deterministic.
Commit message for r170826:
[analyzer] Traverse the Call Graph in topological order.
Modify the call graph by removing the parentless nodes. Instead all
nodes are children of root to ensure they are all reachable. Remove the
tracking of nodes that are "top level" or global. This information is
not used and can be obtained from the Decls stored inside
CallGraphNodes.
Instead of existing ordering hacks, analyze the functions in topological
order over the Call Graph.
Together with the addition of devirtualizable ObjC message sends and
blocks to the call graph, this gives around 6% performance improvement
on several large ObjC benchmarks.
llvm-svn: 170906
We used to not really format them. Now we do:
for (MachineBasicBlock::succ_iterator SI = BB->succ_begin(),
SE = BB->succ_end();
SI != SE; ++SI) {
This is just one example and I am sure we still mess some of them up, but it
is a step forward.
llvm-svn: 170899
CXXScalarValueInitExpr (or an ImplicitValueInitExpr), strip it back down to an
empty pair of parentheses so that the initialization code can tell that we're
performing value-initialization.
llvm-svn: 170867
This is just a minor bit of refactoring, but it is nice cleanup for
the subsequent patch that adds warning support for assigning literals
to weak variables.
llvm-svn: 170863
When we are visiting the extern declaration of 'i' in
static int i = 99;
int foo() {
extern int i;
return i;
}
We should not try to handle it as if it was an function static. That is, we
must consider the written storage class.
Fixing this then exposes that the assert in EmitGlobalVarDeclLValue and the
if leading to its call are not completely accurate. They were passing before
because the second decl was marked as having external storage. I changed them
to check the linkage, which I find easier to understand.
Last but not least, there is something strange going on with cuda and opencl.
My guess is that the linkage computation for these languages needs to be
audited, but I didn't want to change that in this patch so I just updated
the storage classes to keep the current behavior.
Thanks to Reed Kotler for reporting this.
llvm-svn: 170827
Modify the call graph by removing the parentless nodes. Instead all
nodes are children of root to ensure they are all reachable. Remove the
tracking of nodes that are "top level" or global. This information is
not used and can be obtained from the Decls stored inside
CallGraphNodes.
Instead of existing ordering hacks, analyze the functions in topological
order over the Call Graph.
Together with the addition of devirtualizable ObjC message sends and
blocks to the call graph, this gives around 6% performance improvement
on several large ObjC benchmarks.
llvm-svn: 170826
This paves the road for constructing a better function dependency graph.
If we analyze a function before the functions it calls and inlines,
there is more opportunity for optimization.
Note, we add call edges to the called methods that correspond to
function definitions (declarations with bodies).
llvm-svn: 170825
to also remove a trailing space if possible.
For example, removing '__bridge' from:
i = (__bridge I*)p;
should result in:
i = (I*)p;
not:
i = ( I*)p;
rdar://11314821
llvm-svn: 170764
Without this patch comparing two equal versions without patch numbers (4.7 for
example) will result in A < B and B < A.
Patch by Simon Atanasyan.
llvm-svn: 170705
produce a note for that diagnostic either with a different DiagnosticEngine or
after calling DiagnosticEngine::Reset(). That didn't make any sense, and did the
wrong thing if the original diagnostic was suppressed.
llvm-svn: 170636
Instead of using several callbacks to identify the pointer escape event,
checkers now can register for the checkPointerEscape.
Converted the Malloc checker to use the new callback.
SimpleStreamChecker will be converted next.
llvm-svn: 170625
fix any bad objectiveC syntax coming out of
DeclPrinter. This is on going. Also, introduce a new
PrintPolicy and use it as needed when declaration tag
is to be produced via DeclPrinter.
llvm-svn: 170606
modules when getting the decls for a namespace or translation unit.
Otherwise the code-completion results will not be complete.
rdar://12889089
llvm-svn: 170596
too). When instantiating a direct-initializer, if we find it has zero
arguments, produce an empty ParenListExpr rather than returning a null
expression.
llvm-svn: 170490
copy-list-initialization (and doesn't add an additional copy step):
Fill in the ListInitialization bit when creating a CXXConstructExpr. Use it
when instantiating initializers in order to correctly handle instantiation of
copy-list-initialization. Teach TreeTransform that function arguments are
initializations, and so need this special treatment too. Finally, remove some
hacks which were working around SubstInitializer's shortcomings.
llvm-svn: 170489
This does limit these typedefs to being sequences, but no current usage
requires them to be contiguous (we could expand this to a more general
iterator pair range concept at some point).
Also, it'd be nice if SmallVector were constructible directly from an ArrayRef
but this is a bit tricky since ArrayRef depends on SmallVectorBaseImpl for the
inverse conversion. (& generalizing over all range-like things, while nice,
would require some nontrivial SFINAE I haven't thought about yet)
llvm-svn: 170482
use clang's formatter. Currently, formatter is used
to format declaration tags for xml comments. Since formatter
is in flux and its change will break several of the clang comment
tests, only a single tests is formatted using this facility.
Doug has reviewed and approved it for check-in.
llvm-svn: 170467
We used to format initializers like this (with a sort of hacky implementation):
Constructor()
: Val1(A),
Val2(B) {
and now format like this (with a somewhat better solution):
Constructor()
: Val1(A), Val2(B) {
assuming this would not fit on a single line. Also added tests.
As a side effect we now first analyze whether an UnwrappedLine needs to be
split at all. If not, not splitting it is the best solution by definition. As
this should be a very common case in normal code, not exploring the entire
solution space can provide significant speedup.
llvm-svn: 170457
PR 14529 was opened because neither Clang or LLVM was expanding
calls to creal* or cimag* into instructions that just load the
respective complex field. After some discussion, it was not
considered realistic to do this in LLVM because of the platform
specific way complex types are expanded. Thus a way to solve
this in Clang was pursued. GCC does a similar expansion.
This patch adds the feature to Clang by making the creal* and
cimag* functions library builtins and modifying the builtin code
generator to look for the new builtin types.
llvm-svn: 170455
This fixes the storage class of extern decls that are merged with file level
statics. The patch also fixes the linkage computation so that they are
considered internal.
llvm-svn: 170406
This also requires adding support to -cc1as for passing the detecting
PWD down through LLVM's debug info (which in turn required the LLVM
change in r170371).
The test case is weak (we only test the driver behavior) because there
is currently to infrastructure for running cc1as in the test suite. So
those four lines are untested (much like all other lines in that file),
but we have a test for the same pattern using llvm-mc in the LLVM
repository.
llvm-svn: 170373
performance heuristic
After inlining a function with more than 13 basic blocks 32 times, we
are not going to inline it anymore. The idea is that inlining large
functions leads to drastic performance implications. Since the function
has already been inlined, we know that we've analyzed it in many
contexts.
The following metrics are used:
- Large function is a function with more than 13 basic blocks (we
should switch to another metric, like cyclomatic complexity)
- We consider that we've inlined a function many times if it's been
inlined 32 times. This number is configurable with -analyzer-config
max-times-inline-large=xx
This heuristic addresses a performance regression introduced with
inlining on one benchmark. The analyzer on this benchmark became 60
times slower with inlining turned on. The heuristic allows us to analyze
it in 24% of the time. The performance improvements on the other
benchmarks I've tested with are much lower - under 10%, which is
expected.
llvm-svn: 170361
avoided.
This required a minor modification of the memoization as now the
"CurrentPenalty" depends on whether or not we break before the current
token. Therefore, the CurrentPenalty should not be memoized but added
after retrieving a value from memory. This should not affect the runtime
behavior.
llvm-svn: 170337
More specifically:
- Improve formatting of static initializers.
- Fix formatting of lines comments in enums.
- Fix formmating of trailing line comments.
llvm-svn: 170316
incompatibility with how complex values are returned. It is sufficient
to flag all complex types as direct rather than indirect.
A new test case is provided that checks correct IR generation for the
various supported flavors of _Complex.
llvm-svn: 170302
of a member function with parenthesized declarator.
Like this test case:
class Foo {
const char *(baz)() {
return __PRETTY_FUNCTION__;
}
};
llvm-svn: 170233
Nico Weber