spatialGradient: Vectorise inner area

modules/imgproc/src/precomp.hpp

@@ -49,6 +49,7 @@
 #include "opencv2/imgproc/imgproc_c.h"
 #include "opencv2/core/private.hpp"
 #include "opencv2/core/ocl.hpp"
+#include "opencv2/hal.hpp"
 
 #include <math.h>
 #include <assert.h>

modules/imgproc/src/spatialgradient.cpp

@@ -41,6 +41,7 @@
 //M*/
 
 #include "precomp.hpp"
+#include "opencv2/hal/intrin.hpp"
 
 namespace cv
 {
@@ -96,7 +97,6 @@ void spatialGradient( InputArray _src, OutputArray _dx, OutputArray _dy, int ksi
     // No-SSE
     int idx;
 
-
     p_dx[0] = 0;   // Top-left corner
     p_dy[0] = 0;
     p_dx[W-1] = 0; // Top-right corner
@@ -168,6 +168,100 @@ void spatialGradient( InputArray _src, OutputArray _dx, OutputArray _dy, int ksi
     }
 
     // Do Inner area
+#if CV_SIMD128
+    // Characters in variable names have the following meanings:
+    // u: unsigned char
+    // s: signed int
+    //
+    // [row][column]
+    // m: offset -1
+    // n: offset  0
+    // p: offset  1
+    // Example: umn is offset -1 in row and offset 0 in column
+    v_uint8x16 v_umm, v_umn, v_ump,
+               v_unm, v_unn, v_unp,
+               v_upm, v_upn, v_upp;
+    v_uint16x8 v_umm1, v_umm2, v_umn1, v_umn2, v_ump1, v_ump2,
+               v_unm1, v_unm2, v_unn1, v_unn2, v_unp1, v_unp2,
+               v_upm1, v_upm2, v_upn1, v_upn2, v_upp1, v_upp2;
+    v_int16x8 v_smm1, v_smm2, v_smn1, v_smn2, v_smp1, v_smp2,
+              v_snm1, v_snm2, v_snn1, v_snn2, v_snp1, v_snp2,
+              v_spm1, v_spm2, v_spn1, v_spn2, v_spp1, v_spp2,
+              v_two = v_setall_s16(2),
+              v_sdx1, v_sdx2, v_sdy1, v_sdy2;
+    for ( i = 1; i < H - 1; i++ )
+    for ( j = 1; j < W - 1 - 15; j += 16 )
+    {
+        // Load
+        idx = i*W + j;
+        v_umm = v_load(&p_src[idx - W - 1]);
+        v_umn = v_load(&p_src[idx - W]);
+        v_ump = v_load(&p_src[idx - W + 1]);
+        v_unm = v_load(&p_src[idx - 1]);
+        v_unn = v_load(&p_src[idx]);
+        v_unp = v_load(&p_src[idx + 1]);
+        v_upm = v_load(&p_src[idx + W - 1]);
+        v_upn = v_load(&p_src[idx + W]);
+        v_upp = v_load(&p_src[idx + W + 1]);
+
+        // Expand to uint
+        v_expand(v_umm, v_umm1, v_umm2);
+        v_expand(v_umn, v_umn1, v_umn2);
+        v_expand(v_ump, v_ump1, v_ump2);
+        v_expand(v_unm, v_unm1, v_unm2);
+        v_expand(v_unn, v_unn1, v_unn2);
+        v_expand(v_unp, v_unp1, v_unp2);
+        v_expand(v_upm, v_upm1, v_upm2);
+        v_expand(v_upn, v_upn1, v_upn2);
+        v_expand(v_upp, v_upp1, v_upp2);
+
+        // Convert to int
+        v_smm1 = v_reinterpret_as_s16(v_umm1);
+        v_smm2 = v_reinterpret_as_s16(v_umm2);
+        v_smn1 = v_reinterpret_as_s16(v_umn1);
+        v_smn2 = v_reinterpret_as_s16(v_umn2);
+        v_smp1 = v_reinterpret_as_s16(v_ump1);
+        v_smp2 = v_reinterpret_as_s16(v_ump2);
+        v_snm1 = v_reinterpret_as_s16(v_unm1);
+        v_snm2 = v_reinterpret_as_s16(v_unm2);
+        v_snn1 = v_reinterpret_as_s16(v_unn1);
+        v_snn2 = v_reinterpret_as_s16(v_unn2);
+        v_snp1 = v_reinterpret_as_s16(v_unp1);
+        v_snp2 = v_reinterpret_as_s16(v_unp2);
+        v_spm1 = v_reinterpret_as_s16(v_upm1);
+        v_spm2 = v_reinterpret_as_s16(v_upm2);
+        v_spn1 = v_reinterpret_as_s16(v_upn1);
+        v_spn2 = v_reinterpret_as_s16(v_upn2);
+        v_spp1 = v_reinterpret_as_s16(v_upp1);
+        v_spp2 = v_reinterpret_as_s16(v_upp2);
+
+        // dx
+        v_sdx1 = (v_smp1 - v_smm1) + v_two*(v_snp1 - v_snm1) + (v_spp1 - v_spm1);
+        v_sdx2 = (v_smp2 - v_smm2) + v_two*(v_snp2 - v_snm2) + (v_spp2 - v_spm2);
+
+        // dy
+        v_sdy1 = (v_spm1 - v_smm1) + v_two*(v_spn1 - v_smn1) + (v_spp1 - v_smp1);
+        v_sdy2 = (v_spm2 - v_smm2) + v_two*(v_spn2 - v_smn2) + (v_spp2 - v_smp2);
+
+        // Store
+        v_store(&p_dx[idx],   v_sdx1);
+        v_store(&p_dx[idx+8], v_sdx2);
+        v_store(&p_dy[idx],   v_sdy1);
+        v_store(&p_dy[idx+8], v_sdy2);
+    }
+
+    // Cleanup
+    int end_j = j;
+    for ( i = 1; i < H - 1; i++ )
+    for ( j = end_j; j < W - 1; j++ )
+    {
+        idx = i*W + j;
+        p_dx[idx] = -(p_src[idx-W-1] + 2*p_src[idx-1] + p_src[idx+W-1]) +
+                     (p_src[idx-W+1] + 2*p_src[idx+1] + p_src[idx+W+1]);
+        p_dy[idx] = -(p_src[idx-W-1] + 2*p_src[idx-W] + p_src[idx-W+1]) +
+                     (p_src[idx+W-1] + 2*p_src[idx+W] + p_src[idx+W+1]);
+    }
+#else
     for ( i = 1; i < H - 1; i++ )
     for ( j = 1; j < W - 1; j++ )
     {
@@ -177,6 +271,7 @@ void spatialGradient( InputArray _src, OutputArray _dx, OutputArray _dy, int ksi
         p_dy[idx] = -(p_src[idx-W-1] + 2*p_src[idx-W] + p_src[idx-W+1]) +
                      (p_src[idx+W-1] + 2*p_src[idx+W] + p_src[idx+W+1]);
     }
+#endif
 
 }