[DA] Fix direction vectors for weakZeroSrcSIV

[dmgreen]

Summary

Both weakZeroSrcSIV and weakZeroDstSIV are currently giving the same
direction vectors. One of them must be wrong, and it looks like Src
is getting things the wrong way round.

[dmgreen]

Ping :)

[dmgreen]

friendly ping

[bcahoon]

This change looks correct to me. I agree with you that those functions shouldn't return the same direction vectors and that the Src version is wrong.

[dmgreen]

Thanks! I have a couple more DA patches around, but they may not be as straight forward as this one.

lib/Analysis/DependenceAnalysis.cpp

 //    c1 = c2 + a*i
 //
 // we get
 //
 //    (c1 - c2)/a = i
 //
 // If i is not an integer, there's no dependence.
 // If i < 0 or > UB, there's no dependence.
-// If i = 0, the direction is <= and peeling the
+// If i = 0, the direction is >= and peeling the
 // 1st iteration will break the dependence.
-// If i = UB, the direction is >= and peeling the
+// If i = UB, the direction is <= and peeling the
 // last iteration will break the dependence.
 // Otherwise, the direction is *.
 //
 // Can prove independence. Failing that, we can sometimes refine
 // the directions. Can sometimes show that first or last
 // iteration carries all the dependences (so worth peeling).
 //
 // (see also weakZeroDstSIVtest)
bool DependenceInfo::weakZeroSrcSIVtest(const SCEV *DstCoeff,
   Level--;
   Result.Consistent = false;
   const SCEV *Delta = SE->getMinusSCEV(SrcConst, DstConst);
   NewConstraint.setLine(SE->getZero(Delta->getType()), DstCoeff, Delta,
                         CurLoop);
   DEBUG(dbgs() << "\t    Delta = " << *Delta << "\n");
   if (isKnownPredicate(CmpInst::ICMP_EQ, SrcConst, DstConst)) {
     if (Level < CommonLevels) {
-      Result.DV[Level].Direction &= Dependence::DVEntry::LE;
+      Result.DV[Level].Direction &= Dependence::DVEntry::GE;
       Result.DV[Level].PeelFirst = true;
       ++WeakZeroSIVsuccesses;
     }
     return false; // dependences caused by first iteration
   }
   const SCEVConstant *ConstCoeff = dyn_cast<SCEVConstant>(DstCoeff);
   if (!ConstCoeff)
     return false;
if (const SCEV *UpperBound = collectUpperBound(CurLoop, Delta->getType())) {
     if (isKnownPredicate(CmpInst::ICMP_SGT, NewDelta, Product)) {
       ++WeakZeroSIVindependence;
       ++WeakZeroSIVsuccesses;
       return true;
     }
     if (isKnownPredicate(CmpInst::ICMP_EQ, NewDelta, Product)) {
       // dependences caused by last iteration
       if (Level < CommonLevels) {
-        Result.DV[Level].Direction &= Dependence::DVEntry::GE;
+        Result.DV[Level].Direction &= Dependence::DVEntry::LE;
         Result.DV[Level].PeelLast = true;
         ++WeakZeroSIVsuccesses;
       }
       return false;
     }
   }
 
   // check that Delta/SrcCoeff >= 0

test/Analysis/DependenceAnalysis/Coupled.ll

for.body:                                         ; preds = %entry, %for.body
   store i32 %0, i32* %B.addr.01, align 4
   %inc = add nsw i64 %i.02, 1
   %exitcond = icmp ne i64 %inc, 100
   br i1 %exitcond, label %for.body, label %for.end
 
 for.end:                                          ; preds = %for.body
   ret void
 }
+
+;;  for(int i = 0; i < N; i+=1) {
+;;    A[M*N*i] = 1;
+;;    for(int j = 0; j < M; j+=1)
+;;      A[M*N + M*i + j] = 2;
+
+define void @couple_weakzerosiv(i32* noalias nocapture %A, i64 %N, i64 %M) {
+entry:
+  %cmp29 = icmp sgt i64 %N, 0
+  br i1 %cmp29, label %for.body.lr.ph, label %for.cond.cleanup
+
+; CHECK-LABEL: couple_weakzerosiv
+; CHECK: da analyze - none!
+; CHECK: da analyze - output [p>]!
+; CHECK: da analyze - none!
+
+for.body.lr.ph:                                   ; preds = %entry
+  %mul = mul nsw i64 %M, %N
+  br label %for.body.us
+
+for.body.us:                                      ; preds = %for.body.lr.ph, %for.cond.cleanup4.loopexit.us
+  %i.030.us = phi i64 [ %add12.us, %for.cond.cleanup4.loopexit.us ], [ 0, %for.body.lr.ph ]
+  %mul1.us = mul nsw i64 %i.030.us, %mul
+  %arrayidx.us = getelementptr inbounds i32, i32* %A, i64 %mul1.us
+  store i32 1, i32* %arrayidx.us, align 4
+  %mul6.us = mul nsw i64 %i.030.us, %M
+  %add.us = add i64 %mul6.us, %mul
+  br label %for.body5.us
+
+for.body5.us:                                     ; preds = %for.body5.us, %for.body.us
+  %j.028.us = phi i64 [ 0, %for.body.us ], [ %add10.us, %for.body5.us ]
+  %add8.us = add i64 %add.us, %j.028.us
+  %arrayidx9.us = getelementptr inbounds i32, i32* %A, i64 %add8.us
+  store i32 2, i32* %arrayidx9.us, align 4
+  %add10.us = add nuw nsw i64 %j.028.us, 1
+  %exitcond.us = icmp eq i64 %add10.us, %M
+  br i1 %exitcond.us, label %for.cond.cleanup4.loopexit.us, label %for.body5.us
+
+for.cond.cleanup4.loopexit.us:                    ; preds = %for.body5.us
+  %add12.us = add nuw nsw i64 %i.030.us, 1
+  %exitcond31.us = icmp eq i64 %add12.us, %N
+  br i1 %exitcond31.us, label %for.cond.cleanup, label %for.body.us
+
+for.cond.cleanup:                                 ; preds = %for.cond.cleanup4.loopexit.us, %entry
+  ret void
+}

test/Analysis/DependenceAnalysis/WeakZeroDstSIV.ll

 ; RUN: opt < %s -analyze -basicaa -da | FileCheck %s
 
 ; ModuleID = 'WeakZeroDstSIV.bc'
 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-v64:64:64-v128:128:128-a0:0:64-s0:64:64-f80:128:128-n8:16:32:64-S128"
 target triple = "x86_64-apple-macosx10.6.0"
 
 
+;;  for (int i = 0; i < N; i++) {
+;;    A[i] = 1;
+;;    A[0] = 2;
+
+define void @dstzero(i32* nocapture %A, i32 %N) {
+entry:
+  %cmp6 = icmp sgt i32 %N, 0
+  br i1 %cmp6, label %for.body, label %for.cond.cleanup
+
+; CHECK: da analyze - none!
+; CHECK: da analyze - output [p<=|<]!
+; CHECK: da analyze - consistent output [S]!
+
+for.body:                                         ; preds = %entry, %for.body
+  %i.07 = phi i32 [ %add, %for.body ], [ 0, %entry ]
+  %arrayidx = getelementptr inbounds i32, i32* %A, i32 %i.07
+  store i32 0, i32* %arrayidx, align 4
+  store i32 1, i32* %A, align 4
+  %add = add nuw nsw i32 %i.07, 1
+  %exitcond = icmp eq i32 %add, %N
+  br i1 %exitcond, label %for.cond.cleanup, label %for.body
+
+for.cond.cleanup:                                 ; preds = %for.body, %entry
+  ret void
+}
+
+
+
+
 ;;  for (long unsigned i = 0; i < 30; i++) {
 ;;    A[2*i + 10] = i;
 ;;    *B++ = A[10];
 
 define void @weakzerodst0(i32* %A, i32* %B, i64 %n) nounwind uwtable ssp {
 entry:
   br label %for.body
 

test/Analysis/DependenceAnalysis/WeakZeroSrcSIV.ll

 ; RUN: opt < %s -analyze -basicaa -da | FileCheck %s
 
 ; ModuleID = 'WeakZeroSrcSIV.bc'
 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-v64:64:64-v128:128:128-a0:0:64-s0:64:64-f80:128:128-n8:16:32:64-S128"
 target triple = "x86_64-apple-macosx10.6.0"
 
 
+;;  for (int i = 0; i < N; i++) {
+;;    A[0] = 1;
+;;    A[i] = 2;
+
+define void @dstzero(i32* nocapture %A, i32 %N) {
+entry:
+  %cmp6 = icmp sgt i32 %N, 0
+  br i1 %cmp6, label %for.body, label %for.cond.cleanup
+
+; CHECK: da analyze - consistent output [S]!
+; CHECK: da analyze - output [p=>|<]!
+; CHECK: da analyze - none!
+
+for.body:                                         ; preds = %entry, %for.body
+  %i.07 = phi i32 [ %add, %for.body ], [ 0, %entry ]
+  store i32 0, i32* %A, align 4
+  %arrayidx1 = getelementptr inbounds i32, i32* %A, i32 %i.07
+  store i32 1, i32* %arrayidx1, align 4
+  %add = add nuw nsw i32 %i.07, 1
+  %exitcond = icmp eq i32 %add, %N
+  br i1 %exitcond, label %for.cond.cleanup, label %for.body
+
+for.cond.cleanup:                                 ; preds = %for.body, %entry
+  ret void
+}
+
+
 ;;  for (long unsigned i = 0; i < 30; i++) {
 ;;    A[10] = i;
 ;;    *B++ = A[2*i + 10];
 
 define void @weakzerosrc0(i32* %A, i32* %B, i64 %n) nounwind uwtable ssp {
 entry:
   br label %for.body
 
 ; CHECK: da analyze - consistent output [S]!
-; CHECK: da analyze - flow [p<=|<]!
+; CHECK: da analyze - flow [p=>|<]!
 ; CHECK: da analyze - confused!
 ; CHECK: da analyze - none!
 ; CHECK: da analyze - confused!
 ; CHECK: da analyze - none!
 
 for.body:                                         ; preds = %entry, %for.body
   %i.02 = phi i64 [ 0, %entry ], [ %inc, %for.body ]
   %B.addr.01 = phi i32* [ %B, %entry ], [ %incdec.ptr, %for.body ]

 ;;    *B++ = A[n*i + 10];
 
 define void @weakzerosrc1(i32* %A, i32* %B, i64 %n) nounwind uwtable ssp {
 entry:
   %cmp1 = icmp eq i64 %n, 0
   br i1 %cmp1, label %for.end, label %for.body.preheader
 
 ; CHECK: da analyze - consistent output [S]!
-; CHECK: da analyze - flow [p<=|<]!
+; CHECK: da analyze - flow [p=>|<]!
 ; CHECK: da analyze - confused!
 ; CHECK: da analyze - none!
 ; CHECK: da analyze - confused!
 ; CHECK: da analyze - none!
 
 for.body.preheader:                               ; preds = %entry
   br label %for.body
 

 ;;    A[10] = i;
 ;;    *B++ = A[2*i];
 
 define void @weakzerosrc3(i32* %A, i32* %B, i64 %n) nounwind uwtable ssp {
 entry:
   br label %for.body
 
 ; CHECK: da analyze - consistent output [S]!
-; CHECK: da analyze - flow [=>p|<]!
+; CHECK: da analyze - flow [<=p|<]!
 ; CHECK: da analyze - confused!
 ; CHECK: da analyze - none!
 ; CHECK: da analyze - confused!
 ; CHECK: da analyze - none!
 
 for.body:                                         ; preds = %entry, %for.body
   %i.02 = phi i64 [ 0, %entry ], [ %inc, %for.body ]
   %B.addr.01 = phi i32* [ %B, %entry ], [ %incdec.ptr, %for.body ]