This commit should have zero performance effect if SSBOs are supported.
If they aren't (e.g. on all Macs), this commit alters FramebufferManager
to attach a new stencil buffer and VertexManager to draw to it when
bounding box is active. `BBoxRead` gets the pixel data from the buffer
and dumbly loops through it to find the bounding box.
This patch can run Paper Mario: The Thousand-Year Door at almost full
speed (50–60 FPS) without Dual-Core enabled for all common bounding
box-using actions I tested (going through pipes, Plane Mode, Paper
Mode, Prof. Frankly's gate, combat, walking around the overworld, etc.)
on my computer (macOS 10.12.3, 2.8 GHz Intel Core i7, 16 GB 1600 MHz
DDR3, and Intel Iris 1536 MB).
A few more demanding scenes (e.g. the self-building bridge on the way
to Petalburg) slow to ~15% of their speed without this patch (though
they don't run quite at full speed even on master). The slowdown is
caused almost solely by `glReadPixels` in `OGL::BoundingBox::Get`.
Other implementation ideas:
- Use a stencil buffer that's separate from the depth buffer. This would
require ARB_texture_stencil8 / OpenGL 4.4, which isn't available on
macOS.
- Use `glGetTexImage` instead of `glReadPixels`. This is ~5 FPS slower
on my computer, presumably because it has to transfer the entire
combined depth-stencil buffer instead of only the stencil data.
Getting only stencil data from `glGetTexImage` requires
ARB_texture_stencil8 / OpenGL 4.4, which (again) is not available on
macOS.
- Don't use a PBO, and use `glReadPixels` synchronously. This has no
visible performance effect on my computer, and is theoretically
slower.
GCMemcard.h has quite a bit of different classes implemented within it
that could likely be split up into other files to make it a little
easier to read. However, they should be moved into their own folder
first so that they don't clutter up the base HW directory.
Defaulting to SSL verification off, *and* forcing it to be off even
when the emulated software asks us to enable it is very bad behaviour,
inaccurate and insecure.
Because the old option defaulted to off, we have to change the INI
option name to force the new default to be used. Unfortunate,
but without this we cannot ensure our users' security.
LoadPatches was apparently never being called when booting
Wii discs. Maybe this will fix the recent regression with
cheat codes not getting loaded? I don't know how this
managed to work to begin with, though...
(The call was also moved for WADs, just for consistency.)
ES.cpp was becoming pretty huge. This commit splits the ES code into
several files:
* Main ES (launch, UID, current title directory and title ID, etc.)
* Device identity and encryption (ID and cert, keys, encrypt/decrypt)
* Title management (imports, exports, deletions)
* Title contents (open/close/read/seek)
* Title information (titles, stored contents, TMDs)
* Views (for tickets and TMDs)
This prevents truncation when assigning to this member in the
constructor. This isn't size-critical code, so opting for the more
straightforward assignment is fine here.
Advantages:
* Simpler code in general
* No extra volume objects created
* Now actually notices if the disc or partition gets
changed while the core is running
* No longer picks up on disc access done by the GUI
(it used to do so as long as the core was running)
* Gets rid of a Core dependency in DiscIO
There are two performance disadvantages:
* FileMonitor is now a bit slower when used with VolumeDirectory
because FileMonitor now always uses the FileSystemGCWii code
for finding filenames instead of VolumeDirectory finding the
filename on its own and directly hooking into FileMonitor.
But this isn't such a big deal, because it's happening on the
DVD thread, and my currently unmerged file system PR will make
FileSystemGCWii's file finding code about as fast as
VolumeDirectory's.
* FileMonitor's creation of the file system object is now
done on the CPU thread instead of the DVD thread, and
it will be done even if FileMonitor logging is disabled.
This will be fixed in the next commit.
PR #3582 removed VolumeIsValid, then PR #3582 added a call
to VolumeIsValid, then both PRs were merged without either
of them being rebased on top of the other.
There's no point in creating a volume without a blob,
since essentially all the functionality of a volume
requires a blob to be used.
Also, VolumeCreator doesn't support creating volumes
without blobs (it can't even figure out the volume type
unless it gets a blob), so it's currently impossible
for a volume to be created without a blob.
Given none of these are used outside of the DSPEmitter class (nor does
it really make sense to allow them to be used outside of the class),
these should all be made private.
Using DiscIO's NAND content loader is the wrong way to get the ticket
for a title, because it checks whether the TMD is present and the
validity check fails if it isn't. This is not the correct behaviour:
we should just read the ticket from /ticket without caring about TMDs.
* IOS doesn't rely on the number of contents indicated in the TMD.
Instead, it checks whether the contents *do* exist on the NAND.
* Implement ES_GetTMDStoredContents (and the count ioctlv).
* Drop a hack in ES_GetStoredContents, which is unnecessary now that
we do it properly.
This is slightly safer than writing contents to /title directly.
We still cannot rename everything in one go atomically, but this allows
implementing AddTitleCancel very easily.
Also, this ensures that when a title import fails, no incomplete files
will be left in the title directory, which can mess up the system menu.
Regression introduced in e99cd57 / 4935: VideoBackends: Set the maximum
range when the depth range is oversized[1]. The NV_depth_buffer_float
extension is not part of OpenGL 3.0, and requiring it causes a hard
crash when it's not supported (e.g. macOS).
[1]: https://github.com/dolphin-emu/dolphin/pull/4935
Most of the Volume code was written before this
convenience function was added. Let's use it more.
Also deleting m_pReader nullptr checks that are
unnecessary because of Read (which ReadSwapped calls)
already having a nullptr check.
This stops the virtual method call from within the Renderer constructor.
The initialization here for GL had to be moved to VideoBackend, as the
Renderer constructor will not have been executed before the value is
required.
This adds a check to the SSL code to make sure we are using the correct
client certificate and key (and root CA).
Now, instead of silently failing, the user will be notified whenever a
file is missing or when it is invalid, i.e. when the hash does not
match; this is likely to happen for existing users as the program
linked in the network guide extracted the wrong certs :(
This partially restores a hack which causes ES to fake ticket views for
IOS titles.
This is necessary because we still allow users to boot games from the
game list, so, with no way of making sure the required IOSes are
installed beforehand, games may OSPanic() when they try to reload to
some IOS version and just find out that the IOS is not installed
(something which *never* happens on the real console, of course).
A warning is printed in the logs to make sure technical users know the
IOS titles are being faked. To try and keep things accurate in all
other cases, this hack is only active when it is needed (when the
current title is a disc title which was launched from the game list).
Depending upon the desktop colour scheme, the light/dark
GameList backgrounds can cause the always white text
to become unreadble.
Use the common luminance approximation algorithm to
determine whether black text should be used instead.
This adds a hash check for imported contents. IOS does it for security;
we do it for a somewhat different reason, to catch content decryption
bugs before incorrectly decrypted contents get written to the NAND,
which can cause titles to be corrupted.
Either way, we should have been doing this check in all cases.
const, when used on value type parameters in the declaration,
is superfluous. This doesn't really convey any information to take note
of when using the function. This only matters in the definition when you
want to prevent accidental modification.
e.g.
// Header
void CalculateSomething(int lhs, int rhs);
// Definition
void CalculateSomething(const int lhs, const int rhs)
{
// lhs and rhs can't accidentally be modified
}
When the TMD doesn't exist on the NAND, IOS returns -106.
This commit also changes IsValid() to not check for the TMD validity,
since this is not always something we want. (IOS can have different
error codes when the TMD is missing, or even worse, simply assume
that the TMD is valid.)
IOS determines installed titles by looking at /title, not uid.sys,
which is more like a history of installed titles. And it does not care
at all about the installed TMD (or even if it is present at all).
This moves all the byte swapping utilities into a header named Swap.h.
A dedicated header is much more preferable here due to the size of the
code itself. In general usage throughout the codebase, CommonFuncs.h was
generally only included for these functions anyway. These being in their
own header avoids dumping the lesser used utilities into scope. As well
as providing a localized area for more utilities related to byte
swapping in the future (should they be needed). This also makes it nicer
to identify which files depend on the byte swapping utilities in
particular.
Since this is a completely new header, moving the code uncovered a few
indirect includes, as well as making some other inclusions unnecessary.
Not only this is pretty pointless because there is a load and save button on the appropriate panels, but for the breakpoints one, it caused an error while mapping the memory since adding memory breakpoint requires to update the DBAT and this is done too early (right after boot). This also only worked if you had the right panel on making it even more useless because it would fail to laod if you didn't have the right panel on. It's better to just let the user click load and save.
This removes wrappers for ES_DIVerify and ES::LoadWAD. They are not
really useful as we can simply call the ES function directly, and
it is actually somewhat confusing because both functions are static
and are not tied to a particular ES instance.
This allows Dolphin to stay up-to-date about what title is currently
running, which fixes savestates, screenshots, etc. after an ES_Launch.
Same limitation as with MIOS: currently, GameINIs are not reloaded,
because it's a pain with the current config system. It'll happen
when the new config system is done, and this commit makes it much
easier to do!
Some members are shared between ES instances, and they are just global
variables in IOS.
This is more efficient than getting the installed titles or setting the
current active title tons of times for no reason.
This changes ES to keep track of the active title properly,
just like IOS:
* It is NOT changed on resource manager open/close.
* It is reset on IOS reload.
* It is changed by ES_DIVerify and ES_Launch.
IOS stores the active title in a structure like this:
struct ESTitleContext
{
Ticket* ticket;
TMD* tmd;
u32 active;
};
With this commit, we also do keep the Ticket and TMD around. This
makes some of the DI ioctlvs (which return data about the current
active title) trivial to implement in the future.
This fixes the System Menu not being able to see update partitions
and also allows us to change Dolphin's active game info in the future.
Currently, slowmem is used at any time that memory breakpoints are in use. This commit makes it so that whenever the DBAT gets updated, if the address is overllaping any memchecks, it forces the use of slowmem. This allows to keep fastmem for any other cases and noticably increases performance when using memory breakpoints.
This fixes ES_GetTMDView and ES_GetTMDViewSize to return -106
(FS_ENOENT) if the title does not exist (and more specifically when no
TMD exists in the NAND). This allows installed (or not installed) IOSes
to be detected properly.
It makes absolutely no sense to have asserts for what is obviously an
error condition. And they should definitely not cause Dolphin to crash
because it assumes that everything is valid, and Dolphin should not
report those to the user either, as it is very obviously a bug in the
emulated software and there is nothing the user (or we) can do.
This commit replaces all of the request asserts with proper checks
and adds missing checks for some ioctlvs. We still do not check sizes
yet; this will be done later.
Before #4581, an invocation of `SetBlendMode` could invoke
`glBlendEquationSeparate` and `glBlendFuncSeparate` even when it was
setting `glDisable(GL_BLEND)`. I couldn't figure out how to map the old
behavior over to the new BlendingState code, so I changed it to always
call the two blend functions.
Fixes https://bugs.dolphin-emu.org/issues/10120 : "Sonic Adventure 2
Battle: graphics crash when loading first Dark level".
Currently, slowmem is used at any time that memory breakpoints are in use. This commit makes it so that whenever the DBAT gets updated, if the address is overllaping any memchecks, it forces the use of slowmem. This allows to keep fastmem for any other cases and noticably increases performance when using memory breakpoints.
The vector was not constructed with the proper size, which results in a
buffer overflow as we were using memcpy.
This commit fixes that mistake and also uses a safer way of copying the
ticket view data (std::vector::insert instead of memcpy).
This commit fixes ES_Launch to work mostly the same as the real IOS
(except temporary, internal files such as /sys/launch.sys and title
handling; the latter will be handled in a future PR).
First of all, this adds two IOS functions, which correspond to two
IOS syscalls: 0x41 (boot_ppc) and 0x42 (boot_ios).
boot_ios() writes the new version to 0x3140, loads the new kernel,
which then proceeds to reinit IPC and load modules as part of its
boot process. Note that this doesn't include writing to any of the
other constants in the 0x3100 region.
In Dolphin, this is implemented by changing the active IOS
version variable, writing to 0x3140 and resetting all devices. This
has exactly the same effect as the real syscall.
The other syscall, boot_ppc(), writes code to the EXI boot buffer,
pokes all constants to memory before bootstrapping the PPC with a
binary from the NAND.
We skip the low level stuff and just load the DOL to memory (and set
the PPC's PC to 0x3400), which is essentially what IOS does.
The other change is mostly related to how ES_Launch is handled.
With a real IOS, if the launched title type is 00000001 (system) and
the title is not 1-2 (System Menu), ES calls boot_ios().
Otherwise, ES handles the launch as a PPC title. It reads the TMD
to determine the required IOS version. If it is the same, boot_ppc()
is called directly. If not, ES saves the title to launch to the NAND
before launching the new IOS. After the new IOS has finished booting,
it will notice the flag and then launch the requested title.
What this commit does is really just implement this logic into IOS HLE.
The result is a fix for a regression introduced by SetupMemory,
where reloading an IOS would have overwritten some OS constants.
This fixes booting games from the disc channel.
No idea why this wasn't implemented whereas ES_DeleteTicket and
ES_DeleteTitleContent were.
This probably fixes title deletion in old System Menus, and maybe
the new ones as well in some cases; I've seen 4.3 use this ioctlv.
Michael Maltese