root / volk / include / volk / volk_64f_convert_32f_unaligned16.h @ 23914465
History | View | Annotate | Download (2 kB)
| 1 | #ifndef INCLUDED_VOLK_64f_CONVERT_32f_UNALIGNED16_H
|
|---|---|
| 2 | #define INCLUDED_VOLK_64f_CONVERT_32f_UNALIGNED16_H
|
| 3 | |
| 4 | #include <inttypes.h> |
| 5 | #include <stdio.h> |
| 6 | |
| 7 | #if LV_HAVE_SSE2
|
| 8 | #include <emmintrin.h> |
| 9 | /*!
|
| 10 | \brief Converts the double values into float values |
| 11 | \param dVector The converted float vector values |
| 12 | \param fVector The double vector values to be converted |
| 13 | \param num_points The number of points in the two vectors to be converted |
| 14 | */ |
| 15 | static inline void volk_64f_convert_32f_unaligned16_sse2(float* outputVector, const double* inputVector, unsigned int num_points){ |
| 16 | unsigned int number = 0; |
| 17 | |
| 18 | const unsigned int quarterPoints = num_points / 4; |
| 19 | |
| 20 | const double* inputVectorPtr = (const double*)inputVector; |
| 21 | float* outputVectorPtr = outputVector;
|
| 22 | __m128 ret, ret2; |
| 23 | __m128d inputVal1, inputVal2; |
| 24 | |
| 25 | for(;number < quarterPoints; number++){
|
| 26 | inputVal1 = _mm_loadu_pd(inputVectorPtr); inputVectorPtr += 2;
|
| 27 | inputVal2 = _mm_loadu_pd(inputVectorPtr); inputVectorPtr += 2;
|
| 28 | |
| 29 | ret = _mm_cvtpd_ps(inputVal1); |
| 30 | ret2 = _mm_cvtpd_ps(inputVal2); |
| 31 | |
| 32 | ret = _mm_movelh_ps(ret, ret2); |
| 33 | |
| 34 | _mm_storeu_ps(outputVectorPtr, ret); |
| 35 | outputVectorPtr += 4;
|
| 36 | } |
| 37 | |
| 38 | number = quarterPoints * 4;
|
| 39 | for(; number < num_points; number++){
|
| 40 | outputVector[number] = (float)(inputVector[number]);
|
| 41 | } |
| 42 | } |
| 43 | #endif /* LV_HAVE_SSE2 */ |
| 44 | |
| 45 | |
| 46 | #ifdef LV_HAVE_GENERIC
|
| 47 | /*!
|
| 48 | \brief Converts the double values into float values |
| 49 | \param dVector The converted float vector values |
| 50 | \param fVector The double vector values to be converted |
| 51 | \param num_points The number of points in the two vectors to be converted |
| 52 | */ |
| 53 | static inline void volk_64f_convert_32f_unaligned16_generic(float* outputVector, const double* inputVector, unsigned int num_points){ |
| 54 | float* outputVectorPtr = outputVector;
|
| 55 | const double* inputVectorPtr = inputVector; |
| 56 | unsigned int number = 0; |
| 57 | |
| 58 | for(number = 0; number < num_points; number++){ |
| 59 | *outputVectorPtr++ = ((float)(*inputVectorPtr++));
|
| 60 | } |
| 61 | } |
| 62 | #endif /* LV_HAVE_GENERIC */ |
| 63 | |
| 64 | |
| 65 | |
| 66 | |
| 67 | #endif /* INCLUDED_VOLK_64f_CONVERT_32f_UNALIGNED16_H */ |