We should be masking the value and emitting a register copy like we do in non-fast isel. Instead we were just updating the value map and emitting nothing.
After r298928 we started seeing cases where we would create a copy from GR8 to GR32 because the source register in a VK1 to GR32 copy was replaced by the GR8 going into a truncate.
This fixes PR32451.
llvm-svn: 298957
In PPCBoolRetToInt bool value is changed to i32 type. On ppc64 it may introduce an extra zero extension for the return value. This patch changes the integer type to i64 to avoid the zero extension on ppc64.
This patch fixed PR32442.
Differential Revision: https://reviews.llvm.org/D31407
llvm-svn: 298955
Summary:
This patch is a stripped down from features a NetBSD process
code (patch is kept under 2k LOC). This code has assumption that
there is only one thread within a debugged process. The only
debugger trap supported is software breakpoint (TRAP_BRKPT).
The generic platform code requires to add dummy function for
watchpoints etc. These functions are currently empty.
This code is not the final platform support as is and it's treated as
a base to extend, refactor and address issues afterwards.
Supported features:
- handle software breakpoints,
- correctly attach to a tracee,
- support NetBSD specific ptrace(2),
- monitor process termination,
- monitor SIGTRAP events,
- monitor SIGSTOP events,
- monitor other signals events,
- resume the whole process,
- get memory region info perms,
- read memory from tracee,
- write memory to tracee,
- read ELF AUXV,
- x86_64 GPR read and write code
For the generic framework include:
- halt,
- detach,
- signal,
- kill,
- allocatememory,
- deallocatememory,
- update threads,
- getarchitecture,
- getfileloadaddress,
- and others.
This code has preliminary AddThread code.
Out of interest in this patch:
- exec() traps,
- hardware debug register traps,
- single step trap,
- thread creation/termination trap,
- process fork(2), vfork(2) and vfork(2) done traps,
- syscall entry and exit trap,
- threads,
- FPR registers,
- retrieving tracee's thread name,
- non x86_64 support.
This code can be used to start a hello world application and trace it.
This code can be used by other BSD systems as a starting point to get similar
capabilities.
Sponsored by <The NetBSD Foundation>
Reviewers: emaste, joerg, kettenis, labath
Subscribers: mgorny, #lldb
Tags: #lldb
Differential Revision: https://reviews.llvm.org/D31374
llvm-svn: 298953
This is less important than increase threshold for private memory,
but still brings performance improvements in a wide range of tests.
Unrolling more for local memory serves three purposes: it allows
to combine ds operations if offset becomes static, saves registers
used for offsets in case of static offsets, and allows better lds
latency hiding.
Differential Revision: https://reviews.llvm.org/D31412
llvm-svn: 298948
Summary:
This prevents InternalAlloc from being called before the sanitizers
are fully initialized.
Reviewers: kubamracek, kcc
Subscribers: llvm-commits
Differential Revision: https://reviews.llvm.org/D31306
llvm-svn: 298947
Summary:
Now that __thread is no longer used for lsan on darwin, i386 builds
can be enabled.
Reviewers: kcc, kubamracek
Subscribers: danalbert, srhines, mgorny, llvm-commits
Differential Revision: https://reviews.llvm.org/D29995
llvm-svn: 298946
This is incorrect to record region boundaries before scheduling,
it may change after scheduling. As a result second pass may see less
instructions to schedule than it should.
Differential Revision: https://reviews.llvm.org/D31434
llvm-svn: 298945
We currently perform the various fp_to_sint XMM conversion and then transfer to the MMX register (on 32-bit via the stack).
This patch improves support for MOVDQ2Q XMM to MMX transfers and adds the XMM->MMX fp_to_sint direct conversion patterns. The SSE2 specifications are the same as for XMM->XMM and XMM->MMX rounding/exceptions/etc.
Differential Revision: https://reviews.llvm.org/D30868
llvm-svn: 298943
Summary: In the ELF linker, we create the buffer identifier for bitcode files by appending the object name to the archive name. This change makes the COFF linker do the same. Without the change, ThinLTO builds can fail with an error message about multiple ThinLTO modules per object file, caused by object files contained in different archives having the same name.
Reviewers: pcc, ruiu
Reviewed By: pcc
Subscribers: mehdi_amini
Differential Revision: https://reviews.llvm.org/D31402
llvm-svn: 298942
-ffp-contract=fast does not currently work with LTO because it's passed as a
TargetOption to the backend rather than in the IR. This adds it to
FastMathFlags.
This is toward fixing PR25721
Differential Revision: https://reviews.llvm.org/D31164
llvm-svn: 298939
Summary:
bad_function_call is currently an empty class, so any object files using
that class will end up with their own copy of its typeinfo, typeinfo
name and vtable, leading to unnecessary duplication that has to be
resolved by the dynamic linker. Instead, give bad_function_call a key
function and put a definition for that key function in libc++ itself, to
centralize the typeinfo and vtable.
This is consistent with the behavior for other exception classes. The
key functions are defined in libc++ rather than libc++abi since the
class is defined in the libc++ versioning namespace, so ABI
compatibility with libstdc++ is not a concern.
Guard this change behind an ABI macro, since it isn't backwards
compatible (i.e., clients built against the new libc++ headers wouldn't
be able to run against an older libc++ library).
Reviewers: mclow.lists, EricWF
Subscribers: mgorny, cfe-commits
Differential Revision: https://reviews.llvm.org/D27387
llvm-svn: 298937
Previously it was covered by the internalization. It turns out we cannot
run internalizer in FE, it break separate compilation tests. Thus early
inliner gets its own option.
Differential Revision: https://reviews.llvm.org/D31429
llvm-svn: 298935
.gnu.hash happen to contain only 32-bit integers for 32-bit arch,
but the section contents are not uniform array members, so setting
entsize doesn't make much sense. This behavior seems to have been
blindly copied from GNU linkers.
llvm-svn: 298934
Deal with case that initial node is deleted during dag-combine leading
to an assertional failure in promoteIntShiftOp.
Fixes PR32420.
Reviewers: spatel, RKSimon
Subscribers: llvm-commits
Differential Revision: https://reviews.llvm.org/D31403
llvm-svn: 298931
We've had several bugs(PR32256, PR32241) recently that resulted from usages of AH/BH/CH/DH either before or after a copy to/from a mask register.
This ultimately occurs because we create COPY_TO_REGCLASS with VK1 and GR8. Then in CopyToFromAsymmetricReg in X86InstrInfo we find a 32-bit super register for the GR8 to emit the KMOV with. But as these tests are demonstrating, its possible for the GR8 register to be a high register and we end up doing an accidental extra or insert from bits 15:8.
I think the best way forward is to stop making copies directly between mask registers and GR8/GR16. Instead I think we should restrict to only copies between mask registers and GR32/GR64 and use EXTRACT_SUBREG/INSERT_SUBREG to handle the conversion from GR32 to GR16/8 or vice versa.
Unfortunately, this complicates fastisel a bit more now to create the subreg extracts where we used to create GR8 copies. We can probably make a helper function to bring down the repitition.
This does result in KMOVD being used for copies when BWI is available because we don't know the original mask register size. This caused a lot of deltas on tests because we have to split the checks for KMOVD vs KMOVW based on BWI.
Differential Revision: https://reviews.llvm.org/D30968
llvm-svn: 298928
If the value is known, but we cannot increment it, conjure a symbol to
represent the result of the operation based on the operator expression,
not on the sub-expression.
In particular, no longer crash on comparing a result of a LocAsInteger increment
to a constant integer.
rdar://problem/31067356
Differential Revision: https://reviews.llvm.org/D31289
llvm-svn: 298927
We want to check each test on each target, so we need another prefix
when SSE and AVX diverge (as they will if we handle 32-byte and higher).
llvm-svn: 298926
Adjustments should be considered properly; we should copy the unadjusted object
over the whole temporary base region. If the unadjusted object is no longer
available in the Environment, invalidate the temporary base region, and then
copy the adjusted object into the adjusted sub-region of the temporary region.
This fixes a regression introduced by r288263, that caused various
false positives, due to copying only adjusted object into the adjusted region;
the rest of the base region therefore remained undefined.
Before r288263, the adjusted value was copied over the unadjusted region,
which is incorrect, but accidentally worked better due to how region store
disregards compound value bindings to non-base regions.
An additional test machinery is introduced to make sure that despite making
two binds, we only notify checkers once for both of them, without exposing
the partially copied objects.
This fix is a hack over a hack. The proper fix would be to model C++ temporaries
in the CFG, and after that dealing with adjustments would no longer be
necessary, and the values we need would no longer disappear from the
Environment.
rdar://problem/30658168
Differential Revision: https://reviews.llvm.org/D30534
llvm-svn: 298924
Reorder work in PromoteIntBinOp to prevent stale (deleted) nodes from
being used.
Fixes PR32340 and PR32345.
Reviewers: hfinkel, dbabokin
Subscribers: llvm-commits
Differential Revision: https://reviews.llvm.org/D31148
llvm-svn: 298923
Summary: ... which applies a set of `AtomicChange`s on code.
Reviewers: klimek, djasper
Reviewed By: djasper
Subscribers: cfe-commits
Differential Revision: https://reviews.llvm.org/D30777
llvm-svn: 298913
Craig Topper