As noticed on D52965, the SINT_TO_FP i64 to f32 legalization code has been dead for years - protected by an assert.
Differential Revision: https://reviews.llvm.org/D53703
llvm-svn: 345290
KNL is based on a modified Silvermont core so I don't think these features apply. I think the LEA flag is probably also wrong, but I'm less sure as I barely understand the 3 LEA flags we have currently.
Differential Revision: https://reviews.llvm.org/D53671
llvm-svn: 345285
Summary:
This patch moves the last method in `Z3ConstraintManager` to `SMTConstraintManager`: `canReasonAbout()`.
The `canReasonAbout()` method checks if a given `SVal` can be encoded in SMT. I've added a new method to the SMT API to return true if a solver can encode floating-point arithmetics and it was enough to make `canReasonAbout()` solver independent.
As an annoying side-effect, `Z3ConstraintManager` is pretty empty now and only (1) creates the Z3 solver object by calling `CreateZ3Solver()` and (2) instantiates `SMTConstraintManager`. Maybe we can get rid of this class altogether in the future: a `CreateSMTConstraintManager()` method that does (1) and (2) and returns the constraint manager object?
Reviewers: george.karpenkov, NoQ
Reviewed By: george.karpenkov
Subscribers: mehdi_amini, xazax.hun, szepet, a.sidorin, dexonsmith, Szelethus, donat.nagy, dkrupp
Differential Revision: https://reviews.llvm.org/D53694
llvm-svn: 345284
Summary:
Getting an `APSInt` from the model always returned an unsigned integer because of the unused parameter.
This was not breaking any test case because no code relies on the actual value of the integer returned here, but rather it is only used to check if a symbol has more than one solution in `getSymVal`.
Reviewers: NoQ, george.karpenkov
Reviewed By: george.karpenkov
Subscribers: xazax.hun, szepet, a.sidorin, Szelethus, donat.nagy, dkrupp
Differential Revision: https://reviews.llvm.org/D53637
llvm-svn: 345283
Summary:
Currently, Legalizer is trying to lower G_LOAD with a vector type
that has more than two elements due to the incorrect LegalityPredicate.
This patch fixes the issue by removing the multiplication by 8
as `MemDesc.Size` already contains the size in bits.
Reviewers: dsanders, aemerson
Reviewed By: dsanders
Subscribers: rovka, javed.absar, kristof.beyls, llvm-commits
Differential Revision: https://reviews.llvm.org/D53679
llvm-svn: 345282
Summary:
C++14 sized deallocation is disabled by default due to ABI concerns. However, when a user manually enables it then libc++ should take advantage of it since sized deallocation can provide a significant performance win depending on the underlying malloc implementation. (Note that libc++'s definitions of sized delete don't do anything special yet, but users are free to provide their own).
This patch updates __libcpp_deallocate to selectively call sized operator delete when it's available. `__libcpp_deallocate_unsized` should be used when the size of the allocation is unknown.
On Apple this patch makes no attempt to determine if the sized operator delete is unavailable, only that the language feature is enabled. This could cause a compile error when using `std::allocator`, but the same compile error would occur whenever the user calls `new`, so I don't think it's a problem.
Reviewers: ldionne, mclow.lists
Reviewed By: ldionne
Subscribers: rsmith, ckennelly, libcxx-commits, christof
Differential Revision: https://reviews.llvm.org/D53120
llvm-svn: 345281
This patch introduces a new base class for Instruction named InstructionBase.
Class InstructionBase is responsible for tracking data dependencies with the
help of ReadState and WriteState objects. Class Instruction now derives from
InstructionBase, and adds extra information related to the `InstrStage` as well
as the `RCUTokenID`.
ReadState and WriteState objects are no longer unique pointers. This avoids
extra heap allocation and pointer checks that weren't really needed. Now, those
objects are simply stored into SmallVectors. We use a SmallVector instead of a
std::vector because we expect most instructions to only have a very small number
of reads and writes. By using a simple SmallVector we also avoid extra heap
allocations most of the time.
In a debug build, this improves the performance of llvm-mca by roughly 10% (I
still have to verify the impact in performance on a release build).
llvm-svn: 345280
According to the OpenMP standard, In a task generating construct, if no
default clause is present, a variable for which the data-sharing
attribute is not determined by the rules above is firstprivatized.
Compiler tries to implement this, but if the variable is not directly
used in the task context, this variable may not be firstprivatized.
Patch fixes this problem.
llvm-svn: 345277
If we have a 64-bit EXT where one of the operands is a subvector of a 128-bit
vector then in some cases we can eliminate an extract_subvector by converting
to a 128-bit EXT of the 128-bit vector.
Differential Revision: https://reviews.llvm.org/D53582
llvm-svn: 345275
This fixes a bug PlatformDarwin::SDKSupportsModule introduced by
https://reviews.llvm.org/D47889. VersionTuple::tryParse() can deal
with an optional third (micro) component, but the parse will fail when
there are extra characters after the version number (e.g.: trying to
parse the substring "12.0.sdk" out of "iPhoneSimulator12.0.sdk" fails
after that patch). Fixed here by stripping the ".sdk" suffix first.
(Part of) rdar://problem/45041492
Differential Revision https://reviews.llvm.org/D53677
llvm-svn: 345274
This mirrors what we already do for AArch64 as the cores are similar.
As discussed in the review, enabling the machine scheduler causes
more variations in performance changes so it is not enabled for now.
This patch improves LNT scores by a geomean of 1.57% at -O3.
Differential Revision: https://reviews.llvm.org/D53562
llvm-svn: 345272
Using a multiclass reduces duplication, and makes it easier to add new patterns
later. This refactoring does add some new patterns, but as far as I can tell
there's no IR that will end up triggering them so this is effectively NFC.
Differential Revision: https://reviews.llvm.org/D53580
llvm-svn: 345271
Currently a vector move of 0 or -1 will use different instructions depending on
the size of the vector. Using a single instruction (the 128-bit one) for both
gives more opportunity for Machine CSE to eliminate instructions.
Differential Revision: https://reviews.llvm.org/D53579
llvm-svn: 345270
Summary:
If the instruction in the eliminateFrameIndex function is a DBG_VALUE
instruction, it requires special processing. The frame register is set
to VRFrame and the offset is based on the object offset.
The code is similar to the code used in
lib/CodeGen/PrologEpilogInserter.cpp.
Reviewers: tra
Subscribers: jholewinski, llvm-commits
Differential Revision: https://reviews.llvm.org/D53657
llvm-svn: 345269
Summary:
To enable this, TUScheduler has to provide a way to run async tasks without
needing a preamble or AST!
Reviewers: ilya-biryukov
Subscribers: javed.absar, ioeric, MaskRay, jkorous, arphaman, jfb, kadircet, cfe-commits
Differential Revision: https://reviews.llvm.org/D53644
llvm-svn: 345268
The current state of the llc invocation is:
* Running all the passes from dwarfehprepare to stack coloring
(included)
* It runs it from the LLVM IR included in the file
* It *ADDS* the generated MI from ISel to the MI in the MIR file
* The machine verifier doesn't like it.
Differential Revision: https://reviews.llvm.org/D53698
llvm-svn: 345266
Summary:
This function was performing two hash lookups when a new struct type was requested: first checking if it exists and second to insert it. This patch updates the function to perform a single hash lookup in this case by updating the value in the hash table in-place in case the struct type was not there before.
Similar to r345151.
Reviewers: bkramer
Reviewed By: bkramer
Subscribers: llvm-commits
Differential Revision: https://reviews.llvm.org/D53689
llvm-svn: 345264
Summary:
When building with -fvisibility=hidden, some symbols do not get exported from
libc++.dylib. This means that some entities are not explicitly given default
visibility in the source code, and that we rely on the fact -fvisibility=default
is the default. This commit explicitly gives default visibility to those
symbols to avoid being dependent on the command line flags used.
The commit also remove symbols from the dylib -- those symbols do not
actually need to be exported from the dylib and this should not be an
ABI break.
Finally, in the future, we may want to mark the whole std:: namespace as
having hidden visibility (to switch from opt-out to opt-in), in which
case the changes done in this commit will be required.
Reviewers: EricWF
Subscribers: mgorny, christof, dexonsmith, libcxx-commits
Differential Revision: https://reviews.llvm.org/D52662
llvm-svn: 345260
MicrosoftExt allows explicit constructor calls. Prevent lookup of constructor name unless the name has explicit scope.
This avoids a compile-time crash due to confusing a member access for a constructor name.
Test case included. All tests pass.
Differential Revision: https://reviews.llvm.org/D53441
llvm-svn: 345258
As suggested on D52965, this patch moves the i64 to f64 UINT_TO_FP expansion code from LegalizeDAG into TargetLowering and makes it available to LegalizeVectorOps as well.
Not only does this help perform X86 lowering as a true vectorization instead of (partially vectorized) scalar conversions, it avoids the HADDPD op from the scalar code which can be slow on most targets.
The AVX512F does have the vcvtusi2sdq scalar operation but we don't unroll to use it as it seems to only help for the v2f64 case - otherwise the unrolling cost will certainly be too high. My feeling is that we should leave it to the vectorizers - and if it generates the vector UINT_TO_FP we should use it.
Differential Revision: https://reviews.llvm.org/D53649
llvm-svn: 345256
As was already mentioned in comments for D53364, DWARF 5
spec says about DW_LLE_startx_length:
"This is a form of bounded location description that has two unsigned ULEB operands.
The first value is an address index (into the .debug_addr section) that indicates the beginning of the address range
over which the location is valid. The second value is the length of the range. ")
Currently, the length is always parsed as U32.
Patch change the behavior to parse DW_LLE_startx_length as ULEB128 for DWARF 5
and keeps it as U32 for DWARF4+(pre-DWARF5) for compatibility.
Differential revision: https://reviews.llvm.org/D53564
llvm-svn: 345254
I noticed while fixing PR39368 that we don't have generic shuffle costs for broadcast style shuffles.
This patch adds SK_BROADCAST handling, but exposes ARM/AARCH64 lack of handling of this type, which I've added a fix for at the same time.
Differential Revision: https://reviews.llvm.org/D53570
llvm-svn: 345253
With the fix: do not forget to hanlde the DW_RLE_start_end, which seems was
omited/forgotten/removed by mistake.
Original commit message:
The patch implements the support for DW_RLE_base_address and DW_RLE_offset_pair
.debug_rnglists entries
Differential revision: https://reviews.llvm.org/D53140
----
Added : /lldb/trunk/lit/Breakpoint/Inputs/debug_rnglist_offset_pair.yaml
Added : /lldb/trunk/lit/Breakpoint/debug_rnglist_offset_pair.test
Modified : /lldb/trunk/source/Plugins/SymbolFile/DWARF/DWARFDebugInfoEntry.cpp
Modified : /lldb/trunk/source/Plugins/SymbolFile/DWARF/DWARFDebugRanges.cpp
Modified : /lldb/trunk/source/Plugins/SymbolFile/DWARF/DWARFDebugRanges.h
Modified : /lldb/trunk/source/Plugins/SymbolFile/DWARF/SymbolFileDWARF.cpp
Modified : /lldb/trunk/source/Plugins/SymbolFile/DWARF/SymbolFileDWARF.h
llvm-svn: 345251
When SimplifyCFG changes the PHI node into a select instruction, the debug line records becomes ambiguous. It causes the debugger to display unreachable source lines.
Differential Revision: https://reviews.llvm.org/D53287
llvm-svn: 345250
Currently, we always parse the length field of DW_LLE_startx_length entry as U32.
That is correct for pre-standard definition:
https://gcc.gnu.org/wiki/DebugFission - "A start/length entry contains one unsigned LEB128 number
and a 4-byte unsigned value (as would be represented by the form code DW_FORM_const4u). The first
number is an index into the .debug_addr section that selects the beginning offset, and the second
number is the length of the range. ")
But DWARF v5 says: "This is a form of bounded location description that has two unsigned ULEB operands.
The first value is an address index (into the .debug_addr section) that indicates the beginning of the address
range over which the location is valid. The second value is the length of the range."
Fortunately, we can easily handle the difference. No test case because it seems impossible to test
until we will be ready to use DWARF v5 in tests that need to run the executables.
Differential revision: https://reviews.llvm.org/D53646
llvm-svn: 345249
Summary:
The pfm counters are now in the ExegesisTarget rather than the
MCSchedModel (PR39165).
This also compresses the pfm counter tables (PR37068).
Reviewers: RKSimon, gchatelet
Subscribers: mgrang, llvm-commits
Differential Revision: https://reviews.llvm.org/D52932
llvm-svn: 345243
Summary:
Currently clang-format breaks before the next parameter after multiline parameters (also recursively for the parent expressions of multiline parameters). However, it fails to do so for formatted multiline raw string literals:
```
$ cat test.cc
// Examples
// Regular multiline tokens
int x = f(R"(multi
line)", 2);
}
int y = g(h(R"(multi
line)"), 2);
// Formatted multiline tokens
int z = f(R"pb(multi: 1 #
line: 2)pb", 2);
int w = g(h(R"pb(multi: 1 #
line: 2)pb"), 2);
$ clang-format -style=google test.cc
// Examples
// Regular multiline tokens
int x = f(R"(multi
line)",
2);
}
int y = g(h(R"(multi
line)"),
2);
// Formatted multiline tokens
int z = f(R"pb(multi: 1 #
line: 2)pb", 2);
int w = g(h(R"pb(multi: 1 #
line: 2)pb"), 2);
```
This patch addresses this inconsistency by forcing breaking after multiline formatted raw string literals. This requires a little tweak to the indentation chosen for the contents of a formatted raw string literal: in case when that's a parameter and not the last one, the indentation is based off of the uniform indentation of all of the parameters.
Reviewers: sammccall
Reviewed By: sammccall
Subscribers: cfe-commits
Differential Revision: https://reviews.llvm.org/D52448
llvm-svn: 345242
Multiply a is complex operation so just because some bit of the output isn't used doesn't mean that bit of the input isn't used.
We might able to bound it, but it will require some more thought.
llvm-svn: 345241
Simon Pilgrim