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
5 define amdgpu_ps float @raw_tbuffer_load_f32__sgpr_rsrc__vgpr_voffset__sgpr_soffset(ptr addrspace(8) 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
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: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0
15 ; CHECK-NEXT: [[COPY5:%[0-9]+]]:sreg_32 = COPY $sgpr6
16 ; CHECK-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3
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, implicit $exec :: (dereferenceable load (s32) from %ir.rsrc, align 1, addrspace 8)
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.ptr.tbuffer.load.f32(ptr addrspace(8) %rsrc, i32 %voffset, i32 %soffset, i32 78, i32 0)
24 define amdgpu_ps <2 x float> @raw_tbuffer_load_v2f32__sgpr_rsrc__vgpr_voffset__sgpr_soffset(ptr addrspace(8) 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
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: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0
34 ; CHECK-NEXT: [[COPY5:%[0-9]+]]:sreg_32 = COPY $sgpr6
35 ; CHECK-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3
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, implicit $exec :: (dereferenceable load (<2 x s32>) from %ir.rsrc, align 1, addrspace 8)
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.ptr.tbuffer.load.v2f32(ptr addrspace(8) %rsrc, i32 %voffset, i32 %soffset, i32 78, i32 0)
46 define amdgpu_ps <3 x float> @raw_tbuffer_load_v3f32__sgpr_rsrc__vgpr_voffset__sgpr_soffset(ptr addrspace(8) 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
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: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0
56 ; CHECK-NEXT: [[COPY5:%[0-9]+]]:sreg_32 = COPY $sgpr6
57 ; CHECK-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3
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, implicit $exec :: (dereferenceable load (<3 x s32>) from %ir.rsrc, align 1, addrspace 8)
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.ptr.tbuffer.load.v3f32(ptr addrspace(8) %rsrc, i32 %voffset, i32 %soffset, i32 78, i32 0)
70 define amdgpu_ps <4 x float> @raw_tbuffer_load_v4f32__sgpr_rsrc__vgpr_voffset__sgpr_soffset(ptr addrspace(8) 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
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: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0
80 ; CHECK-NEXT: [[COPY5:%[0-9]+]]:sreg_32 = COPY $sgpr6
81 ; CHECK-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3
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, implicit $exec :: (dereferenceable load (<4 x s32>) from %ir.rsrc, align 1, addrspace 8)
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.ptr.tbuffer.load.v4f32(ptr addrspace(8) %rsrc, i32 %voffset, i32 %soffset, i32 78, i32 0)
96 define amdgpu_ps float @raw_tbuffer_load_f32__vgpr_rsrc__sgpr_voffset__vgpr_soffset(ptr addrspace(8) %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
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: [[COPY4:%[0-9]+]]:sreg_32 = COPY $sgpr2
107 ; CHECK-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr4
108 ; CHECK-NEXT: [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3
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
113 ; CHECK-NEXT: successors: %bb.3(0x80000000)
115 ; CHECK-NEXT: [[V_READFIRSTLANE_B32_:%[0-9]+]]:sreg_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
116 ; CHECK-NEXT: [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sreg_32 = V_READFIRSTLANE_B32 [[COPY1]], implicit $exec
117 ; CHECK-NEXT: [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sreg_32 = V_READFIRSTLANE_B32 [[COPY2]], implicit $exec
118 ; CHECK-NEXT: [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sreg_32 = V_READFIRSTLANE_B32 [[COPY3]], implicit $exec
119 ; 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
120 ; CHECK-NEXT: [[COPY7:%[0-9]+]]:vreg_64 = COPY [[REG_SEQUENCE]].sub0_sub1
121 ; CHECK-NEXT: [[COPY8:%[0-9]+]]:vreg_64 = COPY [[REG_SEQUENCE]].sub2_sub3
122 ; CHECK-NEXT: [[COPY9:%[0-9]+]]:sreg_64 = COPY [[REG_SEQUENCE1]].sub0_sub1
123 ; CHECK-NEXT: [[COPY10:%[0-9]+]]:sreg_64 = COPY [[REG_SEQUENCE1]].sub2_sub3
124 ; CHECK-NEXT: [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_32_xm0_xexec = V_CMP_EQ_U64_e64 [[COPY9]], [[COPY7]], implicit $exec
125 ; CHECK-NEXT: [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_32_xm0_xexec = V_CMP_EQ_U64_e64 [[COPY10]], [[COPY8]], implicit $exec
126 ; 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
127 ; CHECK-NEXT: [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sreg_32 = V_READFIRSTLANE_B32 [[COPY5]], implicit $exec
128 ; 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
129 ; 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
130 ; 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
133 ; CHECK-NEXT: successors: %bb.4(0x40000000), %bb.2(0x40000000)
135 ; 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, implicit $exec :: (dereferenceable load (s32) from %ir.rsrc, align 1, addrspace 8)
136 ; CHECK-NEXT: $exec_lo = S_XOR_B32_term $exec_lo, [[S_AND_SAVEEXEC_B32_]], implicit-def $scc
137 ; CHECK-NEXT: SI_WATERFALL_LOOP %bb.2, implicit $exec
140 ; CHECK-NEXT: successors: %bb.5(0x80000000)
142 ; CHECK-NEXT: $exec_lo = S_MOV_B32_term [[S_MOV_B32_]]
145 ; CHECK-NEXT: $vgpr0 = COPY [[TBUFFER_LOAD_FORMAT_X_OFFEN]]
146 ; CHECK-NEXT: SI_RETURN_TO_EPILOG implicit $vgpr0
147 %val = call float @llvm.amdgcn.raw.ptr.tbuffer.load.f32(ptr addrspace(8) %rsrc, i32 %voffset, i32 %soffset, i32 78, i32 0)
151 define amdgpu_ps float @raw_tbuffer_load_f32__sgpr_rsrc__vgpr_voffset__sgpr_soffset_glc(ptr addrspace(8) inreg %rsrc, i32 %voffset, i32 inreg %soffset) {
152 ; CHECK-LABEL: name: raw_tbuffer_load_f32__sgpr_rsrc__vgpr_voffset__sgpr_soffset_glc
153 ; CHECK: bb.1 (%ir-block.0):
154 ; CHECK-NEXT: liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $vgpr0
156 ; CHECK-NEXT: [[COPY:%[0-9]+]]:sreg_32 = COPY $sgpr2
157 ; CHECK-NEXT: [[COPY1:%[0-9]+]]:sreg_32 = COPY $sgpr3
158 ; CHECK-NEXT: [[COPY2:%[0-9]+]]:sreg_32 = COPY $sgpr4
159 ; CHECK-NEXT: [[COPY3:%[0-9]+]]:sreg_32 = COPY $sgpr5
160 ; CHECK-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0
161 ; CHECK-NEXT: [[COPY5:%[0-9]+]]:sreg_32 = COPY $sgpr6
162 ; CHECK-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3
163 ; CHECK-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFEN:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFEN [[COPY4]], [[REG_SEQUENCE]], [[COPY5]], 0, 78, 1, 0, implicit $exec :: (dereferenceable load (s32) from %ir.rsrc, align 1, addrspace 8)
164 ; CHECK-NEXT: $vgpr0 = COPY [[TBUFFER_LOAD_FORMAT_X_OFFEN]]
165 ; CHECK-NEXT: SI_RETURN_TO_EPILOG implicit $vgpr0
166 %val = call float @llvm.amdgcn.raw.ptr.tbuffer.load.f32(ptr addrspace(8) %rsrc, i32 %voffset, i32 %soffset, i32 78, i32 1)
170 define amdgpu_ps float @raw_tbuffer_load_f32__sgpr_rsrc__vgpr_voffset__sgpr_soffset_slc(ptr addrspace(8) inreg %rsrc, i32 %voffset, i32 inreg %soffset) {
171 ; CHECK-LABEL: name: raw_tbuffer_load_f32__sgpr_rsrc__vgpr_voffset__sgpr_soffset_slc
172 ; CHECK: bb.1 (%ir-block.0):
173 ; CHECK-NEXT: liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $vgpr0
175 ; CHECK-NEXT: [[COPY:%[0-9]+]]:sreg_32 = COPY $sgpr2
176 ; CHECK-NEXT: [[COPY1:%[0-9]+]]:sreg_32 = COPY $sgpr3
177 ; CHECK-NEXT: [[COPY2:%[0-9]+]]:sreg_32 = COPY $sgpr4
178 ; CHECK-NEXT: [[COPY3:%[0-9]+]]:sreg_32 = COPY $sgpr5
179 ; CHECK-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0
180 ; CHECK-NEXT: [[COPY5:%[0-9]+]]:sreg_32 = COPY $sgpr6
181 ; CHECK-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3
182 ; CHECK-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFEN:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFEN [[COPY4]], [[REG_SEQUENCE]], [[COPY5]], 0, 78, 2, 0, implicit $exec :: (dereferenceable load (s32) from %ir.rsrc, align 1, addrspace 8)
183 ; CHECK-NEXT: $vgpr0 = COPY [[TBUFFER_LOAD_FORMAT_X_OFFEN]]
184 ; CHECK-NEXT: SI_RETURN_TO_EPILOG implicit $vgpr0
185 %val = call float @llvm.amdgcn.raw.ptr.tbuffer.load.f32(ptr addrspace(8) %rsrc, i32 %voffset, i32 %soffset, i32 78, i32 2)
189 define amdgpu_ps float @raw_tbuffer_load_f32__sgpr_rsrc__vgpr_voffset__sgpr_soffset_slc_glc(ptr addrspace(8) inreg %rsrc, i32 %voffset, i32 inreg %soffset) {
190 ; CHECK-LABEL: name: raw_tbuffer_load_f32__sgpr_rsrc__vgpr_voffset__sgpr_soffset_slc_glc
191 ; CHECK: bb.1 (%ir-block.0):
192 ; CHECK-NEXT: liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $vgpr0
194 ; CHECK-NEXT: [[COPY:%[0-9]+]]:sreg_32 = COPY $sgpr2
195 ; CHECK-NEXT: [[COPY1:%[0-9]+]]:sreg_32 = COPY $sgpr3
196 ; CHECK-NEXT: [[COPY2:%[0-9]+]]:sreg_32 = COPY $sgpr4
197 ; CHECK-NEXT: [[COPY3:%[0-9]+]]:sreg_32 = COPY $sgpr5
198 ; CHECK-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0
199 ; CHECK-NEXT: [[COPY5:%[0-9]+]]:sreg_32 = COPY $sgpr6
200 ; CHECK-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3
201 ; CHECK-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFEN:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFEN [[COPY4]], [[REG_SEQUENCE]], [[COPY5]], 0, 78, 3, 0, implicit $exec :: (dereferenceable load (s32) from %ir.rsrc, align 1, addrspace 8)
202 ; CHECK-NEXT: $vgpr0 = COPY [[TBUFFER_LOAD_FORMAT_X_OFFEN]]
203 ; CHECK-NEXT: SI_RETURN_TO_EPILOG implicit $vgpr0
204 %val = call float @llvm.amdgcn.raw.ptr.tbuffer.load.f32(ptr addrspace(8) %rsrc, i32 %voffset, i32 %soffset, i32 78, i32 3)
208 define amdgpu_ps float @raw_tbuffer_load_f32__sgpr_rsrc__vgpr_voffset__sgpr_soffset_dlc(ptr addrspace(8) inreg %rsrc, i32 %voffset, i32 inreg %soffset) {
209 ; CHECK-LABEL: name: raw_tbuffer_load_f32__sgpr_rsrc__vgpr_voffset__sgpr_soffset_dlc
210 ; CHECK: bb.1 (%ir-block.0):
211 ; CHECK-NEXT: liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $vgpr0
213 ; CHECK-NEXT: [[COPY:%[0-9]+]]:sreg_32 = COPY $sgpr2
214 ; CHECK-NEXT: [[COPY1:%[0-9]+]]:sreg_32 = COPY $sgpr3
215 ; CHECK-NEXT: [[COPY2:%[0-9]+]]:sreg_32 = COPY $sgpr4
216 ; CHECK-NEXT: [[COPY3:%[0-9]+]]:sreg_32 = COPY $sgpr5
217 ; CHECK-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0
218 ; CHECK-NEXT: [[COPY5:%[0-9]+]]:sreg_32 = COPY $sgpr6
219 ; CHECK-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3
220 ; CHECK-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFEN:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFEN [[COPY4]], [[REG_SEQUENCE]], [[COPY5]], 0, 78, 4, 0, implicit $exec :: (dereferenceable load (s32) from %ir.rsrc, align 1, addrspace 8)
221 ; CHECK-NEXT: $vgpr0 = COPY [[TBUFFER_LOAD_FORMAT_X_OFFEN]]
222 ; CHECK-NEXT: SI_RETURN_TO_EPILOG implicit $vgpr0
223 %val = call float @llvm.amdgcn.raw.ptr.tbuffer.load.f32(ptr addrspace(8) %rsrc, i32 %voffset, i32 %soffset, i32 78, i32 4)
227 declare float @llvm.amdgcn.raw.ptr.tbuffer.load.f32(ptr addrspace(8), i32, i32, i32 immarg, i32 immarg) #0
228 declare <2 x float> @llvm.amdgcn.raw.ptr.tbuffer.load.v2f32(ptr addrspace(8), i32, i32, i32 immarg, i32 immarg) #0
229 declare <3 x float> @llvm.amdgcn.raw.ptr.tbuffer.load.v3f32(ptr addrspace(8), i32, i32, i32 immarg, i32 immarg) #0
230 declare <4 x float> @llvm.amdgcn.raw.ptr.tbuffer.load.v4f32(ptr addrspace(8), i32, i32, i32 immarg, i32 immarg) #0
232 attributes #0 = { nounwind readonly }