Uint128_t native #3205
Uint128_t native #3205
Conversation
graydon
requested review from
MonsieurNicolas and
jonjove
and removed request for
MonsieurNicolas
graydon
force-pushed
the
uint128-t-native
branch
2 times, most recently
from
338987b
to
6cb63c0
Compare
|
Updated to fix bug in |
src/herder/TransactionQueue.cpp
Outdated
| Rounding::ROUND_UP); | ||
| opsToFlood = | ||
| mBroadcastOpCarryover + | ||
| bigDivide(opsToFloodLedger, int64_t(cfg.FLOOD_TX_PERIOD_MS), |
There was a problem hiding this comment.
I think it would be safer to just avoid function overloading (as it's way too easy to forget to cast like this), we can just rename the version that takes an uint128 into bigDivide128 or something like that
There was a problem hiding this comment.
fixed at some length -- wound up renaming all the overloads in numeric.h because I couldn't tell in a lot of cases which overload was even being called in which order. Easiest fix was to split them all apart and just check (and possibly rewrite) each callsite to call the version that corresponds to what I think was meant by the operands passed and surrounding code. This should be carefully reviewed, especially by @jonjove !
graydon
force-pushed
the
uint128-t-native
branch
from
6cb63c0
to
42b9bc5
Compare
graydon
force-pushed
the
uint128-t-native
branch
from
42b9bc5
to
9d69b7b
Compare
graydon
force-pushed
the
uint128-t-native
branch
from
9d69b7b
to
177c166
Compare
|
r+ 177c166 |
Uint128_t native Reviewed-by: MonsieurNicolas
|
I had to stop this while getting 18.0.3 out the door, rekicking |
|
@latobarita: retry |
This is an experiment in using compiler-native uint128_t types when they are available (on gcc and clang), as a possible performance improvement over the library type. Need to discuss a bit, but it seems to work.