Now that CorrectTypo knows how to correctly search classes for typo
correction candidates, there is no good reason to only replace an
existing CXXScopeSpecifier if it refers to a namespace. While the actual
enablement was a matter of changing a single comparison, the fallout
from enabling the functionality required a lot more code changes
(including my two previous commits).
llvm-svn: 193020
We should dead-strip atoms only if they are created for COMDAT symbols. If we
remove non-COMDAT atoms from a binary, it will no longer be guaranteed that
the binary will work correctly.
In COFF, you can manipulate the order of section contents in the resulting
binary by section name. For example, if you have four sections
.data$unique_prefix_{a,b,c,d}, it's guaranteed that the contents of A, B, C,
and D will be consecutive in the resulting .data section in that order.
Thus, you can access B's and C's contents by incrementing a pointer pointing
to A until it reached to D. That's why we cannot dead-strip B or C even if
no one is directly referencing to them.
Some object files in the standard library actually use that technique.
llvm-svn: 193017
SCEV currently fails to compute loop counts for nonunit stride
loops. This comes up frequently. It prevents loop optimization and
forces vectorization to insert extra loop checks.
For example:
void foo(int n, int *x) {
for (int i = 0; i < n; i += 3) {
x[i] = i;
x[i+1] = i+1;
x[i+2] = i+2;
}
}
We need to properly handle the case in which limit > INT_MAX-stride. In
the above case: n > INT_MAX-3. In this case the loop counter will step
beyond the limit and overflow at the same time. However, knowing that
signed integer overlow in undefined, we can assume the loop test
behavior is arbitrary after overflow. This obeys both C undefined
behavior rules, and the more strict LLVM poison value rules.
I'm finally fixing this in response to Hal Finkel's persistence.
The most probable reason that we never optimized this before is that
we were being careful to handle case where the developer expected a
side-effect free infinite loop relying on overflow:
for (int i = 0; i < n; i += s) {
++j;
}
return j;
If INT_MAX+1 is a multiple of s and n > INT_MAX-s, then we might
expect an infinite loop. However there are plenty of ways to achieve
this effect without relying on undefined behavior of signed overflow.
llvm-svn: 193015
With this commit, all DIEs created in CompileUnit will be added to parents
inside the same function. Also make getOrCreateTemplateType|Value functions
private.
No functionality change.
llvm-svn: 193002
This is another (final?) stab at making us able to parse our own asm output
on Windows.
Symbols on Windows often contain @'s and ?'s in their names. Our asm parser
didn't like this. ?'s were not allowed, and @'s were intepreted as trying to
reference PLT/GOT/etc.
We can't just add quotes around the bad names, since e.g. for MinGW, we use gas
to assemble, and it doesn't like quotes in some places (notably in .def
directives).
This commit makes us allow ?'s in symbol names, and @'s in symbol names for MS
assembly.
Differential Revision: http://llvm-reviews.chandlerc.com/D1978
llvm-svn: 193000
Removing Host/Atomic.h
This header file was not being copied as part of our public API headers and this in turn was causing any plugin to link against LLDB.framework, since SharingPtr.h depends on it
Out of several possible options to fix this issue, the cleanest one is to revert LLDB to use std::atomic<>, as we are a C++11 project and should take advantage of it
The original rationale for going from std::atomic to Host/Atomic.h was that MSVC++ fails to link in CLR mode when std::atomic is used
This is a very Visual Studio/.net specific issue, which hopefully will be fixed
Until them, to allow Windows development to proceed, we are going with a targeted solution where we #ifdef include the Windows specific calls, and let everyone else use the
proper atomic support, as should be
If there is an unavoidable need for a LLDB-specific atomic header, the right way to go at it would be to make an API/lldb-atomic.h header and #ifdef the Windows dependency there
The FormatManager should not need to conditionalize use of std::atomic<>, as other parts of the LLDB internals are successfully using atomic (Address and IRExecutionUnit), so this
Win-specific hack is limited to SharingPtr
llvm-svn: 192993
Before, clang-format would not adjust leading indents if it found a
structural error (e.g. unmatched {}). It seems, however, that
clang-format has gotten good enough at parsing the code structure that
this hurts in almost all cases. Commonly, while writing code, it is
very useful to be able to correclty indent incomplete if statements or
for loops.
In case this leads to errors that we don't anticipate, we need to find
out and fix those.
This fixed llvm.org/PR17594.
llvm-svn: 192988
This looks ugly and leads to llvm.org/PR17590.
Before (with AlwaysBreakBeforeMultilineStrings):
return
"aaaa"
"bbbb";
After:
return "aaaa"
"bbbb";
llvm-svn: 192984
Specifically, prefer breaking before trailing annotations over breaking
before the first parameter.
Before:
void ffffffffffffffffffffffff(
int aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa,
int aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa) OVERRIDE;
After:
void ffffffffffffffffffffffff(int aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa,
int aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa)
OVERRIDE;
llvm-svn: 192983
PR17168 describes a test case that fails when compiling for debug with
fast-isel. Investigation showed that the test was failing because a DBG_VALUE
machine instruction was placed prior to a PHI.
For this problem to occur requires the following:
* Compile for debug
* Compile with fast-isel
* In a block B, fast-isel must partially succeed before punting to DAG-isel
* B must start with a PHI
* The first unhandled node in the DAG must not generate a machine instruction
* A debug value with an order less than that of that first node exists
When all of these circumstances apply, the existing test that an instruction
was not inserted won't fire. Currently it tests whether the block is empty,
or whether the last instruction generated is a phi. When fast-isel has
partially succeeded, the last instruction generated will not be a phi.
Instead, we need to check whether the current insert position is immediately
following a phi. This patch adds that check, and adds the test case from the
PR as a regression test.
llvm-svn: 192976
Nick Lewycky