core(intrin): drop hasSIMD128 checks

- use compile-time checks instead (`#if CV_SIMD128`) - runtime checks are useless
ywscr · Jun 8, 2019 · 1e9ad54 · 1e9ad54
1 parent 254f88f
commit 1e9ad54
Show file tree

Hide file tree

Showing 14 changed files with 27 additions and 166 deletions.
diff --git a/modules/calib3d/src/stereobm.cpp b/modules/calib3d/src/stereobm.cpp
@@ -204,10 +204,6 @@ prefilterXSobel( const Mat& src, Mat& dst, int ftzero )
         tab[x] = (uchar)(x - OFS < -ftzero ? 0 : x - OFS > ftzero ? ftzero*2 : x - OFS + ftzero);
     uchar val0 = tab[0 + OFS];
 
-#if CV_SIMD128
-    bool useSIMD = hasSIMD128();
-#endif
-
     for( y = 0; y < size.height-1; y += 2 )
     {
         const uchar* srow1 = src.ptr<uchar>(y);
@@ -221,7 +217,6 @@ prefilterXSobel( const Mat& src, Mat& dst, int ftzero )
         x = 1;
 
 #if CV_SIMD128
-        if( useSIMD )
         {
             v_int16x8 ftz = v_setall_s16((short) ftzero);
             v_int16x8 ftz2 = v_setall_s16((short)(ftzero*2));
@@ -268,7 +263,6 @@ prefilterXSobel( const Mat& src, Mat& dst, int ftzero )
         uchar* dptr = dst.ptr<uchar>(y);
         x = 0;
 #if CV_SIMD128
-        if( useSIMD )
         {
             v_uint8x16 val0_16 = v_setall_u8(val0);
             for(; x <= size.width-16; x+=16 )
@@ -594,8 +588,6 @@ findStereoCorrespondenceBM( const Mat& left, const Mat& right,
     mType FILTERED = (mType)((mindisp - 1) << disp_shift);
 
 #if CV_SIMD128
-    bool useSIMD = hasSIMD128();
-    if( useSIMD )
     {
         CV_Assert (ndisp % 8 == 0);
     }
@@ -637,7 +629,6 @@ findStereoCorrespondenceBM( const Mat& left, const Mat& right,
             int lval = lptr[0];
             d = 0;
 #if CV_SIMD128
-            if( useSIMD )
             {
                 v_uint8x16 lv = v_setall_u8((uchar)lval);
 
@@ -706,7 +697,6 @@ findStereoCorrespondenceBM( const Mat& left, const Mat& right,
             int lval = lptr[0];
             d = 0;
 #if CV_SIMD128
-            if( useSIMD )
             {
                 v_uint8x16 lv = v_setall_u8((uchar)lval);
                 for( ; d <= ndisp - 16; d += 16 )
@@ -769,7 +759,6 @@ findStereoCorrespondenceBM( const Mat& left, const Mat& right,
         {
             d = 0;
 #if CV_SIMD128
-            if( useSIMD )
             {
                 for( d = 0; d <= ndisp-8; d += 8 )
                 {
@@ -799,7 +788,6 @@ findStereoCorrespondenceBM( const Mat& left, const Mat& right,
             hsad_sub = hsad0 + MAX(y - wsz2 - 1, -dy0)*ndisp;
             d = 0;
 #if CV_SIMD128
-            if( useSIMD )
             {
                 v_int32x4 d0_4 = v_int32x4(0, 1, 2, 3);
                 v_int32x4 dd_4 = v_setall_s32(4);
@@ -1003,9 +991,6 @@ struct FindStereoCorrespInvoker : public ParallelLoopBody
         validDisparityRect = _validDisparityRect;
         slidingSumBuf = &_slidingSumBuf;
         cost = &_cost;
-#if CV_SIMD128
-        useSIMD = hasSIMD128();
-#endif
     }
 
     void operator()(const Range& range) const CV_OVERRIDE
@@ -1043,7 +1028,7 @@ struct FindStereoCorrespInvoker : public ParallelLoopBody
         Mat cost_i = state->disp12MaxDiff >= 0 ? cost->rowRange(row0, row1) : Mat();
 
 #if CV_SIMD128
-        if( useSIMD && useShorts )
+        if (useShorts)
         {
             if( disp_i.type() == CV_16S)
                 findStereoCorrespondenceBM_SIMD<short>( left_i, right_i, disp_i, cost_i, *state, ptr, row0, rows - row1 );
@@ -1083,7 +1068,6 @@ struct FindStereoCorrespInvoker : public ParallelLoopBody
     size_t stripeBufSize;
     bool useShorts;
     Rect validDisparityRect;
-    bool useSIMD;
 };
 
 class StereoBMImpl CV_FINAL : public StereoBM

diff --git a/modules/calib3d/src/stereosgbm.cpp b/modules/calib3d/src/stereosgbm.cpp
@@ -140,9 +140,6 @@ static void calcPixelCostBT( const Mat& img1, const Mat& img2, int y,
     int width2 = maxX2 - minX2;
     const PixType *row1 = img1.ptr<PixType>(y), *row2 = img2.ptr<PixType>(y);
     PixType *prow1 = buffer + width2*2, *prow2 = prow1 + width*cn*2;
-#if CV_SIMD128
-    bool useSIMD = hasSIMD128();
-#endif
 
     tab += tabOfs;
 
@@ -224,7 +221,7 @@ static void calcPixelCostBT( const Mat& img1, const Mat& img2, int y,
             int u1 = std::max(ul, ur); u1 = std::max(u1, u);
 
         #if CV_SIMD128
-            if( useSIMD )
+            if (true)
             {
                 v_uint8x16 _u  = v_setall_u8((uchar)u), _u0 = v_setall_u8((uchar)u0);
                 v_uint8x16 _u1 = v_setall_u8((uchar)u1);
@@ -304,8 +301,6 @@ static void computeDisparitySGBM( const Mat& img1, const Mat& img2,
         5, 0, 1, 0, 2, 0, 1, 0, 3, 0, 1, 0, 2, 0, 1, 0, 4, 0, 1, 0, 2, 0, 1, 0, 3, 0, 1, 0, 2, 0, 1, 0
     };
     static const v_uint16x8 v_LSB = v_uint16x8(0x1, 0x2, 0x4, 0x8, 0x10, 0x20, 0x40, 0x80);
-
-    bool useSIMD = hasSIMD128();
 #endif
 
     const int ALIGN = 16;
@@ -450,7 +445,7 @@ static void computeDisparitySGBM( const Mat& img1, const Mat& img2,
                                 const CostType* pixSub = pixDiff + std::max(x - (SW2+1)*D, 0);
 
                             #if CV_SIMD128
-                                if( useSIMD )
+                                if (true)
                                 {
                                     for( d = 0; d < D; d += 8 )
                                     {
@@ -547,7 +542,7 @@ static void computeDisparitySGBM( const Mat& img1, const Mat& img2,
                 CostType* Sp = S + x*D;
 
             #if CV_SIMD128
-                if( useSIMD )
+                if (true)
                 {
                     v_int16x8 _P1 = v_setall_s16((short)P1);
 
@@ -681,7 +676,7 @@ static void computeDisparitySGBM( const Mat& img1, const Mat& img2,
                         const CostType* Cp = C + x*D;
 
                     #if CV_SIMD128
-                        if( useSIMD )
+                        if (true)
                         {
                             v_int16x8 _P1 = v_setall_s16((short)P1);
                             v_int16x8 _delta0 = v_setall_s16((short)delta0);
@@ -753,7 +748,7 @@ static void computeDisparitySGBM( const Mat& img1, const Mat& img2,
                     else
                     {
                     #if CV_SIMD128
-                        if( useSIMD )
+                        if (true)
                         {
                             v_int16x8 _minS = v_setall_s16(MAX_COST), _bestDisp = v_setall_s16(-1);
                             v_int16x8 _d8 = v_int16x8(0, 1, 2, 3, 4, 5, 6, 7), _8 = v_setall_s16(8);
@@ -868,7 +863,6 @@ struct CalcVerticalSums: public ParallelLoopBody
         Cbuf = alignedBuf;
         Sbuf = Cbuf + CSBufSize;
         hsumBuf = Sbuf + CSBufSize;
-        useSIMD = hasSIMD128();
     }
 
     void operator()(const Range& range) const CV_OVERRIDE
@@ -957,7 +951,7 @@ struct CalcVerticalSums: public ParallelLoopBody
                                     const CostType* pixSub = pixDiff + std::max(x - (SW2+1)*D, 0);
 
                                 #if CV_SIMD128
-                                    if( useSIMD )
+                                    if (true)
                                     {
                                         for( d = 0; d < D; d += 8 )
                                         {
@@ -1034,7 +1028,7 @@ struct CalcVerticalSums: public ParallelLoopBody
                     CostType* Sp = S + x*D;
 
                 #if CV_SIMD128
-                    if( useSIMD )
+                    if (true)
                     {
                         v_int16x8 _P1 = v_setall_s16((short)P1);
 
@@ -1121,7 +1115,6 @@ struct CalcVerticalSums: public ParallelLoopBody
     size_t LrSize;
     size_t hsumBufNRows;
     int ftzero;
-    bool useSIMD;
 };
 
 struct CalcHorizontalSums: public ParallelLoopBody
@@ -1149,7 +1142,6 @@ struct CalcHorizontalSums: public ParallelLoopBody
         LrSize = 2 * D2;
         Cbuf = alignedBuf;
         Sbuf = Cbuf + CSBufSize;
-        useSIMD = hasSIMD128();
     }
 
     void operator()(const Range& range) const CV_OVERRIDE
@@ -1204,7 +1196,7 @@ struct CalcHorizontalSums: public ParallelLoopBody
                 CostType* Sp = S + x*D;
 
             #if CV_SIMD128
-                if( useSIMD )
+                if (true)
                 {
                     v_int16x8 _P1 = v_setall_s16((short)P1);
 
@@ -1277,7 +1269,7 @@ struct CalcHorizontalSums: public ParallelLoopBody
                 minLr = MAX_COST;
 
             #if CV_SIMD128
-                if( useSIMD )
+                if (true)
                 {
                     v_int16x8 _P1 = v_setall_s16((short)P1);
 
@@ -1424,7 +1416,6 @@ struct CalcHorizontalSums: public ParallelLoopBody
     int INVALID_DISP_SCALED;
     int uniquenessRatio;
     int disp12MaxDiff;
-    bool useSIMD;
 };
 /*
  computes disparity for "roi" in img1 w.r.t. img2 and write it to disp1buf.
@@ -1536,10 +1527,6 @@ struct SGBM3WayMainLoop : public ParallelLoopBody
     int costBufSize, hsumBufNRows;
     int TAB_OFS, ftzero;
 
-#if CV_SIMD128
-    bool useSIMD;
-#endif
-
     PixType* clipTab;
 
     SGBM3WayMainLoop(Mat *_buffers, const Mat& _img1, const Mat& _img2, Mat* _dst_disp, const StereoSGBMParams& params, PixType* _clipTab, int _nstripes, int _stripe_overlap);
@@ -1569,10 +1556,6 @@ buffers(_buffers), img1(&_img1), img2(&_img2), dst_disp(_dst_disp), clipTab(_cli
     hsumBufNRows = SH2*2 + 2;
     TAB_OFS = 256*4;
     ftzero = std::max(params.preFilterCap, 15) | 1;
-
-#if CV_SIMD128
-    useSIMD = hasSIMD128();
-#endif
 }
 
 void getBufferPointers(Mat& buffer, int width, int width1, int D, int num_ch, int SH2, int P2,
@@ -1673,7 +1656,7 @@ void SGBM3WayMainLoop::getRawMatchingCost(CostType* C, // target cost-volume row
                     const CostType* pixSub = pixDiff + std::max(x - (SW2+1)*D, 0);
 
 #if CV_SIMD128
-                    if(useSIMD)
+                    if (true)
                     {
                         v_int16x8 hv_reg;
                         for( d = 0; d < D; d+=8 )
@@ -1734,7 +1717,7 @@ inline void accumulateCostsLeftTop(CostType* leftBuf, CostType* leftBuf_prev, Co
                                    CostType& leftMinCost, CostType& topMinCost, int D, int P1, int P2)
 {
 #if CV_SIMD128
-    if(hasSIMD128())
+    if (true)
     {
         v_int16x8 P1_reg = v_setall_s16(cv::saturate_cast<CostType>(P1));
 
@@ -1846,7 +1829,7 @@ inline void accumulateCostsRight(CostType* rightBuf, CostType* topBuf, CostType*
                                  CostType& rightMinCost, int D, int P1, int P2, int& optimal_disp, CostType& min_cost)
 {
 #if CV_SIMD128
-    if(hasSIMD128())
+    if (true)
     {
         v_int16x8 P1_reg = v_setall_s16(cv::saturate_cast<CostType>(P1));
 
@@ -2012,7 +1995,7 @@ void SGBM3WayMainLoop::operator () (const Range& range) const
             if(uniquenessRatio>0)
             {
 #if CV_SIMD128
-                if(useSIMD)
+                if (true)
                 {
                     horPassCostVolume+=x;
                     int thresh = (100*min_cost)/(100-uniquenessRatio);

diff --git a/modules/core/include/opencv2/core/hal/intrin_avx.hpp b/modules/core/include/opencv2/core/hal/intrin_avx.hpp
@@ -2774,15 +2774,6 @@ inline void v_pack_store(float16_t* ptr, const v_float32x8& a)
 
 inline void v256_cleanup() { _mm256_zeroall(); }
 
-//! @name Check SIMD256 support
-//! @{
-//! @brief Check CPU capability of SIMD operation
-static inline bool hasSIMD256()
-{
-    return (CV_CPU_HAS_SUPPORT_AVX2) ? true : false;
-}
-//! @}
-
 CV_CPU_OPTIMIZATION_HAL_NAMESPACE_END
 
 //! @endcond

diff --git a/modules/core/include/opencv2/core/hal/intrin_cpp.hpp b/modules/core/include/opencv2/core/hal/intrin_cpp.hpp
@@ -2357,16 +2357,6 @@ inline void v_cleanup() {}
 
 //! @}
 
-//! @name Check SIMD support
-//! @{
-//! @brief Check CPU capability of SIMD operation
-static inline bool hasSIMD128()
-{
-    return false;
-}
-
-//! @}
-
 #ifndef CV_DOXYGEN
 CV_CPU_OPTIMIZATION_HAL_NAMESPACE_END
 #endif

diff --git a/modules/core/include/opencv2/core/hal/intrin_neon.hpp b/modules/core/include/opencv2/core/hal/intrin_neon.hpp
@@ -1910,16 +1910,6 @@ inline void v_pack_store(float16_t* ptr, const v_float32x4& v)
 
 inline void v_cleanup() {}
 
-//! @name Check SIMD support
-//! @{
-//! @brief Check CPU capability of SIMD operation
-static inline bool hasSIMD128()
-{
-    return (CV_CPU_HAS_SUPPORT_NEON) ? true : false;
-}
-
-//! @}
-
 CV_CPU_OPTIMIZATION_HAL_NAMESPACE_END
 
 //! @endcond

diff --git a/modules/core/include/opencv2/core/hal/intrin_sse.hpp b/modules/core/include/opencv2/core/hal/intrin_sse.hpp
@@ -3043,16 +3043,6 @@ inline void v_pack_store(float16_t* ptr, const v_float32x4& v)
 
 inline void v_cleanup() {}
 
-//! @name Check SIMD support
-//! @{
-//! @brief Check CPU capability of SIMD operation
-static inline bool hasSIMD128()
-{
-    return (CV_CPU_HAS_SUPPORT_SSE2) ? true : false;
-}
-
-//! @}
-
 CV_CPU_OPTIMIZATION_HAL_NAMESPACE_END
 
 //! @endcond

diff --git a/modules/core/include/opencv2/core/hal/intrin_vsx.hpp b/modules/core/include/opencv2/core/hal/intrin_vsx.hpp
@@ -1355,16 +1355,6 @@ OPENCV_HAL_IMPL_VSX_TRANSPOSE4x4(v_uint32x4, vec_uint4)
 OPENCV_HAL_IMPL_VSX_TRANSPOSE4x4(v_int32x4, vec_int4)
 OPENCV_HAL_IMPL_VSX_TRANSPOSE4x4(v_float32x4, vec_float4)
 
-//! @name Check SIMD support
-//! @{
-//! @brief Check CPU capability of SIMD operation
-static inline bool hasSIMD128()
-{
-    return (CV_CPU_HAS_SUPPORT_VSX) ? true : false;
-}
-
-//! @}
-
 CV_CPU_OPTIMIZATION_HAL_NAMESPACE_END
 
 //! @endcond

diff --git a/modules/features2d/src/fast.cpp b/modules/features2d/src/fast.cpp
@@ -64,7 +64,6 @@ void FAST_t(InputArray _img, std::vector<KeyPoint>& keypoints, int threshold, bo
 #if CV_SIMD128
     const int quarterPatternSize = patternSize/4;
     v_uint8x16 delta = v_setall_u8(0x80), t = v_setall_u8((char)threshold), K16 = v_setall_u8((char)K);
-    bool hasSimd = hasSIMD128();
 #if CV_TRY_AVX2
     Ptr<opt_AVX2::FAST_t_patternSize16_AVX2> fast_t_impl_avx2;
     if(CV_CPU_HAS_SUPPORT_AVX2)
@@ -102,7 +101,6 @@ void FAST_t(InputArray _img, std::vector<KeyPoint>& keypoints, int threshold, bo
         {
             j = 3;
 #if CV_SIMD128
-            if( hasSimd )
             {
                 if( patternSize == 16 )
                 {