mirrors/gnss-sdr
1
0
Fork 0
mirror of https://github.com/gnss-sdr/gnss-sdr synced 2024-06-26 15:03:14 +00:00
Code Issues Releases Wiki Activity

adding AVX2 protokerns

Browse Source 
I haven't found a way to do the rotator part better than with SSE3. Only
the dot product takes real advantage of 256-bit registers. Even tough,
the gain with respect to SSE3 is about 12%.
This commit is contained in:
Carles Fernandez 2016-03-26 01:51:01 +01:00
parent 4411af40b3
commit 751764343c
2 changed files with 643 additions and 4 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

523
src/algorithms/libs/volk_gnsssdr_module/volk_gnsssdr/kernels/volk_gnsssdr/volk_gnsssdr_16ic_x2_rotator_dot_prod_16ic_xn.h
View File

@ -71,11 +71,10 @@
#include <volk_gnsssdr/volk_gnsssdr.h>
#include <volk_gnsssdr/volk_gnsssdr_malloc.h>
#include <volk_gnsssdr/volk_gnsssdr_complex.h>
#include <volk_gnsssdr/saturation_arithmetic.h>
#include <math.h>
#include <stdio.h>
//#include <stdio.h>
#ifdef LV_HAVE_GENERIC
@ -719,7 +718,7 @@ static inline void volk_gnsssdr_16ic_x2_rotator_dot_prod_16ic_xn_u_sse3(lv_16sc_
a = _mm_or_si128(realcacc[n_vec], imagcacc[n_vec]);
_mm_storeu_si128((__m128i*)dotProductVector, a); // Store the results back into the dot product vector
_mm_store_si128((__m128i*)dotProductVector, a); // Store the results back into the dot product vector
dotProduct = lv_cmake(0,0);
for (int i = 0; i < 4; ++i)
{
@ -731,7 +730,7 @@ static inline void volk_gnsssdr_16ic_x2_rotator_dot_prod_16ic_xn_u_sse3(lv_16sc_
volk_gnsssdr_free(realcacc);
volk_gnsssdr_free(imagcacc);
_mm_storeu_ps((float*)two_phase_acc, two_phase_acc_reg);
_mm_store_ps((float*)two_phase_acc, two_phase_acc_reg);
(*phase) = two_phase_acc[0];
for(unsigned int n = sse_iters * 4; n < num_points; n++)
@ -751,6 +750,522 @@ static inline void volk_gnsssdr_16ic_x2_rotator_dot_prod_16ic_xn_u_sse3(lv_16sc_
#endif /* LV_HAVE_SSE3 */
#ifdef LV_HAVE_AVX2
#include <immintrin.h>
static inline void volk_gnsssdr_16ic_x2_rotator_dot_prod_16ic_xn_a_avx2(lv_16sc_t* result, const lv_16sc_t* in_common, const lv_32fc_t phase_inc, lv_32fc_t* phase, const lv_16sc_t** in_a, int num_a_vectors, unsigned int num_points)
{
const unsigned int avx2_iters = num_points / 8;
const lv_16sc_t** _in_a = in_a;
const lv_16sc_t* _in_common = in_common;
lv_16sc_t* _out = result;
lv_16sc_t tmp16;
lv_32fc_t tmp32;
__VOLK_ATTR_ALIGNED(32) lv_16sc_t dotProductVector[8];
lv_16sc_t dotProduct = lv_cmake(0,0);
__m256i* realcacc = (__m256i*)volk_gnsssdr_malloc(num_a_vectors * sizeof(__m256i), volk_gnsssdr_get_alignment());
__m256i* imagcacc = (__m256i*)volk_gnsssdr_malloc(num_a_vectors * sizeof(__m256i), volk_gnsssdr_get_alignment());
for (int n_vec = 0; n_vec < num_a_vectors; n_vec++)
{
realcacc[n_vec] = _mm256_setzero_si256();
imagcacc[n_vec] = _mm256_setzero_si256();
}
const __m256i mask_imag = _mm256_set_epi8(255, 255, 0, 0, 255, 255, 0, 0, 255, 255, 0, 0, 255, 255, 0, 0, 255, 255, 0, 0, 255, 255, 0, 0, 255, 255, 0, 0, 255, 255, 0, 0);
const __m256i mask_real = _mm256_set_epi8(0, 0, 255, 255, 0, 0, 255, 255, 0, 0, 255, 255, 0, 0, 255, 255, 0, 0, 255, 255, 0, 0, 255, 255, 0, 0, 255, 255, 0, 0, 255, 255);
__m128 a, b, two_phase_acc_reg, two_phase_inc_reg;
__m128i c1, c2, result1, result2;
__attribute__((aligned(16))) lv_32fc_t two_phase_inc[2];
two_phase_inc[0] = phase_inc * phase_inc;
two_phase_inc[1] = phase_inc * phase_inc;
two_phase_inc_reg = _mm_load_ps((float*) two_phase_inc);
__attribute__((aligned(16))) lv_32fc_t two_phase_acc[2];
two_phase_acc[0] = (*phase);
two_phase_acc[1] = (*phase) * phase_inc;
two_phase_acc_reg = _mm_load_ps((float*) two_phase_acc);
__m256i a2, b2, c, c_sr, real, imag;
__m128 yl, yh, tmp1, tmp2, tmp3;
for(unsigned int number = 0; number < avx2_iters; number++)
{
a = _mm_set_ps((float)(lv_cimag(_in_common[1])), (float)(lv_creal(_in_common[1])), (float)(lv_cimag(_in_common[0])), (float)(lv_creal(_in_common[0]))); // //load (2 byte imag, 2 byte real) x 2 into 128 bits reg
//complex 32fc multiplication b=a*two_phase_acc_reg
yl = _mm_moveldup_ps(two_phase_acc_reg); // Load yl with cr,cr,dr,dr
yh = _mm_movehdup_ps(two_phase_acc_reg); // Load yh with ci,ci,di,di
tmp1 = _mm_mul_ps(a, yl); // tmp1 = ar*cr,ai*cr,br*dr,bi*dr
a = _mm_shuffle_ps(a, a, 0xB1); // Re-arrange x to be ai,ar,bi,br
tmp2 = _mm_mul_ps(a, yh); // tmp2 = ai*ci,ar*ci,bi*di,br*di
b = _mm_addsub_ps(tmp1, tmp2); // ar*cr-ai*ci, ai*cr+ar*ci, br*dr-bi*di, bi*dr+br*di
c1 = _mm_cvtps_epi32(b); // convert from 32fc to 32ic
//complex 32fc multiplication two_phase_acc_reg=two_phase_acc_reg*two_phase_inc_reg
yl = _mm_moveldup_ps(two_phase_acc_reg); // Load yl with cr,cr,dr,dr
yh = _mm_movehdup_ps(two_phase_acc_reg); // Load yh with ci,ci,di,di
tmp1 = _mm_mul_ps(two_phase_inc_reg, yl); // tmp1 = ar*cr,ai*cr,br*dr,bi*dr
tmp3 = _mm_shuffle_ps(two_phase_inc_reg, two_phase_inc_reg, 0xB1); // Re-arrange x to be ai,ar,bi,br
tmp2 = _mm_mul_ps(tmp3, yh); // tmp2 = ai*ci,ar*ci,bi*di,br*di
two_phase_acc_reg = _mm_addsub_ps(tmp1, tmp2); // ar*cr-ai*ci, ai*cr+ar*ci, br*dr-bi*di, bi*dr+br*di
//next two samples
_in_common += 2;
a = _mm_set_ps((float)(lv_cimag(_in_common[1])), (float)(lv_creal(_in_common[1])), (float)(lv_cimag(_in_common[0])), (float)(lv_creal(_in_common[0]))); // //load (2 byte imag, 2 byte real) x 2 into 128 bits reg
//complex 32fc multiplication b=a*two_phase_acc_reg
yl = _mm_moveldup_ps(two_phase_acc_reg); // Load yl with cr,cr,dr,dr
yh = _mm_movehdup_ps(two_phase_acc_reg); // Load yh with ci,ci,di,di
tmp1 = _mm_mul_ps(a, yl); // tmp1 = ar*cr,ai*cr,br*dr,bi*dr
a = _mm_shuffle_ps(a, a, 0xB1); // Re-arrange x to be ai,ar,bi,br
tmp2 = _mm_mul_ps(a, yh); // tmp2 = ai*ci,ar*ci,bi*di,br*di
b = _mm_addsub_ps(tmp1, tmp2); // ar*cr-ai*ci, ai*cr+ar*ci, br*dr-bi*di, bi*dr+br*di
c2 = _mm_cvtps_epi32(b); // convert from 32fc to 32ic
//complex 32fc multiplication two_phase_acc_reg=two_phase_acc_reg*two_phase_inc_reg
yl = _mm_moveldup_ps(two_phase_acc_reg); // Load yl with cr,cr,dr,dr
yh = _mm_movehdup_ps(two_phase_acc_reg); // Load yh with ci,ci,di,di
tmp1 = _mm_mul_ps(two_phase_inc_reg, yl); // tmp1 = ar*cr,ai*cr,br*dr,bi*dr
tmp3 = _mm_shuffle_ps(two_phase_inc_reg, two_phase_inc_reg, 0xB1); // Re-arrange x to be ai,ar,bi,br
tmp2 = _mm_mul_ps(tmp3, yh); // tmp2 = ai*ci,ar*ci,bi*di,br*di
two_phase_acc_reg = _mm_addsub_ps(tmp1, tmp2); // ar*cr-ai*ci, ai*cr+ar*ci, br*dr-bi*di, bi*dr+br*di
// store four output samples
result1 = _mm_packs_epi32(c1, c2);// convert from 32ic to 16ic
_in_common += 2;
a = _mm_set_ps((float)(lv_cimag(_in_common[1])), (float)(lv_creal(_in_common[1])), (float)(lv_cimag(_in_common[0])), (float)(lv_creal(_in_common[0]))); // //load (2 byte imag, 2 byte real) x 2 into 128 bits reg
//complex 32fc multiplication b=a*two_phase_acc_reg
yl = _mm_moveldup_ps(two_phase_acc_reg); // Load yl with cr,cr,dr,dr
yh = _mm_movehdup_ps(two_phase_acc_reg); // Load yh with ci,ci,di,di
tmp1 = _mm_mul_ps(a, yl); // tmp1 = ar*cr,ai*cr,br*dr,bi*dr
a = _mm_shuffle_ps(a, a, 0xB1); // Re-arrange x to be ai,ar,bi,br
tmp2 = _mm_mul_ps(a, yh); // tmp2 = ai*ci,ar*ci,bi*di,br*di
b = _mm_addsub_ps(tmp1, tmp2); // ar*cr-ai*ci, ai*cr+ar*ci, br*dr-bi*di, bi*dr+br*di
c1 = _mm_cvtps_epi32(b); // convert from 32fc to 32ic
//complex 32fc multiplication two_phase_acc_reg=two_phase_acc_reg*two_phase_inc_reg
yl = _mm_moveldup_ps(two_phase_acc_reg); // Load yl with cr,cr,dr,dr
yh = _mm_movehdup_ps(two_phase_acc_reg); // Load yh with ci,ci,di,di
tmp1 = _mm_mul_ps(two_phase_inc_reg, yl); // tmp1 = ar*cr,ai*cr,br*dr,bi*dr
tmp3 = _mm_shuffle_ps(two_phase_inc_reg, two_phase_inc_reg, 0xB1); // Re-arrange x to be ai,ar,bi,br
tmp2 = _mm_mul_ps(tmp3, yh); // tmp2 = ai*ci,ar*ci,bi*di,br*di
two_phase_acc_reg = _mm_addsub_ps(tmp1, tmp2); // ar*cr-ai*ci, ai*cr+ar*ci, br*dr-bi*di, bi*dr+br*di
//next two samples
_in_common += 2;
a = _mm_set_ps((float)(lv_cimag(_in_common[1])), (float)(lv_creal(_in_common[1])), (float)(lv_cimag(_in_common[0])), (float)(lv_creal(_in_common[0]))); // //load (2 byte imag, 2 byte real) x 2 into 128 bits reg
__builtin_prefetch(_in_common + 16);
//complex 32fc multiplication b=a*two_phase_acc_reg
yl = _mm_moveldup_ps(two_phase_acc_reg); // Load yl with cr,cr,dr,dr
yh = _mm_movehdup_ps(two_phase_acc_reg); // Load yh with ci,ci,di,di
tmp1 = _mm_mul_ps(a, yl); // tmp1 = ar*cr,ai*cr,br*dr,bi*dr
a = _mm_shuffle_ps(a, a, 0xB1); // Re-arrange x to be ai,ar,bi,br
tmp2 = _mm_mul_ps(a, yh); // tmp2 = ai*ci,ar*ci,bi*di,br*di
b = _mm_addsub_ps(tmp1, tmp2); // ar*cr-ai*ci, ai*cr+ar*ci, br*dr-bi*di, bi*dr+br*di
c2 = _mm_cvtps_epi32(b); // convert from 32fc to 32ic
//complex 32fc multiplication two_phase_acc_reg=two_phase_acc_reg*two_phase_inc_reg
yl = _mm_moveldup_ps(two_phase_acc_reg); // Load yl with cr,cr,dr,dr
yh = _mm_movehdup_ps(two_phase_acc_reg); // Load yh with ci,ci,di,di
tmp1 = _mm_mul_ps(two_phase_inc_reg, yl); // tmp1 = ar*cr,ai*cr,br*dr,bi*dr
tmp3 = _mm_shuffle_ps(two_phase_inc_reg, two_phase_inc_reg, 0xB1); // Re-arrange x to be ai,ar,bi,br
tmp2 = _mm_mul_ps(tmp3, yh); // tmp2 = ai*ci,ar*ci,bi*di,br*di
two_phase_acc_reg = _mm_addsub_ps(tmp1, tmp2); // ar*cr-ai*ci, ai*cr+ar*ci, br*dr-bi*di, bi*dr+br*di
// store four output samples
result2 = _mm_packs_epi32(c1, c2);// convert from 32ic to 16ic
_in_common += 2;
b2 = _mm256_insertf128_si256(_mm256_castsi128_si256(result1), (result2), 1);
for (int n_vec = 0; n_vec < num_a_vectors; n_vec++)
{
a2 = _mm256_load_si256((__m256i*)&(_in_a[n_vec][number * 8]));
c = _mm256_mullo_epi16(a2, b2);
c_sr = _mm256_srli_si256(c, 2); // Shift a right by imm8 bytes while shifting in zeros, and store the results in dst.
real = _mm256_subs_epi16(c, c_sr);
c_sr = _mm256_slli_si256(b2, 2);
c = _mm256_mullo_epi16(a2, c_sr);
c_sr = _mm256_slli_si256(a2, 2);
imag = _mm256_mullo_epi16(b2, c_sr);
imag = _mm256_adds_epi16(c, imag);
realcacc[n_vec] = _mm256_adds_epi16(realcacc[n_vec], real);
imagcacc[n_vec] = _mm256_adds_epi16(imagcacc[n_vec], imag);
}
// Regenerate phase
if ((number % 128) == 0)
{
tmp1 = _mm_mul_ps(two_phase_acc_reg, two_phase_acc_reg);
tmp2 = _mm_hadd_ps(tmp1, tmp1);
tmp1 = _mm_shuffle_ps(tmp2, tmp2, 0xD8);
tmp2 = _mm_sqrt_ps(tmp1);
two_phase_acc_reg = _mm_div_ps(two_phase_acc_reg, tmp2);
}
}
for (int n_vec = 0; n_vec < num_a_vectors; n_vec++)
{
realcacc[n_vec] = _mm256_and_si256(realcacc[n_vec], mask_real);
imagcacc[n_vec] = _mm256_and_si256(imagcacc[n_vec], mask_imag);
a2 = _mm256_or_si256(realcacc[n_vec], imagcacc[n_vec]);
_mm256_store_si256((__m256i*)dotProductVector, a2); // Store the results back into the dot product vector
dotProduct = lv_cmake(0,0);
for (int i = 0; i < 8; ++i)
{
dotProduct = lv_cmake(sat_adds16i(lv_creal(dotProduct), lv_creal(dotProductVector[i])),
sat_adds16i(lv_cimag(dotProduct), lv_cimag(dotProductVector[i])));
}
_out[n_vec] = dotProduct;
}
volk_gnsssdr_free(realcacc);
volk_gnsssdr_free(imagcacc);
_mm_store_ps((float*)two_phase_acc, two_phase_acc_reg);
(*phase) = two_phase_acc[0];
for(unsigned int n = avx2_iters * 8; n < num_points; n++)
{
tmp16 = in_common[n];
tmp32 = lv_cmake((float)lv_creal(tmp16), (float)lv_cimag(tmp16)) * (*phase);
tmp16 = lv_cmake((int16_t)rintf(lv_creal(tmp32)), (int16_t)rintf(lv_cimag(tmp32)));
(*phase) *= phase_inc;
for (int n_vec = 0; n_vec < num_a_vectors; n_vec++)
{
lv_16sc_t tmp = tmp16 * in_a[n_vec][n];
_out[n_vec] = lv_cmake(sat_adds16i(lv_creal(_out[n_vec]), lv_creal(tmp)),
sat_adds16i(lv_cimag(_out[n_vec]), lv_cimag(tmp)));
}
}
}
#endif /* LV_HAVE_AVX2 */
#ifdef LV_HAVE_AVX2
#include <immintrin.h>
static inline void volk_gnsssdr_16ic_x2_rotator_dot_prod_16ic_xn_a_avx2_reload(lv_16sc_t* result, const lv_16sc_t* in_common, const lv_32fc_t phase_inc, lv_32fc_t* phase, const lv_16sc_t** in_a, int num_a_vectors, unsigned int num_points)
{
const unsigned int avx2_iters = num_points / 8;
const unsigned int ROTATOR_RELOAD = 128;
const lv_16sc_t** _in_a = in_a;
const lv_16sc_t* _in_common = in_common;
lv_16sc_t* _out = result;
lv_16sc_t tmp16;
lv_32fc_t tmp32;
__VOLK_ATTR_ALIGNED(32) lv_16sc_t dotProductVector[8];
lv_16sc_t dotProduct = lv_cmake(0,0);
__m256i* realcacc = (__m256i*)volk_gnsssdr_malloc(num_a_vectors * sizeof(__m256i), volk_gnsssdr_get_alignment());
__m256i* imagcacc = (__m256i*)volk_gnsssdr_malloc(num_a_vectors * sizeof(__m256i), volk_gnsssdr_get_alignment());
for (int n_vec = 0; n_vec < num_a_vectors; n_vec++)
{
realcacc[n_vec] = _mm256_setzero_si256();
imagcacc[n_vec] = _mm256_setzero_si256();
}
const __m256i mask_imag = _mm256_set_epi8(255, 255, 0, 0, 255, 255, 0, 0, 255, 255, 0, 0, 255, 255, 0, 0, 255, 255, 0, 0, 255, 255, 0, 0, 255, 255, 0, 0, 255, 255, 0, 0);
const __m256i mask_real = _mm256_set_epi8(0, 0, 255, 255, 0, 0, 255, 255, 0, 0, 255, 255, 0, 0, 255, 255, 0, 0, 255, 255, 0, 0, 255, 255, 0, 0, 255, 255, 0, 0, 255, 255);
__m128 a, b, two_phase_acc_reg, two_phase_inc_reg;
__m128i c1, c2, result1, result2;
__attribute__((aligned(16))) lv_32fc_t two_phase_inc[2];
two_phase_inc[0] = phase_inc * phase_inc;
two_phase_inc[1] = phase_inc * phase_inc;
two_phase_inc_reg = _mm_load_ps((float*) two_phase_inc);
__attribute__((aligned(16))) lv_32fc_t two_phase_acc[2];
two_phase_acc[0] = (*phase);
two_phase_acc[1] = (*phase) * phase_inc;
two_phase_acc_reg = _mm_load_ps((float*) two_phase_acc);
__m256i a2, b2, c, c_sr, real, imag;
__m128 yl, yh, tmp1, tmp2, tmp3;
for (unsigned int number = 0; number < avx2_iters / ROTATOR_RELOAD; ++number)
{
for (unsigned int j = 0; j < ROTATOR_RELOAD; j++)
{
a = _mm_set_ps((float)(lv_cimag(_in_common[1])), (float)(lv_creal(_in_common[1])), (float)(lv_cimag(_in_common[0])), (float)(lv_creal(_in_common[0]))); // //load (2 byte imag, 2 byte real) x 2 into 128 bits reg
//complex 32fc multiplication b=a*two_phase_acc_reg
yl = _mm_moveldup_ps(two_phase_acc_reg); // Load yl with cr,cr,dr,dr
yh = _mm_movehdup_ps(two_phase_acc_reg); // Load yh with ci,ci,di,di
tmp1 = _mm_mul_ps(a, yl); // tmp1 = ar*cr,ai*cr,br*dr,bi*dr
a = _mm_shuffle_ps(a, a, 0xB1); // Re-arrange x to be ai,ar,bi,br
tmp2 = _mm_mul_ps(a, yh); // tmp2 = ai*ci,ar*ci,bi*di,br*di
b = _mm_addsub_ps(tmp1, tmp2); // ar*cr-ai*ci, ai*cr+ar*ci, br*dr-bi*di, bi*dr+br*di
c1 = _mm_cvtps_epi32(b); // convert from 32fc to 32ic
//complex 32fc multiplication two_phase_acc_reg=two_phase_acc_reg*two_phase_inc_reg
yl = _mm_moveldup_ps(two_phase_acc_reg); // Load yl with cr,cr,dr,dr
yh = _mm_movehdup_ps(two_phase_acc_reg); // Load yh with ci,ci,di,di
tmp1 = _mm_mul_ps(two_phase_inc_reg, yl); // tmp1 = ar*cr,ai*cr,br*dr,bi*dr
tmp3 = _mm_shuffle_ps(two_phase_inc_reg, two_phase_inc_reg, 0xB1); // Re-arrange x to be ai,ar,bi,br
tmp2 = _mm_mul_ps(tmp3, yh); // tmp2 = ai*ci,ar*ci,bi*di,br*di
two_phase_acc_reg = _mm_addsub_ps(tmp1, tmp2); // ar*cr-ai*ci, ai*cr+ar*ci, br*dr-bi*di, bi*dr+br*di
//next two samples
_in_common += 2;
a = _mm_set_ps((float)(lv_cimag(_in_common[1])), (float)(lv_creal(_in_common[1])), (float)(lv_cimag(_in_common[0])), (float)(lv_creal(_in_common[0]))); // //load (2 byte imag, 2 byte real) x 2 into 128 bits reg
//complex 32fc multiplication b=a*two_phase_acc_reg
yl = _mm_moveldup_ps(two_phase_acc_reg); // Load yl with cr,cr,dr,dr
yh = _mm_movehdup_ps(two_phase_acc_reg); // Load yh with ci,ci,di,di
tmp1 = _mm_mul_ps(a, yl); // tmp1 = ar*cr,ai*cr,br*dr,bi*dr
a = _mm_shuffle_ps(a, a, 0xB1); // Re-arrange x to be ai,ar,bi,br
tmp2 = _mm_mul_ps(a, yh); // tmp2 = ai*ci,ar*ci,bi*di,br*di
b = _mm_addsub_ps(tmp1, tmp2); // ar*cr-ai*ci, ai*cr+ar*ci, br*dr-bi*di, bi*dr+br*di
c2 = _mm_cvtps_epi32(b); // convert from 32fc to 32ic
//complex 32fc multiplication two_phase_acc_reg=two_phase_acc_reg*two_phase_inc_reg
yl = _mm_moveldup_ps(two_phase_acc_reg); // Load yl with cr,cr,dr,dr
yh = _mm_movehdup_ps(two_phase_acc_reg); // Load yh with ci,ci,di,di
tmp1 = _mm_mul_ps(two_phase_inc_reg, yl); // tmp1 = ar*cr,ai*cr,br*dr,bi*dr
tmp3 = _mm_shuffle_ps(two_phase_inc_reg, two_phase_inc_reg, 0xB1); // Re-arrange x to be ai,ar,bi,br
tmp2 = _mm_mul_ps(tmp3, yh); // tmp2 = ai*ci,ar*ci,bi*di,br*di
two_phase_acc_reg = _mm_addsub_ps(tmp1, tmp2); // ar*cr-ai*ci, ai*cr+ar*ci, br*dr-bi*di, bi*dr+br*di
// store four output samples
result1 = _mm_packs_epi32(c1, c2);// convert from 32ic to 16ic
_in_common += 2;
a = _mm_set_ps((float)(lv_cimag(_in_common[1])), (float)(lv_creal(_in_common[1])), (float)(lv_cimag(_in_common[0])), (float)(lv_creal(_in_common[0]))); // //load (2 byte imag, 2 byte real) x 2 into 128 bits reg
//complex 32fc multiplication b=a*two_phase_acc_reg
yl = _mm_moveldup_ps(two_phase_acc_reg); // Load yl with cr,cr,dr,dr
yh = _mm_movehdup_ps(two_phase_acc_reg); // Load yh with ci,ci,di,di
tmp1 = _mm_mul_ps(a, yl); // tmp1 = ar*cr,ai*cr,br*dr,bi*dr
a = _mm_shuffle_ps(a, a, 0xB1); // Re-arrange x to be ai,ar,bi,br
tmp2 = _mm_mul_ps(a, yh); // tmp2 = ai*ci,ar*ci,bi*di,br*di
b = _mm_addsub_ps(tmp1, tmp2); // ar*cr-ai*ci, ai*cr+ar*ci, br*dr-bi*di, bi*dr+br*di
c1 = _mm_cvtps_epi32(b); // convert from 32fc to 32ic
//complex 32fc multiplication two_phase_acc_reg=two_phase_acc_reg*two_phase_inc_reg
yl = _mm_moveldup_ps(two_phase_acc_reg); // Load yl with cr,cr,dr,dr
yh = _mm_movehdup_ps(two_phase_acc_reg); // Load yh with ci,ci,di,di
tmp1 = _mm_mul_ps(two_phase_inc_reg, yl); // tmp1 = ar*cr,ai*cr,br*dr,bi*dr
tmp3 = _mm_shuffle_ps(two_phase_inc_reg, two_phase_inc_reg, 0xB1); // Re-arrange x to be ai,ar,bi,br
tmp2 = _mm_mul_ps(tmp3, yh); // tmp2 = ai*ci,ar*ci,bi*di,br*di
two_phase_acc_reg = _mm_addsub_ps(tmp1, tmp2); // ar*cr-ai*ci, ai*cr+ar*ci, br*dr-bi*di, bi*dr+br*di
//next two samples
_in_common += 2;
a = _mm_set_ps((float)(lv_cimag(_in_common[1])), (float)(lv_creal(_in_common[1])), (float)(lv_cimag(_in_common[0])), (float)(lv_creal(_in_common[0]))); // //load (2 byte imag, 2 byte real) x 2 into 128 bits reg
__builtin_prefetch(_in_common + 16);
//complex 32fc multiplication b=a*two_phase_acc_reg
yl = _mm_moveldup_ps(two_phase_acc_reg); // Load yl with cr,cr,dr,dr
yh = _mm_movehdup_ps(two_phase_acc_reg); // Load yh with ci,ci,di,di
tmp1 = _mm_mul_ps(a, yl); // tmp1 = ar*cr,ai*cr,br*dr,bi*dr
a = _mm_shuffle_ps(a, a, 0xB1); // Re-arrange x to be ai,ar,bi,br
tmp2 = _mm_mul_ps(a, yh); // tmp2 = ai*ci,ar*ci,bi*di,br*di
b = _mm_addsub_ps(tmp1, tmp2); // ar*cr-ai*ci, ai*cr+ar*ci, br*dr-bi*di, bi*dr+br*di
c2 = _mm_cvtps_epi32(b); // convert from 32fc to 32ic
//complex 32fc multiplication two_phase_acc_reg=two_phase_acc_reg*two_phase_inc_reg
yl = _mm_moveldup_ps(two_phase_acc_reg); // Load yl with cr,cr,dr,dr
yh = _mm_movehdup_ps(two_phase_acc_reg); // Load yh with ci,ci,di,di
tmp1 = _mm_mul_ps(two_phase_inc_reg, yl); // tmp1 = ar*cr,ai*cr,br*dr,bi*dr
tmp3 = _mm_shuffle_ps(two_phase_inc_reg, two_phase_inc_reg, 0xB1); // Re-arrange x to be ai,ar,bi,br
tmp2 = _mm_mul_ps(tmp3, yh); // tmp2 = ai*ci,ar*ci,bi*di,br*di
two_phase_acc_reg = _mm_addsub_ps(tmp1, tmp2); // ar*cr-ai*ci, ai*cr+ar*ci, br*dr-bi*di, bi*dr+br*di
// store four output samples
result2 = _mm_packs_epi32(c1, c2);// convert from 32ic to 16ic
_in_common += 2;
b2 = _mm256_insertf128_si256(_mm256_castsi128_si256(result1), (result2), 1);
for (int n_vec = 0; n_vec < num_a_vectors; n_vec++)
{
a2 = _mm256_load_si256((__m256i*)&(_in_a[n_vec][(number * ROTATOR_RELOAD + j) * 8]));
c = _mm256_mullo_epi16(a2, b2);
c_sr = _mm256_srli_si256(c, 2); // Shift a right by imm8 bytes while shifting in zeros, and store the results in dst.
real = _mm256_subs_epi16(c, c_sr);
c_sr = _mm256_slli_si256(b2, 2);
c = _mm256_mullo_epi16(a2, c_sr);
c_sr = _mm256_slli_si256(a2, 2);
imag = _mm256_mullo_epi16(b2, c_sr);
imag = _mm256_adds_epi16(c, imag);
realcacc[n_vec] = _mm256_adds_epi16(realcacc[n_vec], real);
imagcacc[n_vec] = _mm256_adds_epi16(imagcacc[n_vec], imag);
}
}
// regenerate phase
tmp1 = _mm_mul_ps(two_phase_acc_reg, two_phase_acc_reg);
tmp2 = _mm_hadd_ps(tmp1, tmp1);
tmp1 = _mm_shuffle_ps(tmp2, tmp2, 0xD8);
tmp2 = _mm_sqrt_ps(tmp1);
two_phase_acc_reg = _mm_div_ps(two_phase_acc_reg, tmp2);
}
for (unsigned int j = 0; j < avx2_iters % ROTATOR_RELOAD; j++)
{
a = _mm_set_ps((float)(lv_cimag(_in_common[1])), (float)(lv_creal(_in_common[1])), (float)(lv_cimag(_in_common[0])), (float)(lv_creal(_in_common[0]))); // //load (2 byte imag, 2 byte real) x 2 into 128 bits reg
//complex 32fc multiplication b=a*two_phase_acc_reg
yl = _mm_moveldup_ps(two_phase_acc_reg); // Load yl with cr,cr,dr,dr
yh = _mm_movehdup_ps(two_phase_acc_reg); // Load yh with ci,ci,di,di
tmp1 = _mm_mul_ps(a, yl); // tmp1 = ar*cr,ai*cr,br*dr,bi*dr
a = _mm_shuffle_ps(a, a, 0xB1); // Re-arrange x to be ai,ar,bi,br
tmp2 = _mm_mul_ps(a, yh); // tmp2 = ai*ci,ar*ci,bi*di,br*di
b = _mm_addsub_ps(tmp1, tmp2); // ar*cr-ai*ci, ai*cr+ar*ci, br*dr-bi*di, bi*dr+br*di
c1 = _mm_cvtps_epi32(b); // convert from 32fc to 32ic
//complex 32fc multiplication two_phase_acc_reg=two_phase_acc_reg*two_phase_inc_reg
yl = _mm_moveldup_ps(two_phase_acc_reg); // Load yl with cr,cr,dr,dr
yh = _mm_movehdup_ps(two_phase_acc_reg); // Load yh with ci,ci,di,di
tmp1 = _mm_mul_ps(two_phase_inc_reg, yl); // tmp1 = ar*cr,ai*cr,br*dr,bi*dr
tmp3 = _mm_shuffle_ps(two_phase_inc_reg, two_phase_inc_reg, 0xB1); // Re-arrange x to be ai,ar,bi,br
tmp2 = _mm_mul_ps(tmp3, yh); // tmp2 = ai*ci,ar*ci,bi*di,br*di
two_phase_acc_reg = _mm_addsub_ps(tmp1, tmp2); // ar*cr-ai*ci, ai*cr+ar*ci, br*dr-bi*di, bi*dr+br*di
//next two samples
_in_common += 2;
a = _mm_set_ps((float)(lv_cimag(_in_common[1])), (float)(lv_creal(_in_common[1])), (float)(lv_cimag(_in_common[0])), (float)(lv_creal(_in_common[0]))); // //load (2 byte imag, 2 byte real) x 2 into 128 bits reg
//complex 32fc multiplication b=a*two_phase_acc_reg
yl = _mm_moveldup_ps(two_phase_acc_reg); // Load yl with cr,cr,dr,dr
yh = _mm_movehdup_ps(two_phase_acc_reg); // Load yh with ci,ci,di,di
tmp1 = _mm_mul_ps(a, yl); // tmp1 = ar*cr,ai*cr,br*dr,bi*dr
a = _mm_shuffle_ps(a, a, 0xB1); // Re-arrange x to be ai,ar,bi,br
tmp2 = _mm_mul_ps(a, yh); // tmp2 = ai*ci,ar*ci,bi*di,br*di
b = _mm_addsub_ps(tmp1, tmp2); // ar*cr-ai*ci, ai*cr+ar*ci, br*dr-bi*di, bi*dr+br*di
c2 = _mm_cvtps_epi32(b); // convert from 32fc to 32ic
//complex 32fc multiplication two_phase_acc_reg=two_phase_acc_reg*two_phase_inc_reg
yl = _mm_moveldup_ps(two_phase_acc_reg); // Load yl with cr,cr,dr,dr
yh = _mm_movehdup_ps(two_phase_acc_reg); // Load yh with ci,ci,di,di
tmp1 = _mm_mul_ps(two_phase_inc_reg, yl); // tmp1 = ar*cr,ai*cr,br*dr,bi*dr
tmp3 = _mm_shuffle_ps(two_phase_inc_reg, two_phase_inc_reg, 0xB1); // Re-arrange x to be ai,ar,bi,br
tmp2 = _mm_mul_ps(tmp3, yh); // tmp2 = ai*ci,ar*ci,bi*di,br*di
two_phase_acc_reg = _mm_addsub_ps(tmp1, tmp2); // ar*cr-ai*ci, ai*cr+ar*ci, br*dr-bi*di, bi*dr+br*di
// store four output samples
result1 = _mm_packs_epi32(c1, c2);// convert from 32ic to 16ic
_in_common += 2;
a = _mm_set_ps((float)(lv_cimag(_in_common[1])), (float)(lv_creal(_in_common[1])), (float)(lv_cimag(_in_common[0])), (float)(lv_creal(_in_common[0]))); // //load (2 byte imag, 2 byte real) x 2 into 128 bits reg
//complex 32fc multiplication b=a*two_phase_acc_reg
yl = _mm_moveldup_ps(two_phase_acc_reg); // Load yl with cr,cr,dr,dr
yh = _mm_movehdup_ps(two_phase_acc_reg); // Load yh with ci,ci,di,di
tmp1 = _mm_mul_ps(a, yl); // tmp1 = ar*cr,ai*cr,br*dr,bi*dr
a = _mm_shuffle_ps(a, a, 0xB1); // Re-arrange x to be ai,ar,bi,br
tmp2 = _mm_mul_ps(a, yh); // tmp2 = ai*ci,ar*ci,bi*di,br*di
b = _mm_addsub_ps(tmp1, tmp2); // ar*cr-ai*ci, ai*cr+ar*ci, br*dr-bi*di, bi*dr+br*di
c1 = _mm_cvtps_epi32(b); // convert from 32fc to 32ic
//complex 32fc multiplication two_phase_acc_reg=two_phase_acc_reg*two_phase_inc_reg
yl = _mm_moveldup_ps(two_phase_acc_reg); // Load yl with cr,cr,dr,dr
yh = _mm_movehdup_ps(two_phase_acc_reg); // Load yh with ci,ci,di,di
tmp1 = _mm_mul_ps(two_phase_inc_reg, yl); // tmp1 = ar*cr,ai*cr,br*dr,bi*dr
tmp3 = _mm_shuffle_ps(two_phase_inc_reg, two_phase_inc_reg, 0xB1); // Re-arrange x to be ai,ar,bi,br
tmp2 = _mm_mul_ps(tmp3, yh); // tmp2 = ai*ci,ar*ci,bi*di,br*di
two_phase_acc_reg = _mm_addsub_ps(tmp1, tmp2); // ar*cr-ai*ci, ai*cr+ar*ci, br*dr-bi*di, bi*dr+br*di
//next two samples
_in_common += 2;
a = _mm_set_ps((float)(lv_cimag(_in_common[1])), (float)(lv_creal(_in_common[1])), (float)(lv_cimag(_in_common[0])), (float)(lv_creal(_in_common[0]))); // //load (2 byte imag, 2 byte real) x 2 into 128 bits reg
__builtin_prefetch(_in_common + 16);
//complex 32fc multiplication b=a*two_phase_acc_reg
yl = _mm_moveldup_ps(two_phase_acc_reg); // Load yl with cr,cr,dr,dr
yh = _mm_movehdup_ps(two_phase_acc_reg); // Load yh with ci,ci,di,di
tmp1 = _mm_mul_ps(a, yl); // tmp1 = ar*cr,ai*cr,br*dr,bi*dr
a = _mm_shuffle_ps(a, a, 0xB1); // Re-arrange x to be ai,ar,bi,br
tmp2 = _mm_mul_ps(a, yh); // tmp2 = ai*ci,ar*ci,bi*di,br*di
b = _mm_addsub_ps(tmp1, tmp2); // ar*cr-ai*ci, ai*cr+ar*ci, br*dr-bi*di, bi*dr+br*di
c2 = _mm_cvtps_epi32(b); // convert from 32fc to 32ic
//complex 32fc multiplication two_phase_acc_reg=two_phase_acc_reg*two_phase_inc_reg
yl = _mm_moveldup_ps(two_phase_acc_reg); // Load yl with cr,cr,dr,dr
yh = _mm_movehdup_ps(two_phase_acc_reg); // Load yh with ci,ci,di,di
tmp1 = _mm_mul_ps(two_phase_inc_reg, yl); // tmp1 = ar*cr,ai*cr,br*dr,bi*dr
tmp3 = _mm_shuffle_ps(two_phase_inc_reg, two_phase_inc_reg, 0xB1); // Re-arrange x to be ai,ar,bi,br
tmp2 = _mm_mul_ps(tmp3, yh); // tmp2 = ai*ci,ar*ci,bi*di,br*di
two_phase_acc_reg = _mm_addsub_ps(tmp1, tmp2); // ar*cr-ai*ci, ai*cr+ar*ci, br*dr-bi*di, bi*dr+br*di
// store four output samples
result2 = _mm_packs_epi32(c1, c2);// convert from 32ic to 16ic
_in_common += 2;
b2 = _mm256_insertf128_si256(_mm256_castsi128_si256(result1), (result2), 1);
for (int n_vec = 0; n_vec < num_a_vectors; n_vec++)
{
a2 = _mm256_load_si256((__m256i*)&(_in_a[n_vec][((avx2_iters / ROTATOR_RELOAD) * ROTATOR_RELOAD + j) * 8]));
c = _mm256_mullo_epi16(a2, b2);
c_sr = _mm256_srli_si256(c, 2); // Shift a right by imm8 bytes while shifting in zeros, and store the results in dst.
real = _mm256_subs_epi16(c, c_sr);
c_sr = _mm256_slli_si256(b2, 2);
c = _mm256_mullo_epi16(a2, c_sr);
c_sr = _mm256_slli_si256(a2, 2);
imag = _mm256_mullo_epi16(b2, c_sr);
imag = _mm256_adds_epi16(c, imag);
realcacc[n_vec] = _mm256_adds_epi16(realcacc[n_vec], real);
imagcacc[n_vec] = _mm256_adds_epi16(imagcacc[n_vec], imag);
}
}
for (int n_vec = 0; n_vec < num_a_vectors; n_vec++)
{
realcacc[n_vec] = _mm256_and_si256(realcacc[n_vec], mask_real);
imagcacc[n_vec] = _mm256_and_si256(imagcacc[n_vec], mask_imag);
a2 = _mm256_or_si256(realcacc[n_vec], imagcacc[n_vec]);
_mm256_store_si256((__m256i*)dotProductVector, a2); // Store the results back into the dot product vector
dotProduct = lv_cmake(0,0);
for (int i = 0; i < 8; ++i)
{
dotProduct = lv_cmake(sat_adds16i(lv_creal(dotProduct), lv_creal(dotProductVector[i])),
sat_adds16i(lv_cimag(dotProduct), lv_cimag(dotProductVector[i])));
}
_out[n_vec] = dotProduct;
}
volk_gnsssdr_free(realcacc);
volk_gnsssdr_free(imagcacc);
_mm_store_ps((float*)two_phase_acc, two_phase_acc_reg);
(*phase) = two_phase_acc[0];
for(unsigned int n = avx2_iters * 8; n < num_points; n++)
{
tmp16 = in_common[n];
tmp32 = lv_cmake((float)lv_creal(tmp16), (float)lv_cimag(tmp16)) * (*phase);
tmp16 = lv_cmake((int16_t)rintf(lv_creal(tmp32)), (int16_t)rintf(lv_cimag(tmp32)));
(*phase) *= phase_inc;
for (int n_vec = 0; n_vec < num_a_vectors; n_vec++)
{
lv_16sc_t tmp = tmp16 * in_a[n_vec][n];
_out[n_vec] = lv_cmake(sat_adds16i(lv_creal(_out[n_vec]), lv_creal(tmp)),
sat_adds16i(lv_cimag(_out[n_vec]), lv_cimag(tmp)));
}
}
}
#endif /* LV_HAVE_AVX2 */
#ifdef LV_HAVE_NEON
#include <arm_neon.h>

124
src/algorithms/libs/volk_gnsssdr_module/volk_gnsssdr/kernels/volk_gnsssdr/volk_gnsssdr_16ic_x2_rotator_dotprodxnpuppet_16ic.h
View File

@ -193,6 +193,130 @@ static inline void volk_gnsssdr_16ic_x2_rotator_dotprodxnpuppet_16ic_u_sse3(lv_1
#endif // SSE3
#ifdef LV_HAVE_AVX2
static inline void volk_gnsssdr_16ic_x2_rotator_dotprodxnpuppet_16ic_a_avx2(lv_16sc_t* result, const lv_16sc_t* local_code, const lv_16sc_t* in, unsigned int num_points)
{
// phases must be normalized. Phase rotator expects a complex exponential input!
float rem_carrier_phase_in_rad = 0.345;
float phase_step_rad = 0.1;
lv_32fc_t phase[1];
phase[0] = lv_cmake(cos(rem_carrier_phase_in_rad), sin(rem_carrier_phase_in_rad));
lv_32fc_t phase_inc[1];
phase_inc[0] = lv_cmake(cos(phase_step_rad), sin(phase_step_rad));
int num_a_vectors = 3;
lv_16sc_t** in_a = (lv_16sc_t**)volk_gnsssdr_malloc(sizeof(lv_16sc_t*) * num_a_vectors, volk_gnsssdr_get_alignment());
for(unsigned int n = 0; n < num_a_vectors; n++)
{
in_a[n] = (lv_16sc_t*)volk_gnsssdr_malloc(sizeof(lv_16sc_t) * num_points, volk_gnsssdr_get_alignment());
memcpy((lv_16sc_t*)in_a[n], (lv_16sc_t*)in, sizeof(lv_16sc_t) * num_points);
}
volk_gnsssdr_16ic_x2_rotator_dot_prod_16ic_xn_a_avx2(result, local_code, phase_inc[0], phase, (const lv_16sc_t**) in_a, num_a_vectors, num_points);
for(unsigned int n = 0; n < num_a_vectors; n++)
{
volk_gnsssdr_free(in_a[n]);
}
volk_gnsssdr_free(in_a);
}
#endif // AVX2
#ifdef LV_HAVE_AVX2
static inline void volk_gnsssdr_16ic_x2_rotator_dotprodxnpuppet_16ic_a_avx2_reload(lv_16sc_t* result, const lv_16sc_t* local_code, const lv_16sc_t* in, unsigned int num_points)
{
// phases must be normalized. Phase rotator expects a complex exponential input!
float rem_carrier_phase_in_rad = 0.345;
float phase_step_rad = 0.1;
lv_32fc_t phase[1];
phase[0] = lv_cmake(cos(rem_carrier_phase_in_rad), sin(rem_carrier_phase_in_rad));
lv_32fc_t phase_inc[1];
phase_inc[0] = lv_cmake(cos(phase_step_rad), sin(phase_step_rad));
int num_a_vectors = 3;
lv_16sc_t** in_a = (lv_16sc_t**)volk_gnsssdr_malloc(sizeof(lv_16sc_t*) * num_a_vectors, volk_gnsssdr_get_alignment());
for(unsigned int n = 0; n < num_a_vectors; n++)
{
in_a[n] = (lv_16sc_t*)volk_gnsssdr_malloc(sizeof(lv_16sc_t) * num_points, volk_gnsssdr_get_alignment());
memcpy((lv_16sc_t*)in_a[n], (lv_16sc_t*)in, sizeof(lv_16sc_t) * num_points);
}
volk_gnsssdr_16ic_x2_rotator_dot_prod_16ic_xn_a_avx2_reload(result, local_code, phase_inc[0], phase, (const lv_16sc_t**) in_a, num_a_vectors, num_points);
for(unsigned int n = 0; n < num_a_vectors; n++)
{
volk_gnsssdr_free(in_a[n]);
}
volk_gnsssdr_free(in_a);
}
#endif // AVX2
#ifdef LV_HAVE_AVX2
static inline void volk_gnsssdr_16ic_x2_rotator_dotprodxnpuppet_16ic_u_avx2(lv_16sc_t* result, const lv_16sc_t* local_code, const lv_16sc_t* in, unsigned int num_points)
{
// phases must be normalized. Phase rotator expects a complex exponential input!
float rem_carrier_phase_in_rad = 0.345;
float phase_step_rad = 0.1;
lv_32fc_t phase[1];
phase[0] = lv_cmake(cos(rem_carrier_phase_in_rad), sin(rem_carrier_phase_in_rad));
lv_32fc_t phase_inc[1];
phase_inc[0] = lv_cmake(cos(phase_step_rad), sin(phase_step_rad));
int num_a_vectors = 3;
lv_16sc_t** in_a = (lv_16sc_t**)volk_gnsssdr_malloc(sizeof(lv_16sc_t*) * num_a_vectors, volk_gnsssdr_get_alignment());
for(unsigned int n = 0; n < num_a_vectors; n++)
{
in_a[n] = (lv_16sc_t*)volk_gnsssdr_malloc(sizeof(lv_16sc_t) * num_points, volk_gnsssdr_get_alignment());
memcpy((lv_16sc_t*)in_a[n], (lv_16sc_t*)in, sizeof(lv_16sc_t) * num_points);
}
volk_gnsssdr_16ic_x2_rotator_dot_prod_16ic_xn_a_avx2(result, local_code, phase_inc[0], phase, (const lv_16sc_t**) in_a, num_a_vectors, num_points);
for(unsigned int n = 0; n < num_a_vectors; n++)
{
volk_gnsssdr_free(in_a[n]);
}
volk_gnsssdr_free(in_a);
}
#endif // AVX2
#ifdef LV_HAVE_AVX2
static inline void volk_gnsssdr_16ic_x2_rotator_dotprodxnpuppet_16ic_u_avx2_reload(lv_16sc_t* result, const lv_16sc_t* local_code, const lv_16sc_t* in, unsigned int num_points)
{
// phases must be normalized. Phase rotator expects a complex exponential input!
float rem_carrier_phase_in_rad = 0.345;
float phase_step_rad = 0.1;
lv_32fc_t phase[1];
phase[0] = lv_cmake(cos(rem_carrier_phase_in_rad), sin(rem_carrier_phase_in_rad));
lv_32fc_t phase_inc[1];
phase_inc[0] = lv_cmake(cos(phase_step_rad), sin(phase_step_rad));
int num_a_vectors = 3;
lv_16sc_t** in_a = (lv_16sc_t**)volk_gnsssdr_malloc(sizeof(lv_16sc_t*) * num_a_vectors, volk_gnsssdr_get_alignment());
for(unsigned int n = 0; n < num_a_vectors; n++)
{
in_a[n] = (lv_16sc_t*)volk_gnsssdr_malloc(sizeof(lv_16sc_t) * num_points, volk_gnsssdr_get_alignment());
memcpy((lv_16sc_t*)in_a[n], (lv_16sc_t*)in, sizeof(lv_16sc_t) * num_points);
}
volk_gnsssdr_16ic_x2_rotator_dot_prod_16ic_xn_a_avx2_reload(result, local_code, phase_inc[0], phase, (const lv_16sc_t**) in_a, num_a_vectors, num_points);
for(unsigned int n = 0; n < num_a_vectors; n++)
{
volk_gnsssdr_free(in_a[n]);
}
volk_gnsssdr_free(in_a);
}
#endif // AVX2
#ifdef LV_HAVE_NEON
static inline void volk_gnsssdr_16ic_x2_rotator_dotprodxnpuppet_16ic_neon(lv_16sc_t* result, const lv_16sc_t* local_code, const lv_16sc_t* in, unsigned int num_points)
{