This change was a lot bigger than I originally anticipated; among
other things it requires us storing more information in the CFG to
record what block-level expressions need to be evaluated as lvalues.
The big change is that CFGBlocks no longer contain Stmt*'s by
CFGElements. Currently CFGElements just wrap Stmt*, but they also
store a bit indicating whether the block-level expression should be
evalauted as an lvalue. DeclStmts involving the initialization of a
reference require us treating the initialization expression as an
lvalue, even though that information isn't recorded in the AST.
Conceptually this change isn't that complicated, but it required
bubbling up the data through the CFGBuilder, to GRCoreEngine, and
eventually to GRExprEngine.
The addition of CFGElement is also useful for when we want to handle
more control-flow constructs or other data we want to keep in the CFG
that isn't represented well with just a block of statements.
In GRExprEngine, this patch introduces logic for evaluating the
lvalues of references, which currently retrieves the internal "pointer
value" that the reference represents. EvalLoad does a two stage load
to catch null dereferences involving an invalid reference (although
this could possibly be caught earlier during the initialization of a
reference).
Symbols are currently symbolicated using the reference type, instead
of a pointer type, and special handling is required creating
ElementRegions that layer on SymbolicRegions (see the changes to
RegionStoreManager).
Along the way, the DeadStoresChecker also silences warnings involving
dead stores to references. This was the original change I introduced
(which I wrote test cases for) that I realized caused GRExprEngine to
crash.
llvm-svn: 91501
than using its own partial implementation of initialization.
Switched CheckInitializerTypes over to
InitializedEntity/InitializationKind, to help move us closer to
InitializationSequence.
Added InitializedEntity::getName() to retrieve the name of the entity,
for diagnostics that care about such things.
Implemented support for default initialization in
InitializationSequence.
Clean up the determination of the "source expressions" for an
initialization sequence in InitializationSequence::Perform.
Taught CXXConstructExpr to store more location information.
llvm-svn: 91492
in local register allocator. If a reg-reg copy has a phys reg
input and a virt reg output, and this is the last use of the phys
reg, assign the phys reg to the virt reg. If a reg-reg copy has
a phys reg output and we need to reload its spilled input, reload
it directly into the phys reg than passing it through another reg.
Following 76208, there is sometimes no dependency between the def of
a phys reg and its use; this creates a window where that phys reg
can be used for spilling (this is true in linear scan also). This
is bad and needs to be fixed a better way, although 76208 works too
well in practice to be reverted. However, there should normally be
no spilling within inline asm blocks. The patch here goes a long way
towards making this actually be true.
llvm-svn: 91485
remove start/finishGVStub and the BufferState helper class from the
MachineCodeEmitter interface. It has the side-effect of not setting the
indirect global writable and then executable on ARM, but that shouldn't be
necessary.
llvm-svn: 91464
found last time. Instead of trying to modify the IR while iterating over it,
I've change it to keep a list of WeakVH references to dead instructions, and
then delete those instructions later. I also added some special case code to
detect and handle the situation when both operands of a memcpy intrinsic are
referencing the same alloca.
llvm-svn: 91459
non-existing 'isa' field of a non-existing struct type
all related to legacy type definition for 'id' which we have
dropped in clang in favor of a built-in type.
(fixes radar 7470820).
llvm-svn: 91455
on PR5610 (2.185 -> 2.130s). The big issue is that this is making
insanely huge macro argument lists with over a million tokens in it.
The reason that mallco and free are so expensive is that we are
actually going to the kernel to get it, and switching to a bump
pointer allocator won't change this in an interesting way.
llvm-svn: 91449
- This is designed to make it obvious that %clang_cc1 is a "test variable"
which is substituted. It is '%clang_cc1' instead of '%clang -cc1' because it
can be useful to redefine what gets run as 'clang -cc1' (for example, to set
a default target).
llvm-svn: 91446
For hi/odd of an odd-length vector, the last component is undefined. Since
we shuffle with an undef vector, no CodeGen needs to change to support this.
llvm-svn: 91437
1. Use std::equal instead of reinventing it.
2. don't run dtors in destroy_range if element is pod-like.
3. Use isPodLike to decide between memcpy/uninitialized_copy
instead of is_class. isPodLike is more generous in some cases.
llvm-svn: 91427
Ted Kremenek