1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
| ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx,-slow-unaligned-mem-32 | FileCheck %s --check-prefix=ALL --check-prefix=FAST32
; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx,+slow-unaligned-mem-32 | FileCheck %s --check-prefix=ALL --check-prefix=SLOW32
define <4 x float> @merge_2_floats(float* nocapture %p) nounwind readonly {
; ALL-LABEL: merge_2_floats:
; ALL: # %bb.0:
; ALL-NEXT: vmovsd {{.*#+}} xmm0 = mem[0],zero
; ALL-NEXT: retq
%tmp1 = load float, float* %p
%vecins = insertelement <4 x float> undef, float %tmp1, i32 0
%add.ptr = getelementptr float, float* %p, i32 1
%tmp5 = load float, float* %add.ptr
%vecins7 = insertelement <4 x float> %vecins, float %tmp5, i32 1
ret <4 x float> %vecins7
}
; Test-case generated due to a crash when trying to treat loading the first
; two i64s of a <4 x i64> as a load of two i32s.
define <4 x i64> @merge_2_floats_into_4() {
; ALL-LABEL: merge_2_floats_into_4:
; ALL: # %bb.0:
; ALL-NEXT: movq (%rax), %rax
; ALL-NEXT: vmovups (%rax), %xmm0
; ALL-NEXT: retq
%1 = load i64*, i64** undef, align 8
%2 = getelementptr inbounds i64, i64* %1, i64 0
%3 = load i64, i64* %2
%4 = insertelement <4 x i64> undef, i64 %3, i32 0
%5 = load i64*, i64** undef, align 8
%6 = getelementptr inbounds i64, i64* %5, i64 1
%7 = load i64, i64* %6
%8 = insertelement <4 x i64> %4, i64 %7, i32 1
%9 = shufflevector <4 x i64> %8, <4 x i64> undef, <4 x i32> <i32 0, i32 1, i32 4, i32 5>
ret <4 x i64> %9
}
define <4 x float> @merge_4_floats(float* %ptr) {
; ALL-LABEL: merge_4_floats:
; ALL: # %bb.0:
; ALL-NEXT: vmovups (%rdi), %xmm0
; ALL-NEXT: retq
%a = load float, float* %ptr, align 8
%vec = insertelement <4 x float> undef, float %a, i32 0
%idx1 = getelementptr inbounds float, float* %ptr, i64 1
%b = load float, float* %idx1, align 8
%vec2 = insertelement <4 x float> %vec, float %b, i32 1
%idx3 = getelementptr inbounds float, float* %ptr, i64 2
%c = load float, float* %idx3, align 8
%vec4 = insertelement <4 x float> %vec2, float %c, i32 2
%idx5 = getelementptr inbounds float, float* %ptr, i64 3
%d = load float, float* %idx5, align 8
%vec6 = insertelement <4 x float> %vec4, float %d, i32 3
ret <4 x float> %vec6
}
; PR21710 ( http://llvm.org/bugs/show_bug.cgi?id=21710 )
; Make sure that 32-byte vectors are handled efficiently.
; If the target has slow 32-byte accesses, we should still generate
; 16-byte loads.
define <8 x float> @merge_8_floats(float* %ptr) {
; FAST32-LABEL: merge_8_floats:
; FAST32: # %bb.0:
; FAST32-NEXT: vmovups (%rdi), %ymm0
; FAST32-NEXT: retq
;
; SLOW32-LABEL: merge_8_floats:
; SLOW32: # %bb.0:
; SLOW32-NEXT: vmovups (%rdi), %xmm0
; SLOW32-NEXT: vinsertf128 $1, 16(%rdi), %ymm0, %ymm0
; SLOW32-NEXT: retq
%a = load float, float* %ptr, align 4
%vec = insertelement <8 x float> undef, float %a, i32 0
%idx1 = getelementptr inbounds float, float* %ptr, i64 1
%b = load float, float* %idx1, align 4
%vec2 = insertelement <8 x float> %vec, float %b, i32 1
%idx3 = getelementptr inbounds float, float* %ptr, i64 2
%c = load float, float* %idx3, align 4
%vec4 = insertelement <8 x float> %vec2, float %c, i32 2
%idx5 = getelementptr inbounds float, float* %ptr, i64 3
%d = load float, float* %idx5, align 4
%vec6 = insertelement <8 x float> %vec4, float %d, i32 3
%idx7 = getelementptr inbounds float, float* %ptr, i64 4
%e = load float, float* %idx7, align 4
%vec8 = insertelement <8 x float> %vec6, float %e, i32 4
%idx9 = getelementptr inbounds float, float* %ptr, i64 5
%f = load float, float* %idx9, align 4
%vec10 = insertelement <8 x float> %vec8, float %f, i32 5
%idx11 = getelementptr inbounds float, float* %ptr, i64 6
%g = load float, float* %idx11, align 4
%vec12 = insertelement <8 x float> %vec10, float %g, i32 6
%idx13 = getelementptr inbounds float, float* %ptr, i64 7
%h = load float, float* %idx13, align 4
%vec14 = insertelement <8 x float> %vec12, float %h, i32 7
ret <8 x float> %vec14
}
define <4 x double> @merge_4_doubles(double* %ptr) {
; FAST32-LABEL: merge_4_doubles:
; FAST32: # %bb.0:
; FAST32-NEXT: vmovups (%rdi), %ymm0
; FAST32-NEXT: retq
;
; SLOW32-LABEL: merge_4_doubles:
; SLOW32: # %bb.0:
; SLOW32-NEXT: vmovups (%rdi), %xmm0
; SLOW32-NEXT: vinsertf128 $1, 16(%rdi), %ymm0, %ymm0
; SLOW32-NEXT: retq
%a = load double, double* %ptr, align 8
%vec = insertelement <4 x double> undef, double %a, i32 0
%idx1 = getelementptr inbounds double, double* %ptr, i64 1
%b = load double, double* %idx1, align 8
%vec2 = insertelement <4 x double> %vec, double %b, i32 1
%idx3 = getelementptr inbounds double, double* %ptr, i64 2
%c = load double, double* %idx3, align 8
%vec4 = insertelement <4 x double> %vec2, double %c, i32 2
%idx5 = getelementptr inbounds double, double* %ptr, i64 3
%d = load double, double* %idx5, align 8
%vec6 = insertelement <4 x double> %vec4, double %d, i32 3
ret <4 x double> %vec6
}
; PR21771 ( http://llvm.org/bugs/show_bug.cgi?id=21771 )
; Recognize and combine consecutive loads even when the
; first of the combined loads is offset from the base address.
define <4 x double> @merge_4_doubles_offset(double* %ptr) {
; FAST32-LABEL: merge_4_doubles_offset:
; FAST32: # %bb.0:
; FAST32-NEXT: vmovups 32(%rdi), %ymm0
; FAST32-NEXT: retq
;
; SLOW32-LABEL: merge_4_doubles_offset:
; SLOW32: # %bb.0:
; SLOW32-NEXT: vmovups 32(%rdi), %xmm0
; SLOW32-NEXT: vinsertf128 $1, 48(%rdi), %ymm0, %ymm0
; SLOW32-NEXT: retq
%arrayidx4 = getelementptr inbounds double, double* %ptr, i64 4
%arrayidx5 = getelementptr inbounds double, double* %ptr, i64 5
%arrayidx6 = getelementptr inbounds double, double* %ptr, i64 6
%arrayidx7 = getelementptr inbounds double, double* %ptr, i64 7
%e = load double, double* %arrayidx4, align 8
%f = load double, double* %arrayidx5, align 8
%g = load double, double* %arrayidx6, align 8
%h = load double, double* %arrayidx7, align 8
%vecinit4 = insertelement <4 x double> undef, double %e, i32 0
%vecinit5 = insertelement <4 x double> %vecinit4, double %f, i32 1
%vecinit6 = insertelement <4 x double> %vecinit5, double %g, i32 2
%vecinit7 = insertelement <4 x double> %vecinit6, double %h, i32 3
ret <4 x double> %vecinit7
}
|