1 ; NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
2 ; RUN: llc -global-isel -march=amdgcn -mcpu=gfx908 -verify-machineinstrs -amdgpu-atomic-optimizer-strategy=DPP -stop-after=instruction-select < %s | FileCheck -check-prefix=GFX908_GFX11 %s
3 ; RUN: llc -global-isel -march=amdgcn -mcpu=gfx90a -verify-machineinstrs -amdgpu-atomic-optimizer-strategy=DPP -stop-after=instruction-select < %s | FileCheck -check-prefix=GFX90A_GFX940 %s
4 ; RUN: llc -global-isel -march=amdgcn -mcpu=gfx940 -verify-machineinstrs -amdgpu-atomic-optimizer-strategy=DPP -stop-after=instruction-select < %s | FileCheck -check-prefix=GFX90A_GFX940 %s
5 ; RUN: llc -global-isel -march=amdgcn -mcpu=gfx1100 -verify-machineinstrs -amdgpu-atomic-optimizer-strategy=DPP -stop-after=instruction-select < %s | FileCheck -check-prefix=GFX908_GFX11 %s
7 define amdgpu_ps void @global_atomic_fadd_f32_no_rtn_intrinsic(ptr addrspace(1) %ptr, float %data) {
8 ; GFX908_GFX11-LABEL: name: global_atomic_fadd_f32_no_rtn_intrinsic
9 ; GFX908_GFX11: bb.1 (%ir-block.0):
10 ; GFX908_GFX11-NEXT: liveins: $vgpr0, $vgpr1, $vgpr2
11 ; GFX908_GFX11-NEXT: {{ $}}
12 ; GFX908_GFX11-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
13 ; GFX908_GFX11-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
14 ; GFX908_GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1
15 ; GFX908_GFX11-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr2
16 ; GFX908_GFX11-NEXT: GLOBAL_ATOMIC_ADD_F32 [[REG_SEQUENCE]], [[COPY2]], 0, 0, implicit $exec :: (volatile dereferenceable load store (s32) on %ir.ptr, addrspace 1)
17 ; GFX908_GFX11-NEXT: S_ENDPGM 0
19 ; GFX90A_GFX940-LABEL: name: global_atomic_fadd_f32_no_rtn_intrinsic
20 ; GFX90A_GFX940: bb.1 (%ir-block.0):
21 ; GFX90A_GFX940-NEXT: liveins: $vgpr0, $vgpr1, $vgpr2
22 ; GFX90A_GFX940-NEXT: {{ $}}
23 ; GFX90A_GFX940-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
24 ; GFX90A_GFX940-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
25 ; GFX90A_GFX940-NEXT: [[REG_SEQUENCE:%[0-9]+]]:vreg_64_align2 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1
26 ; GFX90A_GFX940-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr2
27 ; GFX90A_GFX940-NEXT: GLOBAL_ATOMIC_ADD_F32 [[REG_SEQUENCE]], [[COPY2]], 0, 0, implicit $exec :: (volatile dereferenceable load store (s32) on %ir.ptr, addrspace 1)
28 ; GFX90A_GFX940-NEXT: S_ENDPGM 0
29 %ret = call float @llvm.amdgcn.global.atomic.fadd.f32.p1.f32(ptr addrspace(1) %ptr, float %data)
33 define amdgpu_ps void @global_atomic_fadd_f32_saddr_no_rtn_intrinsic(ptr addrspace(1) inreg %ptr, float %data) {
34 ; GFX908_GFX11-LABEL: name: global_atomic_fadd_f32_saddr_no_rtn_intrinsic
35 ; GFX908_GFX11: bb.1 (%ir-block.0):
36 ; GFX908_GFX11-NEXT: liveins: $sgpr0, $sgpr1, $vgpr0
37 ; GFX908_GFX11-NEXT: {{ $}}
38 ; GFX908_GFX11-NEXT: [[COPY:%[0-9]+]]:sreg_32 = COPY $sgpr0
39 ; GFX908_GFX11-NEXT: [[COPY1:%[0-9]+]]:sreg_32 = COPY $sgpr1
40 ; GFX908_GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sreg_64 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1
41 ; GFX908_GFX11-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr0
42 ; GFX908_GFX11-NEXT: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 0, implicit $exec
43 ; GFX908_GFX11-NEXT: GLOBAL_ATOMIC_ADD_F32_SADDR [[V_MOV_B32_e32_]], [[COPY2]], [[REG_SEQUENCE]], 0, 0, implicit $exec :: (volatile dereferenceable load store (s32) on %ir.ptr, addrspace 1)
44 ; GFX908_GFX11-NEXT: S_ENDPGM 0
46 ; GFX90A_GFX940-LABEL: name: global_atomic_fadd_f32_saddr_no_rtn_intrinsic
47 ; GFX90A_GFX940: bb.1 (%ir-block.0):
48 ; GFX90A_GFX940-NEXT: liveins: $sgpr0, $sgpr1, $vgpr0
49 ; GFX90A_GFX940-NEXT: {{ $}}
50 ; GFX90A_GFX940-NEXT: [[COPY:%[0-9]+]]:sreg_32 = COPY $sgpr0
51 ; GFX90A_GFX940-NEXT: [[COPY1:%[0-9]+]]:sreg_32 = COPY $sgpr1
52 ; GFX90A_GFX940-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sreg_64 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1
53 ; GFX90A_GFX940-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr0
54 ; GFX90A_GFX940-NEXT: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 0, implicit $exec
55 ; GFX90A_GFX940-NEXT: GLOBAL_ATOMIC_ADD_F32_SADDR [[V_MOV_B32_e32_]], [[COPY2]], [[REG_SEQUENCE]], 0, 0, implicit $exec :: (volatile dereferenceable load store (s32) on %ir.ptr, addrspace 1)
56 ; GFX90A_GFX940-NEXT: S_ENDPGM 0
57 %ret = call float @llvm.amdgcn.global.atomic.fadd.f32.p1.f32(ptr addrspace(1) inreg %ptr, float %data)
61 define amdgpu_ps void @global_atomic_fadd_f32_no_rtn_flat_intrinsic(ptr addrspace(1) %ptr, float %data) {
62 ; GFX908_GFX11-LABEL: name: global_atomic_fadd_f32_no_rtn_flat_intrinsic
63 ; GFX908_GFX11: bb.1 (%ir-block.0):
64 ; GFX908_GFX11-NEXT: liveins: $vgpr0, $vgpr1, $vgpr2
65 ; GFX908_GFX11-NEXT: {{ $}}
66 ; GFX908_GFX11-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
67 ; GFX908_GFX11-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
68 ; GFX908_GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1
69 ; GFX908_GFX11-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr2
70 ; GFX908_GFX11-NEXT: GLOBAL_ATOMIC_ADD_F32 [[REG_SEQUENCE]], [[COPY2]], 0, 0, implicit $exec :: (volatile dereferenceable load store (s32) on %ir.ptr, addrspace 1)
71 ; GFX908_GFX11-NEXT: S_ENDPGM 0
73 ; GFX90A_GFX940-LABEL: name: global_atomic_fadd_f32_no_rtn_flat_intrinsic
74 ; GFX90A_GFX940: bb.1 (%ir-block.0):
75 ; GFX90A_GFX940-NEXT: liveins: $vgpr0, $vgpr1, $vgpr2
76 ; GFX90A_GFX940-NEXT: {{ $}}
77 ; GFX90A_GFX940-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
78 ; GFX90A_GFX940-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
79 ; GFX90A_GFX940-NEXT: [[REG_SEQUENCE:%[0-9]+]]:vreg_64_align2 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1
80 ; GFX90A_GFX940-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr2
81 ; GFX90A_GFX940-NEXT: GLOBAL_ATOMIC_ADD_F32 [[REG_SEQUENCE]], [[COPY2]], 0, 0, implicit $exec :: (volatile dereferenceable load store (s32) on %ir.ptr, addrspace 1)
82 ; GFX90A_GFX940-NEXT: S_ENDPGM 0
83 %ret = call float @llvm.amdgcn.flat.atomic.fadd.f32.p1.f32(ptr addrspace(1) %ptr, float %data)
87 define amdgpu_ps void @global_atomic_fadd_f32_saddr_no_rtn_flat_intrinsic(ptr addrspace(1) inreg %ptr, float %data) {
88 ; GFX908_GFX11-LABEL: name: global_atomic_fadd_f32_saddr_no_rtn_flat_intrinsic
89 ; GFX908_GFX11: bb.1 (%ir-block.0):
90 ; GFX908_GFX11-NEXT: liveins: $sgpr0, $sgpr1, $vgpr0
91 ; GFX908_GFX11-NEXT: {{ $}}
92 ; GFX908_GFX11-NEXT: [[COPY:%[0-9]+]]:sreg_32 = COPY $sgpr0
93 ; GFX908_GFX11-NEXT: [[COPY1:%[0-9]+]]:sreg_32 = COPY $sgpr1
94 ; GFX908_GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sreg_64 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1
95 ; GFX908_GFX11-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr0
96 ; GFX908_GFX11-NEXT: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 0, implicit $exec
97 ; GFX908_GFX11-NEXT: GLOBAL_ATOMIC_ADD_F32_SADDR [[V_MOV_B32_e32_]], [[COPY2]], [[REG_SEQUENCE]], 0, 0, implicit $exec :: (volatile dereferenceable load store (s32) on %ir.ptr, addrspace 1)
98 ; GFX908_GFX11-NEXT: S_ENDPGM 0
100 ; GFX90A_GFX940-LABEL: name: global_atomic_fadd_f32_saddr_no_rtn_flat_intrinsic
101 ; GFX90A_GFX940: bb.1 (%ir-block.0):
102 ; GFX90A_GFX940-NEXT: liveins: $sgpr0, $sgpr1, $vgpr0
103 ; GFX90A_GFX940-NEXT: {{ $}}
104 ; GFX90A_GFX940-NEXT: [[COPY:%[0-9]+]]:sreg_32 = COPY $sgpr0
105 ; GFX90A_GFX940-NEXT: [[COPY1:%[0-9]+]]:sreg_32 = COPY $sgpr1
106 ; GFX90A_GFX940-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sreg_64 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1
107 ; GFX90A_GFX940-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr0
108 ; GFX90A_GFX940-NEXT: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 0, implicit $exec
109 ; GFX90A_GFX940-NEXT: GLOBAL_ATOMIC_ADD_F32_SADDR [[V_MOV_B32_e32_]], [[COPY2]], [[REG_SEQUENCE]], 0, 0, implicit $exec :: (volatile dereferenceable load store (s32) on %ir.ptr, addrspace 1)
110 ; GFX90A_GFX940-NEXT: S_ENDPGM 0
111 %ret = call float @llvm.amdgcn.flat.atomic.fadd.f32.p1.f32(ptr addrspace(1) inreg %ptr, float %data)
115 define amdgpu_ps void @global_atomic_fadd_f32_no_rtn_atomicrmw(ptr addrspace(1) %ptr, float %data) #0 {
116 ; GFX908_GFX11-LABEL: name: global_atomic_fadd_f32_no_rtn_atomicrmw
117 ; GFX908_GFX11: bb.1 (%ir-block.0):
118 ; GFX908_GFX11-NEXT: liveins: $vgpr0, $vgpr1, $vgpr2
119 ; GFX908_GFX11-NEXT: {{ $}}
120 ; GFX908_GFX11-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
121 ; GFX908_GFX11-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
122 ; GFX908_GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1
123 ; GFX908_GFX11-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr2
124 ; GFX908_GFX11-NEXT: GLOBAL_ATOMIC_ADD_F32 [[REG_SEQUENCE]], [[COPY2]], 0, 0, implicit $exec :: (load store syncscope("wavefront") monotonic (s32) on %ir.ptr, addrspace 1)
125 ; GFX908_GFX11-NEXT: S_ENDPGM 0
127 ; GFX90A_GFX940-LABEL: name: global_atomic_fadd_f32_no_rtn_atomicrmw
128 ; GFX90A_GFX940: bb.1 (%ir-block.0):
129 ; GFX90A_GFX940-NEXT: liveins: $vgpr0, $vgpr1, $vgpr2
130 ; GFX90A_GFX940-NEXT: {{ $}}
131 ; GFX90A_GFX940-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
132 ; GFX90A_GFX940-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
133 ; GFX90A_GFX940-NEXT: [[REG_SEQUENCE:%[0-9]+]]:vreg_64_align2 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1
134 ; GFX90A_GFX940-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr2
135 ; GFX90A_GFX940-NEXT: GLOBAL_ATOMIC_ADD_F32 [[REG_SEQUENCE]], [[COPY2]], 0, 0, implicit $exec :: (load store syncscope("wavefront") monotonic (s32) on %ir.ptr, addrspace 1)
136 ; GFX90A_GFX940-NEXT: S_ENDPGM 0
137 %ret = atomicrmw fadd ptr addrspace(1) %ptr, float %data syncscope("wavefront") monotonic
141 define amdgpu_ps void @global_atomic_fadd_f32_saddr_no_rtn_atomicrmw(ptr addrspace(1) inreg %ptr, float %data) #0 {
142 ; GFX90A_GFX940-LABEL: name: global_atomic_fadd_f32_saddr_no_rtn_atomicrmw
143 ; GFX90A_GFX940: bb.1 (%ir-block.0):
144 ; GFX90A_GFX940-NEXT: successors: %bb.2(0x40000000), %bb.5(0x40000000)
145 ; GFX90A_GFX940-NEXT: liveins: $sgpr0, $sgpr1, $vgpr0
146 ; GFX90A_GFX940-NEXT: {{ $}}
147 ; GFX90A_GFX940-NEXT: [[COPY:%[0-9]+]]:sreg_32 = COPY $sgpr0
148 ; GFX90A_GFX940-NEXT: [[COPY1:%[0-9]+]]:sreg_32 = COPY $sgpr1
149 ; GFX90A_GFX940-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sreg_64 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1
150 ; GFX90A_GFX940-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr0
151 ; GFX90A_GFX940-NEXT: [[SI_PS_LIVE:%[0-9]+]]:sreg_64_xexec = SI_PS_LIVE
152 ; GFX90A_GFX940-NEXT: [[SI_IF:%[0-9]+]]:sreg_64_xexec = SI_IF [[SI_PS_LIVE]], %bb.5, implicit-def $exec, implicit-def $scc, implicit $exec
153 ; GFX90A_GFX940-NEXT: S_BRANCH %bb.2
154 ; GFX90A_GFX940-NEXT: {{ $}}
155 ; GFX90A_GFX940-NEXT: bb.2 (%ir-block.5):
156 ; GFX90A_GFX940-NEXT: successors: %bb.3(0x40000000), %bb.4(0x40000000)
157 ; GFX90A_GFX940-NEXT: {{ $}}
158 ; GFX90A_GFX940-NEXT: [[COPY3:%[0-9]+]]:sreg_64 = COPY $exec
159 ; GFX90A_GFX940-NEXT: [[COPY4:%[0-9]+]]:sreg_32 = COPY [[COPY3]].sub0
160 ; GFX90A_GFX940-NEXT: [[COPY5:%[0-9]+]]:sreg_32 = COPY [[COPY3]].sub0
161 ; GFX90A_GFX940-NEXT: [[COPY6:%[0-9]+]]:sreg_32 = COPY [[COPY3]].sub1
162 ; GFX90A_GFX940-NEXT: [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 0
163 ; GFX90A_GFX940-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:sreg_64 = REG_SEQUENCE [[COPY6]], %subreg.sub0, [[S_MOV_B32_]], %subreg.sub1
164 ; GFX90A_GFX940-NEXT: [[COPY7:%[0-9]+]]:sreg_32 = COPY [[REG_SEQUENCE1]].sub0
165 ; GFX90A_GFX940-NEXT: [[S_MOV_B32_1:%[0-9]+]]:sreg_32 = S_MOV_B32 0
166 ; GFX90A_GFX940-NEXT: [[COPY8:%[0-9]+]]:vgpr_32 = COPY [[COPY4]]
167 ; GFX90A_GFX940-NEXT: [[COPY9:%[0-9]+]]:vgpr_32 = COPY [[S_MOV_B32_1]]
168 ; GFX90A_GFX940-NEXT: [[V_MBCNT_LO_U32_B32_e64_:%[0-9]+]]:vgpr_32 = V_MBCNT_LO_U32_B32_e64 [[COPY8]], [[COPY9]], implicit $exec
169 ; GFX90A_GFX940-NEXT: [[COPY10:%[0-9]+]]:vgpr_32 = COPY [[COPY7]]
170 ; GFX90A_GFX940-NEXT: [[V_MBCNT_HI_U32_B32_e64_:%[0-9]+]]:vgpr_32 = V_MBCNT_HI_U32_B32_e64 [[COPY10]], [[V_MBCNT_LO_U32_B32_e64_]], implicit $exec
171 ; GFX90A_GFX940-NEXT: [[S_MOV_B32_2:%[0-9]+]]:sreg_32 = S_MOV_B32 -2147483648
172 ; GFX90A_GFX940-NEXT: [[COPY11:%[0-9]+]]:vgpr_32 = COPY [[S_MOV_B32_2]]
173 ; GFX90A_GFX940-NEXT: [[V_SET_INACTIVE_B32_:%[0-9]+]]:vgpr_32 = V_SET_INACTIVE_B32 [[COPY2]], [[COPY11]], implicit-def dead $scc, implicit $exec
174 ; GFX90A_GFX940-NEXT: [[S_MOV_B32_3:%[0-9]+]]:sreg_32 = S_MOV_B32 2147483648
175 ; GFX90A_GFX940-NEXT: [[COPY12:%[0-9]+]]:vgpr_32 = COPY [[S_MOV_B32_3]]
176 ; GFX90A_GFX940-NEXT: [[V_MOV_B32_dpp:%[0-9]+]]:vgpr_32 = V_MOV_B32_dpp [[COPY12]], [[V_SET_INACTIVE_B32_]], 273, 15, 15, 0, implicit $exec
177 ; GFX90A_GFX940-NEXT: [[V_ADD_F32_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_ADD_F32_e64 0, [[V_SET_INACTIVE_B32_]], 0, [[V_MOV_B32_dpp]], 0, 0, implicit $mode, implicit $exec
178 ; GFX90A_GFX940-NEXT: [[COPY13:%[0-9]+]]:vgpr_32 = COPY [[S_MOV_B32_3]]
179 ; GFX90A_GFX940-NEXT: [[V_MOV_B32_dpp1:%[0-9]+]]:vgpr_32 = V_MOV_B32_dpp [[COPY13]], [[V_ADD_F32_e64_]], 274, 15, 15, 0, implicit $exec
180 ; GFX90A_GFX940-NEXT: [[V_ADD_F32_e64_1:%[0-9]+]]:vgpr_32 = nofpexcept V_ADD_F32_e64 0, [[V_ADD_F32_e64_]], 0, [[V_MOV_B32_dpp1]], 0, 0, implicit $mode, implicit $exec
181 ; GFX90A_GFX940-NEXT: [[COPY14:%[0-9]+]]:vgpr_32 = COPY [[S_MOV_B32_3]]
182 ; GFX90A_GFX940-NEXT: [[V_MOV_B32_dpp2:%[0-9]+]]:vgpr_32 = V_MOV_B32_dpp [[COPY14]], [[V_ADD_F32_e64_1]], 276, 15, 15, 0, implicit $exec
183 ; GFX90A_GFX940-NEXT: [[V_ADD_F32_e64_2:%[0-9]+]]:vgpr_32 = nofpexcept V_ADD_F32_e64 0, [[V_ADD_F32_e64_1]], 0, [[V_MOV_B32_dpp2]], 0, 0, implicit $mode, implicit $exec
184 ; GFX90A_GFX940-NEXT: [[COPY15:%[0-9]+]]:vgpr_32 = COPY [[S_MOV_B32_3]]
185 ; GFX90A_GFX940-NEXT: [[V_MOV_B32_dpp3:%[0-9]+]]:vgpr_32 = V_MOV_B32_dpp [[COPY15]], [[V_ADD_F32_e64_2]], 280, 15, 15, 0, implicit $exec
186 ; GFX90A_GFX940-NEXT: [[V_ADD_F32_e64_3:%[0-9]+]]:vgpr_32 = nofpexcept V_ADD_F32_e64 0, [[V_ADD_F32_e64_2]], 0, [[V_MOV_B32_dpp3]], 0, 0, implicit $mode, implicit $exec
187 ; GFX90A_GFX940-NEXT: [[COPY16:%[0-9]+]]:vgpr_32 = COPY [[S_MOV_B32_3]]
188 ; GFX90A_GFX940-NEXT: [[V_MOV_B32_dpp4:%[0-9]+]]:vgpr_32 = V_MOV_B32_dpp [[COPY16]], [[V_ADD_F32_e64_3]], 322, 10, 15, 0, implicit $exec
189 ; GFX90A_GFX940-NEXT: [[V_ADD_F32_e64_4:%[0-9]+]]:vgpr_32 = nofpexcept V_ADD_F32_e64 0, [[V_ADD_F32_e64_3]], 0, [[V_MOV_B32_dpp4]], 0, 0, implicit $mode, implicit $exec
190 ; GFX90A_GFX940-NEXT: [[COPY17:%[0-9]+]]:vgpr_32 = COPY [[S_MOV_B32_3]]
191 ; GFX90A_GFX940-NEXT: [[V_MOV_B32_dpp5:%[0-9]+]]:vgpr_32 = V_MOV_B32_dpp [[COPY17]], [[V_ADD_F32_e64_4]], 323, 12, 15, 0, implicit $exec
192 ; GFX90A_GFX940-NEXT: [[V_ADD_F32_e64_5:%[0-9]+]]:vgpr_32 = nofpexcept V_ADD_F32_e64 0, [[V_ADD_F32_e64_4]], 0, [[V_MOV_B32_dpp5]], 0, 0, implicit $mode, implicit $exec
193 ; GFX90A_GFX940-NEXT: [[S_MOV_B32_4:%[0-9]+]]:sreg_32 = S_MOV_B32 63
194 ; GFX90A_GFX940-NEXT: [[V_READLANE_B32_:%[0-9]+]]:sreg_32 = V_READLANE_B32 [[V_ADD_F32_e64_5]], [[S_MOV_B32_4]]
195 ; GFX90A_GFX940-NEXT: [[COPY18:%[0-9]+]]:vgpr_32 = COPY [[V_READLANE_B32_]]
196 ; GFX90A_GFX940-NEXT: [[STRICT_WWM:%[0-9]+]]:vgpr_32 = STRICT_WWM [[COPY18]], implicit $exec
197 ; GFX90A_GFX940-NEXT: [[COPY19:%[0-9]+]]:vgpr_32 = COPY [[S_MOV_B32_1]]
198 ; GFX90A_GFX940-NEXT: [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_MBCNT_HI_U32_B32_e64_]], [[COPY19]], implicit $exec
199 ; GFX90A_GFX940-NEXT: [[SI_IF1:%[0-9]+]]:sreg_64_xexec = SI_IF [[V_CMP_EQ_U32_e64_]], %bb.4, implicit-def $exec, implicit-def $scc, implicit $exec
200 ; GFX90A_GFX940-NEXT: S_BRANCH %bb.3
201 ; GFX90A_GFX940-NEXT: {{ $}}
202 ; GFX90A_GFX940-NEXT: bb.3 (%ir-block.35):
203 ; GFX90A_GFX940-NEXT: successors: %bb.4(0x80000000)
204 ; GFX90A_GFX940-NEXT: {{ $}}
205 ; GFX90A_GFX940-NEXT: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 0, implicit $exec
206 ; GFX90A_GFX940-NEXT: GLOBAL_ATOMIC_ADD_F32_SADDR [[V_MOV_B32_e32_]], [[STRICT_WWM]], [[REG_SEQUENCE]], 0, 0, implicit $exec :: (load store syncscope("wavefront") monotonic (s32) on %ir.ptr, addrspace 1)
207 ; GFX90A_GFX940-NEXT: {{ $}}
208 ; GFX90A_GFX940-NEXT: bb.4.Flow:
209 ; GFX90A_GFX940-NEXT: successors: %bb.5(0x80000000)
210 ; GFX90A_GFX940-NEXT: {{ $}}
211 ; GFX90A_GFX940-NEXT: SI_END_CF [[SI_IF1]], implicit-def $exec, implicit-def $scc, implicit $exec
212 ; GFX90A_GFX940-NEXT: {{ $}}
213 ; GFX90A_GFX940-NEXT: bb.5 (%ir-block.37):
214 ; GFX90A_GFX940-NEXT: SI_END_CF [[SI_IF]], implicit-def $exec, implicit-def $scc, implicit $exec
215 ; GFX90A_GFX940-NEXT: S_ENDPGM 0
216 %ret = atomicrmw fadd ptr addrspace(1) %ptr, float %data syncscope("wavefront") monotonic
220 declare float @llvm.amdgcn.global.atomic.fadd.f32.p1.f32(ptr addrspace(1), float)
221 declare float @llvm.amdgcn.flat.atomic.fadd.f32.p1.f32(ptr addrspace(1), float)
223 attributes #0 = {"amdgpu-unsafe-fp-atomics"="true" }