diff options
| author | Tom Rondeau <trondeau@vt.edu> | 2011-01-25 10:37:49 -0500 |
|---|---|---|
| committer | Tom Rondeau <trondeau@vt.edu> | 2011-01-25 10:37:49 -0500 |
| commit | 108a594c0838ad21f93cba6597d1f66af097b157 (patch) | |
| tree | 95dc252161e64647bdcf697b21c7882534344182 /volk/include | |
| parent | 34fc10385fbe41ba5475b21135c802e9660235ad (diff) | |
volk: New volk kernel for conjugate dot products with unaligned buffers.
Note: need to convert this to new naming standard.
Diffstat (limited to 'volk/include')
| -rw-r--r-- | volk/include/volk/Makefile.am | 3 | ||||
| -rw-r--r-- | volk/include/volk/volk_32fc_conjugate_dot_prod_unaligned.h | 144 |
2 files changed, 146 insertions, 1 deletions
diff --git a/volk/include/volk/Makefile.am b/volk/include/volk/Makefile.am index 658974d3a4..00289be1eb 100644 --- a/volk/include/volk/Makefile.am +++ b/volk/include/volk/Makefile.am @@ -66,6 +66,7 @@ volkinclude_HEADERS = \ volk_32f_calc_spectral_noise_floor_aligned16.h \ volk_32fc_atan2_32f_aligned16.h \ volk_32fc_conjugate_dot_prod_aligned16.h \ + volk_32fc_conjugate_dot_prod_unaligned.h \ volk_32fc_deinterleave_32f_aligned16.h \ volk_32fc_deinterleave_64f_aligned16.h \ volk_32fc_deinterleave_real_16s_aligned16.h \ @@ -157,4 +158,4 @@ distclean-local: rm -f Makefile.in rm -f volk_environment_init.h rm -f volk_mktables - rm -f $(BUILT_SOURCES)
\ No newline at end of file + rm -f $(BUILT_SOURCES) diff --git a/volk/include/volk/volk_32fc_conjugate_dot_prod_unaligned.h b/volk/include/volk/volk_32fc_conjugate_dot_prod_unaligned.h new file mode 100644 index 0000000000..ead1573fec --- /dev/null +++ b/volk/include/volk/volk_32fc_conjugate_dot_prod_unaligned.h @@ -0,0 +1,144 @@ +#ifndef INCLUDED_VOLK_32fc_CONJUGATE_DOT_PROD_UNALIGNED_H +#define INCLUDED_VOLK_32fc_CONJUGATE_DOT_PROD_UNALIGNED_H + + +#include<volk/volk_complex.h> + + +#if LV_HAVE_GENERIC + + +static inline void volk_32fc_conjugate_dot_prod_unaligned_generic(lv_32fc_t* result, const lv_32fc_t* input, const lv_32fc_t* taps, unsigned int num_bytes) { + + float * res = (float*) result; + float * in = (float*) input; + float * tp = (float*) taps; + unsigned int n_2_ccomplex_blocks = num_bytes >> 4; + unsigned int isodd = (num_bytes >> 3) &1; + + + + float sum0[2] = {0,0}; + float sum1[2] = {0,0}; + int i = 0; + + + for(i = 0; i < n_2_ccomplex_blocks; ++i) { + + sum0[0] += in[0] * tp[0] + in[1] * tp[1]; + sum0[1] += (-in[0] * tp[1]) + in[1] * tp[0]; + sum1[0] += in[2] * tp[2] + in[3] * tp[3]; + sum1[1] += (-in[2] * tp[3]) + in[3] * tp[2]; + + + in += 4; + tp += 4; + + } + + + res[0] = sum0[0] + sum1[0]; + res[1] = sum0[1] + sum1[1]; + + + + for(i = 0; i < isodd; ++i) { + + + *result += input[(num_bytes >> 3) - 1] * lv_conj(taps[(num_bytes >> 3) - 1]); + + } + /* + for(i = 0; i < num_bytes >> 3; ++i) { + *result += input[i] * conjf(taps[i]); + } + */ +} + +#endif /*LV_HAVE_GENERIC*/ + +#if LV_HAVE_SSE3 + +#include <xmmintrin.h> +#include <pmmintrin.h> +#include <mmintrin.h> + + +static inline void volk_32fc_conjugate_dot_prod_unaligned_sse3(lv_32fc_t* result, const lv_32fc_t* input, const lv_32fc_t* taps, unsigned int num_bytes) { + + static const uint32_t conjugator[4] __attribute__((aligned(16)))= {0x00000000, 0x80000000, 0x00000000, 0x80000000}; + + union HalfMask { + uint32_t intRep[4]; + __m128 vec; + } halfMask; + + union NegMask { + int intRep[4]; + __m128 vec; + } negMask; + + unsigned int offset = 0; + float Rsum=0, Isum=0; + float Im,Re; + + __m128 in1, in2, Rv, fehg, Iv, Rs, Ivm, Is; + __m128 zv = {0,0,0,0}; + + halfMask.intRep[0] = halfMask.intRep[1] = 0xFFFFFFFF; + halfMask.intRep[2] = halfMask.intRep[3] = 0x00000000; + + negMask.intRep[0] = negMask.intRep[2] = 0x80000000; + negMask.intRep[1] = negMask.intRep[3] = 0; + + // main loop + while(num_bytes >= 4*sizeof(float)){ + + in1 = _mm_loadu_ps( (float*) (input+offset) ); + in2 = _mm_loadu_ps( (float*) (taps+offset) ); + Rv = in1*in2; + fehg = _mm_shuffle_ps(in2, in2, _MM_SHUFFLE(2,3,0,1)); + Iv = in1*fehg; + Rs = _mm_hadd_ps( _mm_hadd_ps(Rv, zv) ,zv); + Ivm = _mm_xor_ps( negMask.vec, Iv ); + Is = _mm_hadd_ps( _mm_hadd_ps(Ivm, zv) ,zv); + _mm_store_ss( &Im, Is ); + _mm_store_ss( &Re, Rs ); + num_bytes -= 4*sizeof(float); + offset += 2; + Rsum += Re; + Isum += Im; + } + + // handle the last complex case ... + if(num_bytes > 0){ + + if(num_bytes != 4){ + // bad things are happening + } + + in1 = _mm_loadu_ps( (float*) (input+offset) ); + in2 = _mm_loadu_ps( (float*) (taps+offset) ); + Rv = _mm_and_ps(in1*in2, halfMask.vec); + fehg = _mm_shuffle_ps(in2, in2, _MM_SHUFFLE(2,3,0,1)); + Iv = _mm_and_ps(in1*fehg, halfMask.vec); + Rs = _mm_hadd_ps(_mm_hadd_ps(Rv, zv),zv); + Ivm = _mm_xor_ps( negMask.vec, Iv ); + Is = _mm_hadd_ps(_mm_hadd_ps(Ivm, zv),zv); + _mm_store_ss( &Im, Is ); + _mm_store_ss( &Re, Rs ); + Rsum += Re; + Isum += Im; + } + + result[0] = lv_32fc_init(Rsum,Isum); + return; +} + +#endif /*LV_HAVE_SSE3*/ + + +#endif /*INCLUDED_VOLK_32fc_CONJUGATE_DOT_PROD_UNALIGNED_H*/ + + + |