Commit 1a5efc74 authored by Francesco Mani's avatar Francesco Mani

Merge branch 'ldpc-endian' into nr-coreset-bug-fix

parents 7e3d2cc0 2109c768
...@@ -362,7 +362,7 @@ int test_ldpc(short No_iteration, ...@@ -362,7 +362,7 @@ int test_ldpc(short No_iteration,
printf("\ne %d..%d: ",i,i+15); printf("\ne %d..%d: ",i,i+15);
#endif #endif
if (channel_input[j][i-2*Zc]==0) if (channel_input_optim[j][i-2*Zc]==0)
modulated_input[j][i]=1.0;///sqrt(2); //QPSK modulated_input[j][i]=1.0;///sqrt(2); //QPSK
else else
modulated_input[j][i]=-1.0;///sqrt(2); modulated_input[j][i]=-1.0;///sqrt(2);
...@@ -386,7 +386,7 @@ int test_ldpc(short No_iteration, ...@@ -386,7 +386,7 @@ int test_ldpc(short No_iteration,
else else
channel_output_uncoded[j][i]=0; channel_output_uncoded[j][i]=0;
if (channel_output_uncoded[j][i] != channel_input[j][i-2*Zc]) if (channel_output_uncoded[j][i] != channel_input_optim[j][i-2*Zc])
*errors_bit_uncoded = (*errors_bit_uncoded) + 1; *errors_bit_uncoded = (*errors_bit_uncoded) + 1;
} }
......
...@@ -47,7 +47,7 @@ static inline void nrLDPC_bnProcPc(t_nrLDPC_lut* p_lut, t_nrLDPC_procBuf* p_proc ...@@ -47,7 +47,7 @@ static inline void nrLDPC_bnProcPc(t_nrLDPC_lut* p_lut, t_nrLDPC_procBuf* p_proc
int8_t* bnProcBufRes = p_procBuf->bnProcBufRes; int8_t* bnProcBufRes = p_procBuf->bnProcBufRes;
int8_t* llrRes = p_procBuf->llrRes; int8_t* llrRes = p_procBuf->llrRes;
int8_t* llrProcBuf = p_procBuf->llrProcBuf; int8_t* llrProcBuf = p_procBuf->llrProcBuf;
__m128i* p_bnProcBuf; __m128i* p_bnProcBuf;
__m256i* p_bnProcBufRes; __m256i* p_bnProcBufRes;
__m128i* p_llrProcBuf; __m128i* p_llrProcBuf;
...@@ -1693,7 +1693,7 @@ static inline void nrLDPC_bnProc(t_nrLDPC_lut* p_lut, t_nrLDPC_procBuf* p_procBu ...@@ -1693,7 +1693,7 @@ static inline void nrLDPC_bnProc(t_nrLDPC_lut* p_lut, t_nrLDPC_procBuf* p_procBu
int8_t* bnProcBuf = p_procBuf->bnProcBuf; int8_t* bnProcBuf = p_procBuf->bnProcBuf;
int8_t* bnProcBufRes = p_procBuf->bnProcBufRes; int8_t* bnProcBufRes = p_procBuf->bnProcBufRes;
int8_t* llrRes = p_procBuf->llrRes; int8_t* llrRes = p_procBuf->llrRes;
__m256i* p_bnProcBuf; __m256i* p_bnProcBuf;
__m256i* p_bnProcBufRes; __m256i* p_bnProcBufRes;
__m256i* p_llrRes; __m256i* p_llrRes;
...@@ -2775,24 +2775,35 @@ static inline void nrLDPC_llr2bit(int8_t* out, int8_t* llrOut, uint16_t numLLR) ...@@ -2775,24 +2775,35 @@ static inline void nrLDPC_llr2bit(int8_t* out, int8_t* llrOut, uint16_t numLLR)
} }
/** /**
\brief Performs hard-decision on output LLRs and packs the output in 32 bit values. \brief Performs hard-decision on output LLRs and packs the output in byte aligned output according to TS 38.321 Section 6.1.1.
i = 0,1,2,...
IN[i] : a0, a1, a2, ..., a_{A-1}
OUT[i]: a7,a6,a5,a4,a3,a2,a1,a0|a15,14,...,a8|a23,a22,...,a16|a31,a30,...,a24|...
\param out Pointer hard-decision output, every int8_t contains 8 bits \param out Pointer hard-decision output, every int8_t contains 8 bits
\param llrOut Pointer to output LLRs \param llrOut Pointer to output LLRs
\param numLLR Number of LLRs \param numLLR Number of LLRs
*/ */
static inline void nrLDPC_llr2bitPacked(int8_t* out, int8_t* llrOut, uint16_t numLLR) static inline void nrLDPC_llr2bitPacked(int8_t* out, int8_t* llrOut, uint16_t numLLR)
{ {
/** Vector of indices for shuffling input */
const uint8_t constShuffle_256_epi8[32] __attribute__ ((aligned(32))) = {7,6,5,4,3,2,1,0,15,14,13,12,11,10,9,8,7,6,5,4,3,2,1,0,15,14,13,12,11,10,9,8};
__m256i* p_llrOut = (__m256i*) llrOut; __m256i* p_llrOut = (__m256i*) llrOut;
uint32_t* p_bits = (uint32_t*) out; uint32_t* p_bits = (uint32_t*) out;
__m256i inPerm;
int8_t* p_llrOut8; int8_t* p_llrOut8;
uint32_t bitsTmp = 0; uint32_t bitsTmp = 0;
uint32_t i; uint32_t i;
uint32_t M = numLLR>>5; uint32_t M = numLLR>>5;
uint32_t Mr = numLLR&31; uint32_t Mr = numLLR&31;
const __m256i* p_shuffle = (__m256i*) constShuffle_256_epi8;
for (i=0; i<M; i++) for (i=0; i<M; i++)
{ {
*p_bits++ = _mm256_movemask_epi8(*p_llrOut); // Move LSB to MSB on 8 bits
inPerm = _mm256_shuffle_epi8(*p_llrOut,*p_shuffle);
// Hard decision
*p_bits++ = _mm256_movemask_epi8(inPerm);
p_llrOut++; p_llrOut++;
} }
...@@ -2805,11 +2816,11 @@ static inline void nrLDPC_llr2bitPacked(int8_t* out, int8_t* llrOut, uint16_t nu ...@@ -2805,11 +2816,11 @@ static inline void nrLDPC_llr2bitPacked(int8_t* out, int8_t* llrOut, uint16_t nu
{ {
if (p_llrOut8[i] < 0) if (p_llrOut8[i] < 0)
{ {
bitsTmp |= (1<<i); bitsTmp |= (1<<((7-i) + (16*(i/8))));
} }
else else
{ {
bitsTmp &= (0<<i); bitsTmp |= (0<<((7-i) + (16*(i/8))));
} }
} }
} }
......
...@@ -273,7 +273,7 @@ int ldpc_encoder_optim(unsigned char *test_input,unsigned char *channel_input,sh ...@@ -273,7 +273,7 @@ int ldpc_encoder_optim(unsigned char *test_input,unsigned char *channel_input,sh
if(tinput != NULL) start_meas(tinput); if(tinput != NULL) start_meas(tinput);
for (i=0; i<block_length; i++) { for (i=0; i<block_length; i++) {
c[i] = (test_input[i/8]&(128>>(i&7)))>>(i&7); c[i] = (test_input[i/8]&(128>>(i&7)))>>(7-(i&7));
//printf("c(%d,%d)=%d\n",j,i,temp); //printf("c(%d,%d)=%d\n",j,i,temp);
} }
...@@ -330,7 +330,7 @@ int ldpc_encoder_optim_8seg(unsigned char **test_input,unsigned char **channel_i ...@@ -330,7 +330,7 @@ int ldpc_encoder_optim_8seg(unsigned char **test_input,unsigned char **channel_i
#ifdef __AVX2__ #ifdef __AVX2__
__m256i shufmask = _mm256_set_epi64x(0x0303030303030303, 0x0202020202020202,0x0101010101010101, 0x0000000000000000); __m256i shufmask = _mm256_set_epi64x(0x0303030303030303, 0x0202020202020202,0x0101010101010101, 0x0000000000000000);
__m256i andmask = _mm256_set1_epi64x(0x8040201008040201); // every 8 bits -> 8 bytes, pattern repeats. __m256i andmask = _mm256_set1_epi64x(0x0102040810204080); // every 8 bits -> 8 bytes, pattern repeats.
__m256i zero256 = _mm256_setzero_si256(); __m256i zero256 = _mm256_setzero_si256();
__m256i masks[8]; __m256i masks[8];
register __m256i c256; register __m256i c256;
...@@ -413,7 +413,7 @@ int ldpc_encoder_optim_8seg(unsigned char **test_input,unsigned char **channel_i ...@@ -413,7 +413,7 @@ int ldpc_encoder_optim_8seg(unsigned char **test_input,unsigned char **channel_i
for (i=0; i<block_length; i++) { for (i=0; i<block_length; i++) {
for (j=0; j<n_segments; j++) { for (j=0; j<n_segments; j++) {
temp = (test_input[j][i/8]&(1<<(i&7)))>>(i&7); temp = (test_input[j][i/8]&(128>>(i&7)))>>(7-(i&7));
//printf("c(%d,%d)=%d\n",j,i,temp); //printf("c(%d,%d)=%d\n",j,i,temp);
c[i] |= (temp << j); c[i] |= (temp << j);
} }
...@@ -431,7 +431,7 @@ int ldpc_encoder_optim_8seg(unsigned char **test_input,unsigned char **channel_i ...@@ -431,7 +431,7 @@ int ldpc_encoder_optim_8seg(unsigned char **test_input,unsigned char **channel_i
for (i=(block_length>>5)<<5;i<block_length;i++) { for (i=(block_length>>5)<<5;i<block_length;i++) {
for (j=0; j<n_segments; j++) { for (j=0; j<n_segments; j++) {
temp = (test_input[j][i/8]&(128>>(i&7)))>>(i&7); temp = (test_input[j][i/8]&(128>>(i&7)))>>(7-(i&7));
//printf("c(%d,%d)=%d\n",j,i,temp); //printf("c(%d,%d)=%d\n",j,i,temp);
c[i] |= (temp << j); c[i] |= (temp << j);
} }
...@@ -639,7 +639,7 @@ int ldpc_encoder_optim_8seg_multi(unsigned char **test_input,unsigned char **cha ...@@ -639,7 +639,7 @@ int ldpc_encoder_optim_8seg_multi(unsigned char **test_input,unsigned char **cha
//for (j=0; j<n_segments; j++) { //for (j=0; j<n_segments; j++) {
for (j=macro_segment; j < macro_segment_end; j++) { for (j=macro_segment; j < macro_segment_end; j++) {
temp = (test_input[j][i/8]&(128>>(i&7)))>>(i&7); temp = (test_input[j][i/8]&(128>>(i&7)))>>(7-(i&7));
//printf("c(%d,%d)=%d\n",j,i,temp); //printf("c(%d,%d)=%d\n",j,i,temp);
c[i] |= (temp << (j-macro_segment)); c[i] |= (temp << (j-macro_segment));
} }
......
...@@ -460,7 +460,7 @@ int ldpc_encoder_orig(unsigned char *test_input,unsigned char *channel_input,sho ...@@ -460,7 +460,7 @@ int ldpc_encoder_orig(unsigned char *test_input,unsigned char *channel_input,sho
{ {
//c[i] = test_input[i/8]<<(i%8); //c[i] = test_input[i/8]<<(i%8);
//c[i]=c[i]>>7&1; //c[i]=c[i]>>7&1;
c[i]=(test_input[i/8]&(1<<(i&7)))>>(i&7); c[i]=(test_input[i/8]&(128>>(i&7)))>>(7-(i&7));
} }
// parity check part // parity check part
......
Markdown is supported
0%
or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment