TheTest & test_pack()
{
typedef typename V_RegTrait128<LaneType>::w_reg Rx2;
typedef typename Rx2::lane_type w_type;
Data<Rx2> dataA, dataB;
dataA += std::numeric_limits<LaneType>::is_signed ? -10 : 10;
dataB *= 10;
Rx2 a = dataA, b = dataB;
Data<R> resC = v_pack(a, b);
Data<R> resD = v_rshr_pack<s>(a, b);
Data<R> resE(0);
v_pack_store(resE.d, b);
Data<R> resF(0);
v_rshr_pack_store<s>(resF.d, b);
const int n = Rx2::nlanes;
const w_type add = (w_type)1 << (s - 1);
for (int i = 0; i < n; ++i)
{
EXPECT_EQ(saturate_cast<LaneType>(dataA[i]), resC[i]);
EXPECT_EQ(saturate_cast<LaneType>(dataB[i]), resC[i + n]);
EXPECT_EQ(saturate_cast<LaneType>((dataA[i] + add) >> s), resD[i]);
EXPECT_EQ(saturate_cast<LaneType>((dataB[i] + add) >> s), resD[i + n]);
EXPECT_EQ(saturate_cast<LaneType>(dataB[i]), resE[i]);
EXPECT_EQ((LaneType)0, resE[i + n]);
EXPECT_EQ(saturate_cast<LaneType>((dataB[i] + add) >> s), resF[i]);
EXPECT_EQ((LaneType)0, resF[i + n]);
}
return *this;
}
void log64f( const double *x, double *y, int n )
{
CV_INSTRUMENT_REGION();
const double* const logTab = cv::details::getLogTab64f();
const int64 LOGTAB_MASK2_64F = ((int64)1 << (52 - LOGTAB_SCALE)) - 1;
const double
A7 = 1.0,
A6 = -0.5,
A5 = 0.333333333333333314829616256247390992939472198486328125,
A4 = -0.25,
A3 = 0.2,
A2 = -0.1666666666666666574148081281236954964697360992431640625,
A1 = 0.1428571428571428769682682968777953647077083587646484375,
A0 = -0.125;
int i = 0;
#if CV_SIMD_64F
const int VECSZ = v_float64::nlanes;
const v_float64 vln2 = vx_setall_f64(ln_2);
const v_float64
vA0 = vx_setall_f64(A0), vA1 = vx_setall_f64(A1),
vA2 = vx_setall_f64(A2), vA3 = vx_setall_f64(A3),
vA4 = vx_setall_f64(A4), vA5 = vx_setall_f64(A5),
vA6 = vx_setall_f64(A6), vA7 = vx_setall_f64(A7);
for( ; i < n; i += VECSZ )
{
if( i + VECSZ > n )
{
if( i == 0 || x == y )
break;
i = n - VECSZ;
}
v_int64 h0 = vx_load((const int64*)x + i);
v_int32 yi0 = v_pack(v_shr<52>(h0), vx_setzero_s64());
yi0 = (yi0 & vx_setall_s32(0x7ff)) - vx_setall_s32(1023);
v_int64 xi0 = (h0 & vx_setall_s64(LOGTAB_MASK2_64F)) | vx_setall_s64((int64)1023 << 52);
h0 = v_shr<52 - LOGTAB_SCALE - 1>(h0);
v_int32 idx = v_pack(h0, h0) & vx_setall_s32(LOGTAB_MASK*2);
v_float64 xf0, yf0;
v_lut_deinterleave(logTab, idx, yf0, xf0);
yf0 = v_fma(v_cvt_f64(yi0), vln2, yf0);
v_float64 delta = v_cvt_f64(idx == vx_setall_s32(510))*vx_setall_f64(1./512);
xf0 = v_fma(v_reinterpret_as_f64(xi0) - vx_setall_f64(1.), xf0, delta);
v_float64 xq = xf0*xf0;
v_float64 zf0 = v_fma(xq, vA0, vA2);
v_float64 zf1 = v_fma(xq, vA1, vA3);
zf0 = v_fma(zf0, xq, vA4);
zf1 = v_fma(zf1, xq, vA5);
zf0 = v_fma(zf0, xq, vA6);
zf1 = v_fma(zf1, xq, vA7);
zf1 = v_fma(zf1, xf0, yf0);
zf0 = v_fma(zf0, xq, zf1);
v_store(y + i, zf0);
}
#endif
for( ; i < n; i++ )
{
Cv64suf buf;
int64 i0 = ((const int64*)x)[i];
buf.i = (i0 & LOGTAB_MASK2_64F) | ((int64)1023 << 52);
int idx = (int)(i0 >> (52 - LOGTAB_SCALE - 1)) & (LOGTAB_MASK*2);
double y0 = (((int)(i0 >> 52) & 0x7ff) - 1023) * ln_2 + logTab[idx];
double x0 = (buf.f - 1.)*logTab[idx + 1] + (idx == 510 ? -1./512 : 0.);
double xq = x0*x0;
y[i] = (((A0*xq + A2)*xq + A4)*xq + A6)*xq + (((A1*xq + A3)*xq + A5)*xq + A7)*x0 + y0;
}
}
void operator()(const Range &boundaries) const
{
CV_TRACE_FUNCTION();
Mat dx, dy;
AutoBuffer<short> dxMax(0), dyMax(0);
std::deque<uchar*> stack, borderPeaksLocal;
const int rowStart = max(0, boundaries.start - 1), rowEnd = min(src.rows, boundaries.end + 1);
int *_mag_p, *_mag_a, *_mag_n;
short *_dx, *_dy, *_dx_a = NULL, *_dy_a = NULL, *_dx_n = NULL, *_dy_n = NULL;
uchar *_pmap;
double scale = 1.0;
CV_TRACE_REGION("gradient")
if(needGradient)
{
if (aperture_size == 7)
{
scale = 1 / 16.0;
}
Sobel(src.rowRange(rowStart, rowEnd), dx, CV_16S, 1, 0, aperture_size, scale, 0, BORDER_REPLICATE);
Sobel(src.rowRange(rowStart, rowEnd), dy, CV_16S, 0, 1, aperture_size, scale, 0, BORDER_REPLICATE);
}
else
{
dx = src.rowRange(rowStart, rowEnd);
dy = src2.rowRange(rowStart, rowEnd);
}
CV_TRACE_REGION_NEXT("magnitude");
if(cn > 1)
{
dxMax.allocate(2 * dx.cols);
dyMax.allocate(2 * dy.cols);
_dx_a = (short*)dxMax;
_dx_n = _dx_a + dx.cols;
_dy_a = (short*)dyMax;
_dy_n = _dy_a + dy.cols;
}
// _mag_p: previous row, _mag_a: actual row, _mag_n: next row
#if CV_SIMD128
AutoBuffer<int> buffer(3 * (mapstep * cn + CV_MALLOC_SIMD128));
_mag_p = alignPtr((int*)buffer + 1, CV_MALLOC_SIMD128);
_mag_a = alignPtr(_mag_p + mapstep * cn, CV_MALLOC_SIMD128);
_mag_n = alignPtr(_mag_a + mapstep * cn, CV_MALLOC_SIMD128);
#else
AutoBuffer<int> buffer(3 * (mapstep * cn));
_mag_p = (int*)buffer + 1;
_mag_a = _mag_p + mapstep * cn;
_mag_n = _mag_a + mapstep * cn;
#endif
// For the first time when just 2 rows are filled and for left and right borders
if(rowStart == boundaries.start)
memset(_mag_n - 1, 0, mapstep * sizeof(int));
else
_mag_n[src.cols] = _mag_n[-1] = 0;
_mag_a[src.cols] = _mag_a[-1] = _mag_p[src.cols] = _mag_p[-1] = 0;
// calculate magnitude and angle of gradient, perform non-maxima suppression.
// fill the map with one of the following values:
// 0 - the pixel might belong to an edge
// 1 - the pixel can not belong to an edge
// 2 - the pixel does belong to an edge
for (int i = rowStart; i <= boundaries.end; ++i)
{
// Scroll the ring buffer
std::swap(_mag_n, _mag_a);
std::swap(_mag_n, _mag_p);
if(i < rowEnd)
{
// Next row calculation
_dx = dx.ptr<short>(i - rowStart);
_dy = dy.ptr<short>(i - rowStart);
if (L2gradient)
{
int j = 0, width = src.cols * cn;
#if CV_SIMD128
if (haveSIMD)
{
for ( ; j <= width - 8; j += 8)
{
v_int16x8 v_dx = v_load((const short*)(_dx + j));
v_int16x8 v_dy = v_load((const short*)(_dy + j));
v_int32x4 v_dxp_low, v_dxp_high;
v_int32x4 v_dyp_low, v_dyp_high;
v_expand(v_dx, v_dxp_low, v_dxp_high);
v_expand(v_dy, v_dyp_low, v_dyp_high);
v_store_aligned((int *)(_mag_n + j), v_dxp_low*v_dxp_low+v_dyp_low*v_dyp_low);
v_store_aligned((int *)(_mag_n + j + 4), v_dxp_high*v_dxp_high+v_dyp_high*v_dyp_high);
}
}
#endif
for ( ; j < width; ++j)
_mag_n[j] = int(_dx[j])*_dx[j] + int(_dy[j])*_dy[j];
}
else
{
int j = 0, width = src.cols * cn;
#if CV_SIMD128
if (haveSIMD)
{
for(; j <= width - 8; j += 8)
{
v_int16x8 v_dx = v_load((const short *)(_dx + j));
v_int16x8 v_dy = v_load((const short *)(_dy + j));
v_dx = v_reinterpret_as_s16(v_abs(v_dx));
v_dy = v_reinterpret_as_s16(v_abs(v_dy));
v_int32x4 v_dx_ml, v_dy_ml, v_dx_mh, v_dy_mh;
v_expand(v_dx, v_dx_ml, v_dx_mh);
v_expand(v_dy, v_dy_ml, v_dy_mh);
v_store_aligned((int *)(_mag_n + j), v_dx_ml + v_dy_ml);
v_store_aligned((int *)(_mag_n + j + 4), v_dx_mh + v_dy_mh);
}
}
#endif
for ( ; j < width; ++j)
_mag_n[j] = std::abs(int(_dx[j])) + std::abs(int(_dy[j]));
}
if(cn > 1)
{
std::swap(_dx_n, _dx_a);
std::swap(_dy_n, _dy_a);
for(int j = 0, jn = 0; j < src.cols; ++j, jn += cn)
{
int maxIdx = jn;
for(int k = 1; k < cn; ++k)
if(_mag_n[jn + k] > _mag_n[maxIdx]) maxIdx = jn + k;
_mag_n[j] = _mag_n[maxIdx];
_dx_n[j] = _dx[maxIdx];
_dy_n[j] = _dy[maxIdx];
}
_mag_n[src.cols] = 0;
}
// at the very beginning we do not have a complete ring
// buffer of 3 magnitude rows for non-maxima suppression
if (i <= boundaries.start)
continue;
}
else
{
memset(_mag_n - 1, 0, mapstep * sizeof(int));
if(cn > 1)
{
std::swap(_dx_n, _dx_a);
std::swap(_dy_n, _dy_a);
}
}
// From here actual src row is (i - 1)
// Set left and right border to 1
#if CV_SIMD128
if(haveSIMD)
_pmap = map.ptr<uchar>(i) + CV_MALLOC_SIMD128;
else
#endif
_pmap = map.ptr<uchar>(i) + 1;
_pmap[src.cols] =_pmap[-1] = 1;
if(cn == 1)
{
_dx = dx.ptr<short>(i - rowStart - 1);
_dy = dy.ptr<short>(i - rowStart - 1);
}
else
{
_dx = _dx_a;
_dy = _dy_a;
}
const int TG22 = 13573;
int j = 0;
#if CV_SIMD128
if (haveSIMD)
{
const v_int32x4 v_low = v_setall_s32(low);
const v_int8x16 v_one = v_setall_s8(1);
for (; j <= src.cols - 32; j += 32)
{
v_int32x4 v_m1 = v_load_aligned((const int*)(_mag_a + j));
v_int32x4 v_m2 = v_load_aligned((const int*)(_mag_a + j + 4));
v_int32x4 v_m3 = v_load_aligned((const int*)(_mag_a + j + 8));
v_int32x4 v_m4 = v_load_aligned((const int*)(_mag_a + j + 12));
v_int32x4 v_cmp1 = v_m1 > v_low;
v_int32x4 v_cmp2 = v_m2 > v_low;
v_int32x4 v_cmp3 = v_m3 > v_low;
v_int32x4 v_cmp4 = v_m4 > v_low;
v_m1 = v_load_aligned((const int*)(_mag_a + j + 16));
v_m2 = v_load_aligned((const int*)(_mag_a + j + 20));
v_m3 = v_load_aligned((const int*)(_mag_a + j + 24));
v_m4 = v_load_aligned((const int*)(_mag_a + j + 28));
v_store_aligned((signed char*)(_pmap + j), v_one);
v_store_aligned((signed char*)(_pmap + j + 16), v_one);
v_int16x8 v_cmp80 = v_pack(v_cmp1, v_cmp2);
v_int16x8 v_cmp81 = v_pack(v_cmp3, v_cmp4);
v_cmp1 = v_m1 > v_low;
v_cmp2 = v_m2 > v_low;
v_cmp3 = v_m3 > v_low;
v_cmp4 = v_m4 > v_low;
v_int8x16 v_cmp = v_pack(v_cmp80, v_cmp81);
v_cmp80 = v_pack(v_cmp1, v_cmp2);
v_cmp81 = v_pack(v_cmp3, v_cmp4);
unsigned int mask = v_signmask(v_cmp);
v_cmp = v_pack(v_cmp80, v_cmp81);
mask |= v_signmask(v_cmp) << 16;
if (mask)
{
int k = j;
do
{
int l = trailingZeros32(mask);
k += l;
mask >>= l;
int m = _mag_a[k];
short xs = _dx[k];
short ys = _dy[k];
int x = (int)std::abs(xs);
int y = (int)std::abs(ys) << 15;
int tg22x = x * TG22;
if (y < tg22x)
{
if (m > _mag_a[k - 1] && m >= _mag_a[k + 1])
{
CANNY_CHECK_SIMD(m, high, (_pmap+k), stack);
}
}
else
{
int tg67x = tg22x + (x << 16);
if (y > tg67x)
{
if (m > _mag_p[k] && m >= _mag_n[k])
{
CANNY_CHECK_SIMD(m, high, (_pmap+k), stack);
}
}
else
{
int s = (xs ^ ys) < 0 ? -1 : 1;
if(m > _mag_p[k - s] && m > _mag_n[k + s])
{
CANNY_CHECK_SIMD(m, high, (_pmap+k), stack);
}
}
}
++k;
} while((mask >>= 1));
}
}
if (j <= src.cols - 16)
{
v_int32x4 v_m1 = v_load_aligned((const int*)(_mag_a + j));
v_int32x4 v_m2 = v_load_aligned((const int*)(_mag_a + j + 4));
v_int32x4 v_m3 = v_load_aligned((const int*)(_mag_a + j + 8));
v_int32x4 v_m4 = v_load_aligned((const int*)(_mag_a + j + 12));
v_store_aligned((signed char*)(_pmap + j), v_one);
v_int32x4 v_cmp1 = v_m1 > v_low;
v_int32x4 v_cmp2 = v_m2 > v_low;
v_int32x4 v_cmp3 = v_m3 > v_low;
v_int32x4 v_cmp4 = v_m4 > v_low;
v_int16x8 v_cmp80 = v_pack(v_cmp1, v_cmp2);
v_int16x8 v_cmp81 = v_pack(v_cmp3, v_cmp4);
v_int8x16 v_cmp = v_pack(v_cmp80, v_cmp81);
unsigned int mask = v_signmask(v_cmp);
if (mask)
{
int k = j;
do
{
int l = trailingZeros32(mask);
k += l;
mask >>= l;
int m = _mag_a[k];
short xs = _dx[k];
short ys = _dy[k];
int x = (int)std::abs(xs);
int y = (int)std::abs(ys) << 15;
int tg22x = x * TG22;
if (y < tg22x)
{
if (m > _mag_a[k - 1] && m >= _mag_a[k + 1])
{
CANNY_CHECK_SIMD(m, high, (_pmap+k), stack);
}
}
else
{
int tg67x = tg22x + (x << 16);
if (y > tg67x)
{
if (m > _mag_p[k] && m >= _mag_n[k])
{
CANNY_CHECK_SIMD(m, high, (_pmap+k), stack);
}
}
else
{
int s = (xs ^ ys) < 0 ? -1 : 1;
if(m > _mag_p[k - s] && m > _mag_n[k + s])
{
CANNY_CHECK_SIMD(m, high, (_pmap+k), stack);
}
}
}
++k;
} while((mask >>= 1));
}
j += 16;
}
}
#endif
for (; j < src.cols; j++)
{
int m = _mag_a[j];
if (m > low)
{
short xs = _dx[j];
short ys = _dy[j];
int x = (int)std::abs(xs);
int y = (int)std::abs(ys) << 15;
int tg22x = x * TG22;
if (y < tg22x)
{
if (m > _mag_a[j - 1] && m >= _mag_a[j + 1])
{
CANNY_CHECK(m, high, (_pmap+j), stack);
}
}
else
{
int tg67x = tg22x + (x << 16);
if (y > tg67x)
{
if (m > _mag_p[j] && m >= _mag_n[j])
{
CANNY_CHECK(m, high, (_pmap+j), stack);
}
}
else
{
int s = (xs ^ ys) < 0 ? -1 : 1;
if(m > _mag_p[j - s] && m > _mag_n[j + s])
{
CANNY_CHECK(m, high, (_pmap+j), stack);
}
}
}
}
_pmap[j] = 1;
}
}
