use universal intrinsic in FAST

modules/features2d/src/fast_score.cpp

@@ -42,7 +42,7 @@ The references are:
 */
 
 #include "fast_score.hpp"
-
+#include "opencv2/core/hal/intrin.hpp"
 #define VERIFY_CORNERS 0
 
 namespace cv {
@@ -125,45 +125,48 @@ int cornerScore<16>(const uchar* ptr, const int pixel[], int threshold)
     for( k = 0; k < N; k++ )
         d[k] = (short)(v - ptr[pixel[k]]);
 
-#if CV_SSE2
-    __m128i q0 = _mm_set1_epi16(-1000), q1 = _mm_set1_epi16(1000);
-    for( k = 0; k < 16; k += 8 )
+#if CV_SIMD128
+    if (hasSIMD128())
+    {
+        v_int16x8 q0 = v_setall_s16(-1000), q1 = v_setall_s16(1000);
+        for (k = 0; k < 16; k += 8)
         {
-        __m128i v0 = _mm_loadu_si128((__m128i*)(d+k+1));
-        __m128i v1 = _mm_loadu_si128((__m128i*)(d+k+2));
-        __m128i a = _mm_min_epi16(v0, v1);
-        __m128i b = _mm_max_epi16(v0, v1);
-        v0 = _mm_loadu_si128((__m128i*)(d+k+3));
-        a = _mm_min_epi16(a, v0);
-        b = _mm_max_epi16(b, v0);
-        v0 = _mm_loadu_si128((__m128i*)(d+k+4));
-        a = _mm_min_epi16(a, v0);
-        b = _mm_max_epi16(b, v0);
-        v0 = _mm_loadu_si128((__m128i*)(d+k+5));
-        a = _mm_min_epi16(a, v0);
-        b = _mm_max_epi16(b, v0);
-        v0 = _mm_loadu_si128((__m128i*)(d+k+6));
-        a = _mm_min_epi16(a, v0);
-        b = _mm_max_epi16(b, v0);
-        v0 = _mm_loadu_si128((__m128i*)(d+k+7));
-        a = _mm_min_epi16(a, v0);
-        b = _mm_max_epi16(b, v0);
-        v0 = _mm_loadu_si128((__m128i*)(d+k+8));
-        a = _mm_min_epi16(a, v0);
-        b = _mm_max_epi16(b, v0);
-        v0 = _mm_loadu_si128((__m128i*)(d+k));
-        q0 = _mm_max_epi16(q0, _mm_min_epi16(a, v0));
-        q1 = _mm_min_epi16(q1, _mm_max_epi16(b, v0));
-        v0 = _mm_loadu_si128((__m128i*)(d+k+9));
-        q0 = _mm_max_epi16(q0, _mm_min_epi16(a, v0));
-        q1 = _mm_min_epi16(q1, _mm_max_epi16(b, v0));
+            v_int16x8 v0 = v_load(d + k + 1);
+            v_int16x8 v1 = v_load(d + k + 2);
+            v_int16x8 a = v_min(v0, v1);
+            v_int16x8 b = v_max(v0, v1);
+            v0 = v_load(d + k + 3);
+            a = v_min(a, v0);
+            b = v_max(b, v0);
+            v0 = v_load(d + k + 4);
+            a = v_min(a, v0);
+            b = v_max(b, v0);
+            v0 = v_load(d + k + 5);
+            a = v_min(a, v0);
+            b = v_max(b, v0);
+            v0 = v_load(d + k + 6);
+            a = v_min(a, v0);
+            b = v_max(b, v0);
+            v0 = v_load(d + k + 7);
+            a = v_min(a, v0);
+            b = v_max(b, v0);
+            v0 = v_load(d + k + 8);
+            a = v_min(a, v0);
+            b = v_max(b, v0);
+            v0 = v_load(d + k);
+            q0 = v_max(q0, v_min(a, v0));
+            q1 = v_min(q1, v_max(b, v0));
+            v0 = v_load(d + k + 9);
+            q0 = v_max(q0, v_min(a, v0));
+            q1 = v_min(q1, v_max(b, v0));
         }
-    q0 = _mm_max_epi16(q0, _mm_sub_epi16(_mm_setzero_si128(), q1));
-    q0 = _mm_max_epi16(q0, _mm_unpackhi_epi64(q0, q0));
-    q0 = _mm_max_epi16(q0, _mm_srli_si128(q0, 4));
-    q0 = _mm_max_epi16(q0, _mm_srli_si128(q0, 2));
-    threshold = (short)_mm_cvtsi128_si32(q0) - 1;
-#else
+        q0 = v_max(q0, v_setzero_s16() - q1);
+        threshold = v_reduce_max(q0) - 1;
+    }
+    else
+#endif
+    {
+
         int a0 = threshold;
         for( k = 0; k < 16; k += 2 )
         {
@@ -197,8 +200,8 @@ int cornerScore<16>(const uchar* ptr, const int pixel[], int threshold)
             b0 = std::min(b0, std::max(b, (int)d[k+9]));
         }
 
-    threshold = -b0-1;
-#endif
+        threshold = -b0 - 1;
+    }
 
 #if VERIFY_CORNERS
     testCorner(ptr, pixel, K, N, threshold);
@@ -214,44 +217,46 @@ int cornerScore<12>(const uchar* ptr, const int pixel[], int threshold)
     short d[N + 4];
     for( k = 0; k < N; k++ )
         d[k] = (short)(v - ptr[pixel[k]]);
-#if CV_SSE2
+#if CV_SIMD128
     for( k = 0; k < 4; k++ )
         d[N+k] = d[k];
 #endif
 
-#if CV_SSE2
-    __m128i q0 = _mm_set1_epi16(-1000), q1 = _mm_set1_epi16(1000);
-    for( k = 0; k < 16; k += 8 )
+#if CV_SIMD128
+    if (hasSIMD128())
     {
-        __m128i v0 = _mm_loadu_si128((__m128i*)(d+k+1));
-        __m128i v1 = _mm_loadu_si128((__m128i*)(d+k+2));
-        __m128i a = _mm_min_epi16(v0, v1);
-        __m128i b = _mm_max_epi16(v0, v1);
-        v0 = _mm_loadu_si128((__m128i*)(d+k+3));
-        a = _mm_min_epi16(a, v0);
-        b = _mm_max_epi16(b, v0);
-        v0 = _mm_loadu_si128((__m128i*)(d+k+4));
-        a = _mm_min_epi16(a, v0);
-        b = _mm_max_epi16(b, v0);
-        v0 = _mm_loadu_si128((__m128i*)(d+k+5));
-        a = _mm_min_epi16(a, v0);
-        b = _mm_max_epi16(b, v0);
-        v0 = _mm_loadu_si128((__m128i*)(d+k+6));
-        a = _mm_min_epi16(a, v0);
-        b = _mm_max_epi16(b, v0);
-        v0 = _mm_loadu_si128((__m128i*)(d+k));
-        q0 = _mm_max_epi16(q0, _mm_min_epi16(a, v0));
-        q1 = _mm_min_epi16(q1, _mm_max_epi16(b, v0));
-        v0 = _mm_loadu_si128((__m128i*)(d+k+7));
-        q0 = _mm_max_epi16(q0, _mm_min_epi16(a, v0));
-        q1 = _mm_min_epi16(q1, _mm_max_epi16(b, v0));
+        v_int16x8 q0 = v_setall_s16(-1000), q1 = v_setall_s16(1000);
+        for (k = 0; k < 16; k += 8)
+        {
+            v_int16x8 v0 = v_load(d + k + 1);
+            v_int16x8 v1 = v_load(d + k + 2);
+            v_int16x8 a = v_min(v0, v1);
+            v_int16x8 b = v_max(v0, v1);
+            v0 = v_load(d + k + 3);
+            a = v_min(a, v0);
+            b = v_max(b, v0);
+            v0 = v_load(d + k + 4);
+            a = v_min(a, v0);
+            b = v_max(b, v0);
+            v0 = v_load(d + k + 5);
+            a = v_min(a, v0);
+            b = v_max(b, v0);
+            v0 = v_load(d + k + 6);
+            a = v_min(a, v0);
+            b = v_max(b, v0);
+            v0 = v_load(d + k);
+            q0 = v_max(q0, v_min(a, v0));
+            q1 = v_min(q1, v_max(b, v0));
+            v0 = v_load(d + k + 7);
+            q0 = v_max(q0, v_min(a, v0));
+            q1 = v_min(q1, v_max(b, v0));
+        }
+        q0 = v_max(q0, v_setzero_s16() - q1);
+        threshold = v_reduce_max(q0) - 1;
     }
-    q0 = _mm_max_epi16(q0, _mm_sub_epi16(_mm_setzero_si128(), q1));
-    q0 = _mm_max_epi16(q0, _mm_unpackhi_epi64(q0, q0));
-    q0 = _mm_max_epi16(q0, _mm_srli_si128(q0, 4));
-    q0 = _mm_max_epi16(q0, _mm_srli_si128(q0, 2));
-    threshold = (short)_mm_cvtsi128_si32(q0) - 1;
-#else
+    else
+#endif
+    {
         int a0 = threshold;
         for( k = 0; k < 12; k += 2 )
         {
@@ -282,8 +287,7 @@ int cornerScore<12>(const uchar* ptr, const int pixel[], int threshold)
         }
 
         threshold = -b0-1;
-#endif
-
+    }
 #if VERIFY_CORNERS
     testCorner(ptr, pixel, K, N, threshold);
 #endif
@@ -293,35 +297,37 @@ int cornerScore<12>(const uchar* ptr, const int pixel[], int threshold)
 template<>
 int cornerScore<8>(const uchar* ptr, const int pixel[], int threshold)
 {
-    const int K = 4, N = K*3 + 1;
+    const int K = 4, N = K * 3 + 1;
     int k, v = ptr[0];
     short d[N];
-    for( k = 0; k < N; k++ )
+    for (k = 0; k < N; k++)
         d[k] = (short)(v - ptr[pixel[k]]);
 
-#if CV_SSE2
-    __m128i v0 = _mm_loadu_si128((__m128i*)(d+1));
-    __m128i v1 = _mm_loadu_si128((__m128i*)(d+2));
-    __m128i a = _mm_min_epi16(v0, v1);
-    __m128i b = _mm_max_epi16(v0, v1);
-    v0 = _mm_loadu_si128((__m128i*)(d+3));
-    a = _mm_min_epi16(a, v0);
-    b = _mm_max_epi16(b, v0);
-    v0 = _mm_loadu_si128((__m128i*)(d+4));
-    a = _mm_min_epi16(a, v0);
-    b = _mm_max_epi16(b, v0);
-    v0 = _mm_loadu_si128((__m128i*)(d));
-    __m128i q0 = _mm_min_epi16(a, v0);
-    __m128i q1 = _mm_max_epi16(b, v0);
-    v0 = _mm_loadu_si128((__m128i*)(d+5));
-    q0 = _mm_max_epi16(q0, _mm_min_epi16(a, v0));
-    q1 = _mm_min_epi16(q1, _mm_max_epi16(b, v0));
-    q0 = _mm_max_epi16(q0, _mm_sub_epi16(_mm_setzero_si128(), q1));
-    q0 = _mm_max_epi16(q0, _mm_unpackhi_epi64(q0, q0));
-    q0 = _mm_max_epi16(q0, _mm_srli_si128(q0, 4));
-    q0 = _mm_max_epi16(q0, _mm_srli_si128(q0, 2));
-    threshold = (short)_mm_cvtsi128_si32(q0) - 1;
-#else
+#if CV_SIMD128
+    if (hasSIMD128())
+    {
+        v_int16x8 v0 = v_load(d + 1);
+        v_int16x8 v1 = v_load(d + 2);
+        v_int16x8 a = v_min(v0, v1);
+        v_int16x8 b = v_max(v0, v1);
+        v0 = v_load(d + 3);
+        a = v_min(a, v0);
+        b = v_max(b, v0);
+        v0 = v_load(d + 4);
+        a = v_min(a, v0);
+        b = v_max(b, v0);
+        v0 = v_load(d);
+        v_int16x8 q0 = v_min(a, v0);
+        v_int16x8 q1 = v_max(b, v0);
+        v0 = v_load(d + 5);
+        q0 = v_max(q0, v_min(a, v0));
+        q1 = v_min(q1, v_max(b, v0));
+        q0 = v_max(q0, v_setzero_s16() - q1);
+        threshold = v_reduce_max(q0) - 1;
+    }
+    else
+#endif
+    {
         int a0 = threshold;
         for( k = 0; k < 8; k += 2 )
         {
@@ -348,7 +354,7 @@ int cornerScore<8>(const uchar* ptr, const int pixel[], int threshold)
         }
 
         threshold = -b0-1;
-#endif
+    }
 
 #if VERIFY_CORNERS
     testCorner(ptr, pixel, K, N, threshold);