add a method for hacking on JTIdx's

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31270 91177308-0d34-0410-b5e6-96231b3b80d8

const'ify jump table stuff

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31269 91177308-0d34-0410-b5e6-96231b3b80d8

add a method for deleting dead jump tables.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31268 91177308-0d34-0410-b5e6-96231b3b80d8

add an assert

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31267 91177308-0d34-0410-b5e6-96231b3b80d8

don't print dead jump tables

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31266 91177308-0d34-0410-b5e6-96231b3b80d8

determine whether a change was made

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31265 91177308-0d34-0410-b5e6-96231b3b80d8

implement the BlockHasNoFallThrough hook

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31264 91177308-0d34-0410-b5e6-96231b3b80d8

improve deletion of blocks that just contain branches by knowing that
the pred block doesn't fall through into them if it's a jumptable.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31263 91177308-0d34-0410-b5e6-96231b3b80d8

add another target hook for branch folding.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31262 91177308-0d34-0410-b5e6-96231b3b80d8

Load and stores have not been uniqued properly.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31261 91177308-0d34-0410-b5e6-96231b3b80d8

Split *all* critical edges before isel.  This resolves issues with spill code
being inserted on unsplit critical edges, which introduces (sometimes large
amounts of) partially dead spill code.

This also fixes PR925 + CodeGen/Generic/switch-crit-edge-constant.ll

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31260 91177308-0d34-0410-b5e6-96231b3b80d8

new testcase

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31259 91177308-0d34-0410-b5e6-96231b3b80d8

add option to isCriticalEdge

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31258 91177308-0d34-0410-b5e6-96231b3b80d8

break edges more intelligently

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31257 91177308-0d34-0410-b5e6-96231b3b80d8

Expose a smarter way to break critical edges.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31256 91177308-0d34-0410-b5e6-96231b3b80d8

SplitCriticalEdge checks to see if an edge is critical, don't check twice

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31255 91177308-0d34-0410-b5e6-96231b3b80d8

Fix a serious bug that caused any x86 vector stuff to infinite loop

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31254 91177308-0d34-0410-b5e6-96231b3b80d8

Doh. Must check if GV is constant first before putting it in .cstring.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31253 91177308-0d34-0410-b5e6-96231b3b80d8

Doh. Must check if GV is constant first.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31252 91177308-0d34-0410-b5e6-96231b3b80d8

Emitting raw bytecode is awfully rude. Add -disable-output.
Clarify the test.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31251 91177308-0d34-0410-b5e6-96231b3b80d8

Editorial clean up.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31250 91177308-0d34-0410-b5e6-96231b3b80d8

add a method

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31249 91177308-0d34-0410-b5e6-96231b3b80d8

prepare for a change I'm about to make

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31248 91177308-0d34-0410-b5e6-96231b3b80d8

don't dist internal readme

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31247 91177308-0d34-0410-b5e6-96231b3b80d8

don't dist internal readme's

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31246 91177308-0d34-0410-b5e6-96231b3b80d8

don't distribute internal readme's

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31245 91177308-0d34-0410-b5e6-96231b3b80d8

Make sure all the readme files get distributed.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31244 91177308-0d34-0410-b5e6-96231b3b80d8

Clean up.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31243 91177308-0d34-0410-b5e6-96231b3b80d8

Fix a bug in merged condition handling (CodeGen/Generic/2006-10-27-CondFolding.ll).

Add many fewer CFG edges and PHI node entries.  If there is a switch which has
the same block as multiple destinations, only add that block once as a successor/phi
node (in the jumptable case)

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31242 91177308-0d34-0410-b5e6-96231b3b80d8

new testcase

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31241 91177308-0d34-0410-b5e6-96231b3b80d8

Switch over from SelectionNodeCSEMap to FoldingSet.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31240 91177308-0d34-0410-b5e6-96231b3b80d8

Use doc_code class.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31239 91177308-0d34-0410-b5e6-96231b3b80d8

Clean up

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31238 91177308-0d34-0410-b5e6-96231b3b80d8

Document removeModule().

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31237 91177308-0d34-0410-b5e6-96231b3b80d8

this doesn't occur any more in mason

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31236 91177308-0d34-0410-b5e6-96231b3b80d8

the code in question is now:

        cmpw cr0, r7, r3
        ble cr0, LBB1_5 ;bb25
LBB1_8: ;bb17
        cmpw cr0, r8, r5
        bgt cr0, LBB1_2 ;bb

which is just as good as crnand.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31235 91177308-0d34-0410-b5e6-96231b3b80d8

Document getAlignment()

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31234 91177308-0d34-0410-b5e6-96231b3b80d8

remove debug code

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31233 91177308-0d34-0410-b5e6-96231b3b80d8

Codegen cond&cond with two branches.  This compiles (f.e.) PowerPC/and-branch.ll to:

        cmpwi cr0, r4, 4
        bgt cr0, LBB1_2 ;UnifiedReturnBlock
LBB1_3: ;entry
        cmplwi cr0, r3, 0
        bne cr0, LBB1_2 ;UnifiedReturnBlock

instead of:

        cmpwi cr7, r4, 4
        mfcr r2
        addic r4, r3, -1
        subfe r3, r4, r3
        rlwinm r2, r2, 30, 31, 31
        or r2, r2, r3
        cmplwi cr0, r2, 0
        bne cr0, LBB1_2 ;UnifiedReturnBlock
LBB1_1: ;cond_true

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31232 91177308-0d34-0410-b5e6-96231b3b80d8

new testcase

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31231 91177308-0d34-0410-b5e6-96231b3b80d8

Turn conditions like x<Y|z==q  into multiple blocks.

This compiles Regression/CodeGen/X86/or-branch.ll into:

_foo:
        subl $12, %esp
        call L_bar$stub
        movl 20(%esp), %eax
        movl 16(%esp), %ecx
        cmpl $5, %eax
        jl LBB1_1       #cond_true
LBB1_3: #entry
        testl %ecx, %ecx
        jne LBB1_2      #UnifiedReturnBlock
LBB1_1: #cond_true
        call L_bar$stub
        addl $12, %esp
        ret
LBB1_2: #UnifiedReturnBlock
        addl $12, %esp
        ret

instead of:

_foo:
        subl $12, %esp
        call L_bar$stub
        movl 20(%esp), %eax
        movl 16(%esp), %ecx
        cmpl $4, %eax
        setg %al
        testl %ecx, %ecx
        setne %cl
        testb %cl, %al
        jne LBB1_2      #UnifiedReturnBlock
LBB1_1: #cond_true
        call L_bar$stub
        addl $12, %esp
        ret
LBB1_2: #UnifiedReturnBlock
        addl $12, %esp
        ret

And on ppc to:

        cmpwi cr0, r29, 5
        blt cr0, LBB1_1 ;cond_true
LBB1_3: ;entry
        cmplwi cr0, r30, 0
        bne cr0, LBB1_2 ;UnifiedReturnBlock

instead of:

        cmpwi cr7, r4, 4
        mfcr r2
        addic r4, r3, -1
        subfe r30, r4, r3
        rlwinm r29, r2, 30, 31, 31
        and r2, r29, r30
        cmplwi cr0, r2, 0
        bne cr0, LBB1_2 ;UnifiedReturnBlock

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31230 91177308-0d34-0410-b5e6-96231b3b80d8

new testcase

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31229 91177308-0d34-0410-b5e6-96231b3b80d8

Fixed a significant bug where unpcklpd is incorrectly used to extract element 1 from a v2f64 value.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31228 91177308-0d34-0410-b5e6-96231b3b80d8

Add a new vextract test case.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31227 91177308-0d34-0410-b5e6-96231b3b80d8

Forgot a }

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31226 91177308-0d34-0410-b5e6-96231b3b80d8

Re-added the part where it tries to remove all global variables first.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31225 91177308-0d34-0410-b5e6-96231b3b80d8

SmallVector append not insert.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31224 91177308-0d34-0410-b5e6-96231b3b80d8

Grrr.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31223 91177308-0d34-0410-b5e6-96231b3b80d8

Temp patch for missing functionality.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31222 91177308-0d34-0410-b5e6-96231b3b80d8

Make the Value and Type methods print a newline so it prints nicely in gdb

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31221 91177308-0d34-0410-b5e6-96231b3b80d8

Fix for PR968: expand vector sdiv, udiv, srem, urem.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31220 91177308-0d34-0410-b5e6-96231b3b80d8

MathExtras isn't in the llvm/ADT directory but in the llvm/Support directory.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31219 91177308-0d34-0410-b5e6-96231b3b80d8

Apply editorials.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31218 91177308-0d34-0410-b5e6-96231b3b80d8

Improve cvs ignoring of test results.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31217 91177308-0d34-0410-b5e6-96231b3b80d8

Breakout folding hash set from SelectionDAGCSEMap.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31215 91177308-0d34-0410-b5e6-96231b3b80d8

Initialize CStringSection member var.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31214 91177308-0d34-0410-b5e6-96231b3b80d8

Add a run line to this test.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31211 91177308-0d34-0410-b5e6-96231b3b80d8

Change load PatFrag to ignore indexed load.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31210 91177308-0d34-0410-b5e6-96231b3b80d8

getPreIndexedLoad -> getIndexedLoad.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31209 91177308-0d34-0410-b5e6-96231b3b80d8

Indexed load / store changes.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31208 91177308-0d34-0410-b5e6-96231b3b80d8

Place cstrings in .cstring section.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31207 91177308-0d34-0410-b5e6-96231b3b80d8

Speed up isCString()

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31206 91177308-0d34-0410-b5e6-96231b3b80d8

Save temp. bc files when saveTemps flag is true. Use final output file
name supplied by linker to construct temp bc file names.

Remove tabs.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31205 91177308-0d34-0410-b5e6-96231b3b80d8

Simplify code a bit by changing instances of:
   InsertNewInstBefore(new CastInst(Val, ValTy, Val->GetName()), I)
into:
   InsertCastBefore(Val, ValTy, I)

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31204 91177308-0d34-0410-b5e6-96231b3b80d8

Put cstrings in .cstring section when compiling for Mac OS X.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31203 91177308-0d34-0410-b5e6-96231b3b80d8

Added CStringSection.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31202 91177308-0d34-0410-b5e6-96231b3b80d8

Add isCString() - returns true if a ConstantArray is a CString.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31201 91177308-0d34-0410-b5e6-96231b3b80d8

Fix Transforms/InstCombine/2006-10-26-VectorReassoc.ll

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31200 91177308-0d34-0410-b5e6-96231b3b80d8

new testcase, should not reassoc fp vector

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31199 91177308-0d34-0410-b5e6-96231b3b80d8

Add isFPOrFPVector() method, which indicates if a type is either FP or a
vector of FP types.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31198 91177308-0d34-0410-b5e6-96231b3b80d8

initial support for frame pointers

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31197 91177308-0d34-0410-b5e6-96231b3b80d8

Enclose a case in { and } so that the pickier compilers don't complain.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31196 91177308-0d34-0410-b5e6-96231b3b80d8

For PR950:
Make necessary changes to support DIV -> [SUF]Div. This changes llvm to
have three division instructions: signed, unsigned, floating point. The
bytecode and assembler are bacwards compatible, however.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31195 91177308-0d34-0410-b5e6-96231b3b80d8

Fix 2006-10-25-AddSetCC. A relational operator (like setlt) can never
produce an EQ property.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31193 91177308-0d34-0410-b5e6-96231b3b80d8

Handling of setlt/le/gt/ge is broken at the moment; predsimplify removes
the call the pass2.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31192 91177308-0d34-0410-b5e6-96231b3b80d8

Make the makefile tell us when Intrinsics.gen is being updated.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31191 91177308-0d34-0410-b5e6-96231b3b80d8

This testcase was logically wrong! Fixing, and including the test for the
mistake that I made in the file too.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31190 91177308-0d34-0410-b5e6-96231b3b80d8

Resurrect r1.25.
Fix and comment the "or", "and" and "xor" transformations.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31189 91177308-0d34-0410-b5e6-96231b3b80d8

simplify code

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31188 91177308-0d34-0410-b5e6-96231b3b80d8

New shuffle test case.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31187 91177308-0d34-0410-b5e6-96231b3b80d8

During vector shuffle lowering, we sometimes commute a vector shuffle to try
to match MOVL (movss, movsd, etc.). Don't forget to commute it back and try
unpck* and shufp* if that doesn't pan out.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31186 91177308-0d34-0410-b5e6-96231b3b80d8

X86ISD::PEXTRW 3rd operand type is always target pointer type.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31185 91177308-0d34-0410-b5e6-96231b3b80d8

hide symbols properly

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31184 91177308-0d34-0410-b5e6-96231b3b80d8

Remove -disable-x86-shuffle-opti

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31183 91177308-0d34-0410-b5e6-96231b3b80d8

Fix for PR960. Improves bugpoint so that it removes global variable
initializers as well. This is only a first pass. It can be slow because
it clones the module for each pass. An obvious improvement is not to do that.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31182 91177308-0d34-0410-b5e6-96231b3b80d8

Supply alignment info to linker through LLVMSymbol.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31181 91177308-0d34-0410-b5e6-96231b3b80d8

turn off tail merging for now

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31180 91177308-0d34-0410-b5e6-96231b3b80d8

be more aggressive about matching identical instructions.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31179 91177308-0d34-0410-b5e6-96231b3b80d8

Allow this to pass on ppc

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31178 91177308-0d34-0410-b5e6-96231b3b80d8

Modernize the testcase, allow it to pass on ppc

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31177 91177308-0d34-0410-b5e6-96231b3b80d8

TargetData is not subclassed. So no need to have virtual method.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31173 91177308-0d34-0410-b5e6-96231b3b80d8

Targets should depend on all the intrinsics.td files also, otherwise they
will compute a locally wrong numbering for the intrinsics.  This fixes a
nasty issue where the x86 backend started miscompiling stuff in a 'cvs up'd
build after the altivec intrinsics were added.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31172 91177308-0d34-0410-b5e6-96231b3b80d8

Move getPreferredAlignmentLog from AsmPrinter to TargetData

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31171 91177308-0d34-0410-b5e6-96231b3b80d8

expand ISD::VACOPY

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31170 91177308-0d34-0410-b5e6-96231b3b80d8

Removed extraneous semi-colon; this was prevening the grammar file from
bison'ing correctly.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31169 91177308-0d34-0410-b5e6-96231b3b80d8

Instead of hard coding global prefix, use TargetAsmInfo.
Add LTO destructor.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31168 91177308-0d34-0410-b5e6-96231b3b80d8

visitSwitchCase knows how to insert conditional branches well.  Change
visitBr to just call visitSwitchCase, eliminating duplicate logic.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31167 91177308-0d34-0410-b5e6-96231b3b80d8

Generalize CaseBlock a bit more:

Rename LHSBB/RHSBB to TrueBB/FalseBB.  Allow the RHS value to be null,
in which case the LHS is treated as a bool.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31166 91177308-0d34-0410-b5e6-96231b3b80d8

update comment

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31165 91177308-0d34-0410-b5e6-96231b3b80d8

Fix CodeGen/IA64/ret-0.ll, which has apparently been broken since some of the
isel changes happened months ago.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31164 91177308-0d34-0410-b5e6-96231b3b80d8