1 ; NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
2 ; RUN: llc -march=amdgcn -mcpu=gfx90a -verify-machineinstrs -stop-after=amdgpu-isel < %s | FileCheck -check-prefix=GFX90A_GFX940 %s
3 ; RUN: llc -march=amdgcn -mcpu=gfx940 -verify-machineinstrs -stop-after=amdgpu-isel < %s | FileCheck -check-prefix=GFX90A_GFX940 %s
5 define amdgpu_ps <2 x half> @buffer_atomic_fadd_v2f16_offset_rtn(<2 x half> %val, <4 x i32> inreg %rsrc, i32 inreg %soffset) {
6 ; GFX90A_GFX940-LABEL: name: buffer_atomic_fadd_v2f16_offset_rtn
7 ; GFX90A_GFX940: bb.0 (%ir-block.0):
8 ; GFX90A_GFX940-NEXT: liveins: $vgpr0, $sgpr0, $sgpr1, $sgpr2, $sgpr3, $sgpr4
9 ; GFX90A_GFX940-NEXT: {{ $}}
10 ; GFX90A_GFX940-NEXT: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr4
11 ; GFX90A_GFX940-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr3
12 ; GFX90A_GFX940-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2
13 ; GFX90A_GFX940-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr1
14 ; GFX90A_GFX940-NEXT: [[COPY4:%[0-9]+]]:sgpr_32 = COPY $sgpr0
15 ; GFX90A_GFX940-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr0
16 ; GFX90A_GFX940-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY4]], %subreg.sub0, [[COPY3]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY1]], %subreg.sub3
17 ; GFX90A_GFX940-NEXT: [[BUFFER_ATOMIC_PK_ADD_F16_OFFSET_RTN:%[0-9]+]]:vgpr_32 = BUFFER_ATOMIC_PK_ADD_F16_OFFSET_RTN [[COPY5]], killed [[REG_SEQUENCE]], [[COPY]], 0, 1, implicit $exec :: (volatile dereferenceable load store (s32), align 1, addrspace 8)
18 ; GFX90A_GFX940-NEXT: $vgpr0 = COPY [[BUFFER_ATOMIC_PK_ADD_F16_OFFSET_RTN]]
19 ; GFX90A_GFX940-NEXT: SI_RETURN_TO_EPILOG $vgpr0
20 %ret = call <2 x half> @llvm.amdgcn.raw.buffer.atomic.fadd.v2f16(<2 x half> %val, <4 x i32> %rsrc, i32 0, i32 %soffset, i32 0)
24 define amdgpu_ps <2 x half> @buffer_atomic_fadd_v2f16_offen_rtn(<2 x half> %val, <4 x i32> inreg %rsrc, i32 %voffset, i32 inreg %soffset) {
25 ; GFX90A_GFX940-LABEL: name: buffer_atomic_fadd_v2f16_offen_rtn
26 ; GFX90A_GFX940: bb.0 (%ir-block.0):
27 ; GFX90A_GFX940-NEXT: liveins: $vgpr0, $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr1, $sgpr4
28 ; GFX90A_GFX940-NEXT: {{ $}}
29 ; GFX90A_GFX940-NEXT: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr4
30 ; GFX90A_GFX940-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
31 ; GFX90A_GFX940-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr3
32 ; GFX90A_GFX940-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr2
33 ; GFX90A_GFX940-NEXT: [[COPY4:%[0-9]+]]:sgpr_32 = COPY $sgpr1
34 ; GFX90A_GFX940-NEXT: [[COPY5:%[0-9]+]]:sgpr_32 = COPY $sgpr0
35 ; GFX90A_GFX940-NEXT: [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr0
36 ; GFX90A_GFX940-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY5]], %subreg.sub0, [[COPY4]], %subreg.sub1, [[COPY3]], %subreg.sub2, [[COPY2]], %subreg.sub3
37 ; GFX90A_GFX940-NEXT: [[BUFFER_ATOMIC_PK_ADD_F16_OFFEN_RTN:%[0-9]+]]:vgpr_32 = BUFFER_ATOMIC_PK_ADD_F16_OFFEN_RTN [[COPY6]], [[COPY1]], killed [[REG_SEQUENCE]], [[COPY]], 0, 1, implicit $exec :: (volatile dereferenceable load store (s32), align 1, addrspace 8)
38 ; GFX90A_GFX940-NEXT: $vgpr0 = COPY [[BUFFER_ATOMIC_PK_ADD_F16_OFFEN_RTN]]
39 ; GFX90A_GFX940-NEXT: SI_RETURN_TO_EPILOG $vgpr0
40 %ret = call <2 x half> @llvm.amdgcn.raw.buffer.atomic.fadd.v2f16(<2 x half> %val, <4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 0)
44 define amdgpu_ps <2 x half> @buffer_atomic_fadd_v2f16_idxen_rtn(<2 x half> %val, <4 x i32> inreg %rsrc, i32 %vindex, i32 inreg %soffset) {
45 ; GFX90A_GFX940-LABEL: name: buffer_atomic_fadd_v2f16_idxen_rtn
46 ; GFX90A_GFX940: bb.0 (%ir-block.0):
47 ; GFX90A_GFX940-NEXT: liveins: $vgpr0, $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr1, $sgpr4
48 ; GFX90A_GFX940-NEXT: {{ $}}
49 ; GFX90A_GFX940-NEXT: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr4
50 ; GFX90A_GFX940-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
51 ; GFX90A_GFX940-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr3
52 ; GFX90A_GFX940-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr2
53 ; GFX90A_GFX940-NEXT: [[COPY4:%[0-9]+]]:sgpr_32 = COPY $sgpr1
54 ; GFX90A_GFX940-NEXT: [[COPY5:%[0-9]+]]:sgpr_32 = COPY $sgpr0
55 ; GFX90A_GFX940-NEXT: [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr0
56 ; GFX90A_GFX940-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY5]], %subreg.sub0, [[COPY4]], %subreg.sub1, [[COPY3]], %subreg.sub2, [[COPY2]], %subreg.sub3
57 ; GFX90A_GFX940-NEXT: [[BUFFER_ATOMIC_PK_ADD_F16_IDXEN_RTN:%[0-9]+]]:vgpr_32 = BUFFER_ATOMIC_PK_ADD_F16_IDXEN_RTN [[COPY6]], [[COPY1]], killed [[REG_SEQUENCE]], [[COPY]], 0, 1, implicit $exec :: (volatile dereferenceable load store (s32), align 1, addrspace 8)
58 ; GFX90A_GFX940-NEXT: $vgpr0 = COPY [[BUFFER_ATOMIC_PK_ADD_F16_IDXEN_RTN]]
59 ; GFX90A_GFX940-NEXT: SI_RETURN_TO_EPILOG $vgpr0
60 %ret = call <2 x half> @llvm.amdgcn.struct.buffer.atomic.fadd.v2f16(<2 x half> %val, <4 x i32> %rsrc, i32 %vindex, i32 0, i32 %soffset, i32 0)
64 define amdgpu_ps <2 x half> @buffer_atomic_fadd_v2f16_bothen_rtn(<2 x half> %val, <4 x i32> inreg %rsrc, i32 %vindex, i32 %voffset, i32 inreg %soffset) {
65 ; GFX90A_GFX940-LABEL: name: buffer_atomic_fadd_v2f16_bothen_rtn
66 ; GFX90A_GFX940: bb.0 (%ir-block.0):
67 ; GFX90A_GFX940-NEXT: liveins: $vgpr0, $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr1, $vgpr2, $sgpr4
68 ; GFX90A_GFX940-NEXT: {{ $}}
69 ; GFX90A_GFX940-NEXT: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr4
70 ; GFX90A_GFX940-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2
71 ; GFX90A_GFX940-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr1
72 ; GFX90A_GFX940-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3
73 ; GFX90A_GFX940-NEXT: [[COPY4:%[0-9]+]]:sgpr_32 = COPY $sgpr2
74 ; GFX90A_GFX940-NEXT: [[COPY5:%[0-9]+]]:sgpr_32 = COPY $sgpr1
75 ; GFX90A_GFX940-NEXT: [[COPY6:%[0-9]+]]:sgpr_32 = COPY $sgpr0
76 ; GFX90A_GFX940-NEXT: [[COPY7:%[0-9]+]]:vgpr_32 = COPY $vgpr0
77 ; GFX90A_GFX940-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY6]], %subreg.sub0, [[COPY5]], %subreg.sub1, [[COPY4]], %subreg.sub2, [[COPY3]], %subreg.sub3
78 ; GFX90A_GFX940-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:vreg_64_align2 = REG_SEQUENCE [[COPY2]], %subreg.sub0, [[COPY1]], %subreg.sub1
79 ; GFX90A_GFX940-NEXT: [[BUFFER_ATOMIC_PK_ADD_F16_BOTHEN_RTN:%[0-9]+]]:vgpr_32 = BUFFER_ATOMIC_PK_ADD_F16_BOTHEN_RTN [[COPY7]], killed [[REG_SEQUENCE1]], killed [[REG_SEQUENCE]], [[COPY]], 0, 3, implicit $exec :: (volatile dereferenceable load store (s32), align 1, addrspace 8)
80 ; GFX90A_GFX940-NEXT: $vgpr0 = COPY [[BUFFER_ATOMIC_PK_ADD_F16_BOTHEN_RTN]]
81 ; GFX90A_GFX940-NEXT: SI_RETURN_TO_EPILOG $vgpr0
82 %ret = call <2 x half> @llvm.amdgcn.struct.buffer.atomic.fadd.v2f16(<2 x half> %val, <4 x i32> %rsrc, i32 %vindex, i32 %voffset, i32 %soffset, i32 2)
86 define amdgpu_ps <2 x half> @buffer_ptr_atomic_fadd_v2f16_offset_rtn(<2 x half> %val, ptr addrspace(8) inreg %rsrc, i32 inreg %soffset) {
87 ; GFX90A_GFX940-LABEL: name: buffer_ptr_atomic_fadd_v2f16_offset_rtn
88 ; GFX90A_GFX940: bb.0 (%ir-block.0):
89 ; GFX90A_GFX940-NEXT: liveins: $vgpr0, $sgpr0, $sgpr1, $sgpr2, $sgpr3, $sgpr4
90 ; GFX90A_GFX940-NEXT: {{ $}}
91 ; GFX90A_GFX940-NEXT: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr4
92 ; GFX90A_GFX940-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr3
93 ; GFX90A_GFX940-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2
94 ; GFX90A_GFX940-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr1
95 ; GFX90A_GFX940-NEXT: [[COPY4:%[0-9]+]]:sgpr_32 = COPY $sgpr0
96 ; GFX90A_GFX940-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr0
97 ; GFX90A_GFX940-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[COPY2]], %subreg.sub0, [[COPY1]], %subreg.sub1
98 ; GFX90A_GFX940-NEXT: [[COPY6:%[0-9]+]]:sreg_32 = COPY [[REG_SEQUENCE]].sub1
99 ; GFX90A_GFX940-NEXT: [[COPY7:%[0-9]+]]:sreg_32 = COPY [[REG_SEQUENCE]].sub0
100 ; GFX90A_GFX940-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[COPY4]], %subreg.sub0, [[COPY3]], %subreg.sub1
101 ; GFX90A_GFX940-NEXT: [[COPY8:%[0-9]+]]:sreg_32 = COPY [[REG_SEQUENCE1]].sub1
102 ; GFX90A_GFX940-NEXT: [[COPY9:%[0-9]+]]:sreg_32 = COPY [[REG_SEQUENCE1]].sub0
103 ; GFX90A_GFX940-NEXT: [[REG_SEQUENCE2:%[0-9]+]]:sgpr_128 = REG_SEQUENCE killed [[COPY9]], %subreg.sub0, killed [[COPY8]], %subreg.sub1, killed [[COPY7]], %subreg.sub2, killed [[COPY6]], %subreg.sub3
104 ; GFX90A_GFX940-NEXT: [[BUFFER_ATOMIC_PK_ADD_F16_OFFSET_RTN:%[0-9]+]]:vgpr_32 = BUFFER_ATOMIC_PK_ADD_F16_OFFSET_RTN [[COPY5]], killed [[REG_SEQUENCE2]], [[COPY]], 0, 1, implicit $exec :: (volatile dereferenceable load store (s32) on %ir.rsrc, align 1, addrspace 8)
105 ; GFX90A_GFX940-NEXT: $vgpr0 = COPY [[BUFFER_ATOMIC_PK_ADD_F16_OFFSET_RTN]]
106 ; GFX90A_GFX940-NEXT: SI_RETURN_TO_EPILOG $vgpr0
107 %ret = call <2 x half> @llvm.amdgcn.raw.ptr.buffer.atomic.fadd.v2f16(<2 x half> %val, ptr addrspace(8) %rsrc, i32 0, i32 %soffset, i32 0)
111 define amdgpu_ps <2 x half> @buffer_ptr_atomic_fadd_v2f16_offen_rtn(<2 x half> %val, ptr addrspace(8) inreg %rsrc, i32 %voffset, i32 inreg %soffset) {
112 ; GFX90A_GFX940-LABEL: name: buffer_ptr_atomic_fadd_v2f16_offen_rtn
113 ; GFX90A_GFX940: bb.0 (%ir-block.0):
114 ; GFX90A_GFX940-NEXT: liveins: $vgpr0, $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr1, $sgpr4
115 ; GFX90A_GFX940-NEXT: {{ $}}
116 ; GFX90A_GFX940-NEXT: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr4
117 ; GFX90A_GFX940-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
118 ; GFX90A_GFX940-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr3
119 ; GFX90A_GFX940-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr2
120 ; GFX90A_GFX940-NEXT: [[COPY4:%[0-9]+]]:sgpr_32 = COPY $sgpr1
121 ; GFX90A_GFX940-NEXT: [[COPY5:%[0-9]+]]:sgpr_32 = COPY $sgpr0
122 ; GFX90A_GFX940-NEXT: [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr0
123 ; GFX90A_GFX940-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[COPY3]], %subreg.sub0, [[COPY2]], %subreg.sub1
124 ; GFX90A_GFX940-NEXT: [[COPY7:%[0-9]+]]:sreg_32 = COPY [[REG_SEQUENCE]].sub1
125 ; GFX90A_GFX940-NEXT: [[COPY8:%[0-9]+]]:sreg_32 = COPY [[REG_SEQUENCE]].sub0
126 ; GFX90A_GFX940-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[COPY5]], %subreg.sub0, [[COPY4]], %subreg.sub1
127 ; GFX90A_GFX940-NEXT: [[COPY9:%[0-9]+]]:sreg_32 = COPY [[REG_SEQUENCE1]].sub1
128 ; GFX90A_GFX940-NEXT: [[COPY10:%[0-9]+]]:sreg_32 = COPY [[REG_SEQUENCE1]].sub0
129 ; GFX90A_GFX940-NEXT: [[REG_SEQUENCE2:%[0-9]+]]:sgpr_128 = REG_SEQUENCE killed [[COPY10]], %subreg.sub0, killed [[COPY9]], %subreg.sub1, killed [[COPY8]], %subreg.sub2, killed [[COPY7]], %subreg.sub3
130 ; GFX90A_GFX940-NEXT: [[BUFFER_ATOMIC_PK_ADD_F16_OFFEN_RTN:%[0-9]+]]:vgpr_32 = BUFFER_ATOMIC_PK_ADD_F16_OFFEN_RTN [[COPY6]], [[COPY1]], killed [[REG_SEQUENCE2]], [[COPY]], 0, 1, implicit $exec :: (volatile dereferenceable load store (s32) on %ir.rsrc, align 1, addrspace 8)
131 ; GFX90A_GFX940-NEXT: $vgpr0 = COPY [[BUFFER_ATOMIC_PK_ADD_F16_OFFEN_RTN]]
132 ; GFX90A_GFX940-NEXT: SI_RETURN_TO_EPILOG $vgpr0
133 %ret = call <2 x half> @llvm.amdgcn.raw.ptr.buffer.atomic.fadd.v2f16(<2 x half> %val, ptr addrspace(8) %rsrc, i32 %voffset, i32 %soffset, i32 0)
137 define amdgpu_ps <2 x half> @buffer_ptr_atomic_fadd_v2f16_idxen_rtn(<2 x half> %val, ptr addrspace(8) inreg %rsrc, i32 %vindex, i32 inreg %soffset) {
138 ; GFX90A_GFX940-LABEL: name: buffer_ptr_atomic_fadd_v2f16_idxen_rtn
139 ; GFX90A_GFX940: bb.0 (%ir-block.0):
140 ; GFX90A_GFX940-NEXT: liveins: $vgpr0, $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr1, $sgpr4
141 ; GFX90A_GFX940-NEXT: {{ $}}
142 ; GFX90A_GFX940-NEXT: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr4
143 ; GFX90A_GFX940-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
144 ; GFX90A_GFX940-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr3
145 ; GFX90A_GFX940-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr2
146 ; GFX90A_GFX940-NEXT: [[COPY4:%[0-9]+]]:sgpr_32 = COPY $sgpr1
147 ; GFX90A_GFX940-NEXT: [[COPY5:%[0-9]+]]:sgpr_32 = COPY $sgpr0
148 ; GFX90A_GFX940-NEXT: [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr0
149 ; GFX90A_GFX940-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[COPY3]], %subreg.sub0, [[COPY2]], %subreg.sub1
150 ; GFX90A_GFX940-NEXT: [[COPY7:%[0-9]+]]:sreg_32 = COPY [[REG_SEQUENCE]].sub1
151 ; GFX90A_GFX940-NEXT: [[COPY8:%[0-9]+]]:sreg_32 = COPY [[REG_SEQUENCE]].sub0
152 ; GFX90A_GFX940-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[COPY5]], %subreg.sub0, [[COPY4]], %subreg.sub1
153 ; GFX90A_GFX940-NEXT: [[COPY9:%[0-9]+]]:sreg_32 = COPY [[REG_SEQUENCE1]].sub1
154 ; GFX90A_GFX940-NEXT: [[COPY10:%[0-9]+]]:sreg_32 = COPY [[REG_SEQUENCE1]].sub0
155 ; GFX90A_GFX940-NEXT: [[REG_SEQUENCE2:%[0-9]+]]:sgpr_128 = REG_SEQUENCE killed [[COPY10]], %subreg.sub0, killed [[COPY9]], %subreg.sub1, killed [[COPY8]], %subreg.sub2, killed [[COPY7]], %subreg.sub3
156 ; GFX90A_GFX940-NEXT: [[BUFFER_ATOMIC_PK_ADD_F16_IDXEN_RTN:%[0-9]+]]:vgpr_32 = BUFFER_ATOMIC_PK_ADD_F16_IDXEN_RTN [[COPY6]], [[COPY1]], killed [[REG_SEQUENCE2]], [[COPY]], 0, 1, implicit $exec :: (volatile dereferenceable load store (s32) on %ir.rsrc, align 1, addrspace 8)
157 ; GFX90A_GFX940-NEXT: $vgpr0 = COPY [[BUFFER_ATOMIC_PK_ADD_F16_IDXEN_RTN]]
158 ; GFX90A_GFX940-NEXT: SI_RETURN_TO_EPILOG $vgpr0
159 %ret = call <2 x half> @llvm.amdgcn.struct.ptr.buffer.atomic.fadd.v2f16(<2 x half> %val, ptr addrspace(8) %rsrc, i32 %vindex, i32 0, i32 %soffset, i32 0)
163 define amdgpu_ps <2 x half> @buffer_ptr_atomic_fadd_v2f16_bothen_rtn(<2 x half> %val, ptr addrspace(8) inreg %rsrc, i32 %vindex, i32 %voffset, i32 inreg %soffset) {
164 ; GFX90A_GFX940-LABEL: name: buffer_ptr_atomic_fadd_v2f16_bothen_rtn
165 ; GFX90A_GFX940: bb.0 (%ir-block.0):
166 ; GFX90A_GFX940-NEXT: liveins: $vgpr0, $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr1, $vgpr2, $sgpr4
167 ; GFX90A_GFX940-NEXT: {{ $}}
168 ; GFX90A_GFX940-NEXT: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr4
169 ; GFX90A_GFX940-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2
170 ; GFX90A_GFX940-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr1
171 ; GFX90A_GFX940-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3
172 ; GFX90A_GFX940-NEXT: [[COPY4:%[0-9]+]]:sgpr_32 = COPY $sgpr2
173 ; GFX90A_GFX940-NEXT: [[COPY5:%[0-9]+]]:sgpr_32 = COPY $sgpr1
174 ; GFX90A_GFX940-NEXT: [[COPY6:%[0-9]+]]:sgpr_32 = COPY $sgpr0
175 ; GFX90A_GFX940-NEXT: [[COPY7:%[0-9]+]]:vgpr_32 = COPY $vgpr0
176 ; GFX90A_GFX940-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[COPY4]], %subreg.sub0, [[COPY3]], %subreg.sub1
177 ; GFX90A_GFX940-NEXT: [[COPY8:%[0-9]+]]:sreg_32 = COPY [[REG_SEQUENCE]].sub1
178 ; GFX90A_GFX940-NEXT: [[COPY9:%[0-9]+]]:sreg_32 = COPY [[REG_SEQUENCE]].sub0
179 ; GFX90A_GFX940-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[COPY6]], %subreg.sub0, [[COPY5]], %subreg.sub1
180 ; GFX90A_GFX940-NEXT: [[COPY10:%[0-9]+]]:sreg_32 = COPY [[REG_SEQUENCE1]].sub1
181 ; GFX90A_GFX940-NEXT: [[COPY11:%[0-9]+]]:sreg_32 = COPY [[REG_SEQUENCE1]].sub0
182 ; GFX90A_GFX940-NEXT: [[REG_SEQUENCE2:%[0-9]+]]:sgpr_128 = REG_SEQUENCE killed [[COPY11]], %subreg.sub0, killed [[COPY10]], %subreg.sub1, killed [[COPY9]], %subreg.sub2, killed [[COPY8]], %subreg.sub3
183 ; GFX90A_GFX940-NEXT: [[REG_SEQUENCE3:%[0-9]+]]:vreg_64_align2 = REG_SEQUENCE [[COPY2]], %subreg.sub0, [[COPY1]], %subreg.sub1
184 ; GFX90A_GFX940-NEXT: [[BUFFER_ATOMIC_PK_ADD_F16_BOTHEN_RTN:%[0-9]+]]:vgpr_32 = BUFFER_ATOMIC_PK_ADD_F16_BOTHEN_RTN [[COPY7]], killed [[REG_SEQUENCE3]], killed [[REG_SEQUENCE2]], [[COPY]], 0, 3, implicit $exec :: (volatile dereferenceable load store (s32) on %ir.rsrc, align 1, addrspace 8)
185 ; GFX90A_GFX940-NEXT: $vgpr0 = COPY [[BUFFER_ATOMIC_PK_ADD_F16_BOTHEN_RTN]]
186 ; GFX90A_GFX940-NEXT: SI_RETURN_TO_EPILOG $vgpr0
187 %ret = call <2 x half> @llvm.amdgcn.struct.ptr.buffer.atomic.fadd.v2f16(<2 x half> %val, ptr addrspace(8) %rsrc, i32 %vindex, i32 %voffset, i32 %soffset, i32 2)
191 declare <2 x half> @llvm.amdgcn.raw.buffer.atomic.fadd.v2f16(<2 x half>, <4 x i32>, i32, i32, i32 immarg)
192 declare <2 x half> @llvm.amdgcn.struct.buffer.atomic.fadd.v2f16(<2 x half>, <4 x i32>, i32, i32, i32, i32 immarg)
194 declare <2 x half> @llvm.amdgcn.raw.ptr.buffer.atomic.fadd.v2f16(<2 x half>, ptr addrspace(8), i32, i32, i32 immarg)
195 declare <2 x half> @llvm.amdgcn.struct.ptr.buffer.atomic.fadd.v2f16(<2 x half>, ptr addrspace(8), i32, i32, i32, i32 immarg)