| blob | 849c1d7c192dce514ab33b7486ecd90c0ca17c87 |
1 ; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
2 ; RUN: opt < %s -passes=slp-vectorizer -S | FileCheck %s
3 ; RUN: opt < %s -passes=slp-vectorizer -S -mattr=+avx | FileCheck %s
5 target datalayout = "e-m:o-i64:64-f80:128-n8:16:32:64-S128"
6 target triple = "x86_64-apple-macosx10.10.0"
8 define void @testfunc(ptr nocapture %dest, ptr nocapture readonly %src) {
9 ; CHECK-LABEL: @testfunc(
10 ; CHECK-NEXT: entry:
11 ; CHECK-NEXT: br label [[FOR_BODY:%.*]]
12 ; CHECK: for.body:
13 ; CHECK-NEXT: [[INDVARS_IV:%.*]] = phi i64 [ 0, [[ENTRY:%.*]] ], [ [[INDVARS_IV_NEXT:%.*]], [[FOR_BODY]] ]
14 ; CHECK-NEXT: [[ACC1_056:%.*]] = phi float [ 0.000000e+00, [[ENTRY]] ], [ [[ADD13:%.*]], [[FOR_BODY]] ]
15 ; CHECK-NEXT: [[TMP0:%.*]] = phi <2 x float> [ zeroinitializer, [[ENTRY]] ], [ [[TMP18:%.*]], [[FOR_BODY]] ]
16 ; CHECK-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds float, ptr [[SRC:%.*]], i64 [[INDVARS_IV]]
17 ; CHECK-NEXT: [[TMP1:%.*]] = load float, ptr [[ARRAYIDX]], align 4
18 ; CHECK-NEXT: [[INDVARS_IV_NEXT]] = add nuw nsw i64 [[INDVARS_IV]], 1
19 ; CHECK-NEXT: [[ARRAYIDX2:%.*]] = getelementptr inbounds float, ptr [[DEST:%.*]], i64 [[INDVARS_IV]]
20 ; CHECK-NEXT: store float [[ACC1_056]], ptr [[ARRAYIDX2]], align 4
21 ; CHECK-NEXT: [[TMP2:%.*]] = insertelement <2 x float> poison, float [[TMP1]], i32 0
22 ; CHECK-NEXT: [[SHUFFLE:%.*]] = shufflevector <2 x float> [[TMP2]], <2 x float> poison, <2 x i32> zeroinitializer
23 ; CHECK-NEXT: [[TMP3:%.*]] = fadd <2 x float> [[TMP0]], [[SHUFFLE]]
24 ; CHECK-NEXT: [[SHUFFLE1:%.*]] = shufflevector <2 x float> [[TMP3]], <2 x float> poison, <2 x i32> <i32 1, i32 0>
25 ; CHECK-NEXT: [[TMP4:%.*]] = fmul <2 x float> [[TMP0]], zeroinitializer
26 ; CHECK-NEXT: [[TMP5:%.*]] = fadd <2 x float> [[TMP4]], [[SHUFFLE1]]
27 ; CHECK-NEXT: [[TMP6:%.*]] = fcmp olt <2 x float> [[TMP5]], <float 1.000000e+00, float 1.000000e+00>
28 ; CHECK-NEXT: [[TMP7:%.*]] = select <2 x i1> [[TMP6]], <2 x float> [[TMP5]], <2 x float> <float 1.000000e+00, float 1.000000e+00>
29 ; CHECK-NEXT: [[TMP8:%.*]] = fcmp olt <2 x float> [[TMP7]], <float -1.000000e+00, float -1.000000e+00>
30 ; CHECK-NEXT: [[TMP9:%.*]] = fmul <2 x float> [[TMP7]], zeroinitializer
31 ; CHECK-NEXT: [[TMP10:%.*]] = select <2 x i1> [[TMP8]], <2 x float> <float -0.000000e+00, float -0.000000e+00>, <2 x float> [[TMP9]]
32 ; CHECK-NEXT: [[TMP11:%.*]] = extractelement <2 x float> [[TMP10]], i32 0
33 ; CHECK-NEXT: [[TMP12:%.*]] = extractelement <2 x float> [[TMP10]], i32 1
34 ; CHECK-NEXT: [[ADD13]] = fadd float [[TMP11]], [[TMP12]]
35 ; CHECK-NEXT: [[TMP13:%.*]] = shufflevector <2 x float> [[TMP10]], <2 x float> poison, <2 x i32> <i32 1, i32 poison>
36 ; CHECK-NEXT: [[TMP14:%.*]] = insertelement <2 x float> [[TMP13]], float [[ADD13]], i32 1
37 ; CHECK-NEXT: [[TMP15:%.*]] = fcmp olt <2 x float> [[TMP14]], <float 1.000000e+00, float 1.000000e+00>
38 ; CHECK-NEXT: [[TMP16:%.*]] = select <2 x i1> [[TMP15]], <2 x float> [[TMP14]], <2 x float> <float 1.000000e+00, float 1.000000e+00>
39 ; CHECK-NEXT: [[TMP17:%.*]] = fcmp olt <2 x float> [[TMP16]], <float -1.000000e+00, float -1.000000e+00>
40 ; CHECK-NEXT: [[TMP18]] = select <2 x i1> [[TMP17]], <2 x float> <float -1.000000e+00, float -1.000000e+00>, <2 x float> [[TMP16]]
41 ; CHECK-NEXT: [[EXITCOND:%.*]] = icmp eq i64 [[INDVARS_IV_NEXT]], 32
42 ; CHECK-NEXT: br i1 [[EXITCOND]], label [[FOR_END:%.*]], label [[FOR_BODY]]
43 ; CHECK: for.end:
44 ; CHECK-NEXT: ret void
45 ;
46 entry:
47 br label %for.body
49 for.body:
50 %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ]
51 %acc1.056 = phi float [ 0.000000e+00, %entry ], [ %add13, %for.body ]
52 %s1.055 = phi float [ 0.000000e+00, %entry ], [ %cond.i40, %for.body ]
53 %s0.054 = phi float [ 0.000000e+00, %entry ], [ %cond.i44, %for.body ]
54 %arrayidx = getelementptr inbounds float, ptr %src, i64 %indvars.iv
55 %0 = load float, ptr %arrayidx, align 4
56 %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
57 %arrayidx2 = getelementptr inbounds float, ptr %dest, i64 %indvars.iv
58 store float %acc1.056, ptr %arrayidx2, align 4
59 %add = fadd float %s0.054, %0
60 %add3 = fadd float %s1.055, %0
61 %mul = fmul float %s0.054, 0.000000e+00
62 %add4 = fadd float %mul, %add3
63 %mul5 = fmul float %s1.055, 0.000000e+00
64 %add6 = fadd float %mul5, %add
65 %cmp.i = fcmp olt float %add6, 1.000000e+00
66 %cond.i = select i1 %cmp.i, float %add6, float 1.000000e+00
67 %cmp.i51 = fcmp olt float %cond.i, -1.000000e+00
68 %cmp.i49 = fcmp olt float %add4, 1.000000e+00
69 %cond.i50 = select i1 %cmp.i49, float %add4, float 1.000000e+00
70 %cmp.i47 = fcmp olt float %cond.i50, -1.000000e+00
71 %cond.i.op = fmul float %cond.i, 0.000000e+00
72 %mul10 = select i1 %cmp.i51, float -0.000000e+00, float %cond.i.op
73 %cond.i50.op = fmul float %cond.i50, 0.000000e+00
74 %mul11 = select i1 %cmp.i47, float -0.000000e+00, float %cond.i50.op
75 %add13 = fadd float %mul10, %mul11
77 ; The SLPVectorizer crashed in vectorizeChainsInBlock() because it tried
78 ; to access the second operand of the following cmp after the cmp itself
79 ; was already vectorized and deleted.
80 %cmp.i45 = fcmp olt float %add13, 1.000000e+00
82 %cond.i46 = select i1 %cmp.i45, float %add13, float 1.000000e+00
83 %cmp.i43 = fcmp olt float %cond.i46, -1.000000e+00
84 %cond.i44 = select i1 %cmp.i43, float -1.000000e+00, float %cond.i46
85 %cmp.i41 = fcmp olt float %mul11, 1.000000e+00
86 %cond.i42 = select i1 %cmp.i41, float %mul11, float 1.000000e+00
87 %cmp.i39 = fcmp olt float %cond.i42, -1.000000e+00
88 %cond.i40 = select i1 %cmp.i39, float -1.000000e+00, float %cond.i42
89 %exitcond = icmp eq i64 %indvars.iv.next, 32
90 br i1 %exitcond, label %for.end, label %for.body
92 for.end:
93 ret void
94 }