STEPS 5 + 6 - TOKEN ROM + MUX + RAM-E + ROM-P' + ADDER + WRITE-BACK
=====================================================================
Function: add_token_and_positions  (8,128) -> (9,128)   = input of Encoder 0

Guide architecture (page 2, lower half)
   multiplier out --\
                     MUX --> RAM-E (9 x 128) --> (+) --> out (to Encoder 0)
   TOKEN ROM ------/           ^                ^   |
                               |            ROM-P'  |
                               +--- write back -----+

   Phase 1 FILL    (1152 clocks): MUX writes RAM-E
                   row 0 <- Token ROM (1 x 128),  rows 1..8 <- multiplier (Step 4)
   Phase 2 ADD_POS (1152 clocks): RAM-E[t][j] + ROM-P'[t][j] -> write back + out

   ROM-P' (9 x 128): row 0 = P[0];  rows 1..8 = P[t] + b'
   (b' = patch-embedding bias + the removed mean/std correction, see Step 1 rom_data)

Number format: ap_fixed<25,8> for token, ROM-P', RAM-E and output (|v| < 4)

Folder (own folder next to S1_input_rom_raw_src, S2_s2p, S4_patch_embed_src)
   S5_token_pos_src\
      hls_config.cfg                  top = token_pos
      src\token_pos.h / .cpp          Token ROM, ROM-P', RAM-E (2-port), MUX, adder
      tb\token_pos_tb.cpp             feeds Step 4 HW output, checks 1152 values
      data\token_rom.dat              class token (1 x 128)
      data\rom_p.dat                  ROM-P' (9 x 128)
      data\emb_in.dat                 Step 4 hardware output E' (8 x 128)
      data\token_position_ref.dat     Token_position_output_calculated.csv (9 x 128)

Check
   RAM-E (after add) vs Token_position_output_calculated.csv   expect ~1.9e-05 (tol 1e-04)  [3-encoder ROMs: predicted 1.877e-05]
   This is the FULL chain error: raw input -> ROM -> S2P -> ROM-W' x -> + token/ROM-P'

How to run
   1. Extract to  ...\MRT_Blocks\MRT_Blocks\S5_token_pos_src\   (no double folder)
   2. Vitis: New Component -> HLS
        Name: s5_token_pos   Location: ...\MRT_Blocks\MRT_Blocks  (no workspace switch)
        Config: Existing -> S5_token_pos_src\hls_config.cfg
        Top: token_pos       Part: xc7s50csga324-1  -> Finish
   3. C SIMULATION        expect "*** STEP 5+6 TOKEN + POSITION TEST PASSED ***"
   4. C SYNTHESIS         expect FILL and ADD_POS: II = 1, trip count 1152 each,
                          total ~2310 clocks, DSP 0, a few BRAM (RAM-E, ROM-P', Token ROM)
   5. C/RTL COSIMULATION  expect PASS

Waveform (xsim --gui in ...\s5_token_pos\token_pos\hls\sim\verilog, Tcl console):
   set g [add_wave_group "S56_per_clock"]; set p /apatb_token_pos_top/AESL_inst_token_pos; add_wave -into $g $p/ap_clk; add_wave -into $g $p/ap_start; add_wave -into $g $p/emb_in_read; add_wave -into $g $p/tok_out_write; add_wave -into $g -radix hex $p/tok_out_din; add_wave -into $g $p/ap_done
   expect: first 128 clocks no emb_in_read (token row from Token ROM, MUX = 0),
           then emb_in_read = 1 for 1024 clocks (MUX = 1),
           then tok_out_write = 1 for 1152 clocks (adder + write back).
