From: Alexey Spizhevoy <no@email>
Date: Wed, 2 Feb 2011 07:23:55 +0000 (+0000)
Subject: fixed gpu::sum* on CC1.0, updated some tests
X-Git-Tag: accepted/2.0/20130307.220821~3564
X-Git-Url: http://review.tizen.org/git/?a=commitdiff_plain;h=557dd39f0312cb7254721898b9e2274e4dba40af;p=profile%2Fivi%2Fopencv.git

fixed gpu::sum* on CC1.0, updated some tests
---

diff --git a/modules/gpu/include/opencv2/gpu/gpu.hpp b/modules/gpu/include/opencv2/gpu/gpu.hpp
index df30a8c..0862ee6 100644
--- a/modules/gpu/include/opencv2/gpu/gpu.hpp
+++ b/modules/gpu/include/opencv2/gpu/gpu.hpp
@@ -66,8 +66,14 @@ namespace cv
 
         enum GpuFeature
         {
-            NATIVE_DOUBLE,
-            ATOMICS
+            COMPUTE_10 = 10,
+            COMPUTE_11 = 11,
+            COMPUTE_12 = 12,
+            COMPUTE_13 = 13,
+            COMPUTE_20 = 20,
+            COMPUTE_21 = 21,
+            ATOMICS = COMPUTE_11,
+            NATIVE_DOUBLE = COMPUTE_13
         };
 
         class CV_EXPORTS TargetArchs
diff --git a/modules/gpu/src/cuda/matrix_reductions.cu b/modules/gpu/src/cuda/matrix_reductions.cu
index 398e376..396a9d7 100644
--- a/modules/gpu/src/cuda/matrix_reductions.cu
+++ b/modules/gpu/src/cuda/matrix_reductions.cu
@@ -1394,7 +1394,7 @@ namespace cv { namespace gpu { namespace mathfunc
         smem[tid] = res.x;
         smem[tid + nthreads] = res.y;
         smem[tid + 2 * nthreads] = res.z;
-        smem[tid + 3 * nthreads] = res.z;
+        smem[tid + 3 * nthreads] = res.w;
         __syncthreads();
 
         sumInSmem<nthreads, R>(smem, tid);
@@ -1432,21 +1432,25 @@ namespace cv { namespace gpu { namespace mathfunc
                     src, (typename TypeVec<R, 1>::vec_t*)buf.ptr(0));
             sumPass2Kernel<T, R, threads_x * threads_y><<<1, threads_x * threads_y>>>(
                     (typename TypeVec<R, 1>::vec_t*)buf.ptr(0), grid.x * grid.y);
+            break;
         case 2:
             sumKernel_C2<T, R, IdentityOp<R>, threads_x * threads_y><<<grid, threads>>>(
                     src, (typename TypeVec<R, 2>::vec_t*)buf.ptr(0));
             sumPass2Kernel_C2<T, R, threads_x * threads_y><<<1, threads_x * threads_y>>>(
                     (typename TypeVec<R, 2>::vec_t*)buf.ptr(0), grid.x * grid.y);
+            break;
         case 3:
             sumKernel_C3<T, R, IdentityOp<R>, threads_x * threads_y><<<grid, threads>>>(
                     src, (typename TypeVec<R, 3>::vec_t*)buf.ptr(0));
             sumPass2Kernel_C3<T, R, threads_x * threads_y><<<1, threads_x * threads_y>>>(
                     (typename TypeVec<R, 3>::vec_t*)buf.ptr(0), grid.x * grid.y);
+            break;
         case 4:
             sumKernel_C4<T, R, IdentityOp<R>, threads_x * threads_y><<<grid, threads>>>(
                     src, (typename TypeVec<R, 4>::vec_t*)buf.ptr(0));
             sumPass2Kernel_C4<T, R, threads_x * threads_y><<<1, threads_x * threads_y>>>(
                     (typename TypeVec<R, 4>::vec_t*)buf.ptr(0), grid.x * grid.y);
+            break;
         }
         cudaSafeCall(cudaThreadSynchronize());
 
diff --git a/modules/gpu/src/initialization.cpp b/modules/gpu/src/initialization.cpp
index d11b95c..039302c 100644
--- a/modules/gpu/src/initialization.cpp
+++ b/modules/gpu/src/initialization.cpp
@@ -71,11 +71,7 @@ namespace
 
 CV_EXPORTS bool cv::gpu::TargetArchs::builtWith(cv::gpu::GpuFeature feature)
 {
-    if (feature == NATIVE_DOUBLE)
-        return ::compareToSet(CUDA_ARCH_FEATURES, 13, std::greater_equal<int>());
-    if (feature == ATOMICS)
-        return ::compareToSet(CUDA_ARCH_FEATURES, 11, std::greater_equal<int>());
-    return true;
+    return ::compareToSet(CUDA_ARCH_FEATURES, feature, std::greater_equal<int>());
 }
 
 
diff --git a/tests/gpu/src/arithm.cpp b/tests/gpu/src/arithm.cpp
index 28cbc42..b7501a4 100644
--- a/tests/gpu/src/arithm.cpp
+++ b/tests/gpu/src/arithm.cpp
@@ -947,6 +947,16 @@ struct CV_GpuSumTest: CvTest
                 // sum
                 //
 
+                gen(1 + rand() % 500, 1 + rand() % 500, CV_MAKETYPE(type, 1), src);
+                a = sum(src);
+                b = sum(GpuMat(src));
+                if (abs(a[0] - b[0]) > src.size().area() * max_err)
+                {
+                    ts->printf(CvTS::CONSOLE, "1 cols: %d, rows: %d, expected: %f, actual: %f\n", src.cols, src.rows, a[0], b[0]);
+                    ts->set_failed_test_info(CvTS::FAIL_INVALID_OUTPUT);
+                    return;
+                }
+
                 gen(1 + rand() % 500, 1 + rand() % 500, CV_MAKETYPE(type, 2), src);
                 a = sum(src);
                 b = sum(GpuMat(src));
diff --git a/tests/gpu/src/meanshift.cpp b/tests/gpu/src/meanshift.cpp
index dc39423..d1bd345 100644
--- a/tests/gpu/src/meanshift.cpp
+++ b/tests/gpu/src/meanshift.cpp
@@ -56,7 +56,7 @@ struct CV_GpuMeanShiftTest : public CvTest
         cv::Mat img = cv::imread(std::string(ts->get_data_path()) + "meanshift/cones.png");
         cv::Mat img_template;       
         
-        if (cv::gpu::TargetArchs::hasEqualOrGreater(2, 0) && cv::gpu::DeviceInfo().major() >= 2)
+        if (cv::gpu::TargetArchs::builtWith(cv::gpu::COMPUTE_20) && cv::gpu::DeviceInfo().major() >= 2)
             img_template = cv::imread(std::string(ts->get_data_path()) + "meanshift/con_result.png");
         else
             img_template = cv::imread(std::string(ts->get_data_path()) + "meanshift/con_result_CC1X.png");
@@ -199,7 +199,7 @@ struct CV_GpuMeanShiftProcTest : public CvTest
             cv::Mat spmap_template;
             cv::FileStorage fs;
 
-            if (cv::gpu::TargetArchs::hasEqualOrGreater(2, 0) && cv::gpu::DeviceInfo().major() >= 2)
+            if (cv::gpu::TargetArchs::builtWith(cv::gpu::COMPUTE_20) && cv::gpu::DeviceInfo().major() >= 2)
                 fs.open(std::string(ts->get_data_path()) + "meanshift/spmap.yaml", cv::FileStorage::READ);
             else
                 fs.open(std::string(ts->get_data_path()) + "meanshift/spmap_CC1X.yaml", cv::FileStorage::READ);
diff --git a/tests/gpu/src/mssegmentation.cpp b/tests/gpu/src/mssegmentation.cpp
index ea09b01..15a72f3 100644
--- a/tests/gpu/src/mssegmentation.cpp
+++ b/tests/gpu/src/mssegmentation.cpp
@@ -69,7 +69,7 @@ struct CV_GpuMeanShiftSegmentationTest : public CvTest {
             {
                 stringstream path;
                 path << ts->get_data_path() << "meanshift/cones_segmented_sp10_sr10_minsize" << minsize;
-                if (TargetArchs::hasEqualOrGreater(2, 0) && DeviceInfo().major() >= 2)
+                if (TargetArchs::builtWith(COMPUTE_20) && DeviceInfo().major() >= 2)
                     path << ".png";
                 else
                     path << "_CC1X.png";