Bug hardware implementation of Connex
Bugs manifested on zedboard (Archlinux):
* readDataFromArray() blocks - a simple test that was doing writeData & readData: /home/alarm/OpincaaLLVM/opincaa_standalone_app/25_map/output.cpp - see also standard-output file: STDout_008_8Kelems_NOT_OK_blocks_at_readFrom (NOTE: normally the behavior is the one documented, although once it worked...) But for a number of 1K NUM_ELEMS, writeData() and readData() it functions well.
* on a freshly rebooted zedboard I ran the code 500_MatTranspose and obtained matrix At zero (although I had to obtain main diagonal nonnull; look for "At ="):
* /home/alarm/OpincaaLLVM/opincaa_standalone_app/500_MatTranspose/SIZE_128/STDout_013_1_diag_WHY-NULL-At-BUG, or STDout_034_1diag_only_2nd_run_gives_us_LS_data_nonnull (si STDout_033_1diag_fresh_boot, STDout_034_1diag_only_2nd_run_g[output.cpp](/uploads/fc0b2c5701b7a9b5f288b7159ac5e280/output.cpp)[output.cpp](/uploads/349b592991af2b4ef5e9d155771c579e/output.cpp)ives_us_LS_data_nonnull)
* it's probably an error of the kernel (probably error at read and write instructions, CELL_SH* with 128 or 127 steps to which we add 128+ NOPs to take correct result frin shiftreg) (si probabil nu este o eroare de I/O)
Note: code 500_MatTranspose was tested on simulator (OK, simulator had a small bug at CELL_SH* that I fixed, but I suppose it is correct now).
Alex