Uses arithmetic that can be identified more trivially by compilers for
optimizations. e.g. Rather than shifting the halves of the value and
then swapping and combining them, we can swap them in place.
e.g. for the original swap32 code on x86-64, clang 8.0 would generate:
mov ecx, edi
rol cx, 8
shl ecx, 16
shr edi, 16
rol di, 8
movzx eax, di
or eax, ecx
ret
while GCC 8.3 would generate the ideal:
mov eax, edi
bswap eax
ret
now both generate the same optimal output.
MSVC used to generate the following with the old code:
mov eax, ecx
rol cx, 8
shr eax, 16
rol ax, 8
movzx ecx, cx
movzx eax, ax
shl ecx, 16
or eax, ecx
ret 0
Now MSVC also generates a similar, but equally optimal result as clang/GCC:
bswap ecx
mov eax, ecx
ret 0
====
In the swap64 case, for the original code, clang 8.0 would generate:
mov eax, edi
bswap eax
shl rax, 32
shr rdi, 32
bswap edi
or rax, rdi
ret
(almost there, but still missing the mark)
while, again, GCC 8.3 would generate the more ideal:
mov rax, rdi
bswap rax
ret
now clang also generates the optimal sequence for this fallback as well.
This is a case where MSVC unfortunately falls short, despite the new
code, this one still generates a doozy of an output.
mov r8, rcx
mov r9, rcx
mov rax, 71776119061217280
mov rdx, r8
and r9, rax
and edx, 65280
mov rax, rcx
shr rax, 16
or r9, rax
mov rax, rcx
shr r9, 16
mov rcx, 280375465082880
and rax, rcx
mov rcx, 1095216660480
or r9, rax
mov rax, r8
and rax, rcx
shr r9, 16
or r9, rax
mov rcx, r8
mov rax, r8
shr r9, 8
shl rax, 16
and ecx, 16711680
or rdx, rax
mov eax, -16777216
and rax, r8
shl rdx, 16
or rdx, rcx
shl rdx, 16
or rax, rdx
shl rax, 8
or rax, r9
ret 0
which is pretty unfortunate.
Allows the compiler to inform when the result of a swap function is
being ignored (which is 100% a bug in all usage scenarios). We also mark
them noexcept to allow other functions using them to be able to be
marked as noexcept and play nicely with things that potentially inspect
"nothrowability".
Including every OS' own built-in byte swapping functions is kind of
undesirable, since it adds yet another build path to ensure compilation
succeeds on.
Given we only support clang, GCC, and MSVC for the time being, we can
utilize their built-in functions directly instead of going through the
OS's API functions.
This shrinks the overall code down to just
if (msvc)
use msvc's functions
else if (clang or gcc)
use clang/gcc's builtins
else
use the slow path
The template type here is actually a forwarding reference, not an rvalue
reference in this case, so it's more appropriate to use std::forward to
preserve the value category of the type being moved.
Since C++17, the introduction of deduction guides for locking facilities
means that we no longer need to hardcode the mutex type into the locks
themselves, making it easier to switch mutex types, should it ever be
necessary in the future.
We really don't need to pull in several headers of boost related
machinery just to perform the erase-remove idiom (particularly with
C++20 around the corner, which adds universal container std::erase and
std::erase_if, which we can just use instead).
With this, we don't need to link in anything boost-related into common.
Moves local global state into the Impl class itself and initializes it
at the creation of the instance instead of in the function.
This makes it nicer for weakly-ordered architectures, given the
CreateEntry() class won't need to have atomic loads executed for each
individual call to the CreateEntry class.
This makes the class much more flexible and doesn't make performing
copies with classes that contain a bitfield member a pain.
Given BitField instances are only intended to be used within unions, the
fact the full storage value would be copied isn't a big concern (only
sizeof(union_type) would be copied anyways).
While we're at it, provide defaulted move constructors for consistency.
wwylele / 白疾風Today at 6:14 PM
I doubt the performance of constructing regex everytime the function is called
Is TrimSourcePath only called by logging? if so, you can move the implementation into logging, and cache the regex object into global
This function is probably too specific to be in common anyway
Makes it consistent with the regular standard containers in terms of
size representation. This also gets rid of dependence on our own
type aliases, removing the need for an include.
The necessity of this parameter is dubious at best, and in 2019 probably
offers completely negligible savings as opposed to just leaving this
enabled. This removes it and simplifies the overall interface.
This is compromise for swap type being used in union. A union has deleted default constructor if it has at least one variant member with non-trivial default constructor, and no variant member of T has a default member initializer. In the use case of Bitfield, all variant members will be the swap type on endianness mismatch, which would all have non-trivial default constructor if default value is specified, and non of them can have member initializer
While admirable as a means to ensure immutability, this has the
unfortunate downside of making the class non-movable. std::move cannot
actually perform a move operation if the provided operand has const data
members (std::move acts as an operation to "slide" resources out of an
object instance). Given Barrier contains move-only types such as
std::mutex, this can lead to confusing error messages if an object ever
contained a Barrier instance and said object was attempted to be moved.
This is also unused and superceded by standard functionality. The
standard library provides std::this_thread::sleep_for(), which provides
a much more flexible interface, as different time units can be used with
it.
This is an old function that's no longer necessary. C++11 introduced
proper threading support to the language and a thread ID can be
retrieved via std::this_thread::get_id() if it's ever needed.
* Add CheatEngine; Add support for Gateway cheats; Add Cheat UI
* fix a potential crash on some systems
* fix substr with negative length
* Add Joker to the NonOp comp handling
* Fixup JokerOp
* minor fixup in patchop; add todo for nested loops
* Add comment for PadState member variable in HID
* fix: stol to stoul in parsing cheat file
* fix misplaced parsing of values; fix patchop code
* add missing break
* Make read_func and write_func a template parameter
An old function from Dolphin. This is also unused, and pretty inflexible
when it comes to printing out different data types (for example, one
might not want to print out an array of u8s but a different type
instead. Given we use fmt, there's no need to keep this implementation
of the function around.
This is an unused hold-over from Dolphin that was primarily used to
parse values out of the .ini files. Given we already have libraries that
do this for us, we don't need to keep this around.
Storing signed type causes the following behaviour: extractValue can do overflow/negative left shift. Now it only relies on two implementation-defined behaviours (which are almost always defined as we want): unsigned->signed conversion and signed right shift
Everything from here is completely unused and also written with the
notion of supporting 32-bit architecture variants in mind. Given the
Switch itself is on a 64-bit architecture, we won't be supporting 32-bit
architectures. If we need specific allocation functions in the future,
it's likely more worthwhile to new functions for that purpose.
operator+ for std::string creates an entirely new string, which is kind
of unnecessary here if we just want to append a null terminator to the
existing one.
Reduces the total amount of potential allocations that need to be done
in the logging path.
Instead of using an unsigned int as a parameter and expecting a user to
always pass in the correct values, we can just convert the enum into an
enum class and use that type as the parameter type instead, which makes
the interface more type safe.
We also get rid of the bookkeeping "NUM_" element in the enum by just
using an unordered map. This function is generally low-frequency in
terms of calls (and I'd hope so, considering otherwise would mean we're
slamming the disk with IO all the time) so I'd consider this acceptable
in this case.
First of all they are foundamentally broken. As our convention is that std::string is always UTF-8, these functions assume that the multi-byte character version of TString (std::string) from windows is also in UTF-8, which is almost always wrong. We are not going to build multi-byte character build, and even if we do, this dirty work should be handled by frontend framework early.
We always use unicode internally. Any dirty work of conversion with other codec should be handled by frontend framework (Qt). Further more, ShiftJIS/CP1252 are not special (they are not code set used by 3ds, or any guest/host dependencies we have), so there is no reason to specifically include them
Multi-line doc comments still need the '<' after the ///, otherwise it's
treated as a regular comment and makes the original doc comment broken
in viewers, IDEs, etc. While we're at it, also fix some typos in the
comments.
* Add ZeroMQ external submodule
* ZeroMQ libzmq building on macOS
* Added RPC namespace, settings and logging
* Added request queue handling and new classes
* Add C++ interface to ZeroMQ
* Added start of ZeroMQ RPC Server implementation.
* Request construction and callback request handling
* Read and write memory implementation
* Add ID to request format and send reply
* Add RPC setting to macOS UI
* Fixed initialization order bug and added exception handling
* Working read-write through Python
* Update CMakeLists for libzmq to resolve target name conflict on Windows
* Platform-specific CMake definitions for Windows/non-Windows
* Add comments
* Revert "Add RPC setting to macOS UI"
* Always run RPC server instead of configurable
* Add Python scripting example. Updated .gitignore
* Rename member variables to remove trailing underscore
* Finally got libzmq external project building on macOS
* Add missing dependency during libzmq build
* Adding more missing dependencies [skip ci]
* Only build what is required from libzmq
* Extra length checks on client input
* Call InvalidateCacheRange after memory write
* Revert MinGW change. Fix clang-format. Improve error handling in request/reply. Allow any length of data read/write in Python.
* Re-organized RPC static global state into a proper class. [skip ci]
* Make sure libzmq always builds in Release mode
* Renamed Request to Packet since Request and Reply are the same thing
* Moved request fulfillment out of Packet and into RPCServer
* Change request thread from sleep to condition variable
* Remove non-blocking polling from ZMQ server code. Receive now blocks and terminates properly without sleeping. This change significantly improves script speed.
* Move scripting files to dist/ instead of src/
* C++ code review changes for jroweboy [skip ci]
* Python code review changes for jroweboy [skip ci]
* Add docstrings and tests to citra.py [skip ci]
* Add host OS check for libzmq build
* Revert "Add host OS check for libzmq build"
* Fixed a hang when emulation is stopped and restarted due to improper destruction order of ZMQ objects [skip ci]
* Add scripting directory to archive packaging [skip ci]
* Specify C/CXX compiler variables on MinGW build
* Only specify compiler on Linux mingw
* Use gcc and g++ on Windows mingw
* Specify generator for mingw
* Don't specify toolchain on windows mingw
* Changed citra.py to support Python 3 instead of Python 2
* Fix bug where RPC wouldn't restart after Stop/Start emulation
* Added copyright to headers and reorganized includes and forward declarations
* Added a context menu on the buttons including Clear & Restore Default
* Allow clearing (unsetting) inputs. Added a Clear All button
* Allow restoring a single input to default (instead of all)
These currently aren't used and contain commented out source code that
corresponds to Dolphin's JIT. Given our CPU code is organized quite
differently, we shouldn't be keeping this around (at the moment it just
adds to compile times marginally).
The filter is returned via const reference, so this was making a
pointless copy of the entire filter every time a message was being
pushed into the logger instance.
* Change authentication system to JWT
* Address review comments
* Get rid of global variable, fix some documentations, fix a bug when verificating
* Refactor PostJson to avoid code duplication
* Rename jwt_token, add functionality to request a new JWT when getting a 401
* Take bools by value instead of const reference
* Send request again when JWT is invalid and use forward declarations
* Omit brackets
The minimum clang/GCC versions we support already support this. We can also
remove is_standard_layout(), as fread and fwrite only require the type to be
trivially copyable.
These functions include reloading udp client, testing communication and configuring calibration. I also added a function to common/thread.h to use WaitFor.
These are unused and essentially don't provide much benefit either. If
we ever need rotation functions, these can be introduced in a way that
they don't sit in a common_* header and require a bunch of ifdefing to
simply be available
There's no need to perform the resize separately here, since the
constructor allows presizing the buffer.
Also move the empty string check before the construction of the string
to make the early out more straightforward.
This is equivalent to doing:
push_back(std::string(""));
which is likely not to cause issues, assuming a decent std::string
implementation with small-string optimizations implemented in its
design, however it's still a little unnecessary to copy that buffer
regardless. Instead, we can use emplace_back() to directly construct the
empty string within the std::vector instance, eliminating any possible
overhead from the copy.
We can just use the variant of std::string's replace() function that can
replace an occurrence with N copies of the same character, eliminating
the need to allocate a std::string containing a buffer of spaces.
Allows avoiding constructing std::string instances, since this only
reads an arbitrary sequence of characters.
We can also make ParseFilterRule() internal, since it doesn't depend on
any private instance state of Filter
These can just use a view to a string since its only comparing against
two names in both cases for matches. This avoids constructing
std::string instances where they aren't necessary.
We can just leverage std::unique_ptr to automatically close these for us
in error cases instead of jumping to the end of the function to call
fclose on them.
This avoids a redundant std::string construction if a key doesn't exist
in the map already.
e.g.
data[key] requires constructing a new default instance of the value in
the map (but this is wasteful, since we're already setting something
into the map over top of it).
Note: according to cppreference it is necessary to convert char to unsigned char when using std::tolower and std::toupper, otherwise the behaviour would be undefined.
These operators don't modify internal class state, so they can be made
const member functions. While we're at it, drop the unnecessary inline
keywords. Member functions that are defined in the class declaration are
already inline by default.
This provides the equivalent behavior, but without as much boilerplate.
While we're at it, explicitly default the move constructor, since we
have a move-assignment operator defined.
* Change the logging backend to support multiple sinks through the
Backend Interface
* Add a new set of logging macros to use fmtlib instead.
* Qt: Compile as GUI application on windows to make the console hidden by
default. Add filter configuration and a button to open log location.
* SDL: Migrate to the new logging macros
Additionally, when updating fmtlib, there was a change in fmtlib broke
how the old logging macro was overloaded, so this works around that by
just naming the fmtlib macro impl something different
I decided to overload LogMessage because I don't see a reason to come up with a new function name just for this, but if you guys want me to overload FmtLogMessage instead I'm fine with that.
fmt was updated during the clang-format update, which breaks the previous implementation of FmtLogMessage
Changes were:
* Move definition of FmtLogMessage into log.h to use variadic templates as FMT_VARIADIC was removed
To supplement the change above:
* Move Entry and CreateEntry into log.h
* Add LogEntry in backend.cpp
* Uses PopWait to reduce the amount of busy waiting if there aren't many
new logs
* Opens the log file as shared on windows, letting other programs read
the logs, but not write to them while citra is running
* Flushes the logs to disk if a log >= error arrives
Adds a condition var to SPSCQueue so when a new log is pushed it will
wake the consumer thread that is calling PopWait. This only applies to
to queues with NeedSize=true
* Change the logging backend to support multiple sinks through the
Backend Interface
* Add a new set of logging macros to use fmtlib instead.
* Qt: Compile as GUI application on windows to make the console hidden by
default. Add filter configuration and a button to open log location.
* SDL: Migrate to the new logging macros
swap{16,32,64} are defined as macros on the two, but client code
tries to invoke them as Common::swap{16,32,64}, which naturally
doesn't work. This hack redefines the macros as inline functions
in the Common namespace: the bodies of the functions are the
same as the original macros, but relying on OS-specific
implementation details like this is of course brittle.
Instead of doing complex machinations to keep track of the current bit
index, just unset the lowest bit on each iteration, greatly simplifying
the code.
It is unlikely we will ever use this without first doing a Cast to a signed type.
Fixes 9 "unary minus operator applied to unsigned type, result still unsigned" warnings on MSVC2017.3
* Added support for network with ENet lib,
connecting is possible, but data can't be sent, yet.
* fixup! Added support for network with ENet lib,
* fixup! CLang
* fixup! Added support for network with ENet lib,
* fixup! Added support for network with ENet lib,
* fixup! Clang format
* More fixups!
* Moved ENetHost* and ENetPeer* into pimpl classes
* fixup! Moved ENetHost* and ENetPeer* into pimpl classes
* fixup! Clang again
* fixup! Moved ENetHost* and ENetPeer* into pimpl classes
* fixup! Moved ENetHost* and ENetPeer* into pimpl classes
* fixup! Moved ENetHost* and ENetPeer* into pimpl classes
Modules didn't correctly define their dependencies before, which relied
on the frontends implicitly including every module for linking to
succeed.
Also changed every target_link_libraries call to specify visibility of
dependencies to avoid leaking definitions to dependents when not
necessary.
This removes a dependency inversion between core and common. It's also
the proper place for the file since it makes screen layout decisions
specific to the 3DS.
