int main(int argc, char **argv) {
parse_args(argc, argv);
srslte_tcod_gentable();
srslte_tcod_t tcod;
srslte_tcod_init(&tcod, 6144);
uint32_t st=0, end=187;
if (long_cb) {
st=srslte_cbsegm_cbindex(long_cb);
end=st;
}
for (uint32_t len=st;len<=end;len++) {
long_cb = srslte_cbsegm_cbsize(len);
printf("Checking long_cb=%d\n", long_cb);
for (int i=0;i<long_cb/8;i++) {
input_bytes[i] = rand()%256;
}
srslte_bit_unpack_vector(input_bytes, input_bits, long_cb);
if (SRSLTE_VERBOSE_ISINFO()) {
printf("Input bits:\n");
for (int i=0;i<long_cb/8;i++) {
srslte_vec_fprint_b(stdout, &input_bits[i*8], 8);
}
}
srslte_tcod_encode(&tcod, input_bits, output_bits, long_cb);
srslte_tcod_encode_lut(&tcod, input_bytes, parity, len);
srslte_bit_unpack_vector(parity, parity_bits, 2*(long_cb+4));
for (int i=0;i<long_cb;i++) {
output_bits2[3*i] = input_bits[i];
output_bits2[3*i+1] = parity_bits[i];
output_bits2[3*i+2] = parity_bits[i+long_cb+4];
}
if (SRSLTE_VERBOSE_ISINFO()) {
srslte_vec_fprint_b(stdout, output_bits2, 3*long_cb);
srslte_vec_fprint_b(stdout, output_bits, 3*long_cb);
printf("\n");
}
for (int i=0;i<2*long_cb;i++) {
if (output_bits2[long_cb+i] != output_bits[long_cb+i]) {
printf("error in bit %d, len=%d\n", i, len);
exit(-1);
}
}
}
srslte_tcod_free(&tcod);
printf("Done\n");
exit(0);
}
/**
* Undoes rate matching for LTE Turbo Coder. Expands rate matched buffer to full size buffer.
*
* @param[in] input Input buffer of size in_len
* @param[out] output Output buffer of size 3*srslte_cbsegm_cbsize(cb_idx)+12
* @param[in] cb_idx Code block table index
* @param[in] rv_idx Redundancy Version from DCI control message
* @return Error code
*/
int srslte_rm_turbo_rx_lut_(int16_t *input, int16_t *output, uint32_t in_len, uint32_t cb_idx, uint32_t rv_idx, bool enable_input_tdec)
{
if (rv_idx < 4 && cb_idx < SRSLTE_NOF_TC_CB_SIZES) {
#if SRSLTE_TDEC_EXPECT_INPUT_SB == 1
int cb_len=srslte_cbsegm_cbsize(cb_idx);
int idx = deinter_table_idx_from_sb_len(srslte_tdec_autoimp_get_subblocks(cb_len));
uint16_t *deinter = NULL;
if (idx < 0 || !enable_input_tdec) {
deinter = deinterleaver[cb_idx][rv_idx];
} else if (idx < NOF_DEINTER_TABLE_SB_IDX) {
deinter = deinterleaver_sb[idx][cb_idx][rv_idx];
} else {
fprintf(stderr, "Sub-block size index %d not supported in srslte_rm_turbo_rx_lut()\n", idx);
return -1;
}
#else
uint16_t *deinter = deinterleaver[cb_idx][rv_idx];
#endif
#ifdef LV_HAVE_AVX
return srslte_rm_turbo_rx_lut_avx(input, output, deinter, in_len, cb_idx, rv_idx);
#else
#ifdef LV_HAVE_SSE
return srslte_rm_turbo_rx_lut_sse(input, output, deinter, in_len, cb_idx, rv_idx);
#else
uint32_t out_len = 3*srslte_cbsegm_cbsize(cb_idx)+12;
for (int i=0;i<in_len;i++) {
output[deinter[i%out_len]] += input[i];
}
return 0;
#endif
#endif
} else {
printf("Invalid inputs rv_idx=%d, cb_idx=%d\n", rv_idx, cb_idx);
return SRSLTE_ERROR_INVALID_INPUTS;
}
}
void srslte_rm_turbo_gentables() {
if (!rm_turbo_tables_generated) {
rm_turbo_tables_generated = true;
for (int cb_idx=0;cb_idx<SRSLTE_NOF_TC_CB_SIZES;cb_idx++) {
int cb_len = srslte_cbsegm_cbsize(cb_idx);
int in_len = 3 * cb_len + 12;
int nrows = (in_len / 3 - 1) / NCOLS + 1;
int K_p = nrows * NCOLS;
int ndummy = K_p - in_len / 3;
if (ndummy < 0) {
ndummy = 0;
}
for (int i = 0; i < 4; i++) {
k0_vec[cb_idx][i][0] = nrows * (2 * (uint16_t) ceilf((float) (3 * K_p) / (float) (8 * nrows)) * i + 2);
k0_vec[cb_idx][i][1] = -1;
}
srslte_rm_turbo_gentable_systematic(interleaver_systematic_bits[cb_idx], k0_vec[cb_idx], nrows, ndummy);
srslte_bit_interleaver_init(&bit_interleavers_systematic_bits[cb_idx], interleaver_systematic_bits[cb_idx],
(uint32_t) srslte_cbsegm_cbsize(cb_idx) + 4);
srslte_rm_turbo_gentable_parity(interleaver_parity_bits[cb_idx], k0_vec[cb_idx], in_len / 3, nrows, ndummy);
srslte_bit_interleaver_init(&bit_interleavers_parity_bits[cb_idx], interleaver_parity_bits[cb_idx],
(uint32_t) (srslte_cbsegm_cbsize(cb_idx) + 4) * 2);
for (int i = 0; i < 4; i++) {
srslte_rm_turbo_gentable_receive(deinterleaver[cb_idx][i], in_len, i);
#if SRSLTE_TDEC_EXPECT_INPUT_SB == 1
for (uint32_t s = 0; s < NOF_DEINTER_TABLE_SB_IDX; s++) {
interleave_table_sb(deinterleaver[cb_idx][i], deinterleaver_sb[s][cb_idx][i], cb_idx,
deinter_table_sb_idx[s]);
}
#endif
}
}
}
}
int srslte_rm_turbo_rx_lut_8bit(int8_t *input, int8_t *output, uint32_t in_len, uint32_t cb_idx, uint32_t rv_idx)
{
if (rv_idx < 4 && cb_idx < SRSLTE_NOF_TC_CB_SIZES) {
#if SRSLTE_TDEC_EXPECT_INPUT_SB == 1
int cb_len=srslte_cbsegm_cbsize(cb_idx);
int idx = deinter_table_idx_from_sb_len(srslte_tdec_autoimp_get_subblocks_8bit(cb_len));
uint16_t *deinter = NULL;
if (idx < 0) {
deinter = deinterleaver[cb_idx][rv_idx];
} else if (idx < NOF_DEINTER_TABLE_SB_IDX) {
deinter = deinterleaver_sb[idx][cb_idx][rv_idx];
} else {
fprintf(stderr, "Sub-block size index %d not supported in srslte_rm_turbo_rx_lut()\n", idx);
return -1;
}
#else
uint16_t *deinter = deinterleaver[cb_idx][rv_idx];
#endif
// FIXME: AVX version of rm_turbo_rx_lut not working
// Warning: Need to check if 8-bit sse version is correct
#ifdef LV_HAVE_SSE
return srslte_rm_turbo_rx_lut_sse_8bit(input, output, deinter, in_len, cb_idx, rv_idx);
#else
uint32_t out_len = 3*srslte_cbsegm_cbsize(cb_idx)+12;
for (int i=0;i<in_len;i++) {
output[deinter[i%out_len]] += input[i];
}
return 0;
#endif
} else {
printf("Invalid inputs rv_idx=%d, cb_idx=%d\n", rv_idx, cb_idx);
return SRSLTE_ERROR_INVALID_INPUTS;
}
}
/* Prepare output for sliding window decoder:
* (0..long_cb-1) bits are systematic
* (long_cb..2*long_cb-1) are parity0
* (2*long_cb..3*long_cb-1) are parity1
* then tail bits
*
* Within each block, bits are interleaved every nof_sb
*/
static void interleave_table_sb(uint16_t *in, uint16_t *out, uint32_t cb_idx, uint32_t nof_sb)
{
int long_cb = srslte_cbsegm_cbsize(cb_idx);
int out_len = 3*long_cb+12;
for (int i=0;i<out_len;i++) {
// Do not change tail bit order
if (in[i] < 3*long_cb) {
// align to 32 bytes (warning: must be same alignment as in rm_turbo.c)
out[i] = (in[i]%3)*(long_cb+32)+inter(in[i]/3,nof_sb);
} else {
out[i] = (in[i]-3*long_cb)+3*(long_cb+32);
}
}
}
/**
* Rate matching for LTE Turbo Coder
*
* @param[out] w_buff Preallocated softbuffer
* @param[in] systematic Input code block in a byte array
* @param[in] parity Input code turbo coder parity bits in a byte array
* @param[out] output Rate matched output array of size out_len
* @param out_len Output buffer size to be filled with as many FEC bits as fit
* @param w_offset Start writing to output at this bit offset
* @param cb_idx Code block index. Used to lookup interleaver parameters
* @param rv_idx Redundancy Version Index. Indexed offset of FEC bits to copy
*
* @return Error code
*/
int srslte_rm_turbo_tx_lut(uint8_t *w_buff, uint8_t *systematic, uint8_t *parity, uint8_t *output,
uint32_t cb_idx, uint32_t out_len,
uint32_t w_offset, uint32_t rv_idx)
{
if (rv_idx < 4 && cb_idx < SRSLTE_NOF_TC_CB_SIZES) {
int in_len=3*srslte_cbsegm_cbsize(cb_idx)+12;
/* Sub-block interleaver (5.1.4.1.1) and bit collection */
if (rv_idx == 0) {
// Systematic bits
//srslte_bit_interleave(systematic, w_buff, interleaver_systematic_bits[cb_idx], in_len/3);
srslte_bit_interleaver_run(&bit_interleavers_systematic_bits[cb_idx], systematic, w_buff, 0);
// Parity bits
//srslte_bit_interleave_w_offset(parity, &w_buff[in_len/24], interleaver_parity_bits[cb_idx], 2*in_len/3, 4);
srslte_bit_interleaver_run(&bit_interleavers_parity_bits[cb_idx], parity, &w_buff[in_len/24], 4);
}
/* Bit selection and transmission 5.1.4.1.2 */
int w_len = 0;
int r_ptr = k0_vec[cb_idx][rv_idx][1];
while (w_len < out_len) {
int cp_len = out_len - w_len;
if (cp_len + r_ptr >= in_len) {
cp_len = in_len - r_ptr;
}
srslte_bit_copy(output, w_len+w_offset, w_buff, r_ptr, cp_len);
r_ptr += cp_len;
if (r_ptr >= in_len) {
r_ptr -= in_len;
}
w_len += cp_len;
}
return 0;
} else {
return SRSLTE_ERROR_INVALID_INPUTS;
}
}
int main(int argc, char **argv) {
parse_args(argc, argv);
srslte_tcod_gentable();
srslte_tcod_t tcod;
srslte_tcod_init(&tcod, 6144);
uint32_t st=0, end=187;
if (long_cb) {
st=srslte_cbsegm_cbindex(long_cb);
end=st;
}
for (uint32_t len=st;len<=end;len++) {
long_cb = srslte_cbsegm_cbsize(len);
printf("Checking long_cb=%d\n", long_cb);
for (int i=0;i<long_cb/8;i++) {
input_bytes[i] = rand()%256;
}
srslte_bit_unpack_vector(input_bytes, input_bits, long_cb);
if (SRSLTE_VERBOSE_ISINFO()) {
printf("Input bits:\n");
for (int i=0;i<long_cb/8;i++) {
srslte_vec_fprint_b(stdout, &input_bits[i*8], 8);
}
}
srslte_tcod_encode(&tcod, input_bits, output_bits, long_cb);
srslte_tcod_encode_lut(&tcod, input_bytes, output_bytes, long_cb);
srslte_bit_unpack_vector(output_bytes, output_bits2, 2*long_cb+12);
/* de-Interleace bits for comparison */
for (int i=0;i<long_cb;i++) {
for (int j=0;j<2;j++) {
output_bits3[j*long_cb+i] = output_bits[3*i+j+1];
}
}
// copy tail
memcpy(&output_bits3[2*long_cb], &output_bits[3*long_cb], 12);
if (SRSLTE_VERBOSE_ISINFO()) {
printf("1st encoder\n");
srslte_vec_fprint_b(stdout, output_bits2, long_cb);
srslte_vec_fprint_b(stdout, output_bits3, long_cb);
printf("2nd encoder\n");
srslte_vec_fprint_b(stdout, &output_bits2[long_cb], long_cb);
srslte_vec_fprint_b(stdout, &output_bits3[long_cb], long_cb);
printf("tail\n");
srslte_vec_fprint_b(stdout, &output_bits2[2*long_cb], 12);
srslte_vec_fprint_b(stdout, &output_bits3[2*long_cb], 12);
printf("\n");
}
for (int i=0;i<2*long_cb+12;i++) {
if (output_bits2[i] != output_bits3[i]) {
printf("error in bit %d, len=%d\n", i, len);
exit(-1);
}
}
}
srslte_tcod_free(&tcod);
printf("Done\n");
exit(0);
}
/* Initializes the turbo decoder object */
int srslte_tdec_sse_init(srslte_tdec_sse_t * h, uint32_t max_long_cb)
{
int ret = -1;
bzero(h, sizeof(srslte_tdec_sse_t));
uint32_t len = max_long_cb + SRSLTE_TCOD_TOTALTAIL;
h->max_long_cb = max_long_cb;
h->app1 = srslte_vec_malloc(sizeof(int16_t) * len);
if (!h->app1) {
perror("srslte_vec_malloc");
goto clean_and_exit;
}
h->app2 = srslte_vec_malloc(sizeof(int16_t) * len);
if (!h->app2) {
perror("srslte_vec_malloc");
goto clean_and_exit;
}
h->ext1 = srslte_vec_malloc(sizeof(int16_t) * len);
if (!h->ext1) {
perror("srslte_vec_malloc");
goto clean_and_exit;
}
h->ext2 = srslte_vec_malloc(sizeof(int16_t) * len);
if (!h->ext2) {
perror("srslte_vec_malloc");
goto clean_and_exit;
}
h->syst = srslte_vec_malloc(sizeof(int16_t) * len);
if (!h->syst) {
perror("srslte_vec_malloc");
goto clean_and_exit;
}
h->parity0 = srslte_vec_malloc(sizeof(int16_t) * len);
if (!h->parity0) {
perror("srslte_vec_malloc");
goto clean_and_exit;
}
h->parity1 = srslte_vec_malloc(sizeof(int16_t) * len);
if (!h->parity1) {
perror("srslte_vec_malloc");
goto clean_and_exit;
}
if (map_gen_init(&h->dec, h->max_long_cb)) {
goto clean_and_exit;
}
for (int i=0;i<SRSLTE_NOF_TC_CB_SIZES;i++) {
if (srslte_tc_interl_init(&h->interleaver[i], srslte_cbsegm_cbsize(i)) < 0) {
goto clean_and_exit;
}
srslte_tc_interl_LTE_gen(&h->interleaver[i], srslte_cbsegm_cbsize(i));
}
h->current_cbidx = -1;
ret = 0;
clean_and_exit:if (ret == -1) {
srslte_tdec_sse_free(h);
}
return ret;
}
int srslte_rm_turbo_rx_lut_avx_8bit(int8_t *input, int8_t *output, uint16_t *deinter, uint32_t in_len, uint32_t cb_idx, uint32_t rv_idx)
{
if (rv_idx < 4 && cb_idx < SRSLTE_NOF_TC_CB_SIZES) {
uint32_t out_len = 3*srslte_cbsegm_cbsize(cb_idx)+12;
const __m256i* xPtr = (const __m256i*) input;
const __m256i* lutPtr = (const __m256i*) deinter;
__m256i xVal, lutVal1, lutVal2;
int8_t x;
uint16_t l;
/* Simplify load if we do not need to wrap (ie high rates) */
if (in_len <= out_len) {
for (int i=0;i<in_len/32;i++) {
xVal = _mm256_loadu_si256(xPtr);
xPtr ++;
lutVal1 = _mm256_loadu_si256(lutPtr);
lutPtr ++;
lutVal2 = _mm256_loadu_si256(lutPtr);
lutPtr ++;
SAVE_OUTPUT8(0);
SAVE_OUTPUT8(1);
SAVE_OUTPUT8(2);
SAVE_OUTPUT8(3);
SAVE_OUTPUT8(4);
SAVE_OUTPUT8(5);
SAVE_OUTPUT8(6);
SAVE_OUTPUT8(7);
SAVE_OUTPUT8(8);
SAVE_OUTPUT8(9);
SAVE_OUTPUT8(10);
SAVE_OUTPUT8(11);
SAVE_OUTPUT8(12);
SAVE_OUTPUT8(13);
SAVE_OUTPUT8(14);
SAVE_OUTPUT8(15);
SAVE_OUTPUT8_2(0);
SAVE_OUTPUT8_2(1);
SAVE_OUTPUT8_2(2);
SAVE_OUTPUT8_2(3);
SAVE_OUTPUT8_2(4);
SAVE_OUTPUT8_2(5);
SAVE_OUTPUT8_2(6);
SAVE_OUTPUT8_2(7);
SAVE_OUTPUT8_2(8);
SAVE_OUTPUT8_2(9);
SAVE_OUTPUT8_2(10);
SAVE_OUTPUT8_2(11);
SAVE_OUTPUT8_2(12);
SAVE_OUTPUT8_2(13);
SAVE_OUTPUT8_2(14);
SAVE_OUTPUT8_2(15);
}
for (int i=32*(in_len/32);i<in_len;i++) {
output[deinter[i%out_len]] += input[i];
}
} else {
printf("wraps not implemented!\n");
#ifdef kk
int intCnt = 32;
int inputCnt = 0;
int nwrapps = 0;
while(inputCnt < in_len - 32) {
xVal = _mm256_loadu_si256(xPtr);
xPtr ++;
lutVal1 = _mm256_loadu_si256(lutPtr);
lutPtr ++;
lutVal2 = _mm256_loadu_si256(lutPtr);
lutPtr ++;
SAVE_OUTPUT8(0);
SAVE_OUTPUT8(1);
SAVE_OUTPUT8(2);
SAVE_OUTPUT8(3);
SAVE_OUTPUT8(4);
SAVE_OUTPUT8(5);
SAVE_OUTPUT8(6);
SAVE_OUTPUT8(7);
SAVE_OUTPUT8(8);
SAVE_OUTPUT8(9);
SAVE_OUTPUT8(10);
SAVE_OUTPUT8(11);
SAVE_OUTPUT8(12);
SAVE_OUTPUT8(13);
SAVE_OUTPUT8(14);
SAVE_OUTPUT8(15);
SAVE_OUTPUT8_2(0);
SAVE_OUTPUT8_2(1);
SAVE_OUTPUT8_2(2);
SAVE_OUTPUT8_2(3);
SAVE_OUTPUT8_2(4);
SAVE_OUTPUT8_2(5);
SAVE_OUTPUT8_2(6);
SAVE_OUTPUT8_2(7);
SAVE_OUTPUT8_2(8);
SAVE_OUTPUT8_2(9);
SAVE_OUTPUT8_2(10);
SAVE_OUTPUT8_2(11);
SAVE_OUTPUT8_2(12);
SAVE_OUTPUT8_2(13);
SAVE_OUTPUT8_2(14);
SAVE_OUTPUT8_2(15);
intCnt += 32;
inputCnt += 32;
if (intCnt >= out_len && inputCnt < in_len - 32) {
printf("warning rate matching wrapping remainder %d\n", out_len%32);
/* Copy last elements */
for (int j=(nwrapps+1)*out_len-(out_len%32) ;j<(nwrapps+1)*out_len;j++) {
output[deinter[j%out_len]] += input[j];
inputCnt++;
}
/* And wrap pointers */
nwrapps++;
intCnt = 32;
xPtr = (const __m256i*) &input[nwrapps*out_len];
lutPtr = (const __m256i*) deinter;
}
}
for (int i=inputCnt;i<in_len;i++) {
output[deinter[i%out_len]] += input[i];
}
#endif
}
return 0;
} else {
printf("Invalid inputs rv_idx=%d, cb_idx=%d\n", rv_idx, cb_idx);
return SRSLTE_ERROR_INVALID_INPUTS;
}
}
int srslte_rm_turbo_rx_lut_avx(int16_t *input, int16_t *output, uint16_t *deinter, uint32_t in_len, uint32_t cb_idx, uint32_t rv_idx)
{
if (rv_idx < 4 && cb_idx < SRSLTE_NOF_TC_CB_SIZES) {
uint32_t out_len = 3*srslte_cbsegm_cbsize(cb_idx)+12;
const __m256i* xPtr = (const __m256i*) input;
const __m256i* lutPtr = (const __m256i*) deinter;
__m256i xVal, lutVal;
int16_t x;
uint16_t l;
/* Simplify load if we do not need to wrap (ie high rates) */
if (in_len <= out_len) {
for (int i=0;i<in_len/16;i++) {
xVal = _mm256_loadu_si256(xPtr);
lutVal = _mm256_loadu_si256(lutPtr);
SAVE_OUTPUT(0);
SAVE_OUTPUT(1);
SAVE_OUTPUT(2);
SAVE_OUTPUT(3);
SAVE_OUTPUT(4);
SAVE_OUTPUT(5);
SAVE_OUTPUT(6);
SAVE_OUTPUT(7);
SAVE_OUTPUT(8);
SAVE_OUTPUT(9);
SAVE_OUTPUT(10);
SAVE_OUTPUT(11);
SAVE_OUTPUT(12);
SAVE_OUTPUT(13);
SAVE_OUTPUT(14);
SAVE_OUTPUT(15);
xPtr ++;
lutPtr ++;
}
for (int i=16*(in_len/16);i<in_len;i++) {
output[deinter[i%out_len]] += input[i];
}
} else {
int intCnt = 16;
int inputCnt = 0;
int nwrapps = 0;
while(inputCnt < in_len - 16) {
xVal = _mm256_loadu_si256(xPtr);
lutVal = _mm256_loadu_si256(lutPtr);
SAVE_OUTPUT(0);
SAVE_OUTPUT(1);
SAVE_OUTPUT(2);
SAVE_OUTPUT(3);
SAVE_OUTPUT(4);
SAVE_OUTPUT(5);
SAVE_OUTPUT(6);
SAVE_OUTPUT(7);
SAVE_OUTPUT(8);
SAVE_OUTPUT(9);
SAVE_OUTPUT(10);
SAVE_OUTPUT(11);
SAVE_OUTPUT(12);
SAVE_OUTPUT(13);
SAVE_OUTPUT(14);
SAVE_OUTPUT(15);
xPtr++;
lutPtr++;
intCnt += 16;
inputCnt += 16;
if (intCnt >= out_len && inputCnt < in_len - 16) {
/* Copy last elements */
if ((out_len%16) == 12) {
for (int j=(nwrapps+1)*out_len-12;j<(nwrapps+1)*out_len;j++) {
output[deinter[j%out_len]] += input[j];
inputCnt++;
}
} else {
for (int j=(nwrapps+1)*out_len-4;j<(nwrapps+1)*out_len;j++) {
output[deinter[j%out_len]] += input[j];
inputCnt++;
}
}
/* And wrap pointers */
nwrapps++;
intCnt = 16;
xPtr = (const __m256i*) &input[nwrapps*out_len];
lutPtr = (const __m256i*) deinter;
}
}
for (int i=inputCnt;i<in_len;i++) {
output[deinter[i%out_len]] += input[i];
}
}
return 0;
} else {
printf("Invalid inputs rv_idx=%d, cb_idx=%d\n", rv_idx, cb_idx);
return SRSLTE_ERROR_INVALID_INPUTS;
}
}
int srslte_rm_turbo_rx_lut_sse_8bit(int8_t *input, int8_t *output, uint16_t *deinter, uint32_t in_len, uint32_t cb_idx, uint32_t rv_idx)
{
if (rv_idx < 4 && cb_idx < SRSLTE_NOF_TC_CB_SIZES) {
uint32_t out_len = 3*srslte_cbsegm_cbsize(cb_idx)+12;
const __m128i* xPtr = (const __m128i*) input;
const __m128i* lutPtr = (const __m128i*) deinter;
__m128i xVal, lutVal1, lutVal2;
/* Simplify load if we do not need to wrap (ie high rates) */
if (in_len <= out_len) {
for (int i=0;i<in_len/16;i++) {
xVal = _mm_loadu_si128(xPtr);
xPtr ++;
lutVal1 = _mm_loadu_si128(lutPtr);
lutPtr++;
lutVal2 = _mm_loadu_si128(lutPtr);
lutPtr ++;
for (int j=0;j<8;j++) {
int8_t x = (int8_t) _mm_extract_epi8(xVal, j);
uint16_t l = (uint16_t) _mm_extract_epi16(lutVal1, j);
output[l] += x;
}
for (int j=0;j<8;j++) {
int8_t x = (int8_t) _mm_extract_epi8(xVal, j+8);
uint16_t l = (uint16_t) _mm_extract_epi16(lutVal2, j);
output[l] += x;
}
}
for (int i=16*(in_len/16);i<in_len;i++) {
output[deinter[i%out_len]] += input[i];
}
} else {
int intCnt = 16;
int inputCnt = 0;
int nwrapps = 0;
while(inputCnt < in_len - 16) {
xVal = _mm_loadu_si128(xPtr);
xPtr ++;
lutVal1 = _mm_loadu_si128(lutPtr);
lutPtr++;
lutVal2 = _mm_loadu_si128(lutPtr);
lutPtr ++;
for (int j=0;j<8;j++) {
int8_t x = (int8_t) _mm_extract_epi8(xVal, j);
uint16_t l = (uint16_t) _mm_extract_epi16(lutVal1, j);
output[l] += x;
}
for (int j=0;j<8;j++) {
int8_t x = (int8_t) _mm_extract_epi8(xVal, j+8);
uint16_t l = (uint16_t) _mm_extract_epi16(lutVal2, j);
output[l] += x;
}
intCnt += 16;
inputCnt += 16;
if (intCnt >= out_len && inputCnt < in_len - 16) {
/* Copy last elements */
if ((out_len%16) == 12) {
for (int j=(nwrapps+1)*out_len-12;j<(nwrapps+1)*out_len;j++) {
output[deinter[j%out_len]] += input[j];
inputCnt++;
}
} else {
for (int j=(nwrapps+1)*out_len-4;j<(nwrapps+1)*out_len;j++) {
output[deinter[j%out_len]] += input[j];
inputCnt++;
}
}
/* And wrap pointers */
nwrapps++;
intCnt = 16;
xPtr = (const __m128i*) &input[nwrapps*out_len];
lutPtr = (const __m128i*) deinter;
}
}
for (int i=inputCnt;i<in_len;i++) {
output[deinter[i%out_len]] += input[i];
}
}
return 0;
} else {
printf("Invalid inputs rv_idx=%d, cb_idx=%d\n", rv_idx, cb_idx);
return SRSLTE_ERROR_INVALID_INPUTS;
}
}
int main(int argc, char **argv) {
int i;
uint8_t *rm_bits, *rm_bits2, *rm_bits2_bytes;
short *rm_bits_s;
float *rm_bits_f;
parse_args(argc, argv);
srslte_rm_turbo_gentables();
rm_bits_s = srslte_vec_malloc(sizeof(short) * nof_e_bits);
if (!rm_bits_s) {
perror("malloc");
exit(-1);
}
rm_bits_f = srslte_vec_malloc(sizeof(float) * nof_e_bits);
if (!rm_bits_f) {
perror("malloc");
exit(-1);
}
rm_bits = srslte_vec_malloc(sizeof(uint8_t) * nof_e_bits);
if (!rm_bits) {
perror("malloc");
exit(-1);
}
rm_bits2 = malloc(sizeof(uint8_t) * nof_e_bits);
if (!rm_bits2) {
perror("malloc");
exit(-1);
}
rm_bits2_bytes = malloc(sizeof(uint8_t) * nof_e_bits/8 + 1);
if (!rm_bits2_bytes) {
perror("malloc");
exit(-1);
}
uint32_t st=0, end=188;
if (cb_idx != -1) {
st=cb_idx;
end=cb_idx+1;
}
uint32_t rv_st=0, rv_end=4;
if (rv_idx != -1) {
rv_st=rv_idx;
rv_end=rv_idx+1;
}
for (cb_idx=st;cb_idx<end;cb_idx++) {
for (rv_idx=rv_st;rv_idx<rv_end;rv_idx++) {
uint32_t long_cb_enc = 3*srslte_cbsegm_cbsize(cb_idx)+12;
printf("checking cb_idx=%3d rv_idx=%d...", cb_idx, rv_idx);
for (i = 0; i < long_cb_enc; i++) {
bits[i] = rand() % 2;
}
bzero(buff_b, BUFFSZ * sizeof(uint8_t));
srslte_rm_turbo_tx(buff_b, BUFFSZ, bits, long_cb_enc, rm_bits, nof_e_bits, 0);
if (rv_idx > 0) {
srslte_rm_turbo_tx(buff_b, BUFFSZ, bits, long_cb_enc, rm_bits, nof_e_bits, rv_idx);
}
for (int i=0;i<long_cb_enc/3;i++) {
systematic[i] = bits[3*i];
parity[i] = bits[3*i+1];
parity[i+long_cb_enc/3] = bits[3*i+2];
}
srslte_bit_pack_vector(systematic, systematic_bytes, long_cb_enc/3);
srslte_bit_pack_vector(parity, parity_bytes, 2*long_cb_enc/3);
bzero(buff_b, BUFFSZ * sizeof(uint8_t));
bzero(rm_bits2_bytes, nof_e_bits/8);
srslte_rm_turbo_tx_lut(buff_b, systematic_bytes, parity_bytes, rm_bits2_bytes, cb_idx, nof_e_bits, 0, 0);
if (rv_idx > 0) {
bzero(rm_bits2_bytes, nof_e_bits/8);
srslte_rm_turbo_tx_lut(buff_b, systematic_bytes, parity_bytes, rm_bits2_bytes, cb_idx, nof_e_bits, 0, rv_idx);
}
srslte_bit_unpack_vector(rm_bits2_bytes, rm_bits2, nof_e_bits);
for (int i=0;i<nof_e_bits;i++) {
if (rm_bits2[i] != rm_bits[i]) {
printf("Error in TX bit %d\n", i);
exit(-1);
}
}
printf("OK TX...");
for (int i=0;i<nof_e_bits;i++) {
rm_bits_f[i] = rand()%10-5;
rm_bits_s[i] = (short) rm_bits_f[i];
}
bzero(buff_f, BUFFSZ*sizeof(float));
srslte_rm_turbo_rx(buff_f, BUFFSZ, rm_bits_f, nof_e_bits, bits_f, long_cb_enc, rv_idx, 0);
bzero(bits2_s, long_cb_enc*sizeof(short));
srslte_rm_turbo_rx_lut(rm_bits_s, bits2_s, nof_e_bits, cb_idx, rv_idx);
for (int i=0;i<long_cb_enc;i++) {
if (bits_f[i] != bits2_s[i]) {
printf("error RX in bit %d %f!=%d\n", i, bits_f[i], bits2_s[i]);
exit(-1);
}
}
printf("OK RX\n");
}
}
free(rm_bits);
free(rm_bits2);
free(rm_bits2_bytes);
exit(0);
}
int main(int argc, char **argv) {
uint32_t frame_cnt;
float *llr;
short *llr_s;
uint8_t *llr_c;
uint8_t *data_tx, *data_rx, *data_rx_bytes, *symbols;
uint32_t i, j;
float var[SNR_POINTS];
uint32_t snr_points;
uint32_t errors;
uint32_t coded_length;
struct timeval tdata[3];
float mean_usec;
srslte_tdec_t tdec;
srslte_tcod_t tcod;
parse_args(argc, argv);
if (!seed) {
seed = time(NULL);
}
srand(seed);
if (test_known_data) {
frame_length = KNOWN_DATA_LEN;
} else {
frame_length = srslte_cbsegm_cbsize(srslte_cbsegm_cbindex(frame_length));
}
coded_length = 3 * (frame_length) + SRSLTE_TCOD_TOTALTAIL;
printf(" Frame length: %d\n", frame_length);
if (ebno_db < 100.0) {
printf(" EbNo: %.2f\n", ebno_db);
}
data_tx = srslte_vec_malloc(frame_length * sizeof(uint8_t));
if (!data_tx) {
perror("malloc");
exit(-1);
}
data_rx = srslte_vec_malloc(frame_length * sizeof(uint8_t));
if (!data_rx) {
perror("malloc");
exit(-1);
}
data_rx_bytes = srslte_vec_malloc(frame_length * sizeof(uint8_t));
if (!data_rx_bytes) {
perror("malloc");
exit(-1);
}
symbols = srslte_vec_malloc(coded_length * sizeof(uint8_t));
if (!symbols) {
perror("malloc");
exit(-1);
}
llr = srslte_vec_malloc(coded_length * sizeof(float));
if (!llr) {
perror("malloc");
exit(-1);
}
llr_s = srslte_vec_malloc(coded_length * sizeof(short));
if (!llr_s) {
perror("malloc");
exit(-1);
}
llr_c = srslte_vec_malloc(coded_length * sizeof(uint8_t));
if (!llr_c) {
perror("malloc");
exit(-1);
}
if (srslte_tcod_init(&tcod, frame_length)) {
fprintf(stderr, "Error initiating Turbo coder\n");
exit(-1);
}
if (srslte_tdec_init_manual(&tdec, frame_length, tdec_type)) {
fprintf(stderr, "Error initiating Turbo decoder\n");
exit(-1);
}
srslte_tdec_force_not_sb(&tdec);
float ebno_inc, esno_db;
ebno_inc = (SNR_MAX - SNR_MIN) / SNR_POINTS;
if (ebno_db == 100.0) {
snr_points = SNR_POINTS;
for (i = 0; i < snr_points; i++) {
ebno_db = SNR_MIN + i * ebno_inc;
esno_db = ebno_db + 10 * log10((double) 1 / 3);
var[i] = sqrt(1 / (pow(10, esno_db / 10)));
}
} else {
esno_db = ebno_db + 10 * log10((double) 1 / 3);
var[0] = sqrt(1 / (pow(10, esno_db / 10)));
snr_points = 1;
}
for (i = 0; i < snr_points; i++) {
mean_usec = 0;
errors = 0;
frame_cnt = 0;
while (frame_cnt < nof_frames) {
/* generate data_tx */
for (j = 0; j < frame_length; j++) {
if (test_known_data) {
data_tx[j] = known_data[j];
} else {
data_tx[j] = rand() % 2;
}
}
/* coded BER */
if (test_known_data) {
for (j = 0; j < coded_length; j++) {
symbols[j] = known_data_encoded[j];
}
} else {
srslte_tcod_encode(&tcod, data_tx, symbols, frame_length);
}
for (j = 0; j < coded_length; j++) {
llr[j] = symbols[j] ? 1 : -1;
}
srslte_ch_awgn_f(llr, llr, var[i], coded_length);
for (j=0;j<coded_length;j++) {
llr_s[j] = (int16_t) (100*llr[j]);
}
/* decoder */
srslte_tdec_new_cb(&tdec, frame_length);
uint32_t t;
if (nof_iterations == -1) {
t = MAX_ITERATIONS;
} else {
t = nof_iterations;
}
gettimeofday(&tdata[1], NULL);
for (int k=0;k<nof_repetitions;k++) {
srslte_tdec_run_all(&tdec, llr_s, data_rx_bytes, t, frame_length);
}
gettimeofday(&tdata[2], NULL);
get_time_interval(tdata);
mean_usec = (tdata[0].tv_sec*1e6+tdata[0].tv_usec)/nof_repetitions;
frame_cnt++;
uint32_t errors_this = 0;
srslte_bit_unpack_vector(data_rx_bytes, data_rx, frame_length);
errors_this=srslte_bit_diff(data_tx, data_rx, frame_length);
//printf("error[%d]=%d\n", cb, errors_this);
errors += errors_this;
printf("Eb/No: %2.2f %10d/%d ", SNR_MIN + i * ebno_inc, frame_cnt, nof_frames);
printf("BER: %.2e ", (float) errors / (nof_cb*frame_cnt * frame_length));
printf("%3.1f Mbps (%6.2f usec)", (float) (nof_cb*frame_length) / mean_usec, mean_usec);
printf("\r");
}
printf("\n");
}
printf("\n");
if (snr_points == 1) {
if (errors) {
printf("%d Errors\n", errors/nof_cb);
}
}
if (data_rx_bytes) {
free(data_rx_bytes);
}
free(data_tx);
free(symbols);
free(llr);
free(llr_c);
free(llr_s);
free(data_rx);
srslte_tdec_free(&tdec);
srslte_tcod_free(&tcod);
printf("\n");
printf("Done\n");
exit(0);
}
