so the lookup works as expected after prepending the oso-prepend-path.
This manifested only on Windows, because "/" is not a relative path there.
llvm-svn: 248423
Patch by: simoncook
Unlike BitCasts, AddrSpaceCasts do not always produce an output the same size as its input, which was previously assumed. This fixes cases where two address spaces do not have the same size pointer, as an assertion failure would occur when trying to prove deferenceability. LoopUnswitch is used in the particular test, but LICM also exhibits the same problem.
Differential Revision: http://reviews.llvm.org/D13008
llvm-svn: 248422
This patch changes the order of GEPs generated by Splitting GEPs
pass, specially when one of the GEPs has constant and the base is
loop invariant, then we will generate the GEP with constant first
when beneficial, to expose more cases for LICM.
If originally Splitting GEP generate the following:
do.body.i:
%idxprom.i = sext i32 %shr.i to i64
%2 = bitcast %typeD* %s to i8*
%3 = shl i64 %idxprom.i, 2
%uglygep = getelementptr i8, i8* %2, i64 %3
%uglygep7 = getelementptr i8, i8* %uglygep, i64 1032
...
Now it genereates:
do.body.i:
%idxprom.i = sext i32 %shr.i to i64
%2 = bitcast %typeD* %s to i8*
%3 = shl i64 %idxprom.i, 2
%uglygep = getelementptr i8, i8* %2, i64 1032
%uglygep7 = getelementptr i8, i8* %uglygep, i64 %3
...
For no-loop cases, the original way of generating GEPs seems to
expose more CSE cases, so we don't change the logic for no-loop
cases, and only limit our change to the specific case we are
interested in.
llvm-svn: 248420
Previously we would use a non-const loop variable in the range-based
loop for:
void f(const std::vector<int> &v) {
for (size_t i = 0; i < v.size(); ++i) {
Now we use const auto&.
Note that we'll also want to use a copy at least for simple types.
llvm-svn: 248418
Note: I'm am not trying to describe what "should be"; I'm only describing what is true today.
This came out of my recent question to llvm-dev titled: When can the dominator tree not contain a node for a basic block?
Differential Revision: http://reviews.llvm.org/D13078
llvm-svn: 248417
And remove the switch default, so that the -Wcovered-switch-default
warning will catch new types next time they're added.
Differential Revision: http://reviews.llvm.org/D13096
llvm-svn: 248414
This test used fail intermittently, but now passes consistently on
FreeBSD in local runs. We'll investigate further if it's intermittent
on the FreeBSD buildbot, once it's restored.
llvm.org/pr15039
llvm-svn: 248410
Added new option --cuda-path=<path> which allows
overriding default search paths.
If it's not specified we look for CUDA installation in
/usr/include/cuda and /usr/include/cuda-7.0.
Differential Revision: http://reviews.llvm.org/D12989
llvm-svn: 248408
This updates the Reference.pdf files to say LLVM OpenMP Runtime Library and
also updates the build documentation to show how to build with CMake.
llvm-svn: 248407
Add two new ways of accessing the unsafe stack pointer:
* At a fixed offset from the thread TLS base. This is very similar to
StackProtector cookies, but we plan to extend it to other backends
(ARM in particular) soon. Bionic-side implementation here:
https://android-review.googlesource.com/170988.
* Via a function call, as a fallback for platforms that provide
neither a fixed TLS slot, nor a reasonable TLS implementation (i.e.
not emutls).
This is a re-commit of a change in r248357 that was reverted in
r248358.
llvm-svn: 248405
The BEXTR comments didn't make sense before, we may want to extend the
FP logic transform to work on vectors, and this way is more beautiful.
llvm-svn: 248404
Summary:
This is the first part of fixing bug 24848 https://llvm.org/bugs/show_bug.cgi?id=24848.
When range metadata is provided, it should be used to constant fold comparisons with constant values.
Reviewers: sanjoy, hfinkel
Subscribers: llvm-commits
Differential Revision: http://reviews.llvm.org/D12988
llvm-svn: 248402
Summary:
With this change DWARFASTParserClang::CompleteTypeFromDWARF returns false if
DWARFASTParserClang::ParseChildMembers returns false. Similarly, it returns
false if any base class is of an incomplete type. This helps in cases like
these:
class Foo
{
public:
std::string str;
};
...
Foo f;
If a file with the above code is compiled with a modern clang but without
the -fno-limit-debug-info (or similar) option, then the DWARF has only
a forward declration for std::string. In which case, the type for
"class Foo" cannot be completed. If LLDB does not detect that a child
member has incomplete type, then it wrongly conveys to clang (the LLDB
compiler) that "class Foo" is complete, and consequently crashes due to
an assertion failure in clang when running commands like "p f" or
"frame var f".
Reviewers: clayborg
Subscribers: lldb-commits
Differential Revision: http://reviews.llvm.org/D13066
llvm-svn: 248401
Adjust __global__ functions with DiscardableODR linkage to use
StrongODR linkage instead, so they are visible externally.
Differential Revision: http://reviews.llvm.org/D13067
llvm-svn: 248400
This patch extends llvm-dsymutil's ODR type uniquing machinery to also
resolve forward decls for types defined in clang modules.
http://reviews.llvm.org/D13038
llvm-svn: 248398
This reverts r248388 and fixes the underlying bug: hasAddr64 was initialized
in runOnMachineFunction, but runOnMachineFunction isn't ever called in
CodeGen/WebAssembly/global.ll since that testcase has no functions. The fix
here is to use AsmPrinter's getPointerSize() as needed to determine the
pointer size instead.
llvm-svn: 248394
Our string literal parser copied any source-file new-line characters
into the execution string-literal. This is incorrect if the source-file
new-line character was a \r\n sequence because new-line characters are
merely \n.
llvm-svn: 248392
We took both source locations from the end of the initializer list what
the code below doesn't expect. This can lead to a crash when rendering
the diagnostic (PR24816). Assert that we have more than one element in
a scalar initializer with too many elements.
llvm-svn: 248391
This changes the behavior of AddAligntmentAssumptions to match its
comment. I.e, prove the asserted alignment in the context of the caller,
not the callee.
Thanks to Mehdi Amini for seeing the issue here! Also to Artur Pilipenko
who also saw a fix for the issue.
rdar://22521387
Differential Revision: http://reviews.llvm.org/D12997
llvm-svn: 248390
Invoking a function which returns an aggregate can sometimes be
transformed to return a scalar value. However, this means that we need
to create an insertvalue instruction(s) to recreate the correct
aggregate type. We achieved this by inserting an insertvalue
instruction at the invoke's normal successor. However, this is not
feasible if the normal successor uses the invoke's return value inside a
PHI node.
Instead, split the edge between the invoke and the unwind successor and
create the insertvalue instruction in the new basic block. The new
basic block's successor will be the old invoke successor which leaves
us with IR which is well behaved.
This fixes PR24906.
llvm-svn: 248387
There are lingering issues building the atomic builtins with various versions of gcc. To unblock people we can only include them on Apple platforms where they are more tested.
llvm-svn: 248386
* --results-formatter-options renamed to --results-formatter-option,
with short version of -O
* Multiple --results-formatter-option=OPTION can be specified. The
comma-separating mechanism has been removed.
* XunitFormatter options modified: -n and -r are now short forms of
--ignore-skip-name and --ignore-skip-reason. Those long option
names were tweaked lightly. They also can be specified multiple
times on the command line. The comma-separating, multiple-pattern-
per-option mechanism has been removed.
One can now specify:
dotest.py --results-file stdout -O-ndsym -O-nlldb-mi
for example, to ignore reporting skips for dsym-related or lldb-mi-related
tests in the xUnit report.
llvm-svn: 248384
Summary:
Building the builtins on Darwin platforms is a bit complicated. This is a first-pass implementation of the functionality from clang_darwin.mk into CMake.
When building the builtins on Darwin we have layers of blacklists that we apply based on platform, architecture, and minimum supported OS version.
Reviewers: bogner, filcab, bob.wilson, samsonov
Subscribers: llvm-commits
Differential Revision: http://reviews.llvm.org/D13059
llvm-svn: 248383
This translates to -load name.so in the cc1 command. We can't name the driver
option -load, as that means "link against oad", so instead we follow GCC's lead
and name the option -fplugin.
llvm-svn: 248378
Joerg Sonnenberger