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
|
;=========================== begin_copyright_notice ============================
;
; Copyright (C) 2024 Intel Corporation
;
; SPDX-License-Identifier: MIT
;
;============================ end_copyright_notice =============================
;
; REQUIRES: llvm-14-plus
;
; RUN: igc_opt -platformpvc -debugify -igc-subgroup-reduction-pattern -check-debugify -S < %s 2>&1 | FileCheck %s --check-prefixes=CHECK,CHECK-PVC
; RUN: igc_opt -platformdg2 -debugify -igc-subgroup-reduction-pattern -check-debugify -S < %s 2>&1 | FileCheck %s --check-prefixes=CHECK,CHECK-DG2
; Test if a pattern of repeated ShuffleIndex + op is recognized and replaced with WaveAll.
; Debug-info related check
; CHECK: CheckModuleDebugify: PASS
define i64 @wave_all_add_i64() {
entry:
; CHECK-LABEL: entry:
;
; CHECK-PVC: %0 = call i64 @get_i64()
; CHECK-PVC: [[RESULT:%.*]] = call i64 @llvm.genx.GenISA.WaveAll.i64(i64 %0, i8 0, i32 0)
; CHECK-PVC: ret i64 [[RESULT]]
;
; CHECK-DG2-NOT: @llvm.genx.GenISA.WaveAll
; CHECK-DG2: ret i64 %add9
%0 = call i64 @get_i64()
%simdLaneId16 = call i16 @llvm.genx.GenISA.simdLaneId()
%xor16 = xor i16 %simdLaneId16, 16
%zext16 = zext i16 %xor16 to i32
%1 = bitcast i64 %0 to <2 x i32>
%.scalar = extractelement <2 x i32> %1, i64 0
%.scalar51 = extractelement <2 x i32> %1, i64 1
%simdShuffle = call i32 @llvm.genx.GenISA.WaveShuffleIndex.i32(i32 %.scalar, i32 %zext16, i32 0)
%simdShuffle18 = call i32 @llvm.genx.GenISA.WaveShuffleIndex.i32(i32 %.scalar51, i32 %zext16, i32 0)
%.assembled.vect = insertelement <2 x i32> undef, i32 %simdShuffle, i64 0
%.assembled.vect60 = insertelement <2 x i32> %.assembled.vect, i32 %simdShuffle18, i64 1
%2 = bitcast <2 x i32> %.assembled.vect60 to i64
%add = add nsw i64 %0, %2
%3 = bitcast i64 %add to <2 x i32>
%.scalar52 = extractelement <2 x i32> %3, i64 0
%.scalar53 = extractelement <2 x i32> %3, i64 1
%simdShuffleXor = call i32 @llvm.genx.GenISA.simdShuffleXor.i32(i32 %.scalar52, i32 8)
%simdShuffleXor69 = call i32 @llvm.genx.GenISA.simdShuffleXor.i32(i32 %.scalar53, i32 8)
%.assembled.vect61 = insertelement <2 x i32> undef, i32 %simdShuffleXor, i64 0
%.assembled.vect62 = insertelement <2 x i32> %.assembled.vect61, i32 %simdShuffleXor69, i64 1
%4 = bitcast <2 x i32> %.assembled.vect62 to i64
%add3 = add nsw i64 %add, %4
%5 = bitcast i64 %add3 to <2 x i32>
%.scalar54 = extractelement <2 x i32> %5, i64 0
%.scalar55 = extractelement <2 x i32> %5, i64 1
%simdShuffleXor70 = call i32 @llvm.genx.GenISA.simdShuffleXor.i32(i32 %.scalar54, i32 4)
%simdShuffleXor71 = call i32 @llvm.genx.GenISA.simdShuffleXor.i32(i32 %.scalar55, i32 4)
%.assembled.vect63 = insertelement <2 x i32> undef, i32 %simdShuffleXor70, i64 0
%.assembled.vect64 = insertelement <2 x i32> %.assembled.vect63, i32 %simdShuffleXor71, i64 1
%6 = bitcast <2 x i32> %.assembled.vect64 to i64
%add5 = add nsw i64 %add3, %6
%7 = bitcast i64 %add5 to <2 x i32>
%.scalar56 = extractelement <2 x i32> %7, i64 0
%.scalar57 = extractelement <2 x i32> %7, i64 1
%simdShuffleXor72 = call i32 @llvm.genx.GenISA.simdShuffleXor.i32(i32 %.scalar56, i32 2)
%simdShuffleXor73 = call i32 @llvm.genx.GenISA.simdShuffleXor.i32(i32 %.scalar57, i32 2)
%.assembled.vect65 = insertelement <2 x i32> undef, i32 %simdShuffleXor72, i64 0
%.assembled.vect66 = insertelement <2 x i32> %.assembled.vect65, i32 %simdShuffleXor73, i64 1
%8 = bitcast <2 x i32> %.assembled.vect66 to i64
%add7 = add nsw i64 %add5, %8
%9 = bitcast i64 %add7 to <2 x i32>
%.scalar58 = extractelement <2 x i32> %9, i64 0
%.scalar59 = extractelement <2 x i32> %9, i64 1
%simdShuffleXor74 = call i32 @llvm.genx.GenISA.simdShuffleXor.i32(i32 %.scalar58, i32 1)
%simdShuffleXor75 = call i32 @llvm.genx.GenISA.simdShuffleXor.i32(i32 %.scalar59, i32 1)
%.assembled.vect67 = insertelement <2 x i32> undef, i32 %simdShuffleXor74, i64 0
%.assembled.vect68 = insertelement <2 x i32> %.assembled.vect67, i32 %simdShuffleXor75, i64 1
%10 = bitcast <2 x i32> %.assembled.vect68 to i64
%add9 = add nsw i64 %add7, %10
ret i64 %add9
}
define i64 @mixed_extract_insert_indices() {
entry:
; CHECK-LABEL: entry:
; CHECK-NOT: @llvm.genx.GenISA.WaveClustered
; CHECK: ret i64 %add1
%0 = call i64 @get_i64()
%simdLaneId16 = call i16 @llvm.genx.GenISA.simdLaneId()
%xor1 = xor i16 %simdLaneId16, 1
%zext1 = zext i16 %xor1 to i32
%1 = bitcast i64 %0 to <2 x i32>
%.scalar = extractelement <2 x i32> %1, i64 0
%.scalar51 = extractelement <2 x i32> %1, i64 1
%simdShuffle = call i32 @llvm.genx.GenISA.WaveShuffleIndex.i32(i32 %.scalar, i32 %zext1, i32 0)
%simdShuffle18 = call i32 @llvm.genx.GenISA.WaveShuffleIndex.i32(i32 %.scalar51, i32 %zext1, i32 0)
; COM: Elements inserted in reverse order == invalid pattern.
%.assembled.vect = insertelement <2 x i32> undef, i32 %simdShuffle, i64 1
%.assembled.vect60 = insertelement <2 x i32> %.assembled.vect, i32 %simdShuffle18, i64 0
%2 = bitcast <2 x i32> %.assembled.vect60 to i64
%add1 = add nsw i64 %0, %2
ret i64 %add1
}
define i64 @mixed_shuffle_lanes() {
entry:
; CHECK-LABEL: entry:
; CHECK-NOT: @llvm.genx.GenISA.WaveClustered
; CHECK: ret i64 %add9
%0 = call i64 @get_i64()
%1 = bitcast i64 %0 to <2 x i32>
%.scalar56 = extractelement <2 x i32> %1, i64 0
%.scalar57 = extractelement <2 x i32> %1, i64 1
; COM: Different shuffle indices == invalid pattern.
%simdShuffleXor72 = call i32 @llvm.genx.GenISA.simdShuffleXor.i32(i32 %.scalar56, i32 2)
%simdShuffleXor73 = call i32 @llvm.genx.GenISA.simdShuffleXor.i32(i32 %.scalar57, i32 1)
%.assembled.vect65 = insertelement <2 x i32> undef, i32 %simdShuffleXor72, i64 0
%.assembled.vect66 = insertelement <2 x i32> %.assembled.vect65, i32 %simdShuffleXor73, i64 1
%2 = bitcast <2 x i32> %.assembled.vect66 to i64
%add7 = add nsw i64 %0, %2
%3 = bitcast i64 %add7 to <2 x i32>
%.scalar58 = extractelement <2 x i32> %3, i64 0
%.scalar59 = extractelement <2 x i32> %3, i64 1
; COM: Different shuffle indices == invalid pattern.
%simdShuffleXor74 = call i32 @llvm.genx.GenISA.simdShuffleXor.i32(i32 %.scalar58, i32 1)
%simdShuffleXor75 = call i32 @llvm.genx.GenISA.simdShuffleXor.i32(i32 %.scalar59, i32 2)
%.assembled.vect67 = insertelement <2 x i32> undef, i32 %simdShuffleXor74, i64 0
%.assembled.vect68 = insertelement <2 x i32> %.assembled.vect67, i32 %simdShuffleXor75, i64 1
%4 = bitcast <2 x i32> %.assembled.vect68 to i64
%add9 = add nsw i64 %add7, %4
ret i64 %add9
}
declare i32 @llvm.genx.GenISA.WaveShuffleIndex.i32(i32, i32, i32)
declare i32 @llvm.genx.GenISA.simdShuffleXor.i32(i32, i32)
declare i16 @llvm.genx.GenISA.simdLaneId()
declare i64 @get_i64()
!igc.functions = !{!0, !1, !2}
!0 = !{i64 ()* @wave_all_add_i64, !100}
!1 = !{i64 ()* @mixed_extract_insert_indices, !100}
!2 = !{i64 ()* @mixed_shuffle_lanes, !100}
!100 = !{!101, !102}
!101 = !{!"function_type", i32 0}
!102 = !{!"sub_group_size", i32 32}
|
