Disabling a RUN line that's broken until addrspace roundtrips
through llvm-as|llvm-dis.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@45097 91177308-0d34-0410-b5e6-96231b3b80d8

C and Ocaml bindings for address spaces, for that burgeoning market
for Ocaml-based compilers targeting embedded devices. :)

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@45096 91177308-0d34-0410-b5e6-96231b3b80d8

regenerate.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@45085 91177308-0d34-0410-b5e6-96231b3b80d8

Change the PointerType api for creating pointer types. The old functionality of PointerType::get() has become PointerType::getUnqual(), which returns a pointer in the generic address space. The new prototype of PointerType::get() requires both a type and an address space.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@45082 91177308-0d34-0410-b5e6-96231b3b80d8

Make it clear in the LangRef that allocation instructions only operated on the generic address space. Implement support in the verifier for ensuring this is true.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@45080 91177308-0d34-0410-b5e6-96231b3b80d8

Revert this part of r45073 until the verifier is
changed not to reject invoke of inline asm.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@45077 91177308-0d34-0410-b5e6-96231b3b80d8

don't violate C TBAA rules, use FloatToBits instead.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@45076 91177308-0d34-0410-b5e6-96231b3b80d8

fix a questionable cast, thanks to Mike Stump for pointing this out.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@45075 91177308-0d34-0410-b5e6-96231b3b80d8

Fix the JIT encoding of cmp*ss, which aborts with this assertion currently:
X86CodeEmitter.cpp:378: failed assertion `0 && "Immediate size not set!"'

I *think* this is right, but Evan, please verify.  It also looks like
CMPSDrr and maybe others are missing this info.  Evan, plz investigate.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@45074 91177308-0d34-0410-b5e6-96231b3b80d8

Make instcombine promote inline asm calls to 'nounwind'
calls.  Remove special casing of inline asm from the
inliner.  There is a potential problem: the verifier
rejects invokes of inline asm (not sure why).  If an
asm call is not marked "nounwind" in some .ll, and
instcombine is not run, but the inliner is run, then
an illegal module will be created.  This is bad but
I'm not sure what the best approach is.  I'm tempted
to remove the check in the verifier...

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@45073 91177308-0d34-0410-b5e6-96231b3b80d8

Remove spurious warnings from GCC:

warning: suggest a space before ';' or explicit braces around empty
body in 'for' statement

Patch by Mike Stump (modified slightly by yours truly).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@45071 91177308-0d34-0410-b5e6-96231b3b80d8

Break local interferences in StrongPHIElimination.  One step closer...

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@45070 91177308-0d34-0410-b5e6-96231b3b80d8

A few more comments.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@45069 91177308-0d34-0410-b5e6-96231b3b80d8

Define addString() and lookup() out-of-line to dissuade the C++ compiler from inlining it.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@45068 91177308-0d34-0410-b5e6-96231b3b80d8

Provide GraphTraits and DOTGraphTraits interface for Trie.
Retoss private/public stuff.
Make copy ctor and operator= private.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@45067 91177308-0d34-0410-b5e6-96231b3b80d8

Constify graph arguments

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@45066 91177308-0d34-0410-b5e6-96231b3b80d8

Use references in DF iterators. This eliminates copy-ctor calls on huge objects (graphs)

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@45063 91177308-0d34-0410-b5e6-96231b3b80d8

These are more correctly called signaling NaNs.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@45059 91177308-0d34-0410-b5e6-96231b3b80d8

Make better use of instructions that clear high bits; fix various 2-wide shuffle bugs.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@45058 91177308-0d34-0410-b5e6-96231b3b80d8

__builtin_ia32_movqv4si is now expanded to a shuffle.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@45057 91177308-0d34-0410-b5e6-96231b3b80d8

Start committing working test cases for CellSPU.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@45050 91177308-0d34-0410-b5e6-96231b3b80d8

Actually, MOVPQIto64mr is a dup of MOVPQI2QImr, MOV64toPQIrm is a dup of MOVQI2PQIrm.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@45041 91177308-0d34-0410-b5e6-96231b3b80d8

Fix (mem) <-> low 64-bits of xmm bugs pointed out by David Greene. Mac OS X Leopard assembler recognizes movq.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@45040 91177308-0d34-0410-b5e6-96231b3b80d8

Teach the interpreter to read and write memory in the
endianness of the target not of the host.  Done by the
simple expedient of reversing bytes for primitive types
if the host and target endianness don't match.  This is
correct for integer and pointer types.  I don't know if
it is correct for floating point types.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@45039 91177308-0d34-0410-b5e6-96231b3b80d8

x86-32 long doubles are 4-byte aligned on the stack
for parameter passing (only for that, on Darwin).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@45038 91177308-0d34-0410-b5e6-96231b3b80d8

Fix bsf / bsr jit encoding.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@45037 91177308-0d34-0410-b5e6-96231b3b80d8

Oops. Forgot these.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@45036 91177308-0d34-0410-b5e6-96231b3b80d8

Add explicit keywords, and fix a minor typo that they uncovered.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@45034 91177308-0d34-0410-b5e6-96231b3b80d8

Don't redirect stderr when it isn't needed.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@45033 91177308-0d34-0410-b5e6-96231b3b80d8

Fix a typo in a comment.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@45032 91177308-0d34-0410-b5e6-96231b3b80d8

Fix a typo in a comment.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@45031 91177308-0d34-0410-b5e6-96231b3b80d8

Fix Intel asm syntax for the bsr and bsf instructions.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@45030 91177308-0d34-0410-b5e6-96231b3b80d8

Fix ctlz and cttz. llvm definition requires them to return number of bits in of the src type when value is zero.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@45029 91177308-0d34-0410-b5e6-96231b3b80d8

Bug fix. Must also match ResNo when matching an operand with a user.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@45028 91177308-0d34-0410-b5e6-96231b3b80d8

All MMX shift instructions took a <2 x i32> vector as the shift amount parameter. Change this to be <1 x i64> instead, which matches the assembler instruction.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@45027 91177308-0d34-0410-b5e6-96231b3b80d8

Implement ctlz and cttz with bsr and bsf.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@45024 91177308-0d34-0410-b5e6-96231b3b80d8

Add flags to indicate that there are "never" side effects or that there "may be"
side effects for machine instructions.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@45022 91177308-0d34-0410-b5e6-96231b3b80d8

Make it more clear that some things that can't be done in .td files can
still be done in the LLVM code generator. And update the summary for the
X86 target.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@45013 91177308-0d34-0410-b5e6-96231b3b80d8

Make these loops follow GetGEPOperands() behavior.
Let: %q = GEP %p, X, ...
If %p is a GEP, we can chase baseptr further, only if X==0.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@44999 91177308-0d34-0410-b5e6-96231b3b80d8

Fix typo.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@44997 91177308-0d34-0410-b5e6-96231b3b80d8

Add register pairs to the list to check for local interferences.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@44987 91177308-0d34-0410-b5e6-96231b3b80d8

Remove ugly and horrible code.  It's not necessary for correctness, and can be added back later if it causes code quality issues.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@44986 91177308-0d34-0410-b5e6-96231b3b80d8

Be extra careful with extension use optimation. Now turned on by default.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@44981 91177308-0d34-0410-b5e6-96231b3b80d8

Apply CBE/MSIL patch to autoconf

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@44980 91177308-0d34-0410-b5e6-96231b3b80d8

Add install-libs target which only installs libraries, not tools

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@44979 91177308-0d34-0410-b5e6-96231b3b80d8

Remove this testcase as it will always fail on
platform like Darwin.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@44977 91177308-0d34-0410-b5e6-96231b3b80d8

Do not build CBackend and MSIL regardless of configured targets

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@44973 91177308-0d34-0410-b5e6-96231b3b80d8

Fold some and + shift in x86 addressing mode.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@44970 91177308-0d34-0410-b5e6-96231b3b80d8

Oops. Forgot these.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@44969 91177308-0d34-0410-b5e6-96231b3b80d8

Fix for edge profiling, patch by 'Marc' for PR1857

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@44965 91177308-0d34-0410-b5e6-96231b3b80d8

Cosmetic change.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@44961 91177308-0d34-0410-b5e6-96231b3b80d8

Implicit def instructions, e.g. X86::IMPLICIT_DEF_GR32, are always re-materializable and they should not be spilled.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@44960 91177308-0d34-0410-b5e6-96231b3b80d8

Remove host endianness info from TargetData and
put it in a new header System/Host.h instead.
Instead of getting the endianness from configure,
calculate it directly.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@44959 91177308-0d34-0410-b5e6-96231b3b80d8

Remove a forward-declaration for a non-existant class.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@44955 91177308-0d34-0410-b5e6-96231b3b80d8

Allow vector integer constants to be created with
SelectionDAG::getConstant, in the same way as vector floating-point
constants. This allows the legalize expansion code for @llvm.ctpop and
friends to be usable with vector types.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@44954 91177308-0d34-0410-b5e6-96231b3b80d8

Use vector for child storage instead of map. This will also make
our life during future GraphTraits'ing slightly easier.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@44952 91177308-0d34-0410-b5e6-96231b3b80d8

Revert r44626, which turned off the use of readonly
and readnone for functions with bodies because it
broke llvm-gcc-4.2 bootstrap.  It turns out that,
because of LLVM's array_ref hack, gcc was computing
pure/const attributes wrong (now fixed by turning
off the gcc ipa-pure-const pass).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@44937 91177308-0d34-0410-b5e6-96231b3b80d8

1. "Upgrage" comments.
2. Using zero-extended value of Scale and unsigned division is safe provided
   that Scale doesn't have the sign bit set.
   Previously these 2 instructions:
        %p = bitcast [100 x {i8,i8,i8}]* %x to i8*
        %q = getelementptr i8* %p, i32 -4
   were combined into:
        %q = getelementptr [100 x { i8, i8, i8 }]* %x, i32 0,
               i32 1431655764, i32 0
   what was incorrect.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@44936 91177308-0d34-0410-b5e6-96231b3b80d8

Regenerate.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@44934 91177308-0d34-0410-b5e6-96231b3b80d8

Implement part of review feedback for address spaces.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@44933 91177308-0d34-0410-b5e6-96231b3b80d8

Use shuffles to implement insert_vector_elt for i32, i64, f32, and f64.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@44929 91177308-0d34-0410-b5e6-96231b3b80d8

Add a test case for -optimize-ext-uses.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@44928 91177308-0d34-0410-b5e6-96231b3b80d8

Lower a build_vector with all constants into a constpool load unless it can be done with a move to low part.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@44921 91177308-0d34-0410-b5e6-96231b3b80d8

Add a guard to cxxabi header as other platform may
not support it.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@44914 91177308-0d34-0410-b5e6-96231b3b80d8

Fixed PR1629.
Make lli interpreter correctly call external functions sin()/cos(),
__cxa_guard_acquire() and __cxa_guard_release().

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@44910 91177308-0d34-0410-b5e6-96231b3b80d8

Fix a bug in my previous patch, thanks to Jay Foad for
pointing this out and correcting the patch!

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@44907 91177308-0d34-0410-b5e6-96231b3b80d8

Don't muck with phi nodes; bug fixes.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@44905 91177308-0d34-0410-b5e6-96231b3b80d8

Correct typo for Linux: s/esp/%rsp/

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@44904 91177308-0d34-0410-b5e6-96231b3b80d8

Bit masks conflicted. Needed to bump them by one.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@44903 91177308-0d34-0410-b5e6-96231b3b80d8

Forgot to remove a register from the PHI-union after I'd determined that it
interfered with other registers.  Seems like that might be a good thing to do. :-)

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@44902 91177308-0d34-0410-b5e6-96231b3b80d8

Add (very basic) bindings for ModuleProvider.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@44899 91177308-0d34-0410-b5e6-96231b3b80d8

Bug fix. Only safe to perform extension uses optimization if the source of extension is also defined in the same BB as the extension.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@44896 91177308-0d34-0410-b5e6-96231b3b80d8

Changes from Curtis Dunham implementing lazy cycle detection algorithm.
Changes from me implementing different way of representing points-to anything.
Changes from me that improve slightly on LCD.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@44895 91177308-0d34-0410-b5e6-96231b3b80d8

If deleting a reload instruction due to reuse (value is available in register R and reload is targeting R), make sure to invalidate the kill information of the last kill.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@44894 91177308-0d34-0410-b5e6-96231b3b80d8

Need to grow the indexed map. Added debug statements.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@44892 91177308-0d34-0410-b5e6-96231b3b80d8

Simplify slightly.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@44881 91177308-0d34-0410-b5e6-96231b3b80d8

Remove Trie::Edge class. Now edge labels are stored into nodes itself.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@44880 91177308-0d34-0410-b5e6-96231b3b80d8

More progress on StrongPHIElimination.  Now we actually USE the DomForest!

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@44877 91177308-0d34-0410-b5e6-96231b3b80d8

Blark! How in the world did this work without this?!

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@44874 91177308-0d34-0410-b5e6-96231b3b80d8

- Update the virtual reg to machine instruction map when hoisting.
- Fix subtle bug when creating initially creating this map.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@44873 91177308-0d34-0410-b5e6-96231b3b80d8

Checking for "zero operands" during the "CanHoistInst()" method isn't necessary
because those with side effects will be caught by other checks in here.

Also, simplify the check for a BB in a sub loop.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@44871 91177308-0d34-0410-b5e6-96231b3b80d8

Allow the JIT to encode MMX instructions

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@44869 91177308-0d34-0410-b5e6-96231b3b80d8

Use correct member access operator.
(my test commit as well)

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@44868 91177308-0d34-0410-b5e6-96231b3b80d8

Rename these tests to use the appropriate suffixes.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@44867 91177308-0d34-0410-b5e6-96231b3b80d8

Use not instead of ignore when an exit status is expected to always
be non-zero.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@44866 91177308-0d34-0410-b5e6-96231b3b80d8

Don't redirect stderr when it isn't needed.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@44865 91177308-0d34-0410-b5e6-96231b3b80d8

Fix compilation.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@44864 91177308-0d34-0410-b5e6-96231b3b80d8

Update credits.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@44861 91177308-0d34-0410-b5e6-96231b3b80d8

Add information on address space qualifiers for pointer types and global
declarations to the LangRef.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@44860 91177308-0d34-0410-b5e6-96231b3b80d8

Regenerate.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@44859 91177308-0d34-0410-b5e6-96231b3b80d8

Implement address space attribute for LLVM pointer types. Address spaces are
regions of memory that have a target specific relationship, as described in the
Embedded C Technical Report.

This also implements the 2007-12-11-AddressSpaces test,
which demonstrates how address space attributes can be used in LLVM IR.

In addition, this patch changes the bitcode signature for stores (in a backwards
compatible manner), such that the pointer type, rather than the pointee type, is
encoded. This permits type information in the pointer (e.g. address space) to be
preserved for stores.

LangRef updates are forthcoming.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@44858 91177308-0d34-0410-b5e6-96231b3b80d8

Teach VMCore to constant fold shufflevectors with constant operands.
This allows us to compile:

#include <emmintrin.h>
typedef __m128i VSInt16;
typedef short vSInt16 __attribute__ ((__vector_size__ (16)));
VSInt16 t3() {
  return (VSInt16)((vSInt16)_mm_set1_epi16(6518));
}

into:

_t3:
movaps LCPI1_0, %xmm0
ret

instead of:
_t3:
movl $6518, %eax
movd %eax, %xmm0
pextrw $0, %xmm0, %eax
xorps %xmm0, %xmm0
pinsrw $0, %eax, %xmm0
punpcklwd %xmm0, %xmm0
pshufd $0, %xmm0, %xmm0
ret

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@44856 91177308-0d34-0410-b5e6-96231b3b80d8

Implement constant folding if vector<->vector bitcasts where the number
of source/dest elements changes.  This implements
test/Transforms/InstCombine/bitcast-vector-fold.ll

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@44855 91177308-0d34-0410-b5e6-96231b3b80d8

Don't have APInt.cpp depend upon DerivedTypes.h. This helps with splitting the
Support libraries separately into their own module.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@44852 91177308-0d34-0410-b5e6-96231b3b80d8

Add first and really dirty version of generic Trie structure

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@44851 91177308-0d34-0410-b5e6-96231b3b80d8

significantly simplify some code, no functionality change.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@44850 91177308-0d34-0410-b5e6-96231b3b80d8

refactor some code, no functionality change.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@44849 91177308-0d34-0410-b5e6-96231b3b80d8

Switch over to MachineLoopInfo.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@44838 91177308-0d34-0410-b5e6-96231b3b80d8

Pretty print shuffle mask operand.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@44837 91177308-0d34-0410-b5e6-96231b3b80d8

- Improved v8i16 shuffle lowering. It now uses pshuflw and pshufhw as much as
possible before resorting to pextrw and pinsrw.
- Better codegen for v4i32 shuffles masquerading as v8i16 or v16i8 shuffles.
- Improves (i16 extract_vector_element 0) codegen by recognizing
  (i32 extract_vector_element 0) does not require a pextrw.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@44836 91177308-0d34-0410-b5e6-96231b3b80d8