mirror of https://github.com/gnss-sdr/gnss-sdr
487 lines
16 KiB
C
487 lines
16 KiB
C
/*!
|
|
* \file volk_gnsssdr_8ic_x2_dot_prod_8ic.h
|
|
* \brief VOLK_GNSSSDR kernel: multiplies two 16 bits vectors and accumulates them.
|
|
* \authors <ul>
|
|
* <li> Andres Cecilia, 2014. a.cecilia.luque(at)gmail.com
|
|
* </ul>
|
|
*
|
|
* VOLK_GNSSSDR kernel that multiplies two 16 bits vectors (8 bits the real part
|
|
* and 8 bits the imaginary part) and accumulates them.
|
|
*
|
|
* -----------------------------------------------------------------------------
|
|
*
|
|
* GNSS-SDR is a Global Navigation Satellite System software-defined receiver.
|
|
* This file is part of GNSS-SDR.
|
|
*
|
|
* Copyright (C) 2010-2020 (see AUTHORS file for a list of contributors)
|
|
* SPDX-License-Identifier: GPL-3.0-or-later
|
|
*
|
|
* -----------------------------------------------------------------------------
|
|
*/
|
|
|
|
/*!
|
|
* \page volk_gnsssdr_8ic_x2_dot_prod_8ic
|
|
*
|
|
* \b Overview
|
|
*
|
|
* Multiplies two input complex vectors (8-bit integer each component) and accumulates them,
|
|
* storing the result.
|
|
*
|
|
* <b>Dispatcher Prototype</b>
|
|
* \code
|
|
* void volk_gnsssdr_16ic_x2_dot_prod_16ic(lv_16sc_t* result, const lv_16sc_t* in_a, const lv_16sc_t* in_b, unsigned int num_points);
|
|
* \endcode
|
|
*
|
|
* \b Inputs
|
|
* \li in_a: One of the vectors to be multiplied and accumulated
|
|
* \li in_b: The other vector to be multiplied and accumulated
|
|
* \li num_points: The Number of complex values to be multiplied together, accumulated and stored into \p result
|
|
*
|
|
* \b Outputs
|
|
* \li result: Value of the accumulated result
|
|
*
|
|
*/
|
|
|
|
#ifndef INCLUDED_volk_gnsssdr_8ic_x2_dot_prod_8ic_H
|
|
#define INCLUDED_volk_gnsssdr_8ic_x2_dot_prod_8ic_H
|
|
|
|
#include <volk_gnsssdr/volk_gnsssdr_common.h>
|
|
#include <volk_gnsssdr/volk_gnsssdr_complex.h>
|
|
#include <string.h>
|
|
|
|
#ifdef LV_HAVE_GENERIC
|
|
|
|
static inline void volk_gnsssdr_8ic_x2_dot_prod_8ic_generic(lv_8sc_t* result, const lv_8sc_t* in_a, const lv_8sc_t* in_b, unsigned int num_points)
|
|
{
|
|
/*lv_8sc_t* cPtr = result;
|
|
const lv_8sc_t* aPtr = in_a;
|
|
const lv_8sc_t* bPtr = in_b;
|
|
|
|
for(int number = 0; number < num_points; number++){
|
|
*cPtr += (*aPtr++) * (*bPtr++);
|
|
}*/
|
|
|
|
char* res = (char*)result;
|
|
char* in = (char*)in_a;
|
|
char* tp = (char*)in_b;
|
|
unsigned int n_2_ccomplex_blocks = num_points / 2;
|
|
unsigned int isodd = num_points & 1;
|
|
|
|
char sum0[2] = {0, 0};
|
|
char sum1[2] = {0, 0};
|
|
unsigned int i = 0;
|
|
|
|
for (i = 0; i < n_2_ccomplex_blocks; ++i)
|
|
{
|
|
sum0[0] += in[0] * tp[0] - in[1] * tp[1];
|
|
sum0[1] += in[0] * tp[1] + in[1] * tp[0];
|
|
sum1[0] += in[2] * tp[2] - in[3] * tp[3];
|
|
sum1[1] += in[2] * tp[3] + in[3] * tp[2];
|
|
|
|
in += 4;
|
|
tp += 4;
|
|
}
|
|
|
|
res[0] = sum0[0] + sum1[0];
|
|
res[1] = sum0[1] + sum1[1];
|
|
|
|
// Cleanup if we had an odd number of points
|
|
for (i = 0; i < isodd; ++i)
|
|
{
|
|
*result += in_a[num_points - 1] * in_b[num_points - 1];
|
|
}
|
|
}
|
|
|
|
#endif /*LV_HAVE_GENERIC*/
|
|
|
|
|
|
#ifdef LV_HAVE_SSE2
|
|
#include <emmintrin.h>
|
|
|
|
static inline void volk_gnsssdr_8ic_x2_dot_prod_8ic_u_sse2(lv_8sc_t* result, const lv_8sc_t* in_a, const lv_8sc_t* in_b, unsigned int num_points)
|
|
{
|
|
lv_8sc_t dotProduct;
|
|
memset(&dotProduct, 0x0, 2 * sizeof(char));
|
|
unsigned int number;
|
|
unsigned int i;
|
|
const lv_8sc_t* a = in_a;
|
|
const lv_8sc_t* b = in_b;
|
|
|
|
const unsigned int sse_iters = num_points / 8;
|
|
|
|
if (sse_iters > 0)
|
|
{
|
|
__m128i x, y, mult1, realx, imagx, realy, imagy, realx_mult_realy, imagx_mult_imagy, realx_mult_imagy, imagx_mult_realy, realc, imagc, totalc, realcacc, imagcacc;
|
|
|
|
mult1 = _mm_set_epi8(0, 0xFF, 0, 0xFF, 0, 0xFF, 0, 0xFF, 0, 0xFF, 0, 0xFF, 0, 0xFF, 0, 0xFF);
|
|
realcacc = _mm_setzero_si128();
|
|
imagcacc = _mm_setzero_si128();
|
|
|
|
for (number = 0; number < sse_iters; number++)
|
|
{
|
|
x = _mm_loadu_si128((__m128i*)a);
|
|
y = _mm_loadu_si128((__m128i*)b);
|
|
|
|
imagx = _mm_srli_si128(x, 1);
|
|
imagx = _mm_and_si128(imagx, mult1);
|
|
realx = _mm_and_si128(x, mult1);
|
|
|
|
imagy = _mm_srli_si128(y, 1);
|
|
imagy = _mm_and_si128(imagy, mult1);
|
|
realy = _mm_and_si128(y, mult1);
|
|
|
|
realx_mult_realy = _mm_mullo_epi16(realx, realy);
|
|
imagx_mult_imagy = _mm_mullo_epi16(imagx, imagy);
|
|
realx_mult_imagy = _mm_mullo_epi16(realx, imagy);
|
|
imagx_mult_realy = _mm_mullo_epi16(imagx, realy);
|
|
|
|
realc = _mm_sub_epi16(realx_mult_realy, imagx_mult_imagy);
|
|
imagc = _mm_add_epi16(realx_mult_imagy, imagx_mult_realy);
|
|
|
|
realcacc = _mm_add_epi16(realcacc, realc);
|
|
imagcacc = _mm_add_epi16(imagcacc, imagc);
|
|
|
|
a += 8;
|
|
b += 8;
|
|
}
|
|
|
|
realcacc = _mm_and_si128(realcacc, mult1);
|
|
imagcacc = _mm_and_si128(imagcacc, mult1);
|
|
imagcacc = _mm_slli_si128(imagcacc, 1);
|
|
|
|
totalc = _mm_or_si128(realcacc, imagcacc);
|
|
|
|
__VOLK_ATTR_ALIGNED(16)
|
|
lv_8sc_t dotProductVector[8];
|
|
|
|
_mm_storeu_si128((__m128i*)dotProductVector, totalc); // Store the results back into the dot product vector
|
|
|
|
for (i = 0; i < 8; ++i)
|
|
{
|
|
dotProduct += dotProductVector[i];
|
|
}
|
|
}
|
|
|
|
for (i = sse_iters * 8; i < num_points; ++i)
|
|
{
|
|
dotProduct += (*a++) * (*b++);
|
|
}
|
|
|
|
*result = dotProduct;
|
|
}
|
|
|
|
#endif /*LV_HAVE_SSE2*/
|
|
|
|
|
|
#ifdef LV_HAVE_SSE4_1
|
|
#include <smmintrin.h>
|
|
|
|
static inline void volk_gnsssdr_8ic_x2_dot_prod_8ic_u_sse4_1(lv_8sc_t* result, const lv_8sc_t* in_a, const lv_8sc_t* in_b, unsigned int num_points)
|
|
{
|
|
lv_8sc_t dotProduct;
|
|
memset(&dotProduct, 0x0, 2 * sizeof(char));
|
|
unsigned int number;
|
|
unsigned int i;
|
|
const lv_8sc_t* a = in_a;
|
|
const lv_8sc_t* b = in_b;
|
|
|
|
const unsigned int sse_iters = num_points / 8;
|
|
|
|
if (sse_iters > 0)
|
|
{
|
|
__m128i x, y, mult1, realx, imagx, realy, imagy, realx_mult_realy, imagx_mult_imagy, realx_mult_imagy, imagx_mult_realy, realc, imagc, totalc, realcacc, imagcacc;
|
|
|
|
mult1 = _mm_set_epi8(0, 0xFF, 0, 0xFF, 0, 0xFF, 0, 0xFF, 0, 0xFF, 0, 0xFF, 0, 0xFF, 0, 0xFF);
|
|
realcacc = _mm_setzero_si128();
|
|
imagcacc = _mm_setzero_si128();
|
|
|
|
for (number = 0; number < sse_iters; number++)
|
|
{
|
|
x = _mm_lddqu_si128((__m128i*)a);
|
|
y = _mm_lddqu_si128((__m128i*)b);
|
|
|
|
imagx = _mm_srli_si128(x, 1);
|
|
imagx = _mm_and_si128(imagx, mult1);
|
|
realx = _mm_and_si128(x, mult1);
|
|
|
|
imagy = _mm_srli_si128(y, 1);
|
|
imagy = _mm_and_si128(imagy, mult1);
|
|
realy = _mm_and_si128(y, mult1);
|
|
|
|
realx_mult_realy = _mm_mullo_epi16(realx, realy);
|
|
imagx_mult_imagy = _mm_mullo_epi16(imagx, imagy);
|
|
realx_mult_imagy = _mm_mullo_epi16(realx, imagy);
|
|
imagx_mult_realy = _mm_mullo_epi16(imagx, realy);
|
|
|
|
realc = _mm_sub_epi16(realx_mult_realy, imagx_mult_imagy);
|
|
imagc = _mm_add_epi16(realx_mult_imagy, imagx_mult_realy);
|
|
|
|
realcacc = _mm_add_epi16(realcacc, realc);
|
|
imagcacc = _mm_add_epi16(imagcacc, imagc);
|
|
|
|
a += 8;
|
|
b += 8;
|
|
}
|
|
|
|
imagcacc = _mm_slli_si128(imagcacc, 1);
|
|
|
|
totalc = _mm_blendv_epi8(imagcacc, realcacc, mult1);
|
|
|
|
__VOLK_ATTR_ALIGNED(16)
|
|
lv_8sc_t dotProductVector[8];
|
|
|
|
_mm_storeu_si128((__m128i*)dotProductVector, totalc); // Store the results back into the dot product vector
|
|
|
|
for (i = 0; i < 8; ++i)
|
|
{
|
|
dotProduct += dotProductVector[i];
|
|
}
|
|
}
|
|
|
|
for (i = sse_iters * 8; i < num_points; ++i)
|
|
{
|
|
dotProduct += (*a++) * (*b++);
|
|
}
|
|
|
|
*result = dotProduct;
|
|
}
|
|
|
|
#endif /*LV_HAVE_SSE4_1*/
|
|
|
|
|
|
#ifdef LV_HAVE_SSE2
|
|
#include <emmintrin.h>
|
|
|
|
static inline void volk_gnsssdr_8ic_x2_dot_prod_8ic_a_sse2(lv_8sc_t* result, const lv_8sc_t* in_a, const lv_8sc_t* in_b, unsigned int num_points)
|
|
{
|
|
lv_8sc_t dotProduct;
|
|
memset(&dotProduct, 0x0, 2 * sizeof(char));
|
|
unsigned int number;
|
|
unsigned int i;
|
|
const lv_8sc_t* a = in_a;
|
|
const lv_8sc_t* b = in_b;
|
|
|
|
const unsigned int sse_iters = num_points / 8;
|
|
|
|
if (sse_iters > 0)
|
|
{
|
|
__m128i x, y, mult1, realx, imagx, realy, imagy, realx_mult_realy, imagx_mult_imagy, realx_mult_imagy, imagx_mult_realy, realc, imagc, totalc, realcacc, imagcacc;
|
|
|
|
mult1 = _mm_set_epi8(0, 0xFF, 0, 0xFF, 0, 0xFF, 0, 0xFF, 0, 0xFF, 0, 0xFF, 0, 0xFF, 0, 0xFF);
|
|
realcacc = _mm_setzero_si128();
|
|
imagcacc = _mm_setzero_si128();
|
|
|
|
for (number = 0; number < sse_iters; number++)
|
|
{
|
|
x = _mm_load_si128((__m128i*)a);
|
|
y = _mm_load_si128((__m128i*)b);
|
|
|
|
imagx = _mm_srli_si128(x, 1);
|
|
imagx = _mm_and_si128(imagx, mult1);
|
|
realx = _mm_and_si128(x, mult1);
|
|
|
|
imagy = _mm_srli_si128(y, 1);
|
|
imagy = _mm_and_si128(imagy, mult1);
|
|
realy = _mm_and_si128(y, mult1);
|
|
|
|
realx_mult_realy = _mm_mullo_epi16(realx, realy);
|
|
imagx_mult_imagy = _mm_mullo_epi16(imagx, imagy);
|
|
realx_mult_imagy = _mm_mullo_epi16(realx, imagy);
|
|
imagx_mult_realy = _mm_mullo_epi16(imagx, realy);
|
|
|
|
realc = _mm_sub_epi16(realx_mult_realy, imagx_mult_imagy);
|
|
imagc = _mm_add_epi16(realx_mult_imagy, imagx_mult_realy);
|
|
|
|
realcacc = _mm_add_epi16(realcacc, realc);
|
|
imagcacc = _mm_add_epi16(imagcacc, imagc);
|
|
|
|
a += 8;
|
|
b += 8;
|
|
}
|
|
|
|
realcacc = _mm_and_si128(realcacc, mult1);
|
|
imagcacc = _mm_and_si128(imagcacc, mult1);
|
|
imagcacc = _mm_slli_si128(imagcacc, 1);
|
|
|
|
totalc = _mm_or_si128(realcacc, imagcacc);
|
|
|
|
__VOLK_ATTR_ALIGNED(16)
|
|
lv_8sc_t dotProductVector[8];
|
|
|
|
_mm_store_si128((__m128i*)dotProductVector, totalc); // Store the results back into the dot product vector
|
|
|
|
for (i = 0; i < 8; ++i)
|
|
{
|
|
dotProduct += dotProductVector[i];
|
|
}
|
|
}
|
|
|
|
for (i = sse_iters * 8; i < num_points; ++i)
|
|
{
|
|
dotProduct += (*a++) * (*b++);
|
|
}
|
|
|
|
*result = dotProduct;
|
|
}
|
|
|
|
#endif /*LV_HAVE_SSE2*/
|
|
|
|
#ifdef LV_HAVE_SSE4_1
|
|
#include <smmintrin.h>
|
|
|
|
static inline void volk_gnsssdr_8ic_x2_dot_prod_8ic_a_sse4_1(lv_8sc_t* result, const lv_8sc_t* in_a, const lv_8sc_t* in_b, unsigned int num_points)
|
|
{
|
|
lv_8sc_t dotProduct;
|
|
memset(&dotProduct, 0x0, 2 * sizeof(char));
|
|
unsigned int number;
|
|
unsigned int i;
|
|
const lv_8sc_t* a = in_a;
|
|
const lv_8sc_t* b = in_b;
|
|
|
|
const unsigned int sse_iters = num_points / 8;
|
|
|
|
if (sse_iters > 0)
|
|
{
|
|
__m128i x, y, mult1, realx, imagx, realy, imagy, realx_mult_realy, imagx_mult_imagy, realx_mult_imagy, imagx_mult_realy, realc, imagc, totalc, realcacc, imagcacc;
|
|
|
|
mult1 = _mm_set_epi8(0, 0xFF, 0, 0xFF, 0, 0xFF, 0, 0xFF, 0, 0xFF, 0, 0xFF, 0, 0xFF, 0, 0xFF);
|
|
realcacc = _mm_setzero_si128();
|
|
imagcacc = _mm_setzero_si128();
|
|
|
|
for (number = 0; number < sse_iters; number++)
|
|
{
|
|
x = _mm_load_si128((__m128i*)a);
|
|
y = _mm_load_si128((__m128i*)b);
|
|
|
|
imagx = _mm_srli_si128(x, 1);
|
|
imagx = _mm_and_si128(imagx, mult1);
|
|
realx = _mm_and_si128(x, mult1);
|
|
|
|
imagy = _mm_srli_si128(y, 1);
|
|
imagy = _mm_and_si128(imagy, mult1);
|
|
realy = _mm_and_si128(y, mult1);
|
|
|
|
realx_mult_realy = _mm_mullo_epi16(realx, realy);
|
|
imagx_mult_imagy = _mm_mullo_epi16(imagx, imagy);
|
|
realx_mult_imagy = _mm_mullo_epi16(realx, imagy);
|
|
imagx_mult_realy = _mm_mullo_epi16(imagx, realy);
|
|
|
|
realc = _mm_sub_epi16(realx_mult_realy, imagx_mult_imagy);
|
|
imagc = _mm_add_epi16(realx_mult_imagy, imagx_mult_realy);
|
|
|
|
realcacc = _mm_add_epi16(realcacc, realc);
|
|
imagcacc = _mm_add_epi16(imagcacc, imagc);
|
|
|
|
a += 8;
|
|
b += 8;
|
|
}
|
|
|
|
imagcacc = _mm_slli_si128(imagcacc, 1);
|
|
|
|
totalc = _mm_blendv_epi8(imagcacc, realcacc, mult1);
|
|
|
|
__VOLK_ATTR_ALIGNED(16)
|
|
lv_8sc_t dotProductVector[8];
|
|
|
|
_mm_store_si128((__m128i*)dotProductVector, totalc); // Store the results back into the dot product vector
|
|
|
|
for (i = 0; i < 8; ++i)
|
|
{
|
|
dotProduct += dotProductVector[i];
|
|
}
|
|
}
|
|
|
|
for (i = sse_iters * 8; i < num_points; ++i)
|
|
{
|
|
dotProduct += (*a++) * (*b++);
|
|
}
|
|
|
|
*result = dotProduct;
|
|
}
|
|
|
|
#endif /*LV_HAVE_SSE4_1*/
|
|
|
|
|
|
#ifdef LV_HAVE_ORC
|
|
|
|
extern void volk_gnsssdr_8ic_x2_dot_prod_8ic_a_orc_impl(short* resRealShort, short* resImagShort, const lv_8sc_t* in_a, const lv_8sc_t* in_b, unsigned int num_points);
|
|
static inline void volk_gnsssdr_8ic_x2_dot_prod_8ic_u_orc(lv_8sc_t* result, const lv_8sc_t* in_a, const lv_8sc_t* in_b, unsigned int num_points)
|
|
{
|
|
short resReal = 0;
|
|
char* resRealChar = (char*)&resReal;
|
|
resRealChar++;
|
|
|
|
short resImag = 0;
|
|
char* resImagChar = (char*)&resImag;
|
|
resImagChar++;
|
|
|
|
volk_gnsssdr_8ic_x2_dot_prod_8ic_a_orc_impl(&resReal, &resImag, in_a, in_b, num_points);
|
|
|
|
*result = lv_cmake(*resRealChar, *resImagChar);
|
|
}
|
|
#endif /* LV_HAVE_ORC */
|
|
|
|
|
|
#ifdef LV_HAVE_NEON
|
|
#include <arm_neon.h>
|
|
|
|
static inline void volk_gnsssdr_8ic_x2_dot_prod_8ic_neon(lv_8sc_t* result, const lv_8sc_t* in_a, const lv_8sc_t* in_b, unsigned int num_points)
|
|
{
|
|
lv_8sc_t dotProduct;
|
|
dotProduct = lv_cmake(0, 0);
|
|
*result = lv_cmake(0, 0);
|
|
|
|
const lv_8sc_t* a = in_a;
|
|
const lv_8sc_t* b = in_b;
|
|
// for 2-lane vectors, 1st lane holds the real part,
|
|
// 2nd lane holds the imaginary part
|
|
int8x8x2_t a_val, b_val, c_val, accumulator, tmp_real, tmp_imag;
|
|
__VOLK_ATTR_ALIGNED(16)
|
|
lv_8sc_t accum_result[8] = {lv_cmake(0, 0)};
|
|
accumulator.val[0] = vdup_n_s8(0);
|
|
accumulator.val[1] = vdup_n_s8(0);
|
|
unsigned int number;
|
|
|
|
const unsigned int neon_iters = num_points / 8;
|
|
|
|
for (number = 0; number < neon_iters; ++number)
|
|
{
|
|
a_val = vld2_s8((const int8_t*)a);
|
|
b_val = vld2_s8((const int8_t*)b);
|
|
__VOLK_GNSSSDR_PREFETCH(a + 16);
|
|
__VOLK_GNSSSDR_PREFETCH(b + 16);
|
|
|
|
// multiply the real*real and imag*imag to get real result
|
|
tmp_real.val[0] = vmul_s8(a_val.val[0], b_val.val[0]);
|
|
tmp_real.val[1] = vmul_s8(a_val.val[1], b_val.val[1]);
|
|
|
|
// Multiply cross terms to get the imaginary result
|
|
tmp_imag.val[0] = vmul_s8(a_val.val[0], b_val.val[1]);
|
|
tmp_imag.val[1] = vmul_s8(a_val.val[1], b_val.val[0]);
|
|
|
|
c_val.val[0] = vsub_s8(tmp_real.val[0], tmp_real.val[1]);
|
|
c_val.val[1] = vadd_s8(tmp_imag.val[0], tmp_imag.val[1]);
|
|
|
|
accumulator.val[0] = vadd_s8(accumulator.val[0], c_val.val[0]);
|
|
accumulator.val[1] = vadd_s8(accumulator.val[1], c_val.val[1]);
|
|
|
|
a += 8;
|
|
b += 8;
|
|
}
|
|
vst2_s8((int8_t*)accum_result, accumulator);
|
|
for (number = 0; number < 8; ++number)
|
|
{
|
|
*result += accum_result[number];
|
|
}
|
|
|
|
for (number = neon_iters * 8; number < num_points; ++number)
|
|
{
|
|
dotProduct += (*a++) * (*b++);
|
|
}
|
|
|
|
*result += dotProduct;
|
|
}
|
|
#endif /* LV_HAVE_NEON */
|
|
|
|
#endif /*INCLUDED_volk_gnsssdr_8ic_x2_dot_prod_8ic_H*/
|