Age | Commit message (Collapse) | Author | Files | Lines |
|
|
|
Before this fix, corerun would AV on certain invalid paths e.g.: corerun test.exe\
|
|
This allows inlining Span.Length accross versioning boundaries in R2R images. We are leaking the Span implementation details already because the Span indexer is treated as intrinsic. Marking a few more trivial methods as non-versionable should not do any future harm.
|
|
[RyuJIT/ARM32] Add ADDR and fixup return type correctly
|
|
This fix when parsing Hevrew dat e string formatted as dd/MM/yy. we didn't encounter the same on Windows before because default short date pattern on Windows is dd MMM yyyy which we can parse it successfully.
|
|
|
|
|
|
|
|
Improve const return block placement
|
|
[Arm64] Add GT_MULHI
|
|
[Arm64] Transform AND-CMP to TST
|
|
* Introduce FEATURE_GDBJIT_SYMTAB
This commit adds new feature FEATURE_GDBJIT_SYMTAB. When it's set
ELF file with symtab section will be generated for all JITed methods.
It helps lldb and gdb correctly unwind frames of JITed methods as they
can use assembly language inspection to determine location of saved
registers on stack.
* gdbjit symtab feature: use different symbols for each function in JITed code
|
|
* Re-implemented the ecvt function.
Instead of leveraging snprintf, re-implement the ecvt function according to the paper: https://www.cs.indiana.edu/~dyb/pubs/FP-Printing-PLDI96.pdf
Note:
1. This commit won't fix any existing bug.
2. This is a raw implementation of the paper. The performance on Linux only gain 10%. We could tune the performance further.
Fix #10651
* Resolve a cross platform header file issue.
Fix #10651
* Fixed a minor bug. Improved the performance.
Fix #10651
* Modified code according to code review feedback.
This commit fixed most of the issue found in code review. However, some of the feedback may not be involved due to either little performance improvement or need a POC.
Fix #10651
* Try to fix constexpr compile error on Windows.
Fix #10651
* Fixed a potential overflow bug in BigNum::Compare.
Fix #10651
* Improved multiply 10 operation.
Use shift and add operation to replace actual multiply operation.
Fix #10651
* Remove old _ecvt function.
Fix #10651
* Documented the reason why we do not need m+ and m-.
Fix #10651
* Changed exp > 0 to exp != 0 to remove any confusion.
exp should fall in 1 ~ 2046 for normalized value. Denormalized value has exp = 0.
Fix #10651
* Disable the _ecvt tests.
Fix #10651
* Removed _ecvt tests.
Fix #10651
* Re-implemented LogBase2.
Fix #10651
* Use DWORD and DWORD64 for _BitScanReverse and _BitScanReverse64
Fix #10651
* Fixed x86 compile issue for _BitScanReverse64
x86 does not support _BitScanReverse64 so we have to add additional shift operations to handle it.
Fix #10651
* Implemented BitScanReverse64 and BitScanReverse in pal.h
Fix #10651
* Remove the confusion comment which is unrelated to BitScanReverse.
Fix #10651
* Introduced wmemset to enhance the perf for 0.0
Fix #10651
* Improved the performance of converting 0.0.
Fix #10651
* Renamed ecvt to DoubleToNumberWorker.
Fix #10651
* Updated code according to the code review feedback.
Fix #10651
|
|
* [RyuJIT/ARM32] Implement CreateDictionaryLookupHelper only via run-time helper
Implement CreateDictionaryLookupHelper only via run-time helper
* Add assertion for checking CORINFO_USEHELPER
|
|
This change fixes a bug that causes crashes when secure delegate is
invoked on ARM32 with legacy codegen. Register R4 was loaded with
indirection slot address and immediatelly marked as trashed, so the JIT
compiler reused it before the call that was supposed to have that value
in R4.
The fix was to move the reg trashing after the call.
There was also an incorrect marking of that address as EA_PTRSIZE
while it should be EA_BYREF, so I've fixed that too.
|
|
|
|
Remove redundant zero-initialization of struct temps with GC fields.
Structs with GC pointer fields are fully zero-initialized in the prolog if compInitMem is true.
Therefore, we don't need to insert zero-initialization for the result of newobj or for inlinee locals
when they are structs with GC pointer fields and the basic bock is not in a loop.
|
|
|
|
If the only use of a box is in a call to Type:GetType, remove the box and
obtain the type directly. Get the handle needed for obtaining the type from
the newobj call in the original box, via a new box removal option.
Also add some logging to the type optimizations done in morph.
Closes #13187.
|
|
[Arm64] Allow a%b to a&(b-1) transform
|
|
|
|
Fix core dump generation on Centos.
Issue #13764
|
|
Do TreeNodeInfoInit in buildIntervals
|
|
Binary serialization unused methods and attributes cleanup and misc code cleanup
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
* [GDBJIT] Specify C# language in DWARF debug information
To correctly support C# language in lldb DWARF debug information
should contain correct information about language.
Now we use 0x9e57 constant for specify C# language.
* [GDBJIT] Introduce FEATURE_GDBJIT_LANGID_CS
If this feature is set gdbjit will use 0x9e57 as c# lang id,
otherwise it will use DW_LANG_C89.
|
|
Use AsyncTaskCache in ValueTask.AsTask()
|
|
Add new GetMethodNameFromMetadata interface member to ILGEN.
Also fix warning in the build.
[tfs-changeset: 1673078]
[Hand edited to fix break in SPMI]
|
|
Tweak a few things so that generated constant return blocks get laid out
more optimally:
- Don't set BBF_DONT_REMOVE on them; it's ok to move them around, and
if all references to them get dropped, it's fine to eliminate them.
- Insert them immediately after their lexically-last predecessor when
generating them; this increases the likelihood of using fallthrough
rather than gotos to target them in the face of fgReorderBlocks'
reticence to reorder code that we don't have IBC data for and that
hasn't been marked rare.
- Make fgReorderBlocks slightly more aggressive for a pattern that now
shows up somewhat routinely for returning compound conditionals from
predicate functions.
|
|
We can avoid task allocations for common values by using the same task cache that async methods do. This is important to avoid allocations in certain cases when switching from Task-returning to ValueTask-returning methods. If at some point we change Task.FromResult to use the same cache, this can be reverted.
|
|
Check for calls to `Enum.HasFlag` using the new named intrinsic support
introduced in #13815. Implement a simple recognizer for these named
intrinsics (currently just recognizing `Enum.HasFlag`).
When the call is recognized, optimize if both operands are boxes with
compatible types and both boxes can be removed. The optimization changes the
call to a simple and/compare tree on the underlying enum values.
To accomplish this, generalize the behavior of `gtTryRemoveBoxUpstreamEffects`
to add a "trial removal" mode and to optionally suppress narrowing of the
copy source.
Invoke the optimization during importation (which will catch most cases) and
again during morph (to get the post-inline cases).
Added test cases. Suprisingly there were almost no uses of HasFlag in the
current CoreCLR test suite.
Closes #5626.
|
|
Information about each type is dumped once but during debuginfo's
size calculation it was taken into account several time. Due to it
size of debuginfo section could be in several times bigger than
needed that in some cases leaded to OOM.
|
|
When the cache is only partially populated, MemberInfoCache.Insert will
insert the new members into the cache. But then, it will return the
original list. In subsequent calls, it will return the cached list.
However, the order of elements can be different, which can cause issues
with methods that rely on the order of elements returned by
Type.GetFields (for instance, Attribute.GetHashCode)
|
|
Add String.Create span-based method
|
|
Mirror changes from dotnet/corert
|
|
Marking {ReadOnly}Span with IsReadOnly attribute
|
|
|
|
[tfs-changeset: 1673253]
|
|
|
|
|
|
|