dotnet/coreclr

Treat BitConverter.IsLittleEndian as an intrinsic

Increase the length of allowed parameters in ilasm

* Update BitConverter from CoreFX

Related to #9701

* Use ThrowHelper

Because layout in corefx binary has been changed, it causes arm CI failure.
And skip arm32 tests because test is not working recently.

Signed-off-by: Hyung-Kyu Choi <[email protected]>

Signed-off-by: Hyung-Kyu Choi <[email protected]>

…racking

Revert "Simplified MinOpts GC ref tracking."

The extra Unix specific piece of code in the StackFrameIterator::Filter that
handles the difference in the exception stack unwinding on Unix was not
skipping exception trackers belonging to filter clauses. But that was not
right, since filter funclet stack frames behave the same way on Windows and
Unix. They can be present on the stack when we reach their parent frame if
the filter hasn't finished running yet or they can be gone if the filter
completed running, either succesfully or with unhandled exception.

This change adds skipping of filter funclet related exception trackers at
that place so that the common code processes them.

This fixes the GC hole mentioned in the title that was discovered when
running some tests with GCStress mode 2.

…#9778)

JIT: keep ref count for runtime generic context lookups

gettimeofday() is currently used in DateTime.UtcNow.  Depending on the system, it may have accuracy and precision in line with either clock_gettime(CLOCK_REALTIME) or clock_gettime(CLOCK_REALTIME_COARSE), but in returning its data it's necessarily limited by struct timeval's tv_usec field (microseconds), whereas clock_gettime uses struct timespec which has tv_nsec (nanoseconds).  This commit switches to use clock_gettime(CLOCK_REALTIME) to gain the additional precision (though the FILETIME through which the data is returned only supports 100-nsec precision).  On my machine, there was no measurable impact to throughput.

Optimize span clear

The way this code is structured prevents parameters longer than 128 characters
from being passed. This change simply increases that to 1024 characters, although
a more robust change could be made in the future to account for platform differences.

This change removes NakedThrowHelper function for Unix since it was not used.
It also ifdefs out its upstream callers.

Fix incorrect cast in gtFoldExprConst

Remove OSX FileStream Lock/Unlock

It turns out that the getcontext and setcontext that I have used in my stack
oveflow reporting change are not present e.g. on Alpine Linux or on Android.
So I am replacing their usage with RtlCaptureContext and RtlRestoreContext
instead.

I have also found that the addition of the .cfi_adjust_cfa_offset to the
PROLOG_SAVE_REG_PAIR has broken unwinding of all helpers that use
PROLOG_WITH_TRANSITION_BLOCK, because the PROLOG_STACK_ALLOC macro
updates the CFA offset. So I am fixing that by removing the CFA offset
updating from the PROLOG_STACK_ALLOC and adding explicit one to the
ARM64 CallSignalHandlerWrapper.

Fix code generation for array length check in legacy JIT

The fact that an operation is unsigned affects the operation itself but that doesn't mean that the result of the operation is also unsigned. Constants are stored as ssize_t and the node type is TYP_INT so the result has to be sign extended, not zero extended.

Otherwise code that uses the return of IconValue() without first narrowing it to int will behave incorrectly. Such code does exists, even gtFoldExprConst does this when folding shift operations:

uint a = uint.MaxValue;
uint b = 0;
int r = (int)checked(a + b);
Console.WriteLine((r >> 2).ToString("X"));

The above code prints 3FFFFFFF instead of the expected FFFFFFFF.

This also makes gtFoldExprConst consistent with ValueNumStore::EvalFuncForConstantArgs which evaluates TYP_INT nodes as int and then casts to ssize_t.

* Avoid unnecessary branch in CultureAwareComparer

Determine the CompareOptions up-front instead of during each operation.

Use -f instead of -n when copying from Core_Root in runtest.sh