remove part of code using _m64 type (so old mmx instructions)

2025-01-14 10:07:59 +01:00
parent ba27aefe4e
commit f919246121
10 changed files with 18 additions and 70 deletions
@@ -247,7 +247,7 @@ int main(int argc, char *argv[])

  ccodelte_init();

-  phy_generate_viterbi_tables_lte(false);
+  phy_generate_viterbi_tables_lte();
  lte_frame_parms = &(PHY_config->lte_frame_parms);
  lte_frame_parms->N_RB_DL            = 25;   //50 for 10MHz and 25 for 5 MHz
  lte_frame_parms->Ncp                = 1;
@@ -399,7 +399,7 @@ void phy_viterbi_lte_sse2(int8_t *y, uint8_t *decoded_bytes, uint16_t n);
 */
 void phy_generate_viterbi_tables(void);

-/*!\fn void phy_generate_viterbi_tables_lte(bool dabflag)
+/*!\fn void phy_generate_viterbi_tables_lte(void)
 \brief This routine initializes metric tables for the optimized LTE Viterbi decoder.
 */
 void phy_generate_viterbi_tables_lte(void);
@@ -47,13 +47,10 @@ static int8_t m1_table[64*16*16*16] __attribute__ ((aligned(16)));


 // Set up Viterbi tables for SSE2 implementation
-void phy_generate_viterbi_tables_lte(bool dabflag)
+void phy_generate_viterbi_tables_lte( void )
 {
  uint8_t ccodelte_table_rev[128];
-  if (dabflag)
-    ccodedab_init_inv(ccodelte_table_rev);
-  else
-    ccodelte_init_inv(ccodelte_table_rev);
+  ccodelte_init_inv(ccodelte_table_rev);

  int8_t w[8],in0,in1,in2;
  uint8_t state,index0,index1;
@@ -63,7 +63,7 @@ void generate_qpsk_table(void) {

 void init_lte_top(LTE_DL_FRAME_PARMS *frame_parms) {
  ccodelte_init();
-  phy_generate_viterbi_tables_lte(false);
+  phy_generate_viterbi_tables_lte();
  load_codinglib();
  generate_ul_ref_sigs();
  generate_ul_ref_sigs_rx();
@@ -27,7 +27,6 @@ static const int16_t conjug[8]__attribute__((aligned(16))) = {-1,1,-1,1,-1,1,-1,
 static const int16_t conjug2[8]__attribute__((aligned(16))) = {1,-1,1,-1,1,-1,1,-1} ;

 #define simd_q15_t simde__m128i
-#define simdshort_q15_t simde__m64
 #define set1_int16(a) simde_mm_set1_epi16(a)
 #define setr_int16(a0, a1, a2, a3, a4, a5, a6, a7) simde_mm_setr_epi16(a0, a1, a2, a3, a4, a5, a6, a7 )

@@ -1513,7 +1513,6 @@ const static int16_t tw64c[96] __attribute__((aligned(32))) = {
 -12539,-30273,-20787,-25330,-27244,-18205,-31356,-9512
                                                 };
 #define simd_q15_t simde__m128i
-#define simdshort_q15_t simde__m64
 #define shiftright_int16(a,shift) simde_mm_srai_epi16(a,shift)
 #define mulhi_int16(a,b) simde_mm_mulhrs_epi16 (a,b)
 #define simd256_q15_t simde__m256i
@@ -909,9 +909,6 @@ const static int16_t tw64c[96] __attribute__((aligned(32))) = {
 #ifdef simd_q15_t
 #undef simd_q15_t
 #endif
-#ifdef simdshort_q15_t 
-#undef simdshort_q15_t
-#endif
 #ifdef shiftright_int16
 #undef shiftright_int16
 #endif
@@ -65,50 +65,6 @@ int32_t signal_energy(int32_t *input,uint32_t length)
  return temp;
 }

-int32_t signal_energy_amp_shift(int32_t *input,uint32_t length)
-{
-
-  int32_t i;
-  int32_t temp,temp2;
-  register simde__m64 mm0,mm1,mm2,mm3;
-  simde__m64 *in = (simde__m64 *)input;
-
-  mm0 = simde_mm_setzero_si64();
-  mm3 = simde_mm_setzero_si64();
-
-  for (i=0; i<length>>1; i++) {
-
-    mm1 = in[i];
-    mm2 = mm1;
-    mm1 = simde_m_pmaddwd(mm1,mm1);
-    mm1 = simde_m_psradi(mm1,AMP_SHIFT);// shift any 32 bits blocs of the word by the value shift_p9
-    mm0 = simde_m_paddd(mm0,mm1);// add the two 64 bits words 4 bytes by 4 bytes
-    mm3 = simde_m_paddw(mm3,mm2);// add the two 64 bits words 2 bytes by 2 bytes
-  }
-
-  mm1 = mm0;
-  mm0 = simde_m_psrlqi(mm0,32);
-  mm0 = simde_m_paddd(mm0,mm1);
-  temp = simde_m_to_int(mm0);
-  temp/=length; // this is the average of x^2
-
-
-  // now remove the DC component
-
-
-  mm2 = simde_m_psrlqi(mm3,32);
-  mm2 = simde_m_paddw(mm2,mm3);
-  mm2 = simde_m_pmaddwd(mm2,mm2);
-  mm2 = simde_m_psradi(mm2,AMP_SHIFT); // fixed point representation of elements
-  temp2 = simde_m_to_int(mm2);
-  temp2/=(length*length);
-
-  temp -= temp2;
-
-
-  return((temp>0)?temp:1);
-}
-
 uint32_t signal_energy_nodc(const c16_t *input, uint32_t length)
 {
  // init
@@ -41,7 +41,6 @@
 #include "common/utils/LOG/log.h"

 #define simd_q15_t simde__m128i
-#define simdshort_q15_t simde__m64
 #define shiftright_int16(a,shift) simde_mm_srai_epi16(a,shift)
 #define set1_int16(a) simde_mm_set1_epi16(a)
 #define mulhi_int16(a,b) simde_mm_mulhrs_epi16 (a,b)
@@ -727,8 +726,6 @@ int32_t sub_cpx_vector16(int16_t *x,
 */
 int32_t signal_energy(int32_t *,uint32_t);

-int32_t signal_energy_amp_shift(int32_t *input, uint32_t length);
-
 #ifdef LOCALIZATION
 /*!\fn int32_t signal_energy(int *,uint32_t);
 \brief Computes the signal energy per subcarrier
@@ -194,9 +194,6 @@ void nr_64qam_llr(int32_t *rxdataF_comp, int32_t *ch_mag, int32_t *ch_mag2, int1
  simde__m128i *rxF_128 = (simde__m128i *)rxF;
  simde__m128i *ch_mag_128 = (simde__m128i *)ch_maga;
  simde__m128i *ch_magb_128 = (simde__m128i *)ch_magb;
-
-  simde__m64 *llr64 = (simde__m64 *)llr_32;
-
  // Each iteration does 4 RE (gives 24 16bit-llrs)
  for (int i = 0; i < (nb_re >> 2); i++) {
    simde__m128i xmm0, xmm1, xmm2;
@@ -206,12 +203,18 @@ void nr_64qam_llr(int32_t *rxdataF_comp, int32_t *ch_mag, int32_t *ch_mag2, int1
    xmm2 = simde_mm_abs_epi16(xmm1);
    xmm2 = simde_mm_subs_epi16(*ch_magb_128, xmm2);

-    *llr64++ = simde_mm_set_pi32(simde_mm_extract_epi32(xmm1, 0), simde_mm_extract_epi32(xmm0, 0));
-    *llr64++ = simde_mm_set_pi32(simde_mm_extract_epi32(xmm0, 1), simde_mm_extract_epi32(xmm2, 0));
-    *llr64++ = simde_mm_set_pi32(simde_mm_extract_epi32(xmm2, 1), simde_mm_extract_epi32(xmm1, 1));
-    *llr64++ = simde_mm_set_pi32(simde_mm_extract_epi32(xmm1, 2), simde_mm_extract_epi32(xmm0, 2));
-    *llr64++ = simde_mm_set_pi32(simde_mm_extract_epi32(xmm0, 3), simde_mm_extract_epi32(xmm2, 2));
-    *llr64++ = simde_mm_set_pi32(simde_mm_extract_epi32(xmm2, 3), simde_mm_extract_epi32(xmm1, 3));
+    *llr_32++ = simde_mm_extract_epi32(xmm0, 0);
+    *llr_32++ = simde_mm_extract_epi32(xmm1, 0);
+    *llr_32++ = simde_mm_extract_epi32(xmm2, 0);
+    *llr_32++ = simde_mm_extract_epi32(xmm0, 1);
+    *llr_32++ = simde_mm_extract_epi32(xmm1, 1);
+    *llr_32++ = simde_mm_extract_epi32(xmm2, 1);
+    *llr_32++ = simde_mm_extract_epi32(xmm0, 2);
+    *llr_32++ = simde_mm_extract_epi32(xmm1, 2);
+    *llr_32++ = simde_mm_extract_epi32(xmm2, 2);
+    *llr_32++ = simde_mm_extract_epi32(xmm0, 3);
+    *llr_32++ = simde_mm_extract_epi32(xmm1, 3);
+    *llr_32++ = simde_mm_extract_epi32(xmm2, 3);
    rxF_128++;
    ch_mag_128++;
    ch_magb_128++;
@@ -222,7 +225,7 @@ void nr_64qam_llr(int32_t *rxdataF_comp, int32_t *ch_mag, int32_t *ch_mag2, int1
  int16_t *rxDataF_i16 = (int16_t *)rxF_128;
  int16_t *ch_mag_i16 = (int16_t *)ch_mag_128;
  int16_t *ch_magb_i16 = (int16_t *)ch_magb_128;
-  int16_t *llr_i16 = (int16_t *)llr64;
+  int16_t *llr_i16 = (int16_t *)llr_32;
  for (int i = 0; i < nb_re; i++) {
    int16_t real = rxDataF_i16[2 * i];
    int16_t imag = rxDataF_i16[2 * i + 1];