1 ; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 5
2 ; RUN: opt -mtriple=s390x-unknown-linux -mcpu=z15 -passes=slp-vectorizer %s -S -o - \
5 ; Test vectorization and reassociation of fmul operations. If the loads can
6 ; be vectorized, cases of fewer operands are also profitable to vectorize.
8 define double @fmul_double_4_factors_seq(ptr nocapture noundef readonly %x) {
9 ; CHECK-LABEL: define double @fmul_double_4_factors_seq(
10 ; CHECK-SAME: ptr nocapture noundef readonly [[X:%.*]]) #[[ATTR0:[0-9]+]] {
11 ; CHECK-NEXT: [[ENTRY:.*:]]
12 ; CHECK-NEXT: [[TMP0:%.*]] = load <4 x double>, ptr [[X]], align 8
13 ; CHECK-NEXT: [[TMP1:%.*]] = call reassoc nsz arcp contract afn double @llvm.vector.reduce.fmul.v4f64(double 1.000000e+00, <4 x double> [[TMP0]])
14 ; CHECK-NEXT: ret double [[TMP1]]
17 %0 = load double, ptr %x, align 8
18 %arrayidx1 = getelementptr inbounds double, ptr %x, i64 1
19 %1 = load double, ptr %arrayidx1, align 8
20 %mul = fmul reassoc nsz arcp contract afn double %1, %0
21 %arrayidx2 = getelementptr inbounds double, ptr %x, i64 2
22 %2 = load double, ptr %arrayidx2, align 8
23 %mul3 = fmul reassoc nsz arcp contract afn double %mul, %2
24 %arrayidx4 = getelementptr inbounds double, ptr %x, i64 3
25 %3 = load double, ptr %arrayidx4, align 8
26 %mul5 = fmul reassoc nsz arcp contract afn double %mul3, %3
30 define double @fmul_double_8_factors_nonseq(ptr nocapture noundef readonly %x) {
31 ; CHECK-LABEL: define double @fmul_double_8_factors_nonseq(
32 ; CHECK-SAME: ptr nocapture noundef readonly [[X:%.*]]) #[[ATTR0]] {
33 ; CHECK-NEXT: [[ENTRY:.*:]]
34 ; CHECK-NEXT: [[TMP0:%.*]] = load double, ptr [[X]], align 8
35 ; CHECK-NEXT: [[ARRAYIDX1:%.*]] = getelementptr inbounds double, ptr [[X]], i64 2
36 ; CHECK-NEXT: [[TMP1:%.*]] = load double, ptr [[ARRAYIDX1]], align 8
37 ; CHECK-NEXT: [[ARRAYIDX2:%.*]] = getelementptr inbounds double, ptr [[X]], i64 4
38 ; CHECK-NEXT: [[TMP2:%.*]] = load double, ptr [[ARRAYIDX2]], align 8
39 ; CHECK-NEXT: [[ARRAYIDX4:%.*]] = getelementptr inbounds double, ptr [[X]], i64 6
40 ; CHECK-NEXT: [[TMP3:%.*]] = load double, ptr [[ARRAYIDX4]], align 8
41 ; CHECK-NEXT: [[ARRAYIDX6:%.*]] = getelementptr inbounds double, ptr [[X]], i64 8
42 ; CHECK-NEXT: [[TMP4:%.*]] = load double, ptr [[ARRAYIDX6]], align 8
43 ; CHECK-NEXT: [[ARRAYIDX8:%.*]] = getelementptr inbounds double, ptr [[X]], i64 10
44 ; CHECK-NEXT: [[TMP5:%.*]] = load double, ptr [[ARRAYIDX8]], align 8
45 ; CHECK-NEXT: [[ARRAYIDX10:%.*]] = getelementptr inbounds double, ptr [[X]], i64 12
46 ; CHECK-NEXT: [[TMP6:%.*]] = load double, ptr [[ARRAYIDX10]], align 8
47 ; CHECK-NEXT: [[ARRAYIDX12:%.*]] = getelementptr inbounds double, ptr [[X]], i64 14
48 ; CHECK-NEXT: [[TMP7:%.*]] = load double, ptr [[ARRAYIDX12]], align 8
49 ; CHECK-NEXT: [[TMP8:%.*]] = insertelement <8 x double> poison, double [[TMP1]], i32 0
50 ; CHECK-NEXT: [[TMP9:%.*]] = insertelement <8 x double> [[TMP8]], double [[TMP0]], i32 1
51 ; CHECK-NEXT: [[TMP10:%.*]] = insertelement <8 x double> [[TMP9]], double [[TMP2]], i32 2
52 ; CHECK-NEXT: [[TMP11:%.*]] = insertelement <8 x double> [[TMP10]], double [[TMP3]], i32 3
53 ; CHECK-NEXT: [[TMP12:%.*]] = insertelement <8 x double> [[TMP11]], double [[TMP4]], i32 4
54 ; CHECK-NEXT: [[TMP13:%.*]] = insertelement <8 x double> [[TMP12]], double [[TMP5]], i32 5
55 ; CHECK-NEXT: [[TMP14:%.*]] = insertelement <8 x double> [[TMP13]], double [[TMP6]], i32 6
56 ; CHECK-NEXT: [[TMP15:%.*]] = insertelement <8 x double> [[TMP14]], double [[TMP7]], i32 7
57 ; CHECK-NEXT: [[TMP16:%.*]] = call reassoc nsz arcp contract afn double @llvm.vector.reduce.fmul.v8f64(double 1.000000e+00, <8 x double> [[TMP15]])
58 ; CHECK-NEXT: ret double [[TMP16]]
61 %0 = load double, ptr %x, align 8
62 %arrayidx1 = getelementptr inbounds double, ptr %x, i64 2
63 %1 = load double, ptr %arrayidx1, align 8
64 %mul = fmul reassoc nsz arcp contract afn double %1, %0
65 %arrayidx2 = getelementptr inbounds double, ptr %x, i64 4
66 %2 = load double, ptr %arrayidx2, align 8
67 %mul3 = fmul reassoc nsz arcp contract afn double %mul, %2
68 %arrayidx4 = getelementptr inbounds double, ptr %x, i64 6
69 %3 = load double, ptr %arrayidx4, align 8
70 %mul5 = fmul reassoc nsz arcp contract afn double %mul3, %3
71 %arrayidx6 = getelementptr inbounds double, ptr %x, i64 8
72 %4 = load double, ptr %arrayidx6, align 8
73 %mul7 = fmul reassoc nsz arcp contract afn double %mul5, %4
74 %arrayidx8 = getelementptr inbounds double, ptr %x, i64 10
75 %5 = load double, ptr %arrayidx8, align 8
76 %mul9 = fmul reassoc nsz arcp contract afn double %mul7, %5
77 %arrayidx10 = getelementptr inbounds double, ptr %x, i64 12
78 %6 = load double, ptr %arrayidx10, align 8
79 %mul11 = fmul reassoc nsz arcp contract afn double %mul9, %6
80 %arrayidx12 = getelementptr inbounds double, ptr %x, i64 14
81 %7 = load double, ptr %arrayidx12, align 8
82 %mul13 = fmul reassoc nsz arcp contract afn double %mul11, %7
86 define float @fmul_float_16_factors_nonseq(float noundef %m, ptr nocapture noundef readonly %x) {
87 ; CHECK-LABEL: define float @fmul_float_16_factors_nonseq(
88 ; CHECK-SAME: float noundef [[M:%.*]], ptr nocapture noundef readonly [[X:%.*]]) #[[ATTR0]] {
89 ; CHECK-NEXT: [[ENTRY:.*:]]
90 ; CHECK-NEXT: [[TMP0:%.*]] = load float, ptr [[X]], align 4
91 ; CHECK-NEXT: [[ARRAYIDX1:%.*]] = getelementptr inbounds float, ptr [[X]], i64 2
92 ; CHECK-NEXT: [[TMP1:%.*]] = load float, ptr [[ARRAYIDX1]], align 4
93 ; CHECK-NEXT: [[ARRAYIDX2:%.*]] = getelementptr inbounds float, ptr [[X]], i64 4
94 ; CHECK-NEXT: [[TMP2:%.*]] = load float, ptr [[ARRAYIDX2]], align 4
95 ; CHECK-NEXT: [[ARRAYIDX4:%.*]] = getelementptr inbounds float, ptr [[X]], i64 6
96 ; CHECK-NEXT: [[TMP3:%.*]] = load float, ptr [[ARRAYIDX4]], align 4
97 ; CHECK-NEXT: [[ARRAYIDX6:%.*]] = getelementptr inbounds float, ptr [[X]], i64 8
98 ; CHECK-NEXT: [[TMP4:%.*]] = load float, ptr [[ARRAYIDX6]], align 4
99 ; CHECK-NEXT: [[ARRAYIDX8:%.*]] = getelementptr inbounds float, ptr [[X]], i64 10
100 ; CHECK-NEXT: [[TMP5:%.*]] = load float, ptr [[ARRAYIDX8]], align 4
101 ; CHECK-NEXT: [[ARRAYIDX10:%.*]] = getelementptr inbounds float, ptr [[X]], i64 12
102 ; CHECK-NEXT: [[TMP6:%.*]] = load float, ptr [[ARRAYIDX10]], align 4
103 ; CHECK-NEXT: [[ARRAYIDX12:%.*]] = getelementptr inbounds float, ptr [[X]], i64 14
104 ; CHECK-NEXT: [[TMP7:%.*]] = load float, ptr [[ARRAYIDX12]], align 4
105 ; CHECK-NEXT: [[ARRAYIDX14:%.*]] = getelementptr inbounds float, ptr [[X]], i64 16
106 ; CHECK-NEXT: [[TMP8:%.*]] = load float, ptr [[ARRAYIDX14]], align 4
107 ; CHECK-NEXT: [[ARRAYIDX16:%.*]] = getelementptr inbounds float, ptr [[X]], i64 18
108 ; CHECK-NEXT: [[TMP9:%.*]] = load float, ptr [[ARRAYIDX16]], align 4
109 ; CHECK-NEXT: [[ARRAYIDX18:%.*]] = getelementptr inbounds float, ptr [[X]], i64 20
110 ; CHECK-NEXT: [[TMP10:%.*]] = load float, ptr [[ARRAYIDX18]], align 4
111 ; CHECK-NEXT: [[ARRAYIDX20:%.*]] = getelementptr inbounds float, ptr [[X]], i64 22
112 ; CHECK-NEXT: [[TMP11:%.*]] = load float, ptr [[ARRAYIDX20]], align 4
113 ; CHECK-NEXT: [[ARRAYIDX22:%.*]] = getelementptr inbounds float, ptr [[X]], i64 24
114 ; CHECK-NEXT: [[TMP12:%.*]] = load float, ptr [[ARRAYIDX22]], align 4
115 ; CHECK-NEXT: [[ARRAYIDX24:%.*]] = getelementptr inbounds float, ptr [[X]], i64 26
116 ; CHECK-NEXT: [[TMP13:%.*]] = load float, ptr [[ARRAYIDX24]], align 4
117 ; CHECK-NEXT: [[ARRAYIDX26:%.*]] = getelementptr inbounds float, ptr [[X]], i64 28
118 ; CHECK-NEXT: [[TMP14:%.*]] = load float, ptr [[ARRAYIDX26]], align 4
119 ; CHECK-NEXT: [[ARRAYIDX28:%.*]] = getelementptr inbounds float, ptr [[X]], i64 30
120 ; CHECK-NEXT: [[TMP15:%.*]] = load float, ptr [[ARRAYIDX28]], align 4
121 ; CHECK-NEXT: [[TMP16:%.*]] = insertelement <16 x float> poison, float [[TMP1]], i32 0
122 ; CHECK-NEXT: [[TMP17:%.*]] = insertelement <16 x float> [[TMP16]], float [[TMP0]], i32 1
123 ; CHECK-NEXT: [[TMP18:%.*]] = insertelement <16 x float> [[TMP17]], float [[TMP2]], i32 2
124 ; CHECK-NEXT: [[TMP19:%.*]] = insertelement <16 x float> [[TMP18]], float [[TMP3]], i32 3
125 ; CHECK-NEXT: [[TMP20:%.*]] = insertelement <16 x float> [[TMP19]], float [[TMP4]], i32 4
126 ; CHECK-NEXT: [[TMP21:%.*]] = insertelement <16 x float> [[TMP20]], float [[TMP5]], i32 5
127 ; CHECK-NEXT: [[TMP22:%.*]] = insertelement <16 x float> [[TMP21]], float [[TMP6]], i32 6
128 ; CHECK-NEXT: [[TMP23:%.*]] = insertelement <16 x float> [[TMP22]], float [[TMP7]], i32 7
129 ; CHECK-NEXT: [[TMP24:%.*]] = insertelement <16 x float> [[TMP23]], float [[TMP8]], i32 8
130 ; CHECK-NEXT: [[TMP25:%.*]] = insertelement <16 x float> [[TMP24]], float [[TMP9]], i32 9
131 ; CHECK-NEXT: [[TMP26:%.*]] = insertelement <16 x float> [[TMP25]], float [[TMP10]], i32 10
132 ; CHECK-NEXT: [[TMP27:%.*]] = insertelement <16 x float> [[TMP26]], float [[TMP11]], i32 11
133 ; CHECK-NEXT: [[TMP28:%.*]] = insertelement <16 x float> [[TMP27]], float [[TMP12]], i32 12
134 ; CHECK-NEXT: [[TMP29:%.*]] = insertelement <16 x float> [[TMP28]], float [[TMP13]], i32 13
135 ; CHECK-NEXT: [[TMP30:%.*]] = insertelement <16 x float> [[TMP29]], float [[TMP14]], i32 14
136 ; CHECK-NEXT: [[TMP31:%.*]] = insertelement <16 x float> [[TMP30]], float [[TMP15]], i32 15
137 ; CHECK-NEXT: [[TMP32:%.*]] = call reassoc nsz arcp contract afn float @llvm.vector.reduce.fmul.v16f32(float 1.000000e+00, <16 x float> [[TMP31]])
138 ; CHECK-NEXT: ret float [[TMP32]]
141 %0 = load float, ptr %x, align 4
142 %arrayidx1 = getelementptr inbounds float, ptr %x, i64 2
143 %1 = load float, ptr %arrayidx1, align 4
144 %mul = fmul reassoc nsz arcp contract afn float %1, %0
145 %arrayidx2 = getelementptr inbounds float, ptr %x, i64 4
146 %2 = load float, ptr %arrayidx2, align 4
147 %mul3 = fmul reassoc nsz arcp contract afn float %mul, %2
148 %arrayidx4 = getelementptr inbounds float, ptr %x, i64 6
149 %3 = load float, ptr %arrayidx4, align 4
150 %mul5 = fmul reassoc nsz arcp contract afn float %mul3, %3
151 %arrayidx6 = getelementptr inbounds float, ptr %x, i64 8
152 %4 = load float, ptr %arrayidx6, align 4
153 %mul7 = fmul reassoc nsz arcp contract afn float %mul5, %4
154 %arrayidx8 = getelementptr inbounds float, ptr %x, i64 10
155 %5 = load float, ptr %arrayidx8, align 4
156 %mul9 = fmul reassoc nsz arcp contract afn float %mul7, %5
157 %arrayidx10 = getelementptr inbounds float, ptr %x, i64 12
158 %6 = load float, ptr %arrayidx10, align 4
159 %mul11 = fmul reassoc nsz arcp contract afn float %mul9, %6
160 %arrayidx12 = getelementptr inbounds float, ptr %x, i64 14
161 %7 = load float, ptr %arrayidx12, align 4
162 %mul13 = fmul reassoc nsz arcp contract afn float %mul11, %7
163 %arrayidx14 = getelementptr inbounds float, ptr %x, i64 16
164 %8 = load float, ptr %arrayidx14, align 4
165 %mul15 = fmul reassoc nsz arcp contract afn float %mul13, %8
166 %arrayidx16 = getelementptr inbounds float, ptr %x, i64 18
167 %9 = load float, ptr %arrayidx16, align 4
168 %mul17 = fmul reassoc nsz arcp contract afn float %mul15, %9
169 %arrayidx18 = getelementptr inbounds float, ptr %x, i64 20
170 %10 = load float, ptr %arrayidx18, align 4
171 %mul19 = fmul reassoc nsz arcp contract afn float %mul17, %10
172 %arrayidx20 = getelementptr inbounds float, ptr %x, i64 22
173 %11 = load float, ptr %arrayidx20, align 4
174 %mul21 = fmul reassoc nsz arcp contract afn float %mul19, %11
175 %arrayidx22 = getelementptr inbounds float, ptr %x, i64 24
176 %12 = load float, ptr %arrayidx22, align 4
177 %mul23 = fmul reassoc nsz arcp contract afn float %mul21, %12
178 %arrayidx24 = getelementptr inbounds float, ptr %x, i64 26
179 %13 = load float, ptr %arrayidx24, align 4
180 %mul25 = fmul reassoc nsz arcp contract afn float %mul23, %13
181 %arrayidx26 = getelementptr inbounds float, ptr %x, i64 28
182 %14 = load float, ptr %arrayidx26, align 4
183 %mul27 = fmul reassoc nsz arcp contract afn float %mul25, %14
184 %arrayidx28 = getelementptr inbounds float, ptr %x, i64 30
185 %15 = load float, ptr %arrayidx28, align 4
186 %mul29 = fmul reassoc nsz arcp contract afn float %mul27, %15