Add sparc test for memory operand used in inline asm

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@57348 91177308-0d34-0410-b5e6-96231b3b80d8

This is not failing anymore

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@57347 91177308-0d34-0410-b5e6-96231b3b80d8

Add rudimentary asmprinter support for printing inline asm operands for sparc.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@57346 91177308-0d34-0410-b5e6-96231b3b80d8

Add dummy 'm' inline asm constraint handler for Sparc. I'm not sure, whether it is correct, however :)

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@57345 91177308-0d34-0410-b5e6-96231b3b80d8

Cleanup

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@57344 91177308-0d34-0410-b5e6-96231b3b80d8

Fix copy-and-paste-o.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@57343 91177308-0d34-0410-b5e6-96231b3b80d8

Add a basic intra-procedural escape analysis.  This hasn't be extensively tested yet, but feedback is welcome.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@57342 91177308-0d34-0410-b5e6-96231b3b80d8

Moved guard mutex upwards to guard materializing a function
in getPointerToFunction

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@57340 91177308-0d34-0410-b5e6-96231b3b80d8

Added missing print functions that take a raw_ostream

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@57339 91177308-0d34-0410-b5e6-96231b3b80d8

Fixed definition of llvm_anyptr_ty

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@57338 91177308-0d34-0410-b5e6-96231b3b80d8

Generated files for previous checkin.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@57331 91177308-0d34-0410-b5e6-96231b3b80d8

A file missed from previous checkin.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@57330 91177308-0d34-0410-b5e6-96231b3b80d8

Add a "loses information" return value to APFloat::convert
and APFloat::convertToInteger.  Restore return value to
IEEE754.  Adjust all users accordingly.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@57329 91177308-0d34-0410-b5e6-96231b3b80d8

Rename APFloat::convertToAPInt to bitcastToAPInt to
make it clearer what the function does.  No functional
change.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@57325 91177308-0d34-0410-b5e6-96231b3b80d8

Don't drop alignment on globals when cloning.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@57320 91177308-0d34-0410-b5e6-96231b3b80d8

get CodeGen/Alpha/mul128.ll to work.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@57318 91177308-0d34-0410-b5e6-96231b3b80d8

Align EH tables before label is emitted, not after,
thus aligning the label.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@57310 91177308-0d34-0410-b5e6-96231b3b80d8

(re)Put const weak strings in appropriate section on Darwin.
g++dg/abi/key2.C

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@57309 91177308-0d34-0410-b5e6-96231b3b80d8

dont specialize weak functions and the like

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@57305 91177308-0d34-0410-b5e6-96231b3b80d8

Comment to be explicit that the enumeration values for CondCodes matter.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@57295 91177308-0d34-0410-b5e6-96231b3b80d8

Use template to distinguish between function variants.
GCC 4.4.0 gives an error on the "int" declaration for example
saying that it has already been declared (using the "short"
one). Using templates here allow the compiler to distinguish
between the function to choose.

Also, "llvm/Support/DataTypes.h" was not included, leading to
error messages about not knowing "uint32_t" for example.

Patch by Samuel Tardieu.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@57292 91177308-0d34-0410-b5e6-96231b3b80d8

Add <cstdio> include where needed by gcc-4.4.
Patch by Samuel Tardieu.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@57291 91177308-0d34-0410-b5e6-96231b3b80d8

regenerate

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@57290 91177308-0d34-0410-b5e6-96231b3b80d8

Add missing semicolumns in parser rules, those missing semicolumns
are required to compile with the latest Bison.

Patch by Samuel Tardieu!

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@57289 91177308-0d34-0410-b5e6-96231b3b80d8

Add parentheses to avoid warnings in GCC 4.4.0,
patch by Samuel Tardieu!

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@57288 91177308-0d34-0410-b5e6-96231b3b80d8

fix typo

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@57287 91177308-0d34-0410-b5e6-96231b3b80d8

Fix the case where an instruction is not properly marked as using all registers that alias its inputs.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@57286 91177308-0d34-0410-b5e6-96231b3b80d8

Remove -disable-fast-isel. Use cl::boolOrDefault with -fast-isel
instead.

So now: -fast-isel or -fast-isel=true enable fast-isel, and
-fast-isel=false disables it. Fast-isel is also on by default
with -fast, and off by default otherwise.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@57270 91177308-0d34-0410-b5e6-96231b3b80d8

Fix disagreement about where the attributes are
~0 != ~0U.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@57268 91177308-0d34-0410-b5e6-96231b3b80d8

Fix configure issue where configure
turned "obj-c++" into "obj" in the langs line.
Update configure script.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@57267 91177308-0d34-0410-b5e6-96231b3b80d8

Add MBB successors and physreg Uses in the same order that
SDISel typically adds them in. This makes it a little easier
to compare FastISel output with SDISel output.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@57266 91177308-0d34-0410-b5e6-96231b3b80d8

Avoid emitting redundant materializations of integer constants
for things like null pointers, which at this level aren't
different from regular integer constants.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@57265 91177308-0d34-0410-b5e6-96231b3b80d8

Instead of emitting an implicit use for the super-register of
X86::CL that was used, emit an EXTRACT_SUBREG from the CL
super-register to CL. This more precisely describes how the
CL register is being used.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@57264 91177308-0d34-0410-b5e6-96231b3b80d8

Unconditional branch instruction encoding fix. Needs to use ABI, not AXI, to get the proper opcode bits.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@57262 91177308-0d34-0410-b5e6-96231b3b80d8

need ARM.h for ARMCC definition

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@57261 91177308-0d34-0410-b5e6-96231b3b80d8

Add a testcase for i256 add. i256 isn't fully supported in
codegen right now, but add and subtract work.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@57260 91177308-0d34-0410-b5e6-96231b3b80d8

Add an option to enable StrongPHIElimination, for ease of testing.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@57259 91177308-0d34-0410-b5e6-96231b3b80d8

Encode the conditional execution predicate when JITing.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@57258 91177308-0d34-0410-b5e6-96231b3b80d8

Model hardwired inputs & outputs of x86 8-bit divides correctly.
Fixes local RA miscompilation of gcc.c-torture/execute/20020904-1.c -O0.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@57257 91177308-0d34-0410-b5e6-96231b3b80d8

Use Dan's supperior check

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@57255 91177308-0d34-0410-b5e6-96231b3b80d8

Correctly set attributes when removing args during cloning.  Fixes PR2765

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@57254 91177308-0d34-0410-b5e6-96231b3b80d8

Improve function definition, call and invoke instruction docs.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@57253 91177308-0d34-0410-b5e6-96231b3b80d8

Clarify naming and correct conditional so that CMP and CMN instructions get the Rn operand encoded properly

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@57252 91177308-0d34-0410-b5e6-96231b3b80d8

Fix Opcode values of CMP and CMN

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@57251 91177308-0d34-0410-b5e6-96231b3b80d8

128 mul test, xfailed

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@57250 91177308-0d34-0410-b5e6-96231b3b80d8

No need for |=

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@57249 91177308-0d34-0410-b5e6-96231b3b80d8

Use ADDC if it is valid at any smaller size.  Do it right this time

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@57248 91177308-0d34-0410-b5e6-96231b3b80d8

Use ADDC if it is valid at any smaller size.  fixes test/Codegen/Generic/i128-addsub.ll on x86

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@57247 91177308-0d34-0410-b5e6-96231b3b80d8

Certain patterns involving the "movss" instruction were marked as requiring SSE2, when in reality movss is an SSE1 instruction.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@57246 91177308-0d34-0410-b5e6-96231b3b80d8

add support for running the test suite with valgrind. to run it just type 'make VG=1', as in clang
beware of the 42000 leaks reported by valgrind in the Constant.cpp + Type.cpp files. it needs fixing IMHO

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@57245 91177308-0d34-0410-b5e6-96231b3b80d8

Expand arith on machines without carry flags

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@57243 91177308-0d34-0410-b5e6-96231b3b80d8

Add two forgotten </i>'s.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@57239 91177308-0d34-0410-b5e6-96231b3b80d8

fix filetype suffix

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@57237 91177308-0d34-0410-b5e6-96231b3b80d8

Advance version to 2.5

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@57233 91177308-0d34-0410-b5e6-96231b3b80d8

no need to write the output to the disk

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@57232 91177308-0d34-0410-b5e6-96231b3b80d8

Add test case for ADDC ADDE expansion

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@57228 91177308-0d34-0410-b5e6-96231b3b80d8

Note that ADDC and company don't actually expand yet (missing in legalize

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@57226 91177308-0d34-0410-b5e6-96231b3b80d8

Mark negative-zero-to-int conversion as Inexact,
since -0.0 can't be represented as an int.
This prevents llvm from reducing -0.0 to a
load-and-convert of int 0.  Fixes
ieee.exp/mzero[2356].c in gcc testsuite.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@57224 91177308-0d34-0410-b5e6-96231b3b80d8

Correctly handle calls with no return values. This fixes
2006-01-23-UnionInit on x86-64 when inlining is not enabled.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@57223 91177308-0d34-0410-b5e6-96231b3b80d8

Fix typo, fix PR 2865.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@57221 91177308-0d34-0410-b5e6-96231b3b80d8

Be more precise about which conversions of NaNs
are Inexact.  (These are not Inexact as defined
by IEEE754, but that seems like a reasonable way
to abstract what happens:  information is lost.)

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@57218 91177308-0d34-0410-b5e6-96231b3b80d8

Set UNIVERSAL_ARCH value.
Fixes build failure when target is i386.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@57211 91177308-0d34-0410-b5e6-96231b3b80d8

Fix PR2850 and PR2863. Only generate movddup for 128-bit SSE vector shuffles.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@57210 91177308-0d34-0410-b5e6-96231b3b80d8

Temporarily revert functionality change from my previous patch;
it is too aggressive.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@57203 91177308-0d34-0410-b5e6-96231b3b80d8

Remove interfaces implemented by dead pass from the list of available passes.
Patch By Matthijs Kooijman.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@57202 91177308-0d34-0410-b5e6-96231b3b80d8

Cosmetic.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@57200 91177308-0d34-0410-b5e6-96231b3b80d8

Update function attributes docs.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@57197 91177308-0d34-0410-b5e6-96231b3b80d8

This is an objective-c test, not an objective-c++ one.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@57193 91177308-0d34-0410-b5e6-96231b3b80d8

Mark shortening NaN conversions as Inexact.  PR 2856.
Improve description of unsupported formats.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@57185 91177308-0d34-0410-b5e6-96231b3b80d8

It is possible that all functions in one module are not being
optimized for size. Set OptForSize for each function separately.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@57182 91177308-0d34-0410-b5e6-96231b3b80d8

Don't dereference the end() iterator. Thanks to
ENABLE_EXPENSIVE_CHECKS for finding this.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@57181 91177308-0d34-0410-b5e6-96231b3b80d8

Remove unncessary isDeclaration() checks.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@57179 91177308-0d34-0410-b5e6-96231b3b80d8

Allow scalarrepl to treat an all-zero GEP just as bitcast.

This includes not marking a GEP involving a vector as unsafe, but only when it
has all zero indices. This allows scalarrepl to work in a few more cases.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@57177 91177308-0d34-0410-b5e6-96231b3b80d8

Actually run Obj-C++ tests if llvm-gcc supports.
Before there were two problems: (1) configure
turned "obj-c++" into "obj" in the langs line;
(2) the dejagnu library called it objc++ not
obj-c++.
Now the problem is that some of these tests don't
pass!

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@57167 91177308-0d34-0410-b5e6-96231b3b80d8

Clarify the relationship between byval and readonly/
readnone.  Make clearer that readnone functions do not
dereference pointer arguments.  Do not use the highly
ambiguous "side-effects" in the readonly description
(since such functions can have control flow side-effects,
such as throwing an exception, or looping for ever).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@57166 91177308-0d34-0410-b5e6-96231b3b80d8

reorder #include order, patch by Kenneth Boyd!

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@57148 91177308-0d34-0410-b5e6-96231b3b80d8

Add #include to get alloca, patch by Kenneth Boyd!

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@57147 91177308-0d34-0410-b5e6-96231b3b80d8

Fix shift overflow bug that would occur when a field was a full 32-bits
in tblgen.  This is PR2827, thanks to Waldemar Knorr for tracking this
down.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@57124 91177308-0d34-0410-b5e6-96231b3b80d8

fix an incorrect and extremely confusing error message

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@57123 91177308-0d34-0410-b5e6-96231b3b80d8

make the autoupgrade code for ret attributes dramatically simpler
and actually work.  We can now read the llvm 2.3 bc file from PR2849

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@57122 91177308-0d34-0410-b5e6-96231b3b80d8

A word got optimized out, thanks to Duncan for pointing this out

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@57116 91177308-0d34-0410-b5e6-96231b3b80d8

regenerate

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@57114 91177308-0d34-0410-b5e6-96231b3b80d8

clean ArgTypeListI production: free the PATypeHolder

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@57113 91177308-0d34-0410-b5e6-96231b3b80d8

Emit type-correct constant null. Also fix a typo.

Patch by Robert G. Jakabosky!

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@57110 91177308-0d34-0410-b5e6-96231b3b80d8

Fix weird think-o and unbreak build on all gcc-3.4.x-based platforms (e.g. mingw)

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@57106 91177308-0d34-0410-b5e6-96231b3b80d8

this case is matched now.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@57096 91177308-0d34-0410-b5e6-96231b3b80d8

rewrite bswap matching to be more general, allowing arbitrary
shifting and masking inside a bswap expr.  This allows it to handle
the cases from PR2842, which involve the intermediate 'or'
expressions being shifted, not just the input value.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@57095 91177308-0d34-0410-b5e6-96231b3b80d8

Update VC++ project file

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@57091 91177308-0d34-0410-b5e6-96231b3b80d8

fix a bug where the bswap matcher could match a case involving
ashr.  It should only apply to lshr.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@57089 91177308-0d34-0410-b5e6-96231b3b80d8

wrap some long lines and expand i32 mul's to libcalls, inspired by a
patch by Mikael Lepisto!

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@57077 91177308-0d34-0410-b5e6-96231b3b80d8

CMake: Reflected changes on source file structure. New plugin support
for llvmc2 incomplete.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@57076 91177308-0d34-0410-b5e6-96231b3b80d8

Update the documentation for first-class aggregates changes,
and remove getresult and references thereto.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@57064 91177308-0d34-0410-b5e6-96231b3b80d8

add a note about inline asm

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@57062 91177308-0d34-0410-b5e6-96231b3b80d8

improve description of param/ret attrs

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@57061 91177308-0d34-0410-b5e6-96231b3b80d8

improve descriptions of function attrs

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@57060 91177308-0d34-0410-b5e6-96231b3b80d8

notes are gone.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@57059 91177308-0d34-0410-b5e6-96231b3b80d8

comment cleanups

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@57058 91177308-0d34-0410-b5e6-96231b3b80d8

Make GenLibDeps.pl more robust in the face of broken piping problems.
Patch by Kenneth Boyd!

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@57057 91177308-0d34-0410-b5e6-96231b3b80d8

Ignore loads from and stores to local memory (i.e. allocas)
when deciding whether to mark a function readnone/readonly.
Since the pass is currently run before SROA, this may be
quite helpful.  Requested by Chris on IRC.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@57050 91177308-0d34-0410-b5e6-96231b3b80d8

Allow the construction of SCEVs with SCEVCouldNotCompute operands, by
implementing folding. Fixes PR2857.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@57049 91177308-0d34-0410-b5e6-96231b3b80d8