desired triplet is a sub-target, e.g. thumbv7 vs. arm host). Reverting the
patch isn't quite right either since the previous behavior does not allow the
triplet to be overridden with -march.
llvm-svn: 80742
Add MO flags to simplify the printing of relocations.
Remove the support for printing large code model relocs (which
aren't supported anyway).
llvm-svn: 80691
Add statistics for regular edge profiling, this enables the comparation of the
number of edges inserted by regular and optimal edge profiling.
llvm-svn: 80668
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
tied to different source registers, the TwoAddressInstructionPass needs to
be smarter. Change it to check before replacing a source register whether
that source register is tied to a different destination register, and if so,
defer handling it until a subsequent iteration.
llvm-svn: 80654
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
This fixes leaks from LLVMContext in multithreaded apps.
Since constants are only deleted if they have no uses, it is safe to not delete
a Module on shutdown, as many single-threaded tools do.
Multithreaded apps should however delete the Module before destroying the
Context to ensure that there are no leaks (assuming they use a different context
for each thread).
llvm-svn: 80590
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
changes: SimplifyDemandedBits can't use the builder yet because it
has the wrong insertion point. This fixes a crash building
MultiSource/Benchmarks/PAQ8p
llvm-svn: 80537
instead of CallGraphNode*'s. This also papers over a callgraph
problem where a pass (in this case, MemCpyOpt) introduces a new
function into the module (llvm.memset.i64) but doesn't add it to
the call graph (nor should it, since it is a function pass).
While it might be a good idea for MemCpyOpt to not synthesize
functions in a runOnFunction(), there is no need for FunctionAttrs
to be boneheaded, so fix it there. This fixes an assertion building
176.gcc.
llvm-svn: 80535
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
Shared landing pads run into trouble with SJLJ, as the dispatch table is
mapped to call sites, and merging the pads will throw that off. There needs
to be a one-to-one mapping of landing pad exception table entries to invoke
call points.
Detecting the shared pad during lowering of SJLJ info insn't sufficient, as
the dispatch function may still need separate destinations to properly
handle phi-nodes.
llvm-svn: 80530
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
is itself a bitcast. Since we have gep(bitcast(bitcast(y))) in this
case, just wait for the two bitcasts to get zapped. This prevents
instcombine from confusing some aliasing stuff, and allows it to
directly eliminate the load in the testcase.
llvm-svn: 80508
workslist and is set to insert new instructions before the current one.
Convert a bunch of stuff that used to call InsertNewInstBefore over to
use it, greatly simplifying code and making it more natural.
There is still a lot more to go, but this is a good start.
llvm-svn: 80492
if the operand is not an instruction.
Simplify most uses of AddOperandsToWorkList to use AddValue and
inline it into the one remaining callsite.
llvm-svn: 80488
former looks too much like AddUsersToWorkList and keeps
confusing me.
Remove AddSoonDeadInstToWorklist and change its two callers
to do the same thing in a simpler way.
llvm-svn: 80486
into their callers. simplify ReplaceInstUsesWith. Make
EraseInstFromFunction only add operands to the worklist if
there aren't too many of them (this was a scalability win
for crazy programs that was only infrequently enforced).
Switch more code to using EraseInstFromFunction instead of
duplicating it inline. Change some fcmp/icmp optimizations
to modify fcmp/icmp in place instead of creating a new one
and deleting the old one just to change the predicate.
llvm-svn: 80483
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
This implements the maximum spanning tree algorithm on CFGs according to
weights given by the ProfileEstimator. This is then used to implement Optimal
Edge Profiling.
llvm-svn: 80358
- 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
Add MO flags to simplify the printing of relocations.
Remove the support for printing large code model relocs (which
aren't supported anyway).
llvm-svn: 80278
- Note, this is a gigantic hack, with the sole purpose of unblocking further
work on the assembler (its also possible to test the mathcer more completely
now).
- Despite being a hack, its actually good enough to work over all of 403.gcc
(although some encodings are probably incorrect). This is a testament to the
beauty of X86's MachineInstr, no doubt! ;)
llvm-svn: 80234
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
members that call methods that read the PoisonMemory member.
This fixes potential spurious (though probably otherwise
harmless) poising of unused memory, and fixes the
associated valgrind error.
llvm-svn: 80192
moves. This avoids the need to promote the operands (or implicitly
extend them, a partial register update condition), and can reduce
i8 register pressure. This substantially speeds up code such as
write_hex in lib/Support/raw_ostream.cpp.
subclass-coalesce.ll is too trivial and no longer tests what it was
originally intended to test.
llvm-svn: 80184
- 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
The instructions can be selected directly from the intrinsics. We will need
to add some ARM-specific nodes for VLD/VST of 3 and 4 128-bit vectors, but
those are not yet implemented.
llvm-svn: 80117
leads to partial-register definitions. To help avoid redundant
zero-extensions, also teach the h-register matching patterns that
use movzbl to match anyext as well as zext.
llvm-svn: 80099
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
Torok Edwin