How-to for running code in MPARM environment:

One-to-one mapping:
- Generate the code: ant -f runexample -Dnumber=? mparm
- Copy the generated dir 'systemc' to <MPARM>/MPARM/apps
- Copy the library 'queue_lib' into dir 'systemc'
- Compile the code: make
- Link the binary: ln -sf o-optimize/app.exe TargetMem_<?>.mem
    ? depends on how many processors
- Run: $SWARMDIR/bin/mpsim.x -c <?> --intc=i  -C -S -D
    ? is number of processors


Multi-to-one mapping:
- Modify main.c
    - number_of_processes: each field is number of processes per processor
    - xxx_PROCESSOR: set to a same number if mapping to common processor
    - processor_init(): leave  one for each processor (not necessary)
- system.c
    - macro CONFIGURE_MAXIMUM_TASKS should be two times larger than the 
      maximum number of processes mapped to one processor.
- MAXQUEUE
    - where should we put this macro ?
- If using the dol mapping specification, the processor 1 should be reserved
  in the case of enabling macro QUEUE_BUFF_SHAPER !

MPARM queue_lib spec:
- Always use interrupt
- Always use memcpy
- DMA always enables, the switch is in the Makefile
- Token size is always 32 bits
- Default Queue size is 4, can be redefined


Segment wide Calibration:
- Modify Makefile:
    - enable macro PERFORMANCE_EXTRACT (default disable)
    - uncomment line: 
      CXXSRCS += lib/xmlParser.cpp lib/Performance_Extraction.cpp
- main.c: Map all processes into one processor
- Run: $SWARMDIR/bin/mpsim.x -c 1 --intc=i  -C -S -D  


Communication exploration:
- Modify Makefile:
  - enable one of below macro exclusively:
    - QUEUE_BUFF_IN_PRODUCER (default)
    - QUEUE_BUFF_IN_PRODUCER_DMA
    - QUEUE_BUFF_IN_CONSUMER
    - QUEUE_BUFF_IN_CONSUMER_DMA
    - QUEUE_BUFF_SHAPER

Running mpeg2 decoder on MPARM:
- copy dol.h from other case studies
- add macro _DOL_ETHZ_GEN_ to Makefile
- change queue size to 4096 in system.h
- call mpsim.x with option --s-size=19 (scratchpad size=1Mbyts)

Running mjpeg-2000 decoder on MPARM:
- jpeg.h
    - disable macro VIEWER
    - disable macros VERBOSE and INFO
- scratch_queue.h
    - macro MAXQUEUE > 24, if mapping all to one processor
- system.h
    - macro CONFIGURE_MAXIMUM_TASKS > 9


Tricks:
- CONFIGURE_MAXIMUM_TASKS in system.h
- MAXQUEUE in scratch_queue.h 
- Token size should be multiple of 4 bytes
- W/R operations should be the same size, otherwise token size should be
  gcd(Write_token, Read_token).

BUGS & TODO:
- free() in wrapper not correct
- Iterated port not tested yet
- Shaper not finished!
- Timeslice works?  (Yes)