1; NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
2; RUN: llc -global-isel -mtriple=amdgcn-mesa-mesa3d -mcpu=tahiti -stop-after=instruction-select -verify-machineinstrs -o - %s | FileCheck %s
3; RUN: llc -global-isel -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1010 -mattr=+wavefrontsize64 -stop-after=instruction-select -verify-machineinstrs -o - %s | FileCheck %s
4
5define amdgpu_ps float @struct_tbuffer_load_f32__sgpr_rsrc__vgpr_vindex__vgpr_voffset__sgpr_soffset(<4 x i32> inreg %rsrc, i32 %vindex, i32 %voffset, i32 inreg %soffset) {
6  ; CHECK-LABEL: name: struct_tbuffer_load_f32__sgpr_rsrc__vgpr_vindex__vgpr_voffset__sgpr_soffset
7  ; CHECK: bb.1 (%ir-block.0):
8  ; CHECK:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $vgpr0, $vgpr1
9  ; CHECK:   [[COPY:%[0-9]+]]:sreg_32 = COPY $sgpr2
10  ; CHECK:   [[COPY1:%[0-9]+]]:sreg_32 = COPY $sgpr3
11  ; CHECK:   [[COPY2:%[0-9]+]]:sreg_32 = COPY $sgpr4
12  ; CHECK:   [[COPY3:%[0-9]+]]:sreg_32 = COPY $sgpr5
13  ; CHECK:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0
14  ; CHECK:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr1
15  ; CHECK:   [[COPY6:%[0-9]+]]:sreg_32 = COPY $sgpr6
16  ; CHECK:   [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3
17  ; CHECK:   [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY4]], %subreg.sub0, [[COPY5]], %subreg.sub1
18  ; CHECK:   [[TBUFFER_LOAD_FORMAT_X_BOTHEN:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_BOTHEN [[REG_SEQUENCE1]], [[REG_SEQUENCE]], [[COPY6]], 0, 78, 0, 0, 0, 0, 0, implicit $exec :: (dereferenceable load 4 from custom "TargetCustom7", align 1, addrspace 4)
19  ; CHECK:   $vgpr0 = COPY [[TBUFFER_LOAD_FORMAT_X_BOTHEN]]
20  ; CHECK:   SI_RETURN_TO_EPILOG implicit $vgpr0
21  %val = call float @llvm.amdgcn.struct.tbuffer.load.f32(<4 x i32> %rsrc, i32 %vindex, i32 %voffset, i32 %soffset, i32 78, i32 0)
22  ret float %val
23}
24
25define amdgpu_ps <2 x float> @struct_tbuffer_load_v2f32__sgpr_rsrc__vgpr_vindex__vgpr_voffset__sgpr_soffset(<4 x i32> inreg %rsrc, i32 %vindex, i32 %voffset, i32 inreg %soffset) {
26  ; CHECK-LABEL: name: struct_tbuffer_load_v2f32__sgpr_rsrc__vgpr_vindex__vgpr_voffset__sgpr_soffset
27  ; CHECK: bb.1 (%ir-block.0):
28  ; CHECK:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $vgpr0, $vgpr1
29  ; CHECK:   [[COPY:%[0-9]+]]:sreg_32 = COPY $sgpr2
30  ; CHECK:   [[COPY1:%[0-9]+]]:sreg_32 = COPY $sgpr3
31  ; CHECK:   [[COPY2:%[0-9]+]]:sreg_32 = COPY $sgpr4
32  ; CHECK:   [[COPY3:%[0-9]+]]:sreg_32 = COPY $sgpr5
33  ; CHECK:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0
34  ; CHECK:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr1
35  ; CHECK:   [[COPY6:%[0-9]+]]:sreg_32 = COPY $sgpr6
36  ; CHECK:   [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3
37  ; CHECK:   [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY4]], %subreg.sub0, [[COPY5]], %subreg.sub1
38  ; CHECK:   [[TBUFFER_LOAD_FORMAT_XY_BOTHEN:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_BOTHEN [[REG_SEQUENCE1]], [[REG_SEQUENCE]], [[COPY6]], 0, 78, 0, 0, 0, 0, 0, implicit $exec :: (dereferenceable load 8 from custom "TargetCustom7", align 1, addrspace 4)
39  ; CHECK:   [[COPY7:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XY_BOTHEN]].sub0
40  ; CHECK:   [[COPY8:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XY_BOTHEN]].sub1
41  ; CHECK:   $vgpr0 = COPY [[COPY7]]
42  ; CHECK:   $vgpr1 = COPY [[COPY8]]
43  ; CHECK:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1
44  %val = call <2 x float> @llvm.amdgcn.struct.tbuffer.load.v2f32(<4 x i32> %rsrc, i32 %vindex, i32 %voffset, i32 %soffset, i32 78, i32 0)
45  ret <2 x float> %val
46}
47
48define amdgpu_ps <3 x float> @struct_tbuffer_load_v3f32__sgpr_rsrc__vgpr_vindex__vgpr_voffset__sgpr_soffset(<4 x i32> inreg %rsrc, i32 %vindex, i32 %voffset, i32 inreg %soffset) {
49  ; CHECK-LABEL: name: struct_tbuffer_load_v3f32__sgpr_rsrc__vgpr_vindex__vgpr_voffset__sgpr_soffset
50  ; CHECK: bb.1 (%ir-block.0):
51  ; CHECK:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $vgpr0, $vgpr1
52  ; CHECK:   [[COPY:%[0-9]+]]:sreg_32 = COPY $sgpr2
53  ; CHECK:   [[COPY1:%[0-9]+]]:sreg_32 = COPY $sgpr3
54  ; CHECK:   [[COPY2:%[0-9]+]]:sreg_32 = COPY $sgpr4
55  ; CHECK:   [[COPY3:%[0-9]+]]:sreg_32 = COPY $sgpr5
56  ; CHECK:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0
57  ; CHECK:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr1
58  ; CHECK:   [[COPY6:%[0-9]+]]:sreg_32 = COPY $sgpr6
59  ; CHECK:   [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3
60  ; CHECK:   [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY4]], %subreg.sub0, [[COPY5]], %subreg.sub1
61  ; CHECK:   [[TBUFFER_LOAD_FORMAT_XYZ_BOTHEN:%[0-9]+]]:vreg_96 = TBUFFER_LOAD_FORMAT_XYZ_BOTHEN [[REG_SEQUENCE1]], [[REG_SEQUENCE]], [[COPY6]], 0, 78, 0, 0, 0, 0, 0, implicit $exec :: (dereferenceable load 12 from custom "TargetCustom7", align 1, addrspace 4)
62  ; CHECK:   [[COPY7:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XYZ_BOTHEN]].sub0
63  ; CHECK:   [[COPY8:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XYZ_BOTHEN]].sub1
64  ; CHECK:   [[COPY9:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XYZ_BOTHEN]].sub2
65  ; CHECK:   $vgpr0 = COPY [[COPY7]]
66  ; CHECK:   $vgpr1 = COPY [[COPY8]]
67  ; CHECK:   $vgpr2 = COPY [[COPY9]]
68  ; CHECK:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2
69  %val = call <3 x float> @llvm.amdgcn.struct.tbuffer.load.v3f32(<4 x i32> %rsrc, i32 %vindex, i32 %voffset, i32 %soffset, i32 78, i32 0)
70  ret <3 x float> %val
71}
72
73define amdgpu_ps <4 x float> @struct_tbuffer_load_v4f32__sgpr_rsrc__vgpr_vindex__vgpr_voffset__sgpr_soffset(<4 x i32> inreg %rsrc, i32 %vindex, i32 %voffset, i32 inreg %soffset) {
74  ; CHECK-LABEL: name: struct_tbuffer_load_v4f32__sgpr_rsrc__vgpr_vindex__vgpr_voffset__sgpr_soffset
75  ; CHECK: bb.1 (%ir-block.0):
76  ; CHECK:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $vgpr0, $vgpr1
77  ; CHECK:   [[COPY:%[0-9]+]]:sreg_32 = COPY $sgpr2
78  ; CHECK:   [[COPY1:%[0-9]+]]:sreg_32 = COPY $sgpr3
79  ; CHECK:   [[COPY2:%[0-9]+]]:sreg_32 = COPY $sgpr4
80  ; CHECK:   [[COPY3:%[0-9]+]]:sreg_32 = COPY $sgpr5
81  ; CHECK:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0
82  ; CHECK:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr1
83  ; CHECK:   [[COPY6:%[0-9]+]]:sreg_32 = COPY $sgpr6
84  ; CHECK:   [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3
85  ; CHECK:   [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY4]], %subreg.sub0, [[COPY5]], %subreg.sub1
86  ; CHECK:   [[TBUFFER_LOAD_FORMAT_XYZW_BOTHEN:%[0-9]+]]:vreg_128 = TBUFFER_LOAD_FORMAT_XYZW_BOTHEN [[REG_SEQUENCE1]], [[REG_SEQUENCE]], [[COPY6]], 0, 78, 0, 0, 0, 0, 0, implicit $exec :: (dereferenceable load 16 from custom "TargetCustom7", align 1, addrspace 4)
87  ; CHECK:   [[COPY7:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XYZW_BOTHEN]].sub0
88  ; CHECK:   [[COPY8:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XYZW_BOTHEN]].sub1
89  ; CHECK:   [[COPY9:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XYZW_BOTHEN]].sub2
90  ; CHECK:   [[COPY10:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XYZW_BOTHEN]].sub3
91  ; CHECK:   $vgpr0 = COPY [[COPY7]]
92  ; CHECK:   $vgpr1 = COPY [[COPY8]]
93  ; CHECK:   $vgpr2 = COPY [[COPY9]]
94  ; CHECK:   $vgpr3 = COPY [[COPY10]]
95  ; CHECK:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
96  %val = call <4 x float> @llvm.amdgcn.struct.tbuffer.load.v4f32(<4 x i32> %rsrc, i32 %vindex, i32 %voffset, i32 %soffset, i32 78, i32 0)
97  ret <4 x float> %val
98}
99
100define amdgpu_ps float @struct_tbuffer_load_f32__sgpr_rsrc__vgpr_vindex__vgpr_voffset__sgpr_soffset_vindex0(<4 x i32> inreg %rsrc, i32 %voffset, i32 inreg %soffset) {
101  ; CHECK-LABEL: name: struct_tbuffer_load_f32__sgpr_rsrc__vgpr_vindex__vgpr_voffset__sgpr_soffset_vindex0
102  ; CHECK: bb.1 (%ir-block.0):
103  ; CHECK:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $vgpr0
104  ; CHECK:   [[COPY:%[0-9]+]]:sreg_32 = COPY $sgpr2
105  ; CHECK:   [[COPY1:%[0-9]+]]:sreg_32 = COPY $sgpr3
106  ; CHECK:   [[COPY2:%[0-9]+]]:sreg_32 = COPY $sgpr4
107  ; CHECK:   [[COPY3:%[0-9]+]]:sreg_32 = COPY $sgpr5
108  ; CHECK:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0
109  ; CHECK:   [[COPY5:%[0-9]+]]:sreg_32 = COPY $sgpr6
110  ; CHECK:   [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3
111  ; CHECK:   [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 0
112  ; CHECK:   [[COPY6:%[0-9]+]]:vgpr_32 = COPY [[S_MOV_B32_]]
113  ; CHECK:   [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY6]], %subreg.sub0, [[COPY4]], %subreg.sub1
114  ; CHECK:   [[TBUFFER_LOAD_FORMAT_X_BOTHEN:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_BOTHEN [[REG_SEQUENCE1]], [[REG_SEQUENCE]], [[COPY5]], 0, 78, 0, 0, 0, 0, 0, implicit $exec :: (dereferenceable load 4 from custom "TargetCustom7", align 1, addrspace 4)
115  ; CHECK:   $vgpr0 = COPY [[TBUFFER_LOAD_FORMAT_X_BOTHEN]]
116  ; CHECK:   SI_RETURN_TO_EPILOG implicit $vgpr0
117  %val = call float @llvm.amdgcn.struct.tbuffer.load.f32(<4 x i32> %rsrc, i32 0, i32 %voffset, i32 %soffset, i32 78, i32 0)
118  ret float %val
119}
120
121define amdgpu_ps <4 x float> @struct_tbuffer_load_v4f32__vgpr_rsrc__sgpr_vindex__sgpr_voffset__vgpr_soffset(<4 x i32> %rsrc, i32 inreg %vindex, i32 inreg %voffset, i32 %soffset) {
122  ; CHECK-LABEL: name: struct_tbuffer_load_v4f32__vgpr_rsrc__sgpr_vindex__sgpr_voffset__vgpr_soffset
123  ; CHECK: bb.1 (%ir-block.0):
124  ; CHECK:   successors: %bb.2(0x80000000)
125  ; CHECK:   liveins: $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4
126  ; CHECK:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
127  ; CHECK:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
128  ; CHECK:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr2
129  ; CHECK:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr3
130  ; CHECK:   [[COPY4:%[0-9]+]]:sreg_32 = COPY $sgpr2
131  ; CHECK:   [[COPY5:%[0-9]+]]:sreg_32 = COPY $sgpr3
132  ; CHECK:   [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr4
133  ; CHECK:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3
134  ; CHECK:   [[COPY7:%[0-9]+]]:vgpr_32 = COPY [[COPY4]]
135  ; CHECK:   [[COPY8:%[0-9]+]]:vgpr_32 = COPY [[COPY5]]
136  ; CHECK:   [[COPY9:%[0-9]+]]:vreg_64 = COPY [[REG_SEQUENCE]].sub0_sub1
137  ; CHECK:   [[COPY10:%[0-9]+]]:vreg_64 = COPY [[REG_SEQUENCE]].sub2_sub3
138  ; CHECK:   [[S_MOV_B64_term:%[0-9]+]]:sreg_64_xexec = S_MOV_B64_term $exec
139  ; CHECK: bb.2:
140  ; CHECK:   successors: %bb.3(0x40000000), %bb.2(0x40000000)
141  ; CHECK:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sreg_32_xm0 = V_READFIRSTLANE_B32 [[COPY9]].sub0, implicit $exec
142  ; CHECK:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sreg_32_xm0 = V_READFIRSTLANE_B32 [[COPY9]].sub1, implicit $exec
143  ; CHECK:   [[REG_SEQUENCE1:%[0-9]+]]:sreg_64_xexec = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
144  ; CHECK:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE1]], [[COPY9]], implicit $exec
145  ; CHECK:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sreg_32_xm0 = V_READFIRSTLANE_B32 [[COPY10]].sub0, implicit $exec
146  ; CHECK:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sreg_32_xm0 = V_READFIRSTLANE_B32 [[COPY10]].sub1, implicit $exec
147  ; CHECK:   [[REG_SEQUENCE2:%[0-9]+]]:sreg_64_xexec = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
148  ; CHECK:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[COPY10]], implicit $exec
149  ; CHECK:   [[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
150  ; CHECK:   [[REG_SEQUENCE3:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1, [[V_READFIRSTLANE_B32_2]], %subreg.sub2, [[V_READFIRSTLANE_B32_3]], %subreg.sub3
151  ; CHECK:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sreg_32_xm0 = V_READFIRSTLANE_B32 [[COPY6]], implicit $exec
152  ; CHECK:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]], [[COPY6]], implicit $exec
153  ; CHECK:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U32_e64_]], [[S_AND_B64_]], implicit-def $scc
154  ; CHECK:   [[REG_SEQUENCE4:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY7]], %subreg.sub0, [[COPY8]], %subreg.sub1
155  ; CHECK:   [[TBUFFER_LOAD_FORMAT_XYZW_BOTHEN:%[0-9]+]]:vreg_128 = TBUFFER_LOAD_FORMAT_XYZW_BOTHEN [[REG_SEQUENCE4]], [[REG_SEQUENCE3]], [[V_READFIRSTLANE_B32_4]], 0, 78, 0, 0, 0, 0, 0, implicit $exec :: (dereferenceable load 16 from custom "TargetCustom7", align 1, addrspace 4)
156  ; CHECK:   [[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
157  ; CHECK:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
158  ; CHECK:   S_CBRANCH_EXECNZ %bb.2, implicit $exec
159  ; CHECK: bb.3:
160  ; CHECK:   successors: %bb.4(0x80000000)
161  ; CHECK:   $exec = S_MOV_B64_term [[S_MOV_B64_term]]
162  ; CHECK: bb.4:
163  ; CHECK:   [[COPY11:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XYZW_BOTHEN]].sub0
164  ; CHECK:   [[COPY12:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XYZW_BOTHEN]].sub1
165  ; CHECK:   [[COPY13:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XYZW_BOTHEN]].sub2
166  ; CHECK:   [[COPY14:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XYZW_BOTHEN]].sub3
167  ; CHECK:   $vgpr0 = COPY [[COPY11]]
168  ; CHECK:   $vgpr1 = COPY [[COPY12]]
169  ; CHECK:   $vgpr2 = COPY [[COPY13]]
170  ; CHECK:   $vgpr3 = COPY [[COPY14]]
171  ; CHECK:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
172  %val = call <4 x float> @llvm.amdgcn.struct.tbuffer.load.v4f32(<4 x i32> %rsrc, i32 %vindex, i32 %voffset, i32 %soffset, i32 78, i32 0)
173  ret <4 x float> %val
174}
175
176define amdgpu_ps float @struct_tbuffer_load_f32__sgpr_rsrc__vgpr_vindex__vgpr_voffset__sgpr_soffset_voffset_add4095(<4 x i32> inreg %rsrc, i32 %vindex, i32 %voffset.base, i32 inreg %soffset) {
177  ; CHECK-LABEL: name: struct_tbuffer_load_f32__sgpr_rsrc__vgpr_vindex__vgpr_voffset__sgpr_soffset_voffset_add4095
178  ; CHECK: bb.1 (%ir-block.0):
179  ; CHECK:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $vgpr0, $vgpr1
180  ; CHECK:   [[COPY:%[0-9]+]]:sreg_32 = COPY $sgpr2
181  ; CHECK:   [[COPY1:%[0-9]+]]:sreg_32 = COPY $sgpr3
182  ; CHECK:   [[COPY2:%[0-9]+]]:sreg_32 = COPY $sgpr4
183  ; CHECK:   [[COPY3:%[0-9]+]]:sreg_32 = COPY $sgpr5
184  ; CHECK:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0
185  ; CHECK:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr1
186  ; CHECK:   [[COPY6:%[0-9]+]]:sreg_32 = COPY $sgpr6
187  ; CHECK:   [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3
188  ; CHECK:   [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY4]], %subreg.sub0, [[COPY5]], %subreg.sub1
189  ; CHECK:   [[TBUFFER_LOAD_FORMAT_X_BOTHEN:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_BOTHEN [[REG_SEQUENCE1]], [[REG_SEQUENCE]], [[COPY6]], 4095, 78, 0, 0, 0, 0, 0, implicit $exec :: (dereferenceable load 4 from custom "TargetCustom7" + 4095, align 1, addrspace 4)
190  ; CHECK:   $vgpr0 = COPY [[TBUFFER_LOAD_FORMAT_X_BOTHEN]]
191  ; CHECK:   SI_RETURN_TO_EPILOG implicit $vgpr0
192  %voffset = add i32 %voffset.base, 4095
193  %val = call float @llvm.amdgcn.struct.tbuffer.load.f32(<4 x i32> %rsrc, i32 %vindex, i32 %voffset, i32 %soffset, i32 78, i32 0)
194  ret float %val
195}
196
197declare float @llvm.amdgcn.struct.tbuffer.load.f32(<4 x i32>, i32, i32, i32, i32 immarg, i32 immarg) #0
198declare <2 x float> @llvm.amdgcn.struct.tbuffer.load.v2f32(<4 x i32>, i32, i32, i32, i32 immarg, i32 immarg) #0
199declare <3 x float> @llvm.amdgcn.struct.tbuffer.load.v3f32(<4 x i32>, i32, i32, i32, i32 immarg, i32 immarg) #0
200declare <4 x float> @llvm.amdgcn.struct.tbuffer.load.v4f32(<4 x i32>, i32, i32, i32, i32 immarg, i32 immarg) #0
201
202attributes #0 = { nounwind readonly }
203