• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
2; RUN: opt -S -mcpu=swift -mtriple=thumbv7-apple-ios -basic-aa -slp-vectorizer < %s | FileCheck %s
3
4target datalayout = "e-p:32:32:32-i1:8:32-i8:8:32-i16:16:32-i32:32:32-i64:32:64-f32:32:32-f64:32:64-v64:32:64-v128:32:128-a0:0:32-n32-S32"
5
6%class.Complex = type { double, double }
7
8; Code like this is the result of SROA. Make sure we don't vectorize this
9; because the scalar version of the shl/or are handled by the
10; backend and disappear, the vectorized code stays.
11
12define void @SROAed(%class.Complex* noalias nocapture sret(%class.Complex) %agg.result, [4 x i32] %a.coerce, [4 x i32] %b.coerce) {
13; CHECK-LABEL: @SROAed(
14; CHECK-NEXT:  entry:
15; CHECK-NEXT:    [[A_COERCE_FCA_0_EXTRACT:%.*]] = extractvalue [4 x i32] [[A_COERCE:%.*]], 0
16; CHECK-NEXT:    [[A_SROA_0_0_INSERT_EXT:%.*]] = zext i32 [[A_COERCE_FCA_0_EXTRACT]] to i64
17; CHECK-NEXT:    [[A_COERCE_FCA_1_EXTRACT:%.*]] = extractvalue [4 x i32] [[A_COERCE]], 1
18; CHECK-NEXT:    [[A_SROA_0_4_INSERT_EXT:%.*]] = zext i32 [[A_COERCE_FCA_1_EXTRACT]] to i64
19; CHECK-NEXT:    [[A_SROA_0_4_INSERT_SHIFT:%.*]] = shl nuw i64 [[A_SROA_0_4_INSERT_EXT]], 32
20; CHECK-NEXT:    [[A_SROA_0_4_INSERT_INSERT:%.*]] = or i64 [[A_SROA_0_4_INSERT_SHIFT]], [[A_SROA_0_0_INSERT_EXT]]
21; CHECK-NEXT:    [[TMP0:%.*]] = bitcast i64 [[A_SROA_0_4_INSERT_INSERT]] to double
22; CHECK-NEXT:    [[A_COERCE_FCA_2_EXTRACT:%.*]] = extractvalue [4 x i32] [[A_COERCE]], 2
23; CHECK-NEXT:    [[A_SROA_3_8_INSERT_EXT:%.*]] = zext i32 [[A_COERCE_FCA_2_EXTRACT]] to i64
24; CHECK-NEXT:    [[A_COERCE_FCA_3_EXTRACT:%.*]] = extractvalue [4 x i32] [[A_COERCE]], 3
25; CHECK-NEXT:    [[A_SROA_3_12_INSERT_EXT:%.*]] = zext i32 [[A_COERCE_FCA_3_EXTRACT]] to i64
26; CHECK-NEXT:    [[A_SROA_3_12_INSERT_SHIFT:%.*]] = shl nuw i64 [[A_SROA_3_12_INSERT_EXT]], 32
27; CHECK-NEXT:    [[A_SROA_3_12_INSERT_INSERT:%.*]] = or i64 [[A_SROA_3_12_INSERT_SHIFT]], [[A_SROA_3_8_INSERT_EXT]]
28; CHECK-NEXT:    [[TMP1:%.*]] = bitcast i64 [[A_SROA_3_12_INSERT_INSERT]] to double
29; CHECK-NEXT:    [[B_COERCE_FCA_0_EXTRACT:%.*]] = extractvalue [4 x i32] [[B_COERCE:%.*]], 0
30; CHECK-NEXT:    [[B_SROA_0_0_INSERT_EXT:%.*]] = zext i32 [[B_COERCE_FCA_0_EXTRACT]] to i64
31; CHECK-NEXT:    [[B_COERCE_FCA_1_EXTRACT:%.*]] = extractvalue [4 x i32] [[B_COERCE]], 1
32; CHECK-NEXT:    [[B_SROA_0_4_INSERT_EXT:%.*]] = zext i32 [[B_COERCE_FCA_1_EXTRACT]] to i64
33; CHECK-NEXT:    [[B_SROA_0_4_INSERT_SHIFT:%.*]] = shl nuw i64 [[B_SROA_0_4_INSERT_EXT]], 32
34; CHECK-NEXT:    [[B_SROA_0_4_INSERT_INSERT:%.*]] = or i64 [[B_SROA_0_4_INSERT_SHIFT]], [[B_SROA_0_0_INSERT_EXT]]
35; CHECK-NEXT:    [[TMP2:%.*]] = bitcast i64 [[B_SROA_0_4_INSERT_INSERT]] to double
36; CHECK-NEXT:    [[B_COERCE_FCA_2_EXTRACT:%.*]] = extractvalue [4 x i32] [[B_COERCE]], 2
37; CHECK-NEXT:    [[B_SROA_3_8_INSERT_EXT:%.*]] = zext i32 [[B_COERCE_FCA_2_EXTRACT]] to i64
38; CHECK-NEXT:    [[B_COERCE_FCA_3_EXTRACT:%.*]] = extractvalue [4 x i32] [[B_COERCE]], 3
39; CHECK-NEXT:    [[B_SROA_3_12_INSERT_EXT:%.*]] = zext i32 [[B_COERCE_FCA_3_EXTRACT]] to i64
40; CHECK-NEXT:    [[B_SROA_3_12_INSERT_SHIFT:%.*]] = shl nuw i64 [[B_SROA_3_12_INSERT_EXT]], 32
41; CHECK-NEXT:    [[B_SROA_3_12_INSERT_INSERT:%.*]] = or i64 [[B_SROA_3_12_INSERT_SHIFT]], [[B_SROA_3_8_INSERT_EXT]]
42; CHECK-NEXT:    [[TMP3:%.*]] = bitcast i64 [[B_SROA_3_12_INSERT_INSERT]] to double
43; CHECK-NEXT:    [[ADD:%.*]] = fadd double [[TMP0]], [[TMP2]]
44; CHECK-NEXT:    [[ADD3:%.*]] = fadd double [[TMP1]], [[TMP3]]
45; CHECK-NEXT:    [[RE_I_I:%.*]] = getelementptr inbounds [[CLASS_COMPLEX:%.*]], %class.Complex* [[AGG_RESULT:%.*]], i32 0, i32 0
46; CHECK-NEXT:    store double [[ADD]], double* [[RE_I_I]], align 4
47; CHECK-NEXT:    [[IM_I_I:%.*]] = getelementptr inbounds [[CLASS_COMPLEX]], %class.Complex* [[AGG_RESULT]], i32 0, i32 1
48; CHECK-NEXT:    store double [[ADD3]], double* [[IM_I_I]], align 4
49; CHECK-NEXT:    ret void
50;
51entry:
52  %a.coerce.fca.0.extract = extractvalue [4 x i32] %a.coerce, 0
53  %a.sroa.0.0.insert.ext = zext i32 %a.coerce.fca.0.extract to i64
54  %a.coerce.fca.1.extract = extractvalue [4 x i32] %a.coerce, 1
55  %a.sroa.0.4.insert.ext = zext i32 %a.coerce.fca.1.extract to i64
56  %a.sroa.0.4.insert.shift = shl nuw i64 %a.sroa.0.4.insert.ext, 32
57  %a.sroa.0.4.insert.insert = or i64 %a.sroa.0.4.insert.shift, %a.sroa.0.0.insert.ext
58  %0 = bitcast i64 %a.sroa.0.4.insert.insert to double
59  %a.coerce.fca.2.extract = extractvalue [4 x i32] %a.coerce, 2
60  %a.sroa.3.8.insert.ext = zext i32 %a.coerce.fca.2.extract to i64
61  %a.coerce.fca.3.extract = extractvalue [4 x i32] %a.coerce, 3
62  %a.sroa.3.12.insert.ext = zext i32 %a.coerce.fca.3.extract to i64
63  %a.sroa.3.12.insert.shift = shl nuw i64 %a.sroa.3.12.insert.ext, 32
64  %a.sroa.3.12.insert.insert = or i64 %a.sroa.3.12.insert.shift, %a.sroa.3.8.insert.ext
65  %1 = bitcast i64 %a.sroa.3.12.insert.insert to double
66  %b.coerce.fca.0.extract = extractvalue [4 x i32] %b.coerce, 0
67  %b.sroa.0.0.insert.ext = zext i32 %b.coerce.fca.0.extract to i64
68  %b.coerce.fca.1.extract = extractvalue [4 x i32] %b.coerce, 1
69  %b.sroa.0.4.insert.ext = zext i32 %b.coerce.fca.1.extract to i64
70  %b.sroa.0.4.insert.shift = shl nuw i64 %b.sroa.0.4.insert.ext, 32
71  %b.sroa.0.4.insert.insert = or i64 %b.sroa.0.4.insert.shift, %b.sroa.0.0.insert.ext
72  %2 = bitcast i64 %b.sroa.0.4.insert.insert to double
73  %b.coerce.fca.2.extract = extractvalue [4 x i32] %b.coerce, 2
74  %b.sroa.3.8.insert.ext = zext i32 %b.coerce.fca.2.extract to i64
75  %b.coerce.fca.3.extract = extractvalue [4 x i32] %b.coerce, 3
76  %b.sroa.3.12.insert.ext = zext i32 %b.coerce.fca.3.extract to i64
77  %b.sroa.3.12.insert.shift = shl nuw i64 %b.sroa.3.12.insert.ext, 32
78  %b.sroa.3.12.insert.insert = or i64 %b.sroa.3.12.insert.shift, %b.sroa.3.8.insert.ext
79  %3 = bitcast i64 %b.sroa.3.12.insert.insert to double
80  %add = fadd double %0, %2
81  %add3 = fadd double %1, %3
82  %re.i.i = getelementptr inbounds %class.Complex, %class.Complex* %agg.result, i32 0, i32 0
83  store double %add, double* %re.i.i, align 4
84  %im.i.i = getelementptr inbounds %class.Complex, %class.Complex* %agg.result, i32 0, i32 1
85  store double %add3, double* %im.i.i, align 4
86  ret void
87}
88