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
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
|
;=========================== 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
; CHECK-LABEL: target datalayout
; Those are new struct types, skip check them as their names are not fixed and checking isn't critical
; %__StructSOALayout_ = type <{ i32, i32, %__StructAOSLayout_, i32 }>
; %__StructAOSLayout_ = type <{ <2 x i16> }>
; %__StructSOALayout_.1 = type <{ %__StructAOSLayout_.0, i32, i32 }>
; %__StructAOSLayout_.0 = type <{ i16, i16 }>
; %__StructSOALayout_.2 = type <{ %__StructAOSLayout_.0, %__StructAOSLayout_.0, %__StructAOSLayout_.0 }>
; %__StructSOALayout_.5 = type <{ %__StructAOSLayout_.3, %__StructAOSLayout_.4 }>
; %__StructAOSLayout_.3 = type <{ i8, i8, i8, i8 }>
; %__StructAOSLayout_.4 = type <{ i8, i8, i16 }>
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"
; 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 <2 x i32>; load <2 x i16>; load i32; -> load <4 x i32>
; CHECK-LABEL: define spir_kernel void @test_st
; CHECK: [[T0:%.*]] = load <4 x i32>
; CHECK: {{.*}} = call %__StructSOALayout_ @llvm.genx.GenISA.bitcasttostruct.__StructSOALayout_.v4i32(<4 x i32> [[T0]])
%c0.base = add i64 %conv.i.i.i, 16
%c0.arrayidx = getelementptr inbounds i32, i32 addrspace(1)* %si, i64 %c0.base
%c0.si.2i32 = bitcast i32 addrspace(1)* %c0.arrayidx to <2 x i32> addrspace(1)*
%c0.0 = load <2 x i32>, <2 x i32> addrspace(1)* %c0.si.2i32, align 4
%c0.add = add nuw nsw i64 %c0.base, 2
%c0.arrayidx1 = getelementptr inbounds i32, i32 addrspace(1)* %si, i64 %c0.add
%c0.si.2i16 = bitcast i32 addrspace(1)* %c0.arrayidx1 to <2 x i16> addrspace(1)*
%c0.1 = load <2 x i16>, <2 x i16> addrspace(1)* %c0.si.2i16, align 4
%c0.add.1 = add nuw nsw i64 %c0.base, 3
%c0.arrayidx1.1 = getelementptr inbounds i32, i32 addrspace(1)* %si, i64 %c0.add.1
%c0.2 = load i32, i32 addrspace(1)* %c0.arrayidx1.1, align 4
%c0.1.0 = bitcast <2 x i16> %c0.1 to i32
%c0.0.0 = extractelement <2 x i32> %c0.0, i32 0
%c0.0.1 = extractelement <2 x i32> %c0.0, i32 1
%c0.v.0 = insertelement <4 x i32> undef, i32 %c0.0.0, i32 0
%c0.v.1 = insertelement <4 x i32> %c0.v.0, i32 %c0.0.1, i32 1
%c0.v.2 = insertelement <4 x i32> %c0.v.1, i32 %c0.1.0, i32 2
%c0.v.3 = insertelement <4 x i32> %c0.v.2, i32 %c0.2, i32 2
%c0.outidx = getelementptr inbounds i32, i32 addrspace(1)* %d, i64 %c0.base
%c0.addr.1 = bitcast i32 addrspace(1)* %c0.outidx to <4 x i32> addrspace(1)*
store <4 x i32> %c0.v.2, <4 x i32> addrspace(1)* %c0.addr.1, align 4
; case 1: load i16; load i16, load i64; -> 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, 16
%c1.arrayidx = getelementptr inbounds i32, i32 addrspace(1)* %si, i64 %c1.base
%c1.si.i16 = bitcast i32 addrspace(1)* %c1.arrayidx to i16 addrspace(1)*
%c1.0 = load i16, i16 addrspace(1)* %c1.si.i16, align 4
%c1.arrayidx.1 = getelementptr inbounds i16, i16 addrspace(1)* %c1.si.i16, i64 1
%c1.1 = load i16, i16 addrspace(1)* %c1.arrayidx.1, align 2
%c1.arrayidx.2 = getelementptr inbounds i16, i16 addrspace(1)* %c1.si.i16, i64 2
%c1.si.i64 = bitcast i16 addrspace(1)* %c1.arrayidx.2 to i64 addrspace(1)*
%c1.2 = load i64, i64 addrspace(1)* %c1.si.i64, align 4
%c1.0.0 = insertelement <2 x i16> undef, i16 %c1.0, i32 0
%c1.0.1 = insertelement <2 x i16> %c1.0.0, i16 %c1.1, i32 1
%c1.0.2 = bitcast <2 x i16> %c1.0.1 to i32
%c1.2.0 = trunc i64 %c1.2 to i32
%c1.2.1 = lshr i64 %c1.2, 32
%c1.2.2 = trunc i64 %c1.2.1 to i32
%c1.v.0 = insertelement <3 x i32> undef, i32 %c1.0.2, i32 0
%c1.v.1 = insertelement <3 x i32> %c1.v.0, i32 %c1.2.0, 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 i16; load <2xi32>; load i16 -> load <3 x i32>
; note: <2xi32> is split into {i16, i16, i16, i16}
; This happens for packed/mis-aligned loads
; CHECK-LABEL: c2.base
; CHECK: [[T2:%.*]] = load <3 x i32>
; CHECK: {{.*}} = call %__StructSOALayout_.2 @llvm.genx.GenISA.bitcasttostruct.__StructSOALayout_.2.v3i32(<3 x i32> [[T2]])
; ret void
%c2.base = add i64 %conv.i.i.i, 64
%c2.arrayidx = getelementptr inbounds i32, i32 addrspace(1)* %si, i64 %c2.base
%c2.si.i16 = bitcast i32 addrspace(1)* %c2.arrayidx to i16 addrspace(1)*
%c2.0 = load i16, i16 addrspace(1)* %c2.si.i16, align 4
%c2.arrayidx.1 = getelementptr inbounds i16, i16 addrspace(1)* %c2.si.i16, i64 1
%c2.si.2i32 = bitcast i16 addrspace(1)* %c2.arrayidx.1 to <2 x i32> addrspace(1)*
%c2.1 = load <2 x i32>, <2 x i32> addrspace(1)* %c2.si.2i32, align 2
%c2.arrayidx.2 = getelementptr inbounds i16, i16 addrspace(1)* %c2.si.i16, i64 5
%c2.2 = load i16, i16 addrspace(1)* %c2.arrayidx.2, align 2
%c2.0.0 = insertelement <2 x i16> undef, i16 %c2.0, i32 0
%c2.0.1 = insertelement <2 x i16> %c2.0.0, i16 %c2.2, i32 1
%c2.0.2 = bitcast <2 x i16> %c2.0.1 to i32
%c2.1.0 = extractelement <2 x i32> %c2.1, i64 0
%c2.1.1 = extractelement <2 x i32> %c2.1, i64 1
%c2.v.0 = insertelement <3 x i32> undef, i32 %c2.0.2, i32 0
%c2.v.1 = insertelement <3 x i32> %c2.v.0, i32 %c2.1.0, i32 1
%c2.v.2 = insertelement <3 x i32> %c2.v.1, i32 %c2.1.1, i32 2
%c2.outidx = getelementptr inbounds i32, i32 addrspace(1)* %d, i64 %c2.base
%c2.addr.1 = bitcast i32 addrspace(1)* %c2.outidx to <3 x i32> addrspace(1)*
store <3 x i32> %c2.v.2, <3 x i32> addrspace(1)* %c2.addr.1, align 4
; case 3:
; case 3: load i8; load <2xi8>; load <3xi8>; load <i16> -> load <2 x i32>
; note: <3xi8> is split into {i8, i8, i8} and the first i8 is in the 1st
; AOS struct <{i8, <2xi8>, i8}>, and the remaining two are in the
; second and different AOS struct <{i8, i8, i16}>.
; (This test is actually from OCL conformance basic/vector_creation)
; CHECK-LABEL: c3.base
; CHECK: [[T3:%.*]] = load <2 x i32>
; CHECK: {{.*}} = call %__StructSOALayout_.5 @llvm.genx.GenISA.bitcasttostruct.__StructSOALayout_.5.v2i32(<2 x i32> [[T3]])
; ret void
%c3.base = add i64 %conv.i.i.i, 96
%c3.arrayidx = getelementptr inbounds i32, i32 addrspace(1)* %si, i64 %c3.base
%c3.si.i8 = bitcast i32 addrspace(1)* %c3.arrayidx to i8 addrspace(1)*
%c3.0 = load i8, i8 addrspace(1)* %c3.si.i8, align 4
%c3.arrayidx.1 = getelementptr inbounds i8, i8 addrspace(1)* %c3.si.i8, i64 1
%c3.si.2i8 = bitcast i8 addrspace(1)* %c3.arrayidx.1 to <2 x i8> addrspace(1)*
%c3.1 = load <2 x i8>, <2 x i8> addrspace(1)* %c3.si.2i8, align 1
%c3.arrayidx.2 = getelementptr inbounds i8, i8 addrspace(1)* %c3.si.i8, i64 3
%c3.si.3i8 = bitcast i8 addrspace(1)* %c3.arrayidx.2 to <3 x i8> addrspace(1)*
%c3.2 = load <3 x i8>, <3 x i8> addrspace(1)* %c3.si.3i8, align 1
%c3.arrayidx.3 = getelementptr inbounds i8, i8 addrspace(1)* %c3.si.i8, i64 6
%c3.si.i16.0 = bitcast i8 addrspace(1)* %c3.arrayidx.3 to i16 addrspace(1)*
%c3.3.i16 = load i16, i16 addrspace(1)* %c3.si.i16.0, align 1
%c3.3 = bitcast i16 %c3.3.i16 to <2 x i8>
%c3.1.0 = extractelement <2 x i8> %c3.1, i64 0
%c3.1.1 = extractelement <2 x i8> %c3.1, i64 1
%c3.2.0 = extractelement <3 x i8> %c3.2, i64 0
%c3.2.1 = extractelement <3 x i8> %c3.2, i64 1
%c3.2.2 = extractelement <3 x i8> %c3.2, i64 2
%c3.3.0 = extractelement <2 x i8> %c3.3, i64 0
%c3.3.1 = extractelement <2 x i8> %c3.3, i64 1
%c3.v.0 = insertelement <8 x i8> undef, i8 %c3.0, i32 0
%c3.v.1 = insertelement <8 x i8> %c3.v.0, i8 %c3.1.0, i32 1
%c3.v.2 = insertelement <8 x i8> %c3.v.1, i8 %c3.1.1, i32 2
%c3.v.3 = insertelement <8 x i8> %c3.v.2, i8 %c3.2.0, i32 3
%c3.v.4 = insertelement <8 x i8> %c3.v.3, i8 %c3.2.1, i32 4
%c3.v.5 = insertelement <8 x i8> %c3.v.4, i8 %c3.2.2, i32 5
%c3.v.6 = insertelement <8 x i8> %c3.v.5, i8 %c3.3.0, i32 6
%c3.v.7 = insertelement <8 x i8> %c3.v.6, i8 %c3.3.1, i32 7
%c3.outidx = getelementptr inbounds i32, i32 addrspace(1)* %d, i64 %c3.base
%c3.addr.1 = bitcast i32 addrspace(1)* %c3.outidx to <8 x i8> addrspace(1)*
store <8 x i8> %c3.v.2, <8 x i8> addrspace(1)* %c3.addr.1, align 4
ret void
}
|
