Change the scheduler from adding nodes in allnodes order

to adding them in a determinstic order (bottom up from 
the root) based on the structure of the graph itself.

This updates tests for some random changes, interesting
bits: CodeGen/Blackfin/promote-logic.ll no longer crashes.
I have no idea why, but that's good right?

CodeGen/X86/2009-07-16-LoadFoldingBug.ll also fails, but
now compiles to have one fewer constant pool entry, making
the expected load that was being folded disappear.  Since it
is an unreduced mass of gnast, I just removed it.

This fixes PR6370


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@97023 91177308-0d34-0410-b5e6-96231b3b80d8

5 files changed, 18 insertions, 109 deletions

diff --git a/lib/CodeGen/SelectionDAG/ScheduleDAGSDNodes.cpp b/lib/CodeGen/SelectionDAG/ScheduleDAGSDNodes.cpp
index b51c61bf6d..06e7b8c905 100644
--- a/lib/CodeGen/SelectionDAG/ScheduleDAGSDNodes.cpp
+++ b/lib/CodeGen/SelectionDAG/ScheduleDAGSDNodes.cpp
@@ -218,8 +218,20 @@ void ScheduleDAGSDNodes::BuildSchedUnits() {
   // Check to see if the scheduler cares about latencies.
   bool UnitLatencies = ForceUnitLatencies();
 
-  for (SelectionDAG::allnodes_iterator NI = DAG->allnodes_begin(),
-       E = DAG->allnodes_end(); NI != E; ++NI) {
+  // Add all nodes in depth first order.
+  SmallVector<SDNode*, 64> Worklist;
+  SmallPtrSet<SDNode*, 64> Visited;
+  Worklist.push_back(DAG->getRoot().getNode());
+  Visited.insert(DAG->getRoot().getNode());
+  
+  while (!Worklist.empty()) {
+    SDNode *NI = Worklist.pop_back_val();
+    
+    // Add all operands to the worklist unless they've already been added.
+    for (unsigned i = 0, e = NI->getNumOperands(); i != e; ++i)
+      if (Visited.insert(NI->getOperand(i).getNode()))
+        Worklist.push_back(NI->getOperand(i).getNode());
+  
     if (isPassiveNode(NI))  // Leaf node, e.g. a TargetImmediate.
       continue;
     
diff --git a/test/CodeGen/Blackfin/promote-logic.ll b/test/CodeGen/Blackfin/promote-logic.ll
index 46da56681d..1ac1408290 100644
--- a/test/CodeGen/Blackfin/promote-logic.ll
+++ b/test/CodeGen/Blackfin/promote-logic.ll
@@ -1,5 +1,4 @@
-; RUN: llc < %s -march=bfin > %t
-; XFAIL: *
+; RUN: llc < %s -march=bfin 
 
 ; DAGCombiner::SimplifyBinOpWithSameOpcodeHands can produce an illegal i16 OR
 ; operation after LegalizeOps.
diff --git a/test/CodeGen/MSP430/Inst8rr.ll b/test/CodeGen/MSP430/Inst8rr.ll
index 74feaae4eb..0f5fc12b62 100644
--- a/test/CodeGen/MSP430/Inst8rr.ll
+++ b/test/CodeGen/MSP430/Inst8rr.ll
@@ -10,7 +10,7 @@ define i8 @mov(i8 %a, i8 %b) nounwind {
 
 define i8 @add(i8 %a, i8 %b) nounwind {
 ; CHECK: add:
-; CHECK: add.b	r14, r15
+; CHECK: add.b	r12, r15
 	%1 = add i8 %a, %b
 	ret i8 %1
 }
diff --git a/test/CodeGen/PowerPC/LargeAbsoluteAddr.ll b/test/CodeGen/PowerPC/LargeAbsoluteAddr.ll
index 75374abeb0..b10a996867 100644
--- a/test/CodeGen/PowerPC/LargeAbsoluteAddr.ll
+++ b/test/CodeGen/PowerPC/LargeAbsoluteAddr.ll
@@ -1,7 +1,7 @@
 ; RUN: llc < %s -march=ppc32 -mtriple=powerpc-apple-darwin | \
-; RUN:   grep {stw r4, 32751}
+; RUN:   grep {stw r3, 32751}
 ; RUN: llc < %s -march=ppc64 -mtriple=powerpc-apple-darwin | \
-; RUN:   grep {stw r4, 32751}
+; RUN:   grep {stw r3, 32751}
 ; RUN: llc < %s -march=ppc64 -mtriple=powerpc-apple-darwin | \
 ; RUN:   grep {std r3, 9024}
 
diff --git a/test/CodeGen/X86/2009-07-16-LoadFoldingBug.ll b/test/CodeGen/X86/2009-07-16-LoadFoldingBug.ll
deleted file mode 100644
index e21c8923df..0000000000
--- a/test/CodeGen/X86/2009-07-16-LoadFoldingBug.ll
+++ /dev/null
@@ -1,102 +0,0 @@
-; RUN: llc < %s -mtriple=x86_64-apple-darwin10 | FileCheck %s
-
-; CHECK: _foo:
-; CHECK: pavgw LCPI1_4(%rip)
-
-; rdar://7057804
-
-define void @foo(i16* %out8x8, i16* %in8x8, i32 %lastrow) optsize ssp {
-entry:
-	%0 = call <8 x i16> @llvm.x86.sse2.pmulh.w(<8 x i16> <i16 6518, i16 6518, i16 6518, i16 6518, i16 6518, i16 6518, i16 6518, i16 6518>, <8 x i16> undef) nounwind readnone		; <<8 x i16>> [#uses=2]
-	%1 = call <8 x i16> @llvm.x86.sse2.pcmpeq.w(<8 x i16> %0, <8 x i16> <i16 16384, i16 16384, i16 16384, i16 16384, i16 16384, i16 16384, i16 16384, i16 16384>) nounwind readnone		; <<8 x i16>> [#uses=1]
-	%2 = call <8 x i16> @llvm.x86.sse2.psrli.w(<8 x i16> zeroinitializer, i32 14) nounwind readnone		; <<8 x i16>> [#uses=1]
-	%3 = call <8 x i16> @llvm.x86.sse2.pavg.w(<8 x i16> %2, <8 x i16> zeroinitializer) nounwind readnone		; <<8 x i16>> [#uses=1]
-	%tmp.i.i10 = add <8 x i16> %0, %3		; <<8 x i16>> [#uses=1]
-	%4 = call <8 x i16> @llvm.x86.sse2.psubs.w(<8 x i16> zeroinitializer, <8 x i16> %1) nounwind readnone		; <<8 x i16>> [#uses=1]
-	%5 = call <8 x i16> @llvm.x86.sse2.padds.w(<8 x i16> %tmp.i.i10, <8 x i16> %4) nounwind readnone		; <<8 x i16>> [#uses=3]
-	%6 = call <8 x i16> @llvm.x86.sse2.psubs.w(<8 x i16> %5, <8 x i16> undef) nounwind readnone		; <<8 x i16>> [#uses=1]
-	%7 = call <8 x i16> @llvm.x86.sse2.pmulh.w(<8 x i16> <i16 6518, i16 6518, i16 6518, i16 6518, i16 6518, i16 6518, i16 6518, i16 6518>, <8 x i16> undef) nounwind readnone		; <<8 x i16>> [#uses=2]
-	%8 = call <8 x i16> @llvm.x86.sse2.pcmpeq.w(<8 x i16> %7, <8 x i16> <i16 16384, i16 16384, i16 16384, i16 16384, i16 16384, i16 16384, i16 16384, i16 16384>) nounwind readnone		; <<8 x i16>> [#uses=1]
-	%9 = call <8 x i16> @llvm.x86.sse2.psrli.w(<8 x i16> zeroinitializer, i32 14) nounwind readnone		; <<8 x i16>> [#uses=1]
-	%10 = call <8 x i16> @llvm.x86.sse2.pavg.w(<8 x i16> %9, <8 x i16> zeroinitializer) nounwind readnone		; <<8 x i16>> [#uses=1]
-	%tmp.i.i8 = add <8 x i16> %7, %10		; <<8 x i16>> [#uses=1]
-	%11 = call <8 x i16> @llvm.x86.sse2.psubs.w(<8 x i16> undef, <8 x i16> %8) nounwind readnone		; <<8 x i16>> [#uses=1]
-	%12 = call <8 x i16> @llvm.x86.sse2.padds.w(<8 x i16> %tmp.i.i8, <8 x i16> %11) nounwind readnone		; <<8 x i16>> [#uses=1]
-	%13 = call <8 x i16> @llvm.x86.sse2.padds.w(<8 x i16> undef, <8 x i16> undef) nounwind readnone		; <<8 x i16>> [#uses=1]
-	%14 = call <8 x i16> @llvm.x86.sse2.psubs.w(<8 x i16> %5, <8 x i16> undef) nounwind readnone		; <<8 x i16>> [#uses=1]
-	%15 = call <8 x i16> @llvm.x86.sse2.padds.w(<8 x i16> %5, <8 x i16> undef) nounwind readnone		; <<8 x i16>> [#uses=1]
-	%16 = call <8 x i16> @llvm.x86.sse2.padds.w(<8 x i16> %6, <8 x i16> undef) nounwind readnone		; <<8 x i16>> [#uses=1]
-	%17 = call <8 x i16> @llvm.x86.sse2.padds.w(<8 x i16> %12, <8 x i16> undef) nounwind readnone		; <<8 x i16>> [#uses=1]
-	%18 = call <8 x i16> @llvm.x86.sse2.psubs.w(<8 x i16> %13, <8 x i16> %15) nounwind readnone		; <<8 x i16>> [#uses=1]
-	%19 = call <8 x i16> @llvm.x86.sse2.psubs.w(<8 x i16> undef, <8 x i16> %14) nounwind readnone		; <<8 x i16>> [#uses=2]
-	%20 = call <8 x i16> @llvm.x86.sse2.psubs.w(<8 x i16> undef, <8 x i16> undef) nounwind readnone		; <<8 x i16>> [#uses=4]
-	%21 = call <8 x i16> @llvm.x86.sse2.psubs.w(<8 x i16> undef, <8 x i16> %17) nounwind readnone		; <<8 x i16>> [#uses=1]
-	%22 = bitcast <8 x i16> %21 to <2 x i64>		; <<2 x i64>> [#uses=1]
-	%23 = call <8 x i16> @llvm.x86.sse2.pmulh.w(<8 x i16> <i16 23170, i16 23170, i16 23170, i16 23170, i16 23170, i16 23170, i16 23170, i16 23170>, <8 x i16> undef) nounwind readnone		; <<8 x i16>> [#uses=2]
-	%24 = call <8 x i16> @llvm.x86.sse2.pcmpeq.w(<8 x i16> %23, <8 x i16> <i16 16384, i16 16384, i16 16384, i16 16384, i16 16384, i16 16384, i16 16384, i16 16384>) nounwind readnone		; <<8 x i16>> [#uses=1]
-	%25 = call <8 x i16> @llvm.x86.sse2.psrli.w(<8 x i16> zeroinitializer, i32 14) nounwind readnone		; <<8 x i16>> [#uses=1]
-	%26 = call <8 x i16> @llvm.x86.sse2.pavg.w(<8 x i16> %25, <8 x i16> zeroinitializer) nounwind readnone		; <<8 x i16>> [#uses=1]
-	%tmp.i.i6 = add <8 x i16> %23, %26		; <<8 x i16>> [#uses=1]
-	%27 = call <8 x i16> @llvm.x86.sse2.psubs.w(<8 x i16> undef, <8 x i16> %24) nounwind readnone		; <<8 x i16>> [#uses=1]
-	%28 = call <8 x i16> @llvm.x86.sse2.padds.w(<8 x i16> %tmp.i.i6, <8 x i16> %27) nounwind readnone		; <<8 x i16>> [#uses=1]
-	%29 = call <8 x i16> @llvm.x86.sse2.pmulh.w(<8 x i16> <i16 -23170, i16 -23170, i16 -23170, i16 -23170, i16 -23170, i16 -23170, i16 -23170, i16 -23170>, <8 x i16> undef) nounwind readnone		; <<8 x i16>> [#uses=2]
-	%30 = call <8 x i16> @llvm.x86.sse2.pcmpeq.w(<8 x i16> %29, <8 x i16> <i16 16384, i16 16384, i16 16384, i16 16384, i16 16384, i16 16384, i16 16384, i16 16384>) nounwind readnone		; <<8 x i16>> [#uses=1]
-	%31 = call <8 x i16> @llvm.x86.sse2.psrli.w(<8 x i16> zeroinitializer, i32 14) nounwind readnone		; <<8 x i16>> [#uses=1]
-	%32 = call <8 x i16> @llvm.x86.sse2.pavg.w(<8 x i16> %31, <8 x i16> zeroinitializer) nounwind readnone		; <<8 x i16>> [#uses=1]
-	%tmp.i.i4 = add <8 x i16> %29, %32		; <<8 x i16>> [#uses=1]
-	%33 = call <8 x i16> @llvm.x86.sse2.psubs.w(<8 x i16> undef, <8 x i16> %30) nounwind readnone		; <<8 x i16>> [#uses=1]
-	%34 = call <8 x i16> @llvm.x86.sse2.padds.w(<8 x i16> %tmp.i.i4, <8 x i16> %33) nounwind readnone		; <<8 x i16>> [#uses=1]
-	%35 = call <8 x i16> @llvm.x86.sse2.pmulh.w(<8 x i16> <i16 23170, i16 23170, i16 23170, i16 23170, i16 23170, i16 23170, i16 23170, i16 23170>, <8 x i16> %20) nounwind readnone		; <<8 x i16>> [#uses=2]
-	%tmp.i2.i1 = mul <8 x i16> %20, <i16 23170, i16 23170, i16 23170, i16 23170, i16 23170, i16 23170, i16 23170, i16 23170>		; <<8 x i16>> [#uses=1]
-	%36 = call <8 x i16> @llvm.x86.sse2.pcmpeq.w(<8 x i16> %35, <8 x i16> <i16 16384, i16 16384, i16 16384, i16 16384, i16 16384, i16 16384, i16 16384, i16 16384>) nounwind readnone		; <<8 x i16>> [#uses=1]
-	%37 = call <8 x i16> @llvm.x86.sse2.psrli.w(<8 x i16> %tmp.i2.i1, i32 14) nounwind readnone		; <<8 x i16>> [#uses=1]
-	%38 = call <8 x i16> @llvm.x86.sse2.pavg.w(<8 x i16> %37, <8 x i16> zeroinitializer) nounwind readnone		; <<8 x i16>> [#uses=1]
-	%tmp.i.i2 = add <8 x i16> %35, %38		; <<8 x i16>> [#uses=1]
-	%39 = call <8 x i16> @llvm.x86.sse2.psubs.w(<8 x i16> %19, <8 x i16> %36) nounwind readnone		; <<8 x i16>> [#uses=1]
-	%40 = call <8 x i16> @llvm.x86.sse2.padds.w(<8 x i16> %tmp.i.i2, <8 x i16> %39) nounwind readnone		; <<8 x i16>> [#uses=1]
-	%41 = call <8 x i16> @llvm.x86.sse2.pmulh.w(<8 x i16> <i16 -23170, i16 -23170, i16 -23170, i16 -23170, i16 -23170, i16 -23170, i16 -23170, i16 -23170>, <8 x i16> %20) nounwind readnone		; <<8 x i16>> [#uses=2]
-	%tmp.i2.i = mul <8 x i16> %20, <i16 -23170, i16 -23170, i16 -23170, i16 -23170, i16 -23170, i16 -23170, i16 -23170, i16 -23170>		; <<8 x i16>> [#uses=1]
-	%42 = call <8 x i16> @llvm.x86.sse2.pcmpeq.w(<8 x i16> %41, <8 x i16> <i16 16384, i16 16384, i16 16384, i16 16384, i16 16384, i16 16384, i16 16384, i16 16384>) nounwind readnone		; <<8 x i16>> [#uses=1]
-	%43 = call <8 x i16> @llvm.x86.sse2.psrli.w(<8 x i16> %tmp.i2.i, i32 14) nounwind readnone		; <<8 x i16>> [#uses=1]
-	%44 = call <8 x i16> @llvm.x86.sse2.pavg.w(<8 x i16> %43, <8 x i16> zeroinitializer) nounwind readnone		; <<8 x i16>> [#uses=1]
-	%tmp.i.i = add <8 x i16> %41, %44		; <<8 x i16>> [#uses=1]
-	%45 = call <8 x i16> @llvm.x86.sse2.psubs.w(<8 x i16> %19, <8 x i16> %42) nounwind readnone		; <<8 x i16>> [#uses=1]
-	%46 = call <8 x i16> @llvm.x86.sse2.padds.w(<8 x i16> %tmp.i.i, <8 x i16> %45) nounwind readnone		; <<8 x i16>> [#uses=1]
-	%47 = call <8 x i16> @llvm.x86.sse2.padds.w(<8 x i16> %18, <8 x i16> %16) nounwind readnone		; <<8 x i16>> [#uses=1]
-	%48 = bitcast <8 x i16> %47 to <2 x i64>		; <<2 x i64>> [#uses=1]
-	%49 = bitcast <8 x i16> %28 to <2 x i64>		; <<2 x i64>> [#uses=1]
-	%50 = getelementptr i16* %out8x8, i64 8		; <i16*> [#uses=1]
-	%51 = bitcast i16* %50 to <2 x i64>*		; <<2 x i64>*> [#uses=1]
-	store <2 x i64> %49, <2 x i64>* %51, align 16
-	%52 = bitcast <8 x i16> %40 to <2 x i64>		; <<2 x i64>> [#uses=1]
-	%53 = getelementptr i16* %out8x8, i64 16		; <i16*> [#uses=1]
-	%54 = bitcast i16* %53 to <2 x i64>*		; <<2 x i64>*> [#uses=1]
-	store <2 x i64> %52, <2 x i64>* %54, align 16
-	%55 = getelementptr i16* %out8x8, i64 24		; <i16*> [#uses=1]
-	%56 = bitcast i16* %55 to <2 x i64>*		; <<2 x i64>*> [#uses=1]
-	store <2 x i64> %48, <2 x i64>* %56, align 16
-	%57 = bitcast <8 x i16> %46 to <2 x i64>		; <<2 x i64>> [#uses=1]
-	%58 = getelementptr i16* %out8x8, i64 40		; <i16*> [#uses=1]
-	%59 = bitcast i16* %58 to <2 x i64>*		; <<2 x i64>*> [#uses=1]
-	store <2 x i64> %57, <2 x i64>* %59, align 16
-	%60 = bitcast <8 x i16> %34 to <2 x i64>		; <<2 x i64>> [#uses=1]
-	%61 = getelementptr i16* %out8x8, i64 48		; <i16*> [#uses=1]
-	%62 = bitcast i16* %61 to <2 x i64>*		; <<2 x i64>*> [#uses=1]
-	store <2 x i64> %60, <2 x i64>* %62, align 16
-	%63 = getelementptr i16* %out8x8, i64 56		; <i16*> [#uses=1]
-	%64 = bitcast i16* %63 to <2 x i64>*		; <<2 x i64>*> [#uses=1]
-	store <2 x i64> %22, <2 x i64>* %64, align 16
-	ret void
-}
-
-declare <8 x i16> @llvm.x86.sse2.psubs.w(<8 x i16>, <8 x i16>) nounwind readnone
-
-declare <8 x i16> @llvm.x86.sse2.pmulh.w(<8 x i16>, <8 x i16>) nounwind readnone
-
-declare <8 x i16> @llvm.x86.sse2.pcmpeq.w(<8 x i16>, <8 x i16>) nounwind readnone
-
-declare <8 x i16> @llvm.x86.sse2.pavg.w(<8 x i16>, <8 x i16>) nounwind readnone
-
-declare <8 x i16> @llvm.x86.sse2.padds.w(<8 x i16>, <8 x i16>) nounwind readnone
-
-declare <8 x i16> @llvm.x86.sse2.psrli.w(<8 x i16>, i32) nounwind readnone