Reapply r79708 with the appropriate fix for the case that still requires locking.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@79731 91177308-0d34-0410-b5e6-96231b3b80d8

TOT is 2.7svn for CMake, too

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@79730 91177308-0d34-0410-b5e6-96231b3b80d8

Change SmallString::operator{=,+=} to take a StringRef.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@79729 91177308-0d34-0410-b5e6-96231b3b80d8

Regenerate configure for 2.7svn

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@79726 91177308-0d34-0410-b5e6-96231b3b80d8

TOT is now 2.7svn

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@79725 91177308-0d34-0410-b5e6-96231b3b80d8

revert r79708 + r79711

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@79720 91177308-0d34-0410-b5e6-96231b3b80d8

Make x86 test actually test x86 code generation.  Fix the
construct on ARM, which was breaking by coincidence, and add a similar
testcase for ARM.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@79719 91177308-0d34-0410-b5e6-96231b3b80d8

Add new intrinsics for Neon vldN_lane and vstN_lane operations.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@79716 91177308-0d34-0410-b5e6-96231b3b80d8

Note down a few bits and bobs off the top of my head for ReleaseNotes-2.6.html

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@79715 91177308-0d34-0410-b5e6-96231b3b80d8

Add basic information about SJLJ EH

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@79714 91177308-0d34-0410-b5e6-96231b3b80d8

rename test, make more specific.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@79712 91177308-0d34-0410-b5e6-96231b3b80d8

Actually remove unused static. Previous commit removed trailing
whitespace.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@79711 91177308-0d34-0410-b5e6-96231b3b80d8

Remove unused static.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@79710 91177308-0d34-0410-b5e6-96231b3b80d8

Ease contention on this lock by noticing that all writes to the VTs array will
be of (dynamically) constant values, so races on it are immaterial.  We just need
to ensure that at least one write has completed before return the pointer into it.

With this change, parllc exhibits essentially no overhead on 403.gcc.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@79708 91177308-0d34-0410-b5e6-96231b3b80d8

Add missing RUN line

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@79707 91177308-0d34-0410-b5e6-96231b3b80d8

Reduce the test

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@79703 91177308-0d34-0410-b5e6-96231b3b80d8

Use CHECK-NEXT to make sure we're only getting one copy of each shuffle
instruction.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@79702 91177308-0d34-0410-b5e6-96231b3b80d8

Make unit-at-a-time on by default to match the behavior of llvm-gcc.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@79698 91177308-0d34-0410-b5e6-96231b3b80d8

llvm-mc: Improve handling of implicit alignment for magic section directives
(e.g., .objc_message_refs).
 - Just emit a .align when we see the directive; this isn't exactly what 'as'
   does but in practice it should be ok, at least for now. See FIXME.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@79697 91177308-0d34-0410-b5e6-96231b3b80d8

Kill trailing whitespace.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@79696 91177308-0d34-0410-b5e6-96231b3b80d8

remove the directory too

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@79692 91177308-0d34-0410-b5e6-96231b3b80d8

Revert r79563

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@79691 91177308-0d34-0410-b5e6-96231b3b80d8

revert r79562 + r79563

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@79690 91177308-0d34-0410-b5e6-96231b3b80d8

Make MSVC happy.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@79689 91177308-0d34-0410-b5e6-96231b3b80d8

Fix typo.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@79688 91177308-0d34-0410-b5e6-96231b3b80d8

revert r79631

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@79686 91177308-0d34-0410-b5e6-96231b3b80d8

revert 79631

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@79685 91177308-0d34-0410-b5e6-96231b3b80d8

llvm-mc/Mach-O: Support .o emission for .org and .align.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@79684 91177308-0d34-0410-b5e6-96231b3b80d8

llvm-mc: In a .fill directive, still honor .align even if invalid maximum bytes
count is given (this matches 'as').

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@79683 91177308-0d34-0410-b5e6-96231b3b80d8

Implement APInt <-> APFloat conversion for IEEE 128-bit floats.
This fixes PR2555

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@79677 91177308-0d34-0410-b5e6-96231b3b80d8

Rename ARM "lane_cst" operands to "nohash_imm" since they are used for
several things other than Neon vector lane numbers.  For inline assembly
operands with a "c" print code, check that they really are immediates.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@79676 91177308-0d34-0410-b5e6-96231b3b80d8

Fix a bug where the DWARF emitter in the JIT was not initializing alignment
bytes.  libgcc doesn't seem to mind, but if you pass this DWARF to GDB, it
doesn't like it.

Also make the JIT memory manager to initialize it's memory to garbage in debug
mode, so that it's easier to find bugs like these in the future.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@79674 91177308-0d34-0410-b5e6-96231b3b80d8

Match VTRN, VZIP, and VUZP shuffles.  Restore the tests for these operations,
now using shuffles instead of intrinsics.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@79673 91177308-0d34-0410-b5e6-96231b3b80d8

Add fcopysign instructions

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@79664 91177308-0d34-0410-b5e6-96231b3b80d8

Try again at privatizing the layout info map, with a rewritten patch.
This preserves the existing behavior much more closely than my previous attempt.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@79663 91177308-0d34-0410-b5e6-96231b3b80d8

Expand few nodes until someone will be crazy enough to implement them natively :)

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@79659 91177308-0d34-0410-b5e6-96231b3b80d8

Typo :(

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@79657 91177308-0d34-0410-b5e6-96231b3b80d8

Correct instruction names for subtract-with-borrow

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@79656 91177308-0d34-0410-b5e6-96231b3b80d8

llvm-mc/Mach-O: Support byte and fill value emission.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@79652 91177308-0d34-0410-b5e6-96231b3b80d8

Don't assume that the operand of an inttoptr is an pointer-sized integer.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@79651 91177308-0d34-0410-b5e6-96231b3b80d8

Fix typo. Should check both values of RangeUse for 0. Patch by Marius Wachtler.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@79649 91177308-0d34-0410-b5e6-96231b3b80d8

Handle 'r' inline asm constraint

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@79648 91177308-0d34-0410-b5e6-96231b3b80d8

Fix a comment.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@79643 91177308-0d34-0410-b5e6-96231b3b80d8

Fix a problem noticed by gcc-4.4:
  warning: comparison is always true due to limited range of data type.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@79642 91177308-0d34-0410-b5e6-96231b3b80d8

Change getelementptr folding to use APInt instead of uint64_t for
offset computations. This fixes a truncation bug on targets that
don't have 64-bit pointers.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@79639 91177308-0d34-0410-b5e6-96231b3b80d8

Add some tests for vext.16 and vext.32.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@79638 91177308-0d34-0410-b5e6-96231b3b80d8

Fix -Asserts warning.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@79636 91177308-0d34-0410-b5e6-96231b3b80d8

llvm-mc: Accept .fill size of 8.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@79635 91177308-0d34-0410-b5e6-96231b3b80d8

Fix a typo

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@79634 91177308-0d34-0410-b5e6-96231b3b80d8

Removed profiling test, lli not available on all platforms.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@79633 91177308-0d34-0410-b5e6-96231b3b80d8

Add a pass to do call graph analyis to overlay the autos and frame sections of
leaf functions. This pass will be extended to color other nodes of the call tree
as well in future.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@79631 91177308-0d34-0410-b5e6-96231b3b80d8

Fix build on platforms where stdint.h is not implicitly included in the standard headers.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@79629 91177308-0d34-0410-b5e6-96231b3b80d8

More cpp backend fixes. Now for FP stuff.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@79626 91177308-0d34-0410-b5e6-96231b3b80d8

Fix some typos and use type-based isel for VZIP/VUZP/VTRN

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@79625 91177308-0d34-0410-b5e6-96231b3b80d8

Add lowering of ARM 4-element shuffles to multiple instructios via perfectshuffle-generated table.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@79624 91177308-0d34-0410-b5e6-96231b3b80d8

Add simple cost model to perfect shuffle. Currently we're doing just greedy search for cost instead of (proper) dynamic programming approach.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@79623 91177308-0d34-0410-b5e6-96231b3b80d8

Add nodes & dummy matchers for some v{zip,uzp,trn} instructions

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@79622 91177308-0d34-0410-b5e6-96231b3b80d8

Expand EXTRACT_SUBVECTOR

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@79621 91177308-0d34-0410-b5e6-96231b3b80d8

Provide vext.{16,32}

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@79620 91177308-0d34-0410-b5e6-96231b3b80d8

Use masks not nodes for vector shuffle predicates. Provide set of 'legal' masks, so legalizer won't infinite cycle

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@79619 91177308-0d34-0410-b5e6-96231b3b80d8

Whitespace cleanup

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@79618 91177308-0d34-0410-b5e6-96231b3b80d8

Update CMakeLists.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@79617 91177308-0d34-0410-b5e6-96231b3b80d8

bug 4530: Make debug information static to it is preservered during bitcode linking.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@79616 91177308-0d34-0410-b5e6-96231b3b80d8

Added tests for Profiling Infrastructure.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@79615 91177308-0d34-0410-b5e6-96231b3b80d8

Make for 32-bit only.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@79614 91177308-0d34-0410-b5e6-96231b3b80d8

Tweak svn:ignore

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@79613 91177308-0d34-0410-b5e6-96231b3b80d8

llvm-mc: Start MCAssembler and MCMachOStreamer.

 - Together these form the (Mach-O) back end of the assembler.

 - MCAssembler is the actual assembler backend, which is designed to have a
   reasonable API. This will eventually grow to support multiple object file
   implementations, but for now its Mach-O/i386 only.

 - MCMachOStreamer adapts the MCStreamer "actions" API to the MCAssembler API,
   e.g. converting the various directives into fragments, managing state like
   the current section, and so on.

 - llvm-mc will use the new backend via '-filetype=obj', which may eventually
   be, but is not yet, since I hear that people like assemblers which actually
   assemble.

 - The only thing that works at the moment is changing sections. For the time
   being I have a Python Mach-O dumping tool in test/scripts so this stuff can
   be easily tested, eventually I expect to replace this with a real LLVM tool.

 - More doxyments to come.

I assume that since this stuff doesn't touch any of the things which are part of
2.6 that it is ok to put this in not so long before the freeze, but if someone
objects let me know, I can pull it.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@79612 91177308-0d34-0410-b5e6-96231b3b80d8

llvm-mc: Various section parsing fixes.
 - Add missing flags for various Objective-C sections.

 - Fix names for [non_]lazy_symbol_pointer (these are misspelled in the manual).

 - .symbol_stub does not have the self modifying code flag set (this appears to
   be wrong in the manual?).

 - Add implicit alignment values; not yet used.

Also, call MCStreamer::Finish at the end of a successful parse.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@79611 91177308-0d34-0410-b5e6-96231b3b80d8

Probably should emit bitcode instead of text.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@79607 91177308-0d34-0410-b5e6-96231b3b80d8

Don't install llvm-mc by default.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@79604 91177308-0d34-0410-b5e6-96231b3b80d8

Remove #include <iostream>.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@79603 91177308-0d34-0410-b5e6-96231b3b80d8

Clean up the APInt function getDigit.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@79602 91177308-0d34-0410-b5e6-96231b3b80d8

Correct for recent assert change.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@79601 91177308-0d34-0410-b5e6-96231b3b80d8

Testcase for r79597.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@79598 91177308-0d34-0410-b5e6-96231b3b80d8

Update error messages for '+'. Fix grammar and make the two
negative checks resemble each other.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@79595 91177308-0d34-0410-b5e6-96231b3b80d8

Fix trailing whitespace and 80-col violation.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@79594 91177308-0d34-0410-b5e6-96231b3b80d8

Fix bug with APInt::getBitsNeeded with for base 10 numbers 0-9.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@79593 91177308-0d34-0410-b5e6-96231b3b80d8

Allow '+' to appear in APInt strings, and add more unit tests.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@79592 91177308-0d34-0410-b5e6-96231b3b80d8

Remove Neon intrinsics for VZIP, VUZP, and VTRN.  We will represent these as
vector shuffles.  Temporarily remove the tests for these operations until the
new implementation is working.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@79579 91177308-0d34-0410-b5e6-96231b3b80d8

Re-revert r79555.  Apparently it's not just buildbot weirdness.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@79578 91177308-0d34-0410-b5e6-96231b3b80d8

Add support for including '+' in APFloat strings, more asserts,
and many new unit tests.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@79574 91177308-0d34-0410-b5e6-96231b3b80d8

Reapply r79555 for testing.  Daniel's trying to work out some buildbot weirdnesss.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@79572 91177308-0d34-0410-b5e6-96231b3b80d8

Use FileCheck even though this means testing for something
that has nothing to do with the point of the test, per Chris.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@79569 91177308-0d34-0410-b5e6-96231b3b80d8

--- Reverse-merging r79555 into '.':
U    include/llvm/Target/TargetData.h
U    lib/Target/TargetData.cpp

Temporarily revert 79555. It was causing hangs and test failures.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@79568 91177308-0d34-0410-b5e6-96231b3b80d8

Attempt to comment this code more.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@79567 91177308-0d34-0410-b5e6-96231b3b80d8

Suppress build warning in -Asserts

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@79564 91177308-0d34-0410-b5e6-96231b3b80d8

part of the previous commit for PIC16 ISR implementation.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@79563 91177308-0d34-0410-b5e6-96231b3b80d8

Implement support for ISRs.
Clone functions that are shared between the Main thread and Interrupt thread.
CallSites are changed in AsmPrinter currently. A better solution would have been to modify the legalizer (SoftenFloat) to allow targets to change the name of libcalls for float operations. But that currently breaks other targets.
Also, cloing of automatic variables is done AsmPrinter, a better approach would
be to use the ValueMap in CloneFunction itself.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@79562 91177308-0d34-0410-b5e6-96231b3b80d8

Reduce contention on the Attributes lock by using atomic operations for reference counting rather than locking.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@79560 91177308-0d34-0410-b5e6-96231b3b80d8

Make the StructType->StructLayout table private to TargetData, allowing us to avoid locking on it.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@79555 91177308-0d34-0410-b5e6-96231b3b80d8

Fixed PCMPESTRM128 to have opcode 0x60 instead of 0x62, as specified by the
Intel documentation.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@79554 91177308-0d34-0410-b5e6-96231b3b80d8

Fix an x86 code size regression: prefer RIP-relative addressing
over absolute addressing even in non-PIC mode (unless the address
has an index or something else incompatible), because it has a
smaller encoding.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@79553 91177308-0d34-0410-b5e6-96231b3b80d8

Add a comment explaining why llvm_unreachable_internal doesn't call
the ErrorHandler callback.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@79541 91177308-0d34-0410-b5e6-96231b3b80d8

Fix two APFloat bugs in converting hexadecimal constants.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@79540 91177308-0d34-0410-b5e6-96231b3b80d8

Rename hasNoUnsignedOverflow and hasNoSignedOverflow to hasNoUnsignedWrap
and hasNoSignedWrap, for consistency with the nuw and nsw properties.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@79539 91177308-0d34-0410-b5e6-96231b3b80d8

Clarify a few things in the inbounds description.

Describing inbounds in terms of infinitely precise arithmetic prohibits
the assumption address-space-sized integer arithmetic will wrap, with
a wrapped-around value landing within the same object.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@79538 91177308-0d34-0410-b5e6-96231b3b80d8

Fixed header comment.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@79536 91177308-0d34-0410-b5e6-96231b3b80d8

Fix an obvious copy-n-paste bug.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@79535 91177308-0d34-0410-b5e6-96231b3b80d8

Use FileCheck for the test run where it's appropriate.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@79534 91177308-0d34-0410-b5e6-96231b3b80d8

Various comment and whitespace cleanups.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@79533 91177308-0d34-0410-b5e6-96231b3b80d8