Add a vselect test case.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27557 91177308-0d34-0410-b5e6-96231b3b80d8

Conditional move of vector types.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27556 91177308-0d34-0410-b5e6-96231b3b80d8

New entries

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27555 91177308-0d34-0410-b5e6-96231b3b80d8

Use movaps to do VR128 reg-to-reg copies for now. It's shorter and available for SSE1.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27554 91177308-0d34-0410-b5e6-96231b3b80d8

Update docs for ENABLE_ASSERTIONS -> DISABLE_ASSERTIONS

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27553 91177308-0d34-0410-b5e6-96231b3b80d8

Check for DISABLE_ASSERTIONS, not ENABLE_ASSERTIONS

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27552 91177308-0d34-0410-b5e6-96231b3b80d8

Fix a comment.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27551 91177308-0d34-0410-b5e6-96231b3b80d8

Change the assertions defaults:
1. Assertions now default to on for all builds
2. If you want them disabled you must (a) --disable-assertions to configure
   or DISABLE_ASSERTIONS=1 to make.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27548 91177308-0d34-0410-b5e6-96231b3b80d8

Move the vector instructions to their own subsection.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27545 91177308-0d34-0410-b5e6-96231b3b80d8

properly mark vector selects as expanded to select_cc

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27544 91177308-0d34-0410-b5e6-96231b3b80d8

Add VRRC select support

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27543 91177308-0d34-0410-b5e6-96231b3b80d8

Add code generator support for VSELECT

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27542 91177308-0d34-0410-b5e6-96231b3b80d8

Add a new VSELECT node.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27541 91177308-0d34-0410-b5e6-96231b3b80d8

Disable switch lowering for targets based on the selection dag isel,
letting the code generator handle them directly.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27539 91177308-0d34-0410-b5e6-96231b3b80d8

Implement PowerPC/CodeGen/vec_splat.ll:spltish to use vsplish instead of a
constant pool load.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27538 91177308-0d34-0410-b5e6-96231b3b80d8

add new testcase

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27537 91177308-0d34-0410-b5e6-96231b3b80d8

Change the interface to the predicate that determines if vsplti* can be used.
No functionality changes.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27536 91177308-0d34-0410-b5e6-96231b3b80d8

Misc. project changes.  html + build config changes supplied by Devang.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27535 91177308-0d34-0410-b5e6-96231b3b80d8

Initialize SDOperand values because the gcc 4.0.2 compiler complains about
them.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27534 91177308-0d34-0410-b5e6-96231b3b80d8

Canonicalize vvector_shuffle(x,x) -> vvector_shuffle(x,undef) to enable patterns
to match again :)

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27533 91177308-0d34-0410-b5e6-96231b3b80d8

Add constant replacement for insertelement/vectorshuffle constant exprs

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27532 91177308-0d34-0410-b5e6-96231b3b80d8

Wrap long lines, fix verification error

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27531 91177308-0d34-0410-b5e6-96231b3b80d8

use the right escapes

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27530 91177308-0d34-0410-b5e6-96231b3b80d8

Codegen shufflevector as VVECTOR_SHUFFLE

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27529 91177308-0d34-0410-b5e6-96231b3b80d8

add a sanity check: LegalizeOp should return a value that is the same type
as its input.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27528 91177308-0d34-0410-b5e6-96231b3b80d8

use isValidOperands instead of duplicating checks

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27527 91177308-0d34-0410-b5e6-96231b3b80d8

Regenerate

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27526 91177308-0d34-0410-b5e6-96231b3b80d8

Use isValidOperands instead of duplicating or eliding checks.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27525 91177308-0d34-0410-b5e6-96231b3b80d8

Use the isValidOperands helper instead of duplicating checking code

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27524 91177308-0d34-0410-b5e6-96231b3b80d8

Add methods to check insertelement/extractelement instructions for validity,
check validity when instructions are created.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27523 91177308-0d34-0410-b5e6-96231b3b80d8

Add methods to check insert/extract element instructions for validity

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27522 91177308-0d34-0410-b5e6-96231b3b80d8

regenerate

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27521 91177308-0d34-0410-b5e6-96231b3b80d8

Add insertelement and shufflevector constantexpr support

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27520 91177308-0d34-0410-b5e6-96231b3b80d8

Fix recursion:
1. Make setting OPTIONAL_DIRS use immediate assignment
2. Include Makefile.config before ifeq tests and then Makefile.rules later
   instead of Makefile.common up front. This ensures that the variable
   values are set before the ifeq statements in Makefile.rules are
   evaluated.
With this, recursion into projects/examples is corrected.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27519 91177308-0d34-0410-b5e6-96231b3b80d8

Don't build projects/examples if libs-only was specified.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27518 91177308-0d34-0410-b5e6-96231b3b80d8

INSERT_VECTOR_ELT lowering bug:
  store vector to $esp
  store element to $esp + sizeof(VT) * index
  load  vector from $esp
The bug is VT is the type of the vector element, not the type of the vector!

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27517 91177308-0d34-0410-b5e6-96231b3b80d8

Ensure that the automatic "CVS build" detection works for both
objdir == srcdir and objdir != srcdir.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27516 91177308-0d34-0410-b5e6-96231b3b80d8

Add shufflevector support

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27515 91177308-0d34-0410-b5e6-96231b3b80d8

Stub out shufflevector

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27514 91177308-0d34-0410-b5e6-96231b3b80d8

Add supprot for shufflevector

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27513 91177308-0d34-0410-b5e6-96231b3b80d8

regenerate

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27512 91177308-0d34-0410-b5e6-96231b3b80d8

Parse shufflevector

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27511 91177308-0d34-0410-b5e6-96231b3b80d8

Add shufflevector support, todo, implement better constant folding.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27510 91177308-0d34-0410-b5e6-96231b3b80d8

Add shufflevector reading support

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27509 91177308-0d34-0410-b5e6-96231b3b80d8

new testcase for shufflevector

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27508 91177308-0d34-0410-b5e6-96231b3b80d8

Add a new shufflevector instruction

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27507 91177308-0d34-0410-b5e6-96231b3b80d8

ldmxcsr and stmxcsr.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27506 91177308-0d34-0410-b5e6-96231b3b80d8

Fix int_x86_sse_stmxcsr and int_x86_sse_ldmxcsr. Not directly translated from
gcc builtins. They are not lowered into these intrinsics which take a ptr
argument.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27505 91177308-0d34-0410-b5e6-96231b3b80d8

Get Visual Studio building again.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27504 91177308-0d34-0410-b5e6-96231b3b80d8

Remove section change in function end, preventing override of function's real
section.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27503 91177308-0d34-0410-b5e6-96231b3b80d8

Description for the new shufflevector instruction I'm adding.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27502 91177308-0d34-0410-b5e6-96231b3b80d8

Code clean up.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27501 91177308-0d34-0410-b5e6-96231b3b80d8

Doh!

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27500 91177308-0d34-0410-b5e6-96231b3b80d8

Invert the tests on LLVMGCC_MAJVERS to check for value 4 instead of value
3. This ensures that if llvm-gcc isn't available and consequently the value
of LLVMGCC_MAJVERS is blank, that the old (include runtime) behavior will
persist.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27499 91177308-0d34-0410-b5e6-96231b3b80d8

Added patterns for MOVHPSmr and MOVLPSmr.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27497 91177308-0d34-0410-b5e6-96231b3b80d8

Keep track of an Mac OS X / x86 ABI bug.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27496 91177308-0d34-0410-b5e6-96231b3b80d8

Remove int_x86_sse_storeh_ps and int_x86_sse_storel_ps. These are now lowered
by the frontend.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27495 91177308-0d34-0410-b5e6-96231b3b80d8

Make sure that debug labels are defined within the same section and after the
entry point of a function.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27494 91177308-0d34-0410-b5e6-96231b3b80d8

add a newline at the end of the file

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27492 91177308-0d34-0410-b5e6-96231b3b80d8

Foundation for call frame information.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27491 91177308-0d34-0410-b5e6-96231b3b80d8

THis -> This

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27490 91177308-0d34-0410-b5e6-96231b3b80d8

For PR723:
Configure with the --enable-optimized and --enable-assertions option to
ensure that binary RPM packages are built with the options required by the
feature request.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27489 91177308-0d34-0410-b5e6-96231b3b80d8

For PR723:
1. Don't force debug builds to have assertion checking turned on always.
   Let the default (on) be taken, or overridden by the command line
2. Create two new BuildModes based on assertion checking: Release+Assert
   and Debug-Assert.
3. Ensure that when building a distribution we get a release build with
   assertions enabled, regardless of the tree's configuration.
4. (unrelated) Fix library name generation for llvm-config usage.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27488 91177308-0d34-0410-b5e6-96231b3b80d8

For PR723:
Support detection of a "CVS" directory at configure time to distinguish
whether this is a release build or a "from tree" build. This knowledge is
used to set the defaults for --enable-optimzied and --enable-assertions
options.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27487 91177308-0d34-0410-b5e6-96231b3b80d8

For PR723:
- Provide for the include of ENABLE_ASSERTIONS flag from the configuration.
- Enhance the documentation for ENABLE_OPTIMIZED

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27486 91177308-0d34-0410-b5e6-96231b3b80d8

For PR723:
Print out something at the end of the build that indicates what kind of
build has just completed.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27485 91177308-0d34-0410-b5e6-96231b3b80d8

Ignore stuff built by "make dist".

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27484 91177308-0d34-0410-b5e6-96231b3b80d8

A saner workaround. I hope.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27483 91177308-0d34-0410-b5e6-96231b3b80d8

Temporary workaround for a Mac OSX specific issue.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27482 91177308-0d34-0410-b5e6-96231b3b80d8

Added more shuffle tests

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27481 91177308-0d34-0410-b5e6-96231b3b80d8

We have an assertion that checks that we do not encode null values into the
.bc file if they are supposed to be implicit.  This is cool, except that it
checked *after* constant expr folding: improving constant expr folding could
cause the .bc reader to assert out on old .bc files.  Move the check so that
it checks all simple constants, but no constantexprs.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27480 91177308-0d34-0410-b5e6-96231b3b80d8

Constant fold extractelement(zero, x) -> zero

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27479 91177308-0d34-0410-b5e6-96231b3b80d8

Fix inlining of insert/extract element constantexprs

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27478 91177308-0d34-0410-b5e6-96231b3b80d8

Add code to RemapOperand() to handle Instruction::ExtractElement and
Instruction::InsertElement.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27477 91177308-0d34-0410-b5e6-96231b3b80d8

A MOVPS2SSmr, i.e. _mm_store_ss, encoding bug.
Also MOVPDI2DIrr.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27476 91177308-0d34-0410-b5e6-96231b3b80d8

Remove int_x86_sse_loadh_ps and int_x86_sse_loadl_ps. These are now lowered
by the frontend to shuffles.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27475 91177308-0d34-0410-b5e6-96231b3b80d8

- movlp{s|d} and movhp{s|d} support.
- Normalize shuffle nodes so result vector lower half elements come from the
  first vector, the rest come from the second vector. (Except for the
  exceptions :-).
- Other minor fixes.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27474 91177308-0d34-0410-b5e6-96231b3b80d8

New entries.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27473 91177308-0d34-0410-b5e6-96231b3b80d8

1. If both vector operands of a vector_shuffle are undef, turn it into an undef.
2. A shuffle mask element can also be an undef.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27472 91177308-0d34-0410-b5e6-96231b3b80d8

This may be overconservative, but it lets the new cfe compile

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27471 91177308-0d34-0410-b5e6-96231b3b80d8

Add an item

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27470 91177308-0d34-0410-b5e6-96231b3b80d8

Make sure to return the result in the right type.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27469 91177308-0d34-0410-b5e6-96231b3b80d8

Move this to lib/Analysis.
This reduces Core size, yay.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27468 91177308-0d34-0410-b5e6-96231b3b80d8

Match vpku[hw]um(x,x).
Convert vsldoi(x,x) to work the same way other (x,x) cases work.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27467 91177308-0d34-0410-b5e6-96231b3b80d8

Add testcases for vpku[hw]um(x,x)

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27466 91177308-0d34-0410-b5e6-96231b3b80d8

Add the variable llvmgccmajvers to the site.exp file. This will contain
the major version number of llvm-gcc, as configured.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27465 91177308-0d34-0410-b5e6-96231b3b80d8

Avoid building the runtime libraries if llvm-gcc version 4 is detected.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27464 91177308-0d34-0410-b5e6-96231b3b80d8

Add support for matching vmrg(x,x) patterns

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27463 91177308-0d34-0410-b5e6-96231b3b80d8

Add vmrg(x,x) tests

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27462 91177308-0d34-0410-b5e6-96231b3b80d8

For support of new GCC v4, obtain the full and major versions of the
llvm-gcc command. This will help distinguish which version is being used
so the makefiles can accommodate.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27461 91177308-0d34-0410-b5e6-96231b3b80d8

fix some linking problems with the new gcc

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27460 91177308-0d34-0410-b5e6-96231b3b80d8

Add tests for vmrg[hl]*

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27459 91177308-0d34-0410-b5e6-96231b3b80d8

These intrinsics are now lowered by the FE.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27458 91177308-0d34-0410-b5e6-96231b3b80d8

Pattern match vmrg* instructions, which are now lowered by the CFE into shuffles.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27457 91177308-0d34-0410-b5e6-96231b3b80d8

Infer element types for shuffle masks

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27456 91177308-0d34-0410-b5e6-96231b3b80d8

rename a method, to avoid confusion with llvm intrinsics.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27455 91177308-0d34-0410-b5e6-96231b3b80d8

test vperm promotion

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27454 91177308-0d34-0410-b5e6-96231b3b80d8

remove two done items

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27453 91177308-0d34-0410-b5e6-96231b3b80d8

Lower vperm(x,y, mask) -> shuffle(x,y,mask) if mask is constant.  This allows
us to compile oh-so-realistic stuff like this:

 vec_vperm(A, B, (vector unsigned char){14});

to:
        vspltb v0, v0, 14

instead of:

        vspltisb v0, 14
        vperm v0, v2, v1, v0

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27452 91177308-0d34-0410-b5e6-96231b3b80d8

this is no longer an intrinsic.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27451 91177308-0d34-0410-b5e6-96231b3b80d8