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
| ; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
; RUN: opt < %s -mattr=sse2 -slp-vectorizer -S | FileCheck %s --check-prefix=CHECK --check-prefix=SSE
; RUN: opt < %s -mattr=avx2 -slp-vectorizer -S | FileCheck %s --check-prefix=CHECK --check-prefix=AVX
; TODO:
; With AVX, we are able to vectorize the 1st 4 elements as 256-bit vector ops,
; but the final 2 elements remain scalar. They should get vectorized using
; 128-bit ops identically to what happens with SSE.
target datalayout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-f80:128-n8:16:32:64-S128"
target triple = "x86_64-unknown-linux-gnu"
define void @PR28457(double* noalias nocapture align 32 %q, double* noalias nocapture readonly align 32 %p) {
; SSE-LABEL: @PR28457(
; SSE-NEXT: [[P0:%.*]] = getelementptr inbounds double, double* [[P:%.*]], i64 0
; SSE-NEXT: [[P1:%.*]] = getelementptr inbounds double, double* [[P]], i64 1
; SSE-NEXT: [[P2:%.*]] = getelementptr inbounds double, double* [[P]], i64 2
; SSE-NEXT: [[P3:%.*]] = getelementptr inbounds double, double* [[P]], i64 3
; SSE-NEXT: [[P4:%.*]] = getelementptr inbounds double, double* [[P]], i64 4
; SSE-NEXT: [[P5:%.*]] = getelementptr inbounds double, double* [[P]], i64 5
; SSE-NEXT: [[Q0:%.*]] = getelementptr inbounds double, double* [[Q:%.*]], i64 0
; SSE-NEXT: [[Q1:%.*]] = getelementptr inbounds double, double* [[Q]], i64 1
; SSE-NEXT: [[Q2:%.*]] = getelementptr inbounds double, double* [[Q]], i64 2
; SSE-NEXT: [[Q3:%.*]] = getelementptr inbounds double, double* [[Q]], i64 3
; SSE-NEXT: [[Q4:%.*]] = getelementptr inbounds double, double* [[Q]], i64 4
; SSE-NEXT: [[Q5:%.*]] = getelementptr inbounds double, double* [[Q]], i64 5
; SSE-NEXT: [[TMP1:%.*]] = bitcast double* [[P0]] to <2 x double>*
; SSE-NEXT: [[TMP2:%.*]] = load <2 x double>, <2 x double>* [[TMP1]], align 8
; SSE-NEXT: [[TMP3:%.*]] = bitcast double* [[P2]] to <2 x double>*
; SSE-NEXT: [[TMP4:%.*]] = load <2 x double>, <2 x double>* [[TMP3]], align 8
; SSE-NEXT: [[TMP5:%.*]] = bitcast double* [[P4]] to <2 x double>*
; SSE-NEXT: [[TMP6:%.*]] = load <2 x double>, <2 x double>* [[TMP5]], align 8
; SSE-NEXT: [[TMP7:%.*]] = fadd <2 x double> [[TMP2]], <double 1.000000e+00, double 1.000000e+00>
; SSE-NEXT: [[TMP8:%.*]] = fadd <2 x double> [[TMP4]], <double 1.000000e+00, double 1.000000e+00>
; SSE-NEXT: [[TMP9:%.*]] = fadd <2 x double> [[TMP6]], <double 1.000000e+00, double 1.000000e+00>
; SSE-NEXT: [[TMP10:%.*]] = bitcast double* [[Q0]] to <2 x double>*
; SSE-NEXT: store <2 x double> [[TMP7]], <2 x double>* [[TMP10]], align 8
; SSE-NEXT: [[TMP11:%.*]] = bitcast double* [[Q2]] to <2 x double>*
; SSE-NEXT: store <2 x double> [[TMP8]], <2 x double>* [[TMP11]], align 8
; SSE-NEXT: [[TMP12:%.*]] = bitcast double* [[Q4]] to <2 x double>*
; SSE-NEXT: store <2 x double> [[TMP9]], <2 x double>* [[TMP12]], align 8
; SSE-NEXT: ret void
;
; AVX-LABEL: @PR28457(
; AVX-NEXT: [[P0:%.*]] = getelementptr inbounds double, double* [[P:%.*]], i64 0
; AVX-NEXT: [[P1:%.*]] = getelementptr inbounds double, double* [[P]], i64 1
; AVX-NEXT: [[P2:%.*]] = getelementptr inbounds double, double* [[P]], i64 2
; AVX-NEXT: [[P3:%.*]] = getelementptr inbounds double, double* [[P]], i64 3
; AVX-NEXT: [[P4:%.*]] = getelementptr inbounds double, double* [[P]], i64 4
; AVX-NEXT: [[P5:%.*]] = getelementptr inbounds double, double* [[P]], i64 5
; AVX-NEXT: [[Q0:%.*]] = getelementptr inbounds double, double* [[Q:%.*]], i64 0
; AVX-NEXT: [[Q1:%.*]] = getelementptr inbounds double, double* [[Q]], i64 1
; AVX-NEXT: [[Q2:%.*]] = getelementptr inbounds double, double* [[Q]], i64 2
; AVX-NEXT: [[Q3:%.*]] = getelementptr inbounds double, double* [[Q]], i64 3
; AVX-NEXT: [[Q4:%.*]] = getelementptr inbounds double, double* [[Q]], i64 4
; AVX-NEXT: [[Q5:%.*]] = getelementptr inbounds double, double* [[Q]], i64 5
; AVX-NEXT: [[TMP1:%.*]] = bitcast double* [[P0]] to <4 x double>*
; AVX-NEXT: [[TMP2:%.*]] = load <4 x double>, <4 x double>* [[TMP1]], align 8
; AVX-NEXT: [[D4:%.*]] = load double, double* [[P4]]
; AVX-NEXT: [[D5:%.*]] = load double, double* [[P5]]
; AVX-NEXT: [[TMP3:%.*]] = fadd <4 x double> [[TMP2]], <double 1.000000e+00, double 1.000000e+00, double 1.000000e+00, double 1.000000e+00>
; AVX-NEXT: [[A4:%.*]] = fadd double [[D4]], 1.000000e+00
; AVX-NEXT: [[A5:%.*]] = fadd double [[D5]], 1.000000e+00
; AVX-NEXT: [[TMP4:%.*]] = bitcast double* [[Q0]] to <4 x double>*
; AVX-NEXT: store <4 x double> [[TMP3]], <4 x double>* [[TMP4]], align 8
; AVX-NEXT: store double [[A4]], double* [[Q4]]
; AVX-NEXT: store double [[A5]], double* [[Q5]]
; AVX-NEXT: ret void
;
%p0 = getelementptr inbounds double, double* %p, i64 0
%p1 = getelementptr inbounds double, double* %p, i64 1
%p2 = getelementptr inbounds double, double* %p, i64 2
%p3 = getelementptr inbounds double, double* %p, i64 3
%p4 = getelementptr inbounds double, double* %p, i64 4
%p5 = getelementptr inbounds double, double* %p, i64 5
%q0 = getelementptr inbounds double, double* %q, i64 0
%q1 = getelementptr inbounds double, double* %q, i64 1
%q2 = getelementptr inbounds double, double* %q, i64 2
%q3 = getelementptr inbounds double, double* %q, i64 3
%q4 = getelementptr inbounds double, double* %q, i64 4
%q5 = getelementptr inbounds double, double* %q, i64 5
%d0 = load double, double* %p0
%d1 = load double, double* %p1
%d2 = load double, double* %p2
%d3 = load double, double* %p3
%d4 = load double, double* %p4
%d5 = load double, double* %p5
%a0 = fadd double %d0, 1.0
%a1 = fadd double %d1, 1.0
%a2 = fadd double %d2, 1.0
%a3 = fadd double %d3, 1.0
%a4 = fadd double %d4, 1.0
%a5 = fadd double %d5, 1.0
store double %a0, double* %q0
store double %a1, double* %q1
store double %a2, double* %q2
store double %a3, double* %q3
store double %a4, double* %q4
store double %a5, double* %q5
ret void
}
|