commit bb09e52a7481718ae1fe324d16b99970450908f8 parent 89e0bc567960199dec774e5449e471dd31ef440b Author: Sebastiano Tronto <sebastiano@tronto.net> Date: Fri, 20 Sep 2024 11:42:13 +0200 Added comment on performance issue Diffstat:
| M | src/solvers/h48/gendata_h48.h | | | 13 | +++++++++++++ |
1 file changed, 13 insertions(+), 0 deletions(-)
diff --git a/src/solvers/h48/gendata_h48.h b/src/solvers/h48/gendata_h48.h @@ -284,6 +284,19 @@ gendata_h48h0k4_return_size: return H48_TABLESIZE(0, 4) + INFOSIZE; } +/* +TODO: the following function suffers from a big performance loss (about 40%) +because of all the mutex locking. It is possible to reduce the first lock +(at the beginning of the outermost for loop) by caching the next few items at +once and then looping over the cached elements. But for the second one (inner +loop) the access is non-sequential, so this cannot be done. + +However, in both cases in theory we do not care if we are reading the old +value or the updated one. Depending on the compiler guarantees on concurrent +memory access, we may be fine just removing the locks altogether. + +It is probably possible to solve part of this by using atomics. +*/ STATIC void * gendata_h48h0k4_runthread(void *arg) {