SystemZAsmParser can now handle -debug by printing the operands neatly to the
output stream. Before this patch this lead to an llvm_unreachable().
It seems that now '-mllvm -debug' does not cause any crashes anywhere (at
least not on SPEC).
Review: Ulrich Weigand
https://reviews.llvm.org/D53328
llvm-svn: 345349
In order to print the IR slot number for the memory operand, the DAG pointer
must be passed to SDNode::dump().
The isel-debug.ll test updated to also check for the IR Value reference being
printed correctly.
Review: Ulrich Weigand
https://reviews.llvm.org/D53333
llvm-svn: 345347
For the reproducer feature I need to be able to export and import the
current LLDB configuration. To realize this I've extended the existing
functionality to print settings. With the help of a new formatting
option, we can now write the settings and their values to a file
structured as regular commands.
Concretely the functionality works as follows:
(lldb) settings export -f /path/to/file
This file contains a bunch of settings set commands, followed by the
setting's name and value.
...
settings set use-external-editor false
settings set use-color true
settings set auto-one-line-summaries true
settings set auto-indent true
...
You can import the settings again by either sourcing the file or using
the settings read command.
(lldb) settings read -f /path/to/file
Differential revision: https://reviews.llvm.org/D52651
llvm-svn: 345346
Summary:
This adds support for LSDA (exception table) generation for wasm EH.
Wasm EH mostly follows the structure of Itanium-style exception tables,
with one exception: a call site table entry in wasm EH corresponds to
not a call site but a landing pad.
In wasm EH, the VM is responsible for stack unwinding. After an
exception occurs and the stack is unwound, the control flow is
transferred to wasm 'catch' instruction by the VM, after which the
personality function is called from the compiler-generated code. (Refer
to WasmEHPrepare pass for more information on this part.)
This patch:
- Changes wasm.landingpad.index intrinsic to take a token argument, to
make this 1:1 match with a catchpad instruction
- Stores landingpad index info and catch type info MachineFunction in
before instruction selection
- Lowers wasm.lsda intrinsic to an MCSymbol pointing to the start of an
exception table
- Adds WasmException class with overridden methods for table generation
- Adds support for LSDA section in Wasm object writer
Reviewers: dschuff, sbc100, rnk
Subscribers: mgorny, jgravelle-google, sunfish, llvm-commits
Differential Revision: https://reviews.llvm.org/D52748
llvm-svn: 345345
Generate the FP16FML intrinsics into arm_neon.h (AArch64 only for now).
Add two new type modifiers to NeonEmitter to handle the new prototypes.
Define __ARM_FEATURE_FP16FML when +fp16fml is enabled and guard the
intrinsics with the macro in arm_neon.h.
Based on a patch by Gao Yiling.
Differential Revision: https://reviews.llvm.org/D53633
llvm-svn: 345344
Handle the case where getCurrentFunction() returns nullptr by passing -1 to
printIRSlotNumber(). This will result in <badref> being printed instead of an
assertion failure.
Review: Francis Visoiu Mistrih
https://reviews.llvm.org/D53333
llvm-svn: 345342
Nodes which have only one predecessor and only one successor can not
always be hidden, even if all states are the same.
An additional condition is needed: the predecessor may have only one successor.
This can be seen on this example:
```
A
/ \
B C
\ /
D
```
Nodes B and C can not be hidden even if all nodes in the graph have the
same state.
Differential Revision: https://reviews.llvm.org/D53735
llvm-svn: 345341
Previously, OSDynamicCast was modeled as an identity.
This is not correct: the output of OSDynamicCast may be zero even if the
input was not zero (if the class is not of desired type), and thus the
modeling led to false positives.
Instead, we are doing eager state split:
in one branch, the returned value is identical to the input parameter,
and in the other branch, the returned value is zero.
This patch required a substantial refactoring of canEval infrastructure,
as now it can return different function summaries, and not just true/false.
rdar://45497400
Differential Revision: https://reviews.llvm.org/D53624
llvm-svn: 345338
Add LLVM intrinsics for the ARMv8.2-A FP16FML vector-form instructions. Add a
DAG pattern to define the indexed-form intrinsics in terms of the vector-form
ones, similarly to how the Dot Product intrinsics were implemented.
Based on a patch by Gao Yiling.
Differential Revision: https://reviews.llvm.org/D53632
llvm-svn: 345337
Summary:
Currently InstPrinter ignores if there are mismatches between block/loop
and end markers by skipping the case if ControlFlowStack is empty. I
guess it is better to explicitly error out in this case, because this
signals invalid input.
Reviewers: aardappel
Subscribers: dschuff, sbc100, jgravelle-google, sunfish, llvm-commits
Differential Revision: https://reviews.llvm.org/D53620
llvm-svn: 345333
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
Jonas Devlieghere