Moved some more index operations over to LiveIntervals.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@81605 91177308-0d34-0410-b5e6-96231b3b80d8

Added the WAIT instruction to the Intel tables,
for the purposes of the disassembler.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@81603 91177308-0d34-0410-b5e6-96231b3b80d8

Added CMPS (string comparison) instructions for all
operand widths to the Intel instruction tables, for
the purposes of the disassembler.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@81601 91177308-0d34-0410-b5e6-96231b3b80d8

Remove -new-coalescer-heuristic. It's not useful.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@81600 91177308-0d34-0410-b5e6-96231b3b80d8

80 col violations.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@81598 91177308-0d34-0410-b5e6-96231b3b80d8

fix another GCC bootstrap problem, which manifested as things
like:
foo.s:2412:non-relocatable subtraction expression, "_gomp_tls_key" minus "L1$pb"

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@81596 91177308-0d34-0410-b5e6-96231b3b80d8

fix an embarassing typo that resulted in llvm-gcc bootstrap miscompare
because the sorting wasn't sorting.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@81592 91177308-0d34-0410-b5e6-96231b3b80d8

Added SCAS instructions in their 8, 16, 32, and
64-bit variants for the disassembler.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@81591 91177308-0d34-0410-b5e6-96231b3b80d8

Fix -Asserts warning.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@81580 91177308-0d34-0410-b5e6-96231b3b80d8

Update CMake files.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@81577 91177308-0d34-0410-b5e6-96231b3b80d8

De-bork CMake build. llvm-extract depends on asmparser

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@81574 91177308-0d34-0410-b5e6-96231b3b80d8

Fix llvm-extract's "writing bitcode to a terminal" warning, which wasn't
working. To support this, add an is_displayed() function to raw_ostream,
and generalize Process::StandardOutIsDisplayed and friends in order to
support it.

Also, call RemoveFileOnSignal before creating a file instead of after, so
that the file isn't left behind if the program is interrupted between when
the file is created and RemoveFileOnSignal is called.

While here, add a -S to llvm-extract and port it to IRReader so that it
supports assembly input.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@81568 91177308-0d34-0410-b5e6-96231b3b80d8

Revert array initialization regclass change so that the initialization stays static, not runtime.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@81560 91177308-0d34-0410-b5e6-96231b3b80d8

Update register class references to use the global constant ARM::*RegisterClass names.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@81556 91177308-0d34-0410-b5e6-96231b3b80d8

Added ADC, SUB, SBB, and OR instructions that operate
on rAX and an immediate.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@81551 91177308-0d34-0410-b5e6-96231b3b80d8

Fix pr4820: Don't run llvm-config during "make clean" since it may have
already been removed.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@81547 91177308-0d34-0410-b5e6-96231b3b80d8

Remove an unnecessary -f.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@81546 91177308-0d34-0410-b5e6-96231b3b80d8

Convert more tests to avoid llvm-as.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@81545 91177308-0d34-0410-b5e6-96231b3b80d8

fix pasto

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@81544 91177308-0d34-0410-b5e6-96231b3b80d8

Don't generate Dwarf line table entries for source line 0.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@81542 91177308-0d34-0410-b5e6-96231b3b80d8

fix some fixmes: emit stubs in sorted order.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@81541 91177308-0d34-0410-b5e6-96231b3b80d8

Eliminate more redundant llvm-as calls.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@81540 91177308-0d34-0410-b5e6-96231b3b80d8

Fix this test to test what it was originally intended to test.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@81539 91177308-0d34-0410-b5e6-96231b3b80d8

give densemap iterators real iterator traits.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@81538 91177308-0d34-0410-b5e6-96231b3b80d8

Change tests from "opt %s" to "opt < %s" so that opt doesn't see the
input filename so that opt doesn't print the input filename in the
output so that grep lines in the tests don't unintentionally match
strings in the input filename.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@81537 91177308-0d34-0410-b5e6-96231b3b80d8

Fix PR4948 (and a leak): by not destroying the DwarfException
object, the timer it creates was not being deleted.  Since the
timer belonged to a static timer group, the timer group would
be destroyed on shutdown, and would notice and complain that
not all timers it contained were destroyed.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@81533 91177308-0d34-0410-b5e6-96231b3b80d8

turn on -experimental-asm-printer for x86 / AT&T by default.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@81532 91177308-0d34-0410-b5e6-96231b3b80d8

another random update

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@81531 91177308-0d34-0410-b5e6-96231b3b80d8

reject attempts to take the address of an intrinsic, PR4949.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@81530 91177308-0d34-0410-b5e6-96231b3b80d8

fix a bunch of spurious failures for people whose home directory
is sabre.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@81528 91177308-0d34-0410-b5e6-96231b3b80d8

this test is using invalid "intrinsics".

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@81527 91177308-0d34-0410-b5e6-96231b3b80d8

fix test to not get a moduleid that matches 'br'

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@81526 91177308-0d34-0410-b5e6-96231b3b80d8

default construct MCInst's ctor to 0, which is "PHI" which is invalid for MCInsts.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@81525 91177308-0d34-0410-b5e6-96231b3b80d8

... and fix the REQUIRES_RTTI condition.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@81524 91177308-0d34-0410-b5e6-96231b3b80d8

Fix REQUIRES_RTTI while awake.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@81523 91177308-0d34-0410-b5e6-96231b3b80d8

Make REQUIRES_RTTI work.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@81522 91177308-0d34-0410-b5e6-96231b3b80d8

Eliminate some unnecessary implicit constructors in generated DAG ISel code.

Partial fix for PR4946.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@81518 91177308-0d34-0410-b5e6-96231b3b80d8

Bugfix. Sorry.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@81517 91177308-0d34-0410-b5e6-96231b3b80d8

Make ProfileEstimator even more robust on general CFGs.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@81516 91177308-0d34-0410-b5e6-96231b3b80d8

switch HiddenGVStubs to be a DenseMap instead of a string map, mirroring FnStubs and GVStubs.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@81514 91177308-0d34-0410-b5e6-96231b3b80d8

Fix a bug I introduced in FnStubs generation, switch GVStubs to be a
densemap instead of StringMap to match FnStubs.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@81513 91177308-0d34-0410-b5e6-96231b3b80d8

change FnStubs from being a StringMap<std::string> to being a much
more efficient SmallPtrSet<MCSymbol*>.  This eliminates string
craziness and fixes CodeGen/X86/darwin-quote.ll with the new asmprinter.

Codegen is producing stubs in a nondeterminstic order, but it was doing
this before anyway.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@81511 91177308-0d34-0410-b5e6-96231b3b80d8

printInstruction() no longer prints a \n after itself, do it
for the two instruction MOVPC32r sequence.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@81509 91177308-0d34-0410-b5e6-96231b3b80d8

reimplement X86ATTAsmPrinter::GetGlobalAddressSymbol in terms of
Mangler::getNameWithPrefix.  In addition to avoiding some over
quoting, this also is more efficient because it uses smallvector
instead of std::string thrashing.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@81508 91177308-0d34-0410-b5e6-96231b3b80d8

fix prefix ordering, it's L_foo not _Lfoo

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@81506 91177308-0d34-0410-b5e6-96231b3b80d8

add a new Mangler::getNameWithPrefix API which returns the
(uniqued if unnamed) global variable name with the prefix that
it is supposed to get.  It doesn't do "mangling" in the sense of
adding quotes and hacking on bad characters.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@81505 91177308-0d34-0410-b5e6-96231b3b80d8

convert X86ATTAsmPrinter::GetExternalSymbolSymbol to use SmallString
instead of std::string and Mangler.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@81503 91177308-0d34-0410-b5e6-96231b3b80d8

rearrange some code, export a SmallString version of DecorateCygMingName.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@81502 91177308-0d34-0410-b5e6-96231b3b80d8

more typos

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@81499 91177308-0d34-0410-b5e6-96231b3b80d8

Follow up to 81494. When the folded reload is narrowed to a 32-bit load then change the destination register to a 32-bit one or add a sub-register index.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@81496 91177308-0d34-0410-b5e6-96231b3b80d8

PHI nodes can never reach the asmprinter, assert and die instead of printing
out an illegal "PHINODE" instruction.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@81495 91177308-0d34-0410-b5e6-96231b3b80d8

It's not legal to fold a load from a narrower stack slot into a wider instruction. If done, the instruction does a 64-bit load and that's not
safe. This can happen we a subreg_to_reg 0 has been coalesced. One
exception is when the instruction that folds the load is a move, then we
can simply turn it into a 32-bit load from the stack slot.

rdar://7170444

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@81494 91177308-0d34-0410-b5e6-96231b3b80d8

Make fast-isel try ISD::FNEG before resorting to bitcasts and xors.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@81493 91177308-0d34-0410-b5e6-96231b3b80d8

Reapply r81171 with a fix: don't try to use i64 when it
isn't legal.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@81492 91177308-0d34-0410-b5e6-96231b3b80d8

Fix indentation.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@81484 91177308-0d34-0410-b5e6-96231b3b80d8

Teach lib/VMCore/ConstantFold.cpp how to set the inbounds keyword and
how to fold notionally-out-of-bounds array getelementptr indices instead
of just doing these in lib/Analysis/ConstantFolding.cpp, because it can
be done in a fairly general way without TargetData, and because not all
constants are visited by lib/Analysis/ConstantFolding.cpp. This enables
more constant folding.

Also, set the "inbounds" flag when the getelementptr indices are
one-past-the-end.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@81483 91177308-0d34-0410-b5e6-96231b3b80d8

Factor out the code for checking that all indices in a getelementptr are
within the notional bounds of the static type of the getelementptr (which
is not the same as "inbounds") from GlobalOpt into a utility routine,
and use it in ConstantFold.cpp to check whether there are any mis-behaved
indices.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@81478 91177308-0d34-0410-b5e6-96231b3b80d8

Give these files top-level comments that describe the current code.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@81473 91177308-0d34-0410-b5e6-96231b3b80d8

Fix whitespaces.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@81468 91177308-0d34-0410-b5e6-96231b3b80d8

Fix validation errors.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@81466 91177308-0d34-0410-b5e6-96231b3b80d8

Fix validation errors.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@81465 91177308-0d34-0410-b5e6-96231b3b80d8

Don't swap the operands of a subtraction when trying to create a
post-decrement load/store.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@81464 91177308-0d34-0410-b5e6-96231b3b80d8

Fix uppercaseo.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@81463 91177308-0d34-0410-b5e6-96231b3b80d8

Added the ParseInstruction() hook for target specific assembler directives so
that things like .word can be parsed as target specific.  Moved parsing .word
out of AsmParser.cpp into X86AsmParser.cpp as it is 2 bytes on X86 and 4 bytes
for other targets that support the .word directive.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@81461 91177308-0d34-0410-b5e6-96231b3b80d8

Fit code within 80 columns

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@81459 91177308-0d34-0410-b5e6-96231b3b80d8

Added XOR instructions for rAX and immediates of
various widths.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@81458 91177308-0d34-0410-b5e6-96231b3b80d8

Added MOV instructions between rAX and memory offsets,
including segment offsets and (for 8-bit operands)
absolute offsets.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@81457 91177308-0d34-0410-b5e6-96231b3b80d8

Added a variety of PUSH and POP instructions, including
ones capable of accessing R/M operands instead of just
registers.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@81456 91177308-0d34-0410-b5e6-96231b3b80d8

Exit early if exception handling isn't supported.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@81454 91177308-0d34-0410-b5e6-96231b3b80d8

Allow llvmc to take .bc files as input.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@81452 91177308-0d34-0410-b5e6-96231b3b80d8

Make ProfileEstimator more robust on general CFGs.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@81450 91177308-0d34-0410-b5e6-96231b3b80d8

add another broken version of gcc, thanks to Niels Moller for pointing this out.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@81449 91177308-0d34-0410-b5e6-96231b3b80d8

Distinguish between 'empty' and empty DAG marker.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@81448 91177308-0d34-0410-b5e6-96231b3b80d8

Add a logical 'not' operator to llvmc's TableGen dialect.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@81447 91177308-0d34-0410-b5e6-96231b3b80d8

fix leakage of Module

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@81445 91177308-0d34-0410-b5e6-96231b3b80d8

Add some braces to make newer GCCs happy and update CMakeLists.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@81443 91177308-0d34-0410-b5e6-96231b3b80d8

Cleaned up code by factoring out common portions of edge loading into function.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@81438 91177308-0d34-0410-b5e6-96231b3b80d8

Correctly handle the case where a comparison is created in one BasicBlock and
used by a terminator in another.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@81437 91177308-0d34-0410-b5e6-96231b3b80d8

Comment and whitespace cleanups. No intentional functionality change.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@81436 91177308-0d34-0410-b5e6-96231b3b80d8

Revert part of my r81424 patch. I removed what looked like superfluous padding
from the exception tables. However, Duncan explained why it's a can of worms to
do it the GCC way. I went back to doing it the LLVM way and added Duncan's
explanation so that I don't do this again in the future.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@81434 91177308-0d34-0410-b5e6-96231b3b80d8

Add a test case for r81431.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@81432 91177308-0d34-0410-b5e6-96231b3b80d8

Remove prcontext.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@81427 91177308-0d34-0410-b5e6-96231b3b80d8

Add malloc call utility functions. Patch by Victor Hernandez.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@81426 91177308-0d34-0410-b5e6-96231b3b80d8

Don't hardcode the TType format size. In fact, rework the code so that it's more
like what GCC outputs. The mysterious code to insert padding wasn't in GCC at
all. I modified the TType base offset code to calculate the offset like GCC
does, though.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@81424 91177308-0d34-0410-b5e6-96231b3b80d8

Proper support of non-lazy indirect symbols.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@81422 91177308-0d34-0410-b5e6-96231b3b80d8

Remove the "#if 0" that Noone loved. It wasn't really necessary, because the
code within it was the same inside and out. There's still a problem of the
TypeInfoSize should be the size of the TType format encoding (at least that's
what GCC thinks it should be).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@81417 91177308-0d34-0410-b5e6-96231b3b80d8

MC: Give target specific parsers access to the MCStreamer.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@81416 91177308-0d34-0410-b5e6-96231b3b80d8

Revert r81171 which was causing pr4927.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@81415 91177308-0d34-0410-b5e6-96231b3b80d8

Pull check for SJLJ EH into a boolean and use that.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@81409 91177308-0d34-0410-b5e6-96231b3b80d8

Use the SizeOfEncodedValue function instead of magic variables for the
sizeof(DW_EH_PE_udata4).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@81408 91177308-0d34-0410-b5e6-96231b3b80d8

Add helpful comment.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@81406 91177308-0d34-0410-b5e6-96231b3b80d8

Believe it or not, this is a simplification. :-)

Basically, this patch is working towards removing the hard-coded values that are
output for the CIE. In particular, the CIE augmentation and the CIE augmentation
size. Both of these should be calculated. In the process, I was able to make a
bunch of code simpler.

The encodings for the personality, LSDA, and FDE in the CIE are still not
correct. They should be generated either from target-specific callbacks (blech!)
or grokked from first-principles.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@81404 91177308-0d34-0410-b5e6-96231b3b80d8

Fix double load / store multiple encoding.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@81403 91177308-0d34-0410-b5e6-96231b3b80d8

canonicalize namespace gymnastics

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@81402 91177308-0d34-0410-b5e6-96231b3b80d8

Testcase for (llvm-gcc) 81399.  Adjust an older
test case to allow for different, functionally
identical output.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@81400 91177308-0d34-0410-b5e6-96231b3b80d8

Fix pr4939: Change FPCCToARMCC to translate SETOLE to ARMCC::LS.
See the bug report for details.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@81397 91177308-0d34-0410-b5e6-96231b3b80d8

remove DebugLoc from MCInst and eliminate "Comment printing" from
the MCInst path of the asmprinter.  Instead, pull comment printing
out of the autogenerated asmprinter into each target that uses the
autogenerated asmprinter.  This causes code duplication into each
target, but in a way that will be easier to clean up later when more
asmprinter stuff is commonized into the base AsmPrinter class.

This also fixes an xcore strangeness where it inserted two tabs
before every instruction.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@81396 91177308-0d34-0410-b5e6-96231b3b80d8

Fix a subtle bug in "return;" generation which caused us to miss
a return in one case.  Instead of sprinking return handling code
throughout the asmprinter generator, just treat it like any other
normal statement.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@81395 91177308-0d34-0410-b5e6-96231b3b80d8

Add comment re: clang dependency.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@81393 91177308-0d34-0410-b5e6-96231b3b80d8

Added an abstract superclass, MCDisassembler, for
all disassemblers.

Modified the MemoryObject to support 64-bit address
spaces, regardless of the LLVM process's address
width.

Modified the Target class to allow extraction of a
MCDisassembler.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@81392 91177308-0d34-0410-b5e6-96231b3b80d8