introduced in r106343, but only showed up recently (with a particular compiler &
linker combination) because of the particular check, and because we have no
builtin checking for dereferencing the end of an array, which is truly
unfortunate.
llvm-svn: 106908
The VNInfo.kills vector was almost unused except for all the code keeping it
updated. The few places using it were easily rewritten to check for interval
ends instead.
The two new methods LiveInterval::killedAt and killedInRange are replacements.
This brings us down to 3 independent data structures tracking kills.
llvm-svn: 106905
SCEVUnknown values which are loop-variant, as LSR can't do anything
interesting with these values in any case. This fixes very slow compile
times on loops which have large numbers of such values.
llvm-svn: 106897
for an "i" constraint should get lowered; PR 6309. While
this argument was passed around a lot, this is the only
place it was used, so it goes away from a lot of other
places.
llvm-svn: 106893
Tests for lldb are written as python scripts which take advantage of the script
bridging provided by LLDB.framework to interact with lldb core.
A specific naming pattern is followed by the .py script to be recognized as
a module which implements a test scenario, namely, Test*.py.
To specify the directories where "Test*.py" python test scripts are located,
you need to pass in a list of directory names. By default, the current
working directory is searched if nothing is specified on the command line.
An example:
[14:10:20] johnny:/Volumes/data/lldb/svn/trunk/test $ ./dotest.py -v example
test_choice (TestSequenceFunctions.TestSequenceFunctions) ... ok
test_sample (TestSequenceFunctions.TestSequenceFunctions) ... ok
test_shuffle (TestSequenceFunctions.TestSequenceFunctions) ... ok
----------------------------------------------------------------------
Ran 3 tests in 0.000s
OK
llvm-svn: 106890
a winowed list of checkers that actually do something for a given StmtClass.
As the number of checkers grows, this may potentially significantly reduce
the number of checkers called at any one time. My own measurements show that
for the ~20 registered Checker objects, only ~5 of them respond at any one time
to a give statement. While this isn't a net performance win right now (there
is a minor slowdown on sqlite.3) this improvement does greatly improve debugging
when stepping through the checkers used to evaluate a given statement.
llvm-svn: 106884
are dead, not just the def of this register. I.e., a register could be dead, but
it's subreg isn't.
Testcase to follow with a subsequent patch.
llvm-svn: 106878
with the following instructions. This is done via trickery by considering the
instruction preceding the IT to be the hazard. Care must be taken to ensure
it's the first non-debug instruction, or the presence of debug info will
affect codegen.
Part of the continuing work for rdar://7797940, making ARM code-gen unaffected
by the presence of debug information.
llvm-svn: 106871
Daniel Dunbar