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
|
; NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
; RUN: llc -mtriple=amdgcn -mcpu=gfx90a -verify-machineinstrs -stop-after=finalize-isel < %s | FileCheck -check-prefix=GFX90A_GFX942 %s
; RUN: llc -mtriple=amdgcn -mcpu=gfx942 -verify-machineinstrs -stop-after=finalize-isel < %s | FileCheck -check-prefix=GFX90A_GFX942 %s
define amdgpu_ps void @flat_atomic_fadd_f64_no_rtn_intrinsic(ptr %ptr, double %data) {
; GFX90A_GFX942-LABEL: name: flat_atomic_fadd_f64_no_rtn_intrinsic
; GFX90A_GFX942: bb.0 (%ir-block.0):
; GFX90A_GFX942-NEXT: liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3
; GFX90A_GFX942-NEXT: {{ $}}
; GFX90A_GFX942-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr3
; GFX90A_GFX942-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2
; GFX90A_GFX942-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr1
; GFX90A_GFX942-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX90A_GFX942-NEXT: [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
; GFX90A_GFX942-NEXT: [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
; GFX90A_GFX942-NEXT: [[REG_SEQUENCE:%[0-9]+]]:vreg_64_align2 = REG_SEQUENCE [[COPY1]], %subreg.sub0, [[COPY]], %subreg.sub1
; GFX90A_GFX942-NEXT: [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
; GFX90A_GFX942-NEXT: [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
; GFX90A_GFX942-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:vreg_64_align2 = REG_SEQUENCE [[COPY3]], %subreg.sub0, [[COPY2]], %subreg.sub1
; GFX90A_GFX942-NEXT: [[COPY4:%[0-9]+]]:vreg_64_align2 = COPY [[REG_SEQUENCE1]]
; GFX90A_GFX942-NEXT: [[COPY5:%[0-9]+]]:vreg_64_align2 = COPY [[REG_SEQUENCE]]
; GFX90A_GFX942-NEXT: FLAT_ATOMIC_ADD_F64 killed [[COPY4]], killed [[COPY5]], 0, 0, implicit $exec, implicit $flat_scr :: (load store syncscope("agent") seq_cst (s64) on %ir.ptr)
; GFX90A_GFX942-NEXT: S_ENDPGM 0
%ret = call double @llvm.amdgcn.flat.atomic.fadd.f64.p1.f64(ptr %ptr, double %data)
ret void
}
define amdgpu_ps double @flat_atomic_fadd_f64_rtn_intrinsic(ptr %ptr, double %data) {
; GFX90A_GFX942-LABEL: name: flat_atomic_fadd_f64_rtn_intrinsic
; GFX90A_GFX942: bb.0 (%ir-block.0):
; GFX90A_GFX942-NEXT: liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3
; GFX90A_GFX942-NEXT: {{ $}}
; GFX90A_GFX942-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr3
; GFX90A_GFX942-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2
; GFX90A_GFX942-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr1
; GFX90A_GFX942-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX90A_GFX942-NEXT: [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
; GFX90A_GFX942-NEXT: [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
; GFX90A_GFX942-NEXT: [[REG_SEQUENCE:%[0-9]+]]:vreg_64_align2 = REG_SEQUENCE [[COPY1]], %subreg.sub0, [[COPY]], %subreg.sub1
; GFX90A_GFX942-NEXT: [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
; GFX90A_GFX942-NEXT: [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
; GFX90A_GFX942-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:vreg_64_align2 = REG_SEQUENCE [[COPY3]], %subreg.sub0, [[COPY2]], %subreg.sub1
; GFX90A_GFX942-NEXT: [[COPY4:%[0-9]+]]:vreg_64_align2 = COPY [[REG_SEQUENCE1]]
; GFX90A_GFX942-NEXT: [[COPY5:%[0-9]+]]:vreg_64_align2 = COPY [[REG_SEQUENCE]]
; GFX90A_GFX942-NEXT: [[FLAT_ATOMIC_ADD_F64_RTN:%[0-9]+]]:vreg_64_align2 = FLAT_ATOMIC_ADD_F64_RTN killed [[COPY4]], killed [[COPY5]], 0, 1, implicit $exec, implicit $flat_scr :: (load store syncscope("agent") seq_cst (s64) on %ir.ptr)
; GFX90A_GFX942-NEXT: [[COPY6:%[0-9]+]]:vgpr_32 = COPY [[FLAT_ATOMIC_ADD_F64_RTN]].sub0
; GFX90A_GFX942-NEXT: [[V_READFIRSTLANE_B32_:%[0-9]+]]:sreg_32_xm0 = V_READFIRSTLANE_B32 killed [[COPY6]], implicit $exec
; GFX90A_GFX942-NEXT: [[COPY7:%[0-9]+]]:vgpr_32 = COPY [[FLAT_ATOMIC_ADD_F64_RTN]].sub1
; GFX90A_GFX942-NEXT: [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sreg_32_xm0 = V_READFIRSTLANE_B32 killed [[COPY7]], implicit $exec
; GFX90A_GFX942-NEXT: $sgpr0 = COPY [[V_READFIRSTLANE_B32_]]
; GFX90A_GFX942-NEXT: $sgpr1 = COPY [[V_READFIRSTLANE_B32_1]]
; GFX90A_GFX942-NEXT: SI_RETURN_TO_EPILOG $sgpr0, $sgpr1
%ret = call double @llvm.amdgcn.flat.atomic.fadd.f64.p1.f64(ptr %ptr, double %data)
ret double %ret
}
define amdgpu_ps void @flat_atomic_fadd_f64_no_rtn_atomicrmw(ptr %ptr, double %data) #0 {
; GFX90A_GFX942-LABEL: name: flat_atomic_fadd_f64_no_rtn_atomicrmw
; GFX90A_GFX942: bb.0 (%ir-block.0):
; GFX90A_GFX942-NEXT: liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3
; GFX90A_GFX942-NEXT: {{ $}}
; GFX90A_GFX942-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr3
; GFX90A_GFX942-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2
; GFX90A_GFX942-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr1
; GFX90A_GFX942-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX90A_GFX942-NEXT: [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
; GFX90A_GFX942-NEXT: [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
; GFX90A_GFX942-NEXT: [[REG_SEQUENCE:%[0-9]+]]:vreg_64_align2 = REG_SEQUENCE [[COPY1]], %subreg.sub0, [[COPY]], %subreg.sub1
; GFX90A_GFX942-NEXT: [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
; GFX90A_GFX942-NEXT: [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
; GFX90A_GFX942-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:vreg_64_align2 = REG_SEQUENCE [[COPY3]], %subreg.sub0, [[COPY2]], %subreg.sub1
; GFX90A_GFX942-NEXT: [[COPY4:%[0-9]+]]:vreg_64_align2 = COPY [[REG_SEQUENCE1]]
; GFX90A_GFX942-NEXT: [[COPY5:%[0-9]+]]:vreg_64_align2 = COPY [[REG_SEQUENCE]]
; GFX90A_GFX942-NEXT: FLAT_ATOMIC_ADD_F64 killed [[COPY4]], killed [[COPY5]], 0, 0, implicit $exec, implicit $flat_scr :: (load store syncscope("wavefront") monotonic (s64) on %ir.ptr)
; GFX90A_GFX942-NEXT: S_ENDPGM 0
%ret = atomicrmw fadd ptr %ptr, double %data syncscope("wavefront") monotonic, !noalias.addrspace !1, !amdgpu.no.fine.grained.memory !0
ret void
}
define amdgpu_ps void @flat_atomic_fadd_f64_no_rtn_atomicrmw_noprivate(ptr %ptr, double %data) #0 {
; GFX90A_GFX942-LABEL: name: flat_atomic_fadd_f64_no_rtn_atomicrmw_noprivate
; GFX90A_GFX942: bb.0 (%ir-block.0):
; GFX90A_GFX942-NEXT: liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3
; GFX90A_GFX942-NEXT: {{ $}}
; GFX90A_GFX942-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr3
; GFX90A_GFX942-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2
; GFX90A_GFX942-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr1
; GFX90A_GFX942-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX90A_GFX942-NEXT: [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
; GFX90A_GFX942-NEXT: [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
; GFX90A_GFX942-NEXT: [[REG_SEQUENCE:%[0-9]+]]:vreg_64_align2 = REG_SEQUENCE [[COPY1]], %subreg.sub0, [[COPY]], %subreg.sub1
; GFX90A_GFX942-NEXT: [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
; GFX90A_GFX942-NEXT: [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
; GFX90A_GFX942-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:vreg_64_align2 = REG_SEQUENCE [[COPY3]], %subreg.sub0, [[COPY2]], %subreg.sub1
; GFX90A_GFX942-NEXT: [[COPY4:%[0-9]+]]:vreg_64_align2 = COPY [[REG_SEQUENCE1]]
; GFX90A_GFX942-NEXT: [[COPY5:%[0-9]+]]:vreg_64_align2 = COPY [[REG_SEQUENCE]]
; GFX90A_GFX942-NEXT: FLAT_ATOMIC_ADD_F64 killed [[COPY4]], killed [[COPY5]], 0, 0, implicit $exec, implicit $flat_scr :: (load store syncscope("wavefront") monotonic (s64) on %ir.ptr)
; GFX90A_GFX942-NEXT: S_ENDPGM 0
%ret = atomicrmw fadd ptr %ptr, double %data syncscope("wavefront") monotonic, !noalias.addrspace !1, !amdgpu.no.fine.grained.memory !0
ret void
}
define amdgpu_ps double @flat_atomic_fadd_f64_rtn_atomicrmw(ptr %ptr, double %data) #0 {
; GFX90A_GFX942-LABEL: name: flat_atomic_fadd_f64_rtn_atomicrmw
; GFX90A_GFX942: bb.0 (%ir-block.0):
; GFX90A_GFX942-NEXT: liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3
; GFX90A_GFX942-NEXT: {{ $}}
; GFX90A_GFX942-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr3
; GFX90A_GFX942-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2
; GFX90A_GFX942-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr1
; GFX90A_GFX942-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX90A_GFX942-NEXT: [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
; GFX90A_GFX942-NEXT: [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
; GFX90A_GFX942-NEXT: [[REG_SEQUENCE:%[0-9]+]]:vreg_64_align2 = REG_SEQUENCE [[COPY1]], %subreg.sub0, [[COPY]], %subreg.sub1
; GFX90A_GFX942-NEXT: [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
; GFX90A_GFX942-NEXT: [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
; GFX90A_GFX942-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:vreg_64_align2 = REG_SEQUENCE [[COPY3]], %subreg.sub0, [[COPY2]], %subreg.sub1
; GFX90A_GFX942-NEXT: [[COPY4:%[0-9]+]]:vreg_64_align2 = COPY [[REG_SEQUENCE1]]
; GFX90A_GFX942-NEXT: [[COPY5:%[0-9]+]]:vreg_64_align2 = COPY [[REG_SEQUENCE]]
; GFX90A_GFX942-NEXT: [[FLAT_ATOMIC_ADD_F64_RTN:%[0-9]+]]:vreg_64_align2 = FLAT_ATOMIC_ADD_F64_RTN killed [[COPY4]], killed [[COPY5]], 0, 1, implicit $exec, implicit $flat_scr :: (load store syncscope("wavefront") monotonic (s64) on %ir.ptr)
; GFX90A_GFX942-NEXT: [[COPY6:%[0-9]+]]:vgpr_32 = COPY [[FLAT_ATOMIC_ADD_F64_RTN]].sub0
; GFX90A_GFX942-NEXT: [[V_READFIRSTLANE_B32_:%[0-9]+]]:sreg_32_xm0 = V_READFIRSTLANE_B32 killed [[COPY6]], implicit $exec
; GFX90A_GFX942-NEXT: [[COPY7:%[0-9]+]]:vgpr_32 = COPY [[FLAT_ATOMIC_ADD_F64_RTN]].sub1
; GFX90A_GFX942-NEXT: [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sreg_32_xm0 = V_READFIRSTLANE_B32 killed [[COPY7]], implicit $exec
; GFX90A_GFX942-NEXT: $sgpr0 = COPY [[V_READFIRSTLANE_B32_]]
; GFX90A_GFX942-NEXT: $sgpr1 = COPY [[V_READFIRSTLANE_B32_1]]
; GFX90A_GFX942-NEXT: SI_RETURN_TO_EPILOG $sgpr0, $sgpr1
%ret = atomicrmw fadd ptr %ptr, double %data syncscope("wavefront") monotonic, !noalias.addrspace !1, !amdgpu.no.fine.grained.memory !0
ret double %ret
}
define amdgpu_ps double @flat_atomic_fadd_f64_rtn_atomicrmw__noprivate(ptr %ptr, double %data) #0 {
; GFX90A_GFX942-LABEL: name: flat_atomic_fadd_f64_rtn_atomicrmw__noprivate
; GFX90A_GFX942: bb.0 (%ir-block.0):
; GFX90A_GFX942-NEXT: liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3
; GFX90A_GFX942-NEXT: {{ $}}
; GFX90A_GFX942-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr3
; GFX90A_GFX942-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2
; GFX90A_GFX942-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr1
; GFX90A_GFX942-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX90A_GFX942-NEXT: [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
; GFX90A_GFX942-NEXT: [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
; GFX90A_GFX942-NEXT: [[REG_SEQUENCE:%[0-9]+]]:vreg_64_align2 = REG_SEQUENCE [[COPY1]], %subreg.sub0, [[COPY]], %subreg.sub1
; GFX90A_GFX942-NEXT: [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
; GFX90A_GFX942-NEXT: [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
; GFX90A_GFX942-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:vreg_64_align2 = REG_SEQUENCE [[COPY3]], %subreg.sub0, [[COPY2]], %subreg.sub1
; GFX90A_GFX942-NEXT: [[COPY4:%[0-9]+]]:vreg_64_align2 = COPY [[REG_SEQUENCE1]]
; GFX90A_GFX942-NEXT: [[COPY5:%[0-9]+]]:vreg_64_align2 = COPY [[REG_SEQUENCE]]
; GFX90A_GFX942-NEXT: [[FLAT_ATOMIC_ADD_F64_RTN:%[0-9]+]]:vreg_64_align2 = FLAT_ATOMIC_ADD_F64_RTN killed [[COPY4]], killed [[COPY5]], 0, 1, implicit $exec, implicit $flat_scr :: (load store syncscope("wavefront") monotonic (s64) on %ir.ptr)
; GFX90A_GFX942-NEXT: [[COPY6:%[0-9]+]]:vgpr_32 = COPY [[FLAT_ATOMIC_ADD_F64_RTN]].sub0
; GFX90A_GFX942-NEXT: [[V_READFIRSTLANE_B32_:%[0-9]+]]:sreg_32_xm0 = V_READFIRSTLANE_B32 killed [[COPY6]], implicit $exec
; GFX90A_GFX942-NEXT: [[COPY7:%[0-9]+]]:vgpr_32 = COPY [[FLAT_ATOMIC_ADD_F64_RTN]].sub1
; GFX90A_GFX942-NEXT: [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sreg_32_xm0 = V_READFIRSTLANE_B32 killed [[COPY7]], implicit $exec
; GFX90A_GFX942-NEXT: $sgpr0 = COPY [[V_READFIRSTLANE_B32_]]
; GFX90A_GFX942-NEXT: $sgpr1 = COPY [[V_READFIRSTLANE_B32_1]]
; GFX90A_GFX942-NEXT: SI_RETURN_TO_EPILOG $sgpr0, $sgpr1
%ret = atomicrmw fadd ptr %ptr, double %data syncscope("wavefront") monotonic, !noalias.addrspace !1, !amdgpu.no.fine.grained.memory !0
ret double %ret
}
declare double @llvm.amdgcn.flat.atomic.fadd.f64.p1.f64(ptr, double)
attributes #0 = { nounwind }
!0 = !{}
!1 = !{i32 5, i32 6}
|
