void sqrt64f(const double* src, double* dst, int len)
{
CV_INSTRUMENT_REGION();
int i = 0;
#if CV_SIMD_64F
const int VECSZ = v_float64::nlanes;
for( ; i < len; i += VECSZ*2 )
{
if( i + VECSZ*2 > len )
{
if( i == 0 || src == dst )
break;
i = len - VECSZ*2;
}
v_float64 t0 = vx_load(src + i), t1 = vx_load(src + i + VECSZ);
t0 = v_sqrt(t0);
t1 = v_sqrt(t1);
v_store(dst + i, t0); v_store(dst + i + VECSZ, t1);
}
vx_cleanup();
#endif
for( ; i < len; i++ )
dst[i] = std::sqrt(src[i]);
}
void magnitude64f(const double* x, const double* y, double* mag, int len)
{
CV_INSTRUMENT_REGION();
int i = 0;
#if CV_SIMD_64F
const int VECSZ = v_float64::nlanes;
for( ; i < len; i += VECSZ*2 )
{
if( i + VECSZ*2 > len )
{
if( i == 0 || mag == x || mag == y )
break;
i = len - VECSZ*2;
}
v_float64 x0 = vx_load(x + i), x1 = vx_load(x + i + VECSZ);
v_float64 y0 = vx_load(y + i), y1 = vx_load(y + i + VECSZ);
x0 = v_sqrt(v_muladd(x0, x0, y0*y0));
x1 = v_sqrt(v_muladd(x1, x1, y1*y1));
v_store(mag + i, x0);
v_store(mag + i + VECSZ, x1);
}
vx_cleanup();
#endif
for( ; i < len; i++ )
{
double x0 = x[i], y0 = y[i];
mag[i] = std::sqrt(x0*x0 + y0*y0);
}
}
TheTest & test_sqrt_abs()
{
Data<R> dataA, dataD;
dataD *= -1.0;
R a = dataA, d = dataD;
Data<R> resB = v_sqrt(a), resC = v_invsqrt(a), resE = v_abs(d);
for (int i = 0; i < R::nlanes; ++i)
{
EXPECT_COMPARE_EQ((float)std::sqrt(dataA[i]), (float)resB[i]);
EXPECT_COMPARE_EQ(1/(float)std::sqrt(dataA[i]), (float)resC[i]);
EXPECT_COMPARE_EQ((float)abs(dataA[i]), (float)resE[i]);
}
return *this;
}
double
SpectralDifferenceAudioCurve::processDouble(const double *mag, int increment)
{
double result = 0.0;
const int hs1 = m_lastPerceivedBin + 1;
v_convert(m_tmpbuf, mag, hs1);
v_square(m_tmpbuf, hs1);
v_subtract(m_mag, m_tmpbuf, hs1);
v_abs(m_mag, hs1);
v_sqrt(m_mag, hs1);
for (int i = 0; i < hs1; ++i) {
result += m_mag[i];
}
v_copy(m_mag, m_tmpbuf, hs1);
return result;
}
inline v_float32x4 v_magnitude(const v_float32x4& a, const v_float32x4& b)
{
v_float32x4 x(vmlaq_f32(vmulq_f32(a.val, a.val), b.val, b.val));
return v_sqrt(x);
}
