These are already implemented

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@28990 91177308-0d34-0410-b5e6-96231b3b80d8

0 offsets for memory operands

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@28989 91177308-0d34-0410-b5e6-96231b3b80d8

Split long comment lines.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@28988 91177308-0d34-0410-b5e6-96231b3b80d8

Ugly hack! Add helper functions InsertInFlightSetEntry and
RemoveInFlightSetEntry. They are used in place of direct set operators to
reduce instruction selection function stack size.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@28987 91177308-0d34-0410-b5e6-96231b3b80d8

Allow specifying an explicit list of architectures to build for, e.g.:
make ENABLE_OPTIMIZED=1 UNIVERSAL=1 UNIVERSAL_ARCH="i386 ppc ppc64"

retain the default of building for just i386/ppc.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@28985 91177308-0d34-0410-b5e6-96231b3b80d8

Should just use xorps to clear XMM registers for all data types. pxor is also one byte longer.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@28984 91177308-0d34-0410-b5e6-96231b3b80d8

Describe my newest work.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@28983 91177308-0d34-0410-b5e6-96231b3b80d8

Let X86CompilationCallback pass previous frame and return address to X86CompilationCallback2. Remove alloca hack.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@28982 91177308-0d34-0410-b5e6-96231b3b80d8

Added -disable-cbe to disable C backend testing.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@28981 91177308-0d34-0410-b5e6-96231b3b80d8

Add shift and rotate by 1 instructions / patterns.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@28980 91177308-0d34-0410-b5e6-96231b3b80d8

Always use xorps to clear XMM registers.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@28979 91177308-0d34-0410-b5e6-96231b3b80d8

Move .literal4 and .literal8 support into AsmPrinter.cpp

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@28978 91177308-0d34-0410-b5e6-96231b3b80d8

Add support to print 4-, 8-, and 16- byte constant literals in special
sections. e.g. On Darwin that would be .literal4 and .literal8.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@28977 91177308-0d34-0410-b5e6-96231b3b80d8

Hide x86 symbols

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@28976 91177308-0d34-0410-b5e6-96231b3b80d8

Use hidden visibility to make symbols in an anonymous namespace get
dropped.  This shrinks libllvmgcc.dylib another 67K

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@28975 91177308-0d34-0410-b5e6-96231b3b80d8

Use hidden visibility to reduce codesize

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@28974 91177308-0d34-0410-b5e6-96231b3b80d8

Shave another 27K off libllvmgcc.dylib with visibility hidden

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@28973 91177308-0d34-0410-b5e6-96231b3b80d8

Shrink libllvmgcc.dylib by another 23K

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@28972 91177308-0d34-0410-b5e6-96231b3b80d8

shrink libllvmgcc.dylib another 25K

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@28971 91177308-0d34-0410-b5e6-96231b3b80d8

Mark these two classes as hidden, shrinking libllbmgcc.dylib by 25K

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@28970 91177308-0d34-0410-b5e6-96231b3b80d8

Use hidden visibility to reduce the sizes of some .o files.  This chops 60K off a release llvm-dis.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@28969 91177308-0d34-0410-b5e6-96231b3b80d8

Add support for hidden visibility

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@28968 91177308-0d34-0410-b5e6-96231b3b80d8

not really XFailing these, as only incompleteness is wrong in the graph

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@28967 91177308-0d34-0410-b5e6-96231b3b80d8

change I flag on test

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@28966 91177308-0d34-0410-b5e6-96231b3b80d8

Fix CodeGen/Generic/2006-06-28-SimplifySetCCCrash.ll

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@28965 91177308-0d34-0410-b5e6-96231b3b80d8

Testcase that crashes the ppc backend.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@28964 91177308-0d34-0410-b5e6-96231b3b80d8

Doh.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@28963 91177308-0d34-0410-b5e6-96231b3b80d8

Switch to a very conservative heuristic for determining when loop-unswitching
will be profitable.  This is mainly to remove some cases where excessive
unswitching would result in long compile times and/or huge generated code.

Once someone comes up with a better heuristic that avoids these cases, this
should be switched out.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@28962 91177308-0d34-0410-b5e6-96231b3b80d8

Fix Transforms/InstCombine/2006-06-28-infloop.ll

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@28961 91177308-0d34-0410-b5e6-96231b3b80d8

Infinite loop in instcombine that nate hit.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@28960 91177308-0d34-0410-b5e6-96231b3b80d8

Don't unswitch really large loops even if they are mostly filled with empty
blocks.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@28959 91177308-0d34-0410-b5e6-96231b3b80d8

Oops. Need to keep CP index.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@28958 91177308-0d34-0410-b5e6-96231b3b80d8

Darwin puts float and double literal constants into literal4 and literal8 sections.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@28957 91177308-0d34-0410-b5e6-96231b3b80d8

add some missing externals

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@28955 91177308-0d34-0410-b5e6-96231b3b80d8

Catch more function pointer casting problems
Remove the Function pointer cast in these calls, converting it to
a cast of argument.
%tmp60 = tail call int cast (int (ulong)* %str to int (int)*)( int 10 )
%tmp60 = tail call int cast (int (ulong)* %str to int (int)*)( uint %tmp51 )

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@28953 91177308-0d34-0410-b5e6-96231b3b80d8

Allow EmitConstantPool to be redefined by derived classes.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@28952 91177308-0d34-0410-b5e6-96231b3b80d8

Handle alias sets that have been unified, and thus can have other references
to them.  This fixes a regression in my previous checkin.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@28951 91177308-0d34-0410-b5e6-96231b3b80d8

Don't implement AliasSetTracker::remove in terms of deleteValue.  deleteValue
causes the pointer to be removed from the underlying alias analysis
implementation as well.  This impl of remove is also significantly faster than
the old one.  This fixes:
Regression/Transforms/DeadStoreElimination/2006-06-27-AST-Remove.ll

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@28950 91177308-0d34-0410-b5e6-96231b3b80d8

New testcase, reduced by nate, which crashes DSE

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@28949 91177308-0d34-0410-b5e6-96231b3b80d8

this case isn't handled

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@28948 91177308-0d34-0410-b5e6-96231b3b80d8

Fix for 2006-06-27-DeadSwitchCase.ll

Be more careful when updating Phi nodes after eliminating dead switch cases.  Fix
proposed by Chris.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@28947 91177308-0d34-0410-b5e6-96231b3b80d8

Testcase that Nate found where LoopUnswitch chokes on eliminating a dead case
from a switch instruction.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@28946 91177308-0d34-0410-b5e6-96231b3b80d8

handle the "mov reg1, reg2" case in isMoveInstr

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@28945 91177308-0d34-0410-b5e6-96231b3b80d8

Don't match 64-bit bitfield inserts into rlwimi's.  todo add rldimi. :)

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@28944 91177308-0d34-0410-b5e6-96231b3b80d8

Fix Transforms/DeadArgElim/2006-06-27-struct-ret.ll.  -deadargelim should not
remove the struct return argument of a csret function, even if it is obviously
dead.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@28943 91177308-0d34-0410-b5e6-96231b3b80d8

testcase that crashes dead arg elim.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@28942 91177308-0d34-0410-b5e6-96231b3b80d8

Fix ppc64 jump tables

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@28941 91177308-0d34-0410-b5e6-96231b3b80d8

Pass -Xlinker flags to gcc when it builds the shared object.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@28939 91177308-0d34-0410-b5e6-96231b3b80d8

Remove dead code.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@28938 91177308-0d34-0410-b5e6-96231b3b80d8

Fix -extraflags

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@28937 91177308-0d34-0410-b5e6-96231b3b80d8

Print stubs for external globals right.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@28936 91177308-0d34-0410-b5e6-96231b3b80d8

Implement 64-bit select, bswap, etc.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@28935 91177308-0d34-0410-b5e6-96231b3b80d8

Add a pattern for i64 sra.  Print 8-byte units with a space between the .quad
and the data

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@28934 91177308-0d34-0410-b5e6-96231b3b80d8

Fix rewriting frame offsets with ixaddr instructions, which implicitly shift
the offset two bits to the left.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@28933 91177308-0d34-0410-b5e6-96231b3b80d8

PPC doesn't have bit converts to/from i64

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@28932 91177308-0d34-0410-b5e6-96231b3b80d8

Add 64-bit MTCTR so that indirect calls work.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@28931 91177308-0d34-0410-b5e6-96231b3b80d8

Fix an incorrect store pattern.  This fixes em3d.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@28930 91177308-0d34-0410-b5e6-96231b3b80d8

Implement 64-bit undef, sub, shl/shr, srem/urem

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@28929 91177308-0d34-0410-b5e6-96231b3b80d8

Fix cut-n-pasto in comments.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@28928 91177308-0d34-0410-b5e6-96231b3b80d8

Use i32 for shift amounts instead of i64.  This gets bisort working.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@28927 91177308-0d34-0410-b5e6-96231b3b80d8

Add zextload from i32 -> i64, with this, perimeter works.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@28926 91177308-0d34-0410-b5e6-96231b3b80d8

For PR801:
Refactor the Graph writing code to use a common implementation which is
now in lib/Support/GraphWriter.cpp. This completes the PR.

Patch by Anton Korobeynikov. Thanks, Anton!

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@28925 91177308-0d34-0410-b5e6-96231b3b80d8

De-pessimize the handling of LCSSA Phi nodes in IndVarSimplify.  Hopefully this
will make Shootout-C/nestedloop faster.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@28924 91177308-0d34-0410-b5e6-96231b3b80d8

Print darwin stub stuff correctly in 64-bit mode.  With this, treeadd works in
ppc64 mode!

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@28923 91177308-0d34-0410-b5e6-96231b3b80d8

Fix variable shadowing issue

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@28922 91177308-0d34-0410-b5e6-96231b3b80d8

Implement a bunch of 64-bit cleanliness work.  With this, treeadd builds (but
doesn't work right).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@28921 91177308-0d34-0410-b5e6-96231b3b80d8

Rearrange compares, add ADDI8, add sext from 32-to-64 bit register

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@28920 91177308-0d34-0410-b5e6-96231b3b80d8

Improve PPC64 calling convention support

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@28919 91177308-0d34-0410-b5e6-96231b3b80d8

Remove two more definitions

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@28918 91177308-0d34-0410-b5e6-96231b3b80d8

remove two unused instructions.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@28917 91177308-0d34-0410-b5e6-96231b3b80d8

Fix a stale pointer issue that caused 300.twolf to fail to build on zion
last night.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@28916 91177308-0d34-0410-b5e6-96231b3b80d8

Add a new method.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@28915 91177308-0d34-0410-b5e6-96231b3b80d8

random code cleanups, no functionality change

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@28914 91177308-0d34-0410-b5e6-96231b3b80d8

Make LoopUnswitch able to unswitch loops with live-out values by taking advantage
of LCSSA.  This results several times the number of unswitchings occurring on
tests such and timberwolfmc, unix-tbl, and ldecod.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@28912 91177308-0d34-0410-b5e6-96231b3b80d8

Simplify X86CompilationCallback: always align to 16-byte boundary; don't save EAX/EDX if unnecessary.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@28910 91177308-0d34-0410-b5e6-96231b3b80d8

Add and sort "sections" in debug lines.  This always stepping through
code in sections other than ".text", including weak sections like ctors and
dtors.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@28909 91177308-0d34-0410-b5e6-96231b3b80d8

Added jump table address relocation.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@28908 91177308-0d34-0410-b5e6-96231b3b80d8

Eliminate unneeded parameter.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@28907 91177308-0d34-0410-b5e6-96231b3b80d8

variable_ops instructions such as call can have any number of operands.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@28906 91177308-0d34-0410-b5e6-96231b3b80d8

Whoops, missed a couple more C-style casts.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@28905 91177308-0d34-0410-b5e6-96231b3b80d8

Use C++ style casts instead of C-style casts to shut up compiler warnings
when compiling with -pedantic. Passes regression tests on Linux.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@28904 91177308-0d34-0410-b5e6-96231b3b80d8

Add targets for generating .s file in addition to .o files.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@28903 91177308-0d34-0410-b5e6-96231b3b80d8

Factor a bunch of rules together, no functionality change.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@28902 91177308-0d34-0410-b5e6-96231b3b80d8

Add more anonymous namespaces to make it clear that these are private classes

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@28901 91177308-0d34-0410-b5e6-96231b3b80d8

Remove unneeded libs

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@28900 91177308-0d34-0410-b5e6-96231b3b80d8

For PR811:
Don't both with the "C" and "cc" extensions as they aren't common and they
the "C" extension conflicts with the "c" extension on operating systems
that have case insensitive file names.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@28899 91177308-0d34-0410-b5e6-96231b3b80d8

Add some out-of-line virtual dtors so that the class has a "home", preventing
vtables for (e.g.) Instruction from being emitted into every .o file.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@28898 91177308-0d34-0410-b5e6-96231b3b80d8

fix typo

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@28897 91177308-0d34-0410-b5e6-96231b3b80d8

Add memory operand and int regs

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@28896 91177308-0d34-0410-b5e6-96231b3b80d8

inline asm, at least for floats

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@28895 91177308-0d34-0410-b5e6-96231b3b80d8

fix typos

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@28894 91177308-0d34-0410-b5e6-96231b3b80d8

fix argument problem

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@28893 91177308-0d34-0410-b5e6-96231b3b80d8

Correct returns of 64-bit values, though they seemed to work before...

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@28892 91177308-0d34-0410-b5e6-96231b3b80d8

Fix non-64-bit clean code.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@28891 91177308-0d34-0410-b5e6-96231b3b80d8

Make these predicates correct in 64-bit mode too.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@28890 91177308-0d34-0410-b5e6-96231b3b80d8

Rename OR4 -> OR.  Move some PPC64-specific stuff to the 64-bit file

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@28889 91177308-0d34-0410-b5e6-96231b3b80d8

remove unused flag

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@28888 91177308-0d34-0410-b5e6-96231b3b80d8

add some logical ops

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@28887 91177308-0d34-0410-b5e6-96231b3b80d8

remove some unused patterns

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@28886 91177308-0d34-0410-b5e6-96231b3b80d8

Add some more immediate patterns.  This allows us to compile:

void test6() {
  Y = 0xABCD0123BCDE4567;
}

into:

_test6:
        lis r2, -21555
        lis r3, ha16(_Y)
        ori r2, r2, 291
        rldicr r2, r2, 32, 31
        oris r2, r2, 48350
        ori r2, r2, 17767
        std r2, lo16(_Y)(r3)
        blr

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@28885 91177308-0d34-0410-b5e6-96231b3b80d8