68d37542d9
optimisation: downgrade unknown architecture to a warning
2019-02-02 17:02:17 +11:00
b4db7136eb
optimisation: add lto odr-type-merging and partition=none
2019-01-16 14:33:13 +11:00
fc58c76599
optimisation: disable -fipa-pta
...
GCC encounters an ICE with this option in some projects (eg, edict)
2018-12-10 15:08:14 +11:00
16611ec091
optimisation: group instruction sets by enabled status
2018-09-14 13:00:37 +10:00
99c1ae6afd
optimisation: use more conservative architecture flags
...
We can't enable SSE3 if we want to compile for our Atom test target, so
we explicitly disable this and a few future instruction sets that we
know won't be accessible.
2018-09-13 14:53:06 +10:00
e82c8b7d4a
optimisation: bite the bullet and use O0 by default for debug
...
Og optimisation eliminates too many expressions to be useful for fine
grained debugging so we should make Og the exception rather than O0
2018-09-05 13:17:50 +10:00
305bd4a46e
optimisation: don't enable stack-protector under win32
...
This apparently triggers segfaults when using the standard library so
it's unusable.
2018-08-13 14:56:32 +10:00
43d5fb4523
optimisation: try to enable sse3
...
Effectively all hardware we target includes sse3 instructions at this
point.
2018-03-22 15:08:36 +11:00
6af34bc023
optimisation: add stubs for unsafe maths
2018-01-30 16:26:42 +11:00
712fcdf455
optimisastion: add explanatory note for optimisation flags
2018-01-24 17:33:15 +11:00
630aa5de08
optimisation: switch back to Og for debugging
...
O0 is catastrophically slow for some of our projects. Don't
unconditionally use O0.
2018-01-24 17:28:33 +11:00
8c9ca6a372
optimisation: temporarily disable -ggdb
...
This works around an ICE with libcruft-util's coord classes and GCC's
trunc_int_for_mode.
2018-01-18 11:54:11 +11:00
9cd8c56cdc
optimisation: re-enable fipa-pta under GCC
...
It appears codegen works again? Perhaps we were running into a seperate
issue and fipa-pta always worked.
2018-01-18 11:51:38 +11:00
6576747031
optimsation: prefer -O0 over -Og
...
Og optimises out too many variables to be useful in a large percentages
of the cases. Instead use O0. If we need to revert to Og we'll instead
introduce a new flag.
2017-11-22 17:09:05 +11:00
bd4b8328c2
optimisation: remove deprecated clang optimisation
2017-09-12 14:24:59 +10:00
8a83b07ea0
optimisation: use Og instead of O0 by default
...
O0 is hellishly slow, this is every so slightly less so.
2017-07-04 16:21:00 +10:00
e1ac923f32
optimisation: set gcc-{ar,nm,ranlib} for LTO targets
2017-05-12 15:57:30 +10:00
d106a052a1
nc: protect against multiple invokations
2017-01-25 18:46:41 +11:00
dd58dbde82
optimisation: don't use ENABLE_FOO option names
2017-01-25 18:46:23 +11:00
7fea529319
optimisation: use a more reliable LTO flag test order
2017-01-25 18:45:42 +11:00
7df08f6818
optimisation: use correct capitalisation of Debug
2017-01-23 21:42:34 +11:00
b389d86391
try to cache results as much as possible
2017-01-18 21:43:30 +11:00
a57db559e9
initial import
2017-01-17 19:20:12 +11:00
