1 ; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 3
2 ; RUN: opt -passes=loop-vectorize -mcpu=skx -enable-masked-interleaved-mem-accesses=1 -force-vector-width=4 -force-vector-interleave=1 -S %s | FileCheck %s
3 target datalayout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-f80:128-n8:16:32:64-S128-ni:1-p2:32:8:8:32-ni:2"
4 target triple = "x86_64-apple-macos"
6 ; %l2 load and the preceeding store has a dependency. We should not sink
7 ; that store into the last store (by creating an interleaved store group). This
8 ; means the loaded %l2 will have incorrect value.
9 define void @avoid_sinking_store_across_load(ptr %arr) {
10 ; CHECK-LABEL: define void @avoid_sinking_store_across_load(
11 ; CHECK-SAME: ptr [[ARR:%.*]]) #[[ATTR0:[0-9]+]] {
13 ; CHECK-NEXT: br i1 false, label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
15 ; CHECK-NEXT: br label [[VECTOR_BODY:%.*]]
17 ; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
18 ; CHECK-NEXT: [[VEC_IND:%.*]] = phi <4 x i64> [ <i64 1, i64 4, i64 7, i64 10>, [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[VECTOR_BODY]] ]
19 ; CHECK-NEXT: [[VEC_IND2:%.*]] = phi <4 x i64> [ <i64 4, i64 7, i64 10, i64 13>, [[VECTOR_PH]] ], [ [[VEC_IND_NEXT3:%.*]], [[VECTOR_BODY]] ]
20 ; CHECK-NEXT: [[TMP0:%.*]] = mul i64 [[INDEX]], 3
21 ; CHECK-NEXT: [[OFFSET_IDX:%.*]] = add i64 1, [[TMP0]]
22 ; CHECK-NEXT: [[TMP1:%.*]] = add i64 [[OFFSET_IDX]], 0
23 ; CHECK-NEXT: [[TMP2:%.*]] = add nuw nsw i64 [[TMP1]], 4
24 ; CHECK-NEXT: [[TMP3:%.*]] = getelementptr inbounds i32, ptr [[ARR]], i64 [[TMP2]]
25 ; CHECK-NEXT: [[WIDE_VEC:%.*]] = load <12 x i32>, ptr [[TMP3]], align 4
26 ; CHECK-NEXT: [[STRIDED_VEC:%.*]] = shufflevector <12 x i32> [[WIDE_VEC]], <12 x i32> poison, <4 x i32> <i32 0, i32 3, i32 6, i32 9>
27 ; CHECK-NEXT: [[TMP4:%.*]] = getelementptr inbounds i32, ptr [[ARR]], <4 x i64> [[VEC_IND2]]
28 ; CHECK-NEXT: [[TMP5:%.*]] = add nuw nsw <4 x i64> [[VEC_IND]], splat (i64 2)
29 ; CHECK-NEXT: [[TMP6:%.*]] = getelementptr inbounds i32, ptr [[ARR]], <4 x i64> [[TMP5]]
30 ; CHECK-NEXT: [[TMP7:%.*]] = mul <4 x i32> [[STRIDED_VEC]], splat (i32 25)
31 ; CHECK-NEXT: call void @llvm.masked.scatter.v4i32.v4p0(<4 x i32> [[TMP7]], <4 x ptr> [[TMP6]], i32 4, <4 x i1> splat (i1 true))
32 ; CHECK-NEXT: [[TMP8:%.*]] = extractelement <4 x ptr> [[TMP6]], i32 0
33 ; CHECK-NEXT: [[WIDE_VEC4:%.*]] = load <12 x i32>, ptr [[TMP8]], align 4
34 ; CHECK-NEXT: [[STRIDED_VEC5:%.*]] = shufflevector <12 x i32> [[WIDE_VEC4]], <12 x i32> poison, <4 x i32> <i32 0, i32 3, i32 6, i32 9>
35 ; CHECK-NEXT: [[STRIDED_VEC6:%.*]] = shufflevector <12 x i32> [[WIDE_VEC4]], <12 x i32> poison, <4 x i32> <i32 1, i32 4, i32 7, i32 10>
36 ; CHECK-NEXT: [[TMP9:%.*]] = add <4 x i32> [[STRIDED_VEC6]], [[STRIDED_VEC5]]
37 ; CHECK-NEXT: call void @llvm.masked.scatter.v4i32.v4p0(<4 x i32> [[TMP9]], <4 x ptr> [[TMP4]], i32 4, <4 x i1> splat (i1 true))
38 ; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4
39 ; CHECK-NEXT: [[VEC_IND_NEXT]] = add <4 x i64> [[VEC_IND]], splat (i64 12)
40 ; CHECK-NEXT: [[VEC_IND_NEXT3]] = add <4 x i64> [[VEC_IND2]], splat (i64 12)
41 ; CHECK-NEXT: [[TMP10:%.*]] = icmp eq i64 [[INDEX_NEXT]], 16
42 ; CHECK-NEXT: br i1 [[TMP10]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]
43 ; CHECK: middle.block:
44 ; CHECK-NEXT: br label [[SCALAR_PH]]
46 ; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ 49, [[MIDDLE_BLOCK]] ], [ 1, [[ENTRY:%.*]] ]
47 ; CHECK-NEXT: [[BC_RESUME_VAL1:%.*]] = phi i64 [ 52, [[MIDDLE_BLOCK]] ], [ 4, [[ENTRY]] ]
48 ; CHECK-NEXT: br label [[LOOP:%.*]]
50 ; CHECK-NEXT: [[IV_1:%.*]] = phi i64 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[IV_1_NEXT:%.*]], [[LOOP]] ]
51 ; CHECK-NEXT: [[IV_2:%.*]] = phi i64 [ [[BC_RESUME_VAL1]], [[SCALAR_PH]] ], [ [[IV_2_NEXT:%.*]], [[LOOP]] ]
52 ; CHECK-NEXT: [[IV_1_NEXT]] = add nuw nsw i64 [[IV_1]], 3
53 ; CHECK-NEXT: [[IV_1_PLUS_4:%.*]] = add nuw nsw i64 [[IV_1]], 4
54 ; CHECK-NEXT: [[GEP_IV_1_PLUS_4:%.*]] = getelementptr inbounds i32, ptr [[ARR]], i64 [[IV_1_PLUS_4]]
55 ; CHECK-NEXT: [[L1:%.*]] = load i32, ptr [[GEP_IV_1_PLUS_4]], align 4
56 ; CHECK-NEXT: [[GEP_IV_2:%.*]] = getelementptr inbounds i32, ptr [[ARR]], i64 [[IV_2]]
57 ; CHECK-NEXT: [[IV_1_PLUS_2:%.*]] = add nuw nsw i64 [[IV_1]], 2
58 ; CHECK-NEXT: [[GEP_IV_1_PLUS_2:%.*]] = getelementptr inbounds i32, ptr [[ARR]], i64 [[IV_1_PLUS_2]]
59 ; CHECK-NEXT: [[MUL:%.*]] = mul i32 [[L1]], 25
60 ; CHECK-NEXT: store i32 [[MUL]], ptr [[GEP_IV_1_PLUS_2]], align 4
61 ; CHECK-NEXT: [[L2:%.*]] = load i32, ptr [[GEP_IV_1_PLUS_2]], align 4
62 ; CHECK-NEXT: [[L3:%.*]] = load i32, ptr [[GEP_IV_2]], align 4
63 ; CHECK-NEXT: [[ADD:%.*]] = add i32 [[L3]], [[L2]]
64 ; CHECK-NEXT: store i32 [[ADD]], ptr [[GEP_IV_2]], align 4
65 ; CHECK-NEXT: [[IV_2_NEXT]] = add nuw nsw i64 [[IV_2]], 3
66 ; CHECK-NEXT: [[ICMP:%.*]] = icmp ugt i64 [[IV_2]], 50
67 ; CHECK-NEXT: br i1 [[ICMP]], label [[EXIT:%.*]], label [[LOOP]], !llvm.loop [[LOOP3:![0-9]+]]
69 ; CHECK-NEXT: ret void
75 %iv.1 = phi i64 [ 1, %entry ], [ %iv.1.next, %loop ]
76 %iv.2 = phi i64 [ 4, %entry ], [ %iv.2.next, %loop ]
77 %iv.1.next = add nuw nsw i64 %iv.1, 3
78 %iv.1.plus.4 = add nuw nsw i64 %iv.1, 4
79 %gep.iv.1.plus.4 = getelementptr inbounds i32, ptr %arr, i64 %iv.1.plus.4
80 %l1 = load i32, ptr %gep.iv.1.plus.4
81 %gep.iv.2 = getelementptr inbounds i32, ptr %arr, i64 %iv.2
82 %iv.1.plus.2 = add nuw nsw i64 %iv.1, 2
83 %gep.iv.1.plus.2= getelementptr inbounds i32, ptr %arr, i64 %iv.1.plus.2
84 %mul = mul i32 %l1, 25
85 store i32 %mul, ptr %gep.iv.1.plus.2
86 %l2 = load i32, ptr %gep.iv.1.plus.2
87 %l3 = load i32, ptr %gep.iv.2
88 %add = add i32 %l3 , %l2
89 store i32 %add, ptr %gep.iv.2
90 %iv.2.next = add nuw nsw i64 %iv.2, 3
91 %icmp = icmp ugt i64 %iv.2, 50
92 br i1 %icmp, label %exit, label %loop