Rename createNode -> createSentinal.

Add a new method, destroySentinal, that is used to delete it (instead of
requiring use of delete.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19921 91177308-0d34-0410-b5e6-96231b3b80d8

clean up comments

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19920 91177308-0d34-0410-b5e6-96231b3b80d8

This has been deprecated for long enough, nuke it.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19919 91177308-0d34-0410-b5e6-96231b3b80d8

This file was schizophrenic when it came to representing sizes.  In some
cases it represented them as 'unsigned's, which are not enough for 64-bit
hosts.  In other cases, it represented them as uint64_t's, which are
inefficient for 32-bit hosts.

This patch unifies all of the sizes to use size_t instead.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19918 91177308-0d34-0410-b5e6-96231b3b80d8

There is no reason to include ostream here, include iosfwd instead.

This file was schizophrenic when it came to representing sizes.  In some
cases it represented them as 'unsigneds', which are not enough for 64-bit
hosts.  In other cases, it represented them as uint64_t's, which are
inefficient for 32-bit hosts.

This patch unifies all of the sizes to use size_t instead.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19917 91177308-0d34-0410-b5e6-96231b3b80d8

After reading in a bc file, trim the resultant buffer down to what we
really need.  This reduces 4M of memory consumption reading 176.gcc.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19916 91177308-0d34-0410-b5e6-96231b3b80d8

Finegrainify namespacification

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19915 91177308-0d34-0410-b5e6-96231b3b80d8

first step towards a correct and complete stack.  also add some forms for things that were getting stuck in the nightly tester.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19914 91177308-0d34-0410-b5e6-96231b3b80d8

Due to previous simplifications, we can simplify the data structures being
used here.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19913 91177308-0d34-0410-b5e6-96231b3b80d8

Properly handle volatile.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19912 91177308-0d34-0410-b5e6-96231b3b80d8

Remove some useless map operations.  Loads/stores that are in the same
BB as the load are not included in the Cand* sets at all.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19911 91177308-0d34-0410-b5e6-96231b3b80d8

Before doing expensive global analysis, check to make sure the pointer is
not invalidated on entry and on exit of the block.  This fixes some N^2
behavior in common cases, and speeds up gcc another 5% to 22.35s.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19910 91177308-0d34-0410-b5e6-96231b3b80d8

Fix quotes.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19909 91177308-0d34-0410-b5e6-96231b3b80d8

Unbreak VC++ build

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19908 91177308-0d34-0410-b5e6-96231b3b80d8

Minor simplification/speedup.  Replaces a set lookup with a pointer comparison.
This speeds up 176.gcc from 25.73s to 23.48s, which is 9.5%

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19907 91177308-0d34-0410-b5e6-96231b3b80d8

Eliminate generality that is not buying us anything.  In particular, this
will cause us to miss cases where the input pointer to a load could be value
numbered to another load.  Something like this:

  %X = load int* %P1
  %Y = load int* %P2

Those are obviously the same if P1/P2 are the same.  The code this patch
removes attempts to handle that.  However, since GCSE iterates, this doesn't
actually buy us anything: GCSE will first replace P1 or P2 with the other
one, then the load can be value numbered as equal.

Removing this code speeds up gcse a lot.  On 176.gcc in debug mode, this
speeds up gcse from 29.08s -> 25.73s, a 13% savings.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19906 91177308-0d34-0410-b5e6-96231b3b80d8

If we see:

  %A = alloca int
  %V = load int* %A

value number %V to undef, not 0.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19905 91177308-0d34-0410-b5e6-96231b3b80d8

Memory used is a delta between memuse at the start of the time and the
memuse at the end, thus it is signed.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19904 91177308-0d34-0410-b5e6-96231b3b80d8

Unbreak VC++ build

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19903 91177308-0d34-0410-b5e6-96231b3b80d8

Make sure that we always grow a multiple of 2 operands.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19902 91177308-0d34-0410-b5e6-96231b3b80d8

noteworthy.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19901 91177308-0d34-0410-b5e6-96231b3b80d8

Adjust to changes in instruction interfaces.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19900 91177308-0d34-0410-b5e6-96231b3b80d8

Switchinst takes a hint for the number of cases it will have.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19899 91177308-0d34-0410-b5e6-96231b3b80d8

switchinst ctor now takes a hint for the number of cases that it will have.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19898 91177308-0d34-0410-b5e6-96231b3b80d8

Adjust Valuehandle to hold its operand directly in it.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19897 91177308-0d34-0410-b5e6-96231b3b80d8

Finegrainify namespacification.
Adjust TmpInstruction to work with the new User model.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19896 91177308-0d34-0410-b5e6-96231b3b80d8

add namespace qualifier

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19895 91177308-0d34-0410-b5e6-96231b3b80d8

Adjust to changes in User class and minor changes in instruction ctors.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19894 91177308-0d34-0410-b5e6-96231b3b80d8

Adjust to slight changes in instruction interfaces.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19893 91177308-0d34-0410-b5e6-96231b3b80d8

Adjust to changes in User class.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19892 91177308-0d34-0410-b5e6-96231b3b80d8

Merge InstrTypes.cpp into this file
Adjust to changes in the User class, operand handling is very different.
PHI node and switch statements must handle explicit resizing of operand
lists.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19891 91177308-0d34-0410-b5e6-96231b3b80d8

Adjust to changes in User class.  Aggregate constants now must explicitly
manage their operands.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19890 91177308-0d34-0410-b5e6-96231b3b80d8

This file is now merged into Instructions.cpp

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19889 91177308-0d34-0410-b5e6-96231b3b80d8

Adjust to changes in the User class.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19888 91177308-0d34-0410-b5e6-96231b3b80d8

Adjust to changes in the User class.  Introduce a new UnaryInstruction
class.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19887 91177308-0d34-0410-b5e6-96231b3b80d8

Adjust to user changes.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19886 91177308-0d34-0410-b5e6-96231b3b80d8

Many changes to cope with the User.h changes.  Instructions now generally
directly embed their operands.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19885 91177308-0d34-0410-b5e6-96231b3b80d8

Adjust to User.h changes.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19884 91177308-0d34-0410-b5e6-96231b3b80d8

Instead of storing operands as std::vector<Use>, just maintain a pointer
and num operands in the User class.  this allows us to embed the operands
directly in the subclasses if possible.  For example, for binary operators
we store the two operands in the derived class.

The has several effects:
  1. it improves locality because the operands and instruction are together
  2. it makes accesses to operands faster (one less load) if you access them
     through the derived class pointer.  For example this:

Value *GetBinaryOperatorOp(BinaryOperator *I, int i) {
  return I->getOperand(i);
}

Was compiled to:

_Z19GetBinaryOperatorOpPN4llvm14BinaryOperatorEi:
        movl    4(%esp), %edx
        movl    8(%esp), %eax
        sall    $4, %eax
        movl    24(%edx), %ecx
        addl    %ecx, %eax
        movl    (%eax), %eax
        ret

and is now compiled to:

_Z19GetBinaryOperatorOpPN4llvm14BinaryOperatorEi:
        movl    8(%esp), %eax
        movl    4(%esp), %edx
        sall    $4, %eax
        addl    %edx, %eax
        movl    44(%eax), %eax
        ret

Accesses through "Instruction*" are unmodified.

   3. This reduces memory consumption (by about 3%) by eliminating 1 word of
      vector overhead and a malloc header on a seperate object.
   4. This speeds up gccas about 10% (both debug and release builds) on
      large things (such as 176.gcc).  For example, it takes a debug build
      from 172.9 -> 155.6s and a release gccas from 67.7 -> 61.8s

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19883 91177308-0d34-0410-b5e6-96231b3b80d8

fix ExprMap, partially teach about add long

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19882 91177308-0d34-0410-b5e6-96231b3b80d8

Fix a nasty thinko in my previous commit.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19881 91177308-0d34-0410-b5e6-96231b3b80d8

Alpha doesn't have a native f32 extload instruction.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19880 91177308-0d34-0410-b5e6-96231b3b80d8

implement legalization of truncates whose results and sources need to be
truncated, e.g. (truncate:i8 something:i16) on a 32 or 64-bit RISC.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19879 91177308-0d34-0410-b5e6-96231b3b80d8

Get alpha working with memset/memcpy/memmove

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19878 91177308-0d34-0410-b5e6-96231b3b80d8

Fix some typos in the Makefile.rules.
Patch contributed by Vladimer Merzliakov.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19877 91177308-0d34-0410-b5e6-96231b3b80d8

Hrm, who knows what 'uint' is, but it seems to work sometimes?  Wierd.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19876 91177308-0d34-0410-b5e6-96231b3b80d8

* add some DEBUG statements
* Properly compile this:

struct a {};
int test() {
  struct a b[2];
  if (&b[0] != &b[1])
    abort ();
  return 0;
}

to 'return 0', not abort().

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19875 91177308-0d34-0410-b5e6-96231b3b80d8

Fix ConstProp/2005-01-28-SetCCGEP.ll: indexing over zero sized elements does
not change the address.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19874 91177308-0d34-0410-b5e6-96231b3b80d8

New testcase.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19873 91177308-0d34-0410-b5e6-96231b3b80d8

Add some initial documentation for the SelectionDAG based instruction selectors

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19872 91177308-0d34-0410-b5e6-96231b3b80d8

Do not clean up if the MappedFile was never used or if the client already
closed the file.  This unbreaks the build.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19871 91177308-0d34-0410-b5e6-96231b3b80d8

fix ExprMap and constant check in setcc

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19870 91177308-0d34-0410-b5e6-96231b3b80d8

Get VC++ compiling again

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19869 91177308-0d34-0410-b5e6-96231b3b80d8

Convert some old C-style casts to C++ style.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19868 91177308-0d34-0410-b5e6-96231b3b80d8

move FP into it's own select

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19867 91177308-0d34-0410-b5e6-96231b3b80d8

CopyFromReg produces two values.  Make sure that we remember that both are
legalized, and actually return the correct result when we legalize the chain first.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19866 91177308-0d34-0410-b5e6-96231b3b80d8

These passes are no more.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19865 91177308-0d34-0410-b5e6-96231b3b80d8

Remove this code as it is currently completely broken and unmaintained.
If needed, this can be resurrected from CVS.

Note that several of the interfaces (e.g. the IPModRef ones) are supersumed
by generic AliasAnalysis interfaces that have been written since this code
was developed (and they are not DSA specific).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19864 91177308-0d34-0410-b5e6-96231b3b80d8

Properly close mapped files.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19863 91177308-0d34-0410-b5e6-96231b3b80d8

Mark -parallel pass as `experimental'

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19858 91177308-0d34-0410-b5e6-96231b3b80d8

stack frame fix and zero FP reg fix

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19857 91177308-0d34-0410-b5e6-96231b3b80d8

Floating point instructions like Floating point registers

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19856 91177308-0d34-0410-b5e6-96231b3b80d8

int to float conversion and another setcc

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19855 91177308-0d34-0410-b5e6-96231b3b80d8

Fix grammar

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19854 91177308-0d34-0410-b5e6-96231b3b80d8

teach isel about comparison with constants and zero extending bits

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19853 91177308-0d34-0410-b5e6-96231b3b80d8

Fix some Path bugs

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19852 91177308-0d34-0410-b5e6-96231b3b80d8

perhaps this will let me have calls again

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19851 91177308-0d34-0410-b5e6-96231b3b80d8

minor bug fix

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19850 91177308-0d34-0410-b5e6-96231b3b80d8

minor bug fix

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19849 91177308-0d34-0410-b5e6-96231b3b80d8

added instructions for fp to int to fp moves

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19848 91177308-0d34-0410-b5e6-96231b3b80d8

 initial fp support

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19847 91177308-0d34-0410-b5e6-96231b3b80d8

xfail this.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19846 91177308-0d34-0410-b5e6-96231b3b80d8

XFAIL this for now.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19845 91177308-0d34-0410-b5e6-96231b3b80d8

hum, writing on one machine, testing on another...

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19844 91177308-0d34-0410-b5e6-96231b3b80d8

add some operations, fix others.  should compile several more tests now

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19843 91177308-0d34-0410-b5e6-96231b3b80d8

We can fold promoted and non-promoted loads into divs also!

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19835 91177308-0d34-0410-b5e6-96231b3b80d8

Fold promoted loads into binary ops for FP, allowing us to generate m32 forms
of FP ops.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19834 91177308-0d34-0410-b5e6-96231b3b80d8

problems with bools, and their work arounds

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19833 91177308-0d34-0410-b5e6-96231b3b80d8

Add a dependency to the trace library so that it gets pulled in
automatically.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19828 91177308-0d34-0410-b5e6-96231b3b80d8

Document the --load option.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19822 91177308-0d34-0410-b5e6-96231b3b80d8

more load choices, better add with imm

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19821 91177308-0d34-0410-b5e6-96231b3b80d8

Make -ds-aa more useful, allowing it to be updated as xforms hack on the program.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19818 91177308-0d34-0410-b5e6-96231b3b80d8

Add some methods.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19817 91177308-0d34-0410-b5e6-96231b3b80d8

Clean ups, and taught the instruction selector about immediate forms

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19816 91177308-0d34-0410-b5e6-96231b3b80d8

Alpha JIT prune

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19815 91177308-0d34-0410-b5e6-96231b3b80d8

include prune and JIT prune

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19814 91177308-0d34-0410-b5e6-96231b3b80d8

Pruned includes

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19813 91177308-0d34-0410-b5e6-96231b3b80d8

let configure recognize Alphas

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19811 91177308-0d34-0410-b5e6-96231b3b80d8

let configure recognize Alphas

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19810 91177308-0d34-0410-b5e6-96231b3b80d8

Mark CVS versions different from releases

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19809 91177308-0d34-0410-b5e6-96231b3b80d8

`primitive' has no `a'

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19808 91177308-0d34-0410-b5e6-96231b3b80d8

Do not return true from isSized for things without a size (like functions and
labels) even though they are concrete.  This fixes the DSA regressions from
last night.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19807 91177308-0d34-0410-b5e6-96231b3b80d8

This giant patch speeds up Type::isSized().  Before, this would have to search
large nested types over and over again to determine if they are sized or not.
Now, isSized() is able to make snap decisions about all concrete types, which
are a common occurance (and includes all primitives).

On 177.mesa, this speeds up DSE from 39.5s -> 21.3s and GCSE from
13.2s -> 11.3s, reducing gccas time from 80s -> 61s (this is a debug build).

DSE and GCSE are still too slow on this testcase, but this is a simple
improvement.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19800 91177308-0d34-0410-b5e6-96231b3b80d8

Fix a spurious warning.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19799 91177308-0d34-0410-b5e6-96231b3b80d8

Silence a warning.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19798 91177308-0d34-0410-b5e6-96231b3b80d8

Silence optimized warnings.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19797 91177308-0d34-0410-b5e6-96231b3b80d8

Allow the FP stackifier to completely ignore functions that do not use FP at
all.  This should speed up the X86 backend fairly significantly on integer
codes.  Now if only we didn't have to compute livevar still... ;-)

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19796 91177308-0d34-0410-b5e6-96231b3b80d8

Simplify/speedup the PEI by not having to scan for uses of the callee saved
registers.  This information is computed directly by the register allocator
now.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19795 91177308-0d34-0410-b5e6-96231b3b80d8

Add an accessor.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19794 91177308-0d34-0410-b5e6-96231b3b80d8

Update physregsused info.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19793 91177308-0d34-0410-b5e6-96231b3b80d8