root / volk / include / volk / volk_32f_s32f_multiply_32f_u.h @ 374afd87
History | View | Annotate | Download (3 kB)
| 1 | #ifndef INCLUDED_volk_32f_s32f_multiply_32f_u_H
|
|---|---|
| 2 | #define INCLUDED_volk_32f_s32f_multiply_32f_u_H
|
| 3 | |
| 4 | #include <inttypes.h> |
| 5 | #include <stdio.h> |
| 6 | |
| 7 | #ifdef LV_HAVE_SSE
|
| 8 | #include <xmmintrin.h> |
| 9 | /*!
|
| 10 | \brief Scalar float multiply |
| 11 | \param cVector The vector where the results will be stored |
| 12 | \param aVector One of the vectors to be multiplied |
| 13 | \param scalar the scalar value |
| 14 | \param num_points The number of values in aVector and bVector to be multiplied together and stored into cVector |
| 15 | */ |
| 16 | static inline void volk_32f_s32f_multiply_32f_u_sse(float* cVector, const float* aVector, const float scalar, unsigned int num_points){ |
| 17 | unsigned int number = 0; |
| 18 | const unsigned int quarterPoints = num_points / 4; |
| 19 | |
| 20 | float* cPtr = cVector;
|
| 21 | const float* aPtr = aVector; |
| 22 | |
| 23 | __m128 aVal, bVal, cVal; |
| 24 | bVal = _mm_set_ps1(scalar); |
| 25 | for(;number < quarterPoints; number++){
|
| 26 | |
| 27 | aVal = _mm_loadu_ps(aPtr); |
| 28 | |
| 29 | cVal = _mm_mul_ps(aVal, bVal); |
| 30 | |
| 31 | _mm_storeu_ps(cPtr,cVal); // Store the results back into the C container
|
| 32 | |
| 33 | aPtr += 4;
|
| 34 | cPtr += 4;
|
| 35 | } |
| 36 | |
| 37 | number = quarterPoints * 4;
|
| 38 | for(;number < num_points; number++){
|
| 39 | *cPtr++ = (*aPtr++) * scalar; |
| 40 | } |
| 41 | } |
| 42 | #endif /* LV_HAVE_SSE */ |
| 43 | |
| 44 | #ifdef LV_HAVE_AVX
|
| 45 | #include <immintrin.h> |
| 46 | /*!
|
| 47 | \brief Scalar float multiply |
| 48 | \param cVector The vector where the results will be stored |
| 49 | \param aVector One of the vectors to be multiplied |
| 50 | \param scalar the scalar value |
| 51 | \param num_points The number of values in aVector and bVector to be multiplied together and stored into cVector |
| 52 | */ |
| 53 | static inline void volk_32f_s32f_multiply_32f_u_avx(float* cVector, const float* aVector, const float scalar, unsigned int num_points){ |
| 54 | unsigned int number = 0; |
| 55 | const unsigned int eighthPoints = num_points / 8; |
| 56 | |
| 57 | float* cPtr = cVector;
|
| 58 | const float* aPtr = aVector; |
| 59 | |
| 60 | __m256 aVal, bVal, cVal; |
| 61 | bVal = _mm256_set1_ps(scalar); |
| 62 | for(;number < eighthPoints; number++){
|
| 63 | |
| 64 | aVal = _mm256_loadu_ps(aPtr); |
| 65 | |
| 66 | cVal = _mm256_mul_ps(aVal, bVal); |
| 67 | |
| 68 | _mm256_storeu_ps(cPtr,cVal); // Store the results back into the C container
|
| 69 | |
| 70 | aPtr += 8;
|
| 71 | cPtr += 8;
|
| 72 | } |
| 73 | |
| 74 | number = eighthPoints * 8;
|
| 75 | for(;number < num_points; number++){
|
| 76 | *cPtr++ = (*aPtr++) * scalar; |
| 77 | } |
| 78 | } |
| 79 | #endif /* LV_HAVE_AVX */ |
| 80 | |
| 81 | #ifdef LV_HAVE_GENERIC
|
| 82 | /*!
|
| 83 | \brief Scalar float multiply |
| 84 | \param cVector The vector where the results will be stored |
| 85 | \param aVector One of the vectors to be multiplied |
| 86 | \param scalar the scalar value |
| 87 | \param num_points The number of values in aVector and bVector to be multiplied together and stored into cVector |
| 88 | */ |
| 89 | static inline void volk_32f_s32f_multiply_32f_u_generic(float* cVector, const float* aVector, const float scalar, unsigned int num_points){ |
| 90 | unsigned int number = 0; |
| 91 | const float* inputPtr = aVector; |
| 92 | float* outputPtr = cVector;
|
| 93 | for(number = 0; number < num_points; number++){ |
| 94 | *outputPtr = (*inputPtr) * scalar; |
| 95 | inputPtr++; |
| 96 | outputPtr++; |
| 97 | } |
| 98 | } |
| 99 | #endif /* LV_HAVE_GENERIC */ |
| 100 | |
| 101 | |
| 102 | #endif /* INCLUDED_volk_32f_s32f_multiply_32f_u_H */ |