remove dead enum, make setNodeId public.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42977 91177308-0d34-0410-b5e6-96231b3b80d8

Fix 80 col violation

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42976 91177308-0d34-0410-b5e6-96231b3b80d8

Handle PPC long double in CBackend.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42972 91177308-0d34-0410-b5e6-96231b3b80d8

remove misleading comment.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42970 91177308-0d34-0410-b5e6-96231b3b80d8

If a target doesn't have HasMULHU or HasUMUL_LOHI, ExpandOp would return
without lo/hi set.  Fall through to making a libcall instead.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42969 91177308-0d34-0410-b5e6-96231b3b80d8

Consolidate logic for creating NaNs.  Silence compiler warning.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42966 91177308-0d34-0410-b5e6-96231b3b80d8

Whether arithmetic is supported is a property of the semantics.  Make it
so, and clean up the checks by putting them in an inline function.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42965 91177308-0d34-0410-b5e6-96231b3b80d8

Separate out parsing of decimal number.  Use this to only allocate
memory for the significand once up-front.  Also ignore insignificant
trailing zeroes; this saves unnecessary multiplications later.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42964 91177308-0d34-0410-b5e6-96231b3b80d8

New test case.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42963 91177308-0d34-0410-b5e6-96231b3b80d8

Unbreak x86-64.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42962 91177308-0d34-0410-b5e6-96231b3b80d8

When coalescing an EXTRACT_SUBREG and the dst register is a physical register,
the source register will be coalesced to the super register of the LHS. Properly
merge in the live ranges of the resulting coalesced interval that were part of
the original source interval to the live interval of the super-register.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42961 91177308-0d34-0410-b5e6-96231b3b80d8

Revert 42908 for now.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42960 91177308-0d34-0410-b5e6-96231b3b80d8

Fix type mismatch error in PPC Altivec (only causes
a problem when asserts are on).  From vecLib.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42959 91177308-0d34-0410-b5e6-96231b3b80d8

Disable some compile-time optimizations on PPC
long double.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42958 91177308-0d34-0410-b5e6-96231b3b80d8

Fixing a typo. Found by Kevin André!

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42956 91177308-0d34-0410-b5e6-96231b3b80d8

Clarify that fastcc has a problem with nested function
trampolines, rather than with nested functions themselves.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42955 91177308-0d34-0410-b5e6-96231b3b80d8

Enhance the truncstore optimization code to handle shifted
values and propagate demanded bits through them in simple cases.

This allows this code:
void foo(char *P) {
   strcpy(P, "abc");
}
to compile to:

_foo:
        ldrb r3, [r1]
        ldrb r2, [r1, #+1]
        ldrb r12, [r1, #+2]!
        ldrb r1, [r1, #+1]
        strb r1, [r0, #+3]
        strb r2, [r0, #+1]
        strb r12, [r0, #+2]
        strb r3, [r0]
        bx lr

instead of:

_foo:
        ldrb r3, [r1, #+3]
        ldrb r2, [r1, #+2]
        orr r3, r2, r3, lsl #8
        ldrb r2, [r1, #+1]
        ldrb r1, [r1]
        orr r2, r1, r2, lsl #8
        orr r3, r2, r3, lsl #16
        strb r3, [r0]
        mov r2, r3, lsr #24
        strb r2, [r0, #+3]
        mov r2, r3, lsr #16
        strb r2, [r0, #+2]
        mov r3, r3, lsr #8
        strb r3, [r0, #+1]
        bx lr

testcase here: test/CodeGen/ARM/truncstore-dag-combine.ll

This also helps occasionally for X86 and other cases not involving
unaligned load/stores.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42954 91177308-0d34-0410-b5e6-96231b3b80d8

new testcase

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42953 91177308-0d34-0410-b5e6-96231b3b80d8

Add a simple optimization to simplify the input to
truncate and truncstore instructions, based on the
knowledge that they don't demand the top bits.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42952 91177308-0d34-0410-b5e6-96231b3b80d8

If the power of 5 is exact, and the reciprocal exact, the error is zero not one half-ulps.  This prevents an infinite loop in rare cases.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42950 91177308-0d34-0410-b5e6-96231b3b80d8

Fix test case.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42949 91177308-0d34-0410-b5e6-96231b3b80d8

New tests.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42948 91177308-0d34-0410-b5e6-96231b3b80d8

Local spiller optimization:
Turn this:
movswl  %ax, %eax
movl    %eax, -36(%ebp)
xorl    %edi, -36(%ebp)
into
movswl  %ax, %eax
xorl    %edi, %eax
movl    %eax, -36(%ebp)
by unfolding the load / store xorl into an xorl and a store when we know the
value in the spill slot is available in a register. This doesn't change the
number of instructions but reduce the number of times memory is accessed.

Also unfold some load folding instructions and reuse the value when similar
situation presents itself.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42947 91177308-0d34-0410-b5e6-96231b3b80d8

Change unfoldMemoryOperand(). User is now responsible for passing in the
register used by the unfolded instructions. User can also specify whether to
unfold the load, the store, or both.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42946 91177308-0d34-0410-b5e6-96231b3b80d8

Optionally create a MachineInstr without default implicit operands.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42945 91177308-0d34-0410-b5e6-96231b3b80d8

Correcting the corrections. Bad bad baaad emacs!

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42935 91177308-0d34-0410-b5e6-96231b3b80d8

Corrected many typing errors. And removed 'nest' parameter handling
for fastcc from X86CallingConv.td.  This means that nested functions
are not supported for calling convention 'fastcc'.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42934 91177308-0d34-0410-b5e6-96231b3b80d8

Fix this test to not depend on the assembly output containing something
that includes the string "st". This probably fixes the regression on
Darwin.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42932 91177308-0d34-0410-b5e6-96231b3b80d8

Dest type is always i8 *. This allows some simplification.
Do not filter memmove.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42930 91177308-0d34-0410-b5e6-96231b3b80d8

t

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42929 91177308-0d34-0410-b5e6-96231b3b80d8

Fix ocaml bindings for picky linkers. :)

Thanks to Török Edvin for helping to track this down.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42927 91177308-0d34-0410-b5e6-96231b3b80d8

Due to the new tail call optimization, trampolines can no
longer be created for fastcc functions.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42925 91177308-0d34-0410-b5e6-96231b3b80d8

ppc long double.  Implement fabs and fneg.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42924 91177308-0d34-0410-b5e6-96231b3b80d8

Update.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42922 91177308-0d34-0410-b5e6-96231b3b80d8

don't use intptr_t without including it.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42921 91177308-0d34-0410-b5e6-96231b3b80d8

Fix a bug in my patch last night that broke InstCombine/2007-10-12-Crash.ll

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42920 91177308-0d34-0410-b5e6-96231b3b80d8

Implement i64->ppcf128 conversions.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42919 91177308-0d34-0410-b5e6-96231b3b80d8

Make this compute the correct offset, handling alignment of the element
pointer correctly.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42918 91177308-0d34-0410-b5e6-96231b3b80d8

Did mean to leave this in. INSERT_SUBREG isn't being coalesced yet.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42916 91177308-0d34-0410-b5e6-96231b3b80d8

Remove duplicate comment.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42913 91177308-0d34-0410-b5e6-96231b3b80d8

Implement correctly-rounded decimal->binary conversion, i.e. conversion
from user input strings.

Such conversions are more intricate and subtle than they may appear;
it is unlikely I have got it completely right first time.  I would
appreciate being informed of any bugs and incorrect roundings you
might discover.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42912 91177308-0d34-0410-b5e6-96231b3b80d8

Remove a field that was never used.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42911 91177308-0d34-0410-b5e6-96231b3b80d8

If we're trying to be arbitrary precision, unsigned char clearly won't cut it.  Needed for dec->bin conversions.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42910 91177308-0d34-0410-b5e6-96231b3b80d8

Don't attempt to mask no bits

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42909 91177308-0d34-0410-b5e6-96231b3b80d8

Change the names used for internal labels to use the current
function symbol name instead of a codegen-assigned function
number.

Thanks Evan! :-)

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42908 91177308-0d34-0410-b5e6-96231b3b80d8

Fix some corner cases with vectors in copyToRegs and copyFromRegs.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42907 91177308-0d34-0410-b5e6-96231b3b80d8

Add support to SplitVectorOp for powi, where the second operand
is a scalar integer.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42906 91177308-0d34-0410-b5e6-96231b3b80d8

Mark vector ctpop, cttz, and ctlz as Expand on x86.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42905 91177308-0d34-0410-b5e6-96231b3b80d8

Mark vector pow, ctpop, cttz, and ctlz as Expand on PowerPC.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42904 91177308-0d34-0410-b5e6-96231b3b80d8

Restrict EXTRACT_SUBREG coalescing to avoid negative performance impact.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42903 91177308-0d34-0410-b5e6-96231b3b80d8

Doh.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42901 91177308-0d34-0410-b5e6-96231b3b80d8

EXTRACT_SUBREG test case.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42900 91177308-0d34-0410-b5e6-96231b3b80d8

EXTRACT_SUBREG coalescing support. The coalescer now treats EXTRACT_SUBREG like
(almost) a register copy. However, it always coalesced to the register of the
RHS (the super-register). All uses of the result of a EXTRACT_SUBREG are sub-
register uses which adds subtle complications to load folding, spiller rewrite,
etc.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42899 91177308-0d34-0410-b5e6-96231b3b80d8

Some clean up.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42898 91177308-0d34-0410-b5e6-96231b3b80d8

Cosmetic.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42897 91177308-0d34-0410-b5e6-96231b3b80d8

Fix typos.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42896 91177308-0d34-0410-b5e6-96231b3b80d8

Fold load / store into MOV32to32_ and MOV16to16_.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42895 91177308-0d34-0410-b5e6-96231b3b80d8

Flag MOV32to32_ with EXTRACT_SUBREG. They should not be scheduled apart.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42894 91177308-0d34-0410-b5e6-96231b3b80d8

Added missing -march=x86 flag.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42893 91177308-0d34-0410-b5e6-96231b3b80d8

eliminate warning

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42892 91177308-0d34-0410-b5e6-96231b3b80d8

Fix some 80 column violations.
Fix DecomposeSimpleLinearExpr to handle simple constants better.
Don't nuke gep(bitcast(allocation)) if the bitcast(allocation) will
fold the allocation.  This fixes PR1728 and Instcombine/malloc3.ll

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42891 91177308-0d34-0410-b5e6-96231b3b80d8

testcase for PR1728

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42890 91177308-0d34-0410-b5e6-96231b3b80d8

make operator== work with non-equal sized bitvectors, as long as
the extra bits are all zeros.  This allows  "010" and "010000" to be
treated as equal.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42889 91177308-0d34-0410-b5e6-96231b3b80d8

PPC long double.  Implement a couple more conversions.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42888 91177308-0d34-0410-b5e6-96231b3b80d8

Add intrinsics for sin, cos, and pow. These use llvm_anyfloat_ty, and so
may be overloaded with vector types. And add a testcase for codegen for
these.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42885 91177308-0d34-0410-b5e6-96231b3b80d8

Codegen support for vector intrinsics.

Factor out the code that expands the "nasty scalar code" for unrolling
vectors into a separate routine, teach it how to handle mixed
vector/scalar operands, as seen in powi, and use it for several operators,
including sin, cos, powi, and pow.

Add support in SplitVectorOp for fpow, fpowi and for several unary
operators.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42884 91177308-0d34-0410-b5e6-96231b3b80d8

Implement ppc long double->uint conversion.
Make ppc long double constants print.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42882 91177308-0d34-0410-b5e6-96231b3b80d8

Set ISD::FPOW to Expand.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42881 91177308-0d34-0410-b5e6-96231b3b80d8

Add runtime library names for pow.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42880 91177308-0d34-0410-b5e6-96231b3b80d8

Add an ISD::FPOW node type.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42879 91177308-0d34-0410-b5e6-96231b3b80d8

Add an explicit target triple to make this test behave as expected on
non-Apple hosts. And use the count script instead of wc + grep.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42878 91177308-0d34-0410-b5e6-96231b3b80d8

Provided accessors to internal allocator for ImutAVLTree and ImmutableSet.
Added postfix ++,-- support for ImmutableSet::iterator.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42877 91177308-0d34-0410-b5e6-96231b3b80d8

Add missing argument to PALIGNR

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42874 91177308-0d34-0410-b5e6-96231b3b80d8

Fix an assertion abort on sparc. malloc(0) is allowed to
return NULL.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42871 91177308-0d34-0410-b5e6-96231b3b80d8

Added tail call optimization to the x86 back end. It can be
enabled by passing -tailcallopt to llc.  The optimization is
performed if the following conditions are satisfied:
* caller/callee are fastcc
* elf/pic is disabled OR
  elf/pic enabled + callee is in module + callee has
  visibility protected or hidden

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42870 91177308-0d34-0410-b5e6-96231b3b80d8

Adding myself

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42869 91177308-0d34-0410-b5e6-96231b3b80d8

Next PPC long double bits.  First cut at constants.
No compile-time support for constant operations yet,
just format transformations.  Make readers and
writers work.  Split constants into 2 doubles in
Legalize.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42865 91177308-0d34-0410-b5e6-96231b3b80d8

Lower memcpy if it makes sense.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42864 91177308-0d34-0410-b5e6-96231b3b80d8

make bitvector  &= do the right thing if vectors have mismatched length.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42860 91177308-0d34-0410-b5e6-96231b3b80d8

Add a new use_iterator::atEnd() method, which allows us to shrink
pred_iterator down to a single ivar.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42859 91177308-0d34-0410-b5e6-96231b3b80d8

Added iterators to ImmutableSet.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42851 91177308-0d34-0410-b5e6-96231b3b80d8

Added some doxygen comments to ImmutableSet.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42850 91177308-0d34-0410-b5e6-96231b3b80d8

Removed uninformative assertions that catch problems that will
fire anyway at runtime due to a NULL dereference.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42848 91177308-0d34-0410-b5e6-96231b3b80d8

Removed "height" of an AVL tree node from its Profile.  This is
implicitly captured by using the addresses of its children in the
profile.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42847 91177308-0d34-0410-b5e6-96231b3b80d8

Removed spurious forward declaration to a structure that will no longer be used.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42839 91177308-0d34-0410-b5e6-96231b3b80d8

Added some doxygen comments to a few methods of ImutAVLTree.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42837 91177308-0d34-0410-b5e6-96231b3b80d8

Updated VC++ build system.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42835 91177308-0d34-0410-b5e6-96231b3b80d8

Added preliminary support for iterators in ImutAVLTree.
Implemented ImutAVLTree::isEqual.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42833 91177308-0d34-0410-b5e6-96231b3b80d8

Fix CodeGen/Generic/BasicInstrs.llx on sparc by marking divrem
illegal.  Thanks to gabor for pointing this out!

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42832 91177308-0d34-0410-b5e6-96231b3b80d8

Renamed internal method "Create" of ImutAVLTree to "CreateNode".

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42825 91177308-0d34-0410-b5e6-96231b3b80d8

Correct swapped arguments to getConstant.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42824 91177308-0d34-0410-b5e6-96231b3b80d8

Fix 80-column violations

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42823 91177308-0d34-0410-b5e6-96231b3b80d8

Next PPC long double bits: ppcf128->i32 conversion.
Surprisingly complicated.
Adds getTargetNode for 2 outputs, no inputs (missing).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42822 91177308-0d34-0410-b5e6-96231b3b80d8

Bad choice of variable name.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42821 91177308-0d34-0410-b5e6-96231b3b80d8

Fix an extremely stupid bug that prevented first round of coalescing (physical registers only) from happening.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42820 91177308-0d34-0410-b5e6-96231b3b80d8

Modified XCode project to contain...

ADT/DenseSet.h
ADT/ImmutableMap.h
ADT/ImmutableSet.h

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42816 91177308-0d34-0410-b5e6-96231b3b80d8

Add new MemoryBuffer::getMemBufferCopy method.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42815 91177308-0d34-0410-b5e6-96231b3b80d8

Fix indentation.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42814 91177308-0d34-0410-b5e6-96231b3b80d8

Added implementation of immutable (functional) maps and sets, as
implemented on top of a functional AVL tree.  The AVL balancing code
is inspired by the OCaml implementation of Map, which also uses a functional
AVL tree.

Documentation is currently limited and cleanups are planned, but this code
compiles and has been tested.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42813 91177308-0d34-0410-b5e6-96231b3b80d8

Do not walk invalid iterator.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42812 91177308-0d34-0410-b5e6-96231b3b80d8