Summary: As per title. Also add a facility method to get the name of a basic block from the C API.
Reviewers: bogner, chandlerc, echristo, dblaikie, joker.eph, Wallbraker
Subscribers: llvm-commits
Differential Revision: http://reviews.llvm.org/D16912
llvm-svn: 260309
1) Turns out we weren't correctly uniquing types for C++. We would search our repository for "lldb_private::Process", but yet store just "Process" in the unique type map. Now we store things correctly and correctly unique types.
2) SymbolFileDWARF::CompleteType() can be called at any time in order to complete a C++ or Objective C class. All public inquiries into the SymbolFile go through SymbolVendor, and SymbolVendor correctly takes the module lock before it call the SymbolFile API call, but when we let CompilerType objects out in the wild, they can complete themselves at any time from the expression parser, so the ValueObjects or (SBValue objects in the public API), and many more places. So we now take the module lock when completing a type to avoid two threads being in the SymbolFileDWARF at the same time.
3) If a class has a template member function like:
class A
{
<template T>
void Foo(T t);
};
The DWARF will _only_ contain a DW_TAG_subprogram for "Foo" if anyone specialized it. This would cause a class definition for A inside a.cpp that used a "int" and "float" overload to look like:
class A
{
void Foo(int t);
void Foo(double t);
};
And a version from b.cpp that used a "float" overload to look like:
class A
{
void Foo(float t);
};
And a version from c.cpp that use no overloads to look like:
class A
{
};
Then in an expression if you have two variables, one name "a" from a.cpp in liba.dylib, and one named "b" from b.cpp in libb.dylib, you will get conflicting definitions for "A" and your expression will fail. This all stems from the fact that DWARF _only_ emits template specializations, not generic definitions, and they are only emitted if they are used. There are two solutions to this:
a) When ever you run into ANY class, you must say "just because this class doesn't have templatized member functions, it doesn't mean that any other instances might not have any, so when ever I run into ANY class, I must parse all compile units and parse all instances of class "A" just in case it has member functions that are templatized.". That is really bad because it means you always pull in ALL DWARF that contains most likely exact duplicate definitions of the class "A" and you bloat the memory that the SymbolFileDWARF plug-in uses in LLDB (since you pull in all DIEs from all compile units that contain a "A" definition) uses for little value most of the time.
b) Modify DWARF to emit generic template member function definitions so that you know from looking at any instance of class "A" wether it has template member functions or not. In order to do this, we would have to have the ability to correctly parse a member function template, but there is a compiler bug:
<rdar://problem/24515533> [PR 26553] C++ Debug info should reference DW_TAG_template_type_parameter
This bugs means that not all of the info needed to correctly make a template member function is in the DWARF. The main source of the problem is if we have DWARF for a template instantiation for "int" like: "void A::Foo<int>(T)" the DWARF comes out as "void A::Foo<int>(int)" (it doesn't mention type "T", it resolves the type to the specialized type to "int"). But if you actually have your function defined as "<template T> void Foo(int t)" and you only use T for local variables inside the function call, we can't correctly make the function prototype up in the clang::ASTContext.
So the best we can do for now we just omit all member functions that are templatized from the class definition so that "A" never has any template member functions. This means all defintions of "A" look like:
class A
{
};
And our expressions will work. You won't be able to call template member fucntions in expressions (not a regression, we weren't able to do this before) and if you are stopped in a templatized member function, we won't know that are are in a method of class "A". All things we should fix, but we need <rdar://problem/24515533> fixed first, followed by:
<rdar://problem/24515624> Classes should always include a template subprogram definition, even when no template member functions are used
before we can do anything about it in LLDB.
This bug mainly fixed the following Apple radar:
<rdar://problem/24483905>
llvm-svn: 260308
The problem is that the master's thread state was not saved before entering a
parallel region so it does not remember tasks when it returns.
llvm-svn: 260306
Now the parser supports `%got(sym)` expressions only but `%got(sym + const)`
variant is also valid and accepted by GAS.
Differential Revision: http://reviews.llvm.org/D16885
llvm-svn: 260305
Also remove definitions if provided by clang (3.7+)
This halves the size of builtin.opt.{cedar,barts}.bc
reviewer: tstellard
Signed-off-by: Jan Vesely <jan.vesely@rutgers.edu>
llvm-svn: 260303
Make cl_khr_fp64 define per-device.
This patch does not change the generated Makefile (for llvm 3.6, 3.7)
v2: Make the device defines per LLVM version, 'all' for all versions
reviewer: tstellard
Signed-off-by: Jan Vesely <jan.vesely@rutgers.edu>
llvm-svn: 260302
The -install_name linker flag will use "@rpath/" when supported in CMake
which is the recommended usage for dynamic libraries on Mac OSX.
llvm-svn: 260300
Summary:
`hasNoAliasAttr` is buggy: it checks to see if the called function has
a `noalias` attribute, which is incorrect since functions are not even
allowed to have the `noalias` attribute. The comment on its only
caller, `llvm::isNoAliasFn`, makes it pretty clear that the intention
to do the `noalias` check on the return value, and not the callee.
Unfortunately I couldn't find a way to test this upstream -- fixing
this does not change the observable behavior of any of the passes that
use this. This is not very surprising, since `noalias` does not tell
anything about the contents of the allocated memory (so, e.g., you
still cannot fold loads). I'll be happy to be proven wrong though.
Reviewers: chandlerc, reames
Subscribers: mcrosier, llvm-commits
Differential Revision: http://reviews.llvm.org/D17037
llvm-svn: 260298
While trying to track down what appears to be a LoopVectorizer bug, I noticed that we had no validation of the correctness of calls emitted to @llvm.masked.load and @llvm.masked.store. This meant malformed IR was showing up much much later than it should. Hopefully, having Verifier rules in place will make this easier to isolate.
llvm-svn: 260296
I reinvented this functionality in http://reviews.llvm.org/D16828 because it was
hidden away as a static function. The changes in x86 are not based on a complete
audit. I suspect there are other possible uses there, and there are almost certainly
more potential users in other targets.
llvm-svn: 260295
This basically reverts commit r260073 because it is found that
augmentation strings don't always start with "zR". It is reported
as https://llvm.org/bugs/show_bug.cgi?id=26541.
llvm-svn: 260294
This reverts commit r259578.
There are enough issues with this small patch that it is better to
revert and then commit a fixed version (will be committed shortly).
llvm-svn: 260285
This test fails in the ninja-x64-msvc-RA-centos6 builder, so the
UNSUPPORTED: system-windows condition is insufficient. Removing it
for now; I will investigate how this can be fixed later.
llvm-svn: 260281
We introduced gc.relocates of vector-of-pointer types a couple of weeks back. Somehow, I missed updating the InstCombine rule to account for this. If we hit this code path with a vector-of-pointers gc.relocate, we'd crash on a cast<PointerType>.
I also took the chance to do a bit of code style cleanup.
llvm-svn: 260279
that type from the global scope.
Patch by Sterling Augustine, derived (with permission) from code from Cling by
Vassil Vassilev and Philippe Canal.
llvm-svn: 260278
Summary:
Fix case where a pre-inc/dec load/store would not be formed if the
add/sub that forms the inc/dec part of the operation was the first
instruction in the block being examined.
Reviewers: mcrosier, jmolloy, t.p.northover, junbuml
Subscribers: aemerson, rengolin, mcrosier, llvm-commits
Differential Revision: http://reviews.llvm.org/D16785
llvm-svn: 260275
The non lazy atoms generated in the stubs pass use an image cache to
hold all of the pointers. On arm archs, this is the __got section,
but on x86 archs it should be __nl_symbol_ptr.
rdar://problem/24572729
llvm-svn: 260271
This is the third time it has crossed the 2^16 section limit. We've
already spent time optimizing this file to reduce template
instantiations, and it's not clear that there is anymore low hanging
fruit.
llvm-svn: 260267
This builds on the support being added to LLVM to import and export
registries from DLLs. This will allow us to pick up the registry
entries added in the DLL's copy of FrontendPluginRegistry.
This will allow us to use plugins on Windows using:
$ clang-cl -Xclang -load -Xclang plugin.dll \
-Xclang -add-plugin -Xclang foo
llvm-svn: 260265
Summary:
This patch adds Windows support for a few of the llvm-config commands,
including cflags, ldflags, libs, and system-libs.
Currently llvm-config is untested, so this patch adds tests for the
commands that it fixes as well.
Reviewers: rnk
Subscribers: llvm-commits
Differential Revision: http://reviews.llvm.org/D16762
llvm-svn: 260263
On Windows, the DLL containing the registry will get its own global head
and tail variables, so the entries registered in the DLL will be
invisible to the consumer.
In order to solve this, we need to export a getter function from the
plugin DLL per registry and copy over the data inside it. This patch
adds support for this. This will be used to support clang plugins on
Windows.
llvm-svn: 260261
Richard Smith