mirrors/gnss-sdr
1
0
Fork 0
mirror of https://github.com/gnss-sdr/gnss-sdr synced 2025-10-31 15:23:04 +00:00
Code Issues Releases Wiki Activity

Add NEON intrinsics

Browse Source
This commit is contained in:
Carles Fernandez
2019-08-23 22:29:07 +02:00
parent 11afe63ef3
commit 91ccc8589e
1 changed files with 43 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

43
src/algorithms/libs/volk_gnsssdr_module/volk_gnsssdr/include/volk_gnsssdr/volk_gnsssdr_neon_intrinsics.h
View File

@@ -53,4 +53,47 @@ static inline float32x4_t vsqrtq_f32(float32x4_t q_x)
return vmulq_f32(q_x, q_step_2);
}
/* Magnitude squared for float32x4x2_t */
static inline float32x4_t _vmagnitudesquaredq_f32(float32x4x2_t cmplxValue)
{
float32x4_t iValue, qValue, result;
iValue = vmulq_f32(cmplxValue.val[0], cmplxValue.val[0]); // Square the values
qValue = vmulq_f32(cmplxValue.val[1], cmplxValue.val[1]); // Square the values
result = vaddq_f32(iValue, qValue); // Add the I2 and Q2 values
return result;
}
/* Inverse square root for float32x4_t */
static inline float32x4_t _vinvsqrtq_f32(float32x4_t x)
{
float32x4_t sqrt_reciprocal = vrsqrteq_f32(x);
sqrt_reciprocal = vmulq_f32(vrsqrtsq_f32(vmulq_f32(x, sqrt_reciprocal), sqrt_reciprocal), sqrt_reciprocal);
sqrt_reciprocal = vmulq_f32(vrsqrtsq_f32(vmulq_f32(x, sqrt_reciprocal), sqrt_reciprocal), sqrt_reciprocal);
return sqrt_reciprocal;
}
/* Complex multiplication for float32x4x2_t */
static inline float32x4x2_t _vmultiply_complexq_f32(float32x4x2_t a_val, float32x4x2_t b_val)
{
float32x4x2_t tmp_real;
float32x4x2_t tmp_imag;
float32x4x2_t c_val;
// multiply the real*real and imag*imag to get real result
// a0r*b0r|a1r*b1r|a2r*b2r|a3r*b3r
tmp_real.val[0] = vmulq_f32(a_val.val[0], b_val.val[0]);
// a0i*b0i|a1i*b1i|a2i*b2i|a3i*b3i
tmp_real.val[1] = vmulq_f32(a_val.val[1], b_val.val[1]);
// Multiply cross terms to get the imaginary result
// a0r*b0i|a1r*b1i|a2r*b2i|a3r*b3i
tmp_imag.val[0] = vmulq_f32(a_val.val[0], b_val.val[1]);
// a0i*b0r|a1i*b1r|a2i*b2r|a3i*b3r
tmp_imag.val[1] = vmulq_f32(a_val.val[1], b_val.val[0]);
// combine the products
c_val.val[0] = vsubq_f32(tmp_real.val[0], tmp_real.val[1]);
c_val.val[1] = vaddq_f32(tmp_imag.val[0], tmp_imag.val[1]);
return c_val;
}
#endif /* INCLUDED_VOLK_GNSSSDR_NEON_INTRINSICS_H_ */