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
|
;=========================== begin_copyright_notice ============================
;
; Copyright (C) 2017-2024 Intel Corporation
;
; SPDX-License-Identifier: MIT
;
;============================ end_copyright_notice =============================
; REQUIRES: llvm-14-plus, regkeys
; RUN: igc_opt --opaque-pointers %s -S -inputocl -igc-ldstcombine -regkey=EnableLdStCombine=5 \
; RUN: -platformbmg \
; RUN: | FileCheck %s
target datalayout = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v16:16:16-v24:32:32-v32:32:32-v48:64:64-v64:64:64-v96:128:128-v128:128:128-v192:256:256-v256:256:256-v512:512:512-v1024:1024:1024-n8:16:32"
target triple = "spir64-unknown-unknown"
; CHECK-LABEL: target datalayout
; layout struct generated. The test needs to be adjusted if their names are different
; %__StructSOALayout_ = type <{ float, %__StructAOSLayout_ }>
; %__StructAOSLayout_ = type <{ i8, i8, i8, i8 }>
; %__StructSOALayout_.1 = type <{ i32, %__StructAOSLayout_.0, float }>
; %__StructAOSLayout_.0 = type <{ i16, i16 }>
; %__StructSOALayout_.2 = type <{ %__StructAOSLayout_, %__StructAOSLayout_ }>
; Function Attrs: convergent nounwind
define spir_kernel void @test_st(i32 addrspace(1)* %d, i32 addrspace(1)* %si, float addrspace(1)* %sf, <8 x i32> %r0, <8 x i32> %payloadHeader, <3 x i32> %enqueuedLocalSize, i16 %localIdX, i16 %localIdY, i16 %localIdZ) {
entry:
%scalar27 = extractelement <8 x i32> %payloadHeader, i32 0
%scalar24 = extractelement <3 x i32> %enqueuedLocalSize, i32 0
%scalar17 = extractelement <8 x i32> %r0, i32 1
%mul.i.i.i = mul i32 %scalar24, %scalar17
%localIdX2 = zext i16 %localIdX to i32
%add.i.i.i = add i32 %mul.i.i.i, %localIdX2
%add4.i.i.i = add i32 %add.i.i.i, %scalar27
%conv.i.i.i = zext i32 %add4.i.i.i to i64
; case 0: load float; load i8; load i8; load i8; load i8 --> load <2 x i32>
; CHECK-LABEL: define spir_kernel void @test_st
; CHECK: [[T0:%.*]] = load <2 x i32>
; CHECK: {{.*}} = call %__StructSOALayout_ @llvm.genx.GenISA.bitcasttostruct.__StructSOALayout_.v2i32(<2 x i32> [[T0]])
%c0.arrayidx = getelementptr inbounds float, float addrspace(1)* %sf, i64 %conv.i.i.i
%c0.0 = load float, float addrspace(1)* %c0.arrayidx, align 4
%c0.add = add nuw nsw i64 %conv.i.i.i, 1
%c0.arrayidx1 = getelementptr inbounds float, float addrspace(1)* %sf, i64 %c0.add
%c0.sf.i8 = bitcast float addrspace(1)* %c0.arrayidx1 to i8 addrspace(1)*
%c0.1 = load i8, i8 addrspace(1)* %c0.sf.i8, align 4
%c0.arrayidx1.2 = getelementptr inbounds i8, i8 addrspace(1)* %c0.sf.i8, i64 1
%c0.2 = load i8, i8 addrspace(1)* %c0.arrayidx1.2, align 1
%c0.arrayidx1.3 = getelementptr inbounds i8, i8 addrspace(1)* %c0.sf.i8, i64 2
%c0.3 = load i8, i8 addrspace(1)* %c0.arrayidx1.3, align 1
%c0.arrayidx1.4 = getelementptr inbounds i8, i8 addrspace(1)* %c0.sf.i8, i64 3
%c0.4 = load i8, i8 addrspace(1)* %c0.arrayidx1.4, align 1
%c0.v.0 = insertelement <4 x i8> undef, i8 %c0.1, i32 0
%c0.v.1 = insertelement <4 x i8> %c0.v.0, i8 %c0.2, i32 1
%c0.v.2 = insertelement <4 x i8> %c0.v.1, i8 %c0.3, i32 2
%c0.v.3 = insertelement <4 x i8> %c0.v.2, i8 %c0.4, i32 3
%c0.v1.0 = bitcast <4 x i8> %c0.v.3 to i32
%c0.v1.1 = bitcast float %c0.0 to i32
%c0.v2.0 = insertelement <2 x i32> undef, i32 %c0.v1.0, i32 0
%c0.v2.1 = insertelement <2 x i32> %c0.v2.0, i32 %c0.v1.1, i32 1
%c0.outidx = getelementptr inbounds i32, i32 addrspace(1)* %d, i64 %conv.i.i.i
%c0.addr.0 = bitcast i32 addrspace(1)* %c0.outidx to <2 x i32> addrspace(1)*
store <2 x i32> %c0.v2.1, <2 x i32> addrspace(1)* %c0.addr.0, align 4
; case 1: load i32; load i16; load i16; load float -> load <3 x i32>
; CHECK-LABEL: %c1.base
; CHECK: [[T1:%.*]] = load <3 x i32>
; CHECK: {{.*}} = call %__StructSOALayout_.1 @llvm.genx.GenISA.bitcasttostruct.__StructSOALayout_.1.v3i32(<3 x i32> [[T1]])
%c1.base = add i64 %conv.i.i.i, 64
%c1.arrayidx = getelementptr inbounds i32, i32 addrspace(1)* %si, i64 %c1.base
%c1.0 = load i32, i32 addrspace(1)* %c1.arrayidx, align 4
%c1.add = add nuw nsw i64 %c1.base, 1
%c1.arrayidx1 = getelementptr inbounds i32, i32 addrspace(1)* %si, i64 %c1.add
%c1.si.i16 = bitcast i32 addrspace(1)* %c1.arrayidx1 to i16 addrspace(1)*
%c1.1 = load i16, i16 addrspace(1)* %c1.si.i16, align 4
%c1.arrayidx1.2 = getelementptr inbounds i16, i16 addrspace(1)* %c1.si.i16, i64 1
%c1.2 = load i16, i16 addrspace(1)* %c1.arrayidx1.2, align 2
%c1.add.0 = add nuw nsw i64 %c1.base, 2
%c1.arrayidx1.1 = getelementptr inbounds i32, i32 addrspace(1)* %si, i64 %c1.add.0
%c1.ldaddr = bitcast i32 addrspace(1)* %c1.arrayidx1.1 to float addrspace(1)*
%c1.3 = load float, float addrspace(1)* %c1.ldaddr, align 4
%c1.1.0 = zext i16 %c1.1 to i32
%c1.2.0 = zext i16 %c1.2 to i32
%c1.2.1 = shl i32 %c1.2.0, 16
%c1.2.2 = or i32 %c1.1.0, %c1.2.1
%c1.3.0 = bitcast float %c1.3 to i32
%c1.v.0 = insertelement <3 x i32> undef, i32 %c1.0, i32 0
%c1.v.1 = insertelement <3 x i32> %c1.v.0, i32 %c1.3.0, i32 1
%c1.v.2 = insertelement <3 x i32> %c1.v.1, i32 %c1.2.2, i32 2
%c1.outidx = getelementptr inbounds i32, i32 addrspace(1)* %d, i64 %c1.base
%c1.addr.1 = bitcast i32 addrspace(1)* %c1.outidx to <3 x i32> addrspace(1)*
store <3 x i32> %c1.v.2, <3 x i32> addrspace(1)* %c1.addr.1, align 4
; case 2: load i8; load <2xi16>; load <3xi8> -> load <2 x i32>
; Manually created to handle mis-aligned load that could come from packed struct.
; Note: <2xi16> is split into {i8, i8, i8, i8} and the first three i8's are in the first AOS
; struct <{i8, i8, i8, i8}>, where the last i8 is in the second AOS struct <{i8, <3xi8>}>
; CHECK-LABEL: %c2.base
; CHECK: [[T2:%.*]] = load <2 x i32>
; CHECK: {{.*}} = call %__StructSOALayout_.2 @llvm.genx.GenISA.bitcasttostruct.__StructSOALayout_.2.v2i32(<2 x i32> [[T2]])
; CHECK: ret void
%c2.base = add i64 %conv.i.i.i, 96
%c2.arrayidx = getelementptr inbounds i32, i32 addrspace(1)* %si, i64 %c2.base
%c2.si.i8 = bitcast i32 addrspace(1)* %c2.arrayidx to i8 addrspace(1)*
%c2.0 = load i8, i8 addrspace(1)* %c2.si.i8, align 4
%c2.arrayidx.1 = getelementptr inbounds i8, i8 addrspace(1)* %c2.si.i8, i64 1
%c2.si.i16 = bitcast i8 addrspace(1)* %c2.arrayidx.1 to <2 x i16> addrspace(1)*
%c2.1 = load <2 x i16>, <2 x i16> addrspace(1)* %c2.si.i16, align 1
%c2.arrayidx.2 = getelementptr inbounds i8, i8 addrspace(1)* %c2.si.i8, i64 5
%c2.si.3i8 = bitcast i8 addrspace(1)* %c2.arrayidx.2 to <3 x i8> addrspace(1)*
%c2.2 = load <3 x i8>, <3 x i8> addrspace(1)* %c2.si.3i8, align 1
%c2.1.0 = bitcast <2 x i16> %c2.1 to <4 x i8>
%c2.1.0.0 = extractelement <4 x i8> %c2.1.0, i64 0
%c2.1.0.1 = extractelement <4 x i8> %c2.1.0, i64 1
%c2.1.0.2 = extractelement <4 x i8> %c2.1.0, i64 2
%c2.1.0.3 = extractelement <4 x i8> %c2.1.0, i64 3
%c2.2.0 = extractelement <3 x i8> %c2.2, i64 0
%c2.2.1 = extractelement <3 x i8> %c2.2, i64 1
%c2.2.2 = extractelement <3 x i8> %c2.2, i64 2
%c2.v.0 = insertelement <8 x i8> undef, i8 %c2.0, i32 0
%c2.v.1 = insertelement <8 x i8> %c2.v.0, i8 %c2.1.0.0, i32 1
%c2.v.2 = insertelement <8 x i8> %c2.v.1, i8 %c2.1.0.1, i32 2
%c2.v.3 = insertelement <8 x i8> %c2.v.2, i8 %c2.1.0.2, i32 3
%c2.v.4 = insertelement <8 x i8> %c2.v.3, i8 %c2.1.0.3, i32 4
%c2.v.5 = insertelement <8 x i8> %c2.v.4, i8 %c2.2.0, i32 5
%c2.v.6 = insertelement <8 x i8> %c2.v.5, i8 %c2.2.1, i32 6
%c2.v.7 = insertelement <8 x i8> %c2.v.6, i8 %c2.2.2, i32 7
%c2.outidx = getelementptr inbounds i32, i32 addrspace(1)* %d, i64 %c2.base
%c2.addr.1 = bitcast i32 addrspace(1)* %c2.outidx to <8 x i8> addrspace(1)*
store <8 x i8> %c2.v.2, <8 x i8> addrspace(1)* %c2.addr.1, align 4
ret void
}
|
