[llvm] [SCEV] Introduce IncrementNSUW, replacing NSSW (PR #219559)

[David Sherwood via llvm-commits]

================
@@ -687,93 +685,61 @@ define i32 @reverse_load_with_bounded(ptr %A, ptr %B, i32 %N) {
 ; CHECK:       [[VECTOR_SCEVCHECK]]:
 ; CHECK-NEXT:    [[TMP0:%.*]] = add i32 [[N]], -1
 ; CHECK-NEXT:    [[TMP1:%.*]] = icmp ugt i32 [[TMP0]], 3
-; CHECK-NEXT:    [[TMP2:%.*]] = icmp sgt i32 1, [[N]]
-; CHECK-NEXT:    [[TMP3:%.*]] = or i1 [[TMP1]], [[TMP2]]
-; CHECK-NEXT:    br i1 [[TMP3]], label %[[VEC_EPILOG_SCALAR_PH]], label %[[VECTOR_MAIN_LOOP_ITER_CHECK:.*]]
-; CHECK:       [[VECTOR_MAIN_LOOP_ITER_CHECK]]:
-; CHECK-NEXT:    [[MIN_ITERS_CHECK2:%.*]] = icmp ult i32 [[N]], 16
-; CHECK-NEXT:    br i1 [[MIN_ITERS_CHECK2]], label %[[VEC_EPILOG_PH:.*]], label %[[VECTOR_PH:.*]]
+; CHECK-NEXT:    br i1 [[TMP1]], label %[[VEC_EPILOG_SCALAR_PH]], label %[[VECTOR_PH:.*]]
 ; CHECK:       [[VECTOR_PH]]:
-; CHECK-NEXT:    [[TMP4:%.*]] = and i32 [[N]], 15
+; CHECK-NEXT:    [[TMP4:%.*]] = and i32 [[N]], 3
 ; CHECK-NEXT:    [[N_VEC:%.*]] = sub i32 [[N]], [[TMP4]]
 ; CHECK-NEXT:    br label %[[VECTOR_BODY:.*]]
 ; CHECK:       [[VECTOR_BODY]]:
 ; CHECK-NEXT:    [[INDEX:%.*]] = phi i32 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ]
-; CHECK-NEXT:    [[VEC_PHI:%.*]] = phi <4 x i32> [ zeroinitializer, %[[VECTOR_PH]] ], [ [[TMP20:%.*]], %[[VECTOR_BODY]] ]
-; CHECK-NEXT:    [[VEC_PHI3:%.*]] = phi <4 x i32> [ zeroinitializer, %[[VECTOR_PH]] ], [ [[TMP21:%.*]], %[[VECTOR_BODY]] ]
-; CHECK-NEXT:    [[VEC_PHI4:%.*]] = phi <4 x i32> [ zeroinitializer, %[[VECTOR_PH]] ], [ [[TMP22:%.*]], %[[VECTOR_BODY]] ]
-; CHECK-NEXT:    [[VEC_PHI5:%.*]] = phi <4 x i32> [ zeroinitializer, %[[VECTOR_PH]] ], [ [[TMP23:%.*]], %[[VECTOR_BODY]] ]
+; CHECK-NEXT:    [[VEC_PHI:%.*]] = phi i32 [ 0, %[[VECTOR_PH]] ], [ [[TMP34:%.*]], %[[VECTOR_BODY]] ]
+; CHECK-NEXT:    [[VEC_PHI1:%.*]] = phi i32 [ 0, %[[VECTOR_PH]] ], [ [[TMP35:%.*]], %[[VECTOR_BODY]] ]
+; CHECK-NEXT:    [[VEC_PHI2:%.*]] = phi i32 [ 0, %[[VECTOR_PH]] ], [ [[TMP36:%.*]], %[[VECTOR_BODY]] ]
+; CHECK-NEXT:    [[VEC_PHI3:%.*]] = phi i32 [ 0, %[[VECTOR_PH]] ], [ [[TMP37:%.*]], %[[VECTOR_BODY]] ]
+; CHECK-NEXT:    [[TMP26:%.*]] = add i32 [[INDEX]], 1
+; CHECK-NEXT:    [[INDEX19:%.*]] = add i32 [[INDEX]], 2
+; CHECK-NEXT:    [[TMP18:%.*]] = add i32 [[INDEX]], 3
 ; CHECK-NEXT:    [[TMP5:%.*]] = and i32 [[INDEX]], 3
+; CHECK-NEXT:    [[TMP7:%.*]] = and i32 [[TMP26]], 3
+; CHECK-NEXT:    [[TMP8:%.*]] = and i32 [[INDEX19]], 3
+; CHECK-NEXT:    [[TMP9:%.*]] = and i32 [[TMP18]], 3
 ; CHECK-NEXT:    [[TMP6:%.*]] = getelementptr inbounds i32, ptr [[A]], i32 [[TMP5]]
-; CHECK-NEXT:    [[TMP7:%.*]] = getelementptr inbounds i32, ptr [[TMP6]], i64 4
-; CHECK-NEXT:    [[TMP8:%.*]] = getelementptr inbounds i32, ptr [[TMP6]], i64 8
-; CHECK-NEXT:    [[TMP9:%.*]] = getelementptr inbounds i32, ptr [[TMP6]], i64 12
----------------
david-arm wrote:

This looks like a regression. @fhahn recently added supporting for vectorising these loops.

https://github.com/llvm/llvm-project/pull/219559