[lmi] Helpful compiler options [Was: Micro-optimization in ledger_format]

[Greg Chicares]

On 2019-01-17 02:33, Vadim Zeitlin wrote:
[...]
>  BTW, in principle, PGO should help with making your original (i.e.
> current) code as fast as my version as it would discover that the "global"
> variable inside imbued() is not really shared. It might be interesting to
> try building lmi with PGO, it's known to result in quite significant
> optimizations in many cases.

https://www.phoronix.com/scan.php?page=article&item=gcc-82-pgo&num=3
|
| The Crafty chess benchmark benefited by a very small amount with PGO.
|
| The Stockfish chess benchmark performance advantage was slightly more 
noticeable than Crafty.
|
| The TSCP chess engine saw about a 9% performance boost thanks to GCC's PGO.

A possible improvement on the order of ten percent would be well worth
investigating.

Similarly, a few months ago I tried using gcc's "sanitizers", but IIRC
they don't work with 32-bit msw binaries. (That's yet another reason
why I need to work on multiarch builds.)

We're already using quite a few gcc-7 warning options. Some of them are
turned off for certain files, either for good reasons:

bourn_cast_test.o: gcc_common_extra_warnings += \
  -Wno-double-promotion \

...or just because we haven't gotten around to addressing them:

currency_test.o: gcc_common_extra_warnings += \
  -Wno-useless-cast \

And 'workhorse.make' lists some extra options that we can consider
enabling (making any useful code changes they may suggest) in 2019:

# Consider these later.
#
# -Wdate-time: only for "bit-wise-identical reproducible compilations"

postponed_gcc_common_warnings := \
  -Wdate-time \
  -Wmissing-declarations \
  -Wnull-dereference \
  -Wswitch-enum \
  -Wunsafe-loop-optimizations \

# Consider these later.
#
# -Wsuggest*: use these only occasionally, like -Weffc++

postponed_gcc_cxx_warnings := \
  -Wnon-virtual-dtor \
  -Wsign-promo \
  -Wsuggest-final-methods \
  -Wsuggest-final-types \