Fix large count and negative constant count handling in PPCCTRLoops

This commit fixes an assert that would occur on loops with large constant counts
(like looping for ((uint32_t) -1) iterations on PPC64). The existing code did
not handle counts that it computed to be negative (asserting instead), but
these can be created with valid inputs.

This bug was discovered by bugpoint while I was attempting to isolate a
completely different problem.

Also, in writing test cases for the negative-count problem, I discovered that
the ori/lsi handling was broken (there was a typo which caused the logic that
was supposed to detect these pairs and extract the iteration count to always
fail). This has now also been corrected (and is covered by one of the new test
cases).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@177295 91177308-0d34-0410-b5e6-96231b3b80d8

2 files changed, 104 insertions, 11 deletions

diff --git a/lib/Target/PowerPC/PPCCTRLoops.cpp b/lib/Target/PowerPC/PPCCTRLoops.cpp
index 51658011a9..14e07f8947 100644
--- a/lib/Target/PowerPC/PPCCTRLoops.cpp
+++ b/lib/Target/PowerPC/PPCCTRLoops.cpp
@@ -189,12 +189,23 @@ INITIALIZE_PASS_END(PPCCTRLoops, "ppc-ctr-loops", "PowerPC CTR Loops",
 
 /// isCompareEquals - Returns true if the instruction is a compare equals
 /// instruction with an immediate operand.
-static bool isCompareEqualsImm(const MachineInstr *MI, bool &SignedCmp) {
-  if (MI->getOpcode() == PPC::CMPWI || MI->getOpcode() == PPC::CMPDI) {
+static bool isCompareEqualsImm(const MachineInstr *MI, bool &SignedCmp,
+                               bool &Int64Cmp) {
+  if (MI->getOpcode() == PPC::CMPWI) {
     SignedCmp = true;
+    Int64Cmp = false;
     return true;
-  } else if (MI->getOpcode() == PPC::CMPLWI || MI->getOpcode() == PPC::CMPLDI) {
+  } else if (MI->getOpcode() == PPC::CMPDI) {
+    SignedCmp = true;
+    Int64Cmp = true;
+    return true;
+  } else if (MI->getOpcode() == PPC::CMPLWI) {
+    SignedCmp = false;
+    Int64Cmp = false;
+    return true;
+  } else if (MI->getOpcode() == PPC::CMPLDI) {
     SignedCmp = false;
+    Int64Cmp = true;
     return true;
   }
 
@@ -353,9 +364,9 @@ CountValue *PPCCTRLoops::getTripCount(MachineLoop *L,
          RI = MRI->reg_begin(IV_Opnd->getReg()), RE = MRI->reg_end();
          RI != RE; ++RI) {
       IV_Opnd = &RI.getOperand();
-      bool SignedCmp;
+      bool SignedCmp, Int64Cmp;
       MachineInstr *MI = IV_Opnd->getParent();
-      if (L->contains(MI) && isCompareEqualsImm(MI, SignedCmp) &&
+      if (L->contains(MI) && isCompareEqualsImm(MI, SignedCmp, Int64Cmp) &&
           MI->getOperand(0).getReg() == PredReg) {
 
         OldInsts.push_back(MI);
@@ -387,7 +398,7 @@ CountValue *PPCCTRLoops::getTripCount(MachineLoop *L,
                          DefInstr->getOpcode() == PPC::ORI)) {
           int64_t start = (short) DefInstr->getOperand(2).getImm();
           MachineInstr *DefInstr2 =
-            MRI->getVRegDef(DefInstr->getOperand(0).getReg());
+            MRI->getVRegDef(DefInstr->getOperand(1).getReg());
           if (DefInstr2 && (DefInstr2->getOpcode() == PPC::LIS8 ||
                             DefInstr2->getOpcode() == PPC::LIS)) {
             DEBUG(dbgs() << "  initial constant: " << *DefInstr);
@@ -403,7 +414,12 @@ CountValue *PPCCTRLoops::getTripCount(MachineLoop *L,
             OldInsts.push_back(DefInstr);
             OldInsts.push_back(DefInstr2);
 
-            return new CountValue(count/iv_value);
+            // count/iv_value, the trip count, should be positive here. If it
+            // is negative, that indicates that the counter will wrap.
+            if (Int64Cmp)
+              return new CountValue(count/iv_value);
+            else
+              return new CountValue(uint32_t(count/iv_value));
           }
         } else if (DefInstr && (DefInstr->getOpcode() == PPC::LI8 ||
                                 DefInstr->getOpcode() == PPC::LI)) {
@@ -416,7 +432,10 @@ CountValue *PPCCTRLoops::getTripCount(MachineLoop *L,
 
           OldInsts.push_back(DefInstr);
 
-          return new CountValue(count/iv_value);
+          if (Int64Cmp)
+            return new CountValue(count/iv_value);
+          else
+            return new CountValue(uint32_t(count/iv_value));
         } else if (iv_value == 1 || iv_value == -1) {
           // We can't determine a constant starting value.
           if (ImmVal == 0) {
@@ -601,6 +620,16 @@ bool PPCCTRLoops::convertToCTRLoop(MachineLoop *L) {
     DEBUG(dbgs() << "failed to get trip count!\n");
     return false;
   }
+
+  if (TripCount->isImm()) {
+    DEBUG(dbgs() << "constant trip count: " << TripCount->getImm() << "\n");
+
+    // FIXME: We currently can't form 64-bit constants
+    // (including 32-bit unsigned constants)
+    if (!isInt<32>(TripCount->getImm()))
+      return false;
+  }
+
   // Does the loop contain any invalid instructions?
   if (containsInvalidInstruction(L)) {
     return false;
@@ -671,13 +700,14 @@ bool PPCCTRLoops::convertToCTRLoop(MachineLoop *L) {
     // Put the trip count in a register for transfer into the count register.
 
     int64_t CountImm = TripCount->getImm();
-    assert(!TripCount->isNeg() && "Constant trip count must be positive");
+    if (TripCount->isNeg())
+      CountImm = -CountImm;
 
     CountReg = MF->getRegInfo().createVirtualRegister(RC);
-    if (CountImm > 0xFFFF) {
+    if (abs64(CountImm) > 0x7FFF) {
       BuildMI(*Preheader, InsertPos, dl,
               TII->get(isPPC64 ? PPC::LIS8 : PPC::LIS),
-              CountReg).addImm(CountImm >> 16);
+              CountReg).addImm((CountImm >> 16) & 0xFFFF);
       unsigned CountReg1 = CountReg;
       CountReg = MF->getRegInfo().createVirtualRegister(RC);
       BuildMI(*Preheader, InsertPos, dl,
diff --git a/test/CodeGen/PowerPC/negctr.ll b/test/CodeGen/PowerPC/negctr.ll
new file mode 100644
index 0000000000..9b438a138f
--- /dev/null
+++ b/test/CodeGen/PowerPC/negctr.ll
@@ -0,0 +1,63 @@
+; RUN: llc < %s -mcpu=a2 | FileCheck %s
+target datalayout = "E-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-f128:128:128-v128:128:128-n32:64"
+target triple = "powerpc64-unknown-linux-gnu"
+
+define void @main() #0 {
+entry:
+  br i1 undef, label %for.end, label %for.body
+
+for.body:                                         ; preds = %for.body, %entry
+  %indvars.iv = phi i64 [ %indvars.iv.next, %for.body ], [ 1, %entry ]
+  %indvars.iv.next = add i64 %indvars.iv, 1
+  %lftr.wideiv = trunc i64 %indvars.iv.next to i32
+  %exitcond = icmp eq i32 %lftr.wideiv, 0
+  br i1 %exitcond, label %for.end, label %for.body
+
+; FIXME: We currently can't form the 32-bit unsigned trip count necessary here!
+; CHECK: @main
+; CHECK-NOT: bdnz
+
+for.end:                                          ; preds = %for.body, %entry
+  ret void
+}
+
+define void @main1() #0 {
+entry:
+  br i1 undef, label %for.end, label %for.body
+
+for.body:                                         ; preds = %for.body, %entry
+  %indvars.iv = phi i64 [ %indvars.iv.next, %for.body ], [ 1, %entry ]
+  %indvars.iv.next = add i64 %indvars.iv, 1
+  %exitcond = icmp eq i64 %indvars.iv.next, 0
+  br i1 %exitcond, label %for.end, label %for.body
+
+; CHECK: @main1
+; CHECK: li [[REG:[0-9]+]], -1
+; CHECK: mtctr [[REG]]
+; CHECK: bdnz
+
+for.end:                                          ; preds = %for.body, %entry
+  ret void
+}
+
+define void @main2() #0 {
+entry:
+  br i1 undef, label %for.end, label %for.body
+
+for.body:                                         ; preds = %for.body, %entry
+  %indvars.iv = phi i64 [ %indvars.iv.next, %for.body ], [ 1, %entry ]
+  %indvars.iv.next = add i64 %indvars.iv, 1
+  %exitcond = icmp eq i64 %indvars.iv.next, -100000
+  br i1 %exitcond, label %for.end, label %for.body
+
+; CHECK: @main2
+; CHECK: lis [[REG:[0-9]+]], -2
+; CHECK: ori [[REG2:[0-9]+]], [[REG]], 31071
+; CHECK: mtctr [[REG2]]
+; CHECK: bdnz
+
+for.end:                                          ; preds = %for.body, %entry
+  ret void
+}
+
+attributes #0 = { nounwind "less-precise-fpmad"="false" "no-frame-pointer-elim"="false" "no-frame-pointer-elim-non-leaf"="true" "no-infs-fp-math"="false" "no-nans-fp-math"="false" "unsafe-fp-math"="false" "use-soft-float"="false" }