1; RUN: opt < %s -basicaa -slp-vectorizer -S -mtriple=i386-apple-macosx10.8.0 -mcpu=corei7-avx | FileCheck %s
2
3target datalayout = "e-p:32:32:32-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:32:64-f32:32:32-f64:32:64-v64:64:64-v128:128:128-a0:0:64-f80:128:128-n8:16:32-S128"
4target triple = "i386-apple-macosx10.8.0"
5
6;int test(double *G) {
7;  G[0] = 1+G[5]*4;
8;  G[1] = 6+G[6]*3;
9;  G[2] = 7+G[5]*4;
10;  G[3] = 8+G[6]*4;
11;}
12
13;CHECK-LABEL: @test(
14;CHECK: load <2 x double>
15;CHECK: fadd <2 x double>
16;CHECK: store <2 x double>
17;CHECK: insertelement <2 x double>
18;CHECK: fadd <2 x double>
19;CHECK: store <2 x double>
20;CHECK: ret i32
21
22define i32 @test(double* nocapture %G) {
23entry:
24  %arrayidx = getelementptr inbounds double* %G, i64 5
25  %0 = load double* %arrayidx, align 8
26  %mul = fmul double %0, 4.000000e+00
27  %add = fadd double %mul, 1.000000e+00
28  store double %add, double* %G, align 8
29  %arrayidx2 = getelementptr inbounds double* %G, i64 6
30  %1 = load double* %arrayidx2, align 8
31  %mul3 = fmul double %1, 3.000000e+00
32  %add4 = fadd double %mul3, 6.000000e+00
33  %arrayidx5 = getelementptr inbounds double* %G, i64 1
34  store double %add4, double* %arrayidx5, align 8
35  %add8 = fadd double %mul, 7.000000e+00
36  %arrayidx9 = getelementptr inbounds double* %G, i64 2
37  store double %add8, double* %arrayidx9, align 8
38  %mul11 = fmul double %1, 4.000000e+00
39  %add12 = fadd double %mul11, 8.000000e+00
40  %arrayidx13 = getelementptr inbounds double* %G, i64 3
41  store double %add12, double* %arrayidx13, align 8
42  ret i32 undef
43}
44
45;int foo(double *A, int n) {
46;  A[0] = A[0] * 7.9 * n + 6.0;
47;  A[1] = A[1] * 7.7 * n + 2.0;
48;  A[2] = A[2] * 7.6 * n + 3.0;
49;  A[3] = A[3] * 7.4 * n + 4.0;
50;}
51;CHECK-LABEL: @foo(
52;CHECK: insertelement <2 x double>
53;CHECK: insertelement <2 x double>
54;CHECK-NOT: insertelement <2 x double>
55;CHECK: ret
56define i32 @foo(double* nocapture %A, i32 %n) {
57entry:
58  %0 = load double* %A, align 8
59  %mul = fmul double %0, 7.900000e+00
60  %conv = sitofp i32 %n to double
61  %mul1 = fmul double %conv, %mul
62  %add = fadd double %mul1, 6.000000e+00
63  store double %add, double* %A, align 8
64  %arrayidx3 = getelementptr inbounds double* %A, i64 1
65  %1 = load double* %arrayidx3, align 8
66  %mul4 = fmul double %1, 7.700000e+00
67  %mul6 = fmul double %conv, %mul4
68  %add7 = fadd double %mul6, 2.000000e+00
69  store double %add7, double* %arrayidx3, align 8
70  %arrayidx9 = getelementptr inbounds double* %A, i64 2
71  %2 = load double* %arrayidx9, align 8
72  %mul10 = fmul double %2, 7.600000e+00
73  %mul12 = fmul double %conv, %mul10
74  %add13 = fadd double %mul12, 3.000000e+00
75  store double %add13, double* %arrayidx9, align 8
76  %arrayidx15 = getelementptr inbounds double* %A, i64 3
77  %3 = load double* %arrayidx15, align 8
78  %mul16 = fmul double %3, 7.400000e+00
79  %mul18 = fmul double %conv, %mul16
80  %add19 = fadd double %mul18, 4.000000e+00
81  store double %add19, double* %arrayidx15, align 8
82  ret i32 undef
83}
84
85; int test2(double *G, int k) {
86;   if (k) {
87;     G[0] = 1+G[5]*4;
88;     G[1] = 6+G[6]*3;
89;   } else {
90;     G[2] = 7+G[5]*4;
91;     G[3] = 8+G[6]*3;
92;   }
93; }
94
95; We can't merge the gather sequences because one does not dominate the other.
96; CHECK: test2
97; CHECK: insertelement
98; CHECK: insertelement
99; CHECK: insertelement
100; CHECK: insertelement
101; CHECK: ret
102define i32 @test2(double* nocapture %G, i32 %k) {
103  %1 = icmp eq i32 %k, 0
104  %2 = getelementptr inbounds double* %G, i64 5
105  %3 = load double* %2, align 8
106  %4 = fmul double %3, 4.000000e+00
107  br i1 %1, label %12, label %5
108
109; <label>:5                                       ; preds = %0
110  %6 = fadd double %4, 1.000000e+00
111  store double %6, double* %G, align 8
112  %7 = getelementptr inbounds double* %G, i64 6
113  %8 = load double* %7, align 8
114  %9 = fmul double %8, 3.000000e+00
115  %10 = fadd double %9, 6.000000e+00
116  %11 = getelementptr inbounds double* %G, i64 1
117  store double %10, double* %11, align 8
118  br label %20
119
120; <label>:12                                      ; preds = %0
121  %13 = fadd double %4, 7.000000e+00
122  %14 = getelementptr inbounds double* %G, i64 2
123  store double %13, double* %14, align 8
124  %15 = getelementptr inbounds double* %G, i64 6
125  %16 = load double* %15, align 8
126  %17 = fmul double %16, 3.000000e+00
127  %18 = fadd double %17, 8.000000e+00
128  %19 = getelementptr inbounds double* %G, i64 3
129  store double %18, double* %19, align 8
130  br label %20
131
132; <label>:20                                      ; preds = %12, %5
133  ret i32 undef
134}
135
136
137;int foo(double *A, int n) {
138;  A[0] = A[0] * 7.9 * n + 6.0;
139;  A[1] = A[1] * 7.9 * n + 6.0;
140;  A[2] = A[2] * 7.9 * n + 6.0;
141;  A[3] = A[3] * 7.9 * n + 6.0;
142;}
143;CHECK-LABEL: @foo4(
144;CHECK: insertelement <2 x double>
145;CHECK: insertelement <2 x double>
146;CHECK-NOT: insertelement <2 x double>
147;CHECK: ret
148define i32 @foo4(double* nocapture %A, i32 %n) {
149entry:
150  %0 = load double* %A, align 8
151  %mul = fmul double %0, 7.900000e+00
152  %conv = sitofp i32 %n to double
153  %mul1 = fmul double %conv, %mul
154  %add = fadd double %mul1, 6.000000e+00
155  store double %add, double* %A, align 8
156  %arrayidx3 = getelementptr inbounds double* %A, i64 1
157  %1 = load double* %arrayidx3, align 8
158  %mul4 = fmul double %1, 7.900000e+00
159  %mul6 = fmul double %conv, %mul4
160  %add7 = fadd double %mul6, 6.000000e+00
161  store double %add7, double* %arrayidx3, align 8
162  %arrayidx9 = getelementptr inbounds double* %A, i64 2
163  %2 = load double* %arrayidx9, align 8
164  %mul10 = fmul double %2, 7.900000e+00
165  %mul12 = fmul double %conv, %mul10
166  %add13 = fadd double %mul12, 6.000000e+00
167  store double %add13, double* %arrayidx9, align 8
168  %arrayidx15 = getelementptr inbounds double* %A, i64 3
169  %3 = load double* %arrayidx15, align 8
170  %mul16 = fmul double %3, 7.900000e+00
171  %mul18 = fmul double %conv, %mul16
172  %add19 = fadd double %mul18, 6.000000e+00
173  store double %add19, double* %arrayidx15, align 8
174  ret i32 undef
175}
176
177;int partial_mrg(double *A, int n) {
178;  A[0] = A[0] * n;
179;  A[1] = A[1] * n;
180;  if (n < 4) return 0;
181;  A[2] = A[2] * n;
182;  A[3] = A[3] * (n+4);
183;}
184;CHECK-LABEL: @partial_mrg(
185;CHECK: insertelement <2 x double>
186;CHECK: insertelement <2 x double>
187;CHECK: insertelement <2 x double>
188;CHECK-NOT: insertelement <2 x double>
189;CHECK: ret
190define i32 @partial_mrg(double* nocapture %A, i32 %n) {
191entry:
192  %0 = load double* %A, align 8
193  %conv = sitofp i32 %n to double
194  %mul = fmul double %conv, %0
195  store double %mul, double* %A, align 8
196  %arrayidx2 = getelementptr inbounds double* %A, i64 1
197  %1 = load double* %arrayidx2, align 8
198  %mul4 = fmul double %conv, %1
199  store double %mul4, double* %arrayidx2, align 8
200  %cmp = icmp slt i32 %n, 4
201  br i1 %cmp, label %return, label %if.end
202
203if.end:                                           ; preds = %entry
204  %arrayidx7 = getelementptr inbounds double* %A, i64 2
205  %2 = load double* %arrayidx7, align 8
206  %mul9 = fmul double %conv, %2
207  store double %mul9, double* %arrayidx7, align 8
208  %arrayidx11 = getelementptr inbounds double* %A, i64 3
209  %3 = load double* %arrayidx11, align 8
210  %add = add nsw i32 %n, 4
211  %conv12 = sitofp i32 %add to double
212  %mul13 = fmul double %conv12, %3
213  store double %mul13, double* %arrayidx11, align 8
214  br label %return
215
216return:                                           ; preds = %entry, %if.end
217  ret i32 0
218}
219
220