One more time for indentation.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31604 91177308-0d34-0410-b5e6-96231b3b80d8

Fix indentation problem starting at GEP.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31603 91177308-0d34-0410-b5e6-96231b3b80d8

if lazy compilation is disabled, print an error message and abort if
lazy compilation is ever attempted

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31602 91177308-0d34-0410-b5e6-96231b3b80d8

init ivar

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31601 91177308-0d34-0410-b5e6-96231b3b80d8

allow clients to indicate that they never want lazy compilation.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31600 91177308-0d34-0410-b5e6-96231b3b80d8

PPC supports i32 / i64 pre-inc load / store.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31599 91177308-0d34-0410-b5e6-96231b3b80d8

Don't attempt expensive pre-/post- indexed dag combine if target does not support them.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31598 91177308-0d34-0410-b5e6-96231b3b80d8

Add a mechanism to specify whether a target supports a particular indexed load / store.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31597 91177308-0d34-0410-b5e6-96231b3b80d8

Rename ISD::MemOpAddrMode to ISD::MemIndexedMode

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31596 91177308-0d34-0410-b5e6-96231b3b80d8

Rename ISD::MemOpAddrMode to ISD::MemIndexedMode

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31595 91177308-0d34-0410-b5e6-96231b3b80d8

Allows debugging llc self hosted.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31594 91177308-0d34-0410-b5e6-96231b3b80d8

Merging dwarf info to a single compile unit.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31593 91177308-0d34-0410-b5e6-96231b3b80d8

implement load effective address similar to the alpha backend
remove lea_addri and the now unused memri addressing mode

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31592 91177308-0d34-0410-b5e6-96231b3b80d8

math.h creates ambiguity

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31591 91177308-0d34-0410-b5e6-96231b3b80d8

math.h for C's sake.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31590 91177308-0d34-0410-b5e6-96231b3b80d8

completely revert patrick's enhancement to bugpoint.  Though it makes bugpoint
speed up, it sometimes makes it crash on complex bc files, which isn't very nice.

With this, bugpoint can reduce the 176.gcc failure.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31589 91177308-0d34-0410-b5e6-96231b3b80d8

add a new bugpoint mode -llc-safe.  This uses LLC for both halves of a
miscompilation.  This is useful for working around GCC+CBE bugs or for handling
programs that CBE doesn't support (e.g. inline asm) when searching for
optimizer bugs.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31588 91177308-0d34-0410-b5e6-96231b3b80d8

silence warnings

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31587 91177308-0d34-0410-b5e6-96231b3b80d8

Teach ShrinkDemandedConstant how to handle X+C.  This implements:
add.ll:test33, add.ll:test34, shift-sra.ll:test2

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31586 91177308-0d34-0410-b5e6-96231b3b80d8

new testcases

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31585 91177308-0d34-0410-b5e6-96231b3b80d8

getPostIndexedAddressParts change: passes in load/store instead of its loaded / stored VT.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31584 91177308-0d34-0410-b5e6-96231b3b80d8

Remove M_2_ADDR_FLAG.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31583 91177308-0d34-0410-b5e6-96231b3b80d8

Don't run bugpoint if we can't find a misoptimization.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31582 91177308-0d34-0410-b5e6-96231b3b80d8

Print a usage message if too few arguments to program.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31581 91177308-0d34-0410-b5e6-96231b3b80d8

Add a utility script to find a mis-optimization problem. This sometimes
helps when bugpoint can't find the problem directly because it needs the
set of optimizations that cause the program to fail.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31580 91177308-0d34-0410-b5e6-96231b3b80d8

Should be xfailed for llvmgcc3

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31578 91177308-0d34-0410-b5e6-96231b3b80d8

Should be xfailed for llvmgcc4 and NOT *. * means all platforms regardless of what llvmgcc you use.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31577 91177308-0d34-0410-b5e6-96231b3b80d8

Added indexed store node and patfrag's.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31576 91177308-0d34-0410-b5e6-96231b3b80d8

Divide select methods into groups by SelectionDAG node opcodes (ISD::ADD,
X86ISD::CMP, etc.) instead of SDNode names (add, x86cmp, etc). We now allow
multiple SDNodes to map to the same SelectionDAG node (e.g. store, indexed
store).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31575 91177308-0d34-0410-b5e6-96231b3b80d8

Ensure we don't regress on reading version 6 bytecode.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31574 91177308-0d34-0410-b5e6-96231b3b80d8

Bump the bytecode version number to 7. Implement upgrade of version 6 and
version 6 bytecode.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31573 91177308-0d34-0410-b5e6-96231b3b80d8

Use movl+xchgl instead of pushl+popl.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31572 91177308-0d34-0410-b5e6-96231b3b80d8

Match tblegen changes.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31571 91177308-0d34-0410-b5e6-96231b3b80d8

Always pass the root node to ComplexPattern isel matching function.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31570 91177308-0d34-0410-b5e6-96231b3b80d8

Match more post-indexed ops.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31569 91177308-0d34-0410-b5e6-96231b3b80d8

Should be xfailed for llvm-gcc3.4

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31568 91177308-0d34-0410-b5e6-96231b3b80d8

Document correct opcodes for post 1.9 release.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31566 91177308-0d34-0410-b5e6-96231b3b80d8

Update the instruction opcodes for release 1.9

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31565 91177308-0d34-0410-b5e6-96231b3b80d8

reenable factoring of GEP expressions, being more precise about the
case that it bad to do.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31563 91177308-0d34-0410-b5e6-96231b3b80d8

make this code more efficient by not creating a phi node we are just going to
delete in the first place.  This also makes it simpler.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31562 91177308-0d34-0410-b5e6-96231b3b80d8

Remove redundant <cmath>.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31561 91177308-0d34-0410-b5e6-96231b3b80d8

disable this factoring optzn for GEPs for now, this severely pessimizes some
loops.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31560 91177308-0d34-0410-b5e6-96231b3b80d8

Make sure <cmath> comes first

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31559 91177308-0d34-0410-b5e6-96231b3b80d8

Add backup support for HUGH_VALF.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31553 91177308-0d34-0410-b5e6-96231b3b80d8

initial implementation of addressing mode 2
TODO: fix lea_addri

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31552 91177308-0d34-0410-b5e6-96231b3b80d8

Make it work on Darwin.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31551 91177308-0d34-0410-b5e6-96231b3b80d8

Include llvm/Support/DataTypes.h to define intptr_t.
This fixes the build on OpenBSD and potentially other systems.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31550 91177308-0d34-0410-b5e6-96231b3b80d8

Now can re-enable debug label folding.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31549 91177308-0d34-0410-b5e6-96231b3b80d8

Wasn't handling case of when machine move labels were undefined.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31548 91177308-0d34-0410-b5e6-96231b3b80d8

Split FunctionPassManager_New into FunctionPassManager_New and FunctionPassManagerImpl_New.
FunctionPassManagerImpl_New implements the pass manager.
FunctionPassManager_New is the public interface.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31547 91177308-0d34-0410-b5e6-96231b3b80d8

Split PassManager_New into PassManager_New and PassManagerImpl_New.
PassManagerImpl_New implements the pass manager.
PassManager_New is the public interface.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31546 91177308-0d34-0410-b5e6-96231b3b80d8

Move BasicBlockPassManager_New, FunctionPassManager_New and
ModulePassManager_New class declarations from PassManager.h
to PassManager.cpp

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31545 91177308-0d34-0410-b5e6-96231b3b80d8

- When performing pre-/post- indexed load/store transformation, do not worry
  about whether the new base ptr would be live below the load/store. Let two
  address pass split it back to non-indexed ops.
- Minor tweaks / fixes.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31544 91177308-0d34-0410-b5e6-96231b3b80d8

Fixed a minor bug preventing some pre-indexed load / store transformation.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31543 91177308-0d34-0410-b5e6-96231b3b80d8

For PR950:
This patch converts the old SHR instruction into two instructions,
AShr (Arithmetic) and LShr (Logical). The Shr instructions now are not
dependent on the sign of their operands.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31542 91177308-0d34-0410-b5e6-96231b3b80d8

XFAIL this on Linux until Jim gets Darwin ship shape.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31541 91177308-0d34-0410-b5e6-96231b3b80d8

now that we handle single mbb loops better, one branch is allowed

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31540 91177308-0d34-0410-b5e6-96231b3b80d8

regenerate

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31539 91177308-0d34-0410-b5e6-96231b3b80d8

Fix a bug noticed by Emil Mikulic.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31538 91177308-0d34-0410-b5e6-96231b3b80d8

Fix a obscure post-indexed load / store dag combine bug.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31537 91177308-0d34-0410-b5e6-96231b3b80d8

Refactor all the addressing mode selection stuff into the isel lowering
class, where it can be used for preinc formation.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31536 91177308-0d34-0410-b5e6-96231b3b80d8

correct the (currently unused) pattern for lwzu.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31535 91177308-0d34-0410-b5e6-96231b3b80d8

preincrement case we miss.  xfail until we catch it

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31534 91177308-0d34-0410-b5e6-96231b3b80d8

Beautify.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31533 91177308-0d34-0410-b5e6-96231b3b80d8

Fix the section headings.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31532 91177308-0d34-0410-b5e6-96231b3b80d8

For PR950:
Early commit of the documentation for new conversion instructions for
easier review.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31531 91177308-0d34-0410-b5e6-96231b3b80d8

optimize single MBB loops better.  In particular, produce:

LBB1_57:        #bb207.i
        movl 72(%esp), %ecx
        movb (%ecx,%eax), %cl
        movl 80(%esp), %edx
        movb %cl, 1(%edx,%eax)
        incl %eax
        cmpl $143, %eax
        jne LBB1_57     #bb207.i
        jmp LBB1_64     #cond_next255.i

intead of:

LBB1_57:        #bb207.i
        movl 72(%esp), %ecx
        movb (%ecx,%eax), %cl
        movl 80(%esp), %edx
        movb %cl, 1(%edx,%eax)
        incl %eax
        cmpl $143, %eax
        je LBB1_64      #cond_next255.i
        jmp LBB1_57     #bb207.i

This eliminates a branch per iteration of the loop.  This hurted PPC
particularly, because the extra branch meant another dispatch group for each
iteration of the loop.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31530 91177308-0d34-0410-b5e6-96231b3b80d8

Beautify.
Clarify comments.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31529 91177308-0d34-0410-b5e6-96231b3b80d8

Update new pass managers to use PassManagerAnalysisHelper API.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31526 91177308-0d34-0410-b5e6-96231b3b80d8

Derive new pass managers from PassManagerAnalysisHelper.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31525 91177308-0d34-0410-b5e6-96231b3b80d8

scalarrepl should not split the two elements of the vsiidx array:
int func(vFloat v0, vFloat v1) {
        int ii;
        vSInt32 vsiidx[2];
        vsiidx[0] = _mm_cvttps_epi32(v0);
        vsiidx[1] = _mm_cvttps_epi32(v1);
        ii = ((int *) vsiidx)[4];
        return ii;
}

This fixes Transforms/ScalarRepl/2006-11-07-InvalidArrayPromote.ll

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31524 91177308-0d34-0410-b5e6-96231b3b80d8

scalarrepl should not split the two elements of the vsiidx array:
int func(vFloat v0, vFloat v1) {
int ii;
vSInt32 vsiidx[2];
vsiidx[0] = _mm_cvttps_epi32(v0);
vsiidx[1] = _mm_cvttps_epi32(v1);
ii = ((int *) vsiidx)[4];
return ii;
}

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31523 91177308-0d34-0410-b5e6-96231b3b80d8

Introduce PassManagerAnalysisHelper.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31522 91177308-0d34-0410-b5e6-96231b3b80d8

Add PassManager_New.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31521 91177308-0d34-0410-b5e6-96231b3b80d8

New test case.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31520 91177308-0d34-0410-b5e6-96231b3b80d8

Fixed a bug which causes x86 be to incorrectly match
shuffle v, undef, <2, ?, 3, ?>
to movhlps
It should match to unpckhps instead.

Added proper matching code for
shuffle v, undef, <2, 3, 2, 3>

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31519 91177308-0d34-0410-b5e6-96231b3b80d8

Fix comment.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31518 91177308-0d34-0410-b5e6-96231b3b80d8

Add ModulePassManager_New.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31517 91177308-0d34-0410-b5e6-96231b3b80d8

Accidently reran commit.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31516 91177308-0d34-0410-b5e6-96231b3b80d8

Add FunctionPassManager_New.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31515 91177308-0d34-0410-b5e6-96231b3b80d8

Missed a label map, reverting till covered.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31514 91177308-0d34-0410-b5e6-96231b3b80d8

Add BasicBlockPassManager_New.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31513 91177308-0d34-0410-b5e6-96231b3b80d8

Missed a label map, reverting till covered.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31512 91177308-0d34-0410-b5e6-96231b3b80d8

Optionally allow comparison operations from affect DSGraphs

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31511 91177308-0d34-0410-b5e6-96231b3b80d8

Allow loop detection during debug in forwarding nodes, and revert auxcall patch as it make 176.gcc untenable

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31510 91177308-0d34-0410-b5e6-96231b3b80d8

debug type for DSA TD

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31509 91177308-0d34-0410-b5e6-96231b3b80d8

1. Add a pass to fold debug label instructions so a debug info client can detect
empty ranges.

2. Reorg how MachineDebugInfo maintains changes to debug labels.

3. Have dwarf writer use debug label info to simplify scopes and source line
coorespondence.

4. Revert the merging of compile units until I can get the bugs ironed out.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31507 91177308-0d34-0410-b5e6-96231b3b80d8

add a note from viterbi

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31506 91177308-0d34-0410-b5e6-96231b3b80d8

Enable improved spilling costs by default.  This speeds up viterbi on x86
by 40%, FreeBench/fourinarow by 20%, and many other programs 10-25%.

On PPC, this speeds up fourinarow by 18%, and probably other things as well.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31504 91177308-0d34-0410-b5e6-96231b3b80d8

Got a little ahead of myself here

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31502 91177308-0d34-0410-b5e6-96231b3b80d8

Fix this test

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31501 91177308-0d34-0410-b5e6-96231b3b80d8

Use correct value for float HUGH_VAL.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31500 91177308-0d34-0410-b5e6-96231b3b80d8

Added target hook for post-indexed memory ops transformation.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31499 91177308-0d34-0410-b5e6-96231b3b80d8

Add post-indexed load / store transformations.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31498 91177308-0d34-0410-b5e6-96231b3b80d8

Make some corrections to this test case.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31497 91177308-0d34-0410-b5e6-96231b3b80d8

Ignore the Output dir.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31496 91177308-0d34-0410-b5e6-96231b3b80d8

Add a test case for making sure gdb can get a stack trace with our debug
information.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31495 91177308-0d34-0410-b5e6-96231b3b80d8

Add a new llcbeta option.  This speeds up viterbi from 12.34 to 8.76s on
X86.  If happy, I'll enable this by default.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31493 91177308-0d34-0410-b5e6-96231b3b80d8

Unbreak X86/ELF Debugging. Somehow this line got lost in Jim's cleanup.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31492 91177308-0d34-0410-b5e6-96231b3b80d8

Bumping version number

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31491 91177308-0d34-0410-b5e6-96231b3b80d8