1; NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
2; RUN: llc -global-isel -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1010 -stop-after=instruction-select -verify-machineinstrs -o - %s | FileCheck %s
3; RUN: llc -global-isel -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1100 -stop-after=instruction-select -verify-machineinstrs -o - %s | FileCheck %s
4
5define amdgpu_ps float @raw_tbuffer_load_f32__sgpr_rsrc__vgpr_voffset__sgpr_soffset(<4 x i32> inreg %rsrc, i32 %voffset, i32 inreg %soffset) {
6  ; CHECK-LABEL: name: raw_tbuffer_load_f32__sgpr_rsrc__vgpr_voffset__sgpr_soffset
7  ; CHECK: bb.1 (%ir-block.0):
8  ; CHECK-NEXT:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $vgpr0
9  ; CHECK-NEXT: {{  $}}
10  ; CHECK-NEXT:   [[COPY:%[0-9]+]]:sreg_32 = COPY $sgpr2
11  ; CHECK-NEXT:   [[COPY1:%[0-9]+]]:sreg_32 = COPY $sgpr3
12  ; CHECK-NEXT:   [[COPY2:%[0-9]+]]:sreg_32 = COPY $sgpr4
13  ; CHECK-NEXT:   [[COPY3:%[0-9]+]]:sreg_32 = COPY $sgpr5
14  ; CHECK-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3
15  ; CHECK-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0
16  ; CHECK-NEXT:   [[COPY5:%[0-9]+]]:sreg_32 = COPY $sgpr6
17  ; CHECK-NEXT:   [[TBUFFER_LOAD_FORMAT_X_OFFEN:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFEN [[COPY4]], [[REG_SEQUENCE]], [[COPY5]], 0, 78, 0, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4)
18  ; CHECK-NEXT:   $vgpr0 = COPY [[TBUFFER_LOAD_FORMAT_X_OFFEN]]
19  ; CHECK-NEXT:   SI_RETURN_TO_EPILOG implicit $vgpr0
20  %val = call float @llvm.amdgcn.raw.tbuffer.load.f32(<4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 78, i32 0)
21  ret float %val
22}
23
24define amdgpu_ps <2 x float> @raw_tbuffer_load_v2f32__sgpr_rsrc__vgpr_voffset__sgpr_soffset(<4 x i32> inreg %rsrc, i32 %voffset, i32 inreg %soffset) {
25  ; CHECK-LABEL: name: raw_tbuffer_load_v2f32__sgpr_rsrc__vgpr_voffset__sgpr_soffset
26  ; CHECK: bb.1 (%ir-block.0):
27  ; CHECK-NEXT:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $vgpr0
28  ; CHECK-NEXT: {{  $}}
29  ; CHECK-NEXT:   [[COPY:%[0-9]+]]:sreg_32 = COPY $sgpr2
30  ; CHECK-NEXT:   [[COPY1:%[0-9]+]]:sreg_32 = COPY $sgpr3
31  ; CHECK-NEXT:   [[COPY2:%[0-9]+]]:sreg_32 = COPY $sgpr4
32  ; CHECK-NEXT:   [[COPY3:%[0-9]+]]:sreg_32 = COPY $sgpr5
33  ; CHECK-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3
34  ; CHECK-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0
35  ; CHECK-NEXT:   [[COPY5:%[0-9]+]]:sreg_32 = COPY $sgpr6
36  ; CHECK-NEXT:   [[TBUFFER_LOAD_FORMAT_XY_OFFEN:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_OFFEN [[COPY4]], [[REG_SEQUENCE]], [[COPY5]], 0, 78, 0, 0, 0, implicit $exec :: (dereferenceable load (<2 x s32>), align 1, addrspace 4)
37  ; CHECK-NEXT:   [[COPY6:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XY_OFFEN]].sub0
38  ; CHECK-NEXT:   [[COPY7:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XY_OFFEN]].sub1
39  ; CHECK-NEXT:   $vgpr0 = COPY [[COPY6]]
40  ; CHECK-NEXT:   $vgpr1 = COPY [[COPY7]]
41  ; CHECK-NEXT:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1
42  %val = call <2 x float> @llvm.amdgcn.raw.tbuffer.load.v2f32(<4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 78, i32 0)
43  ret <2 x float> %val
44}
45
46define amdgpu_ps <3 x float> @raw_tbuffer_load_v3f32__sgpr_rsrc__vgpr_voffset__sgpr_soffset(<4 x i32> inreg %rsrc, i32 %voffset, i32 inreg %soffset) {
47  ; CHECK-LABEL: name: raw_tbuffer_load_v3f32__sgpr_rsrc__vgpr_voffset__sgpr_soffset
48  ; CHECK: bb.1 (%ir-block.0):
49  ; CHECK-NEXT:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $vgpr0
50  ; CHECK-NEXT: {{  $}}
51  ; CHECK-NEXT:   [[COPY:%[0-9]+]]:sreg_32 = COPY $sgpr2
52  ; CHECK-NEXT:   [[COPY1:%[0-9]+]]:sreg_32 = COPY $sgpr3
53  ; CHECK-NEXT:   [[COPY2:%[0-9]+]]:sreg_32 = COPY $sgpr4
54  ; CHECK-NEXT:   [[COPY3:%[0-9]+]]:sreg_32 = COPY $sgpr5
55  ; CHECK-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3
56  ; CHECK-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0
57  ; CHECK-NEXT:   [[COPY5:%[0-9]+]]:sreg_32 = COPY $sgpr6
58  ; CHECK-NEXT:   [[TBUFFER_LOAD_FORMAT_XYZ_OFFEN:%[0-9]+]]:vreg_96 = TBUFFER_LOAD_FORMAT_XYZ_OFFEN [[COPY4]], [[REG_SEQUENCE]], [[COPY5]], 0, 78, 0, 0, 0, implicit $exec :: (dereferenceable load (<3 x s32>), align 1, addrspace 4)
59  ; CHECK-NEXT:   [[COPY6:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XYZ_OFFEN]].sub0
60  ; CHECK-NEXT:   [[COPY7:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XYZ_OFFEN]].sub1
61  ; CHECK-NEXT:   [[COPY8:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XYZ_OFFEN]].sub2
62  ; CHECK-NEXT:   $vgpr0 = COPY [[COPY6]]
63  ; CHECK-NEXT:   $vgpr1 = COPY [[COPY7]]
64  ; CHECK-NEXT:   $vgpr2 = COPY [[COPY8]]
65  ; CHECK-NEXT:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2
66  %val = call <3 x float> @llvm.amdgcn.raw.tbuffer.load.v3f32(<4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 78, i32 0)
67  ret <3 x float> %val
68}
69
70define amdgpu_ps <4 x float> @raw_tbuffer_load_v4f32__sgpr_rsrc__vgpr_voffset__sgpr_soffset(<4 x i32> inreg %rsrc, i32 %voffset, i32 inreg %soffset) {
71  ; CHECK-LABEL: name: raw_tbuffer_load_v4f32__sgpr_rsrc__vgpr_voffset__sgpr_soffset
72  ; CHECK: bb.1 (%ir-block.0):
73  ; CHECK-NEXT:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $vgpr0
74  ; CHECK-NEXT: {{  $}}
75  ; CHECK-NEXT:   [[COPY:%[0-9]+]]:sreg_32 = COPY $sgpr2
76  ; CHECK-NEXT:   [[COPY1:%[0-9]+]]:sreg_32 = COPY $sgpr3
77  ; CHECK-NEXT:   [[COPY2:%[0-9]+]]:sreg_32 = COPY $sgpr4
78  ; CHECK-NEXT:   [[COPY3:%[0-9]+]]:sreg_32 = COPY $sgpr5
79  ; CHECK-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3
80  ; CHECK-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0
81  ; CHECK-NEXT:   [[COPY5:%[0-9]+]]:sreg_32 = COPY $sgpr6
82  ; CHECK-NEXT:   [[TBUFFER_LOAD_FORMAT_XYZW_OFFEN:%[0-9]+]]:vreg_128 = TBUFFER_LOAD_FORMAT_XYZW_OFFEN [[COPY4]], [[REG_SEQUENCE]], [[COPY5]], 0, 78, 0, 0, 0, implicit $exec :: (dereferenceable load (<4 x s32>), align 1, addrspace 4)
83  ; CHECK-NEXT:   [[COPY6:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XYZW_OFFEN]].sub0
84  ; CHECK-NEXT:   [[COPY7:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XYZW_OFFEN]].sub1
85  ; CHECK-NEXT:   [[COPY8:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XYZW_OFFEN]].sub2
86  ; CHECK-NEXT:   [[COPY9:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XYZW_OFFEN]].sub3
87  ; CHECK-NEXT:   $vgpr0 = COPY [[COPY6]]
88  ; CHECK-NEXT:   $vgpr1 = COPY [[COPY7]]
89  ; CHECK-NEXT:   $vgpr2 = COPY [[COPY8]]
90  ; CHECK-NEXT:   $vgpr3 = COPY [[COPY9]]
91  ; CHECK-NEXT:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
92  %val = call <4 x float> @llvm.amdgcn.raw.tbuffer.load.v4f32(<4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 78, i32 0)
93  ret <4 x float> %val
94}
95
96define amdgpu_ps float @raw_tbuffer_load_f32__vgpr_rsrc__sgpr_voffset__vgpr_soffset(<4 x i32> %rsrc, i32 inreg %voffset, i32 %soffset) {
97  ; CHECK-LABEL: name: raw_tbuffer_load_f32__vgpr_rsrc__sgpr_voffset__vgpr_soffset
98  ; CHECK: bb.1 (%ir-block.0):
99  ; CHECK-NEXT:   successors: %bb.2(0x80000000)
100  ; CHECK-NEXT:   liveins: $sgpr2, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4
101  ; CHECK-NEXT: {{  $}}
102  ; CHECK-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
103  ; CHECK-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
104  ; CHECK-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr2
105  ; CHECK-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr3
106  ; CHECK-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3
107  ; CHECK-NEXT:   [[COPY4:%[0-9]+]]:sreg_32 = COPY $sgpr2
108  ; CHECK-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr4
109  ; CHECK-NEXT:   [[COPY6:%[0-9]+]]:vgpr_32 = COPY [[COPY4]]
110  ; CHECK-NEXT:   [[S_MOV_B32_:%[0-9]+]]:sreg_32_xm0_xexec = S_MOV_B32 $exec_lo
111  ; CHECK-NEXT: {{  $}}
112  ; CHECK-NEXT: bb.2:
113  ; CHECK-NEXT:   successors: %bb.3(0x80000000)
114  ; CHECK-NEXT: {{  $}}
115  ; CHECK-NEXT:   [[COPY7:%[0-9]+]]:vgpr_32 = COPY [[REG_SEQUENCE]].sub0
116  ; CHECK-NEXT:   [[COPY8:%[0-9]+]]:vgpr_32 = COPY [[REG_SEQUENCE]].sub1
117  ; CHECK-NEXT:   [[COPY9:%[0-9]+]]:vgpr_32 = COPY [[REG_SEQUENCE]].sub2
118  ; CHECK-NEXT:   [[COPY10:%[0-9]+]]:vgpr_32 = COPY [[REG_SEQUENCE]].sub3
119  ; CHECK-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sreg_32 = V_READFIRSTLANE_B32 [[COPY7]], implicit $exec
120  ; CHECK-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sreg_32 = V_READFIRSTLANE_B32 [[COPY8]], implicit $exec
121  ; CHECK-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sreg_32 = V_READFIRSTLANE_B32 [[COPY9]], implicit $exec
122  ; CHECK-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sreg_32 = V_READFIRSTLANE_B32 [[COPY10]], implicit $exec
123  ; 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
124  ; CHECK-NEXT:   [[COPY11:%[0-9]+]]:vreg_64 = COPY [[REG_SEQUENCE]].sub0_sub1
125  ; CHECK-NEXT:   [[COPY12:%[0-9]+]]:vreg_64 = COPY [[REG_SEQUENCE]].sub2_sub3
126  ; CHECK-NEXT:   [[COPY13:%[0-9]+]]:sreg_64 = COPY [[REG_SEQUENCE1]].sub0_sub1
127  ; CHECK-NEXT:   [[COPY14:%[0-9]+]]:sreg_64 = COPY [[REG_SEQUENCE1]].sub2_sub3
128  ; CHECK-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_32_xm0_xexec = V_CMP_EQ_U64_e64 [[COPY13]], [[COPY11]], implicit $exec
129  ; CHECK-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_32_xm0_xexec = V_CMP_EQ_U64_e64 [[COPY14]], [[COPY12]], implicit $exec
130  ; CHECK-NEXT:   [[S_AND_B32_:%[0-9]+]]:sreg_32_xm0_xexec = S_AND_B32 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def dead $scc
131  ; CHECK-NEXT:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sreg_32 = V_READFIRSTLANE_B32 [[COPY5]], implicit $exec
132  ; CHECK-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_32_xm0_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]], [[COPY5]], implicit $exec
133  ; CHECK-NEXT:   [[S_AND_B32_1:%[0-9]+]]:sreg_32_xm0_xexec = S_AND_B32 [[S_AND_B32_]], [[V_CMP_EQ_U32_e64_]], implicit-def dead $scc
134  ; CHECK-NEXT:   [[S_AND_SAVEEXEC_B32_:%[0-9]+]]:sreg_32_xm0_xexec = S_AND_SAVEEXEC_B32 killed [[S_AND_B32_1]], implicit-def $exec, implicit-def $scc, implicit $exec
135  ; CHECK-NEXT: {{  $}}
136  ; CHECK-NEXT: bb.3:
137  ; CHECK-NEXT:   successors: %bb.4(0x40000000), %bb.2(0x40000000)
138  ; CHECK-NEXT: {{  $}}
139  ; CHECK-NEXT:   [[TBUFFER_LOAD_FORMAT_X_OFFEN:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFEN [[COPY6]], [[REG_SEQUENCE1]], [[V_READFIRSTLANE_B32_4]], 0, 78, 0, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4)
140  ; CHECK-NEXT:   $exec_lo = S_XOR_B32_term $exec_lo, [[S_AND_SAVEEXEC_B32_]], implicit-def $scc
141  ; CHECK-NEXT:   SI_WATERFALL_LOOP %bb.2, implicit $exec
142  ; CHECK-NEXT: {{  $}}
143  ; CHECK-NEXT: bb.4:
144  ; CHECK-NEXT:   successors: %bb.5(0x80000000)
145  ; CHECK-NEXT: {{  $}}
146  ; CHECK-NEXT:   $exec_lo = S_MOV_B32_term [[S_MOV_B32_]]
147  ; CHECK-NEXT: {{  $}}
148  ; CHECK-NEXT: bb.5:
149  ; CHECK-NEXT:   $vgpr0 = COPY [[TBUFFER_LOAD_FORMAT_X_OFFEN]]
150  ; CHECK-NEXT:   SI_RETURN_TO_EPILOG implicit $vgpr0
151  %val = call float @llvm.amdgcn.raw.tbuffer.load.f32(<4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 78, i32 0)
152  ret float %val
153}
154
155define amdgpu_ps float @raw_tbuffer_load_f32__sgpr_rsrc__vgpr_voffset__sgpr_soffset_glc(<4 x i32> inreg %rsrc, i32 %voffset, i32 inreg %soffset) {
156  ; CHECK-LABEL: name: raw_tbuffer_load_f32__sgpr_rsrc__vgpr_voffset__sgpr_soffset_glc
157  ; CHECK: bb.1 (%ir-block.0):
158  ; CHECK-NEXT:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $vgpr0
159  ; CHECK-NEXT: {{  $}}
160  ; CHECK-NEXT:   [[COPY:%[0-9]+]]:sreg_32 = COPY $sgpr2
161  ; CHECK-NEXT:   [[COPY1:%[0-9]+]]:sreg_32 = COPY $sgpr3
162  ; CHECK-NEXT:   [[COPY2:%[0-9]+]]:sreg_32 = COPY $sgpr4
163  ; CHECK-NEXT:   [[COPY3:%[0-9]+]]:sreg_32 = COPY $sgpr5
164  ; CHECK-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3
165  ; CHECK-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0
166  ; CHECK-NEXT:   [[COPY5:%[0-9]+]]:sreg_32 = COPY $sgpr6
167  ; CHECK-NEXT:   [[TBUFFER_LOAD_FORMAT_X_OFFEN:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFEN [[COPY4]], [[REG_SEQUENCE]], [[COPY5]], 0, 78, 1, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4)
168  ; CHECK-NEXT:   $vgpr0 = COPY [[TBUFFER_LOAD_FORMAT_X_OFFEN]]
169  ; CHECK-NEXT:   SI_RETURN_TO_EPILOG implicit $vgpr0
170  %val = call float @llvm.amdgcn.raw.tbuffer.load.f32(<4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 78, i32 1)
171  ret float %val
172}
173
174define amdgpu_ps float @raw_tbuffer_load_f32__sgpr_rsrc__vgpr_voffset__sgpr_soffset_slc(<4 x i32> inreg %rsrc, i32 %voffset, i32 inreg %soffset) {
175  ; CHECK-LABEL: name: raw_tbuffer_load_f32__sgpr_rsrc__vgpr_voffset__sgpr_soffset_slc
176  ; CHECK: bb.1 (%ir-block.0):
177  ; CHECK-NEXT:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $vgpr0
178  ; CHECK-NEXT: {{  $}}
179  ; CHECK-NEXT:   [[COPY:%[0-9]+]]:sreg_32 = COPY $sgpr2
180  ; CHECK-NEXT:   [[COPY1:%[0-9]+]]:sreg_32 = COPY $sgpr3
181  ; CHECK-NEXT:   [[COPY2:%[0-9]+]]:sreg_32 = COPY $sgpr4
182  ; CHECK-NEXT:   [[COPY3:%[0-9]+]]:sreg_32 = COPY $sgpr5
183  ; CHECK-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3
184  ; CHECK-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0
185  ; CHECK-NEXT:   [[COPY5:%[0-9]+]]:sreg_32 = COPY $sgpr6
186  ; CHECK-NEXT:   [[TBUFFER_LOAD_FORMAT_X_OFFEN:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFEN [[COPY4]], [[REG_SEQUENCE]], [[COPY5]], 0, 78, 2, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4)
187  ; CHECK-NEXT:   $vgpr0 = COPY [[TBUFFER_LOAD_FORMAT_X_OFFEN]]
188  ; CHECK-NEXT:   SI_RETURN_TO_EPILOG implicit $vgpr0
189  %val = call float @llvm.amdgcn.raw.tbuffer.load.f32(<4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 78, i32 2)
190  ret float %val
191}
192
193define amdgpu_ps float @raw_tbuffer_load_f32__sgpr_rsrc__vgpr_voffset__sgpr_soffset_slc_glc(<4 x i32> inreg %rsrc, i32 %voffset, i32 inreg %soffset) {
194  ; CHECK-LABEL: name: raw_tbuffer_load_f32__sgpr_rsrc__vgpr_voffset__sgpr_soffset_slc_glc
195  ; CHECK: bb.1 (%ir-block.0):
196  ; CHECK-NEXT:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $vgpr0
197  ; CHECK-NEXT: {{  $}}
198  ; CHECK-NEXT:   [[COPY:%[0-9]+]]:sreg_32 = COPY $sgpr2
199  ; CHECK-NEXT:   [[COPY1:%[0-9]+]]:sreg_32 = COPY $sgpr3
200  ; CHECK-NEXT:   [[COPY2:%[0-9]+]]:sreg_32 = COPY $sgpr4
201  ; CHECK-NEXT:   [[COPY3:%[0-9]+]]:sreg_32 = COPY $sgpr5
202  ; CHECK-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3
203  ; CHECK-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0
204  ; CHECK-NEXT:   [[COPY5:%[0-9]+]]:sreg_32 = COPY $sgpr6
205  ; CHECK-NEXT:   [[TBUFFER_LOAD_FORMAT_X_OFFEN:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFEN [[COPY4]], [[REG_SEQUENCE]], [[COPY5]], 0, 78, 3, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4)
206  ; CHECK-NEXT:   $vgpr0 = COPY [[TBUFFER_LOAD_FORMAT_X_OFFEN]]
207  ; CHECK-NEXT:   SI_RETURN_TO_EPILOG implicit $vgpr0
208  %val = call float @llvm.amdgcn.raw.tbuffer.load.f32(<4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 78, i32 3)
209  ret float %val
210}
211
212define amdgpu_ps float @raw_tbuffer_load_f32__sgpr_rsrc__vgpr_voffset__sgpr_soffset_dlc(<4 x i32> inreg %rsrc, i32 %voffset, i32 inreg %soffset) {
213  ; CHECK-LABEL: name: raw_tbuffer_load_f32__sgpr_rsrc__vgpr_voffset__sgpr_soffset_dlc
214  ; CHECK: bb.1 (%ir-block.0):
215  ; CHECK-NEXT:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $vgpr0
216  ; CHECK-NEXT: {{  $}}
217  ; CHECK-NEXT:   [[COPY:%[0-9]+]]:sreg_32 = COPY $sgpr2
218  ; CHECK-NEXT:   [[COPY1:%[0-9]+]]:sreg_32 = COPY $sgpr3
219  ; CHECK-NEXT:   [[COPY2:%[0-9]+]]:sreg_32 = COPY $sgpr4
220  ; CHECK-NEXT:   [[COPY3:%[0-9]+]]:sreg_32 = COPY $sgpr5
221  ; CHECK-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3
222  ; CHECK-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0
223  ; CHECK-NEXT:   [[COPY5:%[0-9]+]]:sreg_32 = COPY $sgpr6
224  ; CHECK-NEXT:   [[TBUFFER_LOAD_FORMAT_X_OFFEN:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFEN [[COPY4]], [[REG_SEQUENCE]], [[COPY5]], 0, 78, 4, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4)
225  ; CHECK-NEXT:   $vgpr0 = COPY [[TBUFFER_LOAD_FORMAT_X_OFFEN]]
226  ; CHECK-NEXT:   SI_RETURN_TO_EPILOG implicit $vgpr0
227  %val = call float @llvm.amdgcn.raw.tbuffer.load.f32(<4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 78, i32 4)
228  ret float %val
229}
230
231declare float @llvm.amdgcn.raw.tbuffer.load.f32(<4 x i32>, i32, i32, i32 immarg, i32 immarg) #0
232declare <2 x float> @llvm.amdgcn.raw.tbuffer.load.v2f32(<4 x i32>, i32, i32, i32 immarg, i32 immarg) #0
233declare <3 x float> @llvm.amdgcn.raw.tbuffer.load.v3f32(<4 x i32>, i32, i32, i32 immarg, i32 immarg) #0
234declare <4 x float> @llvm.amdgcn.raw.tbuffer.load.v4f32(<4 x i32>, i32, i32, i32 immarg, i32 immarg) #0
235
236attributes #0 = { nounwind readonly }
237