[LoopUnrollAnalyzer] Don't treat gep-instructions with simplified offset as simplified.

[mzolotukhin]

Summary

Currently we consider such instructions as simplified, which is incorrect,
because if their user isn't simplified, we can't actually simplify them too.
This biases our estimates of profitability: for instance the analyzer expects
much more gains from unrolling memcpy loops than there actually are.

[mzolotukhin]

Ping!

[mzolotukhin]

Ping!

[mzolotukhin]

Ping!

[chandlerc]

LGTM!

llvm/trunk/lib/Analysis/LoopUnrollAnalyzer.cpp

bool UnrolledInstAnalyzer::simplifyInstWithSCEV(Instruction *I) {
   auto *Offset =
       dyn_cast<SCEVConstant>(SE.getMinusSCEV(ValueAtIteration, Base));
   if (!Offset)
     return false;
   SimplifiedAddress Address;
   Address.Base = Base->getValue();
   Address.Offset = Offset->getValue();
   SimplifiedAddresses[I] = Address;
-  return true;
+  return false;
 }
 
 /// Try to simplify binary operator I.
 ///
 /// TODO: Probably it's worth to hoist the code for estimating the
 /// simplifications effects to a separate class, since we have a very similar
 /// code in InlineCost already.
 bool UnrolledInstAnalyzer::visitBinaryOperator(BinaryOperator &I) {

llvm/trunk/test/Transforms/LoopUnroll/full-unroll-heuristics-cmp.ll

-; RUN: opt < %s -S -loop-unroll -unroll-max-iteration-count-to-analyze=100 -unroll-dynamic-cost-savings-discount=1000 -unroll-threshold=10 -unroll-percent-dynamic-cost-saved-threshold=50 | FileCheck %s
+; RUN: opt < %s -S -loop-unroll -unroll-max-iteration-count-to-analyze=100 -unroll-dynamic-cost-savings-discount=1000 -unroll-threshold=10 -unroll-percent-dynamic-cost-saved-threshold=40 | FileCheck %s
 target datalayout = "e-m:o-i64:64-f80:128-n8:16:32:64-S128"
 
 @known_constant = internal unnamed_addr constant [10 x i32] [i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1], align 16
 
 ; If we can figure out result of comparison on each iteration, we can resolve
 ; the depending branch. That means, that the unrolled version of the loop would
 ; have less code, because we don't need not-taken basic blocks there.
 ; This test checks that this is taken into consideration.

llvm/trunk/test/Transforms/LoopUnroll/full-unroll-heuristics-geps.ll

+; RUN: opt < %s -S -loop-unroll -unroll-max-iteration-count-to-analyze=100 -unroll-dynamic-cost-savings-discount=1000 -unroll-threshold=10 -unroll-percent-dynamic-cost-saved-threshold=40 | FileCheck %s
+target datalayout = "e-m:o-i64:64-f80:128-n8:16:32:64-S128"
+
+; When examining gep-instructions we shouldn't consider them simplified if the
+; corresponding memory access isn't simplified. Doing the opposite might bias
+; our estimate, so that we might decide to unroll even a simple memcpy loop.
+;
+; Thus, the following loop shouldn't be unrolled:
+; CHECK-LABEL: @not_simplified_geps
+; CHECK: br i1 %
+; CHECK: ret void
+define void @not_simplified_geps(i32* noalias %b, i32* noalias %c) {
+entry:
+  br label %for.body
+
+for.body:
+  %iv.0 = phi i64 [ 0, %entry ], [ %iv.1, %for.body ]
+  %arrayidx1 = getelementptr inbounds i32, i32* %b, i64 %iv.0
+  %x1 = load i32, i32* %arrayidx1, align 4
+  %arrayidx2 = getelementptr inbounds i32, i32* %c, i64 %iv.0
+  store i32 %x1, i32* %arrayidx2, align 4
+  %iv.1 = add nuw nsw i64 %iv.0, 1
+  %exitcond = icmp eq i64 %iv.1, 10
+  br i1 %exitcond, label %for.end, label %for.body
+
+for.end:
+  ret void
+}