Index: lib/Transforms/Utils/LoopUnroll.cpp
===================================================================
--- lib/Transforms/Utils/LoopUnroll.cpp
+++ lib/Transforms/Utils/LoopUnroll.cpp
@@ -396,8 +396,19 @@
          "Did not expect runtime trip-count unrolling "
          "and peeling for the same loop");
 
-  if (PeelCount)
-    peelLoop(L, PeelCount, LI, SE, DT, AC, PreserveLCSSA);
+  if (PeelCount) {
+    bool Peeled = peelLoop(L, PeelCount, LI, SE, DT, AC, PreserveLCSSA);
+
+    // Successful peeling may result in a change in the loop preheader/trip
+    // counts. If we later unroll the loop, we want these to be updated.
+    if (Peeled) {
+      BasicBlock *ExitingBlock = L->getExitingBlock();
+      assert(ExitingBlock && "Loop without exiting block?");
+      Preheader = L->getLoopPreheader();
+      TripCount = SE->getSmallConstantTripCount(L, ExitingBlock);
+      TripMultiple = SE->getSmallConstantTripMultiple(L, ExitingBlock);
+    }
+  }
 
   // Loops containing convergent instructions must have a count that divides
   // their TripMultiple.
Index: test/Transforms/LoopUnroll/SystemZ/lit.local.cfg
===================================================================
--- /dev/null
+++ test/Transforms/LoopUnroll/SystemZ/lit.local.cfg
@@ -0,0 +1,2 @@
+if not 'SystemZ' in config.root.targets:
+    config.unsupported = True
Index: test/Transforms/LoopUnroll/SystemZ/pr33437.ll
===================================================================
--- /dev/null
+++ test/Transforms/LoopUnroll/SystemZ/pr33437.ll
@@ -0,0 +1,96 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
+; RUN: opt -loop-unroll -S %s | FileCheck %s
+
+target datalayout = "E-m:e-i1:8:16-i8:8:16-i64:64-f128:64-v128:64-a:8:16-n32:64"
+target triple = "s390x-ibm-linux"
+
+@g_72 = external local_unnamed_addr global i32, align 4
+@g_74 = external local_unnamed_addr global i64, align 8
+@g_126 = external local_unnamed_addr global float, align 4
+@g_202 = external global i64, align 8
+@func_6.l_542 = external unnamed_addr constant [3 x [3 x [10 x i32]]], align 4
+@g_531 = external local_unnamed_addr global i64*, align 8
+@g_623 = external local_unnamed_addr global i32*, align 8
+@g_625 = external local_unnamed_addr global i32*, align 8
+@g_355 = external local_unnamed_addr global [6 x [5 x i32*]], align 8
+
+declare void @llvm.memcpy.p0i8.p0i8.i64(i8* nocapture writeonly, i8* nocapture readonly, i64, i32, i1) #0
+declare zeroext i8 @patatino()
+
+define fastcc void @func_6() unnamed_addr {
+; CHECK-LABEL: @func_6(
+; CHECK-NEXT:  entry:
+; CHECK-NEXT:    br label [[FOR_COND93:%.*]]
+; CHECK:       for.cond93.loopexit:
+; CHECK-NEXT:    ret void
+; CHECK:       for.cond93:
+; CHECK-NEXT:    br label [[FOR_BODY198_PEEL_BEGIN:%.*]]
+; CHECK:       for.body198.peel.begin:
+; CHECK-NEXT:    br label [[FOR_BODY198_PEEL:%.*]]
+; CHECK:       for.body198.peel:
+; CHECK-NEXT:    store i64 0, i64* @g_74, align 8
+; CHECK-NEXT:    tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* undef, i8* bitcast ([3 x [3 x [10 x i32]]]* @func_6.l_542 to i8*), i64 360, i32 4, i1 false)
+; CHECK-NEXT:    store i64* @g_202, i64** @g_531, align 8
+; CHECK-NEXT:    store i8 0, i8* undef, align 1
+; CHECK-NEXT:    store i8 0, i8* undef, align 1
+; CHECK-NEXT:    store i32* null, i32** @g_623, align 8
+; CHECK-NEXT:    store i32* null, i32** @g_625, align 8
+; CHECK-NEXT:    store i32 0, i32* @g_72, align 4
+; CHECK-NEXT:    [[CALL593_PEEL:%.*]] = tail call zeroext i8 @patatino()
+; CHECK-NEXT:    store i64 -1, i64* undef, align 8
+; CHECK-NEXT:    store i64 6707501019708462161, i64* undef, align 8
+; CHECK-NEXT:    store float 0.000000e+00, float* undef, align 4
+; CHECK-NEXT:    store float 0.000000e+00, float* @g_126, align 4
+; CHECK-NEXT:    store i32* null, i32** getelementptr inbounds ([6 x [5 x i32*]], [6 x [5 x i32*]]* @g_355, i64 0, i64 4, i64 4), align 8
+; CHECK-NEXT:    br i1 false, label [[FOR_BODY198_PEEL_NEXT:%.*]], label [[FOR_COND93_LOOPEXIT:%.*]]
+; CHECK:       for.body198.peel.next:
+; CHECK-NEXT:    br label [[FOR_BODY198_PEEL_NEXT1:%.*]]
+; CHECK:       for.body198.peel.next1:
+; CHECK-NEXT:    br label [[FOR_COND93_PEEL_NEWPH:%.*]]
+; CHECK:       for.cond93.peel.newph:
+; CHECK-NEXT:    br label [[FOR_BODY198:%.*]]
+; CHECK:       for.body198:
+; CHECK-NEXT:    store i64 0, i64* @g_74, align 8
+; CHECK-NEXT:    tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* undef, i8* bitcast ([3 x [3 x [10 x i32]]]* @func_6.l_542 to i8*), i64 360, i32 4, i1 false)
+; CHECK-NEXT:    store i64* @g_202, i64** @g_531, align 8
+; CHECK-NEXT:    store i8 0, i8* undef, align 1
+; CHECK-NEXT:    store i8 0, i8* undef, align 1
+; CHECK-NEXT:    store i32* null, i32** @g_623, align 8
+; CHECK-NEXT:    store i32* null, i32** @g_625, align 8
+; CHECK-NEXT:    store i32 0, i32* @g_72, align 4
+; CHECK-NEXT:    [[CALL593:%.*]] = tail call zeroext i8 @patatino()
+; CHECK-NEXT:    store i64 -1, i64* undef, align 8
+; CHECK-NEXT:    store i64 6707501019708462161, i64* undef, align 8
+; CHECK-NEXT:    store float 0.000000e+00, float* undef, align 4
+; CHECK-NEXT:    store float 0.000000e+00, float* @g_126, align 4
+; CHECK-NEXT:    store i32* null, i32** getelementptr inbounds ([6 x [5 x i32*]], [6 x [5 x i32*]]* @g_355, i64 0, i64 4, i64 4), align 8
+; CHECK-NEXT:    br label [[FOR_COND93_LOOPEXIT]]
+;
+entry:
+  br label %for.cond93
+
+for.cond93.loopexit:                              ; preds = %for.body198
+  ret void
+
+for.cond93:                                       ; preds = %entry
+  br label %for.body198
+
+for.body198:                                      ; preds = %for.body198, %for.cond93
+  %l_249.12 = phi i8 [ undef, %for.cond93 ], [ %call593, %for.body198 ]
+  %l_522.01 = phi i32 [ 0, %for.cond93 ], [ 1, %for.body198 ]
+  store i64 0, i64* @g_74, align 8
+  tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* undef, i8* bitcast ([3 x [3 x [10 x i32]]]* @func_6.l_542 to i8*), i64 360, i32 4, i1 false)
+  store i64* @g_202, i64** @g_531, align 8
+  store i8 0, i8* undef, align 1
+  store i8 0, i8* undef, align 1
+  store i32* null, i32** @g_623, align 8
+  store i32* null, i32** @g_625, align 8
+  store i32 0, i32* @g_72, align 4
+  %call593 = tail call zeroext i8 @patatino()
+  store i64 -1, i64* undef, align 8
+  store i64 6707501019708462161, i64* undef, align 8
+  store float 0.000000e+00, float* undef, align 4
+  store float 0.000000e+00, float* @g_126, align 4
+  store i32* null, i32** getelementptr inbounds ([6 x [5 x i32*]], [6 x [5 x i32*]]* @g_355, i64 0, i64 4, i64 4), align 8
+  br i1 false, label %for.body198, label %for.cond93.loopexit
+}