Fixed build warnings. Restoring changes done in 62600, they were lost in 62655.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@62681 91177308-0d34-0410-b5e6-96231b3b80d8

Cleanup whitespace and comments, and tweak some
prototypes, in operand type legalization.  No
functionality change.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@62680 91177308-0d34-0410-b5e6-96231b3b80d8

I accidentally removed this check in an earlier commit, which cause breakage in the pre alloc splitter.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@62678 91177308-0d34-0410-b5e6-96231b3b80d8

Implement LowerOperationWrapper for legalizer.
Also a few signed comparison fixes.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@62665 91177308-0d34-0410-b5e6-96231b3b80d8

CellSPU:
- Ensure that (operation) legalization emits proper FDIV libcall when needed.
- Fix various bugs encountered during llvm-spu-gcc build, along with various
  cleanups.
- Start supporting double precision comparisons for remaining libgcc2 build.
  Discovered interesting DAGCombiner feature, which is currently solved via
  custom lowering (64-bit constants are not legal on CellSPU, but DAGCombiner
  insists on inserting one anyway.)
- Update README.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@62664 91177308-0d34-0410-b5e6-96231b3b80d8

Allow targets to legalize operations (with illegal operands) that produces multiple values. For example, a load with an illegal operand (a load produces two values, a value and chain).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@62663 91177308-0d34-0410-b5e6-96231b3b80d8

Favors generating "not" over "xor -1". For example.
unsigned test(unsigned a) {
  return ~a;
}
llvm used to generate:
movl    $4294967295, %eax
xorl    4(%esp), %eax

Now it generates:
movl      4(%esp), %eax
notl      %eax

It's 3 bytes shorter.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@62661 91177308-0d34-0410-b5e6-96231b3b80d8

Disable on x86_64 until I figure out what's wrong.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@62660 91177308-0d34-0410-b5e6-96231b3b80d8

Make special cases (0 inf nan) work for frem.
Besides APFloat, this involved removing code
from two places that thought they knew the
result of frem(0., x) but were wrong.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@62645 91177308-0d34-0410-b5e6-96231b3b80d8

Be more aggressive about renumbering vregs after splitting them.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@62639 91177308-0d34-0410-b5e6-96231b3b80d8

Encode member accessibility.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@62638 91177308-0d34-0410-b5e6-96231b3b80d8

Appropriately mark fowrad decls.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@62625 91177308-0d34-0410-b5e6-96231b3b80d8

Need compile unit to find location.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@62624 91177308-0d34-0410-b5e6-96231b3b80d8

Calls to fmod, it turns out, are constant-folded by
invoking the host fmod, not by lowering to frem and
constant-folding that.  Fix this so it tests what I
want to test.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@62622 91177308-0d34-0410-b5e6-96231b3b80d8

Don't bother running the assembler, we don't know that it will be configured
for whatever llc defaults to.  This fixes PR3363

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@62619 91177308-0d34-0410-b5e6-96231b3b80d8

Fix PR3243: a LiveVariables bug. When HandlePhysRegKill is checking whether the last reference is also the last def (i.e. dead def), it should also check if last reference is the current machine instruction being processed. This can happen when it is processing a physical register use and setting the current machine instruction as sub-register's last ref.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@62617 91177308-0d34-0410-b5e6-96231b3b80d8

Fix typo.  Patch by Alexei Svitkine.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@62616 91177308-0d34-0410-b5e6-96231b3b80d8

Use "SINT_TO_FP" instead of "UINT_TO_FP" when getting the exponent. This was
causing the limited precision stuff to produce the wrong result for values in
the range [0, 1).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@62615 91177308-0d34-0410-b5e6-96231b3b80d8

Fix struct member's debug info.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@62610 91177308-0d34-0410-b5e6-96231b3b80d8

Add test case for PR3154.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@62604 91177308-0d34-0410-b5e6-96231b3b80d8

indentation...

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@62603 91177308-0d34-0410-b5e6-96231b3b80d8

Need only one set of debug info versions enum.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@62602 91177308-0d34-0410-b5e6-96231b3b80d8

Change TargetInstrInfo::isMoveInstr to return source and destination sub-register indices as well.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@62600 91177308-0d34-0410-b5e6-96231b3b80d8

zap white spaces.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@62598 91177308-0d34-0410-b5e6-96231b3b80d8

Fix global variable's address in a DIE.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@62596 91177308-0d34-0410-b5e6-96231b3b80d8

Enable debug info for enums.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@62594 91177308-0d34-0410-b5e6-96231b3b80d8

Add an IEEE remainder function, which is not
fully implemented yet and not used.  This is
mainly to clarify that APFloat::mod implements
C fmod, not remainder.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@62593 91177308-0d34-0410-b5e6-96231b3b80d8

improve compatibility with various versions of graphviz, patch by
Patrick Boettcher!

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@62592 91177308-0d34-0410-b5e6-96231b3b80d8

eliminate use of uint32_t to improve compatibility with cygwin

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@62591 91177308-0d34-0410-b5e6-96231b3b80d8

Eliminate use of uint32_t to improve compatibility with cygwin

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@62590 91177308-0d34-0410-b5e6-96231b3b80d8

Enable debug info for composite types.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@62589 91177308-0d34-0410-b5e6-96231b3b80d8

Check that the "don't barf on k8" fix is not
accidentally reverted again.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@62587 91177308-0d34-0410-b5e6-96231b3b80d8

Spacing

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@62584 91177308-0d34-0410-b5e6-96231b3b80d8

Temporarily XFAIL until this can be looked at. r62557 is what caused it to start failing.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@62578 91177308-0d34-0410-b5e6-96231b3b80d8

If a vector is empty, you're not allowed to access any
elements, even if it is only to take the address.  Test:
break-anti-dependencies.ll with ENABLE_EXPENSIVE_CHECKS.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@62576 91177308-0d34-0410-b5e6-96231b3b80d8

Refactor code. No functionality change.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@62573 91177308-0d34-0410-b5e6-96231b3b80d8

Testcase for limited precision stuff.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@62572 91177308-0d34-0410-b5e6-96231b3b80d8

Shift types need to match.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@62571 91177308-0d34-0410-b5e6-96231b3b80d8

another fix for PR3354

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@62561 91177308-0d34-0410-b5e6-96231b3b80d8

Add a README entry noticed while investigating PR3216.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@62558 91177308-0d34-0410-b5e6-96231b3b80d8

Fix a dagcombine to not generate loads of non-round integer types,
as its comment says, even in the case where it will be generating
extending loads. This fixes PR3216.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@62557 91177308-0d34-0410-b5e6-96231b3b80d8

Do not use DenseMap because the iterator is invalidated while constructing types. After all there was a reason why std::map was used initially!

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@62555 91177308-0d34-0410-b5e6-96231b3b80d8

Regenerate.

BUILT WITH WRONG VERSION OF AUTOCONF! Somebody please regenerate with an
approved version. Thanks!

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@62554 91177308-0d34-0410-b5e6-96231b3b80d8

Make the Interpreter use libffi if it's available. Patch from Alexei Svitkine!

This requires a rebuild of 'configure' itself. I will be committing that next, but
built with the wrong version of autoconf. Somebody who has the right one, please update
it.

As a side-note, because of the way autoconf works, all built tools will link against
libffi, not just lli. If you know how to fix this, please let me know ...

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@62553 91177308-0d34-0410-b5e6-96231b3b80d8

Make linear scan's trivial coalescer slightly more aggressive.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@62547 91177308-0d34-0410-b5e6-96231b3b80d8

Doxygen-ify comments.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@62546 91177308-0d34-0410-b5e6-96231b3b80d8

Verify debug info.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@62545 91177308-0d34-0410-b5e6-96231b3b80d8

Fix a problem exposed by PR3354: simplifycfg was making a potentially
trapping instruction be executed unconditionally.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@62541 91177308-0d34-0410-b5e6-96231b3b80d8

Remove SDNode's virtual destructor. This makes it impossible for
SDNode subclasses to keep state that requires non-trivial
destructors, however it was already effectively impossible,
since the destructor isn't actually ever called. There currently
aren't any SDNode subclasses affected by this, and in general
it's desireable to keep SDNode objects light-weight.

This eliminates the last virtual member function in the SDNode
class, so it eliminates the need for a vtable pointer, making
SDNode smaller.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@62539 91177308-0d34-0410-b5e6-96231b3b80d8

Move & restructure test per review.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@62538 91177308-0d34-0410-b5e6-96231b3b80d8

convert this to an unfoldable potentially trapping constant expr.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@62536 91177308-0d34-0410-b5e6-96231b3b80d8

improve compatibility with cygwin, patch by Jay Foad!

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@62535 91177308-0d34-0410-b5e6-96231b3b80d8

div/rem by zero and div/rem overflow are both undefined according to
langref.  Constant fold them to undef instead of trying to preserve
the trap.  This fixes PR3354.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@62534 91177308-0d34-0410-b5e6-96231b3b80d8

Fix SelectionDAG::ReplaceAllUsesWith to behave correctly when
uses are added to the From node while it is processing From's
use list, because of automatic local CSE. The fix is to avoid
visiting any new uses.

Fix a few places in the DAGCombiner that assumed that after
a RAUW call, the From node has no users and may be deleted.

This fixes PR3018.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@62533 91177308-0d34-0410-b5e6-96231b3b80d8

Fix PR3353, infinitely jump threading an infinite loop make from switches.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@62529 91177308-0d34-0410-b5e6-96231b3b80d8

compile-time fmod was done incorrectly.  PR 3316.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@62528 91177308-0d34-0410-b5e6-96231b3b80d8

DebugInfo is a lightweight APIs and consumers are expected to use light objects directly. There is no need to support isa<>, dyn_cast<> etc...

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@62527 91177308-0d34-0410-b5e6-96231b3b80d8

Verify Intrinsic::dbg_declare.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@62526 91177308-0d34-0410-b5e6-96231b3b80d8

Add targets to support the installation of clang in isolation.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@62522 91177308-0d34-0410-b5e6-96231b3b80d8

DIVREM isel deficiency: If sign bit is known zero, zero out DX/EDX/RDX instead of sign extending the low part (in AX/EAX/RAX) into it.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@62519 91177308-0d34-0410-b5e6-96231b3b80d8

Fix 80 col violations.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@62518 91177308-0d34-0410-b5e6-96231b3b80d8

Handle ISD::DECLARE with PIC relocation model.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@62516 91177308-0d34-0410-b5e6-96231b3b80d8

Port this test from dejagnu to unit testing.

The way this worked before was to test APInt by running
"lli -force-interpreter=true" knowing the lli uses APInt under the hood to
store its values. Now, we test APInt directly.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@62514 91177308-0d34-0410-b5e6-96231b3b80d8

Fix typo, sentence fragment.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@62512 91177308-0d34-0410-b5e6-96231b3b80d8

Adding your's truly to CREDITS.TXT

(This is also a test commit.)

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@62511 91177308-0d34-0410-b5e6-96231b3b80d8

Temporarily revert r62487. It's causing this error during a release bootstrap of
llvm-gcc. Most likely, it's miscompiling one of the "gen*" programs:

/Volumes/Sandbox/Buildbot/llvm/full-llvm/build/llvm-gcc.obj/./prev-gcc/xgcc -B/Volumes/Sandbox/Buildbot/llvm/full-llvm/build/llvm-gcc.obj/./prev-gcc/ -B/Volumes/Sandbox/Buildbot/llvm/full-llvm/build/llvm-gcc.install/i386-apple-darwin9.6.0/bin/ -c -g -O2 -mdynamic-no-pic -DIN_GCC -W -Wall -Wwrite-strings -Wstrict-prototypes -Wmissing-prototypes -mdynamic-no-pic -DHAVE_CONFIG_H -DGENERATOR_FILE -I. -Ibuild -I../../llvm-gcc.src/gcc -I../../llvm-gcc.src/gcc/build -I../../llvm-gcc.src/gcc/../include -I./../intl -I../../llvm-gcc.src/gcc/../libcpp/include  -I../../llvm-gcc.src/gcc/../libdecnumber -I../libdecnumber -I/Volumes/Sandbox/Buildbot/llvm/full-llvm/build/llvm.obj/include -I/Volumes/Sandbox/Buildbot/llvm/full-llvm/build/llvm.src/include -DENABLE_LLVM -I/Volumes/Sandbox/Buildbot/llvm/full-llvm/build/llvm.obj/../llvm.src/include  -D_DEBUG  -D_GNU_SOURCE -D__STDC_LIMIT_MACROS -D__STDC_CONSTANT_MACROS   -o build/gencondmd.o build/gencondmd.c
../../llvm-gcc.src/gcc/config/i386/mmx.md:926: error: expected '}' before ')' token
../../llvm-gcc.src/gcc/config/i386/mmx.md:926: warning: excess elements in struct initializer
../../llvm-gcc.src/gcc/config/i386/mmx.md:926: warning: (near initialization for 'insn_conditions[4]')
../../llvm-gcc.src/gcc/config/i386/mmx.md:926: error: expected '}' before ')' token
../../llvm-gcc.src/gcc/config/i386/mmx.md:926: error: expected ',' or ';' before ')' token
../../llvm-gcc.src/gcc/config/i386/mmx.md:927: error: expected identifier or '(' before ',' token
../../llvm-gcc.src/gcc/config/i386/sse.md:3458: error: expected identifier or '(' before ',' token
...

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@62506 91177308-0d34-0410-b5e6-96231b3b80d8

Minor tweak to LowerUINT_TO_FP_i32. Bias (after scalar_to_vector) has two uses so we should make it the second source operand of ISD::OR so 2-address pass won't have to be smart about commuting.

%reg1024<def> = MOVSDrm %reg0, 1, %reg0, <cp#0>, Mem:LD(8,8) [ConstantPool + 0]
%reg1025<def> = MOVSD2PDrr %reg1024
%reg1026<def> = MOVDI2PDIrm <fi#-1>, 1, %reg0, 0, Mem:LD(4,16) [FixedStack-1 + 0]
%reg1027<def> = ORPSrr %reg1025<kill>, %reg1026<kill>
%reg1028<def> = MOVPD2SDrr %reg1027<kill>
%reg1029<def> = SUBSDrr %reg1028<kill>, %reg1024<kill>
%reg1030<def> = CVTSD2SSrr %reg1029<kill>
MOVSSmr <fi#0>, 1, %reg0, 0, %reg1030<kill>, Mem:ST(4,4) [FixedStack0 + 0]
%reg1031<def> = LD_Fp32m80 <fi#0>, 1, %reg0, 0, Mem:LD(4,16) [FixedStack0 + 0]
RET %reg1031<kill>, %ST0<imp-use,kill>

The reason 2-addr pass isn't smart enough to commute the ORPSrr is because it can't look pass the MOVSD2PDrr instruction.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@62505 91177308-0d34-0410-b5e6-96231b3b80d8

Now not UINT_TO_FP is legal (it's marked custom), dag combiner won't
optimize it to a SINT_TO_FP when the sign bit is known zero. X86 isel should perform the optimization itself.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@62504 91177308-0d34-0410-b5e6-96231b3b80d8

Fix PR3016, a bug which can occur do to an invalid assumption:
we assumed a CFG structure that would be valid when all code in
the function is reachable, but not all code is necessarily
reachable.  Do a simple, but horrible, CFG walk to check for this
case.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@62487 91177308-0d34-0410-b5e6-96231b3b80d8

reduce indentation by using 'continue', no functionality change.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@62477 91177308-0d34-0410-b5e6-96231b3b80d8

Fix some problems in SpeculativelyExecuteBB.  Basically,
because of dead code, a phi could use the speculated instruction
that was not in "BB2".  Make this check explicit and tighten up
some other corners.  This fixes PR3292.  No testcase becauase this
depends entirely on visitation order of blocks and requires a
sequence of 8 passes to repro.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@62476 91177308-0d34-0410-b5e6-96231b3b80d8

Make this a bit more explicit about which cases need the
check.  No functionality change.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@62474 91177308-0d34-0410-b5e6-96231b3b80d8

Fix rdar://6505632, an llc crash on 483.xalancbmk

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@62470 91177308-0d34-0410-b5e6-96231b3b80d8

Few targets like PIC16 wants libcall generation for illegal type i16.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@62467 91177308-0d34-0410-b5e6-96231b3b80d8

CMake: Add lib/Analysis/CaptureTracking.cpp

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@62462 91177308-0d34-0410-b5e6-96231b3b80d8

BasicAliasAnalysis and FunctionAttrs were both
doing very similar pointer capture analysis.
Factor out the common logic.  The new version
is from FunctionAttrs since it does a better
job than the version in BasicAliasAnalysis

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@62461 91177308-0d34-0410-b5e6-96231b3b80d8

Simplify extract element based on comments from Duncan Sands.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@62459 91177308-0d34-0410-b5e6-96231b3b80d8

Forgot this in the previous checkin: fopen now has nocapture, realloc is
supposed to take two arguments.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@62457 91177308-0d34-0410-b5e6-96231b3b80d8

Fix copy and pasted typos that prevented strtok_r, realloc, getenv, ungetc,
putc, puts, perror, vscanf and vsscanf from getting annotations.

Add annotations for eight printf functions, memalign, pread and pwrite.

On Linux, llvm-gcc sometimes renames strdup, getc, putc, strtok_r, scanf and
sscanf. Match the alternate function names.

Fix a crash annotating opendir.

Don't mark fsetpos's second parameter as nocapture. It's supposed to be
captured.

Do mark fopen's path and mode strings as nocapture. Mark ferror as readonly,
but not fileno which may set errno.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@62456 91177308-0d34-0410-b5e6-96231b3b80d8

add a comment

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@62436 91177308-0d34-0410-b5e6-96231b3b80d8

switch over some other methods from indices to iterators

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@62430 91177308-0d34-0410-b5e6-96231b3b80d8

make comparisons a bist faster

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@62428 91177308-0d34-0410-b5e6-96231b3b80d8

Remove tabs.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@62423 91177308-0d34-0410-b5e6-96231b3b80d8

Refactor code

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@62421 91177308-0d34-0410-b5e6-96231b3b80d8

Testcase for last commit.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@62418 91177308-0d34-0410-b5e6-96231b3b80d8

Extend thi

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@62415 91177308-0d34-0410-b5e6-96231b3b80d8

Fix MatchAddress bug that's preventing negative displacement from being folded in 64-bit mode.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@62413 91177308-0d34-0410-b5e6-96231b3b80d8

Assign argument type to appropriate DIE.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@62412 91177308-0d34-0410-b5e6-96231b3b80d8

Remove dead code.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@62410 91177308-0d34-0410-b5e6-96231b3b80d8

Disable composite type debug info for now.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@62406 91177308-0d34-0410-b5e6-96231b3b80d8

Temporarily revert my last change. It is causing a bootstrap failure.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@62405 91177308-0d34-0410-b5e6-96231b3b80d8

Implement a special algorithm for converting uint_to_fp for i32 values on
X86. This code:

void f() {
  uint32_t x;
  float y = (float)x;
}

used to be:

     movl     %eax, -8(%ebp)
     movl     [2^52 double], -4(%ebp)
     movsd    -8(%ebp), %xmm0
     subsd    [2^52 double], %xmm0
     cvtsd2ss %xmm0, %xmm0

Is now:

   movsd        [2^52 double], %xmm0
   movsd        %xmm0, %xmm1
   movd         %ecx, %xmm2
   orps         %xmm2, %xmm1
   subsd        %xmm0, %xmm1
   cvtsd2ss     %xmm1, %xmm0

This is faster on X86. Note that there's an extra load of %xmm0 into %xmm1. That
will be fixed in a later coalescer fix.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@62404 91177308-0d34-0410-b5e6-96231b3b80d8

CMake: Add lib/Target/IA64/IA64Subtarget.cpp

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@62394 91177308-0d34-0410-b5e6-96231b3b80d8

speed up iterative loop by using iterators. changes direction, but functionally equivalent
if this works out, I'll change the others next.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@62385 91177308-0d34-0410-b5e6-96231b3b80d8

introduce typedef for complicated vector, and use it too

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@62384 91177308-0d34-0410-b5e6-96231b3b80d8

Simplify extract element of a scalar to vector.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@62383 91177308-0d34-0410-b5e6-96231b3b80d8

typo

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@62377 91177308-0d34-0410-b5e6-96231b3b80d8

Fix PPC ISD::Declare isel and eliminate the need for PPCTargetLowering::LowerGlobalAddress to check if isVerifiedDebugInfoDesc() is true. Given the recent changes, it would falsely return true for a lot of GlobalAddressSDNode's.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@62373 91177308-0d34-0410-b5e6-96231b3b80d8

Support for multi-valued options in CommandLine

Makes possible to specify options that take multiple arguments (a-la
-sectalign on Darwin). See documentation for details.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@62372 91177308-0d34-0410-b5e6-96231b3b80d8

Give IA64 a TargetSubtarget subclass, so that it can
implement getSubtargetImpl.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@62369 91177308-0d34-0410-b5e6-96231b3b80d8