When ARMConstantIslandPass cannot find any good locations (i.e., "water") to
place constants, it falls back to inserting unconditional branches to make a
place to put them. My recent change exposed a problem in this area. We may
sometimes append to the same block more than one unconditional branch. The
symptoms of this are that the generated assembly has a branch to an undefined
label and running llc with -debug will cause a seg fault.
This happens more easily since my change to prevent CPEs from moving from
lower to higher addresses as the algorithm iterates, but it could have
happened before. The end of the block may be in range for various constant
pool references, but the insertion point for new CPEs is not right at the end
of the block -- it is at the end of the CPEs that have already been placed
at the end of the block. The insertion point could be out of range. When
that happens, the fallback code will always append another unconditional
branch if the end of the block is in range.
The fix is to only append an unconditional branch if the block does not
already end with one. I also removed a check to see if the constant pool load
instruction is at the end of the block, since that is redundant with
checking if the end of the block is in-range.
There is more to be done here, but I think this fixes the immediate problem.
llvm-svn: 84172
don't bother every time going around the main worklist. This speeds up a
release-asserts opt -std-compile-opts on 403.gcc by about 4% (1.5s). It
seems to speed up the most expensive instances of instcombine by ~10%.
llvm-svn: 84171
instruction (which disqualifies stores, unreachable, etc) and at least the
first operand is a constant. This filters out a lot of obvious cases that
can't be folded. Also, switch the IRBuilder to a TargetFolder, which tries
harder.
llvm-svn: 84170
most of the unsafe boilerplate out of TypeLoc. Create a QualifiedLoc class
to represent the idea that we *might* start representing source locations
of qualifiers. Dealing with qualifiers explicitly like this also lets us
efficiently ignore them in all the concrete cases.
This should make it obvious and easy to add new TypeLoc subclasses.
llvm-svn: 84168
RegionStoreManager::Retrieve() that was intended to handle conflated uses of pointers as integers.
It turns out this isn't needed, and resulted in inconsistent behavior when creating symbolic values on the following test case in 'tests/Analysis/misc-ps.m':
typedef struct _BStruct { void *grue; } BStruct;
void testB_aux(void *ptr);
void testB(BStruct *b) {
{
int *__gruep__ = ((int *)&((b)->grue));
int __gruev__ = *__gruep__;
testB_aux(__gruep__);
}
{
int *__gruep__ = ((int *)&((b)->grue));
int __gruev__ = *__gruep__;
if (~0 != __gruev__) {}
}
}
When the code was analyzed with '-arch x86_64', the value assigned to '__gruev__' be would be a
symbolic integer, but for '-arch i386' the value assigned to '__gruev__' would be a symbolic region
(a blob of memory). With this change the value created is always a symbolic integer.
Since the code being removed was added to support analysis of code calling
OSAtomicCompareAndSwapXXX(), I also modified 'test/Analysis/NSString.m' to analyze the code in both
'-arch i386' and '-arch x86_64', and also added some complementary test cases to test the presence
of leaks when using OSAtomicCompareAndSwap32Barrier()/OSAtomicCompareAndSwap64Barrier() instead of
just their absence. This code change reveals that previously both RegionStore and BasicStore were
handling these cases wrong, and would never cause the analyzer to emit a leak in these cases (false
negatives). Now RegionStore gets it right, but BasicStore still gets it wrong (and hence it has been
disabled temporarily for this test case).
llvm-svn: 84163
struct A { };
struct B : A { };
void f() {
const A& a = B();
}
correctly. (This now does the offset conversion if necessary and calls the destructor when a goes out of scope).
llvm-svn: 84162
header is just the entry block to the loop, and it needn't be at
the top of the loop in the code layout.
Remove the code that suppressed loop alignment for outer loops,
so that outer loops are aligned.
llvm-svn: 84158
the virtual base so they can be reused properly. Don't reuse vcall
slots across a virtual boundary. WIP. I have a testcase, but there
are still things that need to be fixed before the testcase can go in.
llvm-svn: 84120
template, make sure to get the template that corresponds to *this*
declaration of the class template or specialization, rather than the
canonical specialization. Fixes PR5187.
llvm-svn: 84119
to a member operator template. We missed updating this call site when
adding support for function templates; bug exposed by a test for
PR5072.
llvm-svn: 84111
Bob Wilson