diff options
| author | Abhishek Bhowmick <abhowmick22@gmail.com> | 2014-06-12 18:03:56 +0530 |
|---|---|---|
| committer | Tom Rondeau <tom@trondeau.com> | 2014-10-15 10:29:19 -0400 |
| commit | 43934eb5896cf6261c3d9b7662e2ee95fb344dfd (patch) | |
| tree | c9525e8abb1423335229192e4963d0e9454c5b37 | |
| parent | a43d53348d5fabdaeccca1c3ad2c661e1d292e79 (diff) | |
volk: Added avx proto-kernel for fast exp.
| -rw-r--r-- | volk/kernels/volk/volk_32f_expfast_32f.h | 78 |
1 files changed, 78 insertions, 0 deletions
diff --git a/volk/kernels/volk/volk_32f_expfast_32f.h b/volk/kernels/volk/volk_32f_expfast_32f.h index 0826527feb..01ed79aa1a 100644 --- a/volk/kernels/volk/volk_32f_expfast_32f.h +++ b/volk/kernels/volk/volk_32f_expfast_32f.h @@ -12,6 +12,45 @@ #ifndef INCLUDED_volk_32f_expfast_32f_a_H #define INCLUDED_volk_32f_expfast_32f_a_H +#ifdef LV_HAVE_AVX +#include <immintrin.h> +/*! + \brief Computes fast exp (max 7% error) of input vector and stores results in output vector + \param bVector The vector where results will be stored + \param aVector The input vector of floats + \param num_points Number of points for which log is to be computed +*/ +static inline void volk_32f_expfast_32f_a_avx(float* bVector, const float* aVector, unsigned int num_points){ + + float* bPtr = bVector; + const float* aPtr = aVector; + + unsigned int number = 0; + const unsigned int eighthPoints = num_points / 8; + + __m256 aVal, bVal, a, b; + __m256i exp; + a = _mm256_set1_ps(A/Mln2); + b = _mm256_set1_ps(B-C); + + for(;number < eighthPoints; number++){ + aVal = _mm256_load_ps(aPtr); + exp = _mm256_cvtps_epi32(_mm256_add_ps(_mm256_mul_ps(a,aVal), b)); + bVal = _mm256_castsi256_ps(exp); + + _mm256_store_ps(bPtr, bVal); + aPtr += 8; + bPtr += 8; + } + + number = eighthPoints * 8; + for(;number < num_points; number++){ + *bPtr++ = expf(*aPtr++); + } +} + +#endif /* LV_HAVE_AVX for aligned */ + #ifdef LV_HAVE_SSE4_1 #include <smmintrin.h> /*! @@ -76,6 +115,45 @@ static inline void volk_32f_expfast_32f_a_generic(float* bVector, const float* a #ifndef INCLUDED_volk_32f_expfast_32f_u_H #define INCLUDED_volk_32f_expfast_32f_u_H +#ifdef LV_HAVE_AVX +#include <immintrin.h> +/*! + \brief Computes fast exp (max 7% error) of input vector and stores results in output vector + \param bVector The vector where results will be stored + \param aVector The input vector of floats + \param num_points Number of points for which log is to be computed +*/ +static inline void volk_32f_expfast_32f_u_avx(float* bVector, const float* aVector, unsigned int num_points){ + + float* bPtr = bVector; + const float* aPtr = aVector; + + unsigned int number = 0; + const unsigned int eighthPoints = num_points / 8; + + __m256 aVal, bVal, a, b; + __m256i exp; + a = _mm256_set1_ps(A/Mln2); + b = _mm256_set1_ps(B-C); + + for(;number < eighthPoints; number++){ + aVal = _mm256_loadu_ps(aPtr); + exp = _mm256_cvtps_epi32(_mm256_add_ps(_mm256_mul_ps(a,aVal), b)); + bVal = _mm256_castsi256_ps(exp); + + _mm256_storeu_ps(bPtr, bVal); + aPtr += 8; + bPtr += 8; + } + + number = eighthPoints * 8; + for(;number < num_points; number++){ + *bPtr++ = expf(*aPtr++); + } +} + +#endif /* LV_HAVE_AVX for aligned */ + #ifdef LV_HAVE_SSE4_1 #include <smmintrin.h> /*! |