switch the VRBaseMap in the scheduler from an std::map to a DenseMap.  This
speeds up the isel pass from 2.5570s to 2.4722s on kc++ (3.4%).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@33879 91177308-0d34-0410-b5e6-96231b3b80d8

Introduce new UnarySDNode/BinarySDNode/TernarySDNode nodes, which coallocate
their operands with the node itself.  This reduces malloc traffic for operand
lists.  This reduces isel time on kc++ from 2.6164 to 2.5570s, about 2.3%.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@33878 91177308-0d34-0410-b5e6-96231b3b80d8

Add a new SDNode ctor and InitOperands method.  This allows the operands for
an SDNode to be allocated as part of the node itself, instead of being a
separate new[]'d object.  Switch HandleSDNode, LoadSDNode, and StoreSDNode
to use this mechanism.  This saves one heap allocation and free for each node
of this type that is allocated.  This reduces isel time from 2.7638 to 2.6164s
on kc++, which is a 5.6% speedup.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@33877 91177308-0d34-0410-b5e6-96231b3b80d8

eliminate the SDNode::setValueTypes method.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@33876 91177308-0d34-0410-b5e6-96231b3b80d8

eliminate a bunch of duplicate ctors and helper functions.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@33875 91177308-0d34-0410-b5e6-96231b3b80d8

Fix some comments and other minor sundry things.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@33873 91177308-0d34-0410-b5e6-96231b3b80d8

Ensure that visibility and section are properly propagated.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@33872 91177308-0d34-0410-b5e6-96231b3b80d8

Back out last revision which was committed by accident.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@33871 91177308-0d34-0410-b5e6-96231b3b80d8

Make sure that section and visibility are properly propagated.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@33870 91177308-0d34-0410-b5e6-96231b3b80d8

move MorphNode to out of line and merge setNodeOperands into it.  There is
no behavior or performance change here.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@33869 91177308-0d34-0410-b5e6-96231b3b80d8

simplify MorphNodeTo to take a VTList operand.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@33868 91177308-0d34-0410-b5e6-96231b3b80d8

eliminate some extraneous methods in SDNode

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@33867 91177308-0d34-0410-b5e6-96231b3b80d8

Give each selectiondag node class a home for it's vtable and rtti info

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@33866 91177308-0d34-0410-b5e6-96231b3b80d8

Prepare for PR411

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@33865 91177308-0d34-0410-b5e6-96231b3b80d8

Encode small integers more densely in foldingset, avoiding overflowing the SmallVector as often.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@33864 91177308-0d34-0410-b5e6-96231b3b80d8

Switch VAlueMap from std::map to DenseMap.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@33863 91177308-0d34-0410-b5e6-96231b3b80d8

Switch NodeMap from std::map to DenseMap, this speeds up isel by 2.3%

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@33862 91177308-0d34-0410-b5e6-96231b3b80d8

swtich vector-> smallvector, speeding up selectiondag stuff 1%

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@33861 91177308-0d34-0410-b5e6-96231b3b80d8

Switch promoted/expanded ops over to using a DenseMap.  Vector related maps
aren't worth it.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@33860 91177308-0d34-0410-b5e6-96231b3b80d8

Regenerate.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@33859 91177308-0d34-0410-b5e6-96231b3b80d8

For PR1151:
Make llvm-upgrade valgrind clean. Deleting type instances is a really
nasty thing to do to LLVM. This was a hold-over from the re-write.
This fixes test/Assembler/2007-01-22-UpgradeTypeMapInvalidMemory.ll and
potentially many other bugs.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@33858 91177308-0d34-0410-b5e6-96231b3b80d8

switch LegalizedNodes from std::map to a DenseMap.  This speeds up isel
time as a whole on kc++ by 11%.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@33857 91177308-0d34-0410-b5e6-96231b3b80d8

add a version of insert that takes the key and value.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@33856 91177308-0d34-0410-b5e6-96231b3b80d8

For PR1163:
Make the Module's dependent library use a std::vector instead of SetVector
adjust #includes in .cpp files because SetVector.h is no longer included.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@33855 91177308-0d34-0410-b5e6-96231b3b80d8

Make SmallSetVector useful

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@33854 91177308-0d34-0410-b5e6-96231b3b80d8

Eliminate some malloc traffic from LegalizeAllNodesNotLeadingTo, speeding
up isel on kimwitu by 0.7%.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@33853 91177308-0d34-0410-b5e6-96231b3b80d8

Eliminate some std::sets.  This speeds up isel of kimwitu by about 0.9%

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@33852 91177308-0d34-0410-b5e6-96231b3b80d8

Make sure the gccas script is executable.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@33851 91177308-0d34-0410-b5e6-96231b3b80d8

For PR1067:
Update library documentation after removal of libLLVMTransforms.a that
resulted from removal of the LevelRaise pass.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@33850 91177308-0d34-0410-b5e6-96231b3b80d8

Switch SelectionDAG::ReplaceAllUsesOfValueWith to use a SmallSetVector for
the users set (most nodes have 1 or 2 users).  This speeds up the isel pass
3.2% on kimwitu.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@33849 91177308-0d34-0410-b5e6-96231b3b80d8

Various bugfixes

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@33848 91177308-0d34-0410-b5e6-96231b3b80d8

describe SmallSetVector

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@33847 91177308-0d34-0410-b5e6-96231b3b80d8

Convert SetVector to be a true adapter class and add SmallSetVector.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@33846 91177308-0d34-0410-b5e6-96231b3b80d8

remove some dead code

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@33845 91177308-0d34-0410-b5e6-96231b3b80d8

For PR1072:
Removing -raise has neglible positive or negative side effects so we are
opting to remove it. See the PR for comparison details.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@33844 91177308-0d34-0410-b5e6-96231b3b80d8

add an instcombine pass to clean up after heavy-lifting IP passes

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@33843 91177308-0d34-0410-b5e6-96231b3b80d8

null -> pointer, nul -> character :)

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@33841 91177308-0d34-0410-b5e6-96231b3b80d8

Fix some spellos.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@33840 91177308-0d34-0410-b5e6-96231b3b80d8

fix grammar

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@33839 91177308-0d34-0410-b5e6-96231b3b80d8

Serious cleanups.  Make the TOC match the text for the class hierarchy,
move the sections of the class-h around in the right order, make it actually
reflect the classes in LLVM today.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@33838 91177308-0d34-0410-b5e6-96231b3b80d8

Add TOC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@33837 91177308-0d34-0410-b5e6-96231b3b80d8

describe map-like containers

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@33836 91177308-0d34-0410-b5e6-96231b3b80d8

8 buckets is way too small to start out with.  This was only for testing.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@33835 91177308-0d34-0410-b5e6-96231b3b80d8

Clean this up a bit: hide command unless verbose output requested, echo
a useful comment line, add a clean target.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@33834 91177308-0d34-0410-b5e6-96231b3b80d8

Add a test case.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@33833 91177308-0d34-0410-b5e6-96231b3b80d8

ARM callseq_end should have a input flag operand so it would be scheduled right after the call.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@33832 91177308-0d34-0410-b5e6-96231b3b80d8

Fix comments.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@33831 91177308-0d34-0410-b5e6-96231b3b80d8

improve grammar

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@33830 91177308-0d34-0410-b5e6-96231b3b80d8

improve grammar

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@33829 91177308-0d34-0410-b5e6-96231b3b80d8

close run-away tag

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@33828 91177308-0d34-0410-b5e6-96231b3b80d8

fill in the section on Set-like containers.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@33827 91177308-0d34-0410-b5e6-96231b3b80d8

fix validation issues

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@33823 91177308-0d34-0410-b5e6-96231b3b80d8

fix validation problems

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@33822 91177308-0d34-0410-b5e6-96231b3b80d8

Add some notes about choice of container.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@33821 91177308-0d34-0410-b5e6-96231b3b80d8

remove a dead header

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@33820 91177308-0d34-0410-b5e6-96231b3b80d8

Moved the GetTargetRelocation method from PPCMachOWriter to here. It uses
non-Mach-O-specific information.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@33819 91177308-0d34-0410-b5e6-96231b3b80d8

Moved the GetTargetRelocation method to the PPCMachOWriterInfo object. The
PPCMachOWriter is now trivial.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@33818 91177308-0d34-0410-b5e6-96231b3b80d8

Put destructor out-of-line.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@33817 91177308-0d34-0410-b5e6-96231b3b80d8

Pass in non-Mach-O-specific parameters to the GetTargetRelocation method.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@33816 91177308-0d34-0410-b5e6-96231b3b80d8

Added GetTargetRelocation method.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@33815 91177308-0d34-0410-b5e6-96231b3b80d8

Added some accessor methods.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@33814 91177308-0d34-0410-b5e6-96231b3b80d8

Moved MachORelocation to its own header file. Have a call through the
TargetMachOInfo object to the GetTargetRelocation method. Pass into it
non-Mach-O-specific parameters.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@33813 91177308-0d34-0410-b5e6-96231b3b80d8

New file for the MachORelocation structure. It doesn't have to be tied to the
MachOWriter.h file.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@33812 91177308-0d34-0410-b5e6-96231b3b80d8

- Branch max. displacement calculation bug.
- Add debugging info.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@33811 91177308-0d34-0410-b5e6-96231b3b80d8

switch the sched unit map over to use a DenseMap instead of std::map.  This
speeds up isel as a whole time by 2.6%.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@33810 91177308-0d34-0410-b5e6-96231b3b80d8

Switch ComputeTopDownOrdering over to using a densemap.  This speeds up
isel as a whole by 3.3%.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@33809 91177308-0d34-0410-b5e6-96231b3b80d8

__builtin_isunordered test case.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@33808 91177308-0d34-0410-b5e6-96231b3b80d8

For PR1152:
Update documentation to reflect that this isn't a full-blown executable
any more.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@33807 91177308-0d34-0410-b5e6-96231b3b80d8

Pasto

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@33806 91177308-0d34-0410-b5e6-96231b3b80d8

And even more correctly.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@33805 91177308-0d34-0410-b5e6-96231b3b80d8

Format the examples correctly.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@33804 91177308-0d34-0410-b5e6-96231b3b80d8

For PR1152:
Update documentation to reflect addition of gccas functionality into opt.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@33803 91177308-0d34-0410-b5e6-96231b3b80d8

Improves alloca test. Verifies SP restoration.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@33802 91177308-0d34-0410-b5e6-96231b3b80d8

Switch inliner over to use DenseMap instead of std::map for ValueMap.  This
speeds up the inliner 16%.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@33801 91177308-0d34-0410-b5e6-96231b3b80d8

bugfix: SP isn't resetted when function has FP and there is no spills.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@33800 91177308-0d34-0410-b5e6-96231b3b80d8

Switch this back to using an std::map.  DenseMap entries are getting invalidated

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@33799 91177308-0d34-0410-b5e6-96231b3b80d8

Two improvements:
1. Allow -- as well as - options (Bill Wendling)
2. Pass unrecognized options to opt and let it handle the errors, if any
   (Chris Lattner).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@33798 91177308-0d34-0410-b5e6-96231b3b80d8

silence annoying warning in release-asserts build

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@33797 91177308-0d34-0410-b5e6-96231b3b80d8

Remove more malloc thrashing, this speeds up IPSCCP on kimwitu another 6.7%.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@33796 91177308-0d34-0410-b5e6-96231b3b80d8

Another thumb large stack offset codegen bug.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@33795 91177308-0d34-0410-b5e6-96231b3b80d8

Convert an std::set to SmallSet, this speeds up IPSCCP 17% on kimwitu.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@33794 91177308-0d34-0410-b5e6-96231b3b80d8

eliminate a malloc/free for (almost) every GEP processed.  This speeds up
IPSCCP 3.3% on kimwitu.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@33793 91177308-0d34-0410-b5e6-96231b3b80d8

switch hash_map's over to DenseMap in SCCP.  This speeds up SCCP by 30% in
a release-assert build on kimwitu++.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@33792 91177308-0d34-0410-b5e6-96231b3b80d8

add find/erase, add const iterators, fix bugs in iterators.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@33791 91177308-0d34-0410-b5e6-96231b3b80d8

add iterators

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@33790 91177308-0d34-0410-b5e6-96231b3b80d8

Use MBB.empty() instead of MBB.size() for speed.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@33789 91177308-0d34-0410-b5e6-96231b3b80d8

Watch out for empty BB.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@33788 91177308-0d34-0410-b5e6-96231b3b80d8

For PR1152:
Step 2: Make plug compatible shell script to replace gccas binary.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@33787 91177308-0d34-0410-b5e6-96231b3b80d8

For PR1152:
Step 1: Copy gccas functionality to opt. This endows opt with a new
-std-compile-opts option to get the set of optimization passes that
gccas used. It also adds -disable-inlining and -disable-opt which
both apply only if -std-compile-opts is given. The -strip-debug option
was also removed. It just makes sure that "-strip" gets done early and
is mostly there for compatibility with gccas. Finally, a new
-verify-each option will cause the verify pass to be run after each pass.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@33786 91177308-0d34-0410-b5e6-96231b3b80d8

Remove dead code and fix indentation per Chris' review comments.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@33785 91177308-0d34-0410-b5e6-96231b3b80d8

Use short form of BinaryOperator create function.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@33784 91177308-0d34-0410-b5e6-96231b3b80d8

Use short form of binary operator create functions.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@33783 91177308-0d34-0410-b5e6-96231b3b80d8

Put the Shift instructions in the right section.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@33782 91177308-0d34-0410-b5e6-96231b3b80d8

Fix a comment that needed to change after SHIFT patch landed.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@33781 91177308-0d34-0410-b5e6-96231b3b80d8

Ugh. Only meant to do this in thumb mode.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@33780 91177308-0d34-0410-b5e6-96231b3b80d8

bugfix for reid's shift patch.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@33779 91177308-0d34-0410-b5e6-96231b3b80d8

add a note

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@33778 91177308-0d34-0410-b5e6-96231b3b80d8

1. Break long lines to 80 col limit
2. Fix indentation
3. Renumber the instruction opcodes after the Shift became a binary operator.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@33777 91177308-0d34-0410-b5e6-96231b3b80d8

Changes to support making the shift instructions be true BinaryOperators.
This feature is needed in order to support shifts of more than 255 bits
on large integer types.  This changes the syntax for llvm assembly to
make shl, ashr and lshr instructions look like a binary operator:
   shl i32 %X, 1
instead of
   shl i32 %X, i8 1
Additionally, this should help a few passes perform additional optimizations.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@33776 91177308-0d34-0410-b5e6-96231b3b80d8

Fix a bug in getARMCMP (which translate CondCode to ARM specific CC) when the RHS is a constant.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@33775 91177308-0d34-0410-b5e6-96231b3b80d8