PentHertz
/
srsLTE
mirror of https://github.com/PentHertz/srsLTE.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

2rx equalizer working in matlab in gen/sse/avx

This commit is contained in:
Ismael Gomez 2016-09-12 09:57:20 -04:00
parent 5fcb95d54d
commit 67d2b91354
1 changed files with 41 additions and 13 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

54
srslte/lib/mimo/precoding.c
View File

@ -149,41 +149,69 @@ int srslte_predecoding_single_sse(cf_t *y[SRSLTE_MAX_RXANT], cf_t *h[SRSLTE_MAX_
int srslte_predecoding_single_avx(cf_t *y[SRSLTE_MAX_RXANT], cf_t *h[SRSLTE_MAX_RXANT], cf_t *x, int nof_rxant, int nof_symbols, float noise_estimate) {
float *xPtr = (float*) x;
const float *hPtr = (const float*) h;
const float *yPtr = (const float*) y;
const float *hPtr1 = (const float*) h[0];
const float *yPtr1 = (const float*) y[0];
const float *hPtr2 = (const float*) h[1];
const float *yPtr2 = (const float*) y[1];
__m256 conjugator = _mm256_setr_ps(0, -0.f, 0, -0.f, 0, -0.f, 0, -0.f);
__m256 noise = _mm256_set1_ps(noise_estimate);
__m256 h1Val, h2Val, y1Val, y2Val, h12square, h1square, h2square, h1_p, h2_p, h1conj, h2conj, x1Val, x2Val;
__m256 h1Val1, h2Val1, y1Val1, y2Val1, h12square, h1square, h2square, h1_p, h2_p, h1conj1, h2conj1, x1Val, x2Val;
__m256 h1Val2, h2Val2, y1Val2, y2Val2, h1conj2, h2conj2;
printf("using avx\n");
for (int i=0;i<nof_symbols/8;i++) {
y1Val = _mm256_load_ps(yPtr); yPtr+=8;
y2Val = _mm256_load_ps(yPtr); yPtr+=8;
h1Val = _mm256_load_ps(hPtr); hPtr+=8;
h2Val = _mm256_load_ps(hPtr); hPtr+=8;
y1Val1 = _mm256_load_ps(yPtr1); yPtr1+=8;
y2Val1 = _mm256_load_ps(yPtr1); yPtr1+=8;
h1Val1 = _mm256_load_ps(hPtr1); hPtr1+=8;
h2Val1 = _mm256_load_ps(hPtr1); hPtr1+=8;
if (nof_rxant == 2) {
y1Val2 = _mm256_load_ps(yPtr2); yPtr2+=8;
y2Val2 = _mm256_load_ps(yPtr2); yPtr2+=8;
h1Val2 = _mm256_load_ps(hPtr2); hPtr2+=8;
h2Val2 = _mm256_load_ps(hPtr2); hPtr2+=8;
}
__m256 t1 = _mm256_mul_ps(h1Val, h1Val);
__m256 t2 = _mm256_mul_ps(h2Val, h2Val);
__m256 t1 = _mm256_mul_ps(h1Val1, h1Val1);
__m256 t2 = _mm256_mul_ps(h2Val1, h2Val1);
h12square = _mm256_hadd_ps(_mm256_permute2f128_ps(t1, t2, 0x20), _mm256_permute2f128_ps(t1, t2, 0x31));
if (nof_rxant == 2) {
t1 = _mm256_mul_ps(h1Val2, h1Val2);
t2 = _mm256_mul_ps(h2Val2, h2Val2);
h12square = _mm256_add_ps(h12square, _mm256_hadd_ps(_mm256_permute2f128_ps(t1, t2, 0x20), _mm256_permute2f128_ps(t1, t2, 0x31)));
}
if (noise_estimate > 0) {
h12square = _mm256_add_ps(h12square, noise);
}
h1_p = _mm256_permute_ps(h12square, _MM_SHUFFLE(1, 1, 0, 0));
h2_p = _mm256_permute_ps(h12square, _MM_SHUFFLE(3, 3, 2, 2));
h1square = _mm256_permute2f128_ps(h1_p, h2_p, 2<<4);
h2square = _mm256_permute2f128_ps(h1_p, h2_p, 3<<4 | 1);
/* Conjugate channel */
h1conj = _mm256_xor_ps(h1Val, conjugator);
h2conj = _mm256_xor_ps(h2Val, conjugator);
h1conj1 = _mm256_xor_ps(h1Val1, conjugator);
h2conj1 = _mm256_xor_ps(h2Val1, conjugator);
if (nof_rxant == 2) {
h1conj2 = _mm256_xor_ps(h1Val2, conjugator);
h2conj2 = _mm256_xor_ps(h2Val2, conjugator);
}
/* Complex product */
x1Val = PROD_AVX(y1Val, h1conj);
x2Val = PROD_AVX(y2Val, h2conj);
x1Val = PROD_AVX(y1Val1, h1conj1);
x2Val = PROD_AVX(y2Val1, h2conj1);
if (nof_rxant == 2) {
x1Val = _mm256_add_ps(x1Val, PROD_AVX(y1Val2, h1conj2));
x2Val = _mm256_add_ps(x2Val, PROD_AVX(y2Val2, h2conj2));
}
x1Val = _mm256_div_ps(x1Val, h1square);
x2Val = _mm256_div_ps(x2Val, h2square);