This patch reduces the stack memory consumption of the InstCombine
function "isOnlyCopiedFromConstantGlobal() ", that in certain conditions
could overflow the stack because of excessive recursiveness.
For example, in a case like this:
%0 = alloca [50025 x i32], align 4
%1 = getelementptr inbounds [50025 x i32]* %0, i64 0, i64 0
store i32 0, i32* %1
%2 = getelementptr inbounds i32* %1, i64 1
store i32 1, i32* %2
%3 = getelementptr inbounds i32* %2, i64 1
store i32 2, i32* %3
%4 = getelementptr inbounds i32* %3, i64 1
store i32 3, i32* %4
%5 = getelementptr inbounds i32* %4, i64 1
store i32 4, i32* %5
%6 = getelementptr inbounds i32* %5, i64 1
store i32 5, i32* %6
...
This piece of code crashes llvm when trying to apply instcombine on
desktop. On embedded devices this could happen with a much lower limit
of recursiveness. Some instructions (getelementptr and bitcasts) make
the function recursively call itself on their uses, which is what makes
the example above consume so much stack (it becomes a recursive
depth-first tree visit with a very big depth).
The patch changes the algorithm to be semantically equivalent, but
iterative instead of recursive and the visiting order to be from a
depth-first visit to a breadth-first visit (visit all the instructions
of the current level before the ones of the next one).
Now if a lot of memory is required a heap allocation is done instead of
the the stack allocation, avoiding the possible crash.
Reviewed By: rnk
Differential Revision: http://reviews.llvm.org/D4355
Patch by Marcello Maggioni! We don't generally commit large stress test
that look for out of memory conditions, so I didn't request that one be
added to the patch.
llvm-svn: 212133
This macro is sometimes defined manually but isn't (and doesn't need to be) in
llvm-config.h so shouldn't appear in the headers, likewise NDEBUG.
Instead switch them over to LLVM_DUMP_METHOD on the definitions.
llvm-svn: 212130
Matching behavior with DeadArgumentElimination (and leveraging some
now-common infrastructure), keep track of the function from debug info
metadata if arguments are promoted.
This may produce interesting debug info - since the arguments may be
missing or of different types... but at least backtraces, inlining, etc,
will be correct.
llvm-svn: 212128
This change is particularly useful in the code generation as we need
to know which binary operator/identity element we need to combine/initialize
the privatization locations.
+ Print the reduction type for each memory access
+ Adjusted the test cases to comply with the new output format and
to test for the right reduction type
llvm-svn: 212126
The pointer for a class's RTTI data comes right before the VFTable but
has no name. To be properly compatible with this, we do the following:
* Create a single GlobalVariable which holds the contents of the VFTable
_and_ the pointer to the RTTI data.
* Create a GlobalAlias, with appropriate linkage/visibility, that points
just after the RTTI data pointer. This ensures that the VFTable
symbol will always refer to VFTable data.
* Create a Comdat with a "Largest" SelectionKind and stick the private
GlobalVariable in it. By transitivity, the GlobalAlias will be a
member of the Comdat group. Using "Largest" ensures that foreign
definitions without an RTTI data pointer will _not_ be chosen in the
final linked image.
Whether or not we emit RTTI data depends on several things:
* The -fno-rtti flag implies that we should never not emit a pointer to
RTTI data before the VFTable.
* __declspec(dllimport) brings in the VFTable from a remote DLL. Use an
available_externally GlobalVariable to provide a local definition of
the VFTable. This means that we won't have any available_externally
definitions of things like complete object locators. This is
acceptable because they are never directly referenced.
To my knowledge, this completes the implementation of MSVC RTTI code
generation.
Further semantic work should be done to properly support /GR-.
llvm-svn: 212125
The logic for expanding atomics that aren't natively supported in
terms of cmpxchg loops is much simpler to express at the IR level. It
also allows the normal optimisations and CodeGen improvements to help
out with atomics, instead of using a limited set of possible
instructions..
rdar://problem/13496295
llvm-svn: 212119
These flags enable behavior in MSVC that Clang has by default.
/Zc:inline essentially marks all COMDATs as discardable. In LLVM
parlance, this means using linkonce_odr linkage, which is what we
already do.
llvm-svn: 212117
For now I only updated the _alt variants. The main variants are used by
codegen and that will need a bit more work to trigger.
<rdar://problem/17492620>
llvm-svn: 212114
Adding a writemask variant would require a third asm string to be passed to
the template. Generate the AsmString in the template instead.
No change in X86.td.expanded.
llvm-svn: 212113
The former used to crash with a null deref if it was given a not owned pointer,
while the latter returned 0. Now they both return 0. This is still not the best possible
behavior: it is better to print an error report with a stack trace, pointing
to the error in user code, as we do in ASan.
llvm-svn: 212112
- Ported the SWIG wrapper shell scripts to Python so that they would work on Windows too along with other platforms
- Updated CMake handling to fix SWIG errors and manage sym-linking on Windows to liblldb.dll
- More build fixes for Windows
The pending issues are that two Python modules, termios and pexpect are not available on Windows.
These are currently required for the Python command interpreter to be used from within LLDB.
llvm-svn: 212111
Currently, we fail with an error.
Reviewers: rafael
Reviewed By: rafael
Subscribers: rnk, cfe-commits
Differential Revision: http://reviews.llvm.org/D4347
llvm-svn: 212110
to select the slice out of a Mach-O universal file. This also includes
support for -arch all, selecting the host architecture by default from
a universal file and checking if -arch is used with a standard Mach-O
it matches that architecture.
llvm-svn: 212108
There were a few places where we were not catching the possibility of negative
error codes in waitpid() calls. This change fixes those remaining after
the llgs branch fixes to ProcessMonitor.
Change by Shawn Best.
llvm-svn: 212107
There were transforms whose *intent* was to downgrade the linkage of
external objects to have internal linkage.
However, it fired on things with private linkage as well.
llvm-svn: 212104
This is a small targeted fix for pr20119. The code needs quiet a bit of
refactoring and I added some FIXMEs about it, but I want to get the testcase
passing first.
llvm-svn: 212101
copies.
This patch extends the peephole optimization introduced in r190713 to produce
register-coalescer friendly copies when possible.
This extension taught the existing cross-bank copy optimization how to deal
with the instructions that generate cross-bank copies, i.e., insert_subreg,
extract_subreg, reg_sequence, and subreg_to_reg.
E.g.
b = insert_subreg e, A, sub0 <-- cross-bank copy
...
C = copy b.sub0 <-- cross-bank copy
Would produce the following code:
b = insert_subreg e, A, sub0 <-- cross-bank copy
...
C = copy A <-- same-bank copy
This patch also introduces a new helper class for that: ValueTracker.
This class implements the logic to look through the copy related instructions
and get the related source.
For now, the advanced rewriting is disabled by default as we are lacking the
semantic on target specific instructions to catch the motivating examples.
Related to <rdar://problem/12702965>.
llvm-svn: 212100
By default, no functionality change.
Before evicting a local variable, this heuristic tries to find another (set of)
local(s) that can be reassigned to a free color.
In some extreme cases (large basic blocks with tons of local variables), the
compilation time is dominated by the local interference checks that this
heuristic must perform, with no code gen gain.
E.g., the motivating example takes 4 minutes to compile with this heuristic, 12
seconds without.
Improving the situation will likely require to make drastic changes to the
register allocator and/or the interference check framework.
For now, provide this flag to better understand the impact of that heuristic.
<rdar://problem/17444599>
llvm-svn: 212099
This patch removes the dead code, and refines the
getEHResumeBlock() slightly.
The CleanupHackLevel was a hack to the old exception
handling intrinsics, which have several issues with function
inliner.
Since LLVM 3.0, the new landingpad and resume instructions
are added to LLVM IR. With the new exception handling
mechanism, most of the issues are fixed now. We should
always use these instructions to implement the exception
handling code nowadays, and we don't need the hack any more.
Besides, the `CleanupHackLevel` is a compile-time constant,
thus other cases have been considered as dead code for a while.
llvm-svn: 212097
command line option only. Internally we convert them to the "o32" and "n64"
respectively. So we do not need to refer them anywhere after that conversion.
No functional changes.
llvm-svn: 212096
In 32b mode the reference count for block addresses
is not zero. This prevents inlining and constant
folding and causes the test to fail. Changing
the triple allows runnning the test in 64b mode.
The array in foo2 is now local instead of static until
at lower optimization levels the interprocedural constant
propagator is invoked before the global optimizer.
llvm-svn: 212092
llvm r212077 causes this test to fail. We need to reorder some passes and
possibly make other changes to reenable the optimization being tested here.
llvm-svn: 212091
This reverts commit r212085.
This breaks the sanitizer bot... & I thought I'd tried pretty hard not
to do that. Guess I need to try harder.
llvm-svn: 212089
Update the strategy in r212083 to try JIT first and otherwise fall back to the
interpreter. This gives the best of both worlds and still builds fine with no
targets enabled.
Requires supporting changes from LLVM r212086.
llvm-svn: 212087
ForceInterpreter=false shouldn't disable the interpreter completely because it
can still be necessary to interpret if the target doesn't support JIT.
No obvious way to test this in LLVM, but this matches what
LLVMCreateExecutionEngineForModule() does and fixes the clang-interpreter
example in the clang source tree which uses the ExecutionEngine.
llvm-svn: 212086
Originally committed in r211723, reverted in r211724 due to failure
cases found and fixed (ArgumentPromotion: r211872, Inlining: r212065),
and I now believe the invariant actually holds for some reasonable
amount of code (but I'll keep an eye on the buildbots and see what
happens... ).
Original commit message:
PR20038: DebugInfo: Inlined call sites where the caller has debug info
but the call itself has no debug location.
This situation does bad things when inlined, so I've fixed Clang not to
produce inlinable call sites without locations when the caller has debug
info (in the one case where I could find that this occurred). This
updates the PR20038 test case to be what clang now produces, and readds
the assertion that had to be removed due to this bug.
I've also beefed up the debug info verifier to help diagnose these
issues in the future, and I hope to add checks to the inliner to just
assert-fail if it encounters this situation. If, in the future, we
decide we have to cope with this situation, the right thing to do is
probably to just remove all the DebugLocs from the inlined instructions.
llvm-svn: 212085
Fix the build when no targets are enabled. This dependency is incurred by two
unfortunate entries in LinkAllPasses.h included from cc1_main.cpp:
llvm::createJumpInstrTablesPass();
llvm::createCodeGenPreparePass();
llvm-svn: 212084
Fixes the build when no targets are selected, or no native target is built.
This also better matches up with the description/title of the example and
demonstrates how clang can be used to run C++ on constrained environments
without file IO or executable memory permissions (e.g. iOS apps).
A comment is added explaining how to extend the demo with JIT support as
needed.
llvm-svn: 212083
Reid Kleckner