Added RegisterCoalescer to required passes for PBQP.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@78840 91177308-0d34-0410-b5e6-96231b3b80d8

Use WriteAsOperand to print BasicBlock names.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@78838 91177308-0d34-0410-b5e6-96231b3b80d8

Make AsmWriter more careful with formatted_raw_ostream so that
it doesn't leave the underlying stream in unbuffered mode when
the stream was originally buffered.

Also, change WriteAsOperand back to plain raw_ostream. This
lets it work for either formatted_raw_ostream or plain
raw_ostream, so that it doesn't have to force a buffer flush
on a plain raw_ostream.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@78837 91177308-0d34-0410-b5e6-96231b3b80d8

Add a GetBufferSize() member to raw_ostream and use it to
simplify some formatted_raw_ostream code.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@78836 91177308-0d34-0410-b5e6-96231b3b80d8

Generate Neon VTBL and VTBX instructions from the corresponding intrinsics.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@78835 91177308-0d34-0410-b5e6-96231b3b80d8

Use PadToColumn instead of tabs.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@78834 91177308-0d34-0410-b5e6-96231b3b80d8

Fix a few more places to use PadToColumn instead of tabs. And fix
the basic block label printing to check whether a block has a name
before printing a comment character and whitespace for it.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@78830 91177308-0d34-0410-b5e6-96231b3b80d8

PredCC is meant to be 2 bits wide, like PredCC1.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@78829 91177308-0d34-0410-b5e6-96231b3b80d8

Use PadToColumn instead of tabs for aligning comments. Fix one place
that emitted unnecessary whitespace outside of VerboseAsm mode.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@78828 91177308-0d34-0410-b5e6-96231b3b80d8

Enhance the InstrStage object to enable the specification of an Itinerary with overlapping stages. The default is to maintain the current behavior that the "next" stage immediately follows the previous one.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@78827 91177308-0d34-0410-b5e6-96231b3b80d8

Add attempted idiotproofing comment per review.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@78825 91177308-0d34-0410-b5e6-96231b3b80d8

improve win32 path support, patch by Baptiste Lepilleur!

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@78823 91177308-0d34-0410-b5e6-96231b3b80d8

Test for 78821, sort of.  While that bug is nondeterministic,
this test failed consistently on a Darwin build.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@78822 91177308-0d34-0410-b5e6-96231b3b80d8

Fix a nondeterministic bug in APInt::roundToDouble;
when !isSingleWord() but getActiveBits() is small,
we were using the pointer value instead of the low
word of the integer value.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@78821 91177308-0d34-0410-b5e6-96231b3b80d8

Fix whitespace expectations to match the new AsmWriter behavior.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@78819 91177308-0d34-0410-b5e6-96231b3b80d8

Add catch block handling to SjLj exception handling.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@78817 91177308-0d34-0410-b5e6-96231b3b80d8

Make LLVM Assembly dramatically easier to read by aligning the comments,
using formatted_raw_ostream's PadToColumn.

Before:

bb1:            ; preds = %bb
  %2 = sext i32 %i.01 to i64            ; <i64> [#uses=1]
  %3 = getelementptr double* %p, i64 %2         ; <double*> [#uses=1]
  %4 = load double* %3, align 8         ; <double> [#uses=1]
  %5 = fmul double %4, 1.100000e+00             ; <double> [#uses=1]
  %6 = sext i32 %i.01 to i64            ; <i64> [#uses=1]
  %7 = getelementptr double* %p, i64 %6         ; <double*> [#uses=1]

After:

bb1:                                        ; preds = %bb
  %2 = sext i32 %i.01 to i64                ; <i64> [#uses=1]
  %3 = getelementptr double* %p, i64 %2     ; <double*> [#uses=1]
  %4 = load double* %3, align 8             ; <double> [#uses=1]
  %5 = fmul double %4, 1.100000e+00         ; <double> [#uses=1]
  %6 = sext i32 %i.01 to i64                ; <i64> [#uses=1]
  %7 = getelementptr double* %p, i64 %6     ; <double*> [#uses=1]

Several tests required whitespace adjustments.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@78816 91177308-0d34-0410-b5e6-96231b3b80d8

Fix TableGen warnings.  This partly reverts my previous change to this file,
leaving the mayLoad and mayStore settings around only the load/store
instructions where those can't be inferred from the patterns.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@78815 91177308-0d34-0410-b5e6-96231b3b80d8

Fix a missing newline (now that Value*'s operator<< doesn't append one).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@78814 91177308-0d34-0410-b5e6-96231b3b80d8

change CBE to just get TAI now, instead of TM to get TAI.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@78813 91177308-0d34-0410-b5e6-96231b3b80d8

Transform -X/C to X/-C, implementing a README.txt entry.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@78812 91177308-0d34-0410-b5e6-96231b3b80d8

Optimize (x/C)*C to x if the division is exact.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@78811 91177308-0d34-0410-b5e6-96231b3b80d8

Update instcombine's debug output to account for Value*'s operator<<
not appending its own newline.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@78810 91177308-0d34-0410-b5e6-96231b3b80d8

Remove a bunch more now-unnecessary Context arguments.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@78809 91177308-0d34-0410-b5e6-96231b3b80d8

Eliminate a bunch of now unnecessary explicit Context variables.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@78808 91177308-0d34-0410-b5e6-96231b3b80d8

CMake: Added asm file to x86_64 MSVC build.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@78807 91177308-0d34-0410-b5e6-96231b3b80d8

register naming cleanup (s/ip/r12/)

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@78806 91177308-0d34-0410-b5e6-96231b3b80d8

Fix unit test on FreeBSD. We need to make sure there is enough space to save the pointer even if the memory returned from malloc was already aligned.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@78805 91177308-0d34-0410-b5e6-96231b3b80d8

catch a typo and simplify call syntax

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@78804 91177308-0d34-0410-b5e6-96231b3b80d8

here comes the CMAKE part on LLVM_COMPACT_SENTINELS; see also r78628 and r78661

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@78803 91177308-0d34-0410-b5e6-96231b3b80d8

Change TargetAsmInfo to be constructed via TargetRegistry from a Target+Triple
pair instead of from a virtual method on TargetMachine.  This cuts the final
ties of TargetAsmInfo to TargetMachine, meaning that MC can now use
TargetAsmInfo.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@78802 91177308-0d34-0410-b5e6-96231b3b80d8

one last (?) bad x86 triple test.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@78801 91177308-0d34-0410-b5e6-96231b3b80d8

fix some pastos in triple lines.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@78800 91177308-0d34-0410-b5e6-96231b3b80d8

the x86 version of the name is x86-64, not x86_64.  Handle this properly
in getArchTypeForLLVMName.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@78799 91177308-0d34-0410-b5e6-96231b3b80d8

another bogus triple

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@78798 91177308-0d34-0410-b5e6-96231b3b80d8

add support for mingw64 target triples.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@78797 91177308-0d34-0410-b5e6-96231b3b80d8

fix another broken target triple.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@78796 91177308-0d34-0410-b5e6-96231b3b80d8

fix an incorrect target triple.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@78795 91177308-0d34-0410-b5e6-96231b3b80d8

Move immediate constant predicate templates from the Blackfin target to MathExtras.h

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@78793 91177308-0d34-0410-b5e6-96231b3b80d8

add a couple of helpers to the Triple class for decoding
the darwin version string.  This should help consolidate
the variety of weird functions we have scattered around the
codebase that do stuff like this.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@78792 91177308-0d34-0410-b5e6-96231b3b80d8

add nounwind

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@78791 91177308-0d34-0410-b5e6-96231b3b80d8

Shrink Thumb2 movcc instructions.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@78790 91177308-0d34-0410-b5e6-96231b3b80d8

CMake: Fixed sed script for translating library dependencies from
LibDeps.txt format to LLVMLibDeps.cmake format.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@78789 91177308-0d34-0410-b5e6-96231b3b80d8

CMake: Localized dependency on Perl.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@78788 91177308-0d34-0410-b5e6-96231b3b80d8

CMake: Rely on llvm_config again for obtaining the list of required
libraries for an executable.

Now LLVMConfig uses a new system for sorting library dependencies, as
the list of dependent libraries for each entry of FinalLibDeps.txt no
longer is topologically sorted.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@78787 91177308-0d34-0410-b5e6-96231b3b80d8

CMake: updated library dependencies.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@78786 91177308-0d34-0410-b5e6-96231b3b80d8

CMake: cmake/modules/LLVMLibDeps.cmake: Use unix line endings.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@78785 91177308-0d34-0410-b5e6-96231b3b80d8

CMake: Moved hard-coded library dependencies to its own file.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@78784 91177308-0d34-0410-b5e6-96231b3b80d8

Remove another Darwin assembler workaround.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@78779 91177308-0d34-0410-b5e6-96231b3b80d8

80 col violation.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@78778 91177308-0d34-0410-b5e6-96231b3b80d8

Remove an Darwin assembler workaround.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@78777 91177308-0d34-0410-b5e6-96231b3b80d8

Shrink ADDS, ADC, RSB, and SUBS.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@78776 91177308-0d34-0410-b5e6-96231b3b80d8

Add some comments to clarify the arguments to the vtbl and vtbx intrinsics.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@78775 91177308-0d34-0410-b5e6-96231b3b80d8

Testcase for r78770.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@78774 91177308-0d34-0410-b5e6-96231b3b80d8

This logic was accidentally inverted in r78767.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@78773 91177308-0d34-0410-b5e6-96231b3b80d8

CMake: Do not install files with .tmp suffix.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@78769 91177308-0d34-0410-b5e6-96231b3b80d8

CMake: Re-enabled build of llvm-config. Removed recursive invocation
of cmake.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@78768 91177308-0d34-0410-b5e6-96231b3b80d8

Factor out the code for finding an available register for use
in breaking an anti-dependence into a separate function.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@78767 91177308-0d34-0410-b5e6-96231b3b80d8

Test for llvm-gcc patch 78762.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@78763 91177308-0d34-0410-b5e6-96231b3b80d8

Add missing chain operands for VLD* and VST* instructions.
Set "mayLoad" and "mayStore" on the load/store instructions.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@78761 91177308-0d34-0410-b5e6-96231b3b80d8

Add contexts to some of the MVT APIs.  No functionality change yet, just the infrastructure work needed to get the contexts to where they need to be first.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@78759 91177308-0d34-0410-b5e6-96231b3b80d8

llvm-gcc now emits inbounds for this getelementptr.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@78758 91177308-0d34-0410-b5e6-96231b3b80d8

Simplify this code, and use an in-bounds GEP.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@78755 91177308-0d34-0410-b5e6-96231b3b80d8

CMake: target triple for MSVC on Windows 64.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@78753 91177308-0d34-0410-b5e6-96231b3b80d8

llvm-mc/AsmParser: Match hard coded registers (e.g. 'shldl %cl, %eax, %eax')

We now match all of 403.gcc (as emitted by clang). :)

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@78750 91177308-0d34-0410-b5e6-96231b3b80d8

prune #include

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@78749 91177308-0d34-0410-b5e6-96231b3b80d8

prune #includage.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@78748 91177308-0d34-0410-b5e6-96231b3b80d8

fix CodeGen/PowerPC/2007-01-15-AsmDialect.ll, fallout from r78742

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@78747 91177308-0d34-0410-b5e6-96231b3b80d8

change the -x86-asm-syntax=intel/att flag to be in X86TAI
instead of X86 Subtarget.  This elimianates dependencies on
X86Subtarget from X86TAI.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@78746 91177308-0d34-0410-b5e6-96231b3b80d8

Shrinkify Thumb2 r = add sp, imm.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@78745 91177308-0d34-0410-b5e6-96231b3b80d8

second half of commit.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@78744 91177308-0d34-0410-b5e6-96231b3b80d8

pass "is64Bit" flag into PPC TAI ctors instead of a whole targetmachine.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@78743 91177308-0d34-0410-b5e6-96231b3b80d8

eliminate asmflavor from subtarget, PPCTAI is the only client
and each callee knows that it returns.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@78742 91177308-0d34-0410-b5e6-96231b3b80d8

Change the asmprinter to print the comment character before the
"inlineasmstart/end" strings so that the contents of the directive
are separate from the comment character.  This lets elf targets
get #APP/#NOAPP for free even if they don't use "#" as the comment
character.  This also allows hoisting the darwin stuff up to the
shared TAI class.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@78737 91177308-0d34-0410-b5e6-96231b3b80d8

Allow a zero cycle stage to reserve/require a FU without advancing the cycle counter.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@78736 91177308-0d34-0410-b5e6-96231b3b80d8

factorize more darwin TAI stuff.  Note that this gives
darwin/arm support for .no_dead_strip

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@78734 91177308-0d34-0410-b5e6-96231b3b80d8

X86/AsmParser: Mark MOV64GSrm, MOV64FSrm, GS_MOV32rm, FS_MOV32rm as codegen only.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@78733 91177308-0d34-0410-b5e6-96231b3b80d8

factorize darwin ProtectedDirective and SetDirective.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@78732 91177308-0d34-0410-b5e6-96231b3b80d8

Add 'isCodeGenOnly' bit to Instruction .td records.
 - Used to mark fake instructions which don't correspond to an actual machine
   instruction (or are duplicates of a real instruction). This is to be used for
   "special cases" in the .td files, which should be ignored by things like the
   assembler and disassembler. We still need a good solution to handle pervasive
   duplication, like with the Int_ instructions.

 - Set the bit on fake "mov 0" style instructions, which allows turning an
   assembler matcher warning into a hard error.

 - -2 FIXMEs.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@78731 91177308-0d34-0410-b5e6-96231b3b80d8

all darwin targets have .space and .zerofill, pull up.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@78730 91177308-0d34-0410-b5e6-96231b3b80d8

eliminate template from arm TAI

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@78729 91177308-0d34-0410-b5e6-96231b3b80d8

fix a bug I introduced in r78724 that caused failures in:
CodeGen/X86/dll-linkage.ll & CodeGen/X86/mingw-alloca.ll

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@78728 91177308-0d34-0410-b5e6-96231b3b80d8

move LCOMMDirective = "\t.lcomm\t" up to DarwinTAI, eliminate
template in PPC backend for TAI.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@78727 91177308-0d34-0410-b5e6-96231b3b80d8

Fix warnings.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@78725 91177308-0d34-0410-b5e6-96231b3b80d8

eliminate the X86TargetAsmInfo template.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@78724 91177308-0d34-0410-b5e6-96231b3b80d8

Added ADD instructions with rAX as one parameter to the Intel instruction
tables.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@78721 91177308-0d34-0410-b5e6-96231b3b80d8

Shrinkify Thumb2 load / store multiple instructions.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@78717 91177308-0d34-0410-b5e6-96231b3b80d8

llvm-mc/AsmParser: Allow target to specific a comment delimiter, which will be
used to strip hard coded comments out of .td assembly strings.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@78716 91177308-0d34-0410-b5e6-96231b3b80d8

struct -> class

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@78715 91177308-0d34-0410-b5e6-96231b3b80d8

Optimize exact sdiv by a constant power of 2 to ashr.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@78714 91177308-0d34-0410-b5e6-96231b3b80d8

Split EVT into MVT and EVT, the former representing _just_ a primitive type, while
the latter is capable of representing either a primitive or an extended type.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@78713 91177308-0d34-0410-b5e6-96231b3b80d8

StringRef: Add find(char) and find(StringRef).

Also, regroup functions.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@78712 91177308-0d34-0410-b5e6-96231b3b80d8

pass the TargetTriple down from each target ctor to the
LLVMTargetMachine ctor.  It is currently unused.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@78711 91177308-0d34-0410-b5e6-96231b3b80d8

now that JumpTableDirective can differentate picness itself, MIPS TAI
no longer needs a targetmachine to initialize itself.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@78710 91177308-0d34-0410-b5e6-96231b3b80d8

split "JumpTableDirective" (an existing hack) into a PIC and nonPIC
version.  This allows TAI implementations to specify the directive to use
based on the mode being codegen'd for.

The real fix for this is to remove JumpTableDirective, but I don't feel
like diving into the jumptable snarl just now.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@78709 91177308-0d34-0410-b5e6-96231b3b80d8

"TAI::JumpTableDirective" is always null for current arm targets, simplify
the code based on this and make it fall through better.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@78708 91177308-0d34-0410-b5e6-96231b3b80d8

Add convenience functions for creating nsw add operators.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@78707 91177308-0d34-0410-b5e6-96231b3b80d8

llvm-mc/AsmMatcher: Fix two thinkos in determining whether two classes are
related.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@78706 91177308-0d34-0410-b5e6-96231b3b80d8

Use wchar.h to get wchar_t, not ctype.h.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@78703 91177308-0d34-0410-b5e6-96231b3b80d8

Don't set the isexact flag if an sdiv operator has been folded into
something else.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@78702 91177308-0d34-0410-b5e6-96231b3b80d8