Initializing GraphicsWindow layout & children requires cooperation from
the graphics stack: on my system, for example, it causes a Vulkan
context to get created in order to get driver info. This is a slow
operation, and right now it is taking about 60-70% of the Dolphin
startup time on my system.
Move instead to a lazy-initialization model where the constructor
does nothing, instead offloading work to a separate Initialize() method
called before the window is shown.
I would expect this should be done for other larger parts of the UI,
especially the ones where creating widgets ends up triggering large IO
subsystems (I suspect controller configuration might be doing that).
(I'm not super happy with how this is implemented, but right now it's a
one-off, and it's a major complaint users have with the new UI. I
prioritized getting something working quickly...)
Now the detection heuristic has changed, the old value is no longer
valid.
Some example thresholds for known mipmap effects that should trigger:
SMG's lava has a mimimum difference of ~17.8, SMG2's clouds have a
minimum difference of ~14.8, and Wind Waker's foam has a minimum
difference of ~15
Non-triggering examples were tested and all had a calculated difference
lower than 3.
So a value of 14 should lean towards false-negatives instead of
positives, but this is clearly incomplete testing and may require
further tweaks later.
This no longer converts from sRGB to linear for the reference mip
downsample - even if the original mipmap creation tool used an sRGB
colorspace (which isn't really guaranteed, and may even change per
game), this is a "fast" heuristic that's only an estimate anyway.
The average diff is also now stored in a u64, avoiding floating point
calculations in the per-pixel hot loop.
This should speed up the detection significantly, hopefully fixing
jank when loading in new textures.
Normally, SI is polled at a rate defined by the game, and we have to send the pad state to other clients on every poll or else we'll desync. This can result in fairly high bandwidth usage, especially with multiple controllers, mostly due to UDP/IP overhead.
This change introduces an option to reduce the SI poll rate to once per frame, which may introduce up to one frame of additional latency, but will reduce bandwidth usage substantially, which is useful for users on very slow internet connections.
Polling SI less frequently than the game asked for did not seem to cause any problems in my testing, so this should be perfectly safe to do.
Given we now use a base class for the interface, we can make all member
functions, types and constants that aren't directly related to
instructions private.
HID2.LSQE is the Load/store quantize enable bit for non-indexed format
instructions (which are psq_l, psq_lu, psq_st, and psq_stu). If this bit
is not set and any of these instructions are attempted to be executed,
then a program exception is supposed to occur.
This register is defined as "optional reserved" within the aarch64 ABI.
Linux doesn't use it, but we must not modify it on ios or windows.
As we have plenty of registers on aarch64, let's just always skip this one.
This function was duplicated across all the opcode tables: the main info
tables, the interpreter tables, and the x86-64 JIT tables. However, we
can just make the type of the std::array parameter a template type and
get rid of this duplication.
const on a parameter being passed by value in a prototype doesn't actually signify
anything, these are only applicable in the definition, where they make
the opcode parameter immutable.
inline has external linkage, which doesn't really make sense here, given
the function is only used within this translation unit. So we can
replace inline with static.
While we're at it, the code within the function can also be compressed
to a single return statement.
Previously these were required to be built into the executable so that
the JIT portion of the DSP code would build properly, as the
x86-64-specifics were tightly coupled to the DSP common code. As this is
no longer the case, this is no longer necessary.
This adds a base class that is used to replace the concrete instance of
the x64 JIT pointer within DSPCore. This fully removes the direct use
(read: non-ifdefed) usage of x86-64-specifics within the main DSP code.
Said base can also be used for creating JITs for other architectures,
such as AArch64, etc.
This is one of the last things that needed to be done in order to
finally separate the x86-64-specific code from the rest of the common
DSP code. This splits the tables up similar to how it's currently done
for the PowerPC CPU tables.
Now, the tables are split up and within their own relevant source files,
so the main table within the common DSP code acts as the "info" table
that provides specifics about a particular instruction, while the other
tables contain the actual instruction.
With this out of the way, all that's left is to make a general base for
the emitters and we can then replace the x64 JIT pointer in DSPCore with
it, getting all x64 out of the common code once and for all.
While shuffling all the code around, the removal of the DSPEmitter
includes in some places uncovered indirect inclusions, so this also
fixes those as well.
Despite both being documented as read-only registers, only one of them
is truly read-only. An mtspr to HID1 will steamroll bits 0-4 with
bits 0-4 of whatever value is currently in the source register, the rest
of the bits are not modified as bits 5-31 are considered reserved, so
these ignore writes to them.
PVR on the other hand, is truly a read-only register. Attempts to write
to it don't modify the value within it, so we model this behavior.
This makes it much more straightforward to access WiimoteDevice
instances and also keeps the implementation details of accessing those
instances in one spot.
Given as all external accesses to the WiimoteDevice instances go through
this function, we can make the other two private.
Using reinterpret_cast (or a C-styled equivalent) to reinterpret
integers as floating-point values and vice-versa invokes undefined
behavior. Instead, use BitCast, which does this in a well-defined
manner.
According to PEM 3.3.6.1, if a division by zero occurs and FPSCR.ZE is
set, then the result of the instruction operation is unchanged (see
table 3-13). Similarly, if an invalid operation occurs and FPSCR.VE is
set, then the destination should also remain unchanged (see table 3-12).
Hardware also matches this behavior.
We were handling this for other relevant instructions, but we weren't
doing so for the arithmetic instructions. This corrects that.
This also alters our NI_* functions to return an FPResult type, which
allows us to see which kind of exception in particular is set in
exceptional cases. This is necessary for cases like the fdiv
instructions, which requires handling both ZE and VE being potentially
set.
These can be moved into the RegisterColumn constructor, which avoids
potential allocations in the case a std::function would otherwise need
to allocate to hold all of it's captured data.
Also tidy up the inclusion order while we're at it.
Previously the class was intermixing m_ prefixed variables and
non-prefixed ones, which can be misleading. Instead, we make the
prefixing consistent across the board.
Selecting Dummy or Memory Card would pass wrong values to EXI::ChangeDevice and not work as expected
Changing path had no effect until device was changed as it didn't call EXI::ChangeDevice at all
Makes the values strongly-typed and gets more identifiers out of the
global namespace.
We are forced to use anything that is not "None" to mean none, because
X11 is garbage in that it has:
\#define None 0L
Because clearly no one else will ever want to use that identifier for
anything in their own code (and is why you should prefix literally
any and all preprocessor macros you expose to library users in public
headers).
Makes the enum values strongly-typed and prevents the identifiers from
polluting the PowerPC namespace. This also cleans up the parameters of
some functions where we were accepting an ambiguous int type and
expecting the correct values to be passed in.
Now those parameters accept a PowerPC::CPUCore type only, making it
immediately obvious which values should be passed in. It also turns out
we were storing these core types into other structures as plain ints,
which have also been corrected.
As this type is used directly with the configuration code, we need to
provide our own overloaded insertion (<<) and extraction (>>) operators
in order to make it compatible with it. These are fairly trivial to
implement, so there's no issue here.
A minor adjustment to TryParse() was required, as our generic function
was doing the following:
N tmp = 0;
which is problematic, as custom types may not be able to have that
assignment performed (e.g. strongly-typed enums), so we change this to:
N tmp;
which is sufficient, as the value is attempted to be initialized
immediately under that statement.
This changes the identifier to represent the x86-64 DSP emitter. If any
other JITs for the DSP are added in the future, they all can't use the
same generic identifier.
In cases where we just want a random value for a primitive arithmetic
type, we can wrap this in a template to allow convenient direct
assignment instead of keeping declaration and initialization separate
(making it more difficult to use values uninitialized). This also allows
the use of Common::Random with functions such as std::generate, making
it more flexible in how random values can be generated.
This is only ever used internally. Also change the std::string name over
to a const char*, so that we don't need to potentially allocate anything
on the heap at immediate runtime.
Previously, a total of 114 std::string instances would need to construct
(allocating on the heap for larger strings that can't be stored with
small string optimizations). We can just use an array of const char*
strings instead, which allows us to avoid this.
Given JitBase shouldn't include platform specifics, we can generalize this
preprocessor define and allow any JIT to use it to indicate that generated code should be logged.
While we're at it, also move these defines beneath the includes with the
rest of the defines.
Rather than introduce this handling in every system instruction that modifies
the FPSCR directly, we can instead just handle it within the data structure
instead, which avoids duplicating mask handling across instructions.
This also allows handling proper masking from the debugger register
windows themselves without duplicating masking behavior there either.
ChunkFile doesn't use any of the file utilities, so we can drop these
headers to avoid pulling in unnecessary dependencies. This also
uncovered a few indirect inclusions.
This only queries internal state, it doesn't modify it. With minor
adjustments to BTEmu, this also allows us to make its usage instance a
constant reference.
The required version of MSVC already supports [[maybe_unused]], so we
can utilize this here. When GCC 7 and clang 3.9 become hard
requirements, we can eliminate this macro entirely and replace it with
[[maybe_unused]].
UNUSED is quite a generic macro name and has potential to clash with
other libraries, so rename it to DOLPHIN_UNUSED to prevent that, as well
as make its naming consistent with the force inline macro
This is much better as prefixed double underscores are reserved for the
implementation when it comes to identifiers. Another reason its better,
is that, on Windows, where __forceinline is a compiler built-in, with
the previous define, header inclusion software that detects unnecessary
includes will erroneously flag usages of Compiler.h as unnecessary
(despite being necessary on other platforms). So we define a macro
that's used by Windows and other platforms to ensure this doesn't
happen.
Instead of globbing things under an ambiguous Common.h header, move
compiler-specifics over to Compiler.h. This gives us a dedicated home
for anything related to compilers that we want to make functional across
all compilers that we support.
This moves us a little closer to eliminating Common.h entirely.
Rather than have a separate independent variable that we need to keep
track of in conjunction with the JIT code buffer size itself, amend the
analyst code to use the code buffer constant in JitBase.
Now if the size ever changes, then the analyst will automatically adjust
to handle it.
Given the code buffer is something truly common to all JIT
implementations, we can centralize it in the base class and avoid
duplicating it all over the place, while still allowing for differently
sized buffers.
Gets rid of an inclusion dependency with the DSP interpreter, as well as
a header-based dependency on the DSP opcode tables. This also uncovered
an indirect inclusion on the logger within DSPSymbols.cpp
As peculiar as this may be, decrementer exceptions by means of setting
the decrementer's zeroth bit from 0 to 1 is valid behavior by software
(and is defined in Programming Environments for 32-bit Microprocessors
in section 2.3.14.1 -- Decrementer operation). Given it's valid behavior,
it doesn't necessarily make sense to use a panic alert and halt, as this
isn't a condition where everything should be considered in a critical
state.
Instead, change it to an info log, so we still make note of it, but
without potentially tearing down state or halting emulation.
This fixes the The Last Story prototype that GerbilSoft was testing,
because the apploader is a bit more lenient with the max size of DOL
sections when it detects that you're using a devkit console.
Deduplicates code, and gets rid of some problems the old code had
(such as: bad performance when calling native functions, only one
disc showing up for multi-disc games, Wii banners being low-res,
unnecessarily much effort being needed for adding more metadata).
By making the jitted function a private static function of DSPEmitter,
we can allow access to data members within the context of the function
without making them public overall.
This finally makes all data members for the x64 DSP emitter private.
If we don't do this the prompt *may* appear behind the fullscreened window
and thus cause confusion. This happens both with exclusive fullscreen and
borderless fullscreen (e.g. for OpenGL).
This hardware behavior makes sense, as the FI bit is used to signify an
inexact result. An inexact result is a form of value that results during
the rounding phase of denormalization. If any bits of the significand
are lost during said rounding, then the result is considered to be
inexact.
However NaN and infinity are not classed as subnormals and therefore
don't undergo the denormalization step, making loss of precision not
possible (in NaN's case, numerically rounding something that is
literally Not a Number doesn't even make sense).
FR is set to indicate whether or not the last arithmetic or rounding and
conversion instruction that rounded the intermediate result incremented
the fractional portion of the result. Given neither input types would be
affected by this, this should also be unset.
This corrects more of the exceptional case handling for these values to
match hardware.
As suggested here: https://dolp.in/pr7059#pullrequestreview-125401778
More descriptive than having a std::tuple of FS::Mode, and lets us
give names to known triplets of modes (like in ES). Functions that
only forward mode arguments are slightly less verbose now too.
Prevents implicit conversions to types and requires explicitly
specifying them in order to construct instances of them. Given these are
used within emulation code directly, being explicit is always better
than implicit.
As explained within 179d73ac0d, the table
within the Programming Environments Manual for PowerPC lists the FI and
FR bits as cleared for invalid operation cases. So, we amend the
relevant cases here in order to be accurate to hardware.
As explained within commit a08ad82ace, if
an invalid exception occurs and VE is set, then the destination register
should remain unchanged. Ditto for when ZE is set and a zero divide
exception occurs.
This is only used internally, so we don't need to expose it in the
header. This also allows getting rid of inclusion of the byte swapping
utilities in the header as well.
Given they were only made public so that the callback could access class
state, we can simply make the callback a private static function of
CEXIMic, which allows access to members from the callback function
without making all of said members public.
In the PEM manual, within Table 3-12, which lists what should occur for
invalid operation exceptions, the FPSCR.FI and FPSCR.FR bits are listed
as "Cleared" for when FPSCR.VE is unset and set. So we clear these bits
as well to match hardware behavior.
In the PowerPC Microprocessor Family: The Programming Environments
Manual for 32 and 64-bit Microprocessors, in section 3.3.6.1, Table
3-12 lists what should occur if an invalid operation exception occurs in
situations where VE is set and when VE is not set. In the case where VE
is set, it lists the frD as "Unchanged". It also lists the FPRF flags as
"Unchanged".
Further down in Table 3-13, the listings for what should occur when zero
divide exceptions occur is listed, both for when ZE is set, and when it
isn't. When ZE is set, it lists frD as "Unchanged". It also lists the
FPRF flags as "Unchanged" as well.
This also alters the code so that we don't even calculate the result if
we don't need to compute it, making it a little bit less wasteful.
DataBinHeader is not used anywhere in the code other than via Header,
so let's merge them to reduce noise when accessing header fields
(currently we have to do header.hdr which looks silly).
It would make sense for 0x80 and 0xf0c0 to be respectively
sizeof(BkHeader) and sizeof(Header) as Nintendo is signing anything
that comes after the header, including the BkHeader.
spycrab