On 05/20/2016 03:45 PM, Walter Bright wrote:
On 5/20/2016 5:57 AM, ZombineDev wrote:
Walter's PR slows down the compilation with
25-40% according to my tests. I expect that compilation would be
faster if the
whole process is skipped altogether.
The compressor only kicks in if the string is longer than 64 bytes. I
set it pretty low in order to have it kick in often, and hopefully flush
out any bugs in it.
For a production version, the minimum size should be set substantially
larger, based on testing to provide a reasonable balance.
That should speed it up a lot. Also, the algorithm is a bit naively
implemented, I bet it could be speeded up substantially.
Hashing isn't algorithmically cheap, either.
From the measurements shown the work seems Pareto distributed - the
major time spender is the few long symbols, not the many short symbols.
There are a few long symbols that dominate everything else by an order
of magnitude. The one percenters! :o) -- Andrei
