mirrors/gnss-sdr
1
0
Fork 0
mirror of https://github.com/gnss-sdr/gnss-sdr synced 2024-12-14 12:10:34 +00:00
Code Issues Releases Wiki Activity

fix sse implementations

Browse Source
This commit is contained in:
Carles Fernandez 2016-01-16 20:57:55 +01:00
parent a817d49e89
commit 38d4d8aa9a
4 changed files with 18 additions and 16 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

8
src/algorithms/libs/volk_gnsssdr_module/volk_gnsssdr/kernels/volk_gnsssdr/volk_gnsssdr_8i_x2_add_8i.h
View File

@ -70,7 +70,7 @@ static inline void volk_gnsssdr_8i_x2_add_8i_u_sse2(char* cVector, const char* a
cPtr += 16;
}
for(unsigned int i = 0; i<(num_points % 16); ++i)
for(unsigned int i = sse_iters * 16; i < num_points; ++i)
{
*cPtr++ = (*aPtr++) + (*bPtr++);
}
@ -134,14 +134,14 @@ static inline void volk_gnsssdr_8i_x2_add_8i_a_sse2(char* cVector, const char* a
cVal = _mm_add_epi8(aVal, bVal);
_mm_store_si128((__m128i*)cPtr,cVal); // Store the results back into the C container
_mm_store_si128((__m128i*)cPtr, cVal); // Store the results back into the C container
aPtr += 16;
bPtr += 16;
cPtr += 16;
}
for(unsigned int i = 0; i<(num_points % 16); ++i)
for(unsigned int i = sse_iters * 16; i < num_points; ++i)
{
*cPtr++ = (*aPtr++) + (*bPtr++);
}
@ -163,7 +163,7 @@ static inline void volk_gnsssdr_8i_x2_add_8i_a_generic(char* cVector, const char
const char* bPtr= bVector;
unsigned int number = 0;
for(number = 0; number < num_points; number++)
for(; number < num_points; number++)
{
*cPtr++ = (*aPtr++) + (*bPtr++);
}

12
src/algorithms/libs/volk_gnsssdr_module/volk_gnsssdr/kernels/volk_gnsssdr/volk_gnsssdr_8ic_conjugate_8ic.h
View File

@ -75,7 +75,7 @@ static inline void volk_gnsssdr_8ic_conjugate_8ic_u_avx(lv_8sc_t* cVector, const
c += 16;
}
for (unsigned int i = 0; i<(num_points % 16); ++i)
for (unsigned int i = sse_iters * 16; i < num_points; ++i)
{
*c++ = lv_conj(*a++);
}
@ -109,7 +109,7 @@ static inline void volk_gnsssdr_8ic_conjugate_8ic_u_ssse3(lv_8sc_t* cVector, con
c += 8;
}
for (unsigned int i = 0; i<(num_points % 8); ++i)
for (unsigned int i = sse_iters * 8; i < num_points; ++i)
{
*c++ = lv_conj(*a++);
}
@ -146,7 +146,7 @@ static inline void volk_gnsssdr_8ic_conjugate_8ic_u_sse3(lv_8sc_t* cVector, cons
c += 8;
}
for (unsigned int i = 0; i<(num_points % 8); ++i)
for (unsigned int i = sse_iters * 8; i < num_points; ++i)
{
*c++ = lv_conj(*a++);
}
@ -220,7 +220,7 @@ static inline void volk_gnsssdr_8ic_conjugate_8ic_a_avx(lv_8sc_t* cVector, const
c += 16;
}
for (unsigned int i = 0; i<(num_points % 16); ++i)
for (unsigned int i = sse_iters * 16; i < num_points; ++i)
{
*c++ = lv_conj(*a++);
}
@ -254,7 +254,7 @@ static inline void volk_gnsssdr_8ic_conjugate_8ic_a_ssse3(lv_8sc_t* cVector, con
c += 8;
}
for (unsigned int i = 0; i<(num_points % 8); ++i)
for (unsigned int i = sse_iters * 8; i < num_points; ++i)
{
*c++ = lv_conj(*a++);
}
@ -291,7 +291,7 @@ static inline void volk_gnsssdr_8ic_conjugate_8ic_a_sse3(lv_8sc_t* cVector, cons
c += 8;
}
for (unsigned int i = 0; i<(num_points % 8); ++i)
for (unsigned int i = sse_iters * 8; i < num_points; ++i)
{
*c++ = lv_conj(*a++);
}

4
src/algorithms/libs/volk_gnsssdr_module/volk_gnsssdr/kernels/volk_gnsssdr/volk_gnsssdr_8ic_magnitude_squared_8i.h
View File

@ -90,7 +90,7 @@ static inline void volk_gnsssdr_8ic_magnitude_squared_8i_u_sse3(char* magnitudeV
magnitudeVectorPtr += 16;
}
for (unsigned int i = 0; i<(num_points % 16); ++i)
for (unsigned int i = sse_iters * 16; i < num_points; ++i)
{
const char valReal = *complexVectorPtr++;
const char valImag = *complexVectorPtr++;
@ -226,7 +226,7 @@ static inline void volk_gnsssdr_8ic_magnitude_squared_8i_a_sse3(char* magnitudeV
magnitudeVectorPtr += 16;
}
for (unsigned int i = 0; i<(num_points % 16); ++i)
for (unsigned int i = sse_iters * 16; i < num_points; ++i)
{
const char valReal = *complexVectorPtr++;
const char valImag = *complexVectorPtr++;

10
src/algorithms/libs/volk_gnsssdr_module/volk_gnsssdr/kernels/volk_gnsssdr/volk_gnsssdr_8ic_s8ic_multiply_8ic.h
View File

@ -51,6 +51,7 @@
*/
static inline void volk_gnsssdr_8ic_s8ic_multiply_8ic_u_sse3(lv_8sc_t* cVector, const lv_8sc_t* aVector, const lv_8sc_t scalar, unsigned int num_points)
{
unsigned int number = 0;
const unsigned int sse_iters = num_points / 8;
__m128i x, y, mult1, realx, imagx, realy, imagy, realx_mult_realy, imagx_mult_imagy, realx_mult_imagy, imagx_mult_realy, realc, imagc, totalc;
@ -65,7 +66,7 @@ static inline void volk_gnsssdr_8ic_s8ic_multiply_8ic_u_sse3(lv_8sc_t* cVector,
imagy = _mm_and_si128 (imagy, mult1);
realy = _mm_and_si128 (y, mult1);
for(unsigned int number = 0;number < sse_iters; number++)
for(; number < sse_iters; number++)
{
x = _mm_lddqu_si128((__m128i*)a);
@ -92,7 +93,7 @@ static inline void volk_gnsssdr_8ic_s8ic_multiply_8ic_u_sse3(lv_8sc_t* cVector,
c += 8;
}
for (unsigned int i = 0; i<(num_points % 8); ++i)
for (number = sse_iters * 8; number < num_points; ++number)
{
*c++ = (*a++) * scalar;
}
@ -164,6 +165,7 @@ static inline void volk_gnsssdr_8ic_s8ic_multiply_8ic_generic(lv_8sc_t* cVector,
*/
static inline void volk_gnsssdr_8ic_s8ic_multiply_8ic_a_sse3(lv_8sc_t* cVector, const lv_8sc_t* aVector, const lv_8sc_t scalar, unsigned int num_points)
{
unsigned int number = 0;
const unsigned int sse_iters = num_points / 8;
__m128i x, y, mult1, realx, imagx, realy, imagy, realx_mult_realy, imagx_mult_imagy, realx_mult_imagy, imagx_mult_realy, realc, imagc, totalc;
@ -178,7 +180,7 @@ static inline void volk_gnsssdr_8ic_s8ic_multiply_8ic_a_sse3(lv_8sc_t* cVector,
imagy = _mm_and_si128 (imagy, mult1);
realy = _mm_and_si128 (y, mult1);
for(unsigned int number = 0;number < sse_iters; number++)
for(; number < sse_iters; number++)
{
x = _mm_load_si128((__m128i*)a);
@ -205,7 +207,7 @@ static inline void volk_gnsssdr_8ic_s8ic_multiply_8ic_a_sse3(lv_8sc_t* cVector,
c += 8;
}
for (unsigned int i = 0; i<(num_points % 8); ++i)
for (number = sse_iters * 8; number < num_points; ++number)
{
*c++ = (*a++) * scalar;
}