Generating a sink is significantly different behavior from generating a
normal node, and a simple boolean parameter can be rather opaque. Per
offline discussion with Anna, adding new generation methods is the
clearest way to communicate intent.
No functionality change.
llvm-svn: 162215
Forgetting to at least cast the result was giving us Loc/NonLoc problems
in SValBuilder (hitting an assertion). But the standard (both C and C++)
does actually guarantee that && and || will result in the actual values
1 and 0, typed as 'int' in C and 'bool' in C++, and we can easily model that.
PR13461
llvm-svn: 162209
- use InnerMatcher consistently, fix style violations on the way
- doxygenify code snippets across all comments
- start doxygenifying code references in text
- addeed missing Usable as: sections
llvm-svn: 162205
Without this patch, lib.clang_getNumCompletionChunks is called at
each _iteration_ of a 'for chunk in CompletionString' loop. Now we
call it just once.
llvm-svn: 162200
The DAGCombiner tries to optimise a BUILD_VECTOR by checking if it
consists purely of get_vector_elts from one or two source vectors. If
so, it either makes a concat_vectors node or a shufflevector node.
However, it doesn't check the element type width of the underlying
vector, so if you have this sequence:
Node0: v4i16 = ...
Node1: i32 = extract_vector_elt Node0
Node2: i32 = extract_vector_elt Node0
Node3: v16i8 = BUILD_VECTOR Node1, Node2, ...
It will attempt to:
Node0: v4i16 = ...
NewNode1: v16i8 = concat_vectors Node0, ...
Where this is actually invalid because the element width is completely
different. This causes an assertion failure on DAG legalization stage.
Fix:
If output item type of BUILD_VECTOR differs from input item type.
Make concat_vectors based on input element type and then bitcast it to the output vector type. So the case described above will transformed to:
Node0: v4i16 = ...
NewNode1: v8i16 = concat_vectors Node0, ...
NewNode2: v16i8 = bitcast NewNode1
llvm-svn: 162195
this allows for better code generation.
Added a new DAGCombine transformation to convert FMAX and FMIN to FMANC and
FMINC, which are commutative.
For example:
movaps %xmm0, %xmm1
movsd LC(%rip), %xmm0
minsd %xmm1, %xmm0
becomes:
minsd LC(%rip), %xmm0
llvm-svn: 162187
Add these transformations to the existing add/sub ones:
(and (select cc, -1, c), x) -> (select cc, x, (and, x, c))
(or (select cc, 0, c), x) -> (select cc, x, (or, x, c))
(xor (select cc, 0, c), x) -> (select cc, x, (xor, x, c))
The selects can then be transformed to a single predicated instruction
by peephole.
This transformation will make it possible to eliminate the ISD::CAND,
COR, and CXOR custom DAG nodes.
llvm-svn: 162176
In Debug builds, VerifyDiagnosticConsumer checks any files with diagnostics
to make sure we got the chance to parse them for directives (expected-warning
and friends). This check previously relied on every parsed file having a
FileEntry, which broke the cling interpreter's test suite.
This commit changes the extra debug checking to mark a file as unparsed
as soon as we see a diagnostic from that file. At the very end, any files
that are still marked as unparsed are checked for directives, and a fatal
error is emitted (as before) if we find out that there were directives we
missed. -verify directives should always live in actual parsed files, not
in PCH or AST files.
Patch by Andy Gibbs, with slight modifications by me.
llvm-svn: 162171
when you want to find the caller's saved pc, you look up the return address
register and use that. On arm, for instance, this would be the contents of
the link register (lr).
If the eh_frame CIE defines an RA, record that fact in the UnwindPlan.
When we're finding a saved register, if it's the pc, lok for the location
of the return address register instead.
<rdar://problem/12062310>
llvm-svn: 162167