mirrors/gnss-sdr
1
0
Fork 0
mirror of https://github.com/gnss-sdr/gnss-sdr synced 2025-01-29 02:14:51 +00:00
Code Issues Releases Wiki Activity

Add AVX2 implementation

Browse Source
This commit is contained in:
Carles Fernandez 2018-07-20 09:39:37 +02:00
parent f9e4bdfa59
commit 927f4d6b21
No known key found for this signature in database
GPG Key ID: 4C583C52B0C3877D
1 changed files with 67 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

67
src/algorithms/libs/volk_gnsssdr_module/volk_gnsssdr/kernels/volk_gnsssdr/volk_gnsssdr_16ic_convert_32fc.h
View File

@ -179,6 +179,73 @@ static inline void volk_gnsssdr_16ic_convert_32fc_a_axv(lv_32fc_t* outputVector,
}
#endif /* LV_HAVE_AVX */
#ifdef LV_HAVE_AVX2
#include <immintrin.h>
static inline void volk_gnsssdr_16ic_convert_32fc_a_avx2(lv_32fc_t* outputVector, const lv_16sc_t* inputVector, unsigned int num_points)
{
const unsigned int avx_iters = num_points / 8;
unsigned int number = 0;
const int16_t* complexVectorPtr = (int16_t*)inputVector;
float* outputVectorPtr = (float*)outputVector;
__m256 outVal;
__m256i outValInt;
__m128i cplxValue;
for (number = 0; number < avx_iters; number++)
{
cplxValue = _mm_load_si128((__m128i*)complexVectorPtr);
complexVectorPtr += 8;
outValInt = _mm256_cvtepi16_epi32(cplxValue);
outVal = _mm256_cvtepi32_ps(outValInt);
_mm256_store_ps((float*)outputVectorPtr, outVal);
outputVectorPtr += 8;
}
number = avx_iters * 8;
for (; number < num_points * 2; number++)
{
*outputVectorPtr++ = (float)*complexVectorPtr++;
}
}
#endif /* LV_HAVE_AVX2 */
#ifdef LV_HAVE_AVX2
#include <immintrin.h>
static inline void volk_gnsssdr_16ic_convert_32fc_u_avx2(lv_32fc_t* outputVector, const lv_16sc_t* inputVector, unsigned int num_points)
{
const unsigned int avx_iters = num_points / 8;
unsigned int number = 0;
const int16_t* complexVectorPtr = (int16_t*)inputVector;
float* outputVectorPtr = (float*)outputVector;
__m256 outVal;
__m256i outValInt;
__m128i cplxValue;
for (number = 0; number < avx_iters; number++)
{
cplxValue = _mm_loadu_si128((__m128i*)complexVectorPtr);
complexVectorPtr += 8;
outValInt = _mm256_cvtepi16_epi32(cplxValue);
outVal = _mm256_cvtepi32_ps(outValInt);
_mm256_storeu_ps((float*)outputVectorPtr, outVal);
outputVectorPtr += 8;
}
number = avx_iters * 8;
for (; number < num_points * 2; number++)
{
*outputVectorPtr++ = (float)*complexVectorPtr++;
}
}
#endif /* LV_HAVE_AVX2 */
#ifdef LV_HAVE_NEONV7
#include <arm_neon.h>