# Parameter definitions for thorn LoopControl

BOOLEAN verbose "Output some loop information at run time" STEERABLE=always
{
} "no"

BOOLEAN veryverbose "Output detailed debug information at run time" STEERABLE=always
{
} "no"

BOOLEAN selftest "Run a self test with every loop (expensive)" STEERABLE=always
{
} "no"

STRING statistics_filename "File name for LoopControl statistics" STEERABLE=always
{
  ""   :: "disable statistics output"
  ".+" :: "file name"
} "LoopControl-statistics"

SHARES: IO

USES STRING out_dir

PRIVATE:



KEYWORD initial_setup "Initial configuration" STEERABLE=always
{
  "legacy" :: "Like a non-LoopControl loop"
  "tiled"  :: "Basic LoopControl setup"
} "tiled"



# NOTE:
# - Intel chips divide the D1 cache into two, one for each hyperthread.
#   The cache is thus not shared!
#   This is off by default since it seems to affect results on intel processors.
BOOLEAN use_smt_threads "Place SMT threads close together" STEERABLE=always
{
} "no"

BOOLEAN align_with_cachelines "Align innermost loops with cache line size" STEERABLE=always
{
} "yes"



INT tilesize_i "Tile size in i direction (in grid points) for loop tiling" STEERABLE=always
{
  1:* :: ""
} 4

INT tilesize_j "Tile size in j direction (in grid points) for loop tiling" STEERABLE=always
{
  1:* :: ""
} 4

INT tilesize_k "Tile size in k direction (in grid points) for loop tiling" STEERABLE=always
{
  1:* :: ""
} 4



INT loopsize_i "Size of each thread's loop in i direction (in grid points) for multithreading" STEERABLE=always
{
  1:* :: ""
} 8

INT loopsize_j "Size of each thread's loop in j direction (in grid points) for multithreading" STEERABLE=always
{
  1:* :: ""
} 8

INT loopsize_k "Size of each thread's loop in k direction (in grid points) for multithreading" STEERABLE=always
{
  1:* :: ""
} 8