diff options
| author | Tom Rondeau <trondeau@vt.edu> | 2012-01-30 00:22:44 -0500 |
|---|---|---|
| committer | Tom Rondeau <trondeau@vt.edu> | 2012-01-30 00:22:44 -0500 |
| commit | cb458204245632ac7a571bfe96b90d3c55a2f01a (patch) | |
| tree | 4a538c78bf91eb4a4901a1ef2a41586ff2963e48 /volk/include | |
| parent | 83d0bf5d513e516d700e552fa2773dd852a6608a (diff) | |
volk: adding complex to imag kernel.
Diffstat (limited to 'volk/include')
| -rw-r--r-- | volk/include/volk/Makefile.am | 1 | ||||
| -rw-r--r-- | volk/include/volk/volk_32fc_deinterleave_imag_32f_a.h | 68 |
2 files changed, 69 insertions, 0 deletions
diff --git a/volk/include/volk/Makefile.am b/volk/include/volk/Makefile.am index c2502f6e66..c5b99c41bb 100644 --- a/volk/include/volk/Makefile.am +++ b/volk/include/volk/Makefile.am @@ -65,6 +65,7 @@ volkinclude_HEADERS = \ volk_32fc_deinterleave_64f_x2_a.h \ volk_32fc_s32f_deinterleave_real_16i_a.h \ volk_32fc_deinterleave_real_32f_a.h \ + volk_32fc_deinterleave_imag_32f_a.h \ volk_32fc_deinterleave_real_64f_a.h \ volk_32fc_x2_dot_prod_32fc_a.h \ volk_32fc_x2_dot_prod_32fc_u.h \ diff --git a/volk/include/volk/volk_32fc_deinterleave_imag_32f_a.h b/volk/include/volk/volk_32fc_deinterleave_imag_32f_a.h new file mode 100644 index 0000000000..adc4112b95 --- /dev/null +++ b/volk/include/volk/volk_32fc_deinterleave_imag_32f_a.h @@ -0,0 +1,68 @@ +#ifndef INCLUDED_volk_32fc_deinterleave_imag_32f_a_H +#define INCLUDED_volk_32fc_deinterleave_imag_32f_a_H + +#include <inttypes.h> +#include <stdio.h> + +#ifdef LV_HAVE_SSE +#include <xmmintrin.h> +/*! + \brief Deinterleaves the complex vector into Q vector data + \param complexVector The complex input vector + \param qBuffer The Q buffer output data + \param num_points The number of complex data values to be deinterleaved +*/ +static inline void volk_32fc_deinterleave_imag_32f_a_sse(float* qBuffer, const lv_32fc_t* complexVector, unsigned int num_points){ + unsigned int number = 0; + const unsigned int quarterPoints = num_points / 4; + + const float* complexVectorPtr = (const float*)complexVector; + float* qBufferPtr = qBuffer; + + __m128 cplxValue1, cplxValue2, iValue; + for(;number < quarterPoints; number++){ + + cplxValue1 = _mm_load_ps(complexVectorPtr); + complexVectorPtr += 4; + + cplxValue2 = _mm_load_ps(complexVectorPtr); + complexVectorPtr += 4; + + // Arrange in q1q2q3q4 format + iValue = _mm_shuffle_ps(cplxValue1, cplxValue2, _MM_SHUFFLE(3,1,3,1)); + + _mm_store_ps(qBufferPtr, iValue); + + qBufferPtr += 4; + } + + number = quarterPoints * 4; + for(; number < num_points; number++){ + complexVectorPtr++; + *qBufferPtr++ = *complexVectorPtr++; + } +} +#endif /* LV_HAVE_SSE */ + +#ifdef LV_HAVE_GENERIC +/*! + \brief Deinterleaves the complex vector into Q vector data + \param complexVector The complex input vector + \param qBuffer The I buffer output data + \param num_points The number of complex data values to be deinterleaved +*/ +static inline void volk_32fc_deinterleave_imag_32f_a_generic(float* qBuffer, const lv_32fc_t* complexVector, unsigned int num_points){ + unsigned int number = 0; + const float* complexVectorPtr = (float*)complexVector; + float* qBufferPtr = qBuffer; + for(number = 0; number < num_points; number++){ + complexVectorPtr++; + *qBufferPtr++ = *complexVectorPtr++; + } +} +#endif /* LV_HAVE_GENERIC */ + + + + +#endif /* INCLUDED_volk_32fc_deinterleave_imag_32f_a_H */ |