Merge pull request #21024 from anna-khakimova:ak/simd_mul

alalek · alalek · commit 4b6047e746f0 · 2021-11-17T18:41:52.000Z
diff --git a/modules/gapi/perf/common/gapi_core_perf_tests.hpp b/modules/gapi/perf/common/gapi_core_perf_tests.hpp
@@ -32,7 +32,7 @@ namespace opencv_test
     class SubPerfTest : public TestPerfParams<tuple<cv::Size, MatType, int, cv::GCompileArgs>> {};
     class SubCPerfTest : public TestPerfParams<tuple<cv::Size, MatType, int, cv::GCompileArgs>> {};
     class SubRCPerfTest : public TestPerfParams<tuple<cv::Size, MatType, int, cv::GCompileArgs>> {};
-    class MulPerfTest : public TestPerfParams<tuple<cv::Size, MatType, int, cv::GCompileArgs>> {};
+    class MulPerfTest : public TestPerfParams<tuple<compare_f, cv::Size, MatType, int, double, cv::GCompileArgs>> {};
     class MulDoublePerfTest : public TestPerfParams<tuple<cv::Size, MatType, int, cv::GCompileArgs>> {};
     class MulCPerfTest : public TestPerfParams<tuple<cv::Size, MatType, int, cv::GCompileArgs>> {};
     class DivPerfTest : public TestPerfParams<tuple<compare_f, cv::Size, MatType, int, double, cv::GCompileArgs>> {};
diff --git a/modules/gapi/perf/common/gapi_core_perf_tests_inl.hpp b/modules/gapi/perf/common/gapi_core_perf_tests_inl.hpp
@@ -208,19 +208,23 @@ PERF_TEST_P_(SubRCPerfTest, TestPerformance)
 
 PERF_TEST_P_(MulPerfTest, TestPerformance)
 {
-    Size sz = get<0>(GetParam());
-    MatType type = get<1>(GetParam());
-    int dtype = get<2>(GetParam());
-    cv::GCompileArgs compile_args = get<3>(GetParam());
+    compare_f cmpF;
+    cv::Size sz;
+    MatType type = -1;
+    int dtype = -1;
+    double scale = 1.0;
+    cv::GCompileArgs compile_args;
+
+    std::tie(cmpF, sz, type, dtype, scale, compile_args) = GetParam();
 
     initMatsRandU(type, sz, dtype, false);
 
     // OpenCV code ///////////////////////////////////////////////////////////
-    cv::multiply(in_mat1, in_mat2, out_mat_ocv, 1.0, dtype);
+    cv::multiply(in_mat1, in_mat2, out_mat_ocv, scale, dtype);
 
     // G-API code ////////////////////////////////////////////////////////////
     cv::GMat in1, in2, out;
-    out = cv::gapi::mul(in1, in2, 1.0, dtype);
+    out = cv::gapi::mul(in1, in2, scale, dtype);
     cv::GComputation c(GIn(in1, in2), GOut(out));
 
     // Warm-up graph engine:
@@ -234,8 +238,9 @@ PERF_TEST_P_(MulPerfTest, TestPerformance)
     }
 
     // Comparison ////////////////////////////////////////////////////////////
-    // FIXIT unrealiable check: EXPECT_EQ(0, cv::countNonZero(out_mat_gapi != out_mat_ocv));
-    EXPECT_EQ(out_mat_gapi.size(), sz);
+    {
+        EXPECT_TRUE(cmpF(out_mat_gapi, out_mat_ocv));
+    }
 
     SANITY_CHECK_NOTHING();
 }
diff --git a/modules/gapi/perf/cpu/gapi_core_perf_tests_cpu.cpp b/modules/gapi/perf/cpu/gapi_core_perf_tests_cpu.cpp
@@ -46,9 +46,11 @@ INSTANTIATE_TEST_CASE_P(SubRCPerfTestCPU, SubRCPerfTest,
         Values(cv::compile_args(CORE_CPU))));
 
 INSTANTIATE_TEST_CASE_P(MulPerfTestCPU, MulPerfTest,
-    Combine(Values(szSmall128, szVGA, sz720p, sz1080p),
+    Combine(Values(AbsExact().to_compare_f()),
+        Values(szSmall128, szVGA, sz720p, sz1080p),
         Values(CV_8UC1, CV_8UC3, CV_16UC1, CV_16SC1, CV_32FC1),
         Values(-1, CV_8U, CV_16U, CV_32F),
+        Values(2.0),
         Values(cv::compile_args(CORE_CPU))));
 
 INSTANTIATE_TEST_CASE_P(MulDoublePerfTestCPU, MulDoublePerfTest,
diff --git a/modules/gapi/perf/cpu/gapi_core_perf_tests_fluid.cpp b/modules/gapi/perf/cpu/gapi_core_perf_tests_fluid.cpp
@@ -42,11 +42,13 @@ INSTANTIATE_TEST_CASE_P(SubPerfTestFluid, SubPerfTest,
 //         Values(-1, CV_8U, CV_16U, CV_32F),
 //         Values(cv::compile_args(CORE_FLUID))));
 
-// INSTANTIATE_TEST_CASE_P(MulPerfTestFluid, MulPerfTest,
-//     Combine(Values(szSmall128, szVGA, sz720p, sz1080p),
-//         Values(CV_8UC1, CV_8UC3, CV_16UC1, CV_16SC1, CV_32FC1),
-//         Values(-1, CV_8U, CV_16U, CV_32F),
-//         Values(cv::compile_args(CORE_FLUID))));
+ INSTANTIATE_TEST_CASE_P(MulPerfTestFluid, MulPerfTest,
+     Combine(Values(AbsExact().to_compare_f()),
+         Values(szSmall128, szVGA, sz720p, sz1080p),
+         Values(CV_8UC1, CV_8UC3, CV_16UC1, CV_16SC1, CV_32FC1),
+         Values(-1, CV_8U, CV_16U, CV_16S, CV_32F),
+         Values(2.0),
+         Values(cv::compile_args(CORE_FLUID))));
 
 // INSTANTIATE_TEST_CASE_P(MulDoublePerfTestFluid, MulDoublePerfTest,
 //     Combine(Values(szSmall128, szVGA, sz720p, sz1080p),
diff --git a/modules/gapi/perf/gpu/gapi_core_perf_tests_gpu.cpp b/modules/gapi/perf/gpu/gapi_core_perf_tests_gpu.cpp
@@ -44,9 +44,11 @@ INSTANTIATE_TEST_CASE_P(SubRCPerfTestGPU, SubRCPerfTest,
                                 Values(cv::compile_args(CORE_GPU))));
 
 INSTANTIATE_TEST_CASE_P(MulPerfTestGPU, MulPerfTest,
-                        Combine(Values( szSmall128, szVGA, sz720p, sz1080p ),
+                        Combine(Values(AbsExact().to_compare_f()),
+                                Values( szSmall128, szVGA, sz720p, sz1080p ),
                                 Values( CV_8UC1, CV_8UC3, CV_16UC1, CV_16SC1, CV_32FC1 ),
                                 Values( -1, CV_8U, CV_16U, CV_32F ),
+                                Values(2.0),
                                 Values(cv::compile_args(CORE_GPU))));
 
 INSTANTIATE_TEST_CASE_P(MulDoublePerfTestGPU, MulDoublePerfTest,
diff --git a/modules/gapi/src/backends/fluid/gfluidcore.cpp b/modules/gapi/src/backends/fluid/gfluidcore.cpp
@@ -684,9 +684,14 @@ static void run_arithm(Buffer &dst, const View &src1, const View &src2, Arithm a
             break;
         }
         case ARITHM_MULTIPLY:
+        {
+#if CV_SIMD
+            x = mul_simd(in1, in2, out, length, scale);
+#endif
             for (; x < length; ++x)
                 out[x] = mul<DST>(in1[x], in2[x], _scale);
             break;
+        }
         case ARITHM_DIVIDE:
         {
 #if CV_SIMD
@@ -745,13 +750,22 @@ GAPI_FLUID_KERNEL(GFluidMul, cv::gapi::core::GMul, false)
     static void run(const View &src1, const View &src2, double scale, int /*dtype*/, Buffer &dst)
     {
         //      DST     SRC1    SRC2    OP          __VA_ARGS__
-        BINARY_(uchar , uchar , uchar , run_arithm, dst, src1, src2, ARITHM_MULTIPLY, scale);
-        BINARY_(uchar ,  short,  short, run_arithm, dst, src1, src2, ARITHM_MULTIPLY, scale);
-        BINARY_(uchar ,  float,  float, run_arithm, dst, src1, src2, ARITHM_MULTIPLY, scale);
-        BINARY_( short,  short,  short, run_arithm, dst, src1, src2, ARITHM_MULTIPLY, scale);
-        BINARY_( float, uchar , uchar , run_arithm, dst, src1, src2, ARITHM_MULTIPLY, scale);
-        BINARY_( float,  short,  short, run_arithm, dst, src1, src2, ARITHM_MULTIPLY, scale);
-        BINARY_( float,  float,  float, run_arithm, dst, src1, src2, ARITHM_MULTIPLY, scale);
+        BINARY_(uchar,  uchar,  uchar,  run_arithm, dst, src1, src2, ARITHM_MULTIPLY, scale);
+        BINARY_(uchar,  ushort, ushort, run_arithm, dst, src1, src2, ARITHM_MULTIPLY, scale);
+        BINARY_(uchar,  short,  short,  run_arithm, dst, src1, src2, ARITHM_MULTIPLY, scale);
+        BINARY_(uchar,  float,  float,  run_arithm, dst, src1, src2, ARITHM_MULTIPLY, scale);
+        BINARY_(short,  short,  short,  run_arithm, dst, src1, src2, ARITHM_MULTIPLY, scale);
+        BINARY_(short,  ushort, ushort, run_arithm, dst, src1, src2, ARITHM_MULTIPLY, scale);
+        BINARY_(short,  uchar,  uchar,  run_arithm, dst, src1, src2, ARITHM_MULTIPLY, scale);
+        BINARY_(short,  float,  float,  run_arithm, dst, src1, src2, ARITHM_MULTIPLY, scale);
+        BINARY_(ushort, ushort, ushort, run_arithm, dst, src1, src2, ARITHM_MULTIPLY, scale);
+        BINARY_(ushort, uchar,  uchar,  run_arithm, dst, src1, src2, ARITHM_MULTIPLY, scale);
+        BINARY_(ushort, short,  short,  run_arithm, dst, src1, src2, ARITHM_MULTIPLY, scale);
+        BINARY_(ushort, float,  float,  run_arithm, dst, src1, src2, ARITHM_MULTIPLY, scale);
+        BINARY_(float,  uchar,  uchar,  run_arithm, dst, src1, src2, ARITHM_MULTIPLY, scale);
+        BINARY_(float,  ushort, ushort, run_arithm, dst, src1, src2, ARITHM_MULTIPLY, scale);
+        BINARY_(float,  short,  short,  run_arithm, dst, src1, src2, ARITHM_MULTIPLY, scale);
+        BINARY_(float,  float,  float,  run_arithm, dst, src1, src2, ARITHM_MULTIPLY, scale);
 
         CV_Error(cv::Error::StsBadArg, "unsupported combination of types");
     }
diff --git a/modules/gapi/src/backends/fluid/gfluidcore_func.dispatch.cpp b/modules/gapi/src/backends/fluid/gfluidcore_func.dispatch.cpp
@@ -56,6 +56,35 @@ DIV_SIMD(float, float)
 
 #undef DIV_SIMD
 
+
+#define MUL_SIMD(SRC, DST)                                                  \
+int mul_simd(const SRC in1[], const SRC in2[], DST out[],                   \
+             const int length, double _scale)                               \
+{                                                                           \
+    CV_CPU_DISPATCH(mul_simd, (in1, in2, out, length, _scale),              \
+                    CV_CPU_DISPATCH_MODES_ALL);                             \
+}
+
+
+MUL_SIMD(uchar, uchar)
+MUL_SIMD(ushort, uchar)
+MUL_SIMD(short, uchar)
+MUL_SIMD(float, uchar)
+MUL_SIMD(short, short)
+MUL_SIMD(ushort, short)
+MUL_SIMD(uchar, short)
+MUL_SIMD(float, short)
+MUL_SIMD(ushort, ushort)
+MUL_SIMD(uchar, ushort)
+MUL_SIMD(short, ushort)
+MUL_SIMD(float, ushort)
+MUL_SIMD(uchar, float)
+MUL_SIMD(ushort, float)
+MUL_SIMD(short, float)
+MUL_SIMD(float, float)
+
+#undef MUL_SIMD
+
 } // namespace fluid
 } // namespace gapi
 } // namespace cv
diff --git a/modules/gapi/src/backends/fluid/gfluidcore_func.hpp b/modules/gapi/src/backends/fluid/gfluidcore_func.hpp
@@ -37,6 +37,29 @@ DIV_SIMD(float, float)
 
 #undef DIV_SIMD
 
+#define MUL_SIMD(SRC, DST)                                       \
+int mul_simd(const SRC in1[], const SRC in2[], DST out[],        \
+             const int length, double _scale);
+
+MUL_SIMD(uchar, uchar)
+MUL_SIMD(ushort, uchar)
+MUL_SIMD(short, uchar)
+MUL_SIMD(float, uchar)
+MUL_SIMD(short, short)
+MUL_SIMD(ushort, short)
+MUL_SIMD(uchar, short)
+MUL_SIMD(float, short)
+MUL_SIMD(ushort, ushort)
+MUL_SIMD(uchar, ushort)
+MUL_SIMD(short, ushort)
+MUL_SIMD(float, ushort)
+MUL_SIMD(uchar, float)
+MUL_SIMD(ushort, float)
+MUL_SIMD(short, float)
+MUL_SIMD(float, float)
+
+#undef MUL_SIMD
+
 }  // namespace fluid
 }  // namespace gapi
 }  // namespace cv
diff --git a/modules/gapi/src/backends/fluid/gfluidcore_func.simd.hpp b/modules/gapi/src/backends/fluid/gfluidcore_func.simd.hpp
