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
|
; RUN: llc -mtriple=amdgcn--amdhsa -verify-machineinstrs < %s | FileCheck -check-prefix=GCN -check-prefix=CI %s
; RUN: llc -mtriple=amdgcn--amdhsa -mcpu=tonga -verify-machineinstrs < %s | FileCheck -check-prefix=GCN -check-prefix=VI %s
; RUN: llc -mtriple=amdgcn--amdhsa -mcpu=gfx901 -verify-machineinstrs < %s | FileCheck -check-prefix=GCN -check-prefix=GFX9 %s
; DAGCombiner will transform:
; (fabs (f16 bitcast (i16 a))) => (f16 bitcast (and (i16 a), 0x7FFFFFFF))
; unless isFabsFree returns true
; GCN-LABEL: {{^}}s_fabs_free_f16:
; GCN: flat_load_ushort [[VAL:v[0-9]+]],
; GCN: v_and_b32_e32 [[RESULT:v[0-9]+]], 0x7fff, [[VAL]]
; GCN: flat_store_short v{{\[[0-9]+:[0-9]+\]}}, [[RESULT]]
define amdgpu_kernel void @s_fabs_free_f16(half addrspace(1)* %out, i16 %in) {
%bc= bitcast i16 %in to half
%fabs = call half @llvm.fabs.f16(half %bc)
store half %fabs, half addrspace(1)* %out
ret void
}
; GCN-LABEL: {{^}}s_fabs_f16:
; CI: flat_load_ushort [[VAL:v[0-9]+]],
; CI: v_and_b32_e32 [[CVT0:v[0-9]+]], 0x7fff, [[VAL]]
; CI: flat_store_short v{{\[[0-9]+:[0-9]+\]}}, [[RESULT]]
define amdgpu_kernel void @s_fabs_f16(half addrspace(1)* %out, half %in) {
%fabs = call half @llvm.fabs.f16(half %in)
store half %fabs, half addrspace(1)* %out
ret void
}
; FIXME: Should be able to use single and
; GCN-LABEL: {{^}}s_fabs_v2f16:
; CI: s_movk_i32 [[MASK:s[0-9]+]], 0x7fff
; CI: v_and_b32_e32 v{{[0-9]+}}, [[MASK]]
; CI: v_lshlrev_b32_e32 v{{[0-9]+}}, 16,
; CI: v_and_b32_e32 v{{[0-9]+}}, [[MASK]]
; CI: v_or_b32_e32
; VI: flat_load_ushort [[HI:v[0-9]+]]
; VI: flat_load_ushort [[LO:v[0-9]+]]
; VI: v_mov_b32_e32 [[MASK:v[0-9]+]], 0x7fff{{$}}
; VI-DAG: v_and_b32_e32 [[FABS_LO:v[0-9]+]], [[MASK]], [[HI]]
; VI-DAG: v_and_b32_sdwa [[FABS_HI:v[0-9]+]], [[LO]], [[MASK]] dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
; VI-DAG: v_or_b32_e32 v{{[0-9]+}}, [[FABS_HI]], [[FABS_LO]]
; VI: flat_store_dword
; GFX9: s_load_dword [[VAL:s[0-9]+]]
; GFX9: s_and_b32 s{{[0-9]+}}, [[VAL]], 0x7fff7fff
define amdgpu_kernel void @s_fabs_v2f16(<2 x half> addrspace(1)* %out, <2 x half> %in) {
%fabs = call <2 x half> @llvm.fabs.v2f16(<2 x half> %in)
store <2 x half> %fabs, <2 x half> addrspace(1)* %out
ret void
}
; GCN-LABEL: {{^}}s_fabs_v4f16:
; CI: s_movk_i32 [[MASK:s[0-9]+]], 0x7fff
; CI: v_and_b32_e32 v{{[0-9]+}}, [[MASK]]
; CI: v_and_b32_e32 v{{[0-9]+}}, [[MASK]]
; CI: v_and_b32_e32 v{{[0-9]+}}, [[MASK]]
; CI: v_and_b32_e32 v{{[0-9]+}}, [[MASK]]
; VI: v_mov_b32_e32 [[MASK:v[0-9]+]], 0x7fff{{$}}
; VI-DAG: v_and_b32_sdwa v{{[0-9]+}}, v{{[0-9]+}}, [[MASK]] dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
; VI-DAG: v_and_b32_sdwa v{{[0-9]+}}, v{{[0-9]+}}, [[MASK]] dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
; VI-DAG: v_and_b32_e32 v{{[0-9]+}}, [[MASK]], v{{[0-9]+}}
; VI-DAG: v_and_b32_e32 v{{[0-9]+}}, [[MASK]], v{{[0-9]+}}
; VI-DAG: v_or_b32_e32 v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}
; VI: v_or_b32_e32 v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}
; GCN: flat_store_dwordx2
define amdgpu_kernel void @s_fabs_v4f16(<4 x half> addrspace(1)* %out, <4 x half> %in) {
%fabs = call <4 x half> @llvm.fabs.v4f16(<4 x half> %in)
store <4 x half> %fabs, <4 x half> addrspace(1)* %out
ret void
}
; GCN-LABEL: {{^}}fabs_fold_f16:
; GCN: flat_load_ushort [[IN0:v[0-9]+]]
; GCN: flat_load_ushort [[IN1:v[0-9]+]]
; CI-DAG: v_cvt_f32_f16_e32 [[CVT0:v[0-9]+]], [[IN0]]
; CI-DAG: v_cvt_f32_f16_e64 [[ABS_CVT1:v[0-9]+]], |[[IN1]]|
; CI: v_mul_f32_e32 [[RESULT:v[0-9]+]], [[CVT0]], [[ABS_CVT1]]
; CI: v_cvt_f16_f32_e32 [[CVTRESULT:v[0-9]+]], [[RESULT]]
; CI: flat_store_short v{{\[[0-9]+:[0-9]+\]}}, [[CVTRESULT]]
; VI-NOT: and
; VI: v_mul_f16_e64 [[RESULT:v[0-9]+]], |[[IN1]]|, [[IN0]]
; VI: flat_store_short v{{\[[0-9]+:[0-9]+\]}}, [[RESULT]]
define amdgpu_kernel void @fabs_fold_f16(half addrspace(1)* %out, half %in0, half %in1) {
%fabs = call half @llvm.fabs.f16(half %in0)
%fmul = fmul half %fabs, %in1
store half %fmul, half addrspace(1)* %out
ret void
}
; GCN-LABEL: {{^}}v_fabs_v2f16:
; GCN: flat_load_dword [[VAL:v[0-9]+]]
; GCN: v_and_b32_e32 v{{[0-9]+}}, 0x7fff7fff, [[VAL]]
define amdgpu_kernel void @v_fabs_v2f16(<2 x half> addrspace(1)* %out, <2 x half> addrspace(1)* %in) #0 {
%tid = call i32 @llvm.amdgcn.workitem.id.x()
%gep.in = getelementptr inbounds <2 x half>, <2 x half> addrspace(1)* %in, i32 %tid
%gep.out = getelementptr inbounds <2 x half>, <2 x half> addrspace(1)* %in, i32 %tid
%val = load <2 x half>, <2 x half> addrspace(1)* %gep.in, align 2
%fabs = call <2 x half> @llvm.fabs.v2f16(<2 x half> %val)
store <2 x half> %fabs, <2 x half> addrspace(1)* %gep.out
ret void
}
; GCN-LABEL: {{^}}fabs_free_v2f16:
; GCN: s_load_dword [[VAL:s[0-9]+]]
; GCN: s_and_b32 s{{[0-9]+}}, [[VAL]], 0x7fff7fff
define amdgpu_kernel void @fabs_free_v2f16(<2 x half> addrspace(1)* %out, i32 %in) #0 {
%bc = bitcast i32 %in to <2 x half>
%fabs = call <2 x half> @llvm.fabs.v2f16(<2 x half> %bc)
store <2 x half> %fabs, <2 x half> addrspace(1)* %out
ret void
}
; GCN-LABEL: {{^}}v_fabs_fold_v2f16:
; GCN: flat_load_dword [[VAL:v[0-9]+]]
; CI: v_cvt_f32_f16_e32
; CI: v_cvt_f32_f16_e32
; CI: v_mul_f32_e64 v{{[0-9]+}}, |v{{[0-9]+}}|, v{{[0-9]+}}
; CI: v_cvt_f16_f32
; CI: v_mul_f32_e64 v{{[0-9]+}}, |v{{[0-9]+}}|, v{{[0-9]+}}
; CI: v_cvt_f16_f32
; VI: v_lshrrev_b32_e32 v{{[0-9]+}}, 16,
; VI: v_mul_f16_sdwa v{{[0-9]+}}, |v{{[0-9]+}}|, v{{[0-9]+}} dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
; VI: v_mul_f16_e64 v{{[0-9]+}}, |v{{[0-9]+}}|, v{{[0-9]+}}
; GFX9: v_and_b32_e32 [[FABS:v[0-9]+]], 0x7fff7fff, [[VAL]]
; GFX9: v_pk_mul_f16 v{{[0-9]+}}, [[FABS]], v{{[0-9]+$}}
define amdgpu_kernel void @v_fabs_fold_v2f16(<2 x half> addrspace(1)* %out, <2 x half> addrspace(1)* %in) #0 {
%val = load <2 x half>, <2 x half> addrspace(1)* %in
%fabs = call <2 x half> @llvm.fabs.v2f16(<2 x half> %val)
%fmul = fmul <2 x half> %fabs, %val
store <2 x half> %fmul, <2 x half> addrspace(1)* %out
ret void
}
declare half @llvm.fabs.f16(half) #1
declare <2 x half> @llvm.fabs.v2f16(<2 x half>) #1
declare <4 x half> @llvm.fabs.v4f16(<4 x half>) #1
declare i32 @llvm.amdgcn.workitem.id.x() #1
attributes #0 = { nounwind }
attributes #1 = { nounwind readnone }
|
