[LV] Use SCEV for subtraction of src/sink for diff runtime checks.

Instead of expanding the src/sink SCEV expressions and emitting an IR
sub to compute the difference, the subtraction can be directly be
performed by ScalarEvolution. This allows the subtraction to be
simplified by SCEV, which in turn can reduced the number of redundant
runtime check instructions generated.

It also allows to generate checks that are invariant w.r.t. an outer
loop, if he inner loop AddRecs have the same outer loop AddRec as start.

Author: fhahn

llvm/lib/Transforms/Utils/LoopUtils.cpp

@@ -1793,9 +1793,9 @@ Value *llvm::addDiffRuntimeChecks(
     auto *VFTimesUFTimesSize =
         ChkBuilder.CreateMul(GetVF(ChkBuilder, Ty->getScalarSizeInBits()),
                              ConstantInt::get(Ty, IC * C.AccessSize));
-    Value *Sink = Expander.expandCodeFor(C.SinkStart, Ty, Loc);
-    Value *Src = Expander.expandCodeFor(C.SrcStart, Ty, Loc);
-    Value *Diff = ChkBuilder.CreateSub(Sink, Src);
+    auto &SE = *Expander.getSE();
+    Value *Diff = Expander.expandCodeFor(
+        SE.getMinusSCEV(C.SinkStart, C.SrcStart), Ty, Loc);
     Value *IsConflict =
         ChkBuilder.CreateICmpULT(Diff, VFTimesUFTimesSize, "diff.check");
     if (C.NeedsFreeze)

llvm/test/Transforms/LoopVectorize/AArch64/sve-vector-reverse.ll

@@ -117,10 +117,7 @@ define void @vector_reverse_i64(i64 %N, ptr %a, ptr %b) #0 {
 ; CHECK:       vector.memcheck:
 ; CHECK-NEXT:    [[TMP2:%.*]] = call i64 @llvm.vscale.i64()
 ; CHECK-NEXT:    [[TMP3:%.*]] = shl i64 [[TMP2]], 7
-; CHECK-NEXT:    [[TMP4:%.*]] = shl i64 [[N]], 3
-; CHECK-NEXT:    [[TMP5:%.*]] = add i64 [[TMP4]], [[B1]]
-; CHECK-NEXT:    [[TMP6:%.*]] = add i64 [[TMP4]], [[A2]]
-; CHECK-NEXT:    [[TMP7:%.*]] = sub i64 [[TMP5]], [[TMP6]]
+; CHECK-NEXT:    [[TMP7:%.*]] = sub i64 [[B1]], [[A2]]
 ; CHECK-NEXT:    [[DIFF_CHECK:%.*]] = icmp ult i64 [[TMP7]], [[TMP3]]
 ; CHECK-NEXT:    br i1 [[DIFF_CHECK]], label [[SCALAR_PH]], label [[VECTOR_PH:%.*]]
 ; CHECK:       vector.ph:

llvm/test/Transforms/LoopVectorize/runtime-check-small-clamped-bounds.ll

@@ -174,10 +174,8 @@ define void @load_clamped_index_offset_1(ptr %A, ptr %B, i32 %N) {
 ; CHECK-NEXT:    [[TMP6:%.*]] = or i1 [[TMP4]], [[TMP5]]
 ; CHECK-NEXT:    br i1 [[TMP6]], label [[SCALAR_PH]], label [[VECTOR_MEMCHECK:%.*]]
 ; CHECK:       vector.memcheck:
-; CHECK-NEXT:    [[TMP7:%.*]] = add nuw i64 [[B1]], 4
-; CHECK-NEXT:    [[TMP8:%.*]] = add i64 [[A2]], 4
-; CHECK-NEXT:    [[TMP9:%.*]] = sub i64 [[TMP7]], [[TMP8]]
-; CHECK-NEXT:    [[DIFF_CHECK:%.*]] = icmp ult i64 [[TMP9]], 16
+; CHECK-NEXT:    [[TMP7:%.*]] = sub i64 [[B1]], [[A2]]
+; CHECK-NEXT:    [[DIFF_CHECK:%.*]] = icmp ult i64 [[TMP7]], 16
 ; CHECK-NEXT:    br i1 [[DIFF_CHECK]], label [[SCALAR_PH]], label [[VECTOR_PH:%.*]]
 ; CHECK:       vector.ph:
 ; CHECK-NEXT:    [[N_MOD_VF:%.*]] = urem i32 [[TMP0]], 4