Add support for 'special' llvm globals like debug info and static ctors/dtors.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@26628 91177308-0d34-0410-b5e6-96231b3b80d8

fix a pasto

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@26627 91177308-0d34-0410-b5e6-96231b3b80d8

Use $(Verb) instead of @ so that VERBOSE=1 will print these.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@26626 91177308-0d34-0410-b5e6-96231b3b80d8

a couple of miscellaneous things.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@26625 91177308-0d34-0410-b5e6-96231b3b80d8

Fix a really annoying bug in bugpoint that made reducing C++ testcases
almost impossible with the new CFE.  It now guarantees that the static
ctor/dtor list is correctly split between the modules.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@26624 91177308-0d34-0410-b5e6-96231b3b80d8

Automatically pass -emit-llvm to llvmgcc when using %llvmgcc

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@26623 91177308-0d34-0410-b5e6-96231b3b80d8

Pass -emit-llvm automatically to %llvmgcc and %llvmg++ to fix regression
tests with the new f.e.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@26622 91177308-0d34-0410-b5e6-96231b3b80d8

Add #line support for CBE.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@26621 91177308-0d34-0410-b5e6-96231b3b80d8

Fit to 80 columns.
Add support for running static ctor/dtors that aren't handled by __main.
This fixes programs with the JIT and the new CFE, such as HBD.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@26620 91177308-0d34-0410-b5e6-96231b3b80d8

Add a helper method for running static ctors/dtors in the module.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@26619 91177308-0d34-0410-b5e6-96231b3b80d8

add a new helper method.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@26618 91177308-0d34-0410-b5e6-96231b3b80d8

add a new helper method

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@26617 91177308-0d34-0410-b5e6-96231b3b80d8

Get rid of the multiple copies of getStringValue.  Now a Constant:: method.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@26616 91177308-0d34-0410-b5e6-96231b3b80d8

doo de doo

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@26614 91177308-0d34-0410-b5e6-96231b3b80d8

Split the priority function computation and priority queue management out
of the ScheduleDAGList class into a new SchedulingPriorityQueue class.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@26613 91177308-0d34-0410-b5e6-96231b3b80d8

switch from an explicitly managed list of SUnits to a simple vector of sunits

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@26612 91177308-0d34-0410-b5e6-96231b3b80d8

Shrinkify some fields, fit to 80 columns

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@26611 91177308-0d34-0410-b5e6-96231b3b80d8

revert the previous patch, didn't mean to check it in yet

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@26610 91177308-0d34-0410-b5e6-96231b3b80d8

remove "Slot", it is dead

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@26609 91177308-0d34-0410-b5e6-96231b3b80d8

Change the interface for getting a target HazardRecognizer to be more clean.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@26608 91177308-0d34-0410-b5e6-96231b3b80d8

libstdc++-v3 was failing to build.  Needed to handle composite types with empty
members (running into a zero initializer.)

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@26607 91177308-0d34-0410-b5e6-96231b3b80d8

Fix a miscompilation of 188.ammp with the new CFE.  188.ammp is accessing
arrays out of range in a horrible way, but we shouldn't break it anyway.
Details in the comments.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@26606 91177308-0d34-0410-b5e6-96231b3b80d8

add a note

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@26605 91177308-0d34-0410-b5e6-96231b3b80d8

X86ISD::REP_STOS and X86ISD::REP_MOVS now produces a flag.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@26604 91177308-0d34-0410-b5e6-96231b3b80d8

Use rep/stosl; and Count 0x3; rep/stosb for memset with 4 byte aligned dest.
and variable value.
Similarly for memcpy.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@26603 91177308-0d34-0410-b5e6-96231b3b80d8

Two things:
1. Don't emit debug info, or other llvm.metadata to the .cbe.c file.
2. Mark static ctors/dtors as such, so that bugpoint works on C++ code
   compiled with the new CFE.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@26602 91177308-0d34-0410-b5e6-96231b3b80d8

new testcase that should have been added long ago.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@26601 91177308-0d34-0410-b5e6-96231b3b80d8

Use "llvm.metadata" section for debug globals.  Filter out these globals in the
asm printer.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@26599 91177308-0d34-0410-b5e6-96231b3b80d8

Switch to using a numeric id for anchors.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@26598 91177308-0d34-0410-b5e6-96231b3b80d8

Fix ConstantMerge/2006-03-07-DontMergeDiffSections.ll, a problem Jim
hypotheticalized about, where we would incorrectly merge two globals in
different sections.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@26597 91177308-0d34-0410-b5e6-96231b3b80d8

new testcase

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@26596 91177308-0d34-0410-b5e6-96231b3b80d8

add another missing store.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@26595 91177308-0d34-0410-b5e6-96231b3b80d8

add a couple more load/store instrs, add a newline to the end of file.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@26594 91177308-0d34-0410-b5e6-96231b3b80d8

Bitfield support.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@26593 91177308-0d34-0410-b5e6-96231b3b80d8

Don't generate silly matching code like this:

 if (N1.getOpcode() == ISD::ADD &&
     ...)
   if (... &&
       (N1.getNumOperands() == 1 || !isNonImmUse(N1.Val, N10.Val))) &&
       ...)

TableGen knows N1 must have more than one operand.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@26592 91177308-0d34-0410-b5e6-96231b3b80d8

This kinda sorta implements "things that have to lead a dispatch group".

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@26591 91177308-0d34-0410-b5e6-96231b3b80d8

add some new instructions to the classifier.  With this, we correctly insert
a nop into Freebench/neural, which speeds it up from 136->129s (~5.4%).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@26590 91177308-0d34-0410-b5e6-96231b3b80d8

add some comments that describe what we model

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@26588 91177308-0d34-0410-b5e6-96231b3b80d8

Implement a very very simple hazard recognizer for LSU rejects and ctr set/read
flushes

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@26587 91177308-0d34-0410-b5e6-96231b3b80d8

Fix some formatting, when looking for hazards, prefer target nodes over
things like copyfromreg.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@26586 91177308-0d34-0410-b5e6-96231b3b80d8

add a note

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@26585 91177308-0d34-0410-b5e6-96231b3b80d8

Keep VC++ building.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@26584 91177308-0d34-0410-b5e6-96231b3b80d8

add a note

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@26583 91177308-0d34-0410-b5e6-96231b3b80d8

- Emit subsections_via_symbols for Darwin.
- Conditionalize Dwarf debugging output (Darwin only for now).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@26582 91177308-0d34-0410-b5e6-96231b3b80d8

Enable Dwarf debugging info.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@26581 91177308-0d34-0410-b5e6-96231b3b80d8

Teach the alignment handling code to look through constant expr casts and GEPs

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@26580 91177308-0d34-0410-b5e6-96231b3b80d8

new regression test

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@26579 91177308-0d34-0410-b5e6-96231b3b80d8

Teach instcombine to increase the alignment of memset/memcpy/memmove when
the pointer is known to come from either a global variable, alloca or
malloc.  This allows us to compile this:

  P = malloc(28);
  memset(P, 0, 28);

into explicit stores on PPC instead of a memset call.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@26577 91177308-0d34-0410-b5e6-96231b3b80d8

update file comment

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@26573 91177308-0d34-0410-b5e6-96231b3b80d8

Remove some code that doesn't make sense

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@26572 91177308-0d34-0410-b5e6-96231b3b80d8

Make sure command line options are parsed before we try to add the LibPath
(-L options) to TheLinker. Problem noticed by Wink Saville.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@26571 91177308-0d34-0410-b5e6-96231b3b80d8

Remove SUnit::Priority1: it is re-calculated on demand as number of live
range to be generated.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@26570 91177308-0d34-0410-b5e6-96231b3b80d8

Hoist the HazardRecognizer out of the ScheduleDAGList.cpp file to where
targets can implement them.  Make the top-down scheduler non-g5-specific.

Remove the old testing hazard recognizer.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@26569 91177308-0d34-0410-b5e6-96231b3b80d8

Hoist the HazardRecognizer out of the ScheduleDAGList.cpp file to where
targets can implement them.  Make the top-down scheduler non-g5-specific.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@26568 91177308-0d34-0410-b5e6-96231b3b80d8

Comment fixes

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@26567 91177308-0d34-0410-b5e6-96231b3b80d8

Don't depend on the C99 copysign function, implement it ourselves.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@26566 91177308-0d34-0410-b5e6-96231b3b80d8

new testcase I forgot to check in earlier

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@26565 91177308-0d34-0410-b5e6-96231b3b80d8

When a hazard recognizer needs noops to be inserted, do so.  This represents
noops as null pointers in the instruction sequence.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@26564 91177308-0d34-0410-b5e6-96231b3b80d8

add an emitnoop method

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@26563 91177308-0d34-0410-b5e6-96231b3b80d8

implement TII::insertNoop

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@26562 91177308-0d34-0410-b5e6-96231b3b80d8

custom lowered nodes are legal too

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@26561 91177308-0d34-0410-b5e6-96231b3b80d8

add a hook to insert a noop

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@26560 91177308-0d34-0410-b5e6-96231b3b80d8

Implement G5HazardRecognizer as a trivial thing that wants 5 cycles between
copyfromreg nodes.  Clearly useful!

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@26559 91177308-0d34-0410-b5e6-96231b3b80d8

Add basic hazard recognizer support.  noop insertion isn't complete yet though.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@26558 91177308-0d34-0410-b5e6-96231b3b80d8

For transforms the behave differently if main goes away, add an option to prevent bugpoint from removing main

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@26557 91177308-0d34-0410-b5e6-96231b3b80d8

Fix VC++ compilation error.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@26554 91177308-0d34-0410-b5e6-96231b3b80d8

Breathe some life into a comment.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@26553 91177308-0d34-0410-b5e6-96231b3b80d8

Split the list scheduler into top-down and bottom-up pieces.  The priority
function of the top-down scheduler are completely bogus currently, and
having (future) PPC specific in this file is also wrong, but this is a
small incremental step.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@26552 91177308-0d34-0410-b5e6-96231b3b80d8

Add a new scheduling type.  This is, of course, a hack.  Proper factoring
will come later.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@26551 91177308-0d34-0410-b5e6-96231b3b80d8

Move the available queue to being inside the ListSchedule method, since it
bounds its lifetime.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@26550 91177308-0d34-0410-b5e6-96231b3b80d8

add a note

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@26549 91177308-0d34-0410-b5e6-96231b3b80d8

Reinstate this now that the offending opposite xform has been removed.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@26548 91177308-0d34-0410-b5e6-96231b3b80d8

Do not fold (add (shl x, c1), (shl c2, c1)) -> (shl (add x, c2), c1),
we want to canonicalize the other way.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@26547 91177308-0d34-0410-b5e6-96231b3b80d8

print arbitrary constant pool entries

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@26545 91177308-0d34-0410-b5e6-96231b3b80d8

Back out fold (shl (add x, c1), c2) -> (add (shl x, c2), c1<<c2) for now.
It's causing an infinite loop compiling ldecod on x86 / Darwin.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@26544 91177308-0d34-0410-b5e6-96231b3b80d8

Add some simple copysign folds

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@26543 91177308-0d34-0410-b5e6-96231b3b80d8

Codegen copysign[f] into a FCOPYSIGN node

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@26542 91177308-0d34-0410-b5e6-96231b3b80d8

Copysign needs to be expanded everywhere.  Note that Alpha and IA64 should
implement copysign as a native op if they have it.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@26541 91177308-0d34-0410-b5e6-96231b3b80d8

Add a copysign node

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@26540 91177308-0d34-0410-b5e6-96231b3b80d8

add a note for something evan noticed

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@26539 91177308-0d34-0410-b5e6-96231b3b80d8

Make vector narrowing more effective, implementing
Transforms/InstCombine/vec_narrow.ll.  This add support for narrowing
extract_element(insertelement) also.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@26538 91177308-0d34-0410-b5e6-96231b3b80d8

new testcase for vector narrowing.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@26537 91177308-0d34-0410-b5e6-96231b3b80d8

Implemented.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@26536 91177308-0d34-0410-b5e6-96231b3b80d8

fold (mul (add x, c1), c2) -> (add (mul x, c2), c1*c2)
fold (shl (add x, c1), c2) -> (add (shl x, c2), c1<<c2)

This allows us to compile CodeGen/PowerPC/addi-reassoc.ll into:

_test1:
        slwi r2, r4, 4
        add r2, r2, r3
        lwz r3, 36(r2)
        blr
_test2:
        mulli r2, r4, 5
        add r2, r2, r3
        lbz r2, 11(r2)
        extsb r3, r2
        blr

instead of:

_test1:
        addi r2, r4, 2
        slwi r2, r2, 4
        add r2, r3, r2
        lwz r3, 4(r2)
        blr
_test2:
        addi r2, r4, 2
        mulli r2, r2, 5
        add r2, r3, r2
        lbz r2, 1(r2)
        extsb r3, r2
        blr

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@26535 91177308-0d34-0410-b5e6-96231b3b80d8

new testcase

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@26534 91177308-0d34-0410-b5e6-96231b3b80d8

Fix a crash compiling Obsequi

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@26529 91177308-0d34-0410-b5e6-96231b3b80d8

new testcase

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@26525 91177308-0d34-0410-b5e6-96231b3b80d8

Add factoring of multiplications, e.g. turning A*A+A*B into A*(A+B).
Testcase here: Transforms/Reassociate/mulfactor.ll

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@26524 91177308-0d34-0410-b5e6-96231b3b80d8

Add a note

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@26523 91177308-0d34-0410-b5e6-96231b3b80d8

Regenerate

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@26522 91177308-0d34-0410-b5e6-96231b3b80d8

Don't use invalidated iterators!

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@26521 91177308-0d34-0410-b5e6-96231b3b80d8

Add an entry

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@26520 91177308-0d34-0410-b5e6-96231b3b80d8

Canonicalize (X+C1)*C2 -> X*C2+C1*C2
This implements Transforms/InstCombine/add.ll:test31

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@26519 91177308-0d34-0410-b5e6-96231b3b80d8

new testcase

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@26518 91177308-0d34-0410-b5e6-96231b3b80d8

MEMSET / MEMCPY lowering bugs: we can't issue a single WORD / DWORD version of
rep/stos and rep/mov if the count is not a constant. We could do
  rep/stosl; and $count, 3; rep/stosb
For now, I will lower them to memset / memcpy calls. We will revisit this after
a little bit experiment.

Also need to take care of the trailing bytes even if the count is a constant.
Since the max. number of trailing bytes are 3, we will simply issue loads /
stores.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@26517 91177308-0d34-0410-b5e6-96231b3b80d8

Keep Visual Studio happy.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@26516 91177308-0d34-0410-b5e6-96231b3b80d8

Be more conservative with our symbolic alias analysis.  In particular,
don't assume that A[1][0] and A[0][i] can't alias.  "i" might be out of
range, or even negative.  This fixes a miscompilation of 188.ammp (which
does bad pointer tricks) with the new CFE.

Testcase here: Analysis/BasicAA/2006-03-03-BadArraySubscript.ll

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@26515 91177308-0d34-0410-b5e6-96231b3b80d8

new testcase

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@26514 91177308-0d34-0410-b5e6-96231b3b80d8

add a note

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@26513 91177308-0d34-0410-b5e6-96231b3b80d8

Typo

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@26512 91177308-0d34-0410-b5e6-96231b3b80d8