Merge pull request opencv#23929 from CNClareChen:4.x

* Optimize some function with lasx.

Optimize some function with lasx. opencv#23929

This patch optimizes some lasx functions and reduces the runtime of opencv_test_core from 662,238ms to 633603ms on the 3A5000 platform.

### Pull Request Readiness Checklist

See details at https://github.com/opencv/opencv/wiki/How_to_contribute#making-a-good-pull-request

- [x] I agree to contribute to the project under Apache 2 License.
- [x] To the best of my knowledge, the proposed patch is not based on a code under GPL or another license that is incompatible with OpenCV
- [x] The PR is proposed to the proper branch
- [x] There is a reference to the original bug report and related work
- [x] There is accuracy test, performance test and test data in opencv_extra repository, if applicable
      Patch to opencv_extra has the same branch name.
- [x] The feature is well documented and sample code can be built with the project CMake

Author: CNClareChen

cmake/OpenCVCompilerOptimizations.cmake

@@ -53,6 +53,7 @@ list(APPEND CPU_ALL_OPTIMIZATIONS NEON VFPV3 FP16 NEON_DOTPROD)
 list(APPEND CPU_ALL_OPTIMIZATIONS MSA)
 list(APPEND CPU_ALL_OPTIMIZATIONS VSX VSX3)
 list(APPEND CPU_ALL_OPTIMIZATIONS RVV)
+list(APPEND CPU_ALL_OPTIMIZATIONS LSX)
 list(APPEND CPU_ALL_OPTIMIZATIONS LASX)
 list(REMOVE_DUPLICATES CPU_ALL_OPTIMIZATIONS)
 
@@ -397,10 +398,16 @@ elseif(RISCV)
   set(CPU_BASELINE "DETECT" CACHE STRING "${HELP_CPU_BASELINE}")
 
 elseif(LOONGARCH64)
+  ocv_update(CPU_LSX_TEST_FILE "${OpenCV_SOURCE_DIR}/cmake/checks/cpu_lsx.cpp")
   ocv_update(CPU_LASX_TEST_FILE "${OpenCV_SOURCE_DIR}/cmake/checks/cpu_lasx.cpp")
-  ocv_update(CPU_KNOWN_OPTIMIZATIONS "LASX")
+  ocv_update(CPU_KNOWN_OPTIMIZATIONS "LSX;LASX")
+  ocv_update(CPU_LSX_FLAGS_ON "-mlsx")
   ocv_update(CPU_LASX_FLAGS_ON "-mlasx")
-  set(CPU_BASELINE "LASX" CACHE STRING "${HELP_CPU_BASELINE}")
+  if("${CPU_BASELINE_DISABLE}" STREQUAL "LASX")
+    set(CPU_BASELINE "LSX" CACHE  STRING "${HELP_CPU_BASELINE}")
+  else()
+    set(CPU_BASELINE "LASX" CACHE STRING "${HELP_CPU_BASELINE}")
+  endif()
 
 endif()
 

cmake/checks/cpu_lsx.cpp

@@ -0,0 +1,15 @@
+#include <stdio.h>
+#include <lsxintrin.h>
+
+int test()
+{
+    const float src[] = { 0.0f, 1.0f, 2.0f, 3.0f};
+    v4f32 val = (v4f32)__lsx_vld((const float*)(src), 0);
+    return __lsx_vpickve2gr_w(__lsx_vftint_w_s(val), 3);
+}
+
+int main()
+{
+  printf("%d\n", test());
+  return 0;
+}

modules/core/include/opencv2/core/cv_cpu_dispatch.h

@@ -172,6 +172,11 @@
 #  define CV_MSA 1
 #endif
 
+#ifdef CV_CPU_COMPILE_LSX
+#  include <lsxintrin.h>
+#  define CV_LSX 1
+#endif
+
 #ifdef CV_CPU_COMPILE_LASX
 #  include <lasxintrin.h>
 #  define CV_LASX 1
@@ -376,6 +381,10 @@ struct VZeroUpperGuard {
 #  define CV_RVV 0
 #endif
 
+#ifndef CV_LSX
+#  define CV_LSX 0
+#endif
+
 #ifndef CV_LASX
 #  define CV_LASX 0
 #endif

modules/core/include/opencv2/core/cv_cpu_helper.h

@@ -525,6 +525,27 @@
 #endif
 #define __CV_CPU_DISPATCH_CHAIN_RVV(fn, args, mode, ...)  CV_CPU_CALL_RVV(fn, args); __CV_EXPAND(__CV_CPU_DISPATCH_CHAIN_ ## mode(fn, args, __VA_ARGS__))
 
+#if !defined CV_DISABLE_OPTIMIZATION && defined CV_ENABLE_INTRINSICS && defined CV_CPU_COMPILE_LSX
+#  define CV_TRY_LSX 1
+#  define CV_CPU_FORCE_LSX 1
+#  define CV_CPU_HAS_SUPPORT_LSX 1
+#  define CV_CPU_CALL_LSX(fn, args) return (cpu_baseline::fn args)
+#  define CV_CPU_CALL_LSX_(fn, args) return (opt_LSX::fn args)
+#elif !defined CV_DISABLE_OPTIMIZATION && defined CV_ENABLE_INTRINSICS && defined CV_CPU_DISPATCH_COMPILE_LSX
+#  define CV_TRY_LSX 1
+#  define CV_CPU_FORCE_LSX 0
+#  define CV_CPU_HAS_SUPPORT_LSX (cv::checkHardwareSupport(CV_CPU_LSX))
+#  define CV_CPU_CALL_LSX(fn, args) if (CV_CPU_HAS_SUPPORT_LSX) return (opt_LSX::fn args)
+#  define CV_CPU_CALL_LSX_(fn, args) if (CV_CPU_HAS_SUPPORT_LSX) return (opt_LSX::fn args)
+#else
+#  define CV_TRY_LSX 0
+#  define CV_CPU_FORCE_LSX 0
+#  define CV_CPU_HAS_SUPPORT_LSX 0
+#  define CV_CPU_CALL_LSX(fn, args)
+#  define CV_CPU_CALL_LSX_(fn, args)
+#endif
+#define __CV_CPU_DISPATCH_CHAIN_LSX(fn, args, mode, ...)  CV_CPU_CALL_LSX(fn, args); __CV_EXPAND(__CV_CPU_DISPATCH_CHAIN_ ## mode(fn, args, __VA_ARGS__))
+
 #if !defined CV_DISABLE_OPTIMIZATION && defined CV_ENABLE_INTRINSICS && defined CV_CPU_COMPILE_LASX
 #  define CV_TRY_LASX 1
 #  define CV_CPU_FORCE_LASX 1

modules/core/include/opencv2/core/cvdef.h

@@ -281,7 +281,8 @@ namespace cv {
 
 #define CV_CPU_RVV              210
 
-#define CV_CPU_LASX             230
+#define CV_CPU_LSX              230
+#define CV_CPU_LASX             231
 
 // CPU features groups
 #define CV_CPU_AVX512_SKX       256
@@ -342,7 +343,8 @@ enum CpuFeatures {
 
     CPU_RVV             = 210,
 
-    CPU_LASX             = 230,
+    CPU_LSX             = 230,
+    CPU_LASX            = 231,
 
     CPU_AVX512_SKX      = 256, //!< Skylake-X with AVX-512F/CD/BW/DQ/VL
     CPU_AVX512_COMMON   = 257, //!< Common instructions AVX-512F/CD for all CPUs that support AVX-512

modules/core/include/opencv2/core/hal/intrin.hpp

@@ -206,7 +206,7 @@ using namespace CV_CPU_OPTIMIZATION_HAL_NAMESPACE;
 #   undef CV_RVV
 #endif
 
-#if (CV_SSE2 || CV_NEON || CV_VSX || CV_MSA || CV_WASM_SIMD || CV_RVV071) && !defined(CV_FORCE_SIMD128_CPP)
+#if (CV_SSE2 || CV_NEON || CV_VSX || CV_MSA || CV_WASM_SIMD || CV_RVV071 || CV_LSX) && !defined(CV_FORCE_SIMD128_CPP)
 #define CV__SIMD_FORWARD 128
 #include "opencv2/core/hal/intrin_forward.hpp"
 #endif
@@ -242,6 +242,10 @@ using namespace CV_CPU_OPTIMIZATION_HAL_NAMESPACE;
 #include "opencv2/core/hal/intrin_rvv.hpp"
 #endif
 
+#elif CV_LSX && !defined(CV_FORCE_SIMD128_CPP)
+
+#include "opencv2/core/hal/intrin_lsx.hpp"
+
 #elif CV_LASX
     #if !defined(CV_FORCE_SIMD128_CPP)
     #define CV_FORCE_SIMD128_CPP 1