Below is the output of btoblas --help
For a simple test of the system, try the following:
--level1 thread2:12:2

Immutable default args:
--delete_tmp

-----
$ ./btoblas --help

USAGE: btoblas matlab_input_file.m [options]

Options:
  -a [ --precision ] arg (=double)      Set precision type for generated output
                                        [float|double] double is default
  -e [ --empirical_off ]                Disable empirical testing.  Empirical 
                                        testing is enabled by default.
  -c [ --correctness ]                  Enable correctness testing.  
                                        Correctness testing
                                        is disabled by default.  Enabling 
                                        requires a BLAS library
                                        to be present.  (Set path in top level 
                                        make.inc. See 
                                        documentation for further details
  -m [ --use_model ]                    Enable the analytic model when testing 
                                        kernels. If set the compiler will use a
                                        memory model to help with optimization 
                                        choices.  Not recommended for most 
                                        users. 
  -t [ --threshold ] arg (=0.01)        This parameter controls how much 
                                        empirical testing is performed. For 
                                        example, the default of 0.01 says that 
                                        empirical testing will be performed on 
                                        any version that is predicted to be 
                                        within %1 the performance of the best 
                                        predicted version.  A value of 1 will 
                                        rank all versions. A value of 0 will 
                                        select only the best version. Only used
                                        if use_model is on.
  --level1 arg                          Choose thread or cache tiling for outer
                                        level, and input parameter search 
                                        range. Example "thread 2:12:2" or 
                                        "cache 64:512:8".
  --level2 arg                          Choose thread or cache tiling for inner
                                        level, and input parameter search 
                                        range. Example "thread 2:12:2" or 
                                        "cache 64:512:8".
  -r [ --test_param ] arg (=3000:3000:1)
                                        Set parameters for empirical and 
                                        correctness tests as
                                        start:stop:step
  -s [ --search ] arg (=ga)             select the search strategy:
                                          [ga|ex|random|orthogonal|debug|thread
                                        ] 
                                            ga is the default 
                                         ga: genetic algorithm 
                                          orthogonal: search one parameter 
                                                      completely first, then 
                                                      search next.  For example
                                                      find best fusion, and 
                                                      then find best partition 
                                                      strategy
                                          random: randomly search the space
                                          exhaustive: exhaustively search the 
                                                      space (very long run 
                                                      times)
                                        thread: thread search only 
                                        debug: use the versions specified in 
                                        debug.txt 
                                        
  --ga_timelimit arg (=10)              Specifies the number of GA versions to 
                                        be tested
  --empirical_reps arg (=5)             the number of empirical test 
                                        repetitions for each point
  --delete_tmp                          delete *_tmp directory without 
                                        prompting
  --ga_popsize arg (=10)                population size
  --ga_nomaxfuse                        genetic algorithm without max fuse
  --ga_noglobalthread                   genetic algorithm without global thread
                                        search
  --ga_exthread                         genetic algorithm with additional 
                                        exhaustive thread search
-----
Other args not mentioned in r911 help text:

    -p [ --partition_off ]              Disable partitioning
    -b [ --backend ] (=ptr)             [noptr] for VLA C code (not made available
                                            in Lighthouse).