* tiny bug: initial value was considered a valid measurement
* led to slow convergence, if actual measurements are large
* also fixed a bug concerning the minimal/maximal elements

* measures how many jobs with minimal runtime (1cc) can be scheduled per
  second, using a fixed number of threads
* benchmark will record at least 1-8 threads, more while performance
  increases
* also improved convergence and reduced update speed for the other
  measurements

* done: arrayinit, arraysum, basic_test, warraw
* others still missing
* need to unify the tests, basic_test is a good start
* should rename warraw to arrayupdate and remove warraw refs

* now using host name as platform, if unknown

* generate_boot_image.sh requires root rights at many different points
* in it possible that the sudo timeout hits in between, leading to
  aborted execution if the user fails to re-enter the password
* fixed that, asking once, re-using the password
* also fixed and improved several other things:
* fixed zedboard and zc706 devicetrees (needed patching to activate IRQs)
* same with PyNQ, but different approach: have fixed devicetree in pynq
* fixed U-Boot output log redirection

* additional size parameter allows to control size of output image
* if size is larger than 4GiB, the full linux-xlnx build tree will be
  added to the rootfs at /linux-xlnx (1.1GiB)
* modified rc.local script changes owner recursively to 'xilinx'
* fixed some more bugs

* instead of directly formatting and partitioning an SDcard,
  generate_boot_image now prepares a disk image file
* setup via losetup, formatted and mounted via kpartx
* much faster than previous method, including faster writing to SD
* major stylistic improvements, clearer style
* fixed bug in parallel executions

* extended the zedboard version
* fixed devicetrees, now working out-of-the-box
* automated extraction of root FS
* patching root FS for network stuff, removing Jupyter, but keeping the
  autoresizing, which is pretty neat
* PyNQ is non-trivial to rebuild; decided to extract directly from the
  image
* PyNQ image is downloaded from public URL at Digilent (minimal
  footprint in git)
* not enough space to put linux-xlnx on board; maybe increase image
  size? or provide script to automate the process

* it is an incredibly annoying process I'd really like to automate
* first draft seems to work ok, but there is a problem with the devicetrees
* need to continue work next week

* Compose started an HLS task for every PE in a Composition
* problem, if Core does not have a Kernel (imported IP-XACT)
* wanted to fix that in any case, but until now it was just a
  convenience/efficiency thing, now it was a bug

* apparently the interrupt is now shared (SPI) in recent kernels

* latency data now contains min, max and average
* new Json format implemented in scala
* fixed several minor issues in case of errors

* added type clk to the bd pin and a frequency of 125 MHz
* added some debug output
* fixed PS parameters (from PyNQ example design)

* changed to better name reflecting composition, target and freq

* updated Doxygen configs
* added two READMEs for the test builds

* fixed occurrences, renamed the member

* fixes support for 2016.4 and 2017.1
* improved modularity by implementing shared platform::generate
* checking results of runs in Vivado, adding list of logs to check in
  case of error
* cleanup

* compare support was completely broken, a wonder that the code ever
  seemed to work at all
* comparison of the actual Composition was missing; comparing
  Compositions is a bit tricky due to many isomorphisms in the
  definition
* opted to compare area utilization instead: Unlikely that two different
  compositions in the same DSE run will ever have _exactly_ the same
  number of LUTs, FFS, etc.
* should work now, comparison order: area, h-value, frequency

* new object JobExamples contains examples and dump code
* can be used to re-generate directly from the defs
* also added a README.md to mark optional/mandatory params

* previous RE did include many other runme.logs, fixed

* added only synth_1/runme.log, impl_1/runme.log
* tracking all runme.logs just scrambled the output and generated a lot
  of CPU load
* need better way to distinguish the sources; colors?

* missing report now correctly triggers re-evaluation of the core
* moved core.description file into 'ipcore' subdirectory
* fixed problem with relative paths for import file

* was in fact not implemented; fixed