Previously, instructions without a primary patterns wouldn't get their
properties inferred. Now, we use all single-instruction patterns for
inference, including 'def : Pat<>' instances.
This causes a lot of instruction flags to change.
- Many instructions no longer have the UnmodeledSideEffects flag because
their flags are now inferred from a pattern.
- Instructions with intrinsics will get a mayStore flag if they already
have UnmodeledSideEffects and a mayLoad flag if they already have
mayStore. This is because intrinsics properties are linear.
- Instructions with atomic_load patterns get a mayStore flag because
atomic loads can't be reordered. The correct workaround is to create
pseudo-instructions instead of using normal loads. PR13693.
llvm-svn: 162614
Loop Converter Skeleton - array-step-1
Added a check to loop increments - array-step-2b
Added a check on the loop's condition expression - array-step-2c
Finished array matcher - array-step-2
Retrieved matched nodes - array-step-3
Analysis for array loop indices - array-step-4
Added checking for naming and variable scope
Added confidence level and count-only command line args
Added aliased variable elision
Added support for iterator-based loops
Added support for single-iterator loops which call end() repeatedly
Added support for converting array-like containers
llvm-svn: 162610
It's not clear that they should be marked as such, but tbb formation
fails if t2LEApcrelJT is hoisted of of a loop.
This doesn't change the flags on these instructions,
UnmodeledSideEffects was already inferred from the missing pattern.
llvm-svn: 162603
CodeGen option to a LangOpt option. In turn, hoist the guard into the parser
so that we avoid the new (and fairly unstable) Sema/AST/CodeGen logic. This
should restore the behavior of clang to that prior to r158325.
<rdar://problem/12163681>
llvm-svn: 162602
Remaining characters should be discarded once sync() called. If don’t, garbage
characters can be inserted to the front of external buffer in underflow().
Because underflow() copies remaining characters in external buffer to it’s
front. This results wrong characters insertion when seekpos() or seekoff() is
called.
this line should be inserted in sync() just before return:
__extbufnext_ = __extbufend_ = __extbuf_;
2. sync() should use length() rather than out() to calculate offset.
Reversing iterators and calling out() to calculate offset from behind is
working fine in stateless character encoding. However, in stateful encoding,
escape sequences could differ in length. As a result, out() could return wrong
length.
For example, if we have internal buffer converted from this external sequence:
(capital letters mean escape sequence)
… a a a a B b b b b
out() produces this sequence.
b b b b A a a a a
Because out() inserts escape sequence A rather than B, result sequence doesn't
match to external sequence. A and B could have different lengths, result offset
could be wrong value too.
length() method in codecvt is right for calculating offset, but it counts
offset from the beginning of buffer. So it requires another state member
variable to hold state before conversion.
Fixes http://llvm.org/bugs/show_bug.cgi?id=13667
llvm-svn: 162601
Instructions are now only marked as variadic if they use variable_ops in
their ins list.
A variadic SDNode is typically used for call nodes that have the call
arguments as operands.
A variadic MachineInstr can actually encode a variable number of
operands, for example ARM's stm/ldm instructions. A call instruction
does not have to be variadic. The call argument registers are added as
implicit operands.
This change remove the MCID::Variadic flags from most call and return
instructions, allowing us to better verify their operands.
llvm-svn: 162599
The ARM BL and BLX instructions don't have predicate operands, but the
thumb variants tBL and tBLX do.
The argument registers should be added as implicit uses.
llvm-svn: 162593
generated for a given diagnostic to another. Because PathDiagnostics
are specific to a give PathDiagnosticConsumer, store in
a FoldingSet a unique hash for a PathDiagnostic (that will be the same
for the same bug for different PathDiagnosticConsumers) that
stores a list of files generated. This can then be read by the
other PathDiagnosticConsumers.
This fixes breakage in the PLIST-HTML output.
llvm-svn: 162580
ProfileDataTypes.h header.
With this patch the old and new profiling code can exist side-by-side. The new
profiling code will be submitted soon and it only supports insert-edge-profiling
for now and will not depend on ProfileInfo.
Patch by Alastair Murray.
llvm-svn: 162576
Add 'attach <pid>|<process-name>' command to lldb, as well as 'detach' which is an alias of 'process detach'.
Add two completion test cases for "attach" and "detach".
llvm-svn: 162573
the case of multiple edges from one block to another.
A simple example is a switch statement with multiple values to the same
destination. The definition of an edge is modified from a pair of blocks to
a pair of PredBlock and an index into the successors.
Also set the weight correctly when building SelectionDAG from LLVM IR,
especially when converting a Switch.
IntegersSubsetMapping is updated to calculate the weight for each cluster.
llvm-svn: 162572
It is now allowed to explicitly set hasSideEffects, mayStore, and
mayLoad on instructions with patterns.
Verify that the patterns are consistent with the explicit flags.
llvm-svn: 162569
More generally, any time we try to track where a null value came from, we
should show if it came from a function. This usually isn't necessary if
the value is symbolic, but if the value is just a constant we previously
just ignored its origin entirely. Now, we'll step into the function and
recursively add a visitor to the returned expression.
<rdar://problem/12114609>
llvm-svn: 162563
No intended behavior change. This was introduced in r162023. With the fixed
algorithm a Release build of ARMInstPrinter.cpp goes from 16s to 10s on a
2011 MBP.
llvm-svn: 162559
Jakob Stoklund Olesen