         template void matchL1_gpu<short >(const PtrStepSzb& queryDescs, const PtrStepSzb& trainDescs, int k, const PtrStepSzb& mask, const PtrStepSzb& trainIdx, const PtrStepSzb& distance, const PtrStepSzf& allDist, cudaStream_t stream);
         template void matchL1_gpu<int   >(const PtrStepSzb& queryDescs, const PtrStepSzb& trainDescs, int k, const PtrStepSzb& mask, const PtrStepSzb& trainIdx, const PtrStepSzb& distance, const PtrStepSzf& allDist, cudaStream_t stream);
         template void matchL1_gpu<float >(const PtrStepSzb& queryDescs, const PtrStepSzb& trainDescs, int k, const PtrStepSzb& mask, const PtrStepSzb& trainIdx, const PtrStepSzb& distance, const PtrStepSzf& allDist, cudaStream_t stream);
+#endif
 
         template <typename T> void matchL2_gpu(const PtrStepSzb& query, const PtrStepSzb& train, int k, const PtrStepSzb& mask,
             const PtrStepSzb& trainIdx, const PtrStepSzb& distance, const PtrStepSzf& allDist,
@@ -1185,11 +1187,13 @@ namespace cv { namespace gpu { namespace device
                 matchDispatcher<L2Dist>(static_cast< PtrStepSz<T> >(query), static_cast< PtrStepSz<T> >(train), k, WithOutMask(), trainIdx, distance, allDist, stream);
         }
 
+#ifndef OPENCV_TINY_GPU_MODULE
         //template void matchL2_gpu<uchar >(const PtrStepSzb& queryDescs, const PtrStepSzb& trainDescs, int k, const PtrStepSzb& mask, const PtrStepSzb& trainIdx, const PtrStepSzb& distance, const PtrStepSzf& allDist, cudaStream_t stream);
         //template void matchL2_gpu<schar >(const PtrStepSzb& queryDescs, const PtrStepSzb& trainDescs, int k, const PtrStepSzb& mask, const PtrStepSzb& trainIdx, const PtrStepSzb& distance, const PtrStepSzf& allDist, cudaStream_t stream);
         //template void matchL2_gpu<ushort>(const PtrStepSzb& queryDescs, const PtrStepSzb& trainDescs, int k, const PtrStepSzb& mask, const PtrStepSzb& trainIdx, const PtrStepSzb& distance, const PtrStepSzf& allDist, cudaStream_t stream);
         //template void matchL2_gpu<short >(const PtrStepSzb& queryDescs, const PtrStepSzb& trainDescs, int k, const PtrStepSzb& mask, const PtrStepSzb& trainIdx, const PtrStepSzb& distance, const PtrStepSzf& allDist, cudaStream_t stream);
         //template void matchL2_gpu<int   >(const PtrStepSzb& queryDescs, const PtrStepSzb& trainDescs, int k, const PtrStepSzb& mask, const PtrStepSzb& trainIdx, const PtrStepSzb& distance, const PtrStepSzf& allDist, cudaStream_t stream);
+#endif
         template void matchL2_gpu<float >(const PtrStepSzb& queryDescs, const PtrStepSzb& trainDescs, int k, const PtrStepSzb& mask, const PtrStepSzb& trainIdx, const PtrStepSzb& distance, const PtrStepSzf& allDist, cudaStream_t stream);
 
         template <typename T> void matchHamming_gpu(const PtrStepSzb& query, const PtrStepSzb& train, int k, const PtrStepSzb& mask,
@@ -1203,10 +1207,12 @@ namespace cv { namespace gpu { namespace device
         }
 
         template void matchHamming_gpu<uchar >(const PtrStepSzb& queryDescs, const PtrStepSzb& trainDescs, int k, const PtrStepSzb& mask, const PtrStepSzb& trainIdx, const PtrStepSzb& distance, const PtrStepSzf& allDist, cudaStream_t stream);
+#ifndef OPENCV_TINY_GPU_MODULE
         //template void matchHamming_gpu<schar >(const PtrStepSzb& queryDescs, const PtrStepSzb& trainDescs, int k, const PtrStepSzb& mask, const PtrStepSzb& trainIdx, const PtrStepSzb& distance, const PtrStepSzf& allDist, cudaStream_t stream);
         template void matchHamming_gpu<ushort>(const PtrStepSzb& queryDescs, const PtrStepSzb& trainDescs, int k, const PtrStepSzb& mask, const PtrStepSzb& trainIdx, const PtrStepSzb& distance, const PtrStepSzf& allDist, cudaStream_t stream);
         //template void matchHamming_gpu<short >(const PtrStepSzb& queryDescs, const PtrStepSzb& trainDescs, int k, const PtrStepSzb& mask, const PtrStepSzb& trainIdx, const PtrStepSzb& distance, const PtrStepSzf& allDist, cudaStream_t stream);
         template void matchHamming_gpu<int   >(const PtrStepSzb& queryDescs, const PtrStepSzb& trainDescs, int k, const PtrStepSzb& mask, const PtrStepSzb& trainIdx, const PtrStepSzb& distance, const PtrStepSzf& allDist, cudaStream_t stream);
+#endif
 
         template <typename T> void match2L1_gpu(const PtrStepSzb& query, const PtrStepSzb& trains, const PtrStepSz<PtrStepb>& masks,
             const PtrStepSzb& trainIdx, const PtrStepSzb& imgIdx, const PtrStepSzb& distance,
@@ -1218,12 +1224,14 @@ namespace cv { namespace gpu { namespace device
                 match2Dispatcher< L1Dist<T> >(static_cast< PtrStepSz<T> >(query), (const PtrStepSz<T>*)trains.ptr(), trains.cols, WithOutMask(), trainIdx, imgIdx, distance,  stream);
         }
 
+#ifndef OPENCV_TINY_GPU_MODULE
         template void match2L1_gpu<uchar >(const PtrStepSzb& query, const PtrStepSzb& trains, const PtrStepSz<PtrStepb>& masks, const PtrStepSzb& trainIdx, const PtrStepSzb& imgIdx, const PtrStepSzb& distance, cudaStream_t stream);
         //template void match2L1_gpu<schar >(const PtrStepSzb& query, const PtrStepSzb& trains, const PtrStepSz<PtrStepb>& masks, const PtrStepSzb& trainIdx, const PtrStepSzb& imgIdx, const PtrStepSzb& distance, cudaStream_t stream);
         template void match2L1_gpu<ushort>(const PtrStepSzb& query, const PtrStepSzb& trains, const PtrStepSz<PtrStepb>& masks, const PtrStepSzb& trainIdx, const PtrStepSzb& imgIdx, const PtrStepSzb& distance, cudaStream_t stream);
         template void match2L1_gpu<short >(const PtrStepSzb& query, const PtrStepSzb& trains, const PtrStepSz<PtrStepb>& masks, const PtrStepSzb& trainIdx, const PtrStepSzb& imgIdx, const PtrStepSzb& distance, cudaStream_t stream);
         template void match2L1_gpu<int   >(const PtrStepSzb& query, const PtrStepSzb& trains, const PtrStepSz<PtrStepb>& masks, const PtrStepSzb& trainIdx, const PtrStepSzb& imgIdx, const PtrStepSzb& distance, cudaStream_t stream);
         template void match2L1_gpu<float >(const PtrStepSzb& query, const PtrStepSzb& trains, const PtrStepSz<PtrStepb>& masks, const PtrStepSzb& trainIdx, const PtrStepSzb& imgIdx, const PtrStepSzb& distance, cudaStream_t stream);
+#endif
 
         template <typename T> void match2L2_gpu(const PtrStepSzb& query, const PtrStepSzb& trains, const PtrStepSz<PtrStepb>& masks,
             const PtrStepSzb& trainIdx, const PtrStepSzb& imgIdx, const PtrStepSzb& distance,
@@ -1235,12 +1243,14 @@ namespace cv { namespace gpu { namespace device
                 match2Dispatcher<L2Dist>(static_cast< PtrStepSz<T> >(query), (const PtrStepSz<T>*)trains.ptr(), trains.cols, WithOutMask(), trainIdx, imgIdx, distance, stream);
         }
 
+#ifndef OPENCV_TINY_GPU_MODULE
         //template void match2L2_gpu<uchar >(const PtrStepSzb& query, const PtrStepSzb& trains, const PtrStepSz<PtrStepb>& masks, const PtrStepSzb& trainIdx, const PtrStepSzb& imgIdx, const PtrStepSzb& distance, cudaStream_t stream);
         //template void match2L2_gpu<schar >(const PtrStepSzb& query, const PtrStepSzb& trains, const PtrStepSz<PtrStepb>& masks, const PtrStepSzb& trainIdx, const PtrStepSzb& imgIdx, const PtrStepSzb& distance, cudaStream_t stream);
         //template void match2L2_gpu<ushort>(const PtrStepSzb& query, const PtrStepSzb& trains, const PtrStepSz<PtrStepb>& masks, const PtrStepSzb& trainIdx, const PtrStepSzb& imgIdx, const PtrStepSzb& distance, cudaStream_t stream);
         //template void match2L2_gpu<short >(const PtrStepSzb& query, const PtrStepSzb& trains, const PtrStepSz<PtrStepb>& masks, const PtrStepSzb& trainIdx, const PtrStepSzb& imgIdx, const PtrStepSzb& distance, cudaStream_t stream);
         //template void match2L2_gpu<int   >(const PtrStepSzb& query, const PtrStepSzb& trains, const PtrStepSz<PtrStepb>& masks, const PtrStepSzb& trainIdx, const PtrStepSzi& imgIdx, const PtrStepSzb& distance, cudaStream_t stream);
         template void match2L2_gpu<float >(const PtrStepSzb& query, const PtrStepSzb& trains, const PtrStepSz<PtrStepb>& masks, const PtrStepSzb& trainIdx, const PtrStepSzb& imgIdx, const PtrStepSzb& distance, cudaStream_t stream);
+#endif
 
         template <typename T> void match2Hamming_gpu(const PtrStepSzb& query, const PtrStepSzb& trains, const PtrStepSz<PtrStepb>& masks,
             const PtrStepSzb& trainIdx, const PtrStepSzb& imgIdx, const PtrStepSzb& distance,
@@ -1252,11 +1262,13 @@ namespace cv { namespace gpu { namespace device
                 match2Dispatcher<HammingDist>(static_cast< PtrStepSz<T> >(query), (const PtrStepSz<T>*)trains.ptr(), trains.cols, WithOutMask(), trainIdx, imgIdx, distance, stream);
         }
 
+#ifndef OPENCV_TINY_GPU_MODULE
         template void match2Hamming_gpu<uchar >(const PtrStepSzb& query, const PtrStepSzb& trains, const PtrStepSz<PtrStepb>& masks, const PtrStepSzb& trainIdx, const PtrStepSzb& imgIdx, const PtrStepSzb& distance, cudaStream_t stream);
         //template void match2Hamming_gpu<schar >(const PtrStepSzb& query, const PtrStepSzb& trains, const PtrStepSz<PtrStepb>& masks, const PtrStepSzb& trainIdx, const PtrStepSzb& imgIdx, const PtrStepSzb& distance, cudaStream_t stream);
         template void match2Hamming_gpu<ushort>(const PtrStepSzb& query, const PtrStepSzb& trains, const PtrStepSz<PtrStepb>& masks, const PtrStepSzb& trainIdx, const PtrStepSzb& imgIdx, const PtrStepSzb& distance, cudaStream_t stream);
         //template void match2Hamming_gpu<short >(const PtrStepSzb& query, const PtrStepSzb& trains, const PtrStepSz<PtrStepb>& masks, const PtrStepSzb& trainIdx, const PtrStepSzb& imgIdx, const PtrStepSzb& distance, cudaStream_t stream);
         template void match2Hamming_gpu<int   >(const PtrStepSzb& query, const PtrStepSzb& trains, const PtrStepSz<PtrStepb>& masks, const PtrStepSzb& trainIdx, const PtrStepSzb& imgIdx, const PtrStepSzb& distance, cudaStream_t stream);
+#endif
     } // namespace bf_knnmatch
 }}} // namespace cv { namespace gpu { namespace device {
 
diff --git a/modules/gpu/src/cuda/bf_match.cu b/modules/gpu/src/cuda/bf_match.cu
index c2ae48bb30..baaf979e5d 100644
--- a/modules/gpu/src/cuda/bf_match.cu
+++ b/modules/gpu/src/cuda/bf_match.cu
@@ -644,12 +644,14 @@ namespace cv { namespace gpu { namespace device
             }
         }
 
+#ifndef OPENCV_TINY_GPU_MODULE
         template void matchL1_gpu<uchar >(const PtrStepSzb& queryDescs, const PtrStepSzb& trainDescs, const PtrStepSzb& mask, const PtrStepSzi& trainIdx, const PtrStepSzf& distance, cudaStream_t stream);
         //template void matchL1_gpu<schar >(const PtrStepSzb& queryDescs, const PtrStepSzb& trainDescs, const PtrStepSzb& mask, const PtrStepSzi& trainIdx, const PtrStepSzf& distance, cudaStream_t stream);
         template void matchL1_gpu<ushort>(const PtrStepSzb& queryDescs, const PtrStepSzb& trainDescs, const PtrStepSzb& mask, const PtrStepSzi& trainIdx, const PtrStepSzf& distance, cudaStream_t stream);
         template void matchL1_gpu<short >(const PtrStepSzb& queryDescs, const PtrStepSzb& trainDescs, const PtrStepSzb& mask, const PtrStepSzi& trainIdx, const PtrStepSzf& distance, cudaStream_t stream);
         template void matchL1_gpu<int   >(const PtrStepSzb& queryDescs, const PtrStepSzb& trainDescs, const PtrStepSzb& mask, const PtrStepSzi& trainIdx, const PtrStepSzf& distance, cudaStream_t stream);
         template void matchL1_gpu<float >(const PtrStepSzb& queryDescs, const PtrStepSzb& trainDescs, const PtrStepSzb& mask, const PtrStepSzi& trainIdx, const PtrStepSzf& distance, cudaStream_t stream);
+#endif
 
         template <typename T> void matchL2_gpu(const PtrStepSzb& query, const PtrStepSzb& train, const PtrStepSzb& mask,
                                                const PtrStepSzi& trainIdx, const PtrStepSzf& distance,
@@ -669,11 +671,13 @@ namespace cv { namespace gpu { namespace device
             }
         }
 
+#ifndef OPENCV_TINY_GPU_MODULE
         //template void matchL2_gpu<uchar >(const PtrStepSzb& queryDescs, const PtrStepSzb& trainDescs, const PtrStepSzb& mask, const PtrStepSzi& trainIdx, const PtrStepSzf& distance, cudaStream_t stream);
         //template void matchL2_gpu<schar >(const PtrStepSzb& queryDescs, const PtrStepSzb& trainDescs, const PtrStepSzb& mask, const PtrStepSzi& trainIdx, const PtrStepSzf& distance, cudaStream_t stream);
         //template void matchL2_gpu<ushort>(const PtrStepSzb& queryDescs, const PtrStepSzb& trainDescs, const PtrStepSzb& mask, const PtrStepSzi& trainIdx, const PtrStepSzf& distance, cudaStream_t stream);
         //template void matchL2_gpu<short >(const PtrStepSzb& queryDescs, const PtrStepSzb& trainDescs, const PtrStepSzb& mask, const PtrStepSzi& trainIdx, const PtrStepSzf& distance, cudaStream_t stream);
         //template void matchL2_gpu<int   >(const PtrStepSzb& queryDescs, const PtrStepSzb& trainDescs, const PtrStepSzb& mask, const PtrStepSzi& trainIdx, const PtrStepSzf& distance, cudaStream_t stream);
+#endif
         template void matchL2_gpu<float >(const PtrStepSzb& queryDescs, const PtrStepSzb& trainDescs, const PtrStepSzb& mask, const PtrStepSzi& trainIdx, const PtrStepSzf& distance, cudaStream_t stream);
 
         template <typename T> void matchHamming_gpu(const PtrStepSzb& query, const PtrStepSzb& train, const PtrStepSzb& mask,
@@ -695,10 +699,12 @@ namespace cv { namespace gpu { namespace device
         }
 
         template void matchHamming_gpu<uchar >(const PtrStepSzb& queryDescs, const PtrStepSzb& trainDescs, const PtrStepSzb& mask, const PtrStepSzi& trainIdx, const PtrStepSzf& distance, cudaStream_t stream);
+#ifndef OPENCV_TINY_GPU_MODULE
         //template void matchHamming_gpu<schar >(const PtrStepSzb& queryDescs, const PtrStepSzb& trainDescs, const PtrStepSzb& mask, const PtrStepSzi& trainIdx, const PtrStepSzf& distance, cudaStream_t stream);
         template void matchHamming_gpu<ushort>(const PtrStepSzb& queryDescs, const PtrStepSzb& trainDescs, const PtrStepSzb& mask, const PtrStepSzi& trainIdx, const PtrStepSzf& distance, cudaStream_t stream);
         //template void matchHamming_gpu<short >(const PtrStepSzb& queryDescs, const PtrStepSzb& trainDescs, const PtrStepSzb& mask, const PtrStepSzi& trainIdx, const PtrStepSzf& distance, cudaStream_t stream);
         template void matchHamming_gpu<int   >(const PtrStepSzb& queryDescs, const PtrStepSzb& trainDescs, const PtrStepSzb& mask, const PtrStepSzi& trainIdx, const PtrStepSzf& distance, cudaStream_t stream);
+#endif
 
         template <typename T> void matchL1_gpu(const PtrStepSzb& query, const PtrStepSzb& trains, const PtrStepSz<PtrStepb>& masks,
                                                const PtrStepSzi& trainIdx, const PtrStepSzi& imgIdx, const PtrStepSzf& distance,
@@ -718,12 +724,14 @@ namespace cv { namespace gpu { namespace device
             }
         }
 
+#ifndef OPENCV_TINY_GPU_MODULE
         template void matchL1_gpu<uchar >(const PtrStepSzb& query, const PtrStepSzb& trains, const PtrStepSz<PtrStepb>& masks, const PtrStepSzi& trainIdx, const PtrStepSzi& imgIdx, const PtrStepSzf& distance, cudaStream_t stream);
         //template void matchL1_gpu<schar >(const PtrStepSzb& query, const PtrStepSzb& trains, const PtrStepSz<PtrStepb>& masks, const PtrStepSzi& trainIdx, const PtrStepSzi& imgIdx, const PtrStepSzf& distance, cudaStream_t stream);
         template void matchL1_gpu<ushort>(const PtrStepSzb& query, const PtrStepSzb& trains, const PtrStepSz<PtrStepb>& masks, const PtrStepSzi& trainIdx, const PtrStepSzi& imgIdx, const PtrStepSzf& distance, cudaStream_t stream);
         template void matchL1_gpu<short >(const PtrStepSzb& query, const PtrStepSzb& trains, const PtrStepSz<PtrStepb>& masks, const PtrStepSzi& trainIdx, const PtrStepSzi& imgIdx, const PtrStepSzf& distance, cudaStream_t stream);
         template void matchL1_gpu<int   >(const PtrStepSzb& query, const PtrStepSzb& trains, const PtrStepSz<PtrStepb>& masks, const PtrStepSzi& trainIdx, const PtrStepSzi& imgIdx, const PtrStepSzf& distance, cudaStream_t stream);
         template void matchL1_gpu<float >(const PtrStepSzb& query, const PtrStepSzb& trains, const PtrStepSz<PtrStepb>& masks, const PtrStepSzi& trainIdx, const PtrStepSzi& imgIdx, const PtrStepSzf& distance, cudaStream_t stream);
+#endif
 
         template <typename T> void matchL2_gpu(const PtrStepSzb& query, const PtrStepSzb& trains, const PtrStepSz<PtrStepb>& masks,
                                                const PtrStepSzi& trainIdx, const PtrStepSzi& imgIdx, const PtrStepSzf& distance,
@@ -743,12 +751,14 @@ namespace cv { namespace gpu { namespace device
             }
         }
 
+#ifndef OPENCV_TINY_GPU_MODULE
         //template void matchL2_gpu<uchar >(const PtrStepSzb& query, const PtrStepSzb& trains, const PtrStepSz<PtrStepb>& masks, const PtrStepSzi& trainIdx, const PtrStepSzi& imgIdx, const PtrStepSzf& distance, cudaStream_t stream);
         //template void matchL2_gpu<schar >(const PtrStepSzb& query, const PtrStepSzb& trains, const PtrStepSz<PtrStepb>& masks, const PtrStepSzi& trainIdx, const PtrStepSzi& imgIdx, const PtrStepSzf& distance, cudaStream_t stream);
         //template void matchL2_gpu<ushort>(const PtrStepSzb& query, const PtrStepSzb& trains, const PtrStepSz<PtrStepb>& masks, const PtrStepSzi& trainIdx, const PtrStepSzi& imgIdx, const PtrStepSzf& distance, cudaStream_t stream);
         //template void matchL2_gpu<short >(const PtrStepSzb& query, const PtrStepSzb& trains, const PtrStepSz<PtrStepb>& masks, const PtrStepSzi& trainIdx, const PtrStepSzi& imgIdx, const PtrStepSzf& distance, cudaStream_t stream);
         //template void matchL2_gpu<int   >(const PtrStepSzb& query, const PtrStepSzb& trains, const PtrStepSz<PtrStepb>& masks, const PtrStepSzi& trainIdx, const PtrStepSzi& imgIdx, const PtrStepSzf& distance, cudaStream_t stream);
         template void matchL2_gpu<float >(const PtrStepSzb& query, const PtrStepSzb& trains, const PtrStepSz<PtrStepb>& maskCollection, const PtrStepSzi& trainIdx, const PtrStepSzi& imgIdx, const PtrStepSzf& distance, cudaStream_t stream);
+#endif
 
         template <typename T> void matchHamming_gpu(const PtrStepSzb& query, const PtrStepSzb& trains, const PtrStepSz<PtrStepb>& masks,
                                                     const PtrStepSzi& trainIdx, const PtrStepSzi& imgIdx, const PtrStepSzf& distance,
@@ -768,11 +778,13 @@ namespace cv { namespace gpu { namespace device
             }
         }
 
+#ifndef OPENCV_TINY_GPU_MODULE
         template void matchHamming_gpu<uchar >(const PtrStepSzb& query, const PtrStepSzb& trains, const PtrStepSz<PtrStepb>& masks, const PtrStepSzi& trainIdx, const PtrStepSzi& imgIdx, const PtrStepSzf& distance, cudaStream_t stream);
         //template void matchHamming_gpu<schar >(const PtrStepSzb& query, const PtrStepSzb& trains, const PtrStepSz<PtrStepb>& masks, const PtrStepSzi& trainIdx, const PtrStepSzi& imgIdx, const PtrStepSzf& distance, cudaStream_t stream);
         template void matchHamming_gpu<ushort>(const PtrStepSzb& query, const PtrStepSzb& trains, const PtrStepSz<PtrStepb>& masks, const PtrStepSzi& trainIdx, const PtrStepSzi& imgIdx, const PtrStepSzf& distance, cudaStream_t stream);
         //template void matchHamming_gpu<short >(const PtrStepSzb& query, const PtrStepSzb& trains, const PtrStepSz<PtrStepb>& masks, const PtrStepSzi& trainIdx, const PtrStepSzi& imgIdx, const PtrStepSzf& distance, cudaStream_t stream);
         template void matchHamming_gpu<int   >(const PtrStepSzb& query, const PtrStepSzb& trains, const PtrStepSz<PtrStepb>& masks, const PtrStepSzi& trainIdx, const PtrStepSzi& imgIdx, const PtrStepSzf& distance, cudaStream_t stream);
+#endif
     } // namespace bf_match
 }}} // namespace cv { namespace gpu { namespace device {
 
diff --git a/modules/gpu/src/cuda/bf_radius_match.cu b/modules/gpu/src/cuda/bf_radius_match.cu
index d83f9f7f96..8493b4e065 100644
--- a/modules/gpu/src/cuda/bf_radius_match.cu
+++ b/modules/gpu/src/cuda/bf_radius_match.cu
@@ -356,12 +356,14 @@ namespace cv { namespace gpu { namespace device
             }
         }
 
+#ifndef OPENCV_TINY_GPU_MODULE
         template void matchL1_gpu<uchar >(const PtrStepSzb& queryDescs, const PtrStepSzb& trainDescs, float maxDistance, const PtrStepSzb& mask, const PtrStepSzi& trainIdx, const PtrStepSzf& distance, const PtrStepSz<unsigned int>& nMatches, cudaStream_t stream);
         //template void matchL1_gpu<schar >(const PtrStepSzb& queryDescs, const PtrStepSzb& trainDescs, float maxDistance, const PtrStepSzb& mask, const PtrStepSzi& trainIdx, const PtrStepSzf& distance, const PtrStepSz<unsigned int>& nMatches, cudaStream_t stream);
         template void matchL1_gpu<ushort>(const PtrStepSzb& queryDescs, const PtrStepSzb& trainDescs, float maxDistance, const PtrStepSzb& mask, const PtrStepSzi& trainIdx, const PtrStepSzf& distance, const PtrStepSz<unsigned int>& nMatches, cudaStream_t stream);
         template void matchL1_gpu<short >(const PtrStepSzb& queryDescs, const PtrStepSzb& trainDescs, float maxDistance, const PtrStepSzb& mask, const PtrStepSzi& trainIdx, const PtrStepSzf& distance, const PtrStepSz<unsigned int>& nMatches, cudaStream_t stream);
         template void matchL1_gpu<int   >(const PtrStepSzb& queryDescs, const PtrStepSzb& trainDescs, float maxDistance, const PtrStepSzb& mask, const PtrStepSzi& trainIdx, const PtrStepSzf& distance, const PtrStepSz<unsigned int>& nMatches, cudaStream_t stream);
         template void matchL1_gpu<float >(const PtrStepSzb& queryDescs, const PtrStepSzb& trainDescs, float maxDistance, const PtrStepSzb& mask, const PtrStepSzi& trainIdx, const PtrStepSzf& distance, const PtrStepSz<unsigned int>& nMatches, cudaStream_t stream);
+#endif
 
         template <typename T> void matchL2_gpu(const PtrStepSzb& query, const PtrStepSzb& train, float maxDistance, const PtrStepSzb& mask,
             const PtrStepSzi& trainIdx, const PtrStepSzf& distance, const PtrStepSz<unsigned int>& nMatches,
@@ -381,11 +383,13 @@ namespace cv { namespace gpu { namespace device
             }
         }
 
+#ifndef OPENCV_TINY_GPU_MODULE
         //template void matchL2_gpu<uchar >(const PtrStepSzb& queryDescs, const PtrStepSzb& trainDescs, float maxDistance, const PtrStepSzb& mask, const PtrStepSzi& trainIdx, const PtrStepSzf& distance, const PtrStepSz<unsigned int>& nMatches, cudaStream_t stream);
         //template void matchL2_gpu<schar >(const PtrStepSzb& queryDescs, const PtrStepSzb& trainDescs, float maxDistance, const PtrStepSzb& mask, const PtrStepSzi& trainIdx, const PtrStepSzf& distance, const PtrStepSz<unsigned int>& nMatches, cudaStream_t stream);
         //template void matchL2_gpu<ushort>(const PtrStepSzb& queryDescs, const PtrStepSzb& trainDescs, float maxDistance, const PtrStepSzb& mask, const PtrStepSzi& trainIdx, const PtrStepSzf& distance, const PtrStepSz<unsigned int>& nMatches, cudaStream_t stream);
         //template void matchL2_gpu<short >(const PtrStepSzb& queryDescs, const PtrStepSzb& trainDescs, float maxDistance, const PtrStepSzb& mask, const PtrStepSzi& trainIdx, const PtrStepSzf& distance, const PtrStepSz<unsigned int>& nMatches, cudaStream_t stream);
         //template void matchL2_gpu<int   >(const PtrStepSzb& queryDescs, const PtrStepSzb& trainDescs, float maxDistance, const PtrStepSzb& mask, const PtrStepSzi& trainIdx, const PtrStepSzf& distance, const PtrStepSz<unsigned int>& nMatches, cudaStream_t stream);
+#endif
         template void matchL2_gpu<float >(const PtrStepSzb& queryDescs, const PtrStepSzb& trainDescs, float maxDistance, const PtrStepSzb& mask, const PtrStepSzi& trainIdx, const PtrStepSzf& distance, const PtrStepSz<unsigned int>& nMatches, cudaStream_t stream);
 
         template <typename T> void matchHamming_gpu(const PtrStepSzb& query, const PtrStepSzb& train, float maxDistance, const PtrStepSzb& mask,
@@ -407,10 +411,12 @@ namespace cv { namespace gpu { namespace device
         }
 
         template void matchHamming_gpu<uchar >(const PtrStepSzb& queryDescs, const PtrStepSzb& trainDescs, float maxDistance, const PtrStepSzb& mask, const PtrStepSzi& trainIdx, const PtrStepSzf& distance, const PtrStepSz<unsigned int>& nMatches, cudaStream_t stream);
+#ifndef OPENCV_TINY_GPU_MODULE
         //template void matchHamming_gpu<schar >(const PtrStepSzb& queryDescs, const PtrStepSzb& trainDescs, float maxDistance, const PtrStepSzb& mask, const PtrStepSzi& trainIdx, const PtrStepSzf& distance, const PtrStepSz<unsigned int>& nMatches, cudaStream_t stream);
         template void matchHamming_gpu<ushort>(const PtrStepSzb& queryDescs, const PtrStepSzb& trainDescs, float maxDistance, const PtrStepSzb& mask, const PtrStepSzi& trainIdx, const PtrStepSzf& distance, const PtrStepSz<unsigned int>& nMatches, cudaStream_t stream);
         //template void matchHamming_gpu<short >(const PtrStepSzb& queryDescs, const PtrStepSzb& trainDescs, float maxDistance, const PtrStepSzb& mask, const PtrStepSzi& trainIdx, const PtrStepSzf& distance, const PtrStepSz<unsigned int>& nMatches, cudaStream_t stream);
         template void matchHamming_gpu<int   >(const PtrStepSzb& queryDescs, const PtrStepSzb& trainDescs, float maxDistance, const PtrStepSzb& mask, const PtrStepSzi& trainIdx, const PtrStepSzf& distance, const PtrStepSz<unsigned int>& nMatches, cudaStream_t stream);
+#endif
 
         template <typename T> void matchL1_gpu(const PtrStepSzb& query, const PtrStepSzb* trains, int n, float maxDistance, const PtrStepSzb* masks,
             const PtrStepSzi& trainIdx, const PtrStepSzi& imgIdx, const PtrStepSzf& distance, const PtrStepSz<unsigned int>& nMatches,
@@ -421,12 +427,14 @@ namespace cv { namespace gpu { namespace device
                 stream);
         }
 
+#ifndef OPENCV_TINY_GPU_MODULE
         template void matchL1_gpu<uchar >(const PtrStepSzb& query, const PtrStepSzb* trains, int n, float maxDistance, const PtrStepSzb* masks, const PtrStepSzi& trainIdx, const PtrStepSzi& imgIdx, const PtrStepSzf& distance, const PtrStepSz<unsigned int>& nMatches, cudaStream_t stream);
         //template void matchL1_gpu<schar >(const PtrStepSzb& query, const PtrStepSzb* trains, int n, float maxDistance, const PtrStepSzb* masks, const PtrStepSzi& trainIdx, const PtrStepSzi& imgIdx, const PtrStepSzf& distance, const PtrStepSz<unsigned int>& nMatches, cudaStream_t stream);
         template void matchL1_gpu<ushort>(const PtrStepSzb& query, const PtrStepSzb* trains, int n, float maxDistance, const PtrStepSzb* masks, const PtrStepSzi& trainIdx, const PtrStepSzi& imgIdx, const PtrStepSzf& distance, const PtrStepSz<unsigned int>& nMatches, cudaStream_t stream);
         template void matchL1_gpu<short >(const PtrStepSzb& query, const PtrStepSzb* trains, int n, float maxDistance, const PtrStepSzb* masks, const PtrStepSzi& trainIdx, const PtrStepSzi& imgIdx, const PtrStepSzf& distance, const PtrStepSz<unsigned int>& nMatches, cudaStream_t stream);
         template void matchL1_gpu<int   >(const PtrStepSzb& query, const PtrStepSzb* trains, int n, float maxDistance, const PtrStepSzb* masks, const PtrStepSzi& trainIdx, const PtrStepSzi& imgIdx, const PtrStepSzf& distance, const PtrStepSz<unsigned int>& nMatches, cudaStream_t stream);
         template void matchL1_gpu<float >(const PtrStepSzb& query, const PtrStepSzb* trains, int n, float maxDistance, const PtrStepSzb* masks, const PtrStepSzi& trainIdx, const PtrStepSzi& imgIdx, const PtrStepSzf& distance, const PtrStepSz<unsigned int>& nMatches, cudaStream_t stream);
+#endif
 
         template <typename T> void matchL2_gpu(const PtrStepSzb& query, const PtrStepSzb* trains, int n, float maxDistance, const PtrStepSzb* masks,
             const PtrStepSzi& trainIdx, const PtrStepSzi& imgIdx, const PtrStepSzf& distance, const PtrStepSz<unsigned int>& nMatches,
@@ -437,12 +445,14 @@ namespace cv { namespace gpu { namespace device
                 stream);
         }
 
+#ifndef OPENCV_TINY_GPU_MODULE
         //template void matchL2_gpu<uchar >(const PtrStepSzb& query, const PtrStepSzb* trains, int n, float maxDistance, const PtrStepSzb* masks, const PtrStepSzi& trainIdx, const PtrStepSzi& imgIdx, const PtrStepSzf& distance, const PtrStepSz<unsigned int>& nMatches, cudaStream_t stream);
         //template void matchL2_gpu<schar >(const PtrStepSzb& query, const PtrStepSzb* trains, int n, float maxDistance, const PtrStepSzb* masks, const PtrStepSzi& trainIdx, const PtrStepSzi& imgIdx, const PtrStepSzf& distance, const PtrStepSz<unsigned int>& nMatches, cudaStream_t stream);
         //template void matchL2_gpu<ushort>(const PtrStepSzb& query, const PtrStepSzb* trains, int n, float maxDistance, const PtrStepSzb* masks, const PtrStepSzi& trainIdx, const PtrStepSzi& imgIdx, const PtrStepSzf& distance, const PtrStepSz<unsigned int>& nMatches, cudaStream_t stream);
         //template void matchL2_gpu<short >(const PtrStepSzb& query, const PtrStepSzb* trains, int n, float maxDistance, const PtrStepSzb* masks, const PtrStepSzi& trainIdx, const PtrStepSzi& imgIdx, const PtrStepSzf& distance, const PtrStepSz<unsigned int>& nMatches, cudaStream_t stream);
         //template void matchL2_gpu<int   >(const PtrStepSzb& query, const PtrStepSzb* trains, int n, float maxDistance, const PtrStepSzb* masks, const PtrStepSzi& trainIdx, const PtrStepSzi& imgIdx, const PtrStepSzf& distance, const PtrStepSz<unsigned int>& nMatches, cudaStream_t stream);
         template void matchL2_gpu<float >(const PtrStepSzb& query, const PtrStepSzb* trains, int n, float maxDistance, const PtrStepSzb* masks, const PtrStepSzi& trainIdx, const PtrStepSzi& imgIdx, const PtrStepSzf& distance, const PtrStepSz<unsigned int>& nMatches, cudaStream_t stream);
+#endif
 
         template <typename T> void matchHamming_gpu(const PtrStepSzb& query, const PtrStepSzb* trains, int n, float maxDistance, const PtrStepSzb* masks,
             const PtrStepSzi& trainIdx, const PtrStepSzi& imgIdx, const PtrStepSzf& distance, const PtrStepSz<unsigned int>& nMatches,
@@ -453,11 +463,13 @@ namespace cv { namespace gpu { namespace device
                 stream);
         }
 
+#ifndef OPENCV_TINY_GPU_MODULE
         template void matchHamming_gpu<uchar >(const PtrStepSzb& query, const PtrStepSzb* trains, int n, float maxDistance, const PtrStepSzb* masks, const PtrStepSzi& trainIdx, const PtrStepSzi& imgIdx, const PtrStepSzf& distance, const PtrStepSz<unsigned int>& nMatches, cudaStream_t stream);
         //template void matchHamming_gpu<schar >(const PtrStepSzb& query, const PtrStepSzb* trains, int n, float maxDistance, const PtrStepSzb* masks, const PtrStepSzi& trainIdx, const PtrStepSzi& imgIdx, const PtrStepSzf& distance, const PtrStepSz<unsigned int>& nMatches, cudaStream_t stream);
         template void matchHamming_gpu<ushort>(const PtrStepSzb& query, const PtrStepSzb* trains, int n, float maxDistance, const PtrStepSzb* masks, const PtrStepSzi& trainIdx, const PtrStepSzi& imgIdx, const PtrStepSzf& distance, const PtrStepSz<unsigned int>& nMatches, cudaStream_t stream);
         //template void matchHamming_gpu<short >(const PtrStepSzb& query, const PtrStepSzb* trains, int n, float maxDistance, const PtrStepSzb* masks, const PtrStepSzi& trainIdx, const PtrStepSzi& imgIdx, const PtrStepSzf& distance, const PtrStepSz<unsigned int>& nMatches, cudaStream_t stream);
         template void matchHamming_gpu<int   >(const PtrStepSzb& query, const PtrStepSzb* trains, int n, float maxDistance, const PtrStepSzb* masks, const PtrStepSzi& trainIdx, const PtrStepSzi& imgIdx, const PtrStepSzf& distance, const PtrStepSz<unsigned int>& nMatches, cudaStream_t stream);
+#endif
     } // namespace bf_radius_match
 }}} // namespace cv { namespace gpu { namespace device