xref: /llvm-project/llvm/test/Transforms/SLPVectorizer/AMDGPU/phi-result-use-order.ll (revision ea43a30899df5c3c36412392c8f4db79973a1c43)
193a9be0cSAlexey Bataev; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
2e98501e2Sskc7; RUN: opt -passes=slp-vectorizer -S -mtriple=amdgcn-amd-amdhsa -mcpu=gfx906 < %s | FileCheck %s
3e98501e2Sskc7
4e98501e2Sskc7define <4 x half> @phis(i1 %cmp1, <4 x half> %in1, <4 x half> %in2)  {
5e98501e2Sskc7; CHECK-LABEL: @phis(
6e98501e2Sskc7; CHECK-NEXT:  entry:
7*ea43a308SJeffrey Byrnes; CHECK-NEXT:    [[TMP0:%.*]] = shufflevector <4 x half> [[IN1:%.*]], <4 x half> poison, <2 x i32> <i32 0, i32 1>
8*ea43a308SJeffrey Byrnes; CHECK-NEXT:    [[TMP1:%.*]] = shufflevector <4 x half> [[IN1]], <4 x half> poison, <2 x i32> <i32 2, i32 3>
993a9be0cSAlexey Bataev; CHECK-NEXT:    br i1 [[CMP1:%.*]], label [[BB1:%.*]], label [[BB0:%.*]]
10e98501e2Sskc7; CHECK:       bb0:
11*ea43a308SJeffrey Byrnes; CHECK-NEXT:    [[TMP2:%.*]] = shufflevector <4 x half> [[IN2:%.*]], <4 x half> poison, <2 x i32> <i32 0, i32 1>
12*ea43a308SJeffrey Byrnes; CHECK-NEXT:    [[TMP3:%.*]] = shufflevector <4 x half> [[IN2]], <4 x half> poison, <2 x i32> <i32 2, i32 3>
1393a9be0cSAlexey Bataev; CHECK-NEXT:    br label [[BB1]]
14e98501e2Sskc7; CHECK:       bb1:
15*ea43a308SJeffrey Byrnes; CHECK-NEXT:    [[TMP4:%.*]] = phi <2 x half> [ [[TMP0]], [[ENTRY:%.*]] ], [ [[TMP2]], [[BB0]] ]
16*ea43a308SJeffrey Byrnes; CHECK-NEXT:    [[TMP5:%.*]] = phi <2 x half> [ [[TMP1]], [[ENTRY]] ], [ [[TMP3]], [[BB0]] ]
17*ea43a308SJeffrey Byrnes; CHECK-NEXT:    [[TMP6:%.*]] = shufflevector <2 x half> [[TMP4]], <2 x half> poison, <4 x i32> <i32 0, i32 1, i32 poison, i32 poison>
18*ea43a308SJeffrey Byrnes; CHECK-NEXT:    [[TMP7:%.*]] = shufflevector <2 x half> [[TMP5]], <2 x half> poison, <4 x i32> <i32 0, i32 1, i32 poison, i32 poison>
19*ea43a308SJeffrey Byrnes; CHECK-NEXT:    [[TMP8:%.*]] = shufflevector <2 x half> [[TMP4]], <2 x half> [[TMP5]], <4 x i32> <i32 0, i32 1, i32 2, i32 3>
20*ea43a308SJeffrey Byrnes; CHECK-NEXT:    ret <4 x half> [[TMP8]]
2193a9be0cSAlexey Bataev;
22e98501e2Sskc7entry:
23e98501e2Sskc7  %a0 = extractelement <4 x half> %in1, i64 0
24e98501e2Sskc7  %a1 = extractelement <4 x half> %in1, i64 1
25e98501e2Sskc7  %a2 = extractelement <4 x half> %in1, i64 2
26e98501e2Sskc7  %a3 = extractelement <4 x half> %in1, i64 3
27e98501e2Sskc7  br i1 %cmp1, label %bb1, label %bb0
28e98501e2Sskc7
29e98501e2Sskc7bb0:
30e98501e2Sskc7  %b0 = extractelement <4 x half> %in2, i64 0
31e98501e2Sskc7  %b1 = extractelement <4 x half> %in2, i64 1
32e98501e2Sskc7  %b2 = extractelement <4 x half> %in2, i64 2
33e98501e2Sskc7  %b3 = extractelement <4 x half> %in2, i64 3
34e98501e2Sskc7  br label %bb1
35e98501e2Sskc7
36e98501e2Sskc7bb1:
37e98501e2Sskc7  %c0 = phi half [ %a0, %entry ], [ %b0, %bb0 ]
38e98501e2Sskc7  %c1 = phi half [ %a1, %entry ], [ %b1, %bb0 ]
39e98501e2Sskc7  %c2 = phi half [ %a2, %entry ], [ %b2, %bb0 ]
40e98501e2Sskc7  %c3 = phi half [ %a3, %entry ], [ %b3, %bb0 ]
41e98501e2Sskc7
42e98501e2Sskc7  %o0 = insertelement <4 x half> undef, half %c0, i64 0
43e98501e2Sskc7  %o1 = insertelement <4 x half> %o0, half %c1, i64 1
44e98501e2Sskc7  %o2 = insertelement <4 x half> %o1, half %c2, i64 2
45e98501e2Sskc7  %o3 = insertelement <4 x half> %o2, half %c3, i64 3
46e98501e2Sskc7  ret <4 x half> %o3
47e98501e2Sskc7}
48e98501e2Sskc7
49e98501e2Sskc7define <4 x half> @phis_reverse(i1 %cmp1, <4 x half> %in1, <4 x half> %in2)  {
50e98501e2Sskc7; CHECK-LABEL: @phis_reverse(
51e98501e2Sskc7; CHECK-NEXT:  entry:
52*ea43a308SJeffrey Byrnes; CHECK-NEXT:    [[TMP0:%.*]] = shufflevector <4 x half> [[IN1:%.*]], <4 x half> poison, <2 x i32> <i32 2, i32 3>
53*ea43a308SJeffrey Byrnes; CHECK-NEXT:    [[TMP1:%.*]] = shufflevector <4 x half> [[IN1]], <4 x half> poison, <2 x i32> <i32 0, i32 1>
5493a9be0cSAlexey Bataev; CHECK-NEXT:    br i1 [[CMP1:%.*]], label [[BB1:%.*]], label [[BB0:%.*]]
55e98501e2Sskc7; CHECK:       bb0:
56*ea43a308SJeffrey Byrnes; CHECK-NEXT:    [[TMP2:%.*]] = shufflevector <4 x half> [[IN2:%.*]], <4 x half> poison, <2 x i32> <i32 2, i32 3>
57*ea43a308SJeffrey Byrnes; CHECK-NEXT:    [[TMP3:%.*]] = shufflevector <4 x half> [[IN2]], <4 x half> poison, <2 x i32> <i32 0, i32 1>
5893a9be0cSAlexey Bataev; CHECK-NEXT:    br label [[BB1]]
59e98501e2Sskc7; CHECK:       bb1:
60*ea43a308SJeffrey Byrnes; CHECK-NEXT:    [[TMP4:%.*]] = phi <2 x half> [ [[TMP0]], [[ENTRY:%.*]] ], [ [[TMP2]], [[BB0]] ]
61*ea43a308SJeffrey Byrnes; CHECK-NEXT:    [[TMP5:%.*]] = phi <2 x half> [ [[TMP1]], [[ENTRY]] ], [ [[TMP3]], [[BB0]] ]
62*ea43a308SJeffrey Byrnes; CHECK-NEXT:    [[TMP6:%.*]] = shufflevector <2 x half> [[TMP5]], <2 x half> poison, <4 x i32> <i32 0, i32 1, i32 poison, i32 poison>
63*ea43a308SJeffrey Byrnes; CHECK-NEXT:    [[TMP7:%.*]] = shufflevector <2 x half> [[TMP4]], <2 x half> poison, <4 x i32> <i32 0, i32 1, i32 poison, i32 poison>
64*ea43a308SJeffrey Byrnes; CHECK-NEXT:    [[TMP8:%.*]] = shufflevector <4 x half> [[TMP6]], <4 x half> [[TMP7]], <4 x i32> <i32 0, i32 1, i32 4, i32 5>
65*ea43a308SJeffrey Byrnes; CHECK-NEXT:    ret <4 x half> [[TMP8]]
6693a9be0cSAlexey Bataev;
67e98501e2Sskc7entry:
68e98501e2Sskc7  %a0 = extractelement <4 x half> %in1, i64 0
69e98501e2Sskc7  %a1 = extractelement <4 x half> %in1, i64 1
70e98501e2Sskc7  %a2 = extractelement <4 x half> %in1, i64 2
71e98501e2Sskc7  %a3 = extractelement <4 x half> %in1, i64 3
72e98501e2Sskc7  br i1 %cmp1, label %bb1, label %bb0
73e98501e2Sskc7
74e98501e2Sskc7bb0:
75e98501e2Sskc7  %b0 = extractelement <4 x half> %in2, i64 0
76e98501e2Sskc7  %b1 = extractelement <4 x half> %in2, i64 1
77e98501e2Sskc7  %b2 = extractelement <4 x half> %in2, i64 2
78e98501e2Sskc7  %b3 = extractelement <4 x half> %in2, i64 3
79e98501e2Sskc7  br label %bb1
80e98501e2Sskc7
81e98501e2Sskc7bb1:
82e98501e2Sskc7  %c3 = phi half [ %a3, %entry ], [ %b3, %bb0 ]
83e98501e2Sskc7  %c2 = phi half [ %a2, %entry ], [ %b2, %bb0 ]
84e98501e2Sskc7  %c1 = phi half [ %a1, %entry ], [ %b1, %bb0 ]
85e98501e2Sskc7  %c0 = phi half [ %a0, %entry ], [ %b0, %bb0 ]
86e98501e2Sskc7
87e98501e2Sskc7  %o0 = insertelement <4 x half> undef, half %c0, i64 0
88e98501e2Sskc7  %o1 = insertelement <4 x half> %o0, half %c1, i64 1
89e98501e2Sskc7  %o2 = insertelement <4 x half> %o1, half %c2, i64 2
90e98501e2Sskc7  %o3 = insertelement <4 x half> %o2, half %c3, i64 3
91e98501e2Sskc7  ret <4 x half> %o3
92e98501e2Sskc7}
93