I may be willing to sponsor the following projects depending on cost. I also have some proprietary work that would require expert ability to convert Lua tables into optimized memory structures in C for a ML (machine learning) project I am working on.
* Auto trigger the GC before panic exit with out of memory.
o I do this manually and it keeps our long running analyzers from
aborting but it seems like this could be inserted into the
existing allocator right before it panics and exits.
function gc_check_collect_if_needed(max_mem_meg)
if max_mem = nil
max_mem = 1200
end
local meg_used = collectgarbage("count") / 1024
if meg_used > 1500 then
local beg_time = curr_sec()
collectgarbage()
elap(beg_time, "collected garbage before=" ..
math.round(meg_used,2) .. "mb" .. " after=" ..
math.round(collectgarbage("count") / 1024,2) .. "mb elap=")
end
end
* Implement operator to initialize high density array of type with
fact access similar to C eg: InitArray(Double, 30000) - I am willing
to accept non graceful abort if we exceed the space. I am looking
for the following for loop to be able to run extremely fast with
simple pointer arithmetic as close to native c speed as we can get
for simple array of same type. Or failing that have the memory in a
structure so that once I populated it using Lua code that I could
call a C function that could iterate across it with simple pointer
increment.
local maxele = 900000
local arr = InitArray(Double, maxele)
local sum = 0.0
for ndx = 1, maxele do
sum = arr[ndx]
end
I saw a sample of this kind of init in
http://math.nist.gov/scimark2/ array_init() but it didn't seem
to pre-allocate the desired number of elements so at the very
least it is paying extra overhead to grow the array. I know
ahead of time how many elements we will need so I don't want to
pay the realloc overhead.
* Implement enough of the V3 GC to allow use of full 64 bit Memory space
* Implement a += and -= operator. I would almost use them for
integer increments similar to ++ and -- in C.
* produce make script that includes pre-integrated with sockets for at
least the socket client to make HTTP Get calls.
* Produce a method to compile a Lua script with all dependencies after
JIT reduction into single EXE file that can run across all common
windows boxes. Willing to take overhead of multiple copies
optimized for different CPU inside the Exe.
Thanks Joe Ellsworth
CTO & Chief Algorithms scientist at Bayes Analytic LLC
joee@xxxxxxxxxxxxxxxxx
206-601-2985
