Pre-regalloc tale duplication. Work in progress.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@90759 91177308-0d34-0410-b5e6-96231b3b80d8

fixed some typos in method comments, reworded some comments for clarity

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@90754 91177308-0d34-0410-b5e6-96231b3b80d8

Fixes the Atomic implementation if compiled by MSVC compiler.

sys::cas_flag should be long on this platform, InterlockedAdd() is
defined only for the Itanium architecture (according to MSDN).

Patch by Michael Beck!

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@90748 91177308-0d34-0410-b5e6-96231b3b80d8

If BB is empty, insert PHI before end() instead of front().

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@90744 91177308-0d34-0410-b5e6-96231b3b80d8

Some pretty-printing

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@90742 91177308-0d34-0410-b5e6-96231b3b80d8

Truncate the arguments of llvm.frameaddress / llvm.returnaddress intrinsics from i32 to platform's largest native type

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@90741 91177308-0d34-0410-b5e6-96231b3b80d8

Add lowering of returnaddr and frameaddr intrinsics. Shamelessly stolen from x86 :)

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@90740 91177308-0d34-0410-b5e6-96231b3b80d8

Initial codegen support for MSP430 ISRs

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@90739 91177308-0d34-0410-b5e6-96231b3b80d8

Add MSP430 interrupt calling conv. No functionality change yet.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@90738 91177308-0d34-0410-b5e6-96231b3b80d8

Add ability to select hw multiplier mode and select appropriate libcalls.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@90737 91177308-0d34-0410-b5e6-96231b3b80d8

Fix typos. Thanks to John Tytgat for noticing it!

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@90728 91177308-0d34-0410-b5e6-96231b3b80d8

Dynamic stack realignment use of sp register as source/dest register
in "bic sp, sp, #15" leads to unpredicatble behaviour in Thumb2 mode.
Emit the following code instead:
mov r4, sp
bic r4, r4, #15
mov sp, r4

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@90724 91177308-0d34-0410-b5e6-96231b3b80d8

fix PR5698

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@90708 91177308-0d34-0410-b5e6-96231b3b80d8

remove extraneous comma clang warns about

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@90707 91177308-0d34-0410-b5e6-96231b3b80d8

constant fold loads from memcpy's from global constants.  This is important
because clang lowers nontrivial automatic struct/array inits to memcpy from
a global array.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@90698 91177308-0d34-0410-b5e6-96231b3b80d8

add support for forwarding mem intrinsic values to non-local loads.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@90697 91177308-0d34-0410-b5e6-96231b3b80d8

gvn is optimizing this better now.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@90696 91177308-0d34-0410-b5e6-96231b3b80d8

Handle forwarding local memsets to loads.  For example, we optimize this:

short x(short *A) {
  memset(A, 1, sizeof(*A)*100);
  return A[42];
}

to 'return 257' instead of doing the load.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@90695 91177308-0d34-0410-b5e6-96231b3b80d8

Add helper methods for forming shift operations with a constant
shift amount.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@90694 91177308-0d34-0410-b5e6-96231b3b80d8

merge two tests.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@90691 91177308-0d34-0410-b5e6-96231b3b80d8

CheckAtomic.cmake: Put all C++ code inside CHECK_CXX_SOURCE_COMPILES.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@90685 91177308-0d34-0410-b5e6-96231b3b80d8

Fix for atomic intrinsics detection when using MSVC.

Patch by Michael Beck!

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@90683 91177308-0d34-0410-b5e6-96231b3b80d8

Remove old DBG_LABEL code.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@90669 91177308-0d34-0410-b5e6-96231b3b80d8

Remove the unused DisableLegalizeTypes option and related code.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@90668 91177308-0d34-0410-b5e6-96231b3b80d8

Calling InvalidateEntry during the refinement was breaking the bootstrap.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@90656 91177308-0d34-0410-b5e6-96231b3b80d8

Final cleanups:

- Privatize a typedef.
- Call the InvalidateEntry when refining a type.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@90655 91177308-0d34-0410-b5e6-96231b3b80d8

Temporarily revert r90502. It was causing the llvm-gcc bootstrap on PPC to fail.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@90653 91177308-0d34-0410-b5e6-96231b3b80d8

Document that memory use intrinsics may also return Def results.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@90651 91177308-0d34-0410-b5e6-96231b3b80d8

Fix indentation in switch statement.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@90650 91177308-0d34-0410-b5e6-96231b3b80d8

Generalize this optimization to work on equality comparisons between any two
integers that are constant except for a single bit (the same n-th bit in each).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@90646 91177308-0d34-0410-b5e6-96231b3b80d8

More updates to objectsize intrinsic docs.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@90644 91177308-0d34-0410-b5e6-96231b3b80d8

Don't print a space before the : between the file name and line number.
And separate the directory and file name with a '/'.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@90641 91177308-0d34-0410-b5e6-96231b3b80d8

Inline methods which are called only once.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@90640 91177308-0d34-0410-b5e6-96231b3b80d8

Refactor some code. No functionality change.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@90639 91177308-0d34-0410-b5e6-96231b3b80d8

Print newlines after printing labels for debug info, so that the output
isn't cluttered with things like "Llabel47:Llabel48:  movq  (%rsi), %xmm3"

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@90638 91177308-0d34-0410-b5e6-96231b3b80d8

Don't blindly set the debug location for PHI node copies.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@90637 91177308-0d34-0410-b5e6-96231b3b80d8

Make TargetSelectInstruction protected and called from FastISel.cpp
instead of SelectionDAGISel.cpp.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@90636 91177308-0d34-0410-b5e6-96231b3b80d8

Remove the target hook TargetInstrInfo::BlockHasNoFallThrough in favor of
MachineBasicBlock::canFallThrough(), which is target-independent and more
thorough.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@90634 91177308-0d34-0410-b5e6-96231b3b80d8

Simplify this code: don't call AnalyzeBranch before doing simpler checks.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@90633 91177308-0d34-0410-b5e6-96231b3b80d8

The debug information for an LLVM Instruction applies to that Instruction
and that Instruction only. Implement this by setting the "current debug position"
back to Unknown after processing each instruction.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@90632 91177308-0d34-0410-b5e6-96231b3b80d8

Fix this code to use DIScope instead of DICompileUnit, as in r90181.
Don't print "SrcLine"; just print the filename and line number, which
is obvious enough and more informative.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@90631 91177308-0d34-0410-b5e6-96231b3b80d8

Don't print the debug directory; it's often long and uninteresting. Omit
the column number if it is not known. Handle the case of a missing filename
better.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@90630 91177308-0d34-0410-b5e6-96231b3b80d8

Minor code simplification.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@90628 91177308-0d34-0410-b5e6-96231b3b80d8

Remove an unneeded include.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@90627 91177308-0d34-0410-b5e6-96231b3b80d8

Remove now-redundant llvm-as invocations.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@90626 91177308-0d34-0410-b5e6-96231b3b80d8

Remove an unneeded include.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@90625 91177308-0d34-0410-b5e6-96231b3b80d8

Add testcase for PR4262.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@90623 91177308-0d34-0410-b5e6-96231b3b80d8

Print a space between the comment character and the text.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@90621 91177308-0d34-0410-b5e6-96231b3b80d8

Temporarily revert r72620 because r72619 was reverted.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@90619 91177308-0d34-0410-b5e6-96231b3b80d8

In TAG_subrange_type, uppder bound is zero indexed.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@90617 91177308-0d34-0410-b5e6-96231b3b80d8

Fix a bad merge.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@90616 91177308-0d34-0410-b5e6-96231b3b80d8

Update the TargetInstrInfo interfaces so hasLoad/StoreFrom/ToStackSlot
can return a MachineMemOperand.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@90615 91177308-0d34-0410-b5e6-96231b3b80d8

Fix indentation.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@90613 91177308-0d34-0410-b5e6-96231b3b80d8

Use new interfaces to print spill size.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@90611 91177308-0d34-0410-b5e6-96231b3b80d8

Have hasLoad/StoreFrom/ToStackSlot return the relevant MachineMemOperand.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@90608 91177308-0d34-0410-b5e6-96231b3b80d8

Fix up some comments.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@90603 91177308-0d34-0410-b5e6-96231b3b80d8

Fix 80-column violations.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@90601 91177308-0d34-0410-b5e6-96231b3b80d8

OptParser: Emit HelpText field for option groups.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@90599 91177308-0d34-0410-b5e6-96231b3b80d8

Some code cleanup. No functionality change.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@90588 91177308-0d34-0410-b5e6-96231b3b80d8

Avoid creating a metadata slot for all metadata that contains an instruction

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@90581 91177308-0d34-0410-b5e6-96231b3b80d8

Handle recursive PHI's.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@90575 91177308-0d34-0410-b5e6-96231b3b80d8

Fix crasher when N->getElement(n) is NULL

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@90572 91177308-0d34-0410-b5e6-96231b3b80d8

Add a pre-regalloc tail duplication pass.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@90567 91177308-0d34-0410-b5e6-96231b3b80d8

Don't try to be cute with undef optimization here. Let ProcessImplicitDefs handle it.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@90566 91177308-0d34-0410-b5e6-96231b3b80d8

Add note about a subtle bug in this code.  Does not effect the main
architectures that LLVM targets, because they don't use this code.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@90564 91177308-0d34-0410-b5e6-96231b3b80d8

Fix typo and add missing include.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@90557 91177308-0d34-0410-b5e6-96231b3b80d8

Added debug output for inherited passes that are invalidated.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@90553 91177308-0d34-0410-b5e6-96231b3b80d8

Forward -m32/-m64 to the linker.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@90548 91177308-0d34-0410-b5e6-96231b3b80d8

Support -march/-mtune/-mcpu.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@90547 91177308-0d34-0410-b5e6-96231b3b80d8

Fix PR5551 by not ignoring the top level constantexpr when
folding a load from constant.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@90545 91177308-0d34-0410-b5e6-96231b3b80d8

add to cmake

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@90539 91177308-0d34-0410-b5e6-96231b3b80d8

Small and carefully crafted testcase showing a miscompilation by GVN
that I'm working on.  This is manifesting as a miscompile of 255.vortex
on some targets.  No check lines yet because it fails.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@90520 91177308-0d34-0410-b5e6-96231b3b80d8

add the start of a class used to handle phi translation in memdep and
gvn (this is just a skeleton so far).  This will ultimately be used
to fix a nasty miscompilation with GVN.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@90518 91177308-0d34-0410-b5e6-96231b3b80d8

Create yet another helper for Invoke.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@90514 91177308-0d34-0410-b5e6-96231b3b80d8

Teach AsmWriter to write inline (not via a global metadata slot) metadata that contains an instruction

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@90512 91177308-0d34-0410-b5e6-96231b3b80d8

Fix a comment typo.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@90511 91177308-0d34-0410-b5e6-96231b3b80d8

Add some helpers for Invoke to mirror CreateCall helpers.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@90508 91177308-0d34-0410-b5e6-96231b3b80d8

add an assert to make it really clear what this is doing.  Return singularval as
a compile time perf optimization to avoid a load.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@90507 91177308-0d34-0410-b5e6-96231b3b80d8

Also attempt trivial coalescing for live intervals that end in a copy.

The coalescer is supposed to clean these up, but when setting up parameters
for a function call, there may be copies to physregs. If the defining
instruction has been LICM'ed far away, the coalescer won't touch it.

The register allocation hint does not always work - when the register
allocator is backtracking, it clears the hints.

This patch takes care of a few more cases that r90163 missed.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@90502 91177308-0d34-0410-b5e6-96231b3b80d8

- If the reaching definition is an undef and the use is a PHI, add the implicit_def to the end of the source block.
- When reaching value is replaced with another, update the cache as well.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@90501 91177308-0d34-0410-b5e6-96231b3b80d8

Insert composite type DIE into the map before processing type fields. This allows fields to find their context DIE from the map.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@90498 91177308-0d34-0410-b5e6-96231b3b80d8

Add ParseInlineMetadata() which can parses metadata that refers to an instruction.  Extend ParseParameterList() to use this new function so that calls to llvm.dbg.declare can pass inline metadata

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@90497 91177308-0d34-0410-b5e6-96231b3b80d8

remove out of date FIXME.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@90490 91177308-0d34-0410-b5e6-96231b3b80d8

Handle undef values properly.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@90489 91177308-0d34-0410-b5e6-96231b3b80d8

Watch out for PHI instruction with no source operands.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@90488 91177308-0d34-0410-b5e6-96231b3b80d8

Fix a comment typo.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@90487 91177308-0d34-0410-b5e6-96231b3b80d8

Fix ExpandShiftWithUnknownAmountBit, which was completely bogus.
Pointed out by Javier Martinez (who also provided a patch).  Since
this logic is not used on (for example) x86, I guess nobody noticed.
Tested by generating SHL, SRL, SRA on various choices of i64 for all
possible shift amounts, and comparing with gcc.  Since I did this on
x86-32, I had to force the use of ExpandShiftWithUnknownAmountBit.
What I'm saying here is that I don't have a testcase I can add to the
repository.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@90482 91177308-0d34-0410-b5e6-96231b3b80d8

Clean up some loop logic.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@90481 91177308-0d34-0410-b5e6-96231b3b80d8

Add support to emit debug info for virtual functions and virtual base classes.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@90474 91177308-0d34-0410-b5e6-96231b3b80d8

Print a newline after the Args: line so that unrelated errs() output doesn't
end up on the same line.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@90473 91177308-0d34-0410-b5e6-96231b3b80d8

Fix MSVC build.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@90454 91177308-0d34-0410-b5e6-96231b3b80d8

Convert ProfileVerifier to template so it can be used for different types of ProfileInfo.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@90451 91177308-0d34-0410-b5e6-96231b3b80d8

Do not create negative edge weights in ProfileEstimator.
Use integer values for weights to prevent rounding errors.
Make ProfileEstimator more robust in general CFGs.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@90449 91177308-0d34-0410-b5e6-96231b3b80d8

Add an implementation of the delta debugging algorithm.
 - This is a pretty slow / memory intensive implementation, and I will likely
   change it to an iterative model, but it works.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@90447 91177308-0d34-0410-b5e6-96231b3b80d8

Use ProfileInfo-API in ProfileInfo Loader and do more assertions.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@90446 91177308-0d34-0410-b5e6-96231b3b80d8

Converted ProfileInfo to template, added more API for ProfileInfo-preserving.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@90445 91177308-0d34-0410-b5e6-96231b3b80d8

Teach tail duplication to update SSA form. Work in progress.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@90432 91177308-0d34-0410-b5e6-96231b3b80d8

expand note.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@90429 91177308-0d34-0410-b5e6-96231b3b80d8

add a note

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@90428 91177308-0d34-0410-b5e6-96231b3b80d8

Don't pull vector sext through both hands of a logical operation, since doing so prevents the fusion of vector sext and setcc into vsetcc.
Add a testcase for the above transformation.
Fix a bogus use of APInt noticed while tracking this down.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@90423 91177308-0d34-0410-b5e6-96231b3b80d8