This was fixed

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@52090 91177308-0d34-0410-b5e6-96231b3b80d8

add some notes.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@52087 91177308-0d34-0410-b5e6-96231b3b80d8

Added FP instruction formats.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@52086 91177308-0d34-0410-b5e6-96231b3b80d8

Temporarily reverting r52056. It's causing PPC to fail to bootstrap.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@52085 91177308-0d34-0410-b5e6-96231b3b80d8

Added support for FP Registers

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@52079 91177308-0d34-0410-b5e6-96231b3b80d8

Update VC project files, patch by Cédric Venet!

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@52077 91177308-0d34-0410-b5e6-96231b3b80d8

Speculatively execute a block when the the block is the then part of a triangle shape and it contains a single, side effect free, cheap instruction. The branch is eliminated by adding a select instruction. i.e.

Turn
BB:
    %t1 = icmp
    br i1 %t1, label %BB1, label %BB2
BB1:
    %t3 = add %t2, c
    br label BB2
BB2:
=>
BB:
    %t1 = icmp
    %t4 = add %t2, c
    %t3 = select i1 %t1, %t2, %t3

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@52073 91177308-0d34-0410-b5e6-96231b3b80d8

Fix run line.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@52072 91177308-0d34-0410-b5e6-96231b3b80d8

Revert r52046. It broke cbe on x86 / Mac OS X.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@52071 91177308-0d34-0410-b5e6-96231b3b80d8

CodeGen support for insertvalue and extractvalue, and for loads and
stores of aggregate values.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@52069 91177308-0d34-0410-b5e6-96231b3b80d8

Connect successors before creating the DAG node for the branch.  This has
no visible functionality change, but enables a future patch where node creation
will update the CFG if it decides to create an unconditional rather than a conditional branch.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@52067 91177308-0d34-0410-b5e6-96231b3b80d8

get rid of ExtractValueInst::init's Value argument, it is already passed to the UnaryInstruction ctor

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@52064 91177308-0d34-0410-b5e6-96231b3b80d8

Typo.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@52062 91177308-0d34-0410-b5e6-96231b3b80d8

make ExtractValueInst derived from UnaryInstruction

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@52061 91177308-0d34-0410-b5e6-96231b3b80d8

Enable stack coloring by default.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@52057 91177308-0d34-0410-b5e6-96231b3b80d8

PPC preferred loop alignment is 16.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@52056 91177308-0d34-0410-b5e6-96231b3b80d8

LoopSimplify preserves AA.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@52053 91177308-0d34-0410-b5e6-96231b3b80d8

Print debug message only if there are dead passes.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@52052 91177308-0d34-0410-b5e6-96231b3b80d8

Include loop deletion and the ADCE rewrite in the release notes.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@52048 91177308-0d34-0410-b5e6-96231b3b80d8

Testcase for PR2418

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@52047 91177308-0d34-0410-b5e6-96231b3b80d8

Handle assembler identifiers specially in CBE. This fixes PR2418.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@52046 91177308-0d34-0410-b5e6-96231b3b80d8

Tighten up the abstraction slightly.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@52045 91177308-0d34-0410-b5e6-96231b3b80d8

Wrap MVT::ValueType in a struct to get type safety
and better control the abstraction.  Rename the type
to MVT.  To update out-of-tree patches, the main
thing to do is to rename MVT::ValueType to MVT, and
rewrite expressions like MVT::getSizeInBits(VT) in
the form VT.getSizeInBits().  Use VT.getSimpleVT()
to extract a MVT::SimpleValueType for use in switch
statements (you will get an assert failure if VT is
an extended value type - these shouldn't exist after
type legalization).
This results in a small speedup of codegen and no
new testsuite failures (x86-64 linux).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@52044 91177308-0d34-0410-b5e6-96231b3b80d8

As Chris suggested, handle the situation if ShAmt larger than BitWidth,
otherwise, opt might crash.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@52041 91177308-0d34-0410-b5e6-96231b3b80d8

Refine stack slot interval weight computation.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@52040 91177308-0d34-0410-b5e6-96231b3b80d8

Turn stdout into binary mode during bitcode emission.
This is necessary on windows targets, since stdout is in text mode there.
Patch by Julien Lerouge!

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@52038 91177308-0d34-0410-b5e6-96231b3b80d8

Provide hooks for libgcc symbols' address resolution inside lli on mingw32.
Patch by Julien Lerouge!

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@52037 91177308-0d34-0410-b5e6-96231b3b80d8

Added custom isel for MUL, SDIVREM, UDIVREM, SMUL_LOHI and UMUL_LOHI nodes
MUL is not anymore directly matched because its a pseudoinstruction.
LogicI class fixed to zero-extend immediates.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@52036 91177308-0d34-0410-b5e6-96231b3b80d8

Added custom SELECT_CC lowering
Added special isel for ADDE,SUBE and new patterns to match SUBC,ADDC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@52031 91177308-0d34-0410-b5e6-96231b3b80d8

Revert 52002.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@52030 91177308-0d34-0410-b5e6-96231b3b80d8

Expose a public interface to this pass.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@52029 91177308-0d34-0410-b5e6-96231b3b80d8

Let libcall semantics decide whether it knows about functions
with definitions or not.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@52028 91177308-0d34-0410-b5e6-96231b3b80d8

Warn of potential violations of strict aliasing rules.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@52027 91177308-0d34-0410-b5e6-96231b3b80d8

Don't break strict aliasing.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@52026 91177308-0d34-0410-b5e6-96231b3b80d8

Describe loop index split.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@52021 91177308-0d34-0410-b5e6-96231b3b80d8

Wording fixes. Thanks Bill!

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@52017 91177308-0d34-0410-b5e6-96231b3b80d8

Remove debugging code.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@52016 91177308-0d34-0410-b5e6-96231b3b80d8

fix typos, tweak stuff, make it to validate

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@52015 91177308-0d34-0410-b5e6-96231b3b80d8

Use the newly created helper on LiveIntervals.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@52013 91177308-0d34-0410-b5e6-96231b3b80d8

Add a helper for constructing new live ranges that ended from an instruction to the end of its MBB.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@52012 91177308-0d34-0410-b5e6-96231b3b80d8

Move i128 on x86-64 from the codegen section to the x86-specific section.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@52010 91177308-0d34-0410-b5e6-96231b3b80d8

Add a note about improved alignment detection.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@52009 91177308-0d34-0410-b5e6-96231b3b80d8

Add a note mentioning --view-sunit-dags improvements.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@52008 91177308-0d34-0410-b5e6-96231b3b80d8

Expand the bullet item about i128 support and APInt usage in codegen.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@52007 91177308-0d34-0410-b5e6-96231b3b80d8

Add a test case for opt -instcombine bug fix in revision 52003.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@52004 91177308-0d34-0410-b5e6-96231b3b80d8

If BitWidth equals to ShtAmt, the RHSKnownZero[BitWidth-ShiftAmt-1] will
crash the opt. Just fix this.

Test case in llvm/test/Transforms/InstCombine/2008-06-05-ashr-crash.ll

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@52003 91177308-0d34-0410-b5e6-96231b3b80d8

Change the Verifier to support returning first class aggregrates.

Add a testcase for functions returning first class aggregrates.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@52002 91177308-0d34-0410-b5e6-96231b3b80d8

Add a test case for APInt bug fix in r51999.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@52000 91177308-0d34-0410-b5e6-96231b3b80d8

As comments said, for negative value, the arithmetic
over-shift-right should return -1. So here it should be signed-extended,
when bitwidth larger than 64.

test case: llvm/test/ExecutionEngine/2008-06-05-APInt-OverAShr.ll

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@51999 91177308-0d34-0410-b5e6-96231b3b80d8

Escape "$#".

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@51998 91177308-0d34-0410-b5e6-96231b3b80d8

Learn ScalarReplAggregrates how stores and loads of first class aggregrates
work and how to replace them into individual values. Also, when trying to
replace an aggregrate that is used by load or store with a single (large)
integer, don't crash (but don't replace the aggregrate either).

Also adds a testcase for both structs and arrays.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@51997 91177308-0d34-0410-b5e6-96231b3b80d8

Move this to the proper section.

I misread Optimizer Improvements as Other Improvements for some reason.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@51996 91177308-0d34-0410-b5e6-96231b3b80d8

Adding release notes for bindings improvements.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@51995 91177308-0d34-0410-b5e6-96231b3b80d8

Let StructRetPromotion check if all if its users are really calls or invokesn,
not other instructions. This fixes a crash with the added testcase.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@51992 91177308-0d34-0410-b5e6-96231b3b80d8

Let StructRetPromotion check if it's users are really calling it and not
passing its pointer. Fixes test with added testcase.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@51991 91177308-0d34-0410-b5e6-96231b3b80d8

Use use_iterator::getOperandNo instead of CallSite::hasArgument to check if a
function is passed as an argument instead of called. Also do this check a bit
earlier.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@51990 91177308-0d34-0410-b5e6-96231b3b80d8

 * Make CallSite::hasArgument const and let it take a const parameter.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@51989 91177308-0d34-0410-b5e6-96231b3b80d8

more updates and random notes, including changes up through Week-of-Mon-20080324.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@51988 91177308-0d34-0410-b5e6-96231b3b80d8

Update comments and documentation to reflect that GCSE and ValueNumbering are
deprecated by the GVN and GVNPRE passes.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@51983 91177308-0d34-0410-b5e6-96231b3b80d8

Cast because I think I know what I'm doing. There are warnings about converting
from 64-bit to 32-bit.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@51982 91177308-0d34-0410-b5e6-96231b3b80d8

Fix 80 column violations.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@51981 91177308-0d34-0410-b5e6-96231b3b80d8

Remove unused parameter. Some clients might use -Werror when compiling and
cause drama.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@51980 91177308-0d34-0410-b5e6-96231b3b80d8

add llvmc2 notes, stack realignment.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@51979 91177308-0d34-0410-b5e6-96231b3b80d8

various status updates.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@51978 91177308-0d34-0410-b5e6-96231b3b80d8

Start adding 2.3 content.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@51977 91177308-0d34-0410-b5e6-96231b3b80d8

Fix a memcpy lowering bug. Even though the memcpy alignment is smaller than the desired alignment, the frame destination alignment may still be larger than the desired alignment. Don't change its alignment to something smaller.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@51970 91177308-0d34-0410-b5e6-96231b3b80d8

Remove unneeded #include.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@51955 91177308-0d34-0410-b5e6-96231b3b80d8

Fix inst_iterator example.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@51954 91177308-0d34-0410-b5e6-96231b3b80d8

Oops. Should not be enabled by default.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@51953 91177308-0d34-0410-b5e6-96231b3b80d8

Rewrite a bunch of the CBE's inline asm code, giving it the
ability to handle indirect input operands.  This fixes PR2407.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@51952 91177308-0d34-0410-b5e6-96231b3b80d8

Correctly construct live intervals for the copies we inserted into the predecessors of a block containing a PHI.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@51950 91177308-0d34-0410-b5e6-96231b3b80d8

Revert this.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@51949 91177308-0d34-0410-b5e6-96231b3b80d8

Replace two manual loops with calls to CallSite::hasArguments (no functional changes).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@51947 91177308-0d34-0410-b5e6-96231b3b80d8

Add CallSite::hasArgument to allow for seeing if a call passes a certain value as an argument quickly.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@51946 91177308-0d34-0410-b5e6-96231b3b80d8

Add a Name parameter to two of the init methods of GetElementPointer to make the name setting more consistent.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@51945 91177308-0d34-0410-b5e6-96231b3b80d8

Fix example to be valid LLVM assembly.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@51942 91177308-0d34-0410-b5e6-96231b3b80d8

Implement the two constructors in InsertValueInst and ExtractValueInst.

Add a Name argment to two init methods in these classes as well to make things
a bit more consistent.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@51937 91177308-0d34-0410-b5e6-96231b3b80d8

Add a stack slot coloring pass. Not yet enabled.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@51934 91177308-0d34-0410-b5e6-96231b3b80d8

LowerSubregs should not clobber any analysis.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@51933 91177308-0d34-0410-b5e6-96231b3b80d8

Move #include to right place.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@51932 91177308-0d34-0410-b5e6-96231b3b80d8

Register if-converter pass for -debug-pass.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@51931 91177308-0d34-0410-b5e6-96231b3b80d8

More pass manager debugging outputs.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@51930 91177308-0d34-0410-b5e6-96231b3b80d8

Change packed struct layout so that field sizes
are the same as in unpacked structs, only field
positions differ.  This only matters for structs
containing x86 long double or an apint; it may
cause backwards compatibility problems if someone
has bitcode containing a packed struct with a
field of one of those types.
The issue is that only 10 bytes are needed to
hold an x86 long double: the store size is 10
bytes, but the ABI size is 12 or 16 bytes (linux/
darwin) which comes from rounding the store size
up by the alignment.  Because it seemed silly not
to pack an x86 long double into 10 bytes in a
packed struct, this is what was done.  I now
think this was a mistake.  Reserving the ABI size
for an x86 long double field even in a packed
struct makes things more uniform: the ABI size is
now always used when reserving space for a type.
This means that developers are less likely to
make mistakes.  It also makes life easier for the
CBE which otherwise could not represent all LLVM
packed structs (PR2402).
Front-end people might need to adjust the way
they create LLVM structs - see following change
to llvm-gcc.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@51928 91177308-0d34-0410-b5e6-96231b3b80d8

Add #includes required by GCC 4.3, thanks for Zhongxing Xu
for reporting this.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@51926 91177308-0d34-0410-b5e6-96231b3b80d8

Some Mips minor fixes
Added support for mips little endian arch => mipsel

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@51923 91177308-0d34-0410-b5e6-96231b3b80d8

We need to subtract one from this index because live ranges are open at the end.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@51922 91177308-0d34-0410-b5e6-96231b3b80d8

Fix spellnig error

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@51917 91177308-0d34-0410-b5e6-96231b3b80d8

Testcase for LoopIndexSplit and DomFrontier.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@51916 91177308-0d34-0410-b5e6-96231b3b80d8

LoopIndexSplit can sometimes result in cases where a block in its own domfrontier.
Don't crash when we encounter one of these.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@51915 91177308-0d34-0410-b5e6-96231b3b80d8

Expand documentation of StringConstantPrefix.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@51911 91177308-0d34-0410-b5e6-96231b3b80d8

Prevent a crash in debug dumps.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@51910 91177308-0d34-0410-b5e6-96231b3b80d8

Add StringConstantPrefix to control what the
assembler names of string constants look like.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@51909 91177308-0d34-0410-b5e6-96231b3b80d8

Find a better place to output hex constants corresponding to integers.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@51904 91177308-0d34-0410-b5e6-96231b3b80d8

Fixed bug in bad behavior in calculateFrameObjectOffsets,
the solution commited is different from the previous patch to
avoid int and unsigned comparison

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@51899 91177308-0d34-0410-b5e6-96231b3b80d8

Do not run loop-aligner at -fast (e.g. -O0).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@51898 91177308-0d34-0410-b5e6-96231b3b80d8

Revert this patch

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@51897 91177308-0d34-0410-b5e6-96231b3b80d8

nounwindify.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@51893 91177308-0d34-0410-b5e6-96231b3b80d8

"Unable to schedule <A> required by <B>" is more helpful then
"Unable to handle Pass that requires lower level Analysis pass"

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@51892 91177308-0d34-0410-b5e6-96231b3b80d8

Add debugging aid.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@51891 91177308-0d34-0410-b5e6-96231b3b80d8

Fix whitespace in whitespace-significant pseudocode in a comment.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@51890 91177308-0d34-0410-b5e6-96231b3b80d8