[SCEV] Fix trip multiple calculation

[huihuiz]

Summary

If loop bound containing calculations like min(a,b), the Scalar Evolution API getSmallConstantTripMultiple returns 4294967295 "-1" as the trip multiple. The problem is that, SCEV use -1 * umax to represent umin. The multiple constant -1 was returned, and the logic of guarding against huge trip counts was skipped. Because -1 has 32 active bits.

The fix attempt to factor more general cases. First try to get the greatest power of two divisor of trip count expression. In case overflow happens, the trip count expression is still divisible by the greatest power of two divisor returned. Returns 1 if not divisible by 2.

Patch by Huihui Zhang <[e-mail]>

[sanjoy]

Pretty close to lgtm, only very minor comments.

lib/Analysis/ScalarEvolution.cpp

5429 ↗

(On Diff #91381)

Can you please extract out this const ifying change and just land it?

5473 ↗

(On Diff #91381)

Should you be left shifting 1u? Please check, but I think left shifting a positive number into a negative number (in signed type) is undefined behavior.

[huihuiz]

Update based on Sanjoy's suggestions.

Use 1U left shift with unsigned number.

[sanjoy]

LGTM!

llvm/trunk/lib/Analysis/ScalarEvolution.cpp

ScalarEvolution::getSmallConstantTripMultiple(const Loop *L,
   assert(ExitingBlock && "Must pass a non-null exiting block!");
   assert(L->isLoopExiting(ExitingBlock) &&
          "Exiting block must actually branch out of the loop!");
   const SCEV *ExitCount = getExitCount(L, ExitingBlock);
   if (ExitCount == getCouldNotCompute())
     return 1;
 
   // Get the trip count from the BE count by adding 1.
-  const SCEV *TCMul = getAddExpr(ExitCount, getOne(ExitCount->getType()));
-  // FIXME: SCEV distributes multiplication as V1*C1 + V2*C1. We could attempt
-  // to factor simple cases.
-  if (const SCEVMulExpr *Mul = dyn_cast<SCEVMulExpr>(TCMul))
-    TCMul = Mul->getOperand(0);
+  const SCEV *TCExpr = getAddExpr(ExitCount, getOne(ExitCount->getType()));
 
-  const SCEVConstant *MulC = dyn_cast<SCEVConstant>(TCMul);
-  if (!MulC)
-    return 1;
+  const SCEVConstant *TC = dyn_cast<SCEVConstant>(TCExpr);
+  if (!TC)
+    // Attempt to factor more general cases. Returns the greatest power of
+    // two divisor. If overflow happens, the trip count expression is still
+    // divisible by the greatest power of 2 divisor returned.
+    return 1U << std::min((uint32_t)31, GetMinTrailingZeros(TCExpr));
 
-  ConstantInt *Result = MulC->getValue();
+  ConstantInt *Result = TC->getValue();
 
   // Guard against huge trip counts (this requires checking
   // for zero to handle the case where the trip count == -1 and the
   // addition wraps).
   if (!Result || Result->getValue().getActiveBits() > 32 ||
       Result->getValue().getActiveBits() == 0)
     return 1;
 
static void PrintLoopInfo(raw_ostream &OS, ScalarEvolution *SE,
   if (!isa<SCEVCouldNotCompute>(PBT)) {
     OS << "Predicated backedge-taken count is " << *PBT << "\n";
     OS << " Predicates:\n";
     Pred.print(OS, 4);
   } else {
     OS << "Unpredictable predicated backedge-taken count. ";
   }
   OS << "\n";
+
+  if (SE->hasLoopInvariantBackedgeTakenCount(L)) {
+    OS << "Loop ";
+    L->getHeader()->printAsOperand(OS, /*PrintType=*/false);
+    OS << ": ";
+    OS << "Trip multiple is " << SE->getSmallConstantTripMultiple(L) << "\n";
+  }
 }
 
 static StringRef loopDispositionToStr(ScalarEvolution::LoopDisposition LD) {
   switch (LD) {
   case ScalarEvolution::LoopVariant:
     return "Variant";
   case ScalarEvolution::LoopInvariant:
     return "Invariant";

llvm/trunk/test/Analysis/ScalarEvolution/tripmultiple_calculation.ll

+; RUN: opt -S -analyze -scalar-evolution < %s 2>&1 | FileCheck %s
+
+; umin is represented using -1 * umax in scalar evolution. -1 is considered as the
+; constant of the multiply expression (-1 * ((-1 + (-1 * %a)) umax (-1 + (-1 * %b)))).
+; Returns the greatest power of 2 divisor by evaluating the minimal trailing zeros
+; for the trip count expression.
+;
+; int foo(uint32_t a, uint32_t b, uint32_t *c) {
+;   for (uint32_t i = 0; i < (uint32_t)(a < b ? a : b) + 1; i++)
+;     c[i] = i;
+;   return 0;
+; }
+;
+; CHECK: Loop %for.body: Trip multiple is 1
+
+define i32 @foo(i32 %a, i32 %b, i32* %c) {
+entry:
+  %cmp = icmp ult i32 %a, %b
+  %cond = select i1 %cmp, i32 %a, i32 %b
+  %add = add i32 %cond, 1
+  %cmp18 = icmp eq i32 %add, 0
+  br i1 %cmp18, label %for.cond.cleanup, label %for.body.preheader
+
+for.body.preheader:                               ; preds = %entry
+  br label %for.body
+
+for.cond.cleanup.loopexit:                        ; preds = %for.body
+  br label %for.cond.cleanup
+
+for.cond.cleanup:                                 ; preds = %for.cond.cleanup.loopexit, %entry
+  ret i32 0
+
+for.body:                                         ; preds = %for.body.preheader, %for.body
+  %i.09 = phi i32 [ %inc, %for.body ], [ 0, %for.body.preheader ]
+  %arrayidx = getelementptr inbounds i32, i32* %c, i32 %i.09
+  store i32 %i.09, i32* %arrayidx, align 4
+  %inc = add nuw i32 %i.09, 1
+  %cmp1 = icmp ult i32 %inc, %add
+  br i1 %cmp1, label %for.body, label %for.cond.cleanup.loopexit
+}
+
+; Overflow may happen for the multiply expression n * 3, verify that trip
+; multiple is set to 1 if NUW/NSW are not set.
+;
+; __attribute__((noinline)) void a(unsigned n) {
+;   #pragma unroll(3)
+;   for (unsigned i = 0; i != n * 3; ++i)
+;     printf("TEST%u\n", i);
+; }
+; int main() { a(2863311531U); }
+;
+; CHECK: Loop %for.body: Trip multiple is 1
+
+@.str2 = private unnamed_addr constant [8 x i8] c"TEST%u\0A\00", align 1
+
+define void @foo2(i32 %n) {
+entry:
+  %mul = mul i32 %n, 3
+  %cmp4 = icmp eq i32 %mul, 0
+  br i1 %cmp4, label %for.cond.cleanup, label %for.body.preheader
+
+for.body.preheader:                               ; preds = %entry
+  br label %for.body
+
+for.cond.cleanup.loopexit:                        ; preds = %for.body
+  br label %for.cond.cleanup
+
+for.cond.cleanup:                                 ; preds = %for.cond.cleanup.loopexit, %entry
+  ret void
+
+for.body:                                         ; preds = %for.body.preheader, %for.body
+  %i.05 = phi i32 [ %inc, %for.body ], [ 0, %for.body.preheader ]
+  %call = tail call i32 (i8*, ...) @printf(i8* getelementptr inbounds ([8 x i8], [8 x i8]* @.str2, i32 0, i32 0), i32 %i.05)
+  %inc = add nuw i32 %i.05, 1
+  %cmp = icmp eq i32 %inc, %mul
+  br i1 %cmp, label %for.cond.cleanup.loopexit, label %for.body
+}
+
+declare i32 @printf(i8* nocapture readonly, ...)
+
+
+; If we couldn't prove no overflow for the multiply expression 24 * n,
+; returns the greatest power of 2 divisor. If overflows happens
+; the trip count is still divisible by the greatest power of 2 divisor.
+;
+; CHECK: Loop %l3: Trip multiple is 8
+
+declare void @f()
+
+define i32 @foo3(i32 %n) {
+entry:
+  %loop_ctl = mul i32 %n, 24
+  br label %l3
+
+l3:
+  %x.0 = phi i32 [ 0, %entry ], [ %inc, %l3 ]
+  call void @f()
+  %inc = add i32 %x.0, 1
+  %exitcond = icmp eq i32 %inc, %loop_ctl
+  br i1 %exitcond, label %exit, label %l3
+
+exit:
+  ret i32 0
+}
+
+; If the trip count is a constant, verify that we obtained the trip
+; count itself. For huge trip counts, or zero, we return 1.
+;
+; CHECK: Loop %l3: Trip multiple is 3
+
+define i32 @foo4(i32 %n) {
+entry:
+  br label %l3
+
+l3:
+  %x.0 = phi i32 [ 0, %entry ], [ %inc, %l3 ]
+  call void @f()
+  %inc = add i32 %x.0, 1
+  %exitcond = icmp eq i32 %inc, 3
+  br i1 %exitcond, label %exit, label %l3
+
+exit:
+  ret i32 0
+}
+