core(test): intrinsic tests for all dispatched CPU optimizations

- tests for both SIMD128 / SIMD256 - different dispatched + baseline(SIMD128) intrinsics
BRAINSia · Aug 1, 2018 · 3f302ca · 3f302ca
1 parent 67f79aa
commit 3f302ca
Show file tree

Hide file tree

Showing 12 changed files with 527 additions and 426 deletions.
diff --git a/cmake/OpenCVCompilerOptimizations.cmake b/cmake/OpenCVCompilerOptimizations.cmake
@@ -761,24 +761,24 @@ macro(ocv_compiler_optimization_fill_cpu_config)
   endif()
 endmacro()
 
-macro(ocv_add_dispatched_file filename)
+macro(__ocv_add_dispatched_file filename target_src_var src_directory dst_directory precomp_hpp optimizations_var)
   if(NOT OPENCV_INITIAL_PASS)
     set(__codestr "
-#include \"${CMAKE_CURRENT_LIST_DIR}/src/precomp.hpp\"
-#include \"${CMAKE_CURRENT_LIST_DIR}/src/${filename}.simd.hpp\"
+#include \"${src_directory}/${precomp_hpp}\"
+#include \"${src_directory}/${filename}.simd.hpp\"
 ")
 
-    set(__declarations_str "#define CV_CPU_SIMD_FILENAME \"${CMAKE_CURRENT_LIST_DIR}/src/${filename}.simd.hpp\"")
+    set(__declarations_str "#define CV_CPU_SIMD_FILENAME \"${src_directory}/${filename}.simd.hpp\"")
     set(__dispatch_modes "BASELINE")
 
-    set(__optimizations "${ARGN}")
+    set(__optimizations "${${optimizations_var}}")
     if(CV_DISABLE_OPTIMIZATION OR NOT CV_ENABLE_INTRINSICS)
       set(__optimizations "")
     endif()
 
     foreach(OPT ${__optimizations})
       string(TOLOWER "${OPT}" OPT_LOWER)
-      set(__file "${CMAKE_CURRENT_BINARY_DIR}/${filename}.${OPT_LOWER}.cpp")
+      set(__file "${CMAKE_CURRENT_BINARY_DIR}/${dst_directory}${filename}.${OPT_LOWER}.cpp")
       if(EXISTS "${__file}")
         file(READ "${__file}" __content)
       else()
@@ -791,7 +791,11 @@ macro(ocv_add_dispatched_file filename)
       endif()
 
       if(";${CPU_DISPATCH};" MATCHES "${OPT}" OR __CPU_DISPATCH_INCLUDE_ALL)
-        list(APPEND OPENCV_MODULE_${the_module}_SOURCES_DISPATCHED "${__file}")
+        if(EXISTS "${src_directory}/${filename}.${OPT_LOWER}.cpp")
+          message(STATUS "Using overrided ${OPT} source: ${src_directory}/${filename}.${OPT_LOWER}.cpp")
+        else()
+          list(APPEND ${target_src_var} "${__file}")
+        endif()
       endif()
 
       set(__declarations_str "${__declarations_str}
@@ -803,9 +807,11 @@ macro(ocv_add_dispatched_file filename)
 
     set(__declarations_str "${__declarations_str}
 #define CV_CPU_DISPATCH_MODES_ALL ${__dispatch_modes}
+
+#undef CV_CPU_SIMD_FILENAME
 ")
 
-    set(__file "${CMAKE_CURRENT_BINARY_DIR}/${filename}.simd_declarations.hpp")
+    set(__file "${CMAKE_CURRENT_BINARY_DIR}/${dst_directory}${filename}.simd_declarations.hpp")
     if(EXISTS "${__file}")
       file(READ "${__file}" __content)
     endif()
@@ -817,6 +823,17 @@ macro(ocv_add_dispatched_file filename)
   endif()
 endmacro()
 
+macro(ocv_add_dispatched_file filename)
+  set(__optimizations "${ARGN}")
+  if(" ${ARGV1}" STREQUAL " TEST")
+    list(REMOVE_AT __optimizations 0)
+    __ocv_add_dispatched_file("${filename}" "OPENCV_MODULE_${the_module}_TEST_SOURCES_DISPATCHED" "${CMAKE_CURRENT_LIST_DIR}/test" "test/" "test_precomp.hpp" __optimizations)
+  else()
+    __ocv_add_dispatched_file("${filename}" "OPENCV_MODULE_${the_module}_SOURCES_DISPATCHED" "${CMAKE_CURRENT_LIST_DIR}/src" "" "precomp.hpp" __optimizations)
+  endif()
+endmacro()
+
+
 # Workaround to support code which always require all code paths
 macro(ocv_add_dispatched_file_force_all)
   set(__CPU_DISPATCH_INCLUDE_ALL 1)

diff --git a/cmake/OpenCVModule.cmake b/cmake/OpenCVModule.cmake
@@ -1202,6 +1202,9 @@ function(ocv_add_accuracy_tests)
         set(OPENCV_TEST_${the_module}_SOURCES ${test_srcs} ${test_hdrs})
       endif()
 
+      if(OPENCV_MODULE_${the_module}_TEST_SOURCES_DISPATCHED)
+        list(APPEND OPENCV_TEST_${the_module}_SOURCES ${OPENCV_MODULE_${the_module}_TEST_SOURCES_DISPATCHED})
+      endif()
       ocv_compiler_optimization_process_sources(OPENCV_TEST_${the_module}_SOURCES OPENCV_TEST_${the_module}_DEPS ${the_target})
 
       if(NOT BUILD_opencv_world)
@@ -1211,6 +1214,9 @@ function(ocv_add_accuracy_tests)
       source_group("Src" FILES "${${the_target}_pch}")
       ocv_add_executable(${the_target} ${OPENCV_TEST_${the_module}_SOURCES} ${${the_target}_pch})
       ocv_target_include_modules(${the_target} ${test_deps} "${test_path}")
+      if(EXISTS "${CMAKE_CURRENT_BINARY_DIR}/test")
+        ocv_target_include_directories(${the_target} "${CMAKE_CURRENT_BINARY_DIR}/test")
+      endif()
       ocv_target_link_libraries(${the_target} LINK_PRIVATE ${test_deps} ${OPENCV_MODULE_${the_module}_DEPS} ${OPENCV_LINKER_LIBS} ${OPENCV_TEST_${the_module}_DEPS})
       add_dependencies(opencv_tests ${the_target})
 

diff --git a/modules/core/CMakeLists.txt b/modules/core/CMakeLists.txt
@@ -3,6 +3,10 @@ set(the_description "The Core Functionality")
 ocv_add_dispatched_file(mathfuncs_core SSE2 AVX AVX2)
 ocv_add_dispatched_file(stat SSE4_2 AVX2)
 
+# dispatching for accuracy tests
+ocv_add_dispatched_file_force_all(test_intrin128 TEST SSE2 SSE3 SSSE3 SSE4_1 SSE4_2 AVX FP16 AVX2)
+ocv_add_dispatched_file_force_all(test_intrin256 TEST AVX2)
+
 ocv_add_module(core
                OPTIONAL opencv_cudev
                WRAP java python js)

diff --git a/modules/core/include/opencv2/core/hal/intrin.hpp b/modules/core/include/opencv2/core/hal/intrin.hpp
@@ -204,20 +204,6 @@ CV_CPU_OPTIMIZATION_HAL_NAMESPACE_BEGIN
 #define CV_SIMD512_64F 0
 #endif
 
-#if CV_SIMD512
-    #define CV_SIMD 1
-    #define CV_SIMD_64F CV_SIMD512_64F
-    #define CV_SIMD_WIDTH 64
-#elif CV_SIMD256
-    #define CV_SIMD 1
-    #define CV_SIMD_64F CV_SIMD256_64F
-    #define CV_SIMD_WIDTH 32
-#else
-    #define CV_SIMD CV_SIMD128
-    #define CV_SIMD_64F CV_SIMD128_64F
-    #define CV_SIMD_WIDTH 16
-#endif
-
 //==================================================================================================
 
 #define CV_INTRIN_DEFINE_WIDE_INTRIN(typ, vtyp, short_typ, prefix, loadsfx) \
@@ -309,7 +295,15 @@ template<typename _Tp> struct V_RegTraits
 #endif
 #endif
 
-#if CV_SIMD256
+#if CV_SIMD512 && (!defined(CV__SIMD_FORCE_WIDTH) || CV__SIMD_FORCE_WIDTH == 512)
+    #define CV_SIMD 1
+    #define CV_SIMD_64F CV_SIMD512_64F
+    #define CV_SIMD_WIDTH 64
+    // TODO typedef v_uint8 / v_int32 / etc types here
+#elif CV_SIMD256 && (!defined(CV__SIMD_FORCE_WIDTH) || CV__SIMD_FORCE_WIDTH == 256)
+    #define CV_SIMD 1
+    #define CV_SIMD_64F CV_SIMD256_64F
+    #define CV_SIMD_WIDTH 32
     typedef v_uint8x32   v_uint8;
     typedef v_int8x32    v_int8;
     typedef v_uint16x16  v_uint16;
@@ -329,7 +323,10 @@ template<typename _Tp> struct V_RegTraits
     CV_INTRIN_DEFINE_WIDE_INTRIN_ALL_TYPES(v256)
     CV_INTRIN_DEFINE_WIDE_INTRIN(double, v_float64, f64, v256, load)
     inline void vx_cleanup() { v256_cleanup(); }
-#elif CV_SIMD128 || CV_SIMD128_CPP
+#elif (CV_SIMD128 || CV_SIMD128_CPP) && (!defined(CV__SIMD_FORCE_WIDTH) || CV__SIMD_FORCE_WIDTH == 128)
+    #define CV_SIMD CV_SIMD128
+    #define CV_SIMD_64F CV_SIMD128_64F
+    #define CV_SIMD_WIDTH 16
     typedef v_uint8x16  v_uint8;
     typedef v_int8x16   v_int8;
     typedef v_uint16x8  v_uint16;
@@ -380,6 +377,14 @@ inline unsigned int trailingZeros32(unsigned int value) {
 CV_CPU_OPTIMIZATION_HAL_NAMESPACE_END
 #endif
 
+#ifndef CV_SIMD_64F
+#define CV_SIMD_64F 0
+#endif
+
+#ifndef CV_SIMD
+#define CV_SIMD 0
+#endif
+
 } // cv::
 
 //! @endcond

diff --git a/modules/core/test/test_intrin.avx2.cpp b/modules/core/test/test_intrin.avx2.cpp
diff --git a/modules/core/test/test_intrin.cpp b/modules/core/test/test_intrin.cpp
@@ -2,101 +2,100 @@
 // It is subject to the license terms in the LICENSE file found in the top-level directory
 // of this distribution and at http://opencv.org/license.html.
 #include "test_precomp.hpp"
-#include "test_intrin.simd.hpp"
 
-#define CV_CPU_SIMD_FILENAME "test_intrin.simd.hpp"
-#define CV_CPU_DISPATCH_MODE FP16
-#include "opencv2/core/private/cv_cpu_include_simd_declarations.hpp"
+#include "test_intrin128.simd.hpp"
+#include "test_intrin128.simd_declarations.hpp"
 
-#define CV_CPU_DISPATCH_MODE AVX2
-#include "opencv2/core/private/cv_cpu_include_simd_declarations.hpp"
+#undef CV_CPU_DISPATCH_MODES_ALL
 
-namespace opencv_test { namespace hal {
-using namespace CV_CPU_OPTIMIZATION_NAMESPACE;
-
-TEST(hal_intrin, uint8x16)
-{ test_hal_intrin_uint8(); }
-
-TEST(hal_intrin, int8x16)
-{ test_hal_intrin_int8(); }
-
-TEST(hal_intrin, uint16x8)
-{ test_hal_intrin_uint16(); }
-
-TEST(hal_intrin, int16x8)
-{ test_hal_intrin_int16(); }
-
-TEST(hal_intrin, int32x4)
-{ test_hal_intrin_int32(); }
-
-TEST(hal_intrin, uint32x4)
-{ test_hal_intrin_uint32(); }
-
-TEST(hal_intrin, uint64x2)
-{ test_hal_intrin_uint64(); }
-
-TEST(hal_intrin, int64x2)
-{ test_hal_intrin_int64(); }
+#include "opencv2/core/cv_cpu_dispatch.h"
+#include "test_intrin256.simd.hpp"
+#include "test_intrin256.simd_declarations.hpp"
 
-TEST(hal_intrin, float32x4)
-{ test_hal_intrin_float32(); }
 
-TEST(hal_intrin, float64x2)
-{ test_hal_intrin_float64(); }
+namespace opencv_test { namespace hal {
 
-TEST(hal_intrin, float16x8)
+#define CV_CPU_CALL_BASELINE_(fn, args)  CV_CPU_CALL_BASELINE(fn, args)
+
+#define DISPATCH_SIMD128(fn, cpu_opt) do { \
+    CV_CPU_CALL_ ## cpu_opt ## _(fn, ()); \
+    throw SkipTestException("SIMD128 (" #cpu_opt ") is not available or disabled"); \
+} while(0)
+
+#define DISPATCH_SIMD256(fn, cpu_opt) do { \
+    CV_CPU_CALL_ ## cpu_opt ## _(fn, ()); \
+    throw SkipTestException("SIMD256 (" #cpu_opt ") is not available or disabled"); \
+} while(0)
+
+#define DEFINE_SIMD_TESTS(simd_size, cpu_opt) \
+TEST(hal_intrin ## simd_size, uint8x16_ ## cpu_opt)  { DISPATCH_SIMD ## simd_size(test_hal_intrin_uint8, cpu_opt); } \
+TEST(hal_intrin ## simd_size, int8x16_ ## cpu_opt)   { DISPATCH_SIMD ## simd_size(test_hal_intrin_int8, cpu_opt); } \
+TEST(hal_intrin ## simd_size, uint16x8_ ## cpu_opt)  { DISPATCH_SIMD ## simd_size(test_hal_intrin_uint16, cpu_opt); } \
+TEST(hal_intrin ## simd_size, int16x8_ ## cpu_opt)   { DISPATCH_SIMD ## simd_size(test_hal_intrin_int16, cpu_opt); } \
+TEST(hal_intrin ## simd_size, int32x4_ ## cpu_opt)   { DISPATCH_SIMD ## simd_size(test_hal_intrin_int32, cpu_opt); } \
+TEST(hal_intrin ## simd_size, uint32x4_ ## cpu_opt)  { DISPATCH_SIMD ## simd_size(test_hal_intrin_uint32, cpu_opt); } \
+TEST(hal_intrin ## simd_size, uint64x2_ ## cpu_opt)  { DISPATCH_SIMD ## simd_size(test_hal_intrin_uint64, cpu_opt); } \
+TEST(hal_intrin ## simd_size, int64x2_ ## cpu_opt)   { DISPATCH_SIMD ## simd_size(test_hal_intrin_int64, cpu_opt); } \
+TEST(hal_intrin ## simd_size, float32x4_ ## cpu_opt) { DISPATCH_SIMD ## simd_size(test_hal_intrin_float32, cpu_opt); } \
+TEST(hal_intrin ## simd_size, float64x2_ ## cpu_opt) { DISPATCH_SIMD ## simd_size(test_hal_intrin_float64, cpu_opt); } \
+
+namespace intrin128 {
+
+DEFINE_SIMD_TESTS(128, BASELINE)
+
+#if defined CV_CPU_DISPATCH_COMPILE_SSE2 || defined CV_CPU_BASELINE_COMPILE_SSE2
+DEFINE_SIMD_TESTS(128, SSE2)
+#endif
+#if defined CV_CPU_DISPATCH_COMPILE_SSE3 || defined CV_CPU_BASELINE_COMPILE_SSE3
+DEFINE_SIMD_TESTS(128, SSE3)
+#endif
+#if defined CV_CPU_DISPATCH_COMPILE_SSSE3 || defined CV_CPU_BASELINE_COMPILE_SSSE3
+DEFINE_SIMD_TESTS(128, SSSE3)
+#endif
+#if defined CV_CPU_DISPATCH_COMPILE_SSE4_1 || defined CV_CPU_BASELINE_COMPILE_SSE4_1
+DEFINE_SIMD_TESTS(128, SSE4_1)
+#endif
+#if defined CV_CPU_DISPATCH_COMPILE_SSE4_2 || defined CV_CPU_BASELINE_COMPILE_SSE4_2
+DEFINE_SIMD_TESTS(128, SSE4_2)
+#endif
+#if defined CV_CPU_DISPATCH_COMPILE_AVX || defined CV_CPU_BASELINE_COMPILE_AVX
+DEFINE_SIMD_TESTS(128, AVX)
+#endif
+#if defined CV_CPU_DISPATCH_COMPILE_AVX2 || defined CV_CPU_BASELINE_COMPILE_AVX2
+DEFINE_SIMD_TESTS(128, AVX2)
+#endif
+
+TEST(hal_intrin128, float16x8_FP16)
 {
     CV_CPU_CALL_FP16_(test_hal_intrin_float16, ());
     throw SkipTestException("Unsupported hardware: FP16 is not available");
 }
 
-#define DISPATCH_SIMD_MODES AVX2
-#define DISPATCH_SIMD_NAME "SIMD256"
-#define DISPATCH_SIMD(fun)                              \
-    do {                                                \
-        CV_CPU_DISPATCH(fun, (), DISPATCH_SIMD_MODES);  \
-        throw SkipTestException(                        \
-            "Unsupported hardware: "                    \
-            DISPATCH_SIMD_NAME                          \
-            " is not available"                         \
-        );                                              \
-    } while(0)
-
-TEST(hal_intrin256, uint8x32)
-{ DISPATCH_SIMD(test_hal_intrin_uint8); }
-
-TEST(hal_intrin256, int8x32)
-{ DISPATCH_SIMD(test_hal_intrin_int8); }
-
-TEST(hal_intrin256, uint16x16)
-{ DISPATCH_SIMD(test_hal_intrin_uint16); }
+} // namespace intrin128
 
-TEST(hal_intrin256, int16x16)
-{ DISPATCH_SIMD(test_hal_intrin_int16); }
 
-TEST(hal_intrin256, uint32x8)
-{ DISPATCH_SIMD(test_hal_intrin_uint32); }
+namespace intrin256 {
 
-TEST(hal_intrin256, int32x8)
-{ DISPATCH_SIMD(test_hal_intrin_int32); }
 
-TEST(hal_intrin256, uint64x4)
-{ DISPATCH_SIMD(test_hal_intrin_uint64); }
+// Not available due missing C++ backend for SIMD256
+//DEFINE_SIMD_TESTS(256, BASELINE)
 
-TEST(hal_intrin256, int64x4)
-{ DISPATCH_SIMD(test_hal_intrin_int64); }
+//#if defined CV_CPU_DISPATCH_COMPILE_AVX
+//DEFINE_SIMD_TESTS(256, AVX)
+//#endif
 
-TEST(hal_intrin256, float32x8)
-{ DISPATCH_SIMD(test_hal_intrin_float32); }
+#if defined CV_CPU_DISPATCH_COMPILE_AVX2 || defined CV_CPU_BASELINE_COMPILE_AVX2
+DEFINE_SIMD_TESTS(256, AVX2)
+#endif
 
-TEST(hal_intrin256, float64x4)
-{ DISPATCH_SIMD(test_hal_intrin_float64); }
-
-TEST(hal_intrin256, float16x16)
+TEST(hal_intrin256, float16x16_FP16)
 {
-    if (!CV_CPU_HAS_SUPPORT_FP16)
-        throw SkipTestException("Unsupported hardware: FP16 is not available");
-    DISPATCH_SIMD(test_hal_intrin_float16);
+    //CV_CPU_CALL_FP16_(test_hal_intrin_float16, ());
+    CV_CPU_CALL_AVX2_(test_hal_intrin_float16, ());
+    throw SkipTestException("Unsupported hardware: FP16 is not available");
 }
 
+
+} // namespace intrin256
+
 }} // namespace
diff --git a/modules/core/test/test_intrin.fp16.cpp b/modules/core/test/test_intrin.fp16.cpp