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 --check-prefixes=CHECK %s
3; RUN: llc -global-isel -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1010 -mattr=+wavefrontsize64 -stop-after=instruction-select -verify-machineinstrs -o - %s | FileCheck --check-prefixes=CHECK %s
4; RUN: llc -global-isel -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1100 -mattr=+wavefrontsize64 -stop-after=instruction-select -verify-machineinstrs -o - %s | FileCheck --check-prefixes=CHECK %s
5
6define 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) {
7  ; CHECK-LABEL: name: struct_tbuffer_load_f32__sgpr_rsrc__vgpr_vindex__vgpr_voffset__sgpr_soffset
8  ; CHECK: bb.1 (%ir-block.0):
9  ; CHECK-NEXT:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $vgpr0, $vgpr1
10  ; CHECK-NEXT: {{  $}}
11  ; CHECK-NEXT:   [[COPY:%[0-9]+]]:sreg_32 = COPY $sgpr2
12  ; CHECK-NEXT:   [[COPY1:%[0-9]+]]:sreg_32 = COPY $sgpr3
13  ; CHECK-NEXT:   [[COPY2:%[0-9]+]]:sreg_32 = COPY $sgpr4
14  ; CHECK-NEXT:   [[COPY3:%[0-9]+]]:sreg_32 = COPY $sgpr5
15  ; CHECK-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3
16  ; CHECK-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0
17  ; CHECK-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr1
18  ; CHECK-NEXT:   [[COPY6:%[0-9]+]]:sreg_32 = COPY $sgpr6
19  ; CHECK-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY4]], %subreg.sub0, [[COPY5]], %subreg.sub1
20  ; CHECK-NEXT:   [[TBUFFER_LOAD_FORMAT_X_BOTHEN:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_BOTHEN [[REG_SEQUENCE1]], [[REG_SEQUENCE]], [[COPY6]], 0, 78, 0, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4)
21  ; CHECK-NEXT:   $vgpr0 = COPY [[TBUFFER_LOAD_FORMAT_X_BOTHEN]]
22  ; CHECK-NEXT:   SI_RETURN_TO_EPILOG implicit $vgpr0
23  %val = call float @llvm.amdgcn.struct.tbuffer.load.f32(<4 x i32> %rsrc, i32 %vindex, i32 %voffset, i32 %soffset, i32 78, i32 0)
24  ret float %val
25}
26
27define 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) {
28  ; CHECK-LABEL: name: struct_tbuffer_load_v2f32__sgpr_rsrc__vgpr_vindex__vgpr_voffset__sgpr_soffset
29  ; CHECK: bb.1 (%ir-block.0):
30  ; CHECK-NEXT:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $vgpr0, $vgpr1
31  ; CHECK-NEXT: {{  $}}
32  ; CHECK-NEXT:   [[COPY:%[0-9]+]]:sreg_32 = COPY $sgpr2
33  ; CHECK-NEXT:   [[COPY1:%[0-9]+]]:sreg_32 = COPY $sgpr3
34  ; CHECK-NEXT:   [[COPY2:%[0-9]+]]:sreg_32 = COPY $sgpr4
35  ; CHECK-NEXT:   [[COPY3:%[0-9]+]]:sreg_32 = COPY $sgpr5
36  ; CHECK-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3
37  ; CHECK-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0
38  ; CHECK-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr1
39  ; CHECK-NEXT:   [[COPY6:%[0-9]+]]:sreg_32 = COPY $sgpr6
40  ; CHECK-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY4]], %subreg.sub0, [[COPY5]], %subreg.sub1
41  ; CHECK-NEXT:   [[TBUFFER_LOAD_FORMAT_XY_BOTHEN:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_BOTHEN [[REG_SEQUENCE1]], [[REG_SEQUENCE]], [[COPY6]], 0, 78, 0, 0, 0, implicit $exec :: (dereferenceable load (<2 x s32>), align 1, addrspace 4)
42  ; CHECK-NEXT:   [[COPY7:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XY_BOTHEN]].sub0
43  ; CHECK-NEXT:   [[COPY8:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XY_BOTHEN]].sub1
44  ; CHECK-NEXT:   $vgpr0 = COPY [[COPY7]]
45  ; CHECK-NEXT:   $vgpr1 = COPY [[COPY8]]
46  ; CHECK-NEXT:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1
47  %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)
48  ret <2 x float> %val
49}
50
51define 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) {
52  ; CHECK-LABEL: name: struct_tbuffer_load_v3f32__sgpr_rsrc__vgpr_vindex__vgpr_voffset__sgpr_soffset
53  ; CHECK: bb.1 (%ir-block.0):
54  ; CHECK-NEXT:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $vgpr0, $vgpr1
55  ; CHECK-NEXT: {{  $}}
56  ; CHECK-NEXT:   [[COPY:%[0-9]+]]:sreg_32 = COPY $sgpr2
57  ; CHECK-NEXT:   [[COPY1:%[0-9]+]]:sreg_32 = COPY $sgpr3
58  ; CHECK-NEXT:   [[COPY2:%[0-9]+]]:sreg_32 = COPY $sgpr4
59  ; CHECK-NEXT:   [[COPY3:%[0-9]+]]:sreg_32 = COPY $sgpr5
60  ; CHECK-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3
61  ; CHECK-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0
62  ; CHECK-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr1
63  ; CHECK-NEXT:   [[COPY6:%[0-9]+]]:sreg_32 = COPY $sgpr6
64  ; CHECK-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY4]], %subreg.sub0, [[COPY5]], %subreg.sub1
65  ; CHECK-NEXT:   [[TBUFFER_LOAD_FORMAT_XYZ_BOTHEN:%[0-9]+]]:vreg_96 = TBUFFER_LOAD_FORMAT_XYZ_BOTHEN [[REG_SEQUENCE1]], [[REG_SEQUENCE]], [[COPY6]], 0, 78, 0, 0, 0, implicit $exec :: (dereferenceable load (<3 x s32>), align 1, addrspace 4)
66  ; CHECK-NEXT:   [[COPY7:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XYZ_BOTHEN]].sub0
67  ; CHECK-NEXT:   [[COPY8:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XYZ_BOTHEN]].sub1
68  ; CHECK-NEXT:   [[COPY9:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XYZ_BOTHEN]].sub2
69  ; CHECK-NEXT:   $vgpr0 = COPY [[COPY7]]
70  ; CHECK-NEXT:   $vgpr1 = COPY [[COPY8]]
71  ; CHECK-NEXT:   $vgpr2 = COPY [[COPY9]]
72  ; CHECK-NEXT:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2
73  %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)
74  ret <3 x float> %val
75}
76
77define 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) {
78  ; CHECK-LABEL: name: struct_tbuffer_load_v4f32__sgpr_rsrc__vgpr_vindex__vgpr_voffset__sgpr_soffset
79  ; CHECK: bb.1 (%ir-block.0):
80  ; CHECK-NEXT:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $vgpr0, $vgpr1
81  ; CHECK-NEXT: {{  $}}
82  ; CHECK-NEXT:   [[COPY:%[0-9]+]]:sreg_32 = COPY $sgpr2
83  ; CHECK-NEXT:   [[COPY1:%[0-9]+]]:sreg_32 = COPY $sgpr3
84  ; CHECK-NEXT:   [[COPY2:%[0-9]+]]:sreg_32 = COPY $sgpr4
85  ; CHECK-NEXT:   [[COPY3:%[0-9]+]]:sreg_32 = COPY $sgpr5
86  ; CHECK-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3
87  ; CHECK-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0
88  ; CHECK-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr1
89  ; CHECK-NEXT:   [[COPY6:%[0-9]+]]:sreg_32 = COPY $sgpr6
90  ; CHECK-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY4]], %subreg.sub0, [[COPY5]], %subreg.sub1
91  ; CHECK-NEXT:   [[TBUFFER_LOAD_FORMAT_XYZW_BOTHEN:%[0-9]+]]:vreg_128 = TBUFFER_LOAD_FORMAT_XYZW_BOTHEN [[REG_SEQUENCE1]], [[REG_SEQUENCE]], [[COPY6]], 0, 78, 0, 0, 0, implicit $exec :: (dereferenceable load (<4 x s32>), align 1, addrspace 4)
92  ; CHECK-NEXT:   [[COPY7:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XYZW_BOTHEN]].sub0
93  ; CHECK-NEXT:   [[COPY8:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XYZW_BOTHEN]].sub1
94  ; CHECK-NEXT:   [[COPY9:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XYZW_BOTHEN]].sub2
95  ; CHECK-NEXT:   [[COPY10:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XYZW_BOTHEN]].sub3
96  ; CHECK-NEXT:   $vgpr0 = COPY [[COPY7]]
97  ; CHECK-NEXT:   $vgpr1 = COPY [[COPY8]]
98  ; CHECK-NEXT:   $vgpr2 = COPY [[COPY9]]
99  ; CHECK-NEXT:   $vgpr3 = COPY [[COPY10]]
100  ; CHECK-NEXT:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
101  %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)
102  ret <4 x float> %val
103}
104
105define 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) {
106  ; CHECK-LABEL: name: struct_tbuffer_load_f32__sgpr_rsrc__vgpr_vindex__vgpr_voffset__sgpr_soffset_vindex0
107  ; CHECK: bb.1 (%ir-block.0):
108  ; CHECK-NEXT:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $vgpr0
109  ; CHECK-NEXT: {{  $}}
110  ; CHECK-NEXT:   [[COPY:%[0-9]+]]:sreg_32 = COPY $sgpr2
111  ; CHECK-NEXT:   [[COPY1:%[0-9]+]]:sreg_32 = COPY $sgpr3
112  ; CHECK-NEXT:   [[COPY2:%[0-9]+]]:sreg_32 = COPY $sgpr4
113  ; CHECK-NEXT:   [[COPY3:%[0-9]+]]:sreg_32 = COPY $sgpr5
114  ; CHECK-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3
115  ; CHECK-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0
116  ; CHECK-NEXT:   [[COPY5:%[0-9]+]]:sreg_32 = COPY $sgpr6
117  ; CHECK-NEXT:   [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 0
118  ; CHECK-NEXT:   [[COPY6:%[0-9]+]]:vgpr_32 = COPY [[S_MOV_B32_]]
119  ; CHECK-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY6]], %subreg.sub0, [[COPY4]], %subreg.sub1
120  ; CHECK-NEXT:   [[TBUFFER_LOAD_FORMAT_X_BOTHEN:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_BOTHEN [[REG_SEQUENCE1]], [[REG_SEQUENCE]], [[COPY5]], 0, 78, 0, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4)
121  ; CHECK-NEXT:   $vgpr0 = COPY [[TBUFFER_LOAD_FORMAT_X_BOTHEN]]
122  ; CHECK-NEXT:   SI_RETURN_TO_EPILOG implicit $vgpr0
123  %val = call float @llvm.amdgcn.struct.tbuffer.load.f32(<4 x i32> %rsrc, i32 0, i32 %voffset, i32 %soffset, i32 78, i32 0)
124  ret float %val
125}
126
127define 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) {
128  ; CHECK-LABEL: name: struct_tbuffer_load_v4f32__vgpr_rsrc__sgpr_vindex__sgpr_voffset__vgpr_soffset
129  ; CHECK: bb.1 (%ir-block.0):
130  ; CHECK-NEXT:   successors: %bb.2(0x80000000)
131  ; CHECK-NEXT:   liveins: $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4
132  ; CHECK-NEXT: {{  $}}
133  ; CHECK-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
134  ; CHECK-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
135  ; CHECK-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr2
136  ; CHECK-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr3
137  ; CHECK-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3
138  ; CHECK-NEXT:   [[COPY4:%[0-9]+]]:sreg_32 = COPY $sgpr2
139  ; CHECK-NEXT:   [[COPY5:%[0-9]+]]:sreg_32 = COPY $sgpr3
140  ; CHECK-NEXT:   [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr4
141  ; CHECK-NEXT:   [[COPY7:%[0-9]+]]:vgpr_32 = COPY [[COPY4]]
142  ; CHECK-NEXT:   [[COPY8:%[0-9]+]]:vgpr_32 = COPY [[COPY5]]
143  ; CHECK-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
144  ; CHECK-NEXT: {{  $}}
145  ; CHECK-NEXT: bb.2:
146  ; CHECK-NEXT:   successors: %bb.3(0x80000000)
147  ; CHECK-NEXT: {{  $}}
148  ; CHECK-NEXT:   [[COPY9:%[0-9]+]]:vgpr_32 = COPY [[REG_SEQUENCE]].sub0
149  ; CHECK-NEXT:   [[COPY10:%[0-9]+]]:vgpr_32 = COPY [[REG_SEQUENCE]].sub1
150  ; CHECK-NEXT:   [[COPY11:%[0-9]+]]:vgpr_32 = COPY [[REG_SEQUENCE]].sub2
151  ; CHECK-NEXT:   [[COPY12:%[0-9]+]]:vgpr_32 = COPY [[REG_SEQUENCE]].sub3
152  ; CHECK-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sreg_32 = V_READFIRSTLANE_B32 [[COPY9]], implicit $exec
153  ; CHECK-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sreg_32 = V_READFIRSTLANE_B32 [[COPY10]], implicit $exec
154  ; CHECK-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sreg_32 = V_READFIRSTLANE_B32 [[COPY11]], implicit $exec
155  ; CHECK-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sreg_32 = V_READFIRSTLANE_B32 [[COPY12]], implicit $exec
156  ; CHECK-NEXT:   [[REG_SEQUENCE1:%[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
157  ; CHECK-NEXT:   [[COPY13:%[0-9]+]]:vreg_64 = COPY [[REG_SEQUENCE]].sub0_sub1
158  ; CHECK-NEXT:   [[COPY14:%[0-9]+]]:vreg_64 = COPY [[REG_SEQUENCE]].sub2_sub3
159  ; CHECK-NEXT:   [[COPY15:%[0-9]+]]:sreg_64 = COPY [[REG_SEQUENCE1]].sub0_sub1
160  ; CHECK-NEXT:   [[COPY16:%[0-9]+]]:sreg_64 = COPY [[REG_SEQUENCE1]].sub2_sub3
161  ; CHECK-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[COPY15]], [[COPY13]], implicit $exec
162  ; CHECK-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[COPY16]], [[COPY14]], implicit $exec
163  ; CHECK-NEXT:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def dead $scc
164  ; CHECK-NEXT:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sreg_32 = V_READFIRSTLANE_B32 [[COPY6]], implicit $exec
165  ; CHECK-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]], [[COPY6]], implicit $exec
166  ; CHECK-NEXT:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[S_AND_B64_]], [[V_CMP_EQ_U32_e64_]], implicit-def dead $scc
167  ; 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
168  ; CHECK-NEXT: {{  $}}
169  ; CHECK-NEXT: bb.3:
170  ; CHECK-NEXT:   successors: %bb.4(0x40000000), %bb.2(0x40000000)
171  ; CHECK-NEXT: {{  $}}
172  ; CHECK-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY7]], %subreg.sub0, [[COPY8]], %subreg.sub1
173  ; CHECK-NEXT:   [[TBUFFER_LOAD_FORMAT_XYZW_BOTHEN:%[0-9]+]]:vreg_128 = TBUFFER_LOAD_FORMAT_XYZW_BOTHEN [[REG_SEQUENCE2]], [[REG_SEQUENCE1]], [[V_READFIRSTLANE_B32_4]], 0, 78, 0, 0, 0, implicit $exec :: (dereferenceable load (<4 x s32>), align 1, addrspace 4)
174  ; CHECK-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
175  ; CHECK-NEXT:   SI_WATERFALL_LOOP %bb.2, implicit $exec
176  ; CHECK-NEXT: {{  $}}
177  ; CHECK-NEXT: bb.4:
178  ; CHECK-NEXT:   successors: %bb.5(0x80000000)
179  ; CHECK-NEXT: {{  $}}
180  ; CHECK-NEXT:   $exec = S_MOV_B64_term [[S_MOV_B64_]]
181  ; CHECK-NEXT: {{  $}}
182  ; CHECK-NEXT: bb.5:
183  ; CHECK-NEXT:   [[COPY17:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XYZW_BOTHEN]].sub0
184  ; CHECK-NEXT:   [[COPY18:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XYZW_BOTHEN]].sub1
185  ; CHECK-NEXT:   [[COPY19:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XYZW_BOTHEN]].sub2
186  ; CHECK-NEXT:   [[COPY20:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XYZW_BOTHEN]].sub3
187  ; CHECK-NEXT:   $vgpr0 = COPY [[COPY17]]
188  ; CHECK-NEXT:   $vgpr1 = COPY [[COPY18]]
189  ; CHECK-NEXT:   $vgpr2 = COPY [[COPY19]]
190  ; CHECK-NEXT:   $vgpr3 = COPY [[COPY20]]
191  ; CHECK-NEXT:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
192  %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)
193  ret <4 x float> %val
194}
195
196define 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) {
197  ; CHECK-LABEL: name: struct_tbuffer_load_f32__sgpr_rsrc__vgpr_vindex__vgpr_voffset__sgpr_soffset_voffset_add4095
198  ; CHECK: bb.1 (%ir-block.0):
199  ; CHECK-NEXT:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $vgpr0, $vgpr1
200  ; CHECK-NEXT: {{  $}}
201  ; CHECK-NEXT:   [[COPY:%[0-9]+]]:sreg_32 = COPY $sgpr2
202  ; CHECK-NEXT:   [[COPY1:%[0-9]+]]:sreg_32 = COPY $sgpr3
203  ; CHECK-NEXT:   [[COPY2:%[0-9]+]]:sreg_32 = COPY $sgpr4
204  ; CHECK-NEXT:   [[COPY3:%[0-9]+]]:sreg_32 = COPY $sgpr5
205  ; CHECK-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3
206  ; CHECK-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0
207  ; CHECK-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr1
208  ; CHECK-NEXT:   [[COPY6:%[0-9]+]]:sreg_32 = COPY $sgpr6
209  ; CHECK-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY4]], %subreg.sub0, [[COPY5]], %subreg.sub1
210  ; CHECK-NEXT:   [[TBUFFER_LOAD_FORMAT_X_BOTHEN:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_BOTHEN [[REG_SEQUENCE1]], [[REG_SEQUENCE]], [[COPY6]], 4095, 78, 0, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4)
211  ; CHECK-NEXT:   $vgpr0 = COPY [[TBUFFER_LOAD_FORMAT_X_BOTHEN]]
212  ; CHECK-NEXT:   SI_RETURN_TO_EPILOG implicit $vgpr0
213  %voffset = add i32 %voffset.base, 4095
214  %val = call float @llvm.amdgcn.struct.tbuffer.load.f32(<4 x i32> %rsrc, i32 %vindex, i32 %voffset, i32 %soffset, i32 78, i32 0)
215  ret float %val
216}
217
218declare float @llvm.amdgcn.struct.tbuffer.load.f32(<4 x i32>, i32, i32, i32, i32 immarg, i32 immarg) #0
219declare <2 x float> @llvm.amdgcn.struct.tbuffer.load.v2f32(<4 x i32>, i32, i32, i32, i32 immarg, i32 immarg) #0
220declare <3 x float> @llvm.amdgcn.struct.tbuffer.load.v3f32(<4 x i32>, i32, i32, i32, i32 immarg, i32 immarg) #0
221declare <4 x float> @llvm.amdgcn.struct.tbuffer.load.v4f32(<4 x i32>, i32, i32, i32, i32 immarg, i32 immarg) #0
222
223attributes #0 = { nounwind readonly }
224