Several functions (and one variable) were being given external linkage.
Instead, relocate them all to anonymous namespaces to make them
internally linked.
Puts the comment in the header where it's more likely to be seen
initially. We can also remove the TODO, given doing nothing or returning
an error is what is generally done for the JIT interface if the JIT
instance isn't valid.
With 7aa305ea358ee1574f1036493411aa2cdf86458f merged, all that remains
within Profiler.cpp is an unused function that just forwards to the
equivalent function within JitInterface. Given that, we can just remove
the source file.
This global belongs in the JitOptions structure, as it's a conditional
setting (A.K.A. option) that changes the behavior of what the JIT does.
Plus it keeps the scope of the variable constrained to the general area
it's intended to be used and nothing further.
swap32() has a const u8* overload that swaps the data being pointed to as
if it were a 32-bit word. We can just use that instead. It gets rid of
undefined behavior, as we're not type punning a pointer and dereferencing it,
and gets rid of the need to cast entirely.
In both cases of the x64 and AArch64 JITs, these would have const casted
away from them, followed by them being placed within an emitter and
having breakpoint instructions written in them.
In this case, we shouldn't be using const period if we're writing to the
emitted data.
Similar in nature to e28d06353947f65fa8218871d5a488f583ec69ab in which
this same change was applied to the x64 emitter.
There's no real requirement to make this const, and this should also
be decided by the calling code, considering we had places that would
simply cast away the const and carry on
Currently, each player buffers their own inputs and sends them to the
host. The host then relays those inputs to everyone else. Every player
waits on inputs from all players to be buffered before continuing. What
this means is all clients run in lockstep, and the total latency of
inputs cannot be lower than the sum of the 2 highest client ping times
in the game (in 3+ player sessions with people across the world, the
latency can be very high).
Host input authority mode changes it so players no longer buffer their
own inputs, and only send them to the host. The host stores only the
most recent input received from a player. The host then sends inputs
for all pads at the SI poll interval, similar to the existing code. If
a player sends inputs to slowly, their last received input is simply
sent again. If they send too quickly, inputs are dropped. This means
that the host has full control over what inputs are actually read by
the game, hence the name of the mode. Also, because the rate at which
inputs are received by SI is decoupled from the rate at which players
are sending inputs, clients are no longer dependent on each other. They
only care what the host is doing. This means that they can set their
buffer individually based on their latency to the host, rather than the
highest latency between any 2 players, allowing someone with lower ping
to the host to have less latency than someone else.
This is a catch to this: as a necessity of how the host's input sending
works, the host has 0 latency. There isn't a good way to fix this, as
input delay is now solely dependent on the real latency to the host's
server. Having differing latency between players would be considered
unfair for competitive play, but for casual play we don't really care.
For this reason though, combined with the potential for a few inputs to
be dropped on a bad connection, the old mode will remain and this new
mode is entirely optional.
Added an option in General config to enable/disable usage statistics. Added a popup on first open if
the user would like to engage in reporting. Clicking cancel or out of the box opts out. Only
clicking 'Ok' will enable reporting. Also added a new android specific values to report.
At some point SetCRFieldBit was modified to operate on RSCRATCH, but the
function was only partially changed. As such, setting SO, GT or LT would
write the right bit to cr_field, but then cr_field would just get
overwritten with RSCRATCH, undoing the work.
also did these things
fixed crash from joining user that isn't hosting via a direct connection
current game stat can now pass to override the current game in config
uses ip endpoint from dolphin.org
Behaviorally, this belongs within the netplay client. The server will
always transmit a known RTC value, so it doesn't even need a global for
this. Given the client receives the packet containing said RTC value, we can
store it as a member variable and provide an accessor for reading that
value.
This removes another global variable within the netplay code.
The idea of this code was to not unroll loops, but it was completely broken.
So we've unrolled all loops, but only up to the second iteration.
Honestly, a better check would test if we branch to code which is already in the compiling block. But this is out of scope for now.
But testing shows that this unrolling actually improve the performance. So instead of fixing this bug, this check can be dropped.
Most settings which affect determinism will now be synced on NetPlay.
Additionally, there's a strict sync mode which will sync various
enhancements to prevent desync in games that use EFB reads.
This also adds a check for all players having the IPL.bin file, and
doesn't load it for anyone if someone is missing it. This prevents
desyncs caused by mismatched system fonts.
Additionally, the NetPlay window was getting too wide with checkboxes,
so FlowLayout has been introduced to make the checkboxes take up
multiple rows dynamically. However, there's some minor vertical
centering issues I haven't been able to solve, but it's better than a
ridiculously wide window.
This is accomplished by having SConfig::GetDirectoryForRegion no longer
return nullptr, as doing that was kind silly, considering we never
check for nullptr.
Basically everything here was race conditions in Qt callbacks, so I changed the client/server instances to std::shared_ptr and added null checks. It checks that the object exists in the callback, and the shared_ptr ensures it doesn't get destroyed until we're done with it.
MD5 check would also cause a segfault if you quit without cancelling it first, which was pretty silly.
