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
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
|
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc -mtriple=r600 -mcpu=redwood < %s | FileCheck -enable-var-scope -check-prefixes=R600 %s
; BFI_INT Definition pattern from ISA docs
; (y & x) | (z & ~x)
;
define amdgpu_kernel void @bfi_def(ptr addrspace(1) %out, i32 %x, i32 %y, i32 %z) {
; R600-LABEL: bfi_def:
; R600: ; %bb.0: ; %entry
; R600-NEXT: ALU 2, @4, KC0[CB0:0-32], KC1[]
; R600-NEXT: MEM_RAT_CACHELESS STORE_RAW T1.X, T0.X, 1
; R600-NEXT: CF_END
; R600-NEXT: PAD
; R600-NEXT: ALU clause starting at 4:
; R600-NEXT: LSHR * T0.X, KC0[2].Y, literal.x,
; R600-NEXT: 2(2.802597e-45), 0(0.000000e+00)
; R600-NEXT: BFI_INT * T1.X, KC0[2].Z, KC0[2].W, KC0[3].X,
entry:
%0 = xor i32 %x, -1
%1 = and i32 %z, %0
%2 = and i32 %y, %x
%3 = or i32 %1, %2
store i32 %3, ptr addrspace(1) %out
ret void
}
; SHA-256 Ch function
; z ^ (x & (y ^ z))
define amdgpu_kernel void @bfi_sha256_ch(ptr addrspace(1) %out, i32 %x, i32 %y, i32 %z) {
; R600-LABEL: bfi_sha256_ch:
; R600: ; %bb.0: ; %entry
; R600-NEXT: ALU 2, @4, KC0[CB0:0-32], KC1[]
; R600-NEXT: MEM_RAT_CACHELESS STORE_RAW T1.X, T0.X, 1
; R600-NEXT: CF_END
; R600-NEXT: PAD
; R600-NEXT: ALU clause starting at 4:
; R600-NEXT: LSHR * T0.X, KC0[2].Y, literal.x,
; R600-NEXT: 2(2.802597e-45), 0(0.000000e+00)
; R600-NEXT: BFI_INT * T1.X, KC0[2].Z, KC0[2].W, KC0[3].X,
entry:
%0 = xor i32 %y, %z
%1 = and i32 %x, %0
%2 = xor i32 %z, %1
store i32 %2, ptr addrspace(1) %out
ret void
}
; SHA-256 Ma function
; ((x & z) | (y & (x | z)))
define amdgpu_kernel void @bfi_sha256_ma(ptr addrspace(1) %out, i32 %x, i32 %y, i32 %z) {
; R600-LABEL: bfi_sha256_ma:
; R600: ; %bb.0: ; %entry
; R600-NEXT: ALU 3, @4, KC0[CB0:0-32], KC1[]
; R600-NEXT: MEM_RAT_CACHELESS STORE_RAW T0.X, T1.X, 1
; R600-NEXT: CF_END
; R600-NEXT: PAD
; R600-NEXT: ALU clause starting at 4:
; R600-NEXT: XOR_INT * T0.W, KC0[2].Z, KC0[2].W,
; R600-NEXT: BFI_INT * T0.X, PV.W, KC0[3].X, KC0[2].W,
; R600-NEXT: LSHR * T1.X, KC0[2].Y, literal.x,
; R600-NEXT: 2(2.802597e-45), 0(0.000000e+00)
entry:
%0 = and i32 %x, %z
%1 = or i32 %x, %z
%2 = and i32 %y, %1
%3 = or i32 %0, %2
store i32 %3, ptr addrspace(1) %out
ret void
}
define <2 x i32> @v_bitselect_v2i32_pat1(<2 x i32> %a, <2 x i32> %b, <2 x i32> %mask) {
; R600-LABEL: v_bitselect_v2i32_pat1:
; R600: ; %bb.0:
; R600-NEXT: CF_END
; R600-NEXT: PAD
%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
}
define i64 @v_bitselect_i64_pat_0(i64 %a, i64 %b, i64 %mask) {
; R600-LABEL: v_bitselect_i64_pat_0:
; R600: ; %bb.0:
; R600-NEXT: CF_END
; R600-NEXT: PAD
%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
}
define i64 @v_bitselect_i64_pat_1(i64 %a, i64 %b, i64 %mask) {
; R600-LABEL: v_bitselect_i64_pat_1:
; R600: ; %bb.0:
; R600-NEXT: CF_END
; R600-NEXT: PAD
%xor.0 = xor i64 %a, %mask
%and = and i64 %xor.0, %b
%bitselect = xor i64 %and, %mask
ret i64 %bitselect
}
define i64 @v_bitselect_i64_pat_2(i64 %a, i64 %b, i64 %mask) {
; R600-LABEL: v_bitselect_i64_pat_2:
; R600: ; %bb.0:
; R600-NEXT: CF_END
; R600-NEXT: PAD
%xor.0 = xor i64 %a, %mask
%and = and i64 %xor.0, %b
%bitselect = xor i64 %and, %mask
ret i64 %bitselect
}
define i64 @v_bfi_sha256_ma_i64(i64 %x, i64 %y, i64 %z) {
; R600-LABEL: v_bfi_sha256_ma_i64:
; R600: ; %bb.0: ; %entry
; R600-NEXT: CF_END
; R600-NEXT: PAD
entry:
%and0 = and i64 %x, %z
%or0 = or i64 %x, %z
%and1 = and i64 %y, %or0
%or1 = or i64 %and0, %and1
ret i64 %or1
}
define amdgpu_kernel void @s_bitselect_i64_pat_0(i64 %a, i64 %b, i64 %mask) {
; R600-LABEL: s_bitselect_i64_pat_0:
; R600: ; %bb.0:
; R600-NEXT: ALU 9, @4, KC0[CB0:0-32], KC1[]
; R600-NEXT: MEM_RAT_CACHELESS STORE_RAW T0.XY, T1.X, 1
; R600-NEXT: CF_END
; R600-NEXT: PAD
; R600-NEXT: ALU clause starting at 4:
; R600-NEXT: MOV * T0.W, KC0[3].Y,
; R600-NEXT: BFI_INT * T0.W, KC0[2].Y, KC0[2].W, PV.W,
; R600-NEXT: MOV * T1.W, KC0[3].Z,
; R600-NEXT: BFI_INT T1.W, KC0[2].Z, KC0[3].X, PV.W,
; R600-NEXT: ADDC_UINT * T2.W, T0.W, literal.x,
; R600-NEXT: 10(1.401298e-44), 0(0.000000e+00)
; R600-NEXT: ADD_INT * T0.Y, PV.W, PS,
; R600-NEXT: ADD_INT T0.X, T0.W, literal.x,
; R600-NEXT: MOV * T1.X, literal.y,
; R600-NEXT: 10(1.401298e-44), 0(0.000000e+00)
%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, ptr addrspace(1) poison
ret void
}
define amdgpu_kernel void @s_bitselect_i64_pat_1(i64 %a, i64 %b, i64 %mask) {
; R600-LABEL: s_bitselect_i64_pat_1:
; R600: ; %bb.0:
; R600-NEXT: ALU 9, @4, KC0[CB0:0-32], KC1[]
; R600-NEXT: MEM_RAT_CACHELESS STORE_RAW T0.XY, T1.X, 1
; R600-NEXT: CF_END
; R600-NEXT: PAD
; R600-NEXT: ALU clause starting at 4:
; R600-NEXT: MOV * T0.W, KC0[3].Y,
; R600-NEXT: BFI_INT * T0.W, KC0[2].W, KC0[2].Y, PV.W,
; R600-NEXT: MOV * T1.W, KC0[3].Z,
; R600-NEXT: BFI_INT T1.W, KC0[3].X, KC0[2].Z, PV.W,
; R600-NEXT: ADDC_UINT * T2.W, T0.W, literal.x,
; R600-NEXT: 10(1.401298e-44), 0(0.000000e+00)
; R600-NEXT: ADD_INT * T0.Y, PV.W, PS,
; R600-NEXT: ADD_INT T0.X, T0.W, literal.x,
; R600-NEXT: MOV * T1.X, literal.y,
; R600-NEXT: 10(1.401298e-44), 0(0.000000e+00)
%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, ptr addrspace(1) poison
ret void
}
define amdgpu_kernel void @s_bitselect_i64_pat_2(i64 %a, i64 %b, i64 %mask) {
; R600-LABEL: s_bitselect_i64_pat_2:
; R600: ; %bb.0:
; R600-NEXT: ALU 9, @4, KC0[CB0:0-32], KC1[]
; R600-NEXT: MEM_RAT_CACHELESS STORE_RAW T0.XY, T1.X, 1
; R600-NEXT: CF_END
; R600-NEXT: PAD
; R600-NEXT: ALU clause starting at 4:
; R600-NEXT: MOV * T0.W, KC0[3].Y,
; R600-NEXT: BFI_INT * T0.W, KC0[2].W, KC0[2].Y, PV.W,
; R600-NEXT: MOV * T1.W, KC0[3].Z,
; R600-NEXT: BFI_INT T1.W, KC0[3].X, KC0[2].Z, PV.W,
; R600-NEXT: ADDC_UINT * T2.W, T0.W, literal.x,
; R600-NEXT: 10(1.401298e-44), 0(0.000000e+00)
; R600-NEXT: ADD_INT * T0.Y, PV.W, PS,
; R600-NEXT: ADD_INT T0.X, T0.W, literal.x,
; R600-NEXT: MOV * T1.X, literal.y,
; R600-NEXT: 10(1.401298e-44), 0(0.000000e+00)
%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, ptr addrspace(1) poison
ret void
}
define amdgpu_kernel void @s_bfi_sha256_ma_i64(i64 %x, i64 %y, i64 %z) {
; R600-LABEL: s_bfi_sha256_ma_i64:
; R600: ; %bb.0: ; %entry
; R600-NEXT: ALU 9, @4, KC0[CB0:0-32], KC1[]
; R600-NEXT: MEM_RAT_CACHELESS STORE_RAW T0.XY, T1.X, 1
; R600-NEXT: CF_END
; R600-NEXT: PAD
; R600-NEXT: ALU clause starting at 4:
; R600-NEXT: XOR_INT * T0.W, KC0[2].Y, KC0[2].W,
; R600-NEXT: BFI_INT T0.W, PV.W, KC0[3].Y, KC0[2].W,
; R600-NEXT: XOR_INT * T1.W, KC0[2].Z, KC0[3].X,
; R600-NEXT: BFI_INT T1.W, PS, KC0[3].Z, KC0[3].X,
; R600-NEXT: ADDC_UINT * T2.W, PV.W, literal.x,
; R600-NEXT: 10(1.401298e-44), 0(0.000000e+00)
; R600-NEXT: ADD_INT * T0.Y, PV.W, PS,
; R600-NEXT: ADD_INT T0.X, T0.W, literal.x,
; R600-NEXT: MOV * T1.X, literal.y,
; R600-NEXT: 10(1.401298e-44), 0(0.000000e+00)
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, ptr addrspace(1) poison
ret void
}
|
