that get created during loop unswitching, and fix SplitBlockPredecessors'
LCSSA updating code to create new PHIs instead of trying to just move
existing ones.
Also, optimize Loop::verifyLoop, since it gets called a lot. Use
searches on a sorted list of blocks instead of calling the "contains"
function, as is done in other places in the Loop class, since "contains"
does a linear search. Also, don't call verifyLoop from LoopSimplify or
LCSSA, as the PassManager is already calling verifyLoop as part of
LoopInfo's verifyAnalysis.
llvm-svn: 81221
linear scan reg alloc. This fixes a problem I ran into where extracting
a function from a larger file caused the generated code to change (masking
the problem I was trying to debug) because the allocator behaved differently.
This changes the results for two X86 regression checks. stack-color-with-reg
is improved, with one less instruction, but pr3495 is worse, with one more
copy. As far as I can tell, these tests were just getting lucky or unlucky,
so I've changed the expected results.
llvm-svn: 81060
supporting other targets. Changed the code to pass MCAsmInfo to the parser
and the lexer. Then changed the lexer to use CommentString from MCAsmInfo
instead of a literal '#' character.
llvm-svn: 81046
a new class, MachineInstrIndex, which hides arithmetic details from
most clients. This is a step towards allowing the register allocator
to update/insert code during allocation.
llvm-svn: 81040
Constant uniquing tables. This allows distinct ConstantExpr objects
with the same operation and different flags.
Even though a ConstantExpr "a + b" is either always overflowing or
never overflowing (due to being a ConstantExpr), it's still necessary
to be able to represent it both with and without overflow flags at
the same time within the IR, because the safety of the flag may
depend on the context of the use. If the constant really does overflow,
it wouldn't ever be safe to use with the flag set, however the use
may be in code that is never actually executed.
This also makes it possible to merge all the flags tests into a single test.
llvm-svn: 80998
and exact flags. Because ConstantExprs are uniqued, creating an
expression with this flag causes all expressions with the same operands
to have the same flag, which may not be safe. Add, sub, mul, and sdiv
ConstantExprs are usually folded anyway, so the main interesting flag
here is inbounds, and the constant folder already knows how to set the
inbounds flag automatically in most cases, so there isn't an urgent need
for the API support.
This can be reconsidered in the future, but for now just removing these
API bits eliminates a source of potential trouble with little downside.
llvm-svn: 80959
Add or Remove operation complete, and not while building the intermediate tree.
This trades a little bit more memory usage for less accesses to the FoldingSet. On a benchmark for the clang static analyzer, this shaves off another 13% of execution time when using field/array sensitivity.
llvm-svn: 80955
from MCAsmLexer.h in preparation of supporting other targets. Changed the
X86AsmParser code to reflect this by removing AsmLexer::LexPercent and looking
for AsmToken::Percent when parsing in places that used AsmToken::Register.
Then changed X86ATTAsmParser::ParseRegister to parse out registers as an
AsmToken::Percent followed by an AsmToken::Identifier.
llvm-svn: 80929
instead of a bool argument, and to do the dominator check itself.
This makes it eaiser to use when DominatorTree information is
available.
llvm-svn: 80920
ImutAVLTree. This was accidentally left out, and essentially caused
digest caching to be ignored in ImmutableMap and ImmutableSet (this
bug was detected from shark traces that showed ComputeDigest was in
the hot path in the clang static analyzer).
This reduces the running time of the clang static analyzer on an
example benchmark by ~32% for both RegionStore (field-sensitivty) and
BasicStore (without field-sensitivity).
llvm-svn: 80877
avoid reloads by reusing clobbered registers.
This was causing issues in 256.bzip2 when compiled with PIC for
a while (starting at r78217), though the problem has since been masked.
llvm-svn: 80872
Use CallbackVH, instead of WeakVH, to hold MDNode elements.
Use FoldingSetNode to unique MDNodes in a context.
Use CallbackVH hooks to update context's MDNodeSet appropriately.
llvm-svn: 80868
Optimal edge profiling is only possible when blocks with no predecessors get an
virtual edge (BB,0) that counts the execution frequencies of this
function-exiting blocks.
This patch makes the necessary changes before actually enabling optimal edge profiling.
llvm-svn: 80667
This adds a pass to verify the current profile against the flow conditions.
This is very helpful when later on trying to perserve the profiling information
during all passes.
llvm-svn: 80666
for sanity. This didn't turn up any bugs.
Change CallGraphNode to maintain its "callsite" information in the
call edges list as a WeakVH instead of as an instruction*. This fixes
a broad class of dangling pointer bugs, and makes CallGraph have a number
of useful invariants again. This fixes the class of problem indicated
by PR4029 and PR3601.
llvm-svn: 80663
makes an eggregious hack somewhat more palatable. Bringing the LSDA forward
and making it a GV available for reference would be even better, but is
beyond the scope of what I'm looking to solve at this point.
Objective C++ code could generate function names that broke the previous
scheme. This fixes that.
llvm-svn: 80649
SCEVUnknowns, as the non-SCEVUnknown cases in the getSCEVAtScope code
can also end up repeatedly climing through the same expression trees,
which can be unusably slow when the trees are very tall.
Also, add a quick check for SCEV pointer equality to the main
SCEV comparison routine, as the full comparison code can be expensive
in the case of large expression trees.
These fix compile-time problems in some pathlogical cases.
llvm-svn: 80623
stem from the fact that we have two types of passes that need to update it:
1. callgraphscc and module passes that are explicitly aware of it
2. Functionpasses (and loop passes etc) that are interlaced with CGSCC passes
by the CGSCC Passmgr.
In the case of #1, we can reasonably expect the passes to update the call
graph just like any analysis. However, functionpasses are not and generally
should not be CG aware. This has caused us no end of problems, so this takes
a new approach. Logically, the CGSCC Pass manager can rescan every function
after it runs a function pass over it to see if the functionpass made any
updates to the IR that affect the callgraph. This allows it to catch new calls
introduced by the functionpass.
In practice, doing this would be slow. This implementation keeps track of
whether or not the current scc is dirtied by a function pass, and, if so,
delays updating the callgraph until it is actually needed again. This was
we avoid extraneous rescans, but we still have good invariants when the
callgraph is needed.
Step #2 of the "give Callgraph some sane invariants" is to change CallGraphNode
to use a CallBackVH for the callsite entry of the CallGraphNode. This way
we can immediately remove entries from the callgraph when a FunctionPass is
active instead of having dangling pointers. The current pass tries to tolerate
these dangling pointers, but it is just an evil hack.
This is related to PR3601/4835/4029. This also reverts r80541, a hack working
around the sad lack of invariants.
llvm-svn: 80566
indirect function pointer, inline it, then go to delete the body.
The problem is that the callgraph had other references to the function,
though the inliner had no way to know it, so we got a dangling pointer
and an invalid iterator out of the deal.
The fix to this is pretty simple: stop the inliner from deleting the
function by knowing that there are references to it. Do this by making
CallGraphNodes contain a refcount. This requires moving deletion of
available_externally functions to the module-level cleanup sweep where
it belongs.
llvm-svn: 80533
argpromotion and structretpromote. Basically, when replacing
a function, they used the 'changeFunction' api which changes
the entry in the function map (and steals/reuses the callgraph
node).
This has some interesting effects: first, the problem is that it doesn't
update the "callee" edges in any callees of the function in the call graph.
Second, this covers for a major problem in all the CGSCC pass stuff, which
is that it is completely broken when functions are deleted if they *don't*
reuse a CGN. (there is a cute little fixme about this though :).
This patch changes the protocol that CGSCC passes must obey: now the CGSCC
pass manager copies the SCC and preincrements its iterator to avoid passes
invalidating it. This allows CGSCC passes to mutate the current SCC. However
multiple passes may be run on that SCC, so if passes do this, they are now
required to *update* the SCC to be current when they return.
Other less interesting parts of this patch are that it makes passes update
the CG more directly, eliminates changeFunction, and requires clients of
replaceCallSite to specify the new callee CGN if they are changing it.
llvm-svn: 80527
encodings.
- Make some of the values emitted by the FDEs dependent upon the pointer
size. This is in line with how GCC does things. And it has the benefit of
working for Darwin in 64-bit mode now.
llvm-svn: 80428
- I'm still trying to figure out the cleanest way to implement this and match the assembler, currently there are some substantial differences.
llvm-svn: 80347
calls into a function and if the calls bring in arrays, try to merge
them together to reduce stack size. For example, in the testcase
we'd previously end up with 4 allocas, now we end up with 2 allocas.
As described in the comments, this is not really the ideal solution
to this problem, but it is surprisingly effective. For example, on
176.gcc, we end up eliminating 67 arrays at "gccas" time and another
24 at "llvm-ld" time.
One piece of concern that I didn't look into: at -O0 -g with
forced inlining this will almost certainly result in worse debug
info. I think this is acceptable though given that this is a case
of "debugging optimized code", and we don't want debug info to
prevent the optimizer from doing things anyway.
llvm-svn: 80215
A include/llvm/ADT/iterator.cmake
U autoconf/configure.ac
--- Reverse-merging r80161 into '.':
U cmake/config-ix.cmake
--- Reverse-merging r80171 into '.':
U Makefile
--- Reverse-merging r80173 into '.':
U configure
U include/llvm/Config/config.h.in
--- Reverse-merging r80180 into '.':
A include/llvm/ADT/iterator.h.in
Despite common miscomceptions, iterator.h is alive and well. It broke the build
bots for several hours. And yet no one bothered to look at them.
Gabor and Doug, please review your changes and make sure that they actually
build before resubmitting them.
llvm-svn: 80197
- I moved section creation back into AsmParser. I think policy decisions like
this should be pushed higher, not lower, when possible (in addition the
assembler has flags which change this behavior, for example).
llvm-svn: 80162
so get rid of this monstrosity. iterator.h.in is scheduled for deletion in my working copy,
but I wait till I see that configure gets regenerated, as it depends on it. I'll commit
then.
There are still some AC_* tests in the configure.ac dealing with iterators, those can
be zapped probably too.
llvm-svn: 80147
This is a simple AliasAnalysis implementation which works by making
ScalarEvolution queries. ScalarEvolution has a more complete understanding
of arithmetic than BasicAA's collection of ad-hoc checks, so it handles
some cases that BasicAA misses, for example p[i] and p[i+1] within the
same iteration of a loop.
This is currently experimental. It may be that the main use for this pass
will be to help find cases where BasicAA can be profitably extended, or
to help in the development of the overall AliasAnalysis infrastructure,
however it's also possible that it could grow up to become a directly
useful pass.
llvm-svn: 80098
- I haven't really tried to find the "right" way to store the fixups or apply
them, yet. This works, but isn't particularly elegant or fast.
- Still no evaluation support, so we don't actually ever not turn a fixup into
a relocation entry.
llvm-svn: 80089
code hints that it would be a good idea to inline
a function ("inline" keyword). No functional change
yet; FEs do not emit this and inliner does not use it.
llvm-svn: 80063
and introduce a new Instruction::isIdenticalTo which tests for full
identity, including the SubclassOptionalData flags. Also, fix the
Instruction::clone implementations to preserve the SubclassOptionalData
flags. Finally, teach several optimizations how to handle
SubclassOptionalData correctly, given these changes.
This fixes the counterintuitive behavior of isIdenticalTo not comparing
the full value, and clone not returning an identical clone, as well as
some subtle bugs that could be caused by these.
Thanks to Nick Lewycky for reporting this, and for an initial patch!
llvm-svn: 80038
This time there is no additional include of llvm/Config/config.h
at all. Instead I use a hard-coded preprecessor symbol:
LLVM_COMPACTIFY_SENTINELS
(should this work on the self-hosting buildbot, then
cleanups come next)
llvm-svn: 80035
should be forced to 32-bits (.long) even on 64-bit architectures. Darwin wants
these bits to be 64-bits (.quad). However, other platforms may disagree.
This is just the info right now and is part of a work-in-progress which needs
this. We'll add the actual *use* of this soon.
llvm-svn: 80024
This is conventional command-line tool behavior. -f now just means
"enable binary output on terminals".
Add a -f option to llvm-extract and llvm-link, for consistency.
Remove F_Force from raw_fd_ostream and enable overwriting and
truncating by default. Introduce an F_Excl flag to permit users to
enable a failure when the file already exists. This flag is
currently unused.
Update Makefiles and documentation accordingly.
llvm-svn: 79990
U include/llvm/BasicBlock.h
U include/llvm/ADT/ilist_node.h
U include/llvm/ADT/ilist.h
U include/llvm/CodeGen/SelectionDAG.h
U include/llvm/CodeGen/MachineFunction.h
U include/llvm/CodeGen/MachineBasicBlock.h
U include/llvm/Function.h
Revert r79938. It was causing self-hosting build failures.
llvm-svn: 79960
*) introducing new data type and export function of edge info for whole function (preparation for next patch).
*) renaming variables to make clear distinction between data and containers that contain this data.
*) updated comments and whitespaces.
*) made ProfileInfo::MissingValue a double (as it should be...).
(Discussed at http://lists.cs.uiuc.edu/pipermail/llvm-commits/Week-of-Mon-20090817/084955.html.)
llvm-svn: 79940
reduce the size of relevant "ghostly" sentinels
by a pointer.
This attempt now makes the compactification dependent
on the configure variable LLVM_COMPACT_SENTINELS
and should not cause any bootstrap failures for
llvm-gcc any more.
Please note that this is not yet the final version,
and (as settled with Chris) I shall take out the
autofoo/cmake portions in the next days.
This will also lose the assertability on sentinel
dereferencing and operator++, but that seems
an acceptable price to pay for the simplified
build logic.
llvm-svn: 79938
- The indirect table itself isn't being filled in yet.
- This isn't factored properly and is rather FIXMEd, but at the moment I'm more
focused on figuring out what it needs to do.
llvm-svn: 79910
- This is mostly complete, the main thing missing is .indirect_symbol support
(which would be straight-forward, except that the way it is implemented in
'as' makes getting an exact .o match interesting).
llvm-svn: 79899
means that raw_ostream no longer has to #include <iosfwd>. Nothing in llvm
should use raw_os_ostream.h, but llvm-gcc and some unit tests do.
llvm-svn: 79886
- I did not actually remove the configure test itself, someone who can
regenerate configure should do this (see FIXME in autoconf/configure.ac)
llvm-svn: 79881
forcing them down into various .cpp files.
This change also:
1. Renames TimeValue::toString() and Path::toString() to ::str()
for similarity with the STL.
2. Removes all stream insertion support for sys::Path, forcing
clients to call .str().
3. Removes a use of Config/alloca.h from bugpoint, using smallvector
instead.
4. Weans llvm-db off <iostream>
sys::Path really needs to be gutted, but I don't have the desire to
do it at this point.
llvm-svn: 79869
- This also shortens the Format.h implementation, and uses the print buffer
fully (it was wasting a character).
- This manifested as llvm-test failures, because one side effect was that
raw_ostream would write garbage '\x00' values into the output stream if it
happened that the string was at the end of the buffer. This meant that grep
would report 'Binary file matches', which meant the silly pattern matching
llvm-test eventually does would fail. Cute. :)
llvm-svn: 79862
MachineInstr and MachineOperand. This required eliminating a
bunch of stuff that was using DOUT, I hope that bill doesn't
mind me stealing his fun. ;-)
llvm-svn: 79813
instead of as two bools. Use this to add a F_Append flag
which has the obvious behavior.
Other unrelated changes conflated into this patch:
1. REmove EH stuff from llvm-dis and llvm-as, the try blocks
are dead.
2. Simplify the filename inference code in llvm-as/llvm-dis,
because raw_fd_ostream does the right thing with '-'.
3. Switch machine verifier to use raw_ostream instead of ostream
(Which is the thing that needed append in the first place).
llvm-svn: 79807
