Constructing a global std::map requires clang to generate a linear
amount of code to construct the initializer list if the elements are not
constexpr-constructible. std::vector is not constexpr-constructible, so
this code pattern was generating large amounts of code.
Also, because of PR38829, LLVM is pathologically slow on large basic
blocks, and this causes slow compilation. This works around the bug and
reduces code size.
SemaChecking.cpp -debug-info-kind=limited:
time objsize
before: 1m45.023s 9.8M
after: 0m25.205s 6.9M
So, a 42% obj size reduction and 3.2x speedup.
llvm-svn: 345329
This doesn't appear to matter for deserialization purposes, because we
always know what kind of entity (declaration or statement/expression)
we're trying to load, but it makes the llvm-bcanalyzer output a lot less
mysterious.
llvm-svn: 345328
The SystemZ backend can do arithmetic of memory by loading and then extending
one of the operands. Similarly, a load + truncate can be folded into an
operand.
This patch improves the SystemZ TTI cost function to recognize this.
Review: Ulrich Weigand
https://reviews.llvm.org/D52692
llvm-svn: 345327
This isn't the most object-size efficient encoding, but it's the only
one GDB supports for the pre-standard fission format. I've written fixes
for this twice now... - so perhaps this comment will help me remember
why neither of these have been committed and why I shouldn't try to
write a third fix another year from now...
llvm-svn: 345326
Summary: There are too many reasonable cases that would be considered unorderable.
Reviewers: ruiu, espindola, Bigcheese
Reviewed By: ruiu
Subscribers: grimar, emaste, arichardson, llvm-commits
Differential Revision: https://reviews.llvm.org/D53669
llvm-svn: 345322
Enable the DAG optimization that converts vector div/rem with constants into
multiply+shifts sequences by expanding them early. This is needed since
ISD::SMUL_LOHI is 'Custom' lowered on SystemZ, and will therefore not be
available to BuildSDIV after legalization.
Better cost values for these instructions based on how they will be
implemented (a constant divisor is cheaper).
Review: Ulrich Weigand
https://reviews.llvm.org/D53196
llvm-svn: 345321
The artificial dependencies are not real dependencies. In some cases, they
form circuits with bigger MII. However, they are used to schedule instructions
better.
Differential Revision: https://reviews.llvm.org/D53450
llvm-svn: 345319
The required-vector-width attribute was only used for backend testing and has never been generated by clang.
I believe clang is now generating min-legal-vector-width for vector uses in user code.
With this I believe passing -mprefer-vector-width=256 to clang should prevent use of zmm registers in the generated assembly unless the user used a 512-bit intrinsic in their source code.
llvm-svn: 345317
FENTRY_CALL is actually not taking any input / output operands. The
machine verifier complains now because the target description says that:
* It needs 1 unknown output
* It needs 1 or more variable inputs
llvm-svn: 345316
Include all of the store's source vector operands when creating the
MachineMemOperand. Previously, we were missing the first operand,
making the store size seem smaller than it really is.
Differential Revision: https://reviews.llvm.org/D52816
llvm-svn: 345315
This is similar to D53597, but following up with 2 more enums.
After this, all flag enums should be strongly typed all the way
through to the symbol files plugins.
Differential Revision: https://reviews.llvm.org/D53616
llvm-svn: 345314
When we get the `resolve_scope` parameter from the SB API, it's a
`uint32_t`. We then pass it through all of LLDB this way, as a uint32.
This is unfortunate, because it means the user of an API never actually
knows what they're dealing with. We can call it something like
`resolve_scope` and have comments saying "this is a value from the
`SymbolContextItem` enumeration, but it makes more sense to just have it
actually *be* the correct type in the actual C++ type system to begin
with. This way the person reading the code just knows what it is.
The reason to use integers instead of enumerations for flags is because
when you do bitwise operations on enumerations they get promoted to
integers, so it makes it tedious to constantly be casting them back
to the enumeration types, so I've introduced a macro to make this
happen magically. By writing LLDB_MARK_AS_BITMASK_ENUM after defining
an enumeration, it will define overloaded operators so that the
returned type will be the original enum. This should address all
the mechanical issues surrounding using rich enum types directly.
This way, we get a better debugger experience, and new users to
the codebase can get more easily acquainted with the codebase because
their IDE features can help them understand what the types mean.
Differential Revision: https://reviews.llvm.org/D53597
llvm-svn: 345313
We currently had a 2-step process where we had to call
SetBaseClassesForType and DeleteBaseClasses. Every single caller
followed this exact 2-step process, and there was manual memory
management going on with raw pointers. We can do better than this
by storing a vector of unique_ptrs and passing this around.
This makes for a cleaner API, and we only need to call one method
so there is no possibility of a user forgetting to call
DeleteBaseClassSpecifiers.
In addition to this, it also makes for a *simpler* API. Part of
why I wanted to do this is because when I was implementing the native
PDB interface I had to spend some time understanding exactly what I
was deleting and why. ClangAST has significant mental overhead
associated with it, and reducing the API surface can go along
way to making it simpler for people to understand.
Differential Revision: https://reviews.llvm.org/D53590
llvm-svn: 345312
Summary:
This change rejects the shadowing of a capture by a parameter in lambdas in C++17.
```
int main() {
int a;
auto f = [a](int a) { return a; };
}
```
results in:
```
main.cpp:3:20: error: a lambda parameter cannot shadow an explicitly captured entity
auto f = [a](int a) { return a; };
^
main.cpp:3:13: note: variable a is explicitly captured here
auto f = [a](int a) { return a; };
^
```
Reviewers: rsmith
Reviewed By: rsmith
Subscribers: lebedev.ri, erik.pilkington, cfe-commits
Differential Revision: https://reviews.llvm.org/D53595
llvm-svn: 345308
There are some lurking issues with the handling of the SourceManager.
Somehow sometimes we end up extracting completely wrong
portions of the source buffer.
Reverts r344772, r44760, r344758, r344755.
llvm-svn: 345305
storage class.
To be more in line with what GCC does, switch the condition to be based
on the Static Storage duration instead of the storage class.
Change-Id: I8e959d762433cda48855099353bf3c950b9d54b8
llvm-svn: 345302
Similar to how ICC handles CPU-Dispatch on Windows, this patch uses the
resolver function directly to forward the call to the proper function.
This is not nearly as efficient as IFuncs of course, but is still quite
useful for large functions specifically developed for certain
processors.
This is unfortunately still limited to x86, since it depends on
__builtin_cpu_supports and __builtin_cpu_is, which are x86 builtins.
The naming for the resolver/forwarding function for cpu-dispatch was
taken from ICC's implementation, which uses the unmodified name for this
(no mangling additions). This is possible, since cpu-dispatch uses '.A'
for the 'default' version.
In 'target' multiversioning, this function keeps the '.resolver'
extension in order to keep the default function keeping the default
mangling.
Change-Id: I4731555a39be26c7ad59a2d8fda6fa1a50f73284
Differential Revision: https://reviews.llvm.org/D53586
llvm-svn: 345298
Summary:
When -faligned-allocation is specified in C++03 libc++ defines std::align_val_t as an unscoped enumeration type (because Clang didn't provide scoped enumerations as an extension until 8.0).
Unfortunately Clang confuses the `align_val_t` overloads of delete with the sized deallocation overloads which aren't enabled. This caused Clang to call the aligned deallocation function as if it were the sized deallocation overload.
For example: https://godbolt.org/z/xXJELh
This patch fixes the confusion.
Reviewers: rsmith, EricWF
Reviewed By: EricWF
Subscribers: cfe-commits
Differential Revision: https://reviews.llvm.org/D53508
llvm-svn: 345296
Replacing BinaryOperator::isFNeg(...) to avoid regressions when we
separate FNeg from the FSub IR instruction.
Differential Revision: https://reviews.llvm.org/D53650
llvm-svn: 345295
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
Jonas Paulsson