Changed to using LLVM streams.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31955 91177308-0d34-0410-b5e6-96231b3b80d8

Changed to using llvm streams.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31954 91177308-0d34-0410-b5e6-96231b3b80d8

Added helper function to transition to using llvm streams.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31953 91177308-0d34-0410-b5e6-96231b3b80d8

Fix JIT encoding bugs for shift / rotate by one ops.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31952 91177308-0d34-0410-b5e6-96231b3b80d8

testcase for pr1016

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31951 91177308-0d34-0410-b5e6-96231b3b80d8

Fix PR1016

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31950 91177308-0d34-0410-b5e6-96231b3b80d8

Removed #include <iostream> and replaced streams with llvm streams.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31949 91177308-0d34-0410-b5e6-96231b3b80d8

Preliminary support for inline asm in the cbe.  The target specific ugliness
is still in Writer, but issolated to a single function. This might be split
into something in each target directory.  This is sufficient to get through
archie and an strcpy impl on linux-x86.

Module level asm is not handled.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31948 91177308-0d34-0410-b5e6-96231b3b80d8

Change MachineInstr ctor's to take a TargetInstrDescriptor reference instead
of opcode and number of operands.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31947 91177308-0d34-0410-b5e6-96231b3b80d8

remove dead tag

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31944 91177308-0d34-0410-b5e6-96231b3b80d8

Fix the dag combiner bug corresponding to PR1014.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31943 91177308-0d34-0410-b5e6-96231b3b80d8

this doesn't pass with CVS HEAD.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31942 91177308-0d34-0410-b5e6-96231b3b80d8

Fix PR1014 and InstCombine/2006-11-27-XorBug.ll.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31941 91177308-0d34-0410-b5e6-96231b3b80d8

Add 3 more test cases for ptrtoint, fptoui and fptosi

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31940 91177308-0d34-0410-b5e6-96231b3b80d8

testcase for PR1014

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31939 91177308-0d34-0410-b5e6-96231b3b80d8

When truncating to bool, it is necessary to & with 1 for all casts that
can result in a bool. Previously PtrToInt, FPToUI and FPToSI were missing
this operation.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31938 91177308-0d34-0410-b5e6-96231b3b80d8

Protect against null streams.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31937 91177308-0d34-0410-b5e6-96231b3b80d8

Removed #include <iostream> and replaced with llvm_* streams.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31936 91177308-0d34-0410-b5e6-96231b3b80d8

Alter these tests to ensure they match a "test $1, X" X86 instruction that
is now generated by the LLVM backend for "trunc to bool" instructions.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31935 91177308-0d34-0410-b5e6-96231b3b80d8

Add a test to ensure a bitcast/and/trunc combination eliminates the bitcast.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31934 91177308-0d34-0410-b5e6-96231b3b80d8

For better or worse, load from i1 is assumed to be zero extended.  Do not
form a load from i1 from larger loads that may not be zext'd.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31933 91177308-0d34-0410-b5e6-96231b3b80d8

If a brcond condition is promoted, make sure to zero extend it, even if not
expanded into BR_CC.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31932 91177308-0d34-0410-b5e6-96231b3b80d8

For PR950:
The long awaited CAST patch. This introduces 12 new instructions into LLVM
to replace the cast instruction. Corresponding changes throughout LLVM are
provided. This passes llvm-test, llvm/test, and SPEC CPUINT2000 with the
exception of 175.vpr which fails only on a slight floating point output
difference.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31931 91177308-0d34-0410-b5e6-96231b3b80d8

Don't use bitcast yet!

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31930 91177308-0d34-0410-b5e6-96231b3b80d8

Make this test case easier to read. No functional changes.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31929 91177308-0d34-0410-b5e6-96231b3b80d8

Removed #include <iostream> and replace with llvm_* streams.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31927 91177308-0d34-0410-b5e6-96231b3b80d8

Add method that will take a function pointer so that it can handle things
like "llvm_cerr << std::flush";

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31926 91177308-0d34-0410-b5e6-96231b3b80d8

Remove #include <iostream> and use llvm_* streams instead.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31925 91177308-0d34-0410-b5e6-96231b3b80d8

Replace #include <iostream> with llvm_* streams.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31924 91177308-0d34-0410-b5e6-96231b3b80d8

Removed #include <iostream> and replaced with llvm_* streams.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31923 91177308-0d34-0410-b5e6-96231b3b80d8

Removed #include <iostream> and used the llvm_cerr/DOUT streams instead.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31922 91177308-0d34-0410-b5e6-96231b3b80d8

fix storing bools to mem and unordered FP ops

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31920 91177308-0d34-0410-b5e6-96231b3b80d8

Make the absolute/relative tolerance information easier to read/understand.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31908 91177308-0d34-0410-b5e6-96231b3b80d8

Add newline at end of file.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31902 91177308-0d34-0410-b5e6-96231b3b80d8

Only show the first few lines of difference.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31901 91177308-0d34-0410-b5e6-96231b3b80d8

Add an option to enable lli (interpreter mode) testing.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31899 91177308-0d34-0410-b5e6-96231b3b80d8

Fix some typos.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31897 91177308-0d34-0410-b5e6-96231b3b80d8

Promote GEP ubyte indices to uint. Backwards compatibility for 1.2 and
older features will be dropped soon and these test cases must not rely
on the upgrade capability.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31896 91177308-0d34-0410-b5e6-96231b3b80d8

Update to new predicate simplifier VRP design. Fixes PR966 and PR967.
Remove predicate simplifier from default gcc3 pipeline. New design is too
slow to enable by default.
Add new testcases for problems encountered in development.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31895 91177308-0d34-0410-b5e6-96231b3b80d8

Stop early if there is no mis-optimization.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31893 91177308-0d34-0410-b5e6-96231b3b80d8

Documented the --disable-compression option in gccld and gccas.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31892 91177308-0d34-0410-b5e6-96231b3b80d8

Added the --disable-compression option which controls whether the
generated bytecode uses compression.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31891 91177308-0d34-0410-b5e6-96231b3b80d8

This xform is handled by FoldOpIntoPhi in visitCastInst in a more elegant way.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31889 91177308-0d34-0410-b5e6-96231b3b80d8

Refactored *GVRequiresExtraLoad() to Subtarget method.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31887 91177308-0d34-0410-b5e6-96231b3b80d8

Test for PR1013.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31886 91177308-0d34-0410-b5e6-96231b3b80d8

new testcase

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31885 91177308-0d34-0410-b5e6-96231b3b80d8

in ppc64-mode, don't allocate the 32-bit version of r13 either.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31884 91177308-0d34-0410-b5e6-96231b3b80d8

Global label not handled correctly.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31883 91177308-0d34-0410-b5e6-96231b3b80d8

r13 is the thread pointer on darwin/ppc64, don't allocate it.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31882 91177308-0d34-0410-b5e6-96231b3b80d8

Fix PR1005

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31881 91177308-0d34-0410-b5e6-96231b3b80d8

Fix PR1001, patch by Nikhil Patil!

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31880 91177308-0d34-0410-b5e6-96231b3b80d8

Fix codegen for x86-64 on systems (like ppc or i386) that don't have 64-bit
features autodetected.  This fixes PR1010 and Regression/CodeGen/X86/xmm-r64.ll
on non-x86-64 hosts.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31879 91177308-0d34-0410-b5e6-96231b3b80d8

Fix PR1011 and CodeGen/Generic/2006-11-20-DAGCombineCrash.ll

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31878 91177308-0d34-0410-b5e6-96231b3b80d8

Testcase for PR1011

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31877 91177308-0d34-0410-b5e6-96231b3b80d8

setOperand should not zap the operand list or add implicit operands to an
instruction.  Doing so breaks the FP stackifier, the alpha branch selector
the sparc fpmover.

This fixes PR1012 and CodeGen/X86/fp-stack-compare.ll

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31876 91177308-0d34-0410-b5e6-96231b3b80d8

Testcase for pr1012

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31875 91177308-0d34-0410-b5e6-96231b3b80d8

We should perform extra load for dllimported calls

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31874 91177308-0d34-0410-b5e6-96231b3b80d8

Fixed typo

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31871 91177308-0d34-0410-b5e6-96231b3b80d8

Adding 1.9 release and removing references to cfrontend

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31870 91177308-0d34-0410-b5e6-96231b3b80d8

Changing cfrontend dir to llvm-gcc3.4 as the example.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31869 91177308-0d34-0410-b5e6-96231b3b80d8

For PR950:
First in a series of patches to convert SetCondInst into ICmpInst and
FCmpInst using only two opcodes and having the instructions contain their
predicate value. Nothing uses these classes yet. More patches to follow.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31867 91177308-0d34-0410-b5e6-96231b3b80d8

For PR950:
Retain the signedness of the old integer types in a new TypeInfo structure
so that it can be used in the grammar to implement auto-upgrade of things
that depended on signedness of types. This doesn't implement any new
functionality in the AsmParser, its just plumbing for future changes.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31866 91177308-0d34-0410-b5e6-96231b3b80d8

Corrections from review.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31865 91177308-0d34-0410-b5e6-96231b3b80d8

Don't transform in another bad case: if the block is empty, it should be
simplified before we do this xform so that our cost model is accurate.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31864 91177308-0d34-0410-b5e6-96231b3b80d8

Fix another case we *don't* want to do this xform.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31861 91177308-0d34-0410-b5e6-96231b3b80d8

Update a reference to the old setcc instruction.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31860 91177308-0d34-0410-b5e6-96231b3b80d8

For PR950:
Documentation preview of the upcoming icmp and fcmp instructions that
will replace the various setcc instructions.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31859 91177308-0d34-0410-b5e6-96231b3b80d8

MachineInstr includes this

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31858 91177308-0d34-0410-b5e6-96231b3b80d8

Implement operator<< for machine basic blocks to make it easier to dump them.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31857 91177308-0d34-0410-b5e6-96231b3b80d8

make the previous change more aggressive, moving any block with no fallthrough.
This speeds up yacr2 by 7% on a core2.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31856 91177308-0d34-0410-b5e6-96231b3b80d8

Minor code layout tweak: if we have something like this:

  if (cond) goto BB2
BB1:
    ...
    return;
BB2:
   ...

Move BB1 to the end of the function so that the code falls through in the
non-return case.  This has the effect of moving assert (and other no-return
call) bodies and return blocks out of loops.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31855 91177308-0d34-0410-b5e6-96231b3b80d8

Do not convert massive blocks on phi nodes into select statements.  Instead
only do these transformations if there are a small number of phi's.
This speeds up Ptrdist/ks from 2.35s to 2.19s on my mac pro.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31853 91177308-0d34-0410-b5e6-96231b3b80d8

converting massive blocks of phis into selects like this is silly.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31852 91177308-0d34-0410-b5e6-96231b3b80d8

Make the awk filter print out which files its eliminating.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31851 91177308-0d34-0410-b5e6-96231b3b80d8

This is a utility for preparing patches against LLVM. It runs cvs diff with
the correct options, places the patch content in order that makes sense for
review, and filters cruft out of the patch (like diffs in *.cvs files). It
produces both a NAME.patch.raw (unfiltered) and NAME.patch (filtered) file.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31850 91177308-0d34-0410-b5e6-96231b3b80d8

Provide a list of gccld optimization switches.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31849 91177308-0d34-0410-b5e6-96231b3b80d8

Fix bugs in this. What was I thinking??? :)

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31848 91177308-0d34-0410-b5e6-96231b3b80d8

Movind newly-intorudced dll* linkage stuff & windows CC to "supported"
list

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31847 91177308-0d34-0410-b5e6-96231b3b80d8

fix bug

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31846 91177308-0d34-0410-b5e6-96231b3b80d8

first draft of 1.9 release notes

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31845 91177308-0d34-0410-b5e6-96231b3b80d8

Fail even if opt doesn't print anything.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31844 91177308-0d34-0410-b5e6-96231b3b80d8

Add a simple test to make sure getModRefInfo is 1/2 way sane.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31842 91177308-0d34-0410-b5e6-96231b3b80d8

Have ConstantExprs upgrade opcodes the same way as instructions.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31841 91177308-0d34-0410-b5e6-96231b3b80d8

on ppc64, float arguments take 8-byte stack slots not 4-byte stack slots.
Also, valist should create a pointer RC reg class value, not a GPRC value.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31840 91177308-0d34-0410-b5e6-96231b3b80d8

make sure to safe LR8 in the right stack slot for PPC64

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31839 91177308-0d34-0410-b5e6-96231b3b80d8

Pretty print 'rldicr r2, r2, 2, 61' as 'sldi r2, r2, 2'.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31838 91177308-0d34-0410-b5e6-96231b3b80d8

Rewrite the branch selector to be correct in the face of large functions.
The algorithm it used before wasn't 100% correct, we now use an iterative
expansion model.  This fixes assembler errors when compiling 403.gcc with
tail merging enabled.

Change the way the branch selector works overall: Now, the isel generates
PPC::BCC instructions (as it used to) directly, and these BCC instructions
are emitted to the output or jitted directly if branches don't need
expansion.  Only if branches need expansion are instructions rewritten
and created.  This should make branch select faster, and eliminates the
Bxx instructions from the .td file.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31837 91177308-0d34-0410-b5e6-96231b3b80d8

add encoding for BCC, after finally wrestling strange ppc/tblgen endianness
issues to the ground.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31836 91177308-0d34-0410-b5e6-96231b3b80d8

convert PPC::BCC to use the 'pred' operand instead of separate predicate
value and CR reg #.  This requires swapping the order of these everywhere
that touches BCC and requires us to write custom matching logic for
PPCcondbranch :(

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31835 91177308-0d34-0410-b5e6-96231b3b80d8

rename PPC::COND_BRANCH to PPC::BCC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31834 91177308-0d34-0410-b5e6-96231b3b80d8

start using PPC predicates more consistently.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31833 91177308-0d34-0410-b5e6-96231b3b80d8

For unsigned 8-bit division. Use movzbw to set the lower 8 bits of AX while
clearing the upper 8-bits instead of issuing two instructions. This also
eliminates the need to target the AH register which can be problematic on
x86-64.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31832 91177308-0d34-0410-b5e6-96231b3b80d8

Hopefully a good crack at making debugging work on intel -disable-fp-elim.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31830 91177308-0d34-0410-b5e6-96231b3b80d8

Add a x86-64 test case: movb %sil, %ah is illegal.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31829 91177308-0d34-0410-b5e6-96231b3b80d8

Assert unhandled case.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31828 91177308-0d34-0410-b5e6-96231b3b80d8

Case sensitive not.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31827 91177308-0d34-0410-b5e6-96231b3b80d8

Check for MaxAlign.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31826 91177308-0d34-0410-b5e6-96231b3b80d8

Tests to verify PowerPC ABI.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31825 91177308-0d34-0410-b5e6-96231b3b80d8

1. Ignore the -disable-fp-elim when the routine is a leaf.
2. Offsets on 64-bit stores are still in bytes.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31824 91177308-0d34-0410-b5e6-96231b3b80d8

Typo. Fix the nightly tests.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31823 91177308-0d34-0410-b5e6-96231b3b80d8