Merge remote-tracking branch 'origin/resource_mapping_optim' into integration_2025_w07 (!3127)
Optimizations of PDSCH Resource Mapping in nr_dlsch.c/nr_modulation.c These changes add SIMD optimizations for Neon/AVX2/AVX512 in the PDSCH transmit path. The timing improvements are listed here based on the nr_dlsim -e25 -R273 -b273 -s30 -x "layers" -y 4 -z 4 -P benchmark with "layers" 2,3,4 and comparing "PHY proc tx": 273 PRBS, mcs25, 64QAM peafowl (gcc11,AMD EPYC 9374F) 2-layer, 4 TX : 431 us (develop 565 us) 3-layer, 4 TX : 692 us (develop 849 us) 4-layer, 4 TX : 963 us (develop 1172 us) stupix (gcc10, Xeon Gold 6354) 2-layer, 4 TX : 568 us (develop 652 us) 3-layer, 4 TX : 901 us (develop 1030 us) 4-layer, 4 TX : 1250 us (develop 1396 us) matix (gcc14, Ryzen 9 PRO 7945) 2-layer, 4 TX : 317 us (develop 505 us) 3-layer, 4 TX : 538 us (develop 779 us) 4-layer, 4 TX : 767 us (develop 1233 us)
Showing
This diff is collapsed.
This diff is collapsed.
Please register or sign in to comment