[Razor12911]

for -l, lower is better, better ratio because more repetitions are more likely to be found, but too low will also increase size, just find a value which is reasonable to smallest size. Bare in mind, lower you go, means the higher ram is going to be used.
-a is just an accelerator, higher means more speed during compression but more memory is used.

But do read mate, took this from srep.exe itself

SREP 3.93a beta: huge-dictionary LZ77 preprocessor   (c) Bulat.Ziganshin********c
om
October 11, 2014    homepage: http://freearc.org/research/SREP39.aspx

Usage: SREP [options] infile [outfile]
   infile/outfile can be specified as "-" for stdin/stdout
   "SREP [options] somefile" compresses data from somefile to somefile.srep
   "SREP [options] somefile.srep" decompresses data back to somefile
   "SREP [options]" compresses and "SREP -d [options]" decompresses data from st
din to stdout

Options are:
   -m0: only in-memory compression (REP algorithm)
   -m1: fixed-window content-defined chunking with matches checked by VMAC
   -m2: order-1 content-defined chunking with matches checked by VMAC
   -m3: check matches by VMAC digest (compression memory = 7-8% of filesize)
   -m4: check matches by rereading old data (compression memory = 3-4% of filesi
ze)
   -m5/-mx: rereading with byte-accurate matches (compression memory = 7-9% of f
ilesize)
   -l: minimum LZ match length, default 0
   -c: size of hash chunk, by default as small as required to find all these LZ
matches
   -aX[/Y]: alloc X bytes of those Y bits will be set per L input bytes for comp
ression accelerator
            Y=0/1/2/4/8/16/32/64, -a0 is slowest but requires least memory
   -ia-: disable I/O acceleration to reduce memory usage (-m5* only)
   -tN: use N compression threads (only for -m1/-m2)
   -dBYTES: dictionary size for in-memory compression (REP algorithm), default 5
12mb
   -dhBYTES/-dcN/-dlN: size of hash / size of hash chunk / minimum match length
for in-memory compression

   -m1..-m5: index-LZ - list of matches saved at the end of compressed file
   -m1f..-m5f: future-LZ - decompression dictionary will hold only future matche
s
   -m1o..-m5o: I/O LZ - output file used as decompression dictionary
   -memBYTES: amount of RAM used by future-LZ/index-LZ decompression (extra goes
 into VM file)
      -mem75% AKA -mem75p means "use no more than 75% of RAM" - that's by defaul
t
      -mem600mb means itself
      -mem75%-600mb means "use no more than 75% of RAM minus 600 mb"
   -mBYTES: don't store matches larger than BYTES on future-LZ/index-LZ decompre
ssion

   -d: decompression (for -m0o..m5o requires only 24 mb of memory besides of OS
I/O buffers)
   -i: print info about compressed file: srep -i datafile.srep
   -delete: delete source file after successful (de)compression
   -sBYTES: explicitly specify filesize (for compression from stdin), default 25
gb
   -bBYTES: change compression block size, default 8mb
   -index=FILENAME: read/write index of compressed data into separate file
   -temp=[FILENAME]: keep uncompressed data in the file in stdin-to-stdout mode,
 default srep-data.tmp
   -vmfile=FILENAME: temporary file used by Virtual Memory manager, default srep
-virtual-memory.tmp
   -vmblock=BYTES: size of one block in VM temporary file, default 8mb

   -hash=vmac(default)/siphash/md5/sha1/sha512: store hash checksums in every bl
ock
   -hash-: don't store/check block checksums
   -mmap: use memory-mapped files for match checking
   -slp[+/-/]: force/disable/try(default) large pages support (2mb/4mb)
   -pc[max_offset]: display performance counters [for matches closer than max_of
fset]
   -s: save printed stats from overwriting; -s+/-s-/-sX.Y: update stats every X.
Y seconds
   -v[0..2]: verbosity level
   -rem...: command-line remark