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
|
; RUN: llc -march=amdgcn -verify-machineinstrs < %s | FileCheck -check-prefix=SI -check-prefix=FUNC %s
declare i32 @llvm.amdgcn.workitem.id.x() #1
declare float @llvm.fabs.f32(float) #1
declare float @llvm.fma.f32(float, float, float) nounwind readnone
; FUNC-LABEL: @commute_add_imm_fabs_f32
; SI: buffer_load_dword [[X:v[0-9]+]], {{v\[[0-9]+:[0-9]+\]}}, {{s\[[0-9]+:[0-9]+\]}}, 0 addr64{{$}}
; SI: v_add_f32_e64 [[REG:v[0-9]+]], |[[X]]|, 2.0
; SI: buffer_store_dword [[REG]]
define amdgpu_kernel void @commute_add_imm_fabs_f32(float addrspace(1)* %out, float addrspace(1)* %in) #0 {
%tid = call i32 @llvm.amdgcn.workitem.id.x() #1
%gep.0 = getelementptr float, float addrspace(1)* %in, i32 %tid
%x = load float, float addrspace(1)* %gep.0
%x.fabs = call float @llvm.fabs.f32(float %x) #1
%z = fadd float 2.0, %x.fabs
store float %z, float addrspace(1)* %out
ret void
}
; FUNC-LABEL: @commute_mul_imm_fneg_fabs_f32
; SI: buffer_load_dword [[X:v[0-9]+]], {{v\[[0-9]+:[0-9]+\]}}, {{s\[[0-9]+:[0-9]+\]}}, 0 addr64{{$}}
; SI: v_mul_f32_e64 [[REG:v[0-9]+]], |[[X]]|, -4.0
; SI: buffer_store_dword [[REG]]
define amdgpu_kernel void @commute_mul_imm_fneg_fabs_f32(float addrspace(1)* %out, float addrspace(1)* %in) #0 {
%tid = call i32 @llvm.amdgcn.workitem.id.x() #1
%gep.0 = getelementptr float, float addrspace(1)* %in, i32 %tid
%x = load float, float addrspace(1)* %gep.0
%x.fabs = call float @llvm.fabs.f32(float %x) #1
%x.fneg.fabs = fsub float -0.000000e+00, %x.fabs
%z = fmul float 4.0, %x.fneg.fabs
store float %z, float addrspace(1)* %out
ret void
}
; FUNC-LABEL: @commute_mul_imm_fneg_f32
; SI: buffer_load_dword [[X:v[0-9]+]], {{v\[[0-9]+:[0-9]+\]}}, {{s\[[0-9]+:[0-9]+\]}}, 0 addr64{{$}}
; SI: v_mul_f32_e32 [[REG:v[0-9]+]], -4.0, [[X]]
; SI: buffer_store_dword [[REG]]
define amdgpu_kernel void @commute_mul_imm_fneg_f32(float addrspace(1)* %out, float addrspace(1)* %in) #0 {
%tid = call i32 @llvm.amdgcn.workitem.id.x() #1
%gep.0 = getelementptr float, float addrspace(1)* %in, i32 %tid
%x = load float, float addrspace(1)* %gep.0
%x.fneg = fsub float -0.000000e+00, %x
%z = fmul float 4.0, %x.fneg
store float %z, float addrspace(1)* %out
ret void
}
; FIXME: Should use SGPR for literal.
; FUNC-LABEL: @commute_add_lit_fabs_f32
; SI: buffer_load_dword [[X:v[0-9]+]], {{v\[[0-9]+:[0-9]+\]}}, {{s\[[0-9]+:[0-9]+\]}}, 0 addr64{{$}}
; SI: s_mov_b32 [[K:s[0-9]+]], 0x44800000
; SI: v_add_f32_e64 [[REG:v[0-9]+]], |[[X]]|, [[K]]
; SI: buffer_store_dword [[REG]]
define amdgpu_kernel void @commute_add_lit_fabs_f32(float addrspace(1)* %out, float addrspace(1)* %in) #0 {
%tid = call i32 @llvm.amdgcn.workitem.id.x() #1
%gep.0 = getelementptr float, float addrspace(1)* %in, i32 %tid
%x = load float, float addrspace(1)* %gep.0
%x.fabs = call float @llvm.fabs.f32(float %x) #1
%z = fadd float 1024.0, %x.fabs
store float %z, float addrspace(1)* %out
ret void
}
; FUNC-LABEL: @commute_add_fabs_f32
; SI-DAG: buffer_load_dword [[X:v[0-9]+]], {{v\[[0-9]+:[0-9]+\]}}, {{s\[[0-9]+:[0-9]+\]}}, 0 addr64 glc{{$}}
; SI-DAG: buffer_load_dword [[Y:v[0-9]+]], {{v\[[0-9]+:[0-9]+\]}}, {{s\[[0-9]+:[0-9]+\]}}, 0 addr64 offset:4
; SI: v_add_f32_e64 [[REG:v[0-9]+]], [[X]], |[[Y]]|
; SI: buffer_store_dword [[REG]]
define amdgpu_kernel void @commute_add_fabs_f32(float addrspace(1)* %out, float addrspace(1)* %in) #0 {
%tid = call i32 @llvm.amdgcn.workitem.id.x() #1
%gep.0 = getelementptr float, float addrspace(1)* %in, i32 %tid
%gep.1 = getelementptr float, float addrspace(1)* %gep.0, i32 1
%x = load volatile float, float addrspace(1)* %gep.0
%y = load volatile float, float addrspace(1)* %gep.1
%y.fabs = call float @llvm.fabs.f32(float %y) #1
%z = fadd float %x, %y.fabs
store float %z, float addrspace(1)* %out
ret void
}
; FUNC-LABEL: @commute_mul_fneg_f32
; SI-DAG: buffer_load_dword [[X:v[0-9]+]], {{v\[[0-9]+:[0-9]+\]}}, {{s\[[0-9]+:[0-9]+\]}}, 0 addr64 glc{{$}}
; SI-DAG: buffer_load_dword [[Y:v[0-9]+]], {{v\[[0-9]+:[0-9]+\]}}, {{s\[[0-9]+:[0-9]+\]}}, 0 addr64 offset:4
; SI: v_mul_f32_e64 [[REG:v[0-9]+]], [[X]], -[[Y]]
; SI: buffer_store_dword [[REG]]
define amdgpu_kernel void @commute_mul_fneg_f32(float addrspace(1)* %out, float addrspace(1)* %in) #0 {
%tid = call i32 @llvm.amdgcn.workitem.id.x() #1
%gep.0 = getelementptr float, float addrspace(1)* %in, i32 %tid
%gep.1 = getelementptr float, float addrspace(1)* %gep.0, i32 1
%x = load volatile float, float addrspace(1)* %gep.0
%y = load volatile float, float addrspace(1)* %gep.1
%y.fneg = fsub float -0.000000e+00, %y
%z = fmul float %x, %y.fneg
store float %z, float addrspace(1)* %out
ret void
}
; FUNC-LABEL: @commute_mul_fabs_fneg_f32
; SI-DAG: buffer_load_dword [[X:v[0-9]+]], {{v\[[0-9]+:[0-9]+\]}}, {{s\[[0-9]+:[0-9]+\]}}, 0 addr64 glc{{$}}
; SI-DAG: buffer_load_dword [[Y:v[0-9]+]], {{v\[[0-9]+:[0-9]+\]}}, {{s\[[0-9]+:[0-9]+\]}}, 0 addr64 offset:4
; SI: v_mul_f32_e64 [[REG:v[0-9]+]], [[X]], -|[[Y]]|
; SI: buffer_store_dword [[REG]]
define amdgpu_kernel void @commute_mul_fabs_fneg_f32(float addrspace(1)* %out, float addrspace(1)* %in) #0 {
%tid = call i32 @llvm.amdgcn.workitem.id.x() #1
%gep.0 = getelementptr float, float addrspace(1)* %in, i32 %tid
%gep.1 = getelementptr float, float addrspace(1)* %gep.0, i32 1
%x = load volatile float, float addrspace(1)* %gep.0
%y = load volatile float, float addrspace(1)* %gep.1
%y.fabs = call float @llvm.fabs.f32(float %y) #1
%y.fabs.fneg = fsub float -0.000000e+00, %y.fabs
%z = fmul float %x, %y.fabs.fneg
store float %z, float addrspace(1)* %out
ret void
}
; There's no reason to commute this.
; FUNC-LABEL: @commute_mul_fabs_x_fabs_y_f32
; SI-DAG: buffer_load_dword [[X:v[0-9]+]], {{v\[[0-9]+:[0-9]+\]}}, {{s\[[0-9]+:[0-9]+\]}}, 0 addr64 glc{{$}}
; SI-DAG: buffer_load_dword [[Y:v[0-9]+]], {{v\[[0-9]+:[0-9]+\]}}, {{s\[[0-9]+:[0-9]+\]}}, 0 addr64 offset:4
; SI: v_mul_f32_e64 [[REG:v[0-9]+]], |[[X]]|, |[[Y]]|
; SI: buffer_store_dword [[REG]]
define amdgpu_kernel void @commute_mul_fabs_x_fabs_y_f32(float addrspace(1)* %out, float addrspace(1)* %in) #0 {
%tid = call i32 @llvm.amdgcn.workitem.id.x() #1
%gep.0 = getelementptr float, float addrspace(1)* %in, i32 %tid
%gep.1 = getelementptr float, float addrspace(1)* %gep.0, i32 1
%x = load volatile float, float addrspace(1)* %gep.0
%y = load volatile float, float addrspace(1)* %gep.1
%x.fabs = call float @llvm.fabs.f32(float %x) #1
%y.fabs = call float @llvm.fabs.f32(float %y) #1
%z = fmul float %x.fabs, %y.fabs
store float %z, float addrspace(1)* %out
ret void
}
; FUNC-LABEL: @commute_mul_fabs_x_fneg_fabs_y_f32
; SI-DAG: buffer_load_dword [[X:v[0-9]+]], {{v\[[0-9]+:[0-9]+\]}}, {{s\[[0-9]+:[0-9]+\]}}, 0 addr64 glc{{$}}
; SI-DAG: buffer_load_dword [[Y:v[0-9]+]], {{v\[[0-9]+:[0-9]+\]}}, {{s\[[0-9]+:[0-9]+\]}}, 0 addr64 offset:4
; SI: v_mul_f32_e64 [[REG:v[0-9]+]], |[[X]]|, -|[[Y]]|
; SI: buffer_store_dword [[REG]]
define amdgpu_kernel void @commute_mul_fabs_x_fneg_fabs_y_f32(float addrspace(1)* %out, float addrspace(1)* %in) #0 {
%tid = call i32 @llvm.amdgcn.workitem.id.x() #1
%gep.0 = getelementptr float, float addrspace(1)* %in, i32 %tid
%gep.1 = getelementptr float, float addrspace(1)* %gep.0, i32 1
%x = load volatile float, float addrspace(1)* %gep.0
%y = load volatile float, float addrspace(1)* %gep.1
%x.fabs = call float @llvm.fabs.f32(float %x) #1
%y.fabs = call float @llvm.fabs.f32(float %y) #1
%y.fabs.fneg = fsub float -0.000000e+00, %y.fabs
%z = fmul float %x.fabs, %y.fabs.fneg
store float %z, float addrspace(1)* %out
ret void
}
; Make sure we commute the multiply part for the constant in src0 even
; though we have negate modifier on src2.
; SI-LABEL: {{^}}fma_a_2.0_neg_b_f32
; SI-DAG: buffer_load_dword [[R1:v[0-9]+]], {{v\[[0-9]+:[0-9]+\]}}, {{s\[[0-9]+:[0-9]+\]}}, 0 addr64 glc{{$}}
; SI-DAG: buffer_load_dword [[R2:v[0-9]+]], {{v\[[0-9]+:[0-9]+\]}}, {{s\[[0-9]+:[0-9]+\]}}, 0 addr64 offset:4
; SI: v_fma_f32 [[RESULT:v[0-9]+]], [[R1]], 2.0, |[[R2]]|
; SI: buffer_store_dword [[RESULT]]
define amdgpu_kernel void @fma_a_2.0_neg_b_f32(float addrspace(1)* %out, float addrspace(1)* %in) {
%tid = call i32 @llvm.amdgcn.workitem.id.x() nounwind readnone
%gep.0 = getelementptr float, float addrspace(1)* %out, i32 %tid
%gep.1 = getelementptr float, float addrspace(1)* %gep.0, i32 1
%gep.out = getelementptr float, float addrspace(1)* %out, i32 %tid
%r1 = load volatile float, float addrspace(1)* %gep.0
%r2 = load volatile float, float addrspace(1)* %gep.1
%r2.fabs = call float @llvm.fabs.f32(float %r2)
%r3 = tail call float @llvm.fma.f32(float %r1, float 2.0, float %r2.fabs)
store float %r3, float addrspace(1)* %gep.out
ret void
}
attributes #0 = { nounwind }
attributes #1 = { nounwind readnone }
|
