Initial checking of a perfect shuffle generation program for 4-element
Altivec vectors.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27736 91177308-0d34-0410-b5e6-96231b3b80d8

Better way to splat v2f64

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27735 91177308-0d34-0410-b5e6-96231b3b80d8

movduprm, movshduprm bugs

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27734 91177308-0d34-0410-b5e6-96231b3b80d8

Encoding bugs

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27733 91177308-0d34-0410-b5e6-96231b3b80d8

Can't fold loads into alias vector SSE ops used for scalar operation. The load
address has to be 16-byte aligned but the values aren't spilled to 128-bit
locations.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27732 91177308-0d34-0410-b5e6-96231b3b80d8

Implement a TODO: have the legalizer canonicalize a bunch of operations to
one type (v4i32) so that we don't have to write patterns for each type, and
so that more CSE opportunities are exposed.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27731 91177308-0d34-0410-b5e6-96231b3b80d8

Add support for promoting stores from one legal type to another, allowing us
to write one pattern for vector stores instead of 4.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27730 91177308-0d34-0410-b5e6-96231b3b80d8

Make the BUILD_VECTOR lowering code much more aggressive w.r.t constant vectors.
Remove some done items from the todo list.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27729 91177308-0d34-0410-b5e6-96231b3b80d8

Fix a bug in the 'shuffle(undef,x,mask) -> shuffle(x, undef,mask')' xform
Make the insert/extract elt -> shuffle code more aggressive.

This fixes CodeGen/PowerPC/vec_shuffle.ll

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27728 91177308-0d34-0410-b5e6-96231b3b80d8

Canonicalize shuffle(undef,x,mask) -> shuffle(x, undef,mask').

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27727 91177308-0d34-0410-b5e6-96231b3b80d8

Fix a crash when faced with a shuffle vector that has an undef in its mask.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27726 91177308-0d34-0410-b5e6-96231b3b80d8

Add patterns for matching vnots with bit converted inputs.  Most of these will
go away when I start using evan's binop type canonicalizer

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27725 91177308-0d34-0410-b5e6-96231b3b80d8

Add a new vnot_conv predicate for matching vnot's where the allones vector is
bitconverted from some other type.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27724 91177308-0d34-0410-b5e6-96231b3b80d8

Make these predicates return true for bit_convert(buildvector)'s as well as
buildvectors.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27723 91177308-0d34-0410-b5e6-96231b3b80d8

More encoding bugs

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27722 91177308-0d34-0410-b5e6-96231b3b80d8

pslldrm, psrawrm, etc. encoding bug

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27721 91177308-0d34-0410-b5e6-96231b3b80d8

hsubp{s|d} encoding bug

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27720 91177308-0d34-0410-b5e6-96231b3b80d8

Silly bug

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27719 91177308-0d34-0410-b5e6-96231b3b80d8

Do not use movs{h|l}dup for a shuffle with a single non-undef node.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27718 91177308-0d34-0410-b5e6-96231b3b80d8

significant cleanups to code that uses insert/extractelt heavily.  This builds
maximal shuffles out of them where possible.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27717 91177308-0d34-0410-b5e6-96231b3b80d8

Added SSE (and other) entries to foldMemoryOperand().

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27716 91177308-0d34-0410-b5e6-96231b3b80d8

Some clean up

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27715 91177308-0d34-0410-b5e6-96231b3b80d8

Allow undef in a shuffle mask

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27714 91177308-0d34-0410-b5e6-96231b3b80d8

Move these ctors out of line

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27713 91177308-0d34-0410-b5e6-96231b3b80d8

These instructions always return a packed vector.  Improve the class definitions to expose this fact.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27712 91177308-0d34-0410-b5e6-96231b3b80d8

Last few SSE3 intrinsics.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27711 91177308-0d34-0410-b5e6-96231b3b80d8

Teach scalarrepl to promote unions of vectors and floats, producing
insert/extractelement operations.  This implements
Transforms/ScalarRepl/vector_promote.ll

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27710 91177308-0d34-0410-b5e6-96231b3b80d8

New testcase, checking to see we can turn this code:

void test(vector float *F, float f) {
  vector float G = *F + *F;
  *((float*)&G) = f;
  *F = G + G;
}

void test2(vector float *F, float f) {
  vector float G = *F + *F;
  ((float*)&G)[2] = f;
  *F = G + G;
}

void test3(vector float *F, float *f) {
  vector float G = *F + *F;
  *f = ((float*)&G)[2];
}

void test4(vector float *F, float *f) {
  vector float G = *F + *F;
  *f = *((float*)&G);
}

into insert/extract element operations with no memory traffic.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27709 91177308-0d34-0410-b5e6-96231b3b80d8

Adding back vector instructions to keep in mainline.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27704 91177308-0d34-0410-b5e6-96231b3b80d8

Removing unimplemented vector instructions from language referrence.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27703 91177308-0d34-0410-b5e6-96231b3b80d8

Correct the Superclasses list for GlobalVariable and Function to indicate
that they are "Constant" as they derive from GlobalValue. Also, fix some of
the wording where it mentions this.
Patch inspired by Nai Xia.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27701 91177308-0d34-0410-b5e6-96231b3b80d8

Allow an option, -with-externals, to be specified that provides the location
of the external tests. This turns into --with-externals option to configure.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27700 91177308-0d34-0410-b5e6-96231b3b80d8

Misc. SSE2 intrinsics: clflush, lfench, mfence

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27699 91177308-0d34-0410-b5e6-96231b3b80d8

We were not adjusting the frame size to ensure proper alignment when alloca /
vla are present in the function. This causes a crash when a leaf function
allocates space on the stack used to store / load with 128-bit SSE
instructions.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27698 91177308-0d34-0410-b5e6-96231b3b80d8

New entry

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27697 91177308-0d34-0410-b5e6-96231b3b80d8

Don't print out the install command for Intrinsics.gen unless VERBOSE mode.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27696 91177308-0d34-0410-b5e6-96231b3b80d8

Make this assertion better

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27695 91177308-0d34-0410-b5e6-96231b3b80d8

Force a specific config, because this test fails in certain configs otherwise.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27694 91177308-0d34-0410-b5e6-96231b3b80d8

Move the rest of the PPCTargetLowering::LowerOperation cases out into
separate functions, for simplicity and code clarity.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27693 91177308-0d34-0410-b5e6-96231b3b80d8

Pull the VECTOR_SHUFFLE and BUILD_VECTOR lowering code out into separate
functions, which makes the code much cleaner :)

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27692 91177308-0d34-0410-b5e6-96231b3b80d8

Implement value #'ing for vector operations, implementing
Regression/Transforms/GCSE/vectorops.ll

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27691 91177308-0d34-0410-b5e6-96231b3b80d8

new testcase, vector operations should be CSE'd

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27690 91177308-0d34-0410-b5e6-96231b3b80d8

Remove this test, there is no need to test GCC's bugs

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27689 91177308-0d34-0410-b5e6-96231b3b80d8

Ahem. HEAD -> 1.8cvs not 1.7 (I'm an idiot).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27687 91177308-0d34-0410-b5e6-96231b3b80d8

remove the "cvs" part of the version number for the release branch.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27686 91177308-0d34-0410-b5e6-96231b3b80d8

pcmpeq* and pcmpgt* intrinsics.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27685 91177308-0d34-0410-b5e6-96231b3b80d8

psll*, psrl*, and psra* intrinsics.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27684 91177308-0d34-0410-b5e6-96231b3b80d8

Remove the .cvsignore file so this directory can be pruned.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27683 91177308-0d34-0410-b5e6-96231b3b80d8

Remove .cvsignore so that this directory can be pruned.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27682 91177308-0d34-0410-b5e6-96231b3b80d8

My addition of the xfail marker threw off the line #. move it.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27678 91177308-0d34-0410-b5e6-96231b3b80d8

Use quotes properly so that the possibility of a null variable set is
eliminated. This can happen, for example, if LLVM is configured without
llvm-gcc in which case things like LLVMGCC_VERSION will be empty. In
such cases, deja-gnu fails with:
can't read "llvmgcc_version": no such variable
because it sees:
set llvmgcc_version
instead of:
set llvmgcc_version ""

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27676 91177308-0d34-0410-b5e6-96231b3b80d8

from the linux kernel

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27674 91177308-0d34-0410-b5e6-96231b3b80d8

Fix this regex to match what llvmgcc4 produces also

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27673 91177308-0d34-0410-b5e6-96231b3b80d8

Handle some kernel code than ends in [0 x sbyte].  I think this is safe

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27672 91177308-0d34-0410-b5e6-96231b3b80d8

Expand some code with temporary variables to rid ourselves of the warning
about "dereferencing type-punned pointer will break strict-aliasing rules"

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27671 91177308-0d34-0410-b5e6-96231b3b80d8

Fix an incorrect prototype for this intrinsic, fixing
CFrontend/2003-08-18-SigSetJmp.c  with llvm-gcc3.  This is part of PR733.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27670 91177308-0d34-0410-b5e6-96231b3b80d8

Try xfailing this

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27669 91177308-0d34-0410-b5e6-96231b3b80d8

Doh. PANDrm, etc. are not commutable.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27668 91177308-0d34-0410-b5e6-96231b3b80d8

fix this for the more restrictive linkage

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27667 91177308-0d34-0410-b5e6-96231b3b80d8

These tests are now xfailed for llvmgcc4.  This is PR735, unlikely to be
resolved before 1.7 :(

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27666 91177308-0d34-0410-b5e6-96231b3b80d8

Only look at .ll files in this directory

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27665 91177308-0d34-0410-b5e6-96231b3b80d8

This file is an invalid C file, test that it is properly rejected

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27664 91177308-0d34-0410-b5e6-96231b3b80d8

Another case where a dead cast was causing the test to spuriously
fail with the new front-end.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27663 91177308-0d34-0410-b5e6-96231b3b80d8

Don't get confused by dead casts.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27662 91177308-0d34-0410-b5e6-96231b3b80d8

Fix an accidental commit.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27661 91177308-0d34-0410-b5e6-96231b3b80d8

This test fails and I don't know why, xfail it until andrew gets a chance to
look at it.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27660 91177308-0d34-0410-b5e6-96231b3b80d8

Oops, move misplaced test

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27659 91177308-0d34-0410-b5e6-96231b3b80d8

These are only XFAILs with llvmgcc3, they pass with llvmgcc4

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27658 91177308-0d34-0410-b5e6-96231b3b80d8

Force non-darwin targets to use a static relo model.  This fixes PR734,
tested by CodeGen/Generic/vector.ll

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27657 91177308-0d34-0410-b5e6-96231b3b80d8

Add a run with an unusual target triple, revert the patch that sent output to
dev null as it broke the test and doesn't add anything.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27656 91177308-0d34-0410-b5e6-96231b3b80d8

Update a count, this test now passes.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27655 91177308-0d34-0410-b5e6-96231b3b80d8

add a note, move an altivec todo to the altivec list.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27654 91177308-0d34-0410-b5e6-96231b3b80d8

linear -> constant time

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27652 91177308-0d34-0410-b5e6-96231b3b80d8

Add the README files to the distribution.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27651 91177308-0d34-0410-b5e6-96231b3b80d8

Add missing things to the distribution.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27650 91177308-0d34-0410-b5e6-96231b3b80d8

Distribute the Stacker samples directory.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27648 91177308-0d34-0410-b5e6-96231b3b80d8

psad, pmax, pmin intrinsics.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27647 91177308-0d34-0410-b5e6-96231b3b80d8

pmin, pmax, and psad intrinsics.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27646 91177308-0d34-0410-b5e6-96231b3b80d8

Various SSE2 packed integer intrinsics: pmulhuw, pavgw, etc.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27645 91177308-0d34-0410-b5e6-96231b3b80d8

X86 SSE2 supports v8i16 multiplication

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27644 91177308-0d34-0410-b5e6-96231b3b80d8

Update

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27643 91177308-0d34-0410-b5e6-96231b3b80d8

pmul*, pmadd*, and pavg* intrinsics.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27642 91177308-0d34-0410-b5e6-96231b3b80d8

one too many l's

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27641 91177308-0d34-0410-b5e6-96231b3b80d8

quick update

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27640 91177308-0d34-0410-b5e6-96231b3b80d8

padds{b|w}, paddus{b|w}, psubs{b|w}, psubus{b|w} intrinsics.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27639 91177308-0d34-0410-b5e6-96231b3b80d8

Naming inconsistency.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27638 91177308-0d34-0410-b5e6-96231b3b80d8

SSE / SSE2 conversion intrinsics.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27637 91177308-0d34-0410-b5e6-96231b3b80d8

Fix a typo in the name of a file.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27636 91177308-0d34-0410-b5e6-96231b3b80d8

Added the ability to xfail based on llvmgcc version

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27635 91177308-0d34-0410-b5e6-96231b3b80d8

Update my credits. Can I add "Build breaker" to my entry?

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27634 91177308-0d34-0410-b5e6-96231b3b80d8

All "integer" logical ops (pand, por, pxor) are now promoted to v2i64.
Clean up and fix various logical ops issues.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27633 91177308-0d34-0410-b5e6-96231b3b80d8

Promote vector AND, OR, and XOR

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27632 91177308-0d34-0410-b5e6-96231b3b80d8

Don't dump the llc assembly output to stdout.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27631 91177308-0d34-0410-b5e6-96231b3b80d8

Make sure CVS versions of yacc and lex files get distributed.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27630 91177308-0d34-0410-b5e6-96231b3b80d8

Distribute CVS versions of lex files

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27629 91177308-0d34-0410-b5e6-96231b3b80d8

Distribute the lex/yacc files from the cvs version.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27628 91177308-0d34-0410-b5e6-96231b3b80d8

Additional files to distribute.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27627 91177308-0d34-0410-b5e6-96231b3b80d8

Include the win32 and Xcode directories in the distribution.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27626 91177308-0d34-0410-b5e6-96231b3b80d8

Get rid of a signed/unsigned compare warning.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27625 91177308-0d34-0410-b5e6-96231b3b80d8

Add a new way to match vector constants, which make it easier to bang bits of
different types.

Codegen spltw(0x7FFFFFFF) and spltw(0x80000000) without a constant pool load,
implementing PowerPC/vec_constants.ll:test1.  This compiles:

typedef float vf __attribute__ ((vector_size (16)));
typedef int vi __attribute__ ((vector_size (16)));
void test(vi *P1, vi *P2, vf *P3) {
  *P1 &= (vi){0x80000000,0x80000000,0x80000000,0x80000000};
  *P2 &= (vi){0x7FFFFFFF,0x7FFFFFFF,0x7FFFFFFF,0x7FFFFFFF};
  *P3 = vec_abs((vector float)*P3);
}

to:

_test:
        mfspr r2, 256
        oris r6, r2, 49152
        mtspr 256, r6
        vspltisw v0, -1
        vslw v0, v0, v0
        lvx v1, 0, r3
        vand v1, v1, v0
        stvx v1, 0, r3
        lvx v1, 0, r4
        vandc v1, v1, v0
        stvx v1, 0, r4
        lvx v1, 0, r5
        vandc v0, v1, v0
        stvx v0, 0, r5
        mtspr 256, r2
        blr

instead of (with two constant pool entries):

_test:
        mfspr r2, 256
        oris r6, r2, 49152
        mtspr 256, r6
        li r6, lo16(LCPI1_0)
        lis r7, ha16(LCPI1_0)
        li r8, lo16(LCPI1_1)
        lis r9, ha16(LCPI1_1)
        lvx v0, r7, r6
        lvx v1, 0, r3
        vand v0, v1, v0
        stvx v0, 0, r3
        lvx v0, r9, r8
        lvx v1, 0, r4
        vand v1, v1, v0
        stvx v1, 0, r4
        lvx v1, 0, r5
        vand v0, v1, v0
        stvx v0, 0, r5
        mtspr 256, r2
        blr

GCC produces (with 2 cp entries):

_test:
        mfspr r0,256
        stw r0,-4(r1)
        oris r0,r0,0xc00c
        mtspr 256,r0
        lis r2,ha16(LC0)
        lis r9,ha16(LC1)
        la r2,lo16(LC0)(r2)
        lvx v0,0,r3
        lvx v1,0,r5
        la r9,lo16(LC1)(r9)
        lwz r12,-4(r1)
        lvx v12,0,r2
        lvx v13,0,r9
        vand v0,v0,v12
        stvx v0,0,r3
        vspltisw v0,-1
        vslw v12,v0,v0
        vandc v1,v1,v12
        stvx v1,0,r5
        lvx v0,0,r4
        vand v0,v0,v13
        stvx v0,0,r4
        mtspr 256,r12
        blr

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27624 91177308-0d34-0410-b5e6-96231b3b80d8