Improves the accuracy of FMADDS and other single precision FMA operations
This is accomplished by using an error-free transformation
It also thoroughly explains the quirks and difficulty of these operations
This fixes Mario Strikers and is necessary for fully fixing 1080 Avalanche
For single precision inputs it should be equivalent to a 32-bit FMA
Number of instructions stays the same, but we remove the false
dependency on the input registers.
Before:
0x7a1b037a sbcs w26, w27, w27
After:
0x5a9f23fa csetm w26, lo
Another instance where we needlessly materialized constant zero in a
register. We can just write the carry flag directly.
Before:
0x5280001a mov w26, #0x0 ; =0
0x1a1f035a adc w26, w26, wzr
After:
0x1a9f37fa cset w26, hs
For thread safety, we shouldn't return any pointers or references that
can be used to mutate the state of the PPCSymbolDB. This should be the
final part of making PPCSymbolDB thread safe unless I've missed
something.
9395238 added locking in some PPCSymbolDB functions that access member
variables, but far from all. To ensure thread safety, this commit adds
the missing locking.
9395238 added a mutex to PPCSymbolDB, and made functions return with an
"empty" result if called while the mutex is locked. This new behavior
has the potential to affect not only less important call sites like the
symbol printing mentioned in a comment, but also the JIT deciding if it
should HLE a function.
A later commit in this pull request decreases the amount of lock
contention, reducing the performance impact of this commit.
Fix some common anti-patterns with these data structures.
- You can dereference the iterator returned by `find` to access the
underlying value directly, without an extra `operator[]`/`at`.
- Rather than checking for an element before insertion/deletion, you can
just do the operation and if needed check the return value to
determine if the insertion/deletion succeeded.
Cleanup loading code and reduce amount of signals.
On boot. allow previously loaded map to be kept, if its filename matches. Useful for restarting a game with a large symbol map.
Notes are separate from function symbols, and can be searched separately.
Unlike functions, notes of different length can overlap each other.
In the instruction window, a note will always display over the function symbol.
On real hardware, stswi and stswx don't trigger any of the special
behavior for uncached unaligned writes that was implemented in 543ed8a.
This is confirmed by a hwtest (a new commit in
https://github.com/dolphin-emu/hwtests/pull/42).
This change fixes Dolphin's stswi and stswx implementations so they stop
triggering the special behavior, bringing them back to the behavior they
had before 543ed8a. No games are known to be affected, but Extrems has
reported that it affects homebrew they've made.
Previously, PerformanceTracker registered a callback to be updated on
emulation state changes. PerformanceTrackers live in a global variable
(g_perf_metrics) within libvideocommon. The callback was stored in a
global variable in libcore. This created a race condition at shutdown
between these libraries, when the PerfTracker's destructor tried to
unregister the callback.
Notify the PerfTracker directly from libcore, without callbacks, since
Core.cpp already references g_perf_metrics explicitly. Also rename
Core::CallOnStateChangedCallbacks to NotifyStateChanged to better
reflect what it's doing.
bbf72e7 made a change where you can pass `false` to certain MemChecks
functions to get them to skip performing an "update" step. It was then
up to the caller to call the Update function later.
This commit changes the implementation so that, instead of the caller
passing in a boolean that controls whether a function calls Update, the
function now returns an object that on destruction will call Update.
Callers that are fine with Update being called right away can skip
storing the object in a variable and thereby call Update immediately,
and callers that want to call Update later can keep the object around.
This new design reduces the risk that someone will forget calling
Update.