generate wrong code pretty much anywhere AFAICT.
A case that hits the bug reproducibly is impossible,
but the situation was like this:
Addr = ...
Store -> Addr
Addr2 = GEP , 0, 0
Store -> Addr2
Handling the first store, the code changed replaced Addr
with a sunkaddr and deleted Addr, but not its table
entry.  Code in OptimizedBlock replaced Addr2 with a
bitcast; if that happened to reuse the memory of Addr,
the old table entry was erroneously found when handling
the second store.

llvm-svn: 100044

llvm-svn: 99948

e.g., llvm.memcpy.i32(i8*, i8*, i32, i32) -> llvm.memcpy.p0i8.p0i8.i32(i8*, i8*, i32, i32, i1)
A update of langref will occur in a subsequent checkin.

llvm-svn: 99928

llvm-svn: 99917

spotted by baldrick -- thanks\!

llvm-svn: 99914

pointer.  There was also a SmallPtrSet whose settiness wasn't being used, so I
changed it to a SmallVector.

llvm-svn: 99713

llvm-svn: 99711

MemoizedResultChart.

llvm-svn: 99710

of the previous load - it's usually important.  For example, we don't want
to blindly turn an unaligned load into an aligned one.

llvm-svn: 99699

llvm-svn: 99580

llvm-svn: 99567

llvm-svn: 99564

llvm-svn: 99488

llvm-svn: 99451

llvm-svn: 99418

I have audited all getOperandNo calls now, fixing
hidden assumptions. CallSite related uglyness will
be eliminated successively.

Note this patch has a long and griveous history,
for all the back-and-forths have a look at
CallSite.h's log.

llvm-svn: 99399

llvm-svn: 99398

llvm-svn: 99395

llvm-svn: 99394

for the noinline attribute, and make the inliner refuse to
inline a call site when the call site is marked noinline even
if the callee isn't.  This fixes PR6682.

llvm-svn: 99341

llvm-svn: 99324

Teach simplify libcall to transform __strcpy_chk to __memcpy_chk to enable optimizations down stream.

llvm-svn: 99282

if Function::hasAddressTaken works as advertised

also included some cosmetic cleanups

llvm-svn: 99276

llvm-svn: 99263

llvm-svn: 99224

llvm-svn: 99171

we can reapply the InvokeInst operand reordering patch. (see r98957).

llvm-svn: 99170

and use this (as well as getCalledValue) to access the callee,
instead of {g|s}etOperand(0).

llvm-svn: 99084

so that the SCEVExpander doesn't retain a dangling pointer as its
insert position. The dangling pointer in this case wasn't ever used
to insert new instructions, but it was causing trouble with
SCEVExpander's code for automatically advancing its insert position
past debug intrinsics.

This fixes use-after-free errors that valgrind noticed in
test/Transforms/IndVarSimplify/2007-06-06-DeleteDanglesPtr.ll and
test/Transforms/IndVarSimplify/exit_value_tests.ll.

llvm-svn: 99036

back out r98957, it broke http://smooshlab.apple.com:8010/builders/clang-x86_64-darwin10-fnt/builds/703 in the nightly test suite

llvm-svn: 98958

This time I did a self-hosted bootstrap on Linux x86-64,
with no problems. Let's see how darwin 64-bit self-hosting
goes. At the first sign of failure I'll back this out.

Maybe the valgrind bots give me a hint of what may be wrong
(it at all).

llvm-svn: 98957

str[r]chr returns its pointer argument so we cannot mark it as nocapture. Thanks to Duncan for spotting my mistake.

llvm-svn: 98671

llvm-svn: 98663

llvm-svn: 98584

llvm-svn: 98581

In "empty" bb, the return instruction may not be first instruction, if dbg value intrinsics are present in this bb. Use terminator to find return instructions. 

llvm-svn: 98565

llvm-svn: 98491

llvm-svn: 98490

llvm-svn: 98376

to unify the almost identical code in CodeGenPrepare and InstCombineCalls.

llvm-svn: 98338