+add SSE4.1 optimizations of function BgrToYuv444pV2.

ermig1979 · Sep 19, 2023 · e2e3a2f · e2e3a2f
1 parent 1ac70e2
commit e2e3a2f
Show file tree

Hide file tree

Showing 5 changed files with 65 additions and 11 deletions.
diff --git a/docs/2023.html b/docs/2023.html
@@ -39,7 +39,7 @@ <h5>New features</h5>
 <ul>
  <li>Base implementation, SSE4.1, AVX2, AVX-512BW optimizations of function BgrToYuv420pV2.</li>
  <li>Base implementation, SSE4.1, AVX2 optimizations of function BgrToYuv422pV2.</li>
- <li>Base implementation of function BgrToYuv444pV2.</li>
+ <li>Base implementation, SSE4.1 optimizations of function BgrToYuv444pV2.</li>
 </ul>
 <ul>
  <li>Error in AVX-512BW optimizations of function SynetSoftmaxLayerForward.</li>

diff --git a/src/Simd/SimdLib.cpp b/src/Simd/SimdLib.cpp
@@ -1722,11 +1722,11 @@ SIMD_API void SimdBgrToYuv444pV2(const uint8_t* bgr, size_t bgrStride, size_t wi
 //        Avx2::BgrToYuv444pV2(bgr, bgrStride, width, height, y, yStride, u, uStride, v, vStride, yuvType);
 //    else
 //#endif
-//#ifdef SIMD_SSE41_ENABLE
-//    if (Sse41::Enable && width >= Sse41::A)
-//        Sse41::BgrToYuv444pV2(bgr, bgrStride, width, height, y, yStride, u, uStride, v, vStride, yuvType);
-//    else
-//#endif
+#ifdef SIMD_SSE41_ENABLE
+    if (Sse41::Enable && width >= Sse41::A)
+        Sse41::BgrToYuv444pV2(bgr, bgrStride, width, height, y, yStride, u, uStride, v, vStride, yuvType);
+    else
+#endif
 //#ifdef SIMD_NEON_ENABLE
 //    if (Neon::Enable && width >= Neon::A)
 //        Neon::BgrToYuv444pV2(bgr, bgrStride, width, height, y, yStride, u, uStride, v, vStride, yuvType);

diff --git a/src/Simd/SimdSse41.h b/src/Simd/SimdSse41.h
@@ -159,6 +159,9 @@ namespace Simd
 
         void BgrToYuv444p(const uint8_t* bgr, size_t width, size_t height, size_t bgrStride, uint8_t* y, size_t yStride, uint8_t* u, size_t uStride, uint8_t* v, size_t vStride);
 
+        void BgrToYuv444pV2(const uint8_t* bgr, size_t bgrStride, size_t width, size_t height,
+            uint8_t* y, size_t yStride, uint8_t* u, size_t uStride, uint8_t* v, size_t vStride, SimdYuvType yuvType);
+
         void Binarization(const uint8_t* src, size_t srcStride, size_t width, size_t height,
             uint8_t value, uint8_t positive, uint8_t negative, uint8_t* dst, size_t dstStride, SimdCompareType compareType);
 

diff --git a/src/Simd/SimdSse41BgrToYuv.cpp b/src/Simd/SimdSse41BgrToYuv.cpp
@@ -363,6 +363,57 @@ namespace Simd
             default:
                 assert(0);
             }
+#endif
+        }
+
+        //-------------------------------------------------------------------------------------------------
+
+        template <class T> SIMD_INLINE void BgrToYuv444pV2(const uint8_t* bgr, uint8_t* y, uint8_t* u, uint8_t* v)
+        {
+            __m128i blue, green, red;
+            LoadBgr<false>((__m128i*)bgr, blue, green, red);
+            _mm_storeu_si128((__m128i*)y, BgrToY8<T>(blue, green, red));
+            _mm_storeu_si128((__m128i*)u, BgrToU8<T>(blue, green, red));
+            _mm_storeu_si128((__m128i*)v, BgrToV8<T>(blue, green, red));
+        }
+
+        template <class T>  void BgrToYuv444pV2(const uint8_t* bgr, size_t bgrStride, size_t width, size_t height, uint8_t* y, size_t yStride,
+            uint8_t* u, size_t uStride, uint8_t* v, size_t vStride)
+        {
+            assert(width >= A);
+
+            size_t widthA = AlignLo(width, A);
+            for (size_t row = 0; row < height; row += 1)
+            {
+                for (size_t col = 0; col < widthA; col += A)
+                    BgrToYuv444pV2<T>(bgr + col * 3, y + col, u + col, v + col);
+                if (width != widthA)
+                {
+                    size_t col = width - A;
+                    BgrToYuv444pV2<T>(bgr + col * 3, y + col, u + col, v + col);
+                }
+                y += yStride;
+                u += uStride;
+                v += vStride;
+                bgr += bgrStride;
+            }
+        }
+
+        void BgrToYuv444pV2(const uint8_t* bgr, size_t bgrStride, size_t width, size_t height, uint8_t* y, size_t yStride,
+            uint8_t* u, size_t uStride, uint8_t* v, size_t vStride, SimdYuvType yuvType)
+        {
+#if defined(SIMD_X86_ENABLE) && defined(NDEBUG) && defined(_MSC_VER) && _MSC_VER <= 1900
+            Base::BgrToYuv444pV2(bgr, bgrStride, width, height, y, yStride, u, uStride, v, vStride, yuvType);
+#else
+            switch (yuvType)
+            {
+            case SimdYuvBt601: BgrToYuv444pV2<Base::Bt601>(bgr, bgrStride, width, height, y, yStride, u, uStride, v, vStride); break;
+            case SimdYuvBt709: BgrToYuv444pV2<Base::Bt709>(bgr, bgrStride, width, height, y, yStride, u, uStride, v, vStride); break;
+            case SimdYuvBt2020: BgrToYuv444pV2<Base::Bt2020>(bgr, bgrStride, width, height, y, yStride, u, uStride, v, vStride); break;
+            case SimdYuvTrect871: BgrToYuv444pV2<Base::Trect871>(bgr, bgrStride, width, height, y, yStride, u, uStride, v, vStride); break;
+            default:
+                assert(0);
+            }
 #endif
         }
     }

diff --git a/src/Test/TestAnyToYuv.cpp b/src/Test/TestAnyToYuv.cpp
@@ -563,11 +563,11 @@ namespace Test
 
         result = result && AnyToYuvV2AutoTest(View::Bgr24, 1, 1, FUNC_YUV2(Simd::Base::BgrToYuv444pV2), FUNC_YUV2(SimdBgrToYuv444pV2));
 
-//#ifdef SIMD_SSE41_ENABLE
-//        if (Simd::Sse41::Enable)
-//            result = result && AnyToYuvV2AutoTest(View::Bgr24, 1, 1, FUNC_YUV2(Simd::Sse41::BgrToYuv444pV2), FUNC_YUV2(SimdBgrToYuv444pV2));
-//#endif 
-//
+#ifdef SIMD_SSE41_ENABLE
+        if (Simd::Sse41::Enable)
+            result = result && AnyToYuvV2AutoTest(View::Bgr24, 1, 1, FUNC_YUV2(Simd::Sse41::BgrToYuv444pV2), FUNC_YUV2(SimdBgrToYuv444pV2));
+#endif 
+
 //#ifdef SIMD_AVX2_ENABLE
 //        if (Simd::Avx2::Enable)
 //            result = result && AnyToYuvV2AutoTest(View::Bgr24, 1, 1, FUNC_YUV2(Simd::Avx2::BgrToYuv444pV2), FUNC_YUV2(SimdBgrToYuv444pV2));