vixl has an assert check to ensure the register sizes are the same for
ubfm.
32-bit Inline syscalls hit this for both arguments and return value.
VCastFromGPR would have hit this but there aren't any x86 instructions
that move 8-bit and 16-bit values in to a vector register.
It doesn't make any sense anymore to have specific instruction names
set for UND versus a nullptr string anymore.
Was useful when we could use it to determine the difference between
undefined from the start versus set in the tables but with unknown
decoding. Which is an edge case.
Now instead just zero initialize the data, which means it is an unknown
type and nullptr name. Which works for use.
Improves initialization time of the InitializeInfoTables function from
423 microseconds to 37 microseconds.
try_emplace() will return the existing entry in the map if it
exists already, so we don't need to perform a find() and then
try_emplace(). We can just use try_emplace() by itself.
Given the calls are all internal, we can store the function pointers
directly and call them with the this pointer instead of indirecting
through std::bind and std::function.
We can create the table at compile-time, since we already have the
function labels available ahead of time.
This also centralizes the op table in one read-only location
In quite a few places we have a raw primitive to represent an IR node's
ID. This can make reading some bits of the API (or the passes) a little
confusing to take in, since there's no meaningful type name for some
data structure members.
We can provide an alias that communicates this directly to the reader.
This also has the nice benefit of providing a single point of definition
for node IDs which can allow for easier changes in the future (e.g.
making Node IDs strongly-typed etc).
In the case of storing a thread state without a guest signal then this
would be filled with garbage data.
This would then cause a crash on thread restart.
Only happens using gdbstub
Migrates lingering instances of the old logger over to fmt where
applicable. This allows removing some of the old defines and functions.
The only remaining usages of the printf-based variant of the logger is
in Tests/LinuxSyscalls/Syscalls.cpp for the strace handling.
The CompileService was spinning up with the incoming thread mask and
then setting the mask once running.
Instead set the mask, which the thread inherits, then set it back once
it is created.
If a SIGBUS is received in compile service code then we weren't handling
it correctly. Instead we would fail the JIT space check and hand it off
to the guest.
When compiling code and linking blocks, disable all signals then
reenable once back in the dispatcher.
This adds 15-40 microseconds on to each of these steps and some
additional branching overhead.
This is a problem where our allocator is not reentrant safe and when
receiving a signal we need to jump to a new code region and compile.
Depending on the mutexes that are currently live, we will just stall out
forever.
This is /very/ commonly picked up when attempting to run pressure-vessel
Fixes#1217
Instead of throwing an error and closing down FEX. Instead pass the
SIGILL to the guest application.
On unhandled instruction implementation the instruction, we instead emit
a _Break IR op at that location.
A _Break IR op will ensure the context state is synchronized at the
point of of the fault and has fairly low overhead. We branch to the
dispatcher which does the SRA spilling.
Tested this with an application that attempts an AVX512 instruction,
catches the fault, and continues onward.
With #1383 in place, we also won't pass spurious ERROR_AND_DIE to the guest anymore.
By default std::stack uses the heavyweight std::deque container for storage,
but std::vector is perfectly suitable for our purposes and it has better
performance characteristics.
Only the interpreter requires this currently since the stack register
doesn't match.
The JITs correctly set up their stack register and matches what the API
wants.
This is guaranteed to work. signal + rt_sigreturn is a handshake with
the kernel to return the data that the kernel wants.
Only time this doesn't align is when the guest signal handler longjumps
out of the signal handler. Which leaves a dangling entry in the
SignalFrame stack on the interpreter.
These are equivalent to the REX prefix's RXB bits, except that they're
in 1's complement form.
These are trivial to handle and fix usages of the upper range of
registers for ModRM encoded fields.
To ensure that we have coverage for this, I've altered the BEXTR test to
make use of R14 and R15.
In a syscall microbench this improves performance by ~19% on my
Snapdragon 888.
Going from ~9.6 million syscalls per second to ~11.5 million.
Macbook Pro is less effective here due to high syscall overhead due to
VM. Going form 7.2M/s to 7.6M/s, ~6% improvement
We can also inline some 32-bit syscalls but that will need some more
work which isn't done yet. Even though the op in the JIT supports it.
Ryan Houdek