1 ; NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
2 ; RUN: llc -global-isel -mtriple=amdgcn-mesa-mesa3d -stop-after=regbankselect -regbankselect-fast -o - %s | FileCheck %s
3 ; RUN: llc -global-isel -mtriple=amdgcn-mesa-mesa3d -stop-after=regbankselect -regbankselect-greedy -o - %s | FileCheck %s
6 define amdgpu_ps void @struct_buffer_store__sgpr_rsrc__vgpr_val__vgpr_vindex__vgpr_voffset__sgpr_soffset(<4 x i32> inreg %rsrc, float %val, i32 %vindex, i32 %voffset, i32 inreg %soffset) {
7 ; CHECK-LABEL: name: struct_buffer_store__sgpr_rsrc__vgpr_val__vgpr_vindex__vgpr_voffset__sgpr_soffset
8 ; CHECK: bb.1 (%ir-block.0):
9 ; CHECK-NEXT: liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $vgpr0, $vgpr1, $vgpr2
11 ; CHECK-NEXT: [[COPY:%[0-9]+]]:sgpr(s32) = COPY $sgpr2
12 ; CHECK-NEXT: [[COPY1:%[0-9]+]]:sgpr(s32) = COPY $sgpr3
13 ; CHECK-NEXT: [[COPY2:%[0-9]+]]:sgpr(s32) = COPY $sgpr4
14 ; CHECK-NEXT: [[COPY3:%[0-9]+]]:sgpr(s32) = COPY $sgpr5
15 ; CHECK-NEXT: [[BUILD_VECTOR:%[0-9]+]]:sgpr(<4 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32)
16 ; CHECK-NEXT: [[COPY4:%[0-9]+]]:vgpr(s32) = COPY $vgpr0
17 ; CHECK-NEXT: [[COPY5:%[0-9]+]]:vgpr(s32) = COPY $vgpr1
18 ; CHECK-NEXT: [[COPY6:%[0-9]+]]:vgpr(s32) = COPY $vgpr2
19 ; CHECK-NEXT: [[COPY7:%[0-9]+]]:sgpr(s32) = COPY $sgpr6
20 ; CHECK-NEXT: G_AMDGPU_BUFFER_STORE [[COPY4]](s32), [[BUILD_VECTOR]](<4 x s32>), [[COPY5]](s32), [[COPY6]], [[COPY7]], 0, 0, -1 :: (dereferenceable store (s32), align 1, addrspace 4)
21 ; CHECK-NEXT: S_ENDPGM 0
22 call void @llvm.amdgcn.struct.buffer.store.f32(float %val, <4 x i32> %rsrc, i32 %vindex, i32 %voffset, i32 %soffset, i32 0)
26 ; Copies for VGPR arguments
27 define amdgpu_ps void @struct_buffer_store__sgpr_rsrc__sgpr_val__sgpr_vindex__sgpr_voffset__sgpr_soffset(<4 x i32> inreg %rsrc, float inreg %val, i32 inreg %vindex, i32 inreg %voffset, i32 inreg %soffset) {
28 ; CHECK-LABEL: name: struct_buffer_store__sgpr_rsrc__sgpr_val__sgpr_vindex__sgpr_voffset__sgpr_soffset
29 ; CHECK: bb.1 (%ir-block.0):
30 ; CHECK-NEXT: liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9
32 ; CHECK-NEXT: [[COPY:%[0-9]+]]:sgpr(s32) = COPY $sgpr2
33 ; CHECK-NEXT: [[COPY1:%[0-9]+]]:sgpr(s32) = COPY $sgpr3
34 ; CHECK-NEXT: [[COPY2:%[0-9]+]]:sgpr(s32) = COPY $sgpr4
35 ; CHECK-NEXT: [[COPY3:%[0-9]+]]:sgpr(s32) = COPY $sgpr5
36 ; CHECK-NEXT: [[BUILD_VECTOR:%[0-9]+]]:sgpr(<4 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32)
37 ; CHECK-NEXT: [[COPY4:%[0-9]+]]:sgpr(s32) = COPY $sgpr6
38 ; CHECK-NEXT: [[COPY5:%[0-9]+]]:sgpr(s32) = COPY $sgpr7
39 ; CHECK-NEXT: [[COPY6:%[0-9]+]]:sgpr(s32) = COPY $sgpr8
40 ; CHECK-NEXT: [[COPY7:%[0-9]+]]:sgpr(s32) = COPY $sgpr9
41 ; CHECK-NEXT: [[COPY8:%[0-9]+]]:vgpr(s32) = COPY [[COPY4]](s32)
42 ; CHECK-NEXT: [[COPY9:%[0-9]+]]:vgpr(s32) = COPY [[COPY5]](s32)
43 ; CHECK-NEXT: [[COPY10:%[0-9]+]]:vgpr(s32) = COPY [[COPY6]](s32)
44 ; CHECK-NEXT: G_AMDGPU_BUFFER_STORE [[COPY8]](s32), [[BUILD_VECTOR]](<4 x s32>), [[COPY9]](s32), [[COPY10]], [[COPY7]], 0, 0, -1 :: (dereferenceable store (s32), align 1, addrspace 4)
45 ; CHECK-NEXT: S_ENDPGM 0
46 call void @llvm.amdgcn.struct.buffer.store.f32(float %val, <4 x i32> %rsrc, i32 %vindex, i32 %voffset, i32 %soffset, i32 0)
51 define amdgpu_ps void @struct_buffer_store__vgpr_rsrc__vgpr_val__vgpr_vindex__vgpr_voffset__sgpr_soffset(<4 x i32> %rsrc, float %val, i32 %vindex, i32 %voffset, i32 inreg %soffset) {
52 ; CHECK-LABEL: name: struct_buffer_store__vgpr_rsrc__vgpr_val__vgpr_vindex__vgpr_voffset__sgpr_soffset
53 ; CHECK: bb.1 (%ir-block.0):
54 ; CHECK-NEXT: successors: %bb.2(0x80000000)
55 ; CHECK-NEXT: liveins: $sgpr2, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6
57 ; CHECK-NEXT: [[COPY:%[0-9]+]]:vgpr(s32) = COPY $vgpr0
58 ; CHECK-NEXT: [[COPY1:%[0-9]+]]:vgpr(s32) = COPY $vgpr1
59 ; CHECK-NEXT: [[COPY2:%[0-9]+]]:vgpr(s32) = COPY $vgpr2
60 ; CHECK-NEXT: [[COPY3:%[0-9]+]]:vgpr(s32) = COPY $vgpr3
61 ; CHECK-NEXT: [[BUILD_VECTOR:%[0-9]+]]:vgpr(<4 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32)
62 ; CHECK-NEXT: [[COPY4:%[0-9]+]]:vgpr(s32) = COPY $vgpr4
63 ; CHECK-NEXT: [[COPY5:%[0-9]+]]:vgpr(s32) = COPY $vgpr5
64 ; CHECK-NEXT: [[COPY6:%[0-9]+]]:vgpr(s32) = COPY $vgpr6
65 ; CHECK-NEXT: [[COPY7:%[0-9]+]]:sgpr(s32) = COPY $sgpr2
66 ; CHECK-NEXT: [[DEF:%[0-9]+]]:sreg_64_xexec = IMPLICIT_DEF
67 ; CHECK-NEXT: [[UV:%[0-9]+]]:vreg_64(s64), [[UV1:%[0-9]+]]:vreg_64(s64) = G_UNMERGE_VALUES [[BUILD_VECTOR]](<4 x s32>)
68 ; CHECK-NEXT: [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
71 ; CHECK-NEXT: successors: %bb.3(0x40000000), %bb.2(0x40000000)
73 ; CHECK-NEXT: [[PHI:%[0-9]+]]:sreg_64_xexec = PHI [[DEF]], %bb.1, %14, %bb.2
74 ; CHECK-NEXT: [[V_READFIRSTLANE_B32_:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[UV]].sub0(s64), implicit $exec
75 ; CHECK-NEXT: [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[UV]].sub1(s64), implicit $exec
76 ; CHECK-NEXT: [[MV:%[0-9]+]]:sreg_64_xexec(s64) = G_MERGE_VALUES [[V_READFIRSTLANE_B32_]](s32), [[V_READFIRSTLANE_B32_1]](s32)
77 ; CHECK-NEXT: [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[MV]](s64), [[UV]](s64), implicit $exec
78 ; CHECK-NEXT: [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[UV1]].sub0(s64), implicit $exec
79 ; CHECK-NEXT: [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[UV1]].sub1(s64), implicit $exec
80 ; CHECK-NEXT: [[MV1:%[0-9]+]]:sreg_64_xexec(s64) = G_MERGE_VALUES [[V_READFIRSTLANE_B32_2]](s32), [[V_READFIRSTLANE_B32_3]](s32)
81 ; CHECK-NEXT: [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[MV1]](s64), [[UV1]](s64), implicit $exec
82 ; CHECK-NEXT: [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_1]], [[V_CMP_EQ_U64_e64_]], implicit-def $scc
83 ; CHECK-NEXT: [[BUILD_VECTOR1:%[0-9]+]]:sgpr(<4 x s32>) = G_BUILD_VECTOR [[V_READFIRSTLANE_B32_]](s32), [[V_READFIRSTLANE_B32_1]](s32), [[V_READFIRSTLANE_B32_2]](s32), [[V_READFIRSTLANE_B32_3]](s32)
84 ; CHECK-NEXT: [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_]], implicit-def $exec, implicit-def $scc, implicit $exec
85 ; CHECK-NEXT: G_AMDGPU_BUFFER_STORE [[COPY4]](s32), [[BUILD_VECTOR1]](<4 x s32>), [[COPY5]](s32), [[COPY6]], [[COPY7]], 0, 0, -1 :: (dereferenceable store (s32), align 1, addrspace 4)
86 ; CHECK-NEXT: $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
87 ; CHECK-NEXT: SI_WATERFALL_LOOP %bb.2, implicit $exec
90 ; CHECK-NEXT: successors: %bb.4(0x80000000)
92 ; CHECK-NEXT: $exec = S_MOV_B64_term [[S_MOV_B64_]]
95 ; CHECK-NEXT: S_ENDPGM 0
96 call void @llvm.amdgcn.struct.buffer.store.f32(float %val, <4 x i32> %rsrc, i32 %vindex, i32 %voffset, i32 %soffset, i32 0)
100 ; Waterfall for soffset
101 define amdgpu_ps void @struct_buffer_store__sgpr_rsrc__vgpr_val__vgpr_vindex__vgpr_voffset__vgpr_soffset(<4 x i32> inreg %rsrc, float %val, i32 %vindex, i32 %voffset, i32 %soffset) {
102 ; CHECK-LABEL: name: struct_buffer_store__sgpr_rsrc__vgpr_val__vgpr_vindex__vgpr_voffset__vgpr_soffset
103 ; CHECK: bb.1 (%ir-block.0):
104 ; CHECK-NEXT: successors: %bb.2(0x80000000)
105 ; CHECK-NEXT: liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $vgpr0, $vgpr1, $vgpr2, $vgpr3
107 ; CHECK-NEXT: [[COPY:%[0-9]+]]:sgpr(s32) = COPY $sgpr2
108 ; CHECK-NEXT: [[COPY1:%[0-9]+]]:sgpr(s32) = COPY $sgpr3
109 ; CHECK-NEXT: [[COPY2:%[0-9]+]]:sgpr(s32) = COPY $sgpr4
110 ; CHECK-NEXT: [[COPY3:%[0-9]+]]:sgpr(s32) = COPY $sgpr5
111 ; CHECK-NEXT: [[BUILD_VECTOR:%[0-9]+]]:sgpr(<4 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32)
112 ; CHECK-NEXT: [[COPY4:%[0-9]+]]:vgpr(s32) = COPY $vgpr0
113 ; CHECK-NEXT: [[COPY5:%[0-9]+]]:vgpr(s32) = COPY $vgpr1
114 ; CHECK-NEXT: [[COPY6:%[0-9]+]]:vgpr(s32) = COPY $vgpr2
115 ; CHECK-NEXT: [[COPY7:%[0-9]+]]:vgpr_32(s32) = COPY $vgpr3
116 ; CHECK-NEXT: [[DEF:%[0-9]+]]:sreg_64_xexec = IMPLICIT_DEF
117 ; CHECK-NEXT: [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
120 ; CHECK-NEXT: successors: %bb.3(0x40000000), %bb.2(0x40000000)
122 ; CHECK-NEXT: [[PHI:%[0-9]+]]:sreg_64_xexec = PHI [[DEF]], %bb.1, %14, %bb.2
123 ; CHECK-NEXT: [[V_READFIRSTLANE_B32_:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[COPY7]](s32), implicit $exec
124 ; CHECK-NEXT: [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_]](s32), [[COPY7]](s32), implicit $exec
125 ; CHECK-NEXT: [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[V_CMP_EQ_U32_e64_]], implicit-def $exec, implicit-def $scc, implicit $exec
126 ; CHECK-NEXT: G_AMDGPU_BUFFER_STORE [[COPY4]](s32), [[BUILD_VECTOR]](<4 x s32>), [[COPY5]](s32), [[COPY6]], [[V_READFIRSTLANE_B32_]], 0, 0, -1 :: (dereferenceable store (s32), align 1, addrspace 4)
127 ; CHECK-NEXT: $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
128 ; CHECK-NEXT: SI_WATERFALL_LOOP %bb.2, implicit $exec
131 ; CHECK-NEXT: successors: %bb.4(0x80000000)
133 ; CHECK-NEXT: $exec = S_MOV_B64_term [[S_MOV_B64_]]
136 ; CHECK-NEXT: S_ENDPGM 0
137 call void @llvm.amdgcn.struct.buffer.store.f32(float %val, <4 x i32> %rsrc, i32 %vindex, i32 %voffset, i32 %soffset, i32 0)
141 ; Waterfall for rsrc and soffset
142 define amdgpu_ps void @struct_buffer_store__vgpr_rsrc__vgpr_val__vgpr_vindex__vgpr_voffset__vgpr_soffset(<4 x i32> %rsrc, float %val, i32 %vindex, i32 %voffset, i32 %soffset) {
143 ; CHECK-LABEL: name: struct_buffer_store__vgpr_rsrc__vgpr_val__vgpr_vindex__vgpr_voffset__vgpr_soffset
144 ; CHECK: bb.1 (%ir-block.0):
145 ; CHECK-NEXT: successors: %bb.2(0x80000000)
146 ; CHECK-NEXT: liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7
148 ; CHECK-NEXT: [[COPY:%[0-9]+]]:vgpr(s32) = COPY $vgpr0
149 ; CHECK-NEXT: [[COPY1:%[0-9]+]]:vgpr(s32) = COPY $vgpr1
150 ; CHECK-NEXT: [[COPY2:%[0-9]+]]:vgpr(s32) = COPY $vgpr2
151 ; CHECK-NEXT: [[COPY3:%[0-9]+]]:vgpr(s32) = COPY $vgpr3
152 ; CHECK-NEXT: [[BUILD_VECTOR:%[0-9]+]]:vgpr(<4 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32)
153 ; CHECK-NEXT: [[COPY4:%[0-9]+]]:vgpr(s32) = COPY $vgpr4
154 ; CHECK-NEXT: [[COPY5:%[0-9]+]]:vgpr(s32) = COPY $vgpr5
155 ; CHECK-NEXT: [[COPY6:%[0-9]+]]:vgpr(s32) = COPY $vgpr6
156 ; CHECK-NEXT: [[COPY7:%[0-9]+]]:vgpr_32(s32) = COPY $vgpr7
157 ; CHECK-NEXT: [[DEF:%[0-9]+]]:sreg_64_xexec = IMPLICIT_DEF
158 ; CHECK-NEXT: [[UV:%[0-9]+]]:vreg_64(s64), [[UV1:%[0-9]+]]:vreg_64(s64) = G_UNMERGE_VALUES [[BUILD_VECTOR]](<4 x s32>)
159 ; CHECK-NEXT: [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
162 ; CHECK-NEXT: successors: %bb.3(0x40000000), %bb.2(0x40000000)
164 ; CHECK-NEXT: [[PHI:%[0-9]+]]:sreg_64_xexec = PHI [[DEF]], %bb.1, %14, %bb.2
165 ; CHECK-NEXT: [[V_READFIRSTLANE_B32_:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[UV]].sub0(s64), implicit $exec
166 ; CHECK-NEXT: [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[UV]].sub1(s64), implicit $exec
167 ; CHECK-NEXT: [[MV:%[0-9]+]]:sreg_64_xexec(s64) = G_MERGE_VALUES [[V_READFIRSTLANE_B32_]](s32), [[V_READFIRSTLANE_B32_1]](s32)
168 ; CHECK-NEXT: [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[MV]](s64), [[UV]](s64), implicit $exec
169 ; CHECK-NEXT: [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[UV1]].sub0(s64), implicit $exec
170 ; CHECK-NEXT: [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[UV1]].sub1(s64), implicit $exec
171 ; CHECK-NEXT: [[MV1:%[0-9]+]]:sreg_64_xexec(s64) = G_MERGE_VALUES [[V_READFIRSTLANE_B32_2]](s32), [[V_READFIRSTLANE_B32_3]](s32)
172 ; CHECK-NEXT: [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[MV1]](s64), [[UV1]](s64), implicit $exec
173 ; CHECK-NEXT: [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_1]], [[V_CMP_EQ_U64_e64_]], implicit-def $scc
174 ; CHECK-NEXT: [[BUILD_VECTOR1:%[0-9]+]]:sgpr(<4 x s32>) = G_BUILD_VECTOR [[V_READFIRSTLANE_B32_]](s32), [[V_READFIRSTLANE_B32_1]](s32), [[V_READFIRSTLANE_B32_2]](s32), [[V_READFIRSTLANE_B32_3]](s32)
175 ; CHECK-NEXT: [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[COPY7]](s32), implicit $exec
176 ; CHECK-NEXT: [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]](s32), [[COPY7]](s32), implicit $exec
177 ; CHECK-NEXT: [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U32_e64_]], [[S_AND_B64_]], implicit-def $scc
178 ; CHECK-NEXT: [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_1]], implicit-def $exec, implicit-def $scc, implicit $exec
179 ; CHECK-NEXT: G_AMDGPU_BUFFER_STORE [[COPY4]](s32), [[BUILD_VECTOR1]](<4 x s32>), [[COPY5]](s32), [[COPY6]], [[V_READFIRSTLANE_B32_4]], 0, 0, -1 :: (dereferenceable store (s32), align 1, addrspace 4)
180 ; CHECK-NEXT: $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
181 ; CHECK-NEXT: SI_WATERFALL_LOOP %bb.2, implicit $exec
184 ; CHECK-NEXT: successors: %bb.4(0x80000000)
186 ; CHECK-NEXT: $exec = S_MOV_B64_term [[S_MOV_B64_]]
189 ; CHECK-NEXT: S_ENDPGM 0
190 call void @llvm.amdgcn.struct.buffer.store.f32(float %val, <4 x i32> %rsrc, i32 %vindex, i32 %voffset, i32 %soffset, i32 0)
194 declare void @llvm.amdgcn.struct.buffer.store.f32(float, <4 x i32>, i32, i32, i32, i32 immarg)