When there are multiple tied use-def pairs on an inline asm instruction,
the tied uses must appear in the same order as the defs.
It is possible to write an LLVM IR inline asm instruction that breaks
this constraint, but there is no reason for a front end to emit the
operands out of order.
The gnu inline asm syntax specifies tied operands as a single read/write
constraint "+r", so ouf of order operands are not possible.
llvm-svn: 162878
inlined function.
This resolves retain count checker false positives that are caused by
inlining ObjC and other methods. Essentially, if we are passing an
object to a method with "delegate" in the selector or a function pointer
as another argument, we should stop tracking the other parameters/return
value as far as the retain count checker is concerned.
llvm-svn: 162876
Tombstones and full hash collisions are rare, mark the "empty"
and "no collision" paths as likely. The bug in simplifycfg
that prevented the hints from being picked during selfhost
up was fixed recently :)
llvm-svn: 162874
For normal instructions, isTied() is set automatically by addOperand(),
based on MCInstrDesc, but inline asm has tied operands outside the
descriptor.
llvm-svn: 162869
This heuristic addresses the case when a pointer (or ref) is passed
to a function, which initializes the variable (or sets it to something
other than '0'). On the branch where the inlined function does not
set the value, we report use of undefined value (or NULL pointer
dereference). The access happens in the caller and the path
through the callee would get pruned away with regular path pruning. To
solve this issue, we previously disabled diagnostic pruning completely
on undefined and null pointer dereference checks, which entailed very
verbose diagnostics in most cases. Furthermore, not all of the
undef value checks had the diagnostic pruning disabled.
This patch implements the following heuristic: if we pass a pointer (or
ref) to the region (on which the error is reported) into a function and
it's value is either undef or 'NULL' (and is a pointer), do not prune
the function.
llvm-svn: 162863
Ordered memory operations are more constrained than volatile loads and
stores because they must be ordered with respect to all other memory
operations.
llvm-svn: 162861
Make breakpoint setting by file and line much more efficient by only looking for inlined breakpoint locations if we are setting a breakpoint in anything but a source implementation file. Implementing this complex for a many reasons. Turns out that parsing compile units lazily had some issues with respect to how we need to do things with DWARF in .o files. So the fixes in the checkin for this makes these changes:
- Add a new setting called "target.inline-breakpoint-strategy" which can be set to "never", "always", or "headers". "never" will never try and set any inlined breakpoints (fastest). "always" always looks for inlined breakpoint locations (slowest, but most accurate). "headers", which is the default setting, will only look for inlined breakpoint locations if the breakpoint is set in what are consudered to be header files, which is realy defined as "not in an implementation source file".
- modify the breakpoint setting by file and line to check the current "target.inline-breakpoint-strategy" setting and act accordingly
- Modify compile units to be able to get their language and other info lazily. This allows us to create compile units from the debug map and not have to fill all of the details in, and then lazily discover this information as we go on debuggging. This is needed to avoid parsing all .o files when setting breakpoints in implementation only files (no inlines). Otherwise we would need to parse the .o file, the object file (mach-o in our case) and the symbol file (DWARF in the object file) just to see what the compile unit was.
- modify the "SymbolFileDWARFDebugMap" to subclass lldb_private::Module so that the virtual "GetObjectFile()" and "GetSymbolVendor()" functions can be intercepted when the .o file contenst are later lazilly needed. Prior to this fix, when we first instantiated the "SymbolFileDWARFDebugMap" class, we would also make modules, object files and symbol files for every .o file in the debug map because we needed to fix up the sections in the .o files with information that is in the executable debug map. Now we lazily do this in the DebugMapModule::GetObjectFile()
Cleaned up header includes a bit as well.
llvm-svn: 162860
This means the same as LoadInst/StoreInst::isUnordered(), and implies
!isVolatile().
Atomic loads and stored are also ordered, and this is the right method
to check if it is safe to reorder memory operations. Ordered atomics
can't be reordered wrt normal loads and stores, which is a stronger
constraint than volatile.
llvm-svn: 162859
It is technically allowed to move a normal load across a volatile load,
but probably not a good idea.
It is not allowed to move a load across an atomic load with
Ordering > Monotonic, and we model those with MOVolatile as well.
I recently removed the mayStore flag from atomic load instructions, so
they don't need a pseudo-opcode. This patch makes up for the difference.
llvm-svn: 162857
This lets the user run the program from a different directory and still have the
.gcda files show up in the correct place.
<rdar://problem/12179524>
llvm-svn: 162855
We need to reserve space for the mandatory traceback fields,
though leaving them as zero is appropriate for now.
Although the ABI calls for these fields to be filled in fully, no
compiler on Linux currently does this, and GDB does not read these
fields. GDB uses the first word of zeroes during exception handling to
find the end of the function and the size field, allowing it to compute
the beginning of the function. DWARF information is used for everything
else. We need the extra 8 bytes of pad so the size field is found in
the right place.
As a comparison, GCC fills in a few of the fields -- language, number
of saved registers -- but ignores the rest. IBM's proprietary OSes do
make use of the full traceback table facility.
Patch by Bill Schmidt.
llvm-svn: 162854
to define all macros for MIPS targets. Remove redundant virtual function
getArchDefines(). Two virtual functions for this task are really too much.
llvm-svn: 162853
CheckLValueConstantExpression.
Richard pointed out that using the address of a TLS variable is ok in a
core C++11 constant expression, as long as it isn't part of the eventual
result of constant expression evaluation. Having the check in
CheckLValueConstantExpression accomplishes this.
llvm-svn: 162850
The operands on an INLINEASM machine instruction are divided into groups
headed by immediate flag operands. Verify this structure.
Extract verifyTiedOperands(), and only call it for non-inlineasm
instructions.
llvm-svn: 162849
Summary:
The problem was with the following sequence:
#pragma push_macro("long")
#undef long
#pragma pop_macro("long")
in case when "long" didn't represent a macro.
Fixed crash and removed code duplication for #undef/pop_macro case. Added regression tests.
Reviewers: doug.gregor, klimek
Reviewed By: doug.gregor
CC: cfe-commits, chapuni
Differential Revision: http://llvm-reviews.chandlerc.com/D31
llvm-svn: 162845
This disables malloc-specific optimization when -fno-builtin (or -ffreestanding)
is specified. This has been a problem for a long time but became more severe
with the recent memory builtin improvements.
Since the memory builtin functions are used everywhere, this required passing
TLI in many places. This means that functions that now have an optional TLI
argument, like RecursivelyDeleteTriviallyDeadFunctions, won't remove dead
mallocs anymore if the TLI argument is missing. I've updated most passes to do
the right thing.
Fixes PR13694 and probably others.
llvm-svn: 162841
This makes Clang produce an error for code such as:
__thread int x;
int *p = &x;
The lvalue of a thread-local variable cannot be evaluated at compile
time.
llvm-svn: 162835