I know there is already #3521, but it currently needs a rebase and I
needed to add something to IPC_HLE_Device properly, that is, without
putting everything in the header, so this commit cleans up
IPC_HLE_Device first. (And only IPC_HLE_Device: the rest will still
be handled by #3521.)
Also fixes a few indirect includes (removing unused header includes
from IPC_HLE_Device.h broke building)
This is something that was copy-pasted across the IPC_HLE code
(because it's often used). Since all of the duplicated pieces of code
do the same thing as the previous EnqueueReply, except that they also
write to command_address + 0 and + 8 (to write the correct reply type),
this commit changes EnqueueReply to do that instead of having it
duplicated all over IPC HLE.
When the emulated BT device is created, m_HCIEndpoint (which is a
CtrlBuffer)'s m_cmd_address is not initialised to 0. So it ends up
being a random value. This is normally not an issue… but the
emulated Bluetooth code relies on m_cmd_address to know whether the
HCI endpoint is still valid.
This is a problem with ES_Launch, because the bt_emu class is
destructed and re-constructed, and while m_cmd_address is still
uninitialised, the ES_Launch code disconnects all Wii remotes,
which triggers a HCI event and hence the bug.
Instead of resetting two command buffers, now we only have to call
vkResetCommandPool once at the start of a frame.
NV's recommends using one pool per frame/thread. May offer a very small
boost in performance on some systems.
%n writes to a pointer that's provided as a parameter.
We didn't have a custom implementation of this before,
meaning that %n would trigger a write to the host
memory instead of the emulated memory!
The bounds checks in IOCtl were using 0x200 as the size of
m_Registers, which is more than the actual size, 0x200 / 4.
This commit turns m_Registers into an std::array to allow
for a correct and obvious way of getting its size.
anv seems to set this to zero, which is fine according to the spec, but
we were using it as a maximum, which was resulting in a swap chain
without any buffers being created.
This fixes the screenshot stutter, as this needs more than a frame.
So we won't stall on the png writing at all until emulation stops or
a new screenshot is requested.
This increase the performance of good backends a bit, but slows down the bads one a lot.
Let's fix those backends instead of forcing stupid memcpy in the common code.
