Watch out for a constant offset cancelling out a base register, forming

a zero. This situation arrises in Fortran code with induction variables
that start at 1 instead of 0. This fixes PR7651.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@108424 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/Transforms/Scalar/LoopStrengthReduce.cpp b/lib/Transforms/Scalar/LoopStrengthReduce.cpp
index a250a88c994730c0e498932ce5f3b9f2c3b786b3..1f9b4156b9cd6fc1e6c369c6cc124ed9b42d7fa9 100644 (file)
--- a/lib/Transforms/Scalar/LoopStrengthReduce.cpp
+++ b/lib/Transforms/Scalar/LoopStrengthReduce.cpp
@@ -2362,7 +2362,7 @@ void LSRInstance::GenerateConstantOffsets(LSRUse &LU, unsigned LUIdx,
                                           Formula Base) {
   // TODO: For now, just add the min and max offset, because it usually isn't
   // worthwhile looking at everything inbetween.
-  SmallVector<int64_t, 4> Worklist;
+  SmallVector<int64_t, 2> Worklist;
   Worklist.push_back(LU.MinOffset);
   if (LU.MaxOffset != LU.MinOffset)
     Worklist.push_back(LU.MaxOffset);
@@ -2376,7 +2376,14 @@ void LSRInstance::GenerateConstantOffsets(LSRUse &LU, unsigned LUIdx,
       F.AM.BaseOffs = (uint64_t)Base.AM.BaseOffs - *I;
       if (isLegalUse(F.AM, LU.MinOffset - *I, LU.MaxOffset - *I,
                      LU.Kind, LU.AccessTy, TLI)) {
-        F.BaseRegs[i] = SE.getAddExpr(G, SE.getConstant(G->getType(), *I));
+        // Add the offset to the base register.
+        const SCEV *NewG = SE.getAddExpr(G, SE.getConstant(G->getType(), *I));
+        // If it cancelled out, drop the base register, otherwise update it.
+        if (NewG->isZero()) {
+          std::swap(F.BaseRegs[i], F.BaseRegs.back());
+          F.BaseRegs.pop_back();
+        } else
+          F.BaseRegs[i] = NewG;
 
         (void)InsertFormula(LU, LUIdx, F);
       }

diff --git a/test/CodeGen/X86/lsr-i386.ll b/test/CodeGen/X86/lsr-i386.ll
new file mode 100644 (file)
index 0000000..02baf20
--- /dev/null
+++ b/test/CodeGen/X86/lsr-i386.ll
@@ -0,0 +1,44 @@
+; RUN: llc -march=x86 < %s | FileCheck %s
+target datalayout = "e-p:32:32:32-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:32:64-f32:32:32-f64:32:64-v64:64:64-v128:128:128-a0:0:64-f80:32:32-n8:16:32"
+target triple = "i386-pc-linux-gnu"
+; PR7651
+
+; CHECK: align
+; CHECK: align
+; CHECK: align
+; CHECK: movl  $0, (%e
+; CHECK-NEXT: addl  $4, %e
+; CHECK-NEXT: decl  %e
+; CHECK-NEXT: jne
+
+%struct.anon = type { [72 x i32], i32 }
+
+@mp2grad_ = external global %struct.anon
+
+define void @chomp2g_setup_(i32 %n, i32 %m) nounwind {
+entry:
+  br label %bb1
+
+bb1:                                              ; preds = %bb6, %bb
+  %indvar11 = phi i32 [ %indvar.next12, %bb6 ], [ 0, %entry ] ; <i32> [#uses=2]
+  %tmp21 = add i32 %indvar11, 1                   ; <i32> [#uses=1]
+  %t = load i32* getelementptr inbounds (%struct.anon* @mp2grad_, i32 0, i32 1)
+  %tmp15 = mul i32 %n, %t                      ; <i32> [#uses=1]
+  %tmp16 = add i32 %tmp21, %tmp15                 ; <i32> [#uses=1]
+  %tmp17 = shl i32 %tmp16, 3                      ; <i32> [#uses=1]
+  %tmp18 = add i32 %tmp17, -8                     ; <i32> [#uses=1]
+  br label %bb2
+
+bb2:                                              ; preds = %bb2, %bb2.preheader
+  %indvar = phi i32 [ 0, %bb1 ], [ %indvar.next, %bb2 ] ; <i32> [#uses=2]
+  %tmp19 = add i32 %tmp18, %indvar                ; <i32> [#uses=1]
+  %scevgep = getelementptr %struct.anon* @mp2grad_, i32 0, i32 0, i32 %tmp19 ; <i32*> [#uses=1]
+  store i32 0, i32* %scevgep
+  %indvar.next = add i32 %indvar, 1               ; <i32> [#uses=1]
+  %c = icmp ne i32 %indvar.next, %m
+  br i1 %c, label %bb2, label %bb6
+
+bb6:                                              ; preds = %bb2, %bb1
+  %indvar.next12 = add i32 %indvar11, 1           ; <i32> [#uses=1]
+  br label %bb1
+}