Don't carry over the dispatchsetup hack from the old system.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@141040 91177308-0d34-0410-b5e6-96231b3b80d8

ARM parsing/encoding for VCMP/VCMPE.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@141038 91177308-0d34-0410-b5e6-96231b3b80d8

Fix typo in comments.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@141032 91177308-0d34-0410-b5e6-96231b3b80d8

Check-pointing the new SjLj EH lowering.

This code will replace the version in ARMAsmPrinter.cpp. It creates a new
machine basic block, which is the dispatch for the return from a longjmp
call. It then shoves the address of that machine basic block into the correct
place in the function context so that the EH runtime will jump to it directly
instead of having to go through a compare-and-jump-to-the-dispatch bit. This
should be more efficient in the common case.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@141031 91177308-0d34-0410-b5e6-96231b3b80d8

Move CHECK after entry label.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@141030 91177308-0d34-0410-b5e6-96231b3b80d8

Add support for 64-bit logical NOR.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@141029 91177308-0d34-0410-b5e6-96231b3b80d8

Add support for 64-bit count leading ones and zeros instructions.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@141028 91177308-0d34-0410-b5e6-96231b3b80d8

Move the grabbing of the jump buffer into the caller function, eliminating the need for returning a std::pair.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@141026 91177308-0d34-0410-b5e6-96231b3b80d8

ARM assembly parsing and encoding for VMRS/FMSTAT.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@141025 91177308-0d34-0410-b5e6-96231b3b80d8

Add support for 64-bit divide instructions.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@141024 91177308-0d34-0410-b5e6-96231b3b80d8

Add C api for Instruction->eraseFromParent().

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@141023 91177308-0d34-0410-b5e6-96231b3b80d8

Update test for 141010.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@141022 91177308-0d34-0410-b5e6-96231b3b80d8

Thumb2 ADD/SUB can take SP as a destination register.

It's documented as a separate instruction to line up with the Thumb1
encodings, for which it really is a distinct instruction encoding.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@141020 91177308-0d34-0410-b5e6-96231b3b80d8

Clean up MipsInstrInfo::copyPhysReg and handle copies from and to 64-bit integer
registers.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@141019 91177308-0d34-0410-b5e6-96231b3b80d8

Add support for 64-bit integer multiply instructions.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@141017 91177308-0d34-0410-b5e6-96231b3b80d8

Add definitions of instructions which move values between 64-bit integer
registers and 64-bit HI and LO registers. Fix encoding of the 32-bit versions
of the instructions.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@141015 91177308-0d34-0410-b5e6-96231b3b80d8

Find the strip tool that works with the specified SDKROOT.  rdar://10165908

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@141013 91177308-0d34-0410-b5e6-96231b3b80d8

Tidy up a bit. Formatting.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@141010 91177308-0d34-0410-b5e6-96231b3b80d8

Add support for MOVBE and RDRAND instructions for the assembler and disassembler. Includes feature flag checking, but no instrinsic support. Fixes PR10832, PR11026 and PR11027.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@141007 91177308-0d34-0410-b5e6-96231b3b80d8

Whitespace.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@141005 91177308-0d34-0410-b5e6-96231b3b80d8

Typo.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@141004 91177308-0d34-0410-b5e6-96231b3b80d8

Add the returns_twice attribute to LLVM.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@141001 91177308-0d34-0410-b5e6-96231b3b80d8

Treat VEX.vvvv as a 3-bit field outside of 64-bit mode. Prevents access to registers xmm8-xmm15 outside 64-bit mode.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@140997 91177308-0d34-0410-b5e6-96231b3b80d8

Test updates that were supposed to go with r140993.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@140994 91177308-0d34-0410-b5e6-96231b3b80d8

Fix VEX disassembling to ignore REX.RXBW bits in 32-bit mode.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@140993 91177308-0d34-0410-b5e6-96231b3b80d8

Reapply r140979 with fix! We never did get a testcase, but careful review of the
logic by David Meyer revealed this bug.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@140992 91177308-0d34-0410-b5e6-96231b3b80d8

attempt to fix ocaml bindings: landing pads

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@140991 91177308-0d34-0410-b5e6-96231b3b80d8

Revert r140979 due to reports of bootstrap failure.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@140980 91177308-0d34-0410-b5e6-96231b3b80d8

Add one more case we compute a max trip count.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@140979 91177308-0d34-0410-b5e6-96231b3b80d8

Fix some Intel syntax disassembly issues with instructions that implicitly use AL/AX/EAX/RAX such as ADD/SUB/ADC/SUBB/XOR/OR/AND/CMP/MOV/TEST.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@140974 91177308-0d34-0410-b5e6-96231b3b80d8

Special case disassembler handling of REX.B prefix on NOP instruction to decode as XCHG R8D, EAX instead. Fixes PR10344.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@140971 91177308-0d34-0410-b5e6-96231b3b80d8

Add a new icmp+select optz'n. Also shows off the load(cst) folding added in
r140966.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@140969 91177308-0d34-0410-b5e6-96231b3b80d8

Enhance a couple places where we were doing constant folding of instructions,
but not load instructions. Noticed by inspection.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@140966 91177308-0d34-0410-b5e6-96231b3b80d8

Fix typo in r140954.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@140962 91177308-0d34-0410-b5e6-96231b3b80d8

Make canonicalization of ImmutableSetRef::asImmutableSet() semi-explicit.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@140959 91177308-0d34-0410-b5e6-96231b3b80d8

Fix disassembling of INVEPT and INVVPID to take operands

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@140955 91177308-0d34-0410-b5e6-96231b3b80d8

Fix disassembler handling of CRC32 which is an odd instruction that uses 0xf2 as an opcode extension and allows the opsize prefix. This necessitated adding IC_XD_OPSIZE and IC_64BIT_XD_OPSIZE contexts. Unfortunately, this increases the size of the disassembler tables. Fixes PR10702.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@140954 91177308-0d34-0410-b5e6-96231b3b80d8

Revert r140924 "Attempt to fix dynamic stack realignment for thumb1 functions."
to appease nightly testers.  Not quite there yet.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@140953 91177308-0d34-0410-b5e6-96231b3b80d8

Moved type construction out of the loop and added an assert on the legality of the type. Formatted lines to the 80 char limit.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@140952 91177308-0d34-0410-b5e6-96231b3b80d8

Move TableGen's parser and entry point into a library

This is the first step towards splitting LLVM and Clang's tblgen executables.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@140951 91177308-0d34-0410-b5e6-96231b3b80d8

No one should be using the method directly. Assert if they do.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@140947 91177308-0d34-0410-b5e6-96231b3b80d8

Add a convenience method to tell if two things are equal.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@140946 91177308-0d34-0410-b5e6-96231b3b80d8

Use the ARMConstantPoolMBB class to handle the MBB values.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@140943 91177308-0d34-0410-b5e6-96231b3b80d8

Add ARMConstantPoolMBB to hold an MBB value in the constant pool.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@140942 91177308-0d34-0410-b5e6-96231b3b80d8

Remove dead code.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@140941 91177308-0d34-0410-b5e6-96231b3b80d8

Remove now dead methods and ivar.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@140940 91177308-0d34-0410-b5e6-96231b3b80d8

Use the new ARMConstantPoolSymbol class to handle external symbols.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@140939 91177308-0d34-0410-b5e6-96231b3b80d8

Add an ARMConstantPool class for external symbols. This will split out the support for external symbols from the base class.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@140938 91177308-0d34-0410-b5e6-96231b3b80d8

Remove now dead methods and ivar from ARMConstantPoolValue.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@140937 91177308-0d34-0410-b5e6-96231b3b80d8

Switch over to using ARMConstantPoolConstant for global variables, functions,
and block addresses.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@140936 91177308-0d34-0410-b5e6-96231b3b80d8

Some more refactoring.

* Add a couple of Create methods to the ARMConstantPoolConstant class,
* Add its own version of getExistingMachineCPValue, and
* Modify hasSameValue to return false if the object isn't an ARMConstantPoolConstant.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@140935 91177308-0d34-0410-b5e6-96231b3b80d8

Add a Create method that accepts 'kind' and 'pcadj' arguments.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@140934 91177308-0d34-0410-b5e6-96231b3b80d8

Refactoring: Separate out the ARM constant pool Constant from the ARM constant
pool value.

It's not used right now, but will be soon.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@140933 91177308-0d34-0410-b5e6-96231b3b80d8

Subtarget getFeatureBits() returns a uint64_t, not unsigned.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@140928 91177308-0d34-0410-b5e6-96231b3b80d8

Attempt to fix dynamic stack realignment for thumb1 functions.  It is in fact
useful if an optimization assumes the stack has been realigned.  Credit to
Eli for his assistance.
rdar://10043857

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@140924 91177308-0d34-0410-b5e6-96231b3b80d8

Inlining and unrolling heuristics should be aware of free truncs.

We want heuristics to be based on accurate data, but more importantly
we don't want llvm to behave randomly. A benign trunc inserted by an
upstream pass should not cause a wild swings in optimization
level. See PR11034. It's a general problem with threshold-based
heuristics, but we can make it less bad.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@140919 91177308-0d34-0410-b5e6-96231b3b80d8

whitespace

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@140916 91177308-0d34-0410-b5e6-96231b3b80d8

Add Windows x64 stack walking support. Patch by Aaron Ballman!

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@140906 91177308-0d34-0410-b5e6-96231b3b80d8

Use precomputed BitVector for CodeGenRegisterClass::hasSubClass().

All the sub-class bit vectors are computed when first creating the
register bank.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@140905 91177308-0d34-0410-b5e6-96231b3b80d8

Filecheck-ize.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@140904 91177308-0d34-0410-b5e6-96231b3b80d8

Add new line at end of file.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@140903 91177308-0d34-0410-b5e6-96231b3b80d8

When inferring the pointer alignment, if the global doesn't have an initializer
and the alignment is 0 (i.e., it's defined globally in one file and declared in
another file) it could get an alignment which is larger than the ABI allows for
that type, resulting in aligned moves being used for unaligned loads.

For instance, in file A.c:

   struct S s;

In file B.c:
   struct {
     // something long
   };
   extern S s;

   void foo() {
     struct S p = s;
     // ...
   }

this copy is a 'memcpy' which is turned into a series of 'movaps' instructions
on X86. But this is wrong, because 'struct S' has alignment of 4, not 16.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@140902 91177308-0d34-0410-b5e6-96231b3b80d8

Promote comment to doxycomment. Adjust whitespace. No functionality change.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@140899 91177308-0d34-0410-b5e6-96231b3b80d8

Store sub-class lists as a bit vector.

This uses less memory and it reduces the complexity of sub-class
operations:

- hasSubClassEq() and friends become O(1) instead of O(N).

- getCommonSubClass() becomes O(N) instead of O(N^2).

In the future, TableGen will infer register classes.  This makes it
cheap to add them.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@140898 91177308-0d34-0410-b5e6-96231b3b80d8

Extract a slightly more general BitVector printer.

This one can also print 32-bit groups.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@140897 91177308-0d34-0410-b5e6-96231b3b80d8

Move getCommonSubClass() into TRI.

It will soon need the context.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@140896 91177308-0d34-0410-b5e6-96231b3b80d8

Compute lists of super-classes in CodeGenRegisterClass.

Use these lists instead of computing them on the fly in
RegisterInfoEmitter.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@140895 91177308-0d34-0410-b5e6-96231b3b80d8

Correct for my over-eager delete finger.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@140892 91177308-0d34-0410-b5e6-96231b3b80d8

Add definition of MipsELFObjectWriter.

Patch by Reed Kotler at Mips Technologies.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@140891 91177308-0d34-0410-b5e6-96231b3b80d8

Register the MC object streamer.

Patch by Reed Kotler at Mips Technologies.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@140887 91177308-0d34-0410-b5e6-96231b3b80d8

Register Asm backend. Add functions to MipsAsmBackend.

Patch by Reed Kotler at Mips Technologies.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@140886 91177308-0d34-0410-b5e6-96231b3b80d8

Add MCELFObjectTargetWriter and MCAsmBackend classes.

Patch by Reed Kotler at Mips Technologies.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@140885 91177308-0d34-0410-b5e6-96231b3b80d8

Test More Complicated Lists

Test of indexing lists of lists of lists works.  This also exercises
some operators.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@140884 91177308-0d34-0410-b5e6-96231b3b80d8

Test VarListElementInit:: resolveListElementReference

Add a TableGen test to check if indexing lists of lists works.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@140883 91177308-0d34-0410-b5e6-96231b3b80d8

Implement VarListElementInit:: resolveListElementReference

Implement VarListElementInit:: resolveListElementReference so that
lists of lists can be indexed.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@140882 91177308-0d34-0410-b5e6-96231b3b80d8

Update CMake build.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@140879 91177308-0d34-0410-b5e6-96231b3b80d8

Initial implementation of MipsMCCodeEmitter.

Patch by Reed Kotler at Mips Technologies.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@140878 91177308-0d34-0410-b5e6-96231b3b80d8

Don't modify constant in-place.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@140875 91177308-0d34-0410-b5e6-96231b3b80d8

Tracing or debug-printing a newly formed instruction should not crash.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@140874 91177308-0d34-0410-b5e6-96231b3b80d8

whitespace

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@140873 91177308-0d34-0410-b5e6-96231b3b80d8

Remove unnecessary checking of register operands.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@140872 91177308-0d34-0410-b5e6-96231b3b80d8

Add definitions of Mips64 rotate instructions.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@140870 91177308-0d34-0410-b5e6-96231b3b80d8

float comparison to double 'zero' constant can just be a float 'zero.'

InstCombine was incorrectly considering the conversion of the constant
zero to be unsafe.

We want to transform:
define float @bar(float %x) nounwind readnone optsize ssp {
  %conv = fpext float %x to double
  %cmp = fcmp olt double %conv, 0.000000e+00
  %conv1 = zext i1 %cmp to i32
  %conv2 = sitofp i32 %conv1 to float
  ret float %conv2
}

Into:
define float @bar(float %x) nounwind readnone optsize ssp {
  %cmp = fcmp olt float %x, 0.000000e+00   ; <---- This
  %conv1 = zext i1 %cmp to i32
  %conv2 = sitofp i32 %conv1 to float
  ret float %conv2
}

rdar://10215914

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@140869 91177308-0d34-0410-b5e6-96231b3b80d8

Constify 'isLSDA' and move a method out-of-line.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@140868 91177308-0d34-0410-b5e6-96231b3b80d8

Tidy up. Trailing whitespace.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@140865 91177308-0d34-0410-b5e6-96231b3b80d8

ARM Darwin default relocation model is PIC.

This matches clang, so default options in llc and friends are now closer to
clang's defaults.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@140863 91177308-0d34-0410-b5e6-96231b3b80d8

isCommutable should be 0 for DSUBu.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@140862 91177308-0d34-0410-b5e6-96231b3b80d8

ARM Fixup valus for movt/movw are for the whole value.

Remove an assert that was expecting only the relevant 16bit portion for
the fixup being handled. Also kill some dead code in the T2 portion.

rdar://9653509

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@140861 91177308-0d34-0410-b5e6-96231b3b80d8

Check values of immediate operands.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@140860 91177308-0d34-0410-b5e6-96231b3b80d8

Fix a bug in compare_numeric().

Thanks to Alexandru Dura and Jonas Paulsson for finding it.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@140859 91177308-0d34-0410-b5e6-96231b3b80d8

Add forgotten tests that the cleanup flag is cleared if there
is a catch-all landingpad clause.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@140858 91177308-0d34-0410-b5e6-96231b3b80d8

MCJIT initialization TargetData

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@140856 91177308-0d34-0410-b5e6-96231b3b80d8

PTX: Various stylistic and code readability changes recommended by Jim Grosbach.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@140855 91177308-0d34-0410-b5e6-96231b3b80d8

PTX: Add programmable rounding mode specifier for int <-> fp conversion instrs.
Also take this opportunity to clean up the rounding mode pass.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@140854 91177308-0d34-0410-b5e6-96231b3b80d8

Inlining often produces landingpad instructions with repeated
catch or repeated filter clauses.  Teach instcombine a bunch
of tricks for simplifying landingpad clauses.  Currently the
code only recognizes the GNU C++ and Ada personality functions,
but that doesn't stop it doing a bunch of "generic" transforms
which are hopefully fine for any real-world personality function.
If these "generic" transforms turn out not to be generic, they
can always be conditioned on the personality function.  Probably
someone should add the ObjC++ personality function.  I didn't as
I don't know anything about it.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@140852 91177308-0d34-0410-b5e6-96231b3b80d8

some 3.0 API notes

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@140851 91177308-0d34-0410-b5e6-96231b3b80d8

Comment grammar fixes.

thanks to Duncan.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@140850 91177308-0d34-0410-b5e6-96231b3b80d8

PTX: Attempt to cleanup/unify the handling of FP rounding modes. This requires
us to manually provide Pat<> definitions for all FP instruction patterns.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@140849 91177308-0d34-0410-b5e6-96231b3b80d8

Instead of crashing when MCAsmInfo is NULL, add an assert.

This helps with porting code from 2.9 to 3.0 as TargetSelect.h changed location,
and if you include the old one by accident you will trigger this assert.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@140848 91177308-0d34-0410-b5e6-96231b3b80d8

Mips64 shift instructions.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@140841 91177308-0d34-0410-b5e6-96231b3b80d8