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
179
180
181
182
183
184
185
186
187
|
; RUN: llc -march=amdgcn -verify-machineinstrs < %s | FileCheck -allow-deprecated-dag-overlap -enable-var-scope -check-prefixes=GCN,FUNC %s
; RUN: llc -march=amdgcn -mcpu=tonga -mattr=-flat-for-global -verify-machineinstrs < %s | FileCheck -allow-deprecated-dag-overlap -enable-var-scope -check-prefixes=GCN,FUNC %s
; RUN: llc -march=r600 -mcpu=redwood < %s | FileCheck -allow-deprecated-dag-overlap -enable-var-scope -check-prefixes=R600,FUNC %s
; BFI_INT Definition pattern from ISA docs
; (y & x) | (z & ~x)
;
; FUNC-LABEL: {{^}}bfi_def:
; R600: BFI_INT
; GCN: v_bfi_b32
define amdgpu_kernel void @bfi_def(i32 addrspace(1)* %out, i32 %x, i32 %y, i32 %z) {
entry:
%0 = xor i32 %x, -1
%1 = and i32 %z, %0
%2 = and i32 %y, %x
%3 = or i32 %1, %2
store i32 %3, i32 addrspace(1)* %out
ret void
}
; SHA-256 Ch function
; z ^ (x & (y ^ z))
; FUNC-LABEL: {{^}}bfi_sha256_ch:
; R600: BFI_INT
; GCN: v_bfi_b32
define amdgpu_kernel void @bfi_sha256_ch(i32 addrspace(1)* %out, i32 %x, i32 %y, i32 %z) {
entry:
%0 = xor i32 %y, %z
%1 = and i32 %x, %0
%2 = xor i32 %z, %1
store i32 %2, i32 addrspace(1)* %out
ret void
}
; SHA-256 Ma function
; ((x & z) | (y & (x | z)))
; FUNC-LABEL: {{^}}bfi_sha256_ma:
; R600: XOR_INT * [[DST:T[0-9]+\.[XYZW]]], KC0[2].Z, KC0[2].W
; R600: BFI_INT * {{T[0-9]+\.[XYZW]}}, {{[[DST]]|PV\.[XYZW]}}, KC0[3].X, KC0[2].W
; GCN: v_xor_b32_e32 [[DST:v[0-9]+]], {{s[0-9]+, v[0-9]+}}
; GCN: v_bfi_b32 {{v[0-9]+}}, [[DST]], {{s[0-9]+, v[0-9]+}}
define amdgpu_kernel void @bfi_sha256_ma(i32 addrspace(1)* %out, i32 %x, i32 %y, i32 %z) {
entry:
%0 = and i32 %x, %z
%1 = or i32 %x, %z
%2 = and i32 %y, %1
%3 = or i32 %0, %2
store i32 %3, i32 addrspace(1)* %out
ret void
}
; FUNC-LABEL: {{^}}v_bitselect_v2i32_pat1:
; GCN: s_waitcnt
; GCN-NEXT: v_bfi_b32 v0, v2, v0, v4
; GCN-NEXT: v_bfi_b32 v1, v3, v1, v5
; GCN-NEXT: s_setpc_b64
define <2 x i32> @v_bitselect_v2i32_pat1(<2 x i32> %a, <2 x i32> %b, <2 x i32> %mask) {
%xor.0 = xor <2 x i32> %a, %mask
%and = and <2 x i32> %xor.0, %b
%bitselect = xor <2 x i32> %and, %mask
ret <2 x i32> %bitselect
}
; FUNC-LABEL: {{^}}v_bitselect_i64_pat_0:
; GCN: s_waitcnt
; GCN-NEXT: v_bfi_b32 v1, v1, v3, v5
; GCN-NEXT: v_bfi_b32 v0, v0, v2, v4
; GCN-NEXT: s_setpc_b64
define i64 @v_bitselect_i64_pat_0(i64 %a, i64 %b, i64 %mask) {
%and0 = and i64 %a, %b
%not.a = xor i64 %a, -1
%and1 = and i64 %not.a, %mask
%bitselect = or i64 %and0, %and1
ret i64 %bitselect
}
; FUNC-LABEL: {{^}}v_bitselect_i64_pat_1:
; GCN: s_waitcnt
; GCN-NEXT: v_bfi_b32 v1, v3, v1, v5
; GCN-NEXT: v_bfi_b32 v0, v2, v0, v4
; GCN-NEXT: s_setpc_b64
define i64 @v_bitselect_i64_pat_1(i64 %a, i64 %b, i64 %mask) {
%xor.0 = xor i64 %a, %mask
%and = and i64 %xor.0, %b
%bitselect = xor i64 %and, %mask
ret i64 %bitselect
}
; FUNC-LABEL: {{^}}v_bitselect_i64_pat_2:
; GCN: s_waitcnt
; GCN-DAG: v_bfi_b32 v0, v2, v0, v4
; GCN-DAG: v_bfi_b32 v1, v3, v1, v5
; GCN-NEXT: s_setpc_b64
define i64 @v_bitselect_i64_pat_2(i64 %a, i64 %b, i64 %mask) {
%xor.0 = xor i64 %a, %mask
%and = and i64 %xor.0, %b
%bitselect = xor i64 %and, %mask
ret i64 %bitselect
}
; FUNC-LABEL: {{^}}v_bfi_sha256_ma_i64:
; GCN-DAG: v_xor_b32_e32 v1, v1, v3
; GCN-DAG: v_xor_b32_e32 v0, v0, v2
; GCN-DAG: v_bfi_b32 v1, v1, v5, v3
; GCN-DAG: v_bfi_b32 v0, v0, v4, v2
define i64 @v_bfi_sha256_ma_i64(i64 %x, i64 %y, i64 %z) {
entry:
%and0 = and i64 %x, %z
%or0 = or i64 %x, %z
%and1 = and i64 %y, %or0
%or1 = or i64 %and0, %and1
ret i64 %or1
}
; FIXME: Should leave as 64-bit SALU ops
; FUNC-LABEL: {{^}}s_bitselect_i64_pat_0:
; GCN: v_mov_b32_e32 v{{[0-9]+}}, s
; GCN: v_mov_b32_e32 v{{[0-9]+}}, s
; GCN: v_mov_b32_e32 v{{[0-9]+}}, s
; GCN-DAG: v_mov_b32_e32 v{{[0-9]+}}, s
; GCN-DAG: v_bfi_b32
; GCN-DAG: v_bfi_b32
define amdgpu_kernel void @s_bitselect_i64_pat_0(i64 %a, i64 %b, i64 %mask) {
%and0 = and i64 %a, %b
%not.a = xor i64 %a, -1
%and1 = and i64 %not.a, %mask
%bitselect = or i64 %and0, %and1
%scalar.use = add i64 %bitselect, 10
store i64 %scalar.use, i64 addrspace(1)* undef
ret void
}
; FUNC-LABEL: {{^}}s_bitselect_i64_pat_1:
; GCN: v_mov_b32_e32 v{{[0-9]+}}, s
; GCN: v_mov_b32_e32 v{{[0-9]+}}, s
; GCN-DAG: v_mov_b32_e32 v{{[0-9]+}}, s
; GCN-DAG: v_mov_b32_e32 v{{[0-9]+}}, s
; GCN-DAG: v_bfi_b32
; GCN-DAG: v_bfi_b32
define amdgpu_kernel void @s_bitselect_i64_pat_1(i64 %a, i64 %b, i64 %mask) {
%xor.0 = xor i64 %a, %mask
%and = and i64 %xor.0, %b
%bitselect = xor i64 %and, %mask
%scalar.use = add i64 %bitselect, 10
store i64 %scalar.use, i64 addrspace(1)* undef
ret void
}
; FUNC-LABEL: {{^}}s_bitselect_i64_pat_2:
; GCN: v_mov_b32_e32 v{{[0-9]+}}, s
; GCN: v_mov_b32_e32 v{{[0-9]+}}, s
; GCN-DAG: v_bfi_b32
; GCN-DAG: v_mov_b32_e32 v{{[0-9]+}}, s
; GCN-DAG: v_mov_b32_e32 v{{[0-9]+}}, s
; GCN-DAG: v_bfi_b32
define amdgpu_kernel void @s_bitselect_i64_pat_2(i64 %a, i64 %b, i64 %mask) {
%xor.0 = xor i64 %a, %mask
%and = and i64 %xor.0, %b
%bitselect = xor i64 %and, %mask
%scalar.use = add i64 %bitselect, 10
store i64 %scalar.use, i64 addrspace(1)* undef
ret void
}
; FUNC-LABEL: {{^}}s_bfi_sha256_ma_i64:
; GCN: v_mov_b32_e32 v{{[0-9]+}}, s
; GCN-DAG: v_xor_b32
; GCN-DAG: v_mov_b32_e32 v{{[0-9]+}}, s
; GCN-DAG: v_xor_b32
; GCN-DAG: v_bfi_b32
; GCN: v_bfi_b32
define amdgpu_kernel void @s_bfi_sha256_ma_i64(i64 %x, i64 %y, i64 %z) {
entry:
%and0 = and i64 %x, %z
%or0 = or i64 %x, %z
%and1 = and i64 %y, %or0
%or1 = or i64 %and0, %and1
%scalar.use = add i64 %or1, 10
store i64 %scalar.use, i64 addrspace(1)* undef
ret void
}
|
