1; NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
2; RUN: llc -global-isel -mtriple=amdgcn-mesa-mesa3d -mcpu=fiji -stop-after=instruction-select -verify-machineinstrs -o - %s | FileCheck -check-prefix=UNPACKED %s
3; RUN: llc -global-isel -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx810 -stop-after=instruction-select -verify-machineinstrs -o - %s | FileCheck -check-prefix=PACKED %s
4
5define amdgpu_ps half @struct_buffer_load_format_f16__sgpr_rsrc__vgpr_vindex__vgpr_voffset__sgpr_soffset(<4 x i32> inreg %rsrc, i32 %vindex, i32 %voffset, i32 inreg %soffset) {
6  ; UNPACKED-LABEL: name: struct_buffer_load_format_f16__sgpr_rsrc__vgpr_vindex__vgpr_voffset__sgpr_soffset
7  ; UNPACKED: bb.1 (%ir-block.0):
8  ; UNPACKED-NEXT:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $vgpr0, $vgpr1
9  ; UNPACKED-NEXT: {{  $}}
10  ; UNPACKED-NEXT:   [[COPY:%[0-9]+]]:sreg_32 = COPY $sgpr2
11  ; UNPACKED-NEXT:   [[COPY1:%[0-9]+]]:sreg_32 = COPY $sgpr3
12  ; UNPACKED-NEXT:   [[COPY2:%[0-9]+]]:sreg_32 = COPY $sgpr4
13  ; UNPACKED-NEXT:   [[COPY3:%[0-9]+]]:sreg_32 = COPY $sgpr5
14  ; UNPACKED-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3
15  ; UNPACKED-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0
16  ; UNPACKED-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr1
17  ; UNPACKED-NEXT:   [[COPY6:%[0-9]+]]:sreg_32 = COPY $sgpr6
18  ; UNPACKED-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY4]], %subreg.sub0, [[COPY5]], %subreg.sub1
19  ; UNPACKED-NEXT:   [[BUFFER_LOAD_FORMAT_D16_X_gfx80_BOTHEN:%[0-9]+]]:vgpr_32 = BUFFER_LOAD_FORMAT_D16_X_gfx80_BOTHEN [[REG_SEQUENCE1]], [[REG_SEQUENCE]], [[COPY6]], 0, 0, 0, 0, implicit $exec :: (dereferenceable load (s16), align 1, addrspace 4)
20  ; UNPACKED-NEXT:   $vgpr0 = COPY [[BUFFER_LOAD_FORMAT_D16_X_gfx80_BOTHEN]]
21  ; UNPACKED-NEXT:   SI_RETURN_TO_EPILOG implicit $vgpr0
22  ; PACKED-LABEL: name: struct_buffer_load_format_f16__sgpr_rsrc__vgpr_vindex__vgpr_voffset__sgpr_soffset
23  ; PACKED: bb.1 (%ir-block.0):
24  ; PACKED-NEXT:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $vgpr0, $vgpr1
25  ; PACKED-NEXT: {{  $}}
26  ; PACKED-NEXT:   [[COPY:%[0-9]+]]:sreg_32 = COPY $sgpr2
27  ; PACKED-NEXT:   [[COPY1:%[0-9]+]]:sreg_32 = COPY $sgpr3
28  ; PACKED-NEXT:   [[COPY2:%[0-9]+]]:sreg_32 = COPY $sgpr4
29  ; PACKED-NEXT:   [[COPY3:%[0-9]+]]:sreg_32 = COPY $sgpr5
30  ; PACKED-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3
31  ; PACKED-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0
32  ; PACKED-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr1
33  ; PACKED-NEXT:   [[COPY6:%[0-9]+]]:sreg_32 = COPY $sgpr6
34  ; PACKED-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY4]], %subreg.sub0, [[COPY5]], %subreg.sub1
35  ; PACKED-NEXT:   [[BUFFER_LOAD_FORMAT_D16_X_BOTHEN:%[0-9]+]]:vgpr_32 = BUFFER_LOAD_FORMAT_D16_X_BOTHEN [[REG_SEQUENCE1]], [[REG_SEQUENCE]], [[COPY6]], 0, 0, 0, 0, implicit $exec :: (dereferenceable load (s16), align 1, addrspace 4)
36  ; PACKED-NEXT:   $vgpr0 = COPY [[BUFFER_LOAD_FORMAT_D16_X_BOTHEN]]
37  ; PACKED-NEXT:   SI_RETURN_TO_EPILOG implicit $vgpr0
38  %val = call half @llvm.amdgcn.struct.buffer.load.format.f16(<4 x i32> %rsrc, i32 %vindex, i32 %voffset, i32 %soffset, i32 0)
39  ret half %val
40}
41
42define amdgpu_ps <2 x half> @struct_buffer_load_format_v2f16__sgpr_rsrc__vgpr_vindex__vgpr_voffset__sgpr_soffset(<4 x i32> inreg %rsrc, i32 %vindex, i32 %voffset, i32 inreg %soffset) {
43  ; UNPACKED-LABEL: name: struct_buffer_load_format_v2f16__sgpr_rsrc__vgpr_vindex__vgpr_voffset__sgpr_soffset
44  ; UNPACKED: bb.1 (%ir-block.0):
45  ; UNPACKED-NEXT:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $vgpr0, $vgpr1
46  ; UNPACKED-NEXT: {{  $}}
47  ; UNPACKED-NEXT:   [[COPY:%[0-9]+]]:sreg_32 = COPY $sgpr2
48  ; UNPACKED-NEXT:   [[COPY1:%[0-9]+]]:sreg_32 = COPY $sgpr3
49  ; UNPACKED-NEXT:   [[COPY2:%[0-9]+]]:sreg_32 = COPY $sgpr4
50  ; UNPACKED-NEXT:   [[COPY3:%[0-9]+]]:sreg_32 = COPY $sgpr5
51  ; UNPACKED-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3
52  ; UNPACKED-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0
53  ; UNPACKED-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr1
54  ; UNPACKED-NEXT:   [[COPY6:%[0-9]+]]:sreg_32 = COPY $sgpr6
55  ; UNPACKED-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY4]], %subreg.sub0, [[COPY5]], %subreg.sub1
56  ; UNPACKED-NEXT:   [[BUFFER_LOAD_FORMAT_D16_XY_gfx80_BOTHEN:%[0-9]+]]:vreg_64 = BUFFER_LOAD_FORMAT_D16_XY_gfx80_BOTHEN [[REG_SEQUENCE1]], [[REG_SEQUENCE]], [[COPY6]], 0, 0, 0, 0, implicit $exec :: (dereferenceable load (<2 x s16>), align 1, addrspace 4)
57  ; UNPACKED-NEXT:   [[COPY7:%[0-9]+]]:vgpr_32 = COPY [[BUFFER_LOAD_FORMAT_D16_XY_gfx80_BOTHEN]].sub0
58  ; UNPACKED-NEXT:   [[COPY8:%[0-9]+]]:vgpr_32 = COPY [[BUFFER_LOAD_FORMAT_D16_XY_gfx80_BOTHEN]].sub1
59  ; UNPACKED-NEXT:   [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 65535
60  ; UNPACKED-NEXT:   [[COPY9:%[0-9]+]]:vgpr_32 = COPY [[S_MOV_B32_]]
61  ; UNPACKED-NEXT:   [[V_AND_B32_e64_:%[0-9]+]]:vgpr_32 = V_AND_B32_e64 [[COPY7]], [[COPY9]], implicit $exec
62  ; UNPACKED-NEXT:   [[COPY10:%[0-9]+]]:vgpr_32 = COPY [[S_MOV_B32_]]
63  ; UNPACKED-NEXT:   [[V_AND_B32_e64_1:%[0-9]+]]:vgpr_32 = V_AND_B32_e64 [[COPY8]], [[COPY10]], implicit $exec
64  ; UNPACKED-NEXT:   [[S_MOV_B32_1:%[0-9]+]]:sreg_32 = S_MOV_B32 16
65  ; UNPACKED-NEXT:   [[COPY11:%[0-9]+]]:vgpr_32 = COPY [[S_MOV_B32_1]]
66  ; UNPACKED-NEXT:   [[V_LSHLREV_B32_e64_:%[0-9]+]]:vgpr_32 = V_LSHLREV_B32_e64 [[COPY11]], [[V_AND_B32_e64_1]], implicit $exec
67  ; UNPACKED-NEXT:   [[V_OR_B32_e64_:%[0-9]+]]:vgpr_32 = V_OR_B32_e64 [[V_AND_B32_e64_]], [[V_LSHLREV_B32_e64_]], implicit $exec
68  ; UNPACKED-NEXT:   $vgpr0 = COPY [[V_OR_B32_e64_]]
69  ; UNPACKED-NEXT:   SI_RETURN_TO_EPILOG implicit $vgpr0
70  ; PACKED-LABEL: name: struct_buffer_load_format_v2f16__sgpr_rsrc__vgpr_vindex__vgpr_voffset__sgpr_soffset
71  ; PACKED: bb.1 (%ir-block.0):
72  ; PACKED-NEXT:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $vgpr0, $vgpr1
73  ; PACKED-NEXT: {{  $}}
74  ; PACKED-NEXT:   [[COPY:%[0-9]+]]:sreg_32 = COPY $sgpr2
75  ; PACKED-NEXT:   [[COPY1:%[0-9]+]]:sreg_32 = COPY $sgpr3
76  ; PACKED-NEXT:   [[COPY2:%[0-9]+]]:sreg_32 = COPY $sgpr4
77  ; PACKED-NEXT:   [[COPY3:%[0-9]+]]:sreg_32 = COPY $sgpr5
78  ; PACKED-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3
79  ; PACKED-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0
80  ; PACKED-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr1
81  ; PACKED-NEXT:   [[COPY6:%[0-9]+]]:sreg_32 = COPY $sgpr6
82  ; PACKED-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY4]], %subreg.sub0, [[COPY5]], %subreg.sub1
83  ; PACKED-NEXT:   [[BUFFER_LOAD_FORMAT_D16_XY_BOTHEN:%[0-9]+]]:vgpr_32 = BUFFER_LOAD_FORMAT_D16_XY_BOTHEN [[REG_SEQUENCE1]], [[REG_SEQUENCE]], [[COPY6]], 0, 0, 0, 0, implicit $exec :: (dereferenceable load (<2 x s16>), align 1, addrspace 4)
84  ; PACKED-NEXT:   $vgpr0 = COPY [[BUFFER_LOAD_FORMAT_D16_XY_BOTHEN]]
85  ; PACKED-NEXT:   SI_RETURN_TO_EPILOG implicit $vgpr0
86  %val = call <2 x half> @llvm.amdgcn.struct.buffer.load.format.v2f16(<4 x i32> %rsrc, i32 %vindex, i32 %voffset, i32 %soffset, i32 0)
87  ret <2 x half> %val
88}
89
90; FIXME: Crashes
91; define amdgpu_ps <3 x half> @struct_buffer_load_format_v3f16__sgpr_rsrc__vgpr_vindex__vgpr_voffset__sgpr_soffset(<4 x i32> inreg %rsrc, i32 %vindex, i32 %voffset, i32 inreg %soffset) {
92;   %val = call <3 x half> @llvm.amdgcn.struct.buffer.load.format.v3f16(<4 x i32> %rsrc, i32 %vindex, i32 %voffset, i32 %soffset, i32 0)
93;   ret <3 x half> %val
94; }
95
96define amdgpu_ps <4 x half> @struct_buffer_load_format_v4f16__sgpr_rsrc__vgpr_vindex__vgpr_voffset__sgpr_soffset(<4 x i32> inreg %rsrc, i32 %vindex, i32 %voffset, i32 inreg %soffset) {
97  ; UNPACKED-LABEL: name: struct_buffer_load_format_v4f16__sgpr_rsrc__vgpr_vindex__vgpr_voffset__sgpr_soffset
98  ; UNPACKED: bb.1 (%ir-block.0):
99  ; UNPACKED-NEXT:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $vgpr0, $vgpr1
100  ; UNPACKED-NEXT: {{  $}}
101  ; UNPACKED-NEXT:   [[COPY:%[0-9]+]]:sreg_32 = COPY $sgpr2
102  ; UNPACKED-NEXT:   [[COPY1:%[0-9]+]]:sreg_32 = COPY $sgpr3
103  ; UNPACKED-NEXT:   [[COPY2:%[0-9]+]]:sreg_32 = COPY $sgpr4
104  ; UNPACKED-NEXT:   [[COPY3:%[0-9]+]]:sreg_32 = COPY $sgpr5
105  ; UNPACKED-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3
106  ; UNPACKED-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0
107  ; UNPACKED-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr1
108  ; UNPACKED-NEXT:   [[COPY6:%[0-9]+]]:sreg_32 = COPY $sgpr6
109  ; UNPACKED-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY4]], %subreg.sub0, [[COPY5]], %subreg.sub1
110  ; UNPACKED-NEXT:   [[BUFFER_LOAD_FORMAT_D16_XYZW_gfx80_BOTHEN:%[0-9]+]]:vreg_128 = BUFFER_LOAD_FORMAT_D16_XYZW_gfx80_BOTHEN [[REG_SEQUENCE1]], [[REG_SEQUENCE]], [[COPY6]], 0, 0, 0, 0, implicit $exec :: (dereferenceable load (<4 x s16>), align 1, addrspace 4)
111  ; UNPACKED-NEXT:   [[COPY7:%[0-9]+]]:vgpr_32 = COPY [[BUFFER_LOAD_FORMAT_D16_XYZW_gfx80_BOTHEN]].sub0
112  ; UNPACKED-NEXT:   [[COPY8:%[0-9]+]]:vgpr_32 = COPY [[BUFFER_LOAD_FORMAT_D16_XYZW_gfx80_BOTHEN]].sub1
113  ; UNPACKED-NEXT:   [[COPY9:%[0-9]+]]:vgpr_32 = COPY [[BUFFER_LOAD_FORMAT_D16_XYZW_gfx80_BOTHEN]].sub2
114  ; UNPACKED-NEXT:   [[COPY10:%[0-9]+]]:vgpr_32 = COPY [[BUFFER_LOAD_FORMAT_D16_XYZW_gfx80_BOTHEN]].sub3
115  ; UNPACKED-NEXT:   [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 65535
116  ; UNPACKED-NEXT:   [[COPY11:%[0-9]+]]:vgpr_32 = COPY [[S_MOV_B32_]]
117  ; UNPACKED-NEXT:   [[V_AND_B32_e64_:%[0-9]+]]:vgpr_32 = V_AND_B32_e64 [[COPY7]], [[COPY11]], implicit $exec
118  ; UNPACKED-NEXT:   [[COPY12:%[0-9]+]]:vgpr_32 = COPY [[S_MOV_B32_]]
119  ; UNPACKED-NEXT:   [[V_AND_B32_e64_1:%[0-9]+]]:vgpr_32 = V_AND_B32_e64 [[COPY8]], [[COPY12]], implicit $exec
120  ; UNPACKED-NEXT:   [[S_MOV_B32_1:%[0-9]+]]:sreg_32 = S_MOV_B32 16
121  ; UNPACKED-NEXT:   [[COPY13:%[0-9]+]]:vgpr_32 = COPY [[S_MOV_B32_1]]
122  ; UNPACKED-NEXT:   [[V_LSHLREV_B32_e64_:%[0-9]+]]:vgpr_32 = V_LSHLREV_B32_e64 [[COPY13]], [[V_AND_B32_e64_1]], implicit $exec
123  ; UNPACKED-NEXT:   [[V_OR_B32_e64_:%[0-9]+]]:vgpr_32 = V_OR_B32_e64 [[V_AND_B32_e64_]], [[V_LSHLREV_B32_e64_]], implicit $exec
124  ; UNPACKED-NEXT:   [[COPY14:%[0-9]+]]:vgpr_32 = COPY [[S_MOV_B32_]]
125  ; UNPACKED-NEXT:   [[V_AND_B32_e64_2:%[0-9]+]]:vgpr_32 = V_AND_B32_e64 [[COPY9]], [[COPY14]], implicit $exec
126  ; UNPACKED-NEXT:   [[COPY15:%[0-9]+]]:vgpr_32 = COPY [[S_MOV_B32_]]
127  ; UNPACKED-NEXT:   [[V_AND_B32_e64_3:%[0-9]+]]:vgpr_32 = V_AND_B32_e64 [[COPY10]], [[COPY15]], implicit $exec
128  ; UNPACKED-NEXT:   [[COPY16:%[0-9]+]]:vgpr_32 = COPY [[S_MOV_B32_1]]
129  ; UNPACKED-NEXT:   [[V_LSHLREV_B32_e64_1:%[0-9]+]]:vgpr_32 = V_LSHLREV_B32_e64 [[COPY16]], [[V_AND_B32_e64_3]], implicit $exec
130  ; UNPACKED-NEXT:   [[V_OR_B32_e64_1:%[0-9]+]]:vgpr_32 = V_OR_B32_e64 [[V_AND_B32_e64_2]], [[V_LSHLREV_B32_e64_1]], implicit $exec
131  ; UNPACKED-NEXT:   $vgpr0 = COPY [[V_OR_B32_e64_]]
132  ; UNPACKED-NEXT:   $vgpr1 = COPY [[V_OR_B32_e64_1]]
133  ; UNPACKED-NEXT:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1
134  ; PACKED-LABEL: name: struct_buffer_load_format_v4f16__sgpr_rsrc__vgpr_vindex__vgpr_voffset__sgpr_soffset
135  ; PACKED: bb.1 (%ir-block.0):
136  ; PACKED-NEXT:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $vgpr0, $vgpr1
137  ; PACKED-NEXT: {{  $}}
138  ; PACKED-NEXT:   [[COPY:%[0-9]+]]:sreg_32 = COPY $sgpr2
139  ; PACKED-NEXT:   [[COPY1:%[0-9]+]]:sreg_32 = COPY $sgpr3
140  ; PACKED-NEXT:   [[COPY2:%[0-9]+]]:sreg_32 = COPY $sgpr4
141  ; PACKED-NEXT:   [[COPY3:%[0-9]+]]:sreg_32 = COPY $sgpr5
142  ; PACKED-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3
143  ; PACKED-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0
144  ; PACKED-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr1
145  ; PACKED-NEXT:   [[COPY6:%[0-9]+]]:sreg_32 = COPY $sgpr6
146  ; PACKED-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY4]], %subreg.sub0, [[COPY5]], %subreg.sub1
147  ; PACKED-NEXT:   [[BUFFER_LOAD_FORMAT_D16_XYZW_BOTHEN:%[0-9]+]]:vreg_64 = BUFFER_LOAD_FORMAT_D16_XYZW_BOTHEN [[REG_SEQUENCE1]], [[REG_SEQUENCE]], [[COPY6]], 0, 0, 0, 0, implicit $exec :: (dereferenceable load (<4 x s16>), align 1, addrspace 4)
148  ; PACKED-NEXT:   [[COPY7:%[0-9]+]]:vgpr_32 = COPY [[BUFFER_LOAD_FORMAT_D16_XYZW_BOTHEN]].sub0
149  ; PACKED-NEXT:   [[COPY8:%[0-9]+]]:vgpr_32 = COPY [[BUFFER_LOAD_FORMAT_D16_XYZW_BOTHEN]].sub1
150  ; PACKED-NEXT:   $vgpr0 = COPY [[COPY7]]
151  ; PACKED-NEXT:   $vgpr1 = COPY [[COPY8]]
152  ; PACKED-NEXT:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1
153  %val = call <4 x half> @llvm.amdgcn.struct.buffer.load.format.v4f16(<4 x i32> %rsrc, i32 %vindex, i32 %voffset, i32 %soffset, i32 0)
154  ret <4 x half> %val
155}
156
157; Waterfall for rsrc and soffset, copy for voffset
158define amdgpu_ps <4 x half> @struct_buffer_load_format_v4f16__vpr_rsrc__sgpr_vindex__sgpr_voffset__vgpr_soffset(<4 x i32> %rsrc, i32 inreg %vindex, i32 inreg %voffset, i32 %soffset) {
159  ; UNPACKED-LABEL: name: struct_buffer_load_format_v4f16__vpr_rsrc__sgpr_vindex__sgpr_voffset__vgpr_soffset
160  ; UNPACKED: bb.1 (%ir-block.0):
161  ; UNPACKED-NEXT:   successors: %bb.2(0x80000000)
162  ; UNPACKED-NEXT:   liveins: $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4
163  ; UNPACKED-NEXT: {{  $}}
164  ; UNPACKED-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
165  ; UNPACKED-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
166  ; UNPACKED-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr2
167  ; UNPACKED-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr3
168  ; UNPACKED-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3
169  ; UNPACKED-NEXT:   [[COPY4:%[0-9]+]]:sreg_32 = COPY $sgpr2
170  ; UNPACKED-NEXT:   [[COPY5:%[0-9]+]]:sreg_32 = COPY $sgpr3
171  ; UNPACKED-NEXT:   [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr4
172  ; UNPACKED-NEXT:   [[COPY7:%[0-9]+]]:vgpr_32 = COPY [[COPY4]]
173  ; UNPACKED-NEXT:   [[COPY8:%[0-9]+]]:vgpr_32 = COPY [[COPY5]]
174  ; UNPACKED-NEXT:   [[COPY9:%[0-9]+]]:vreg_64 = COPY [[REG_SEQUENCE]].sub0_sub1
175  ; UNPACKED-NEXT:   [[COPY10:%[0-9]+]]:vreg_64 = COPY [[REG_SEQUENCE]].sub2_sub3
176  ; UNPACKED-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
177  ; UNPACKED-NEXT: {{  $}}
178  ; UNPACKED-NEXT: bb.2:
179  ; UNPACKED-NEXT:   successors: %bb.3(0x40000000), %bb.2(0x40000000)
180  ; UNPACKED-NEXT: {{  $}}
181  ; UNPACKED-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sreg_32_xm0 = V_READFIRSTLANE_B32 [[COPY9]].sub0, implicit $exec
182  ; UNPACKED-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sreg_32_xm0 = V_READFIRSTLANE_B32 [[COPY9]].sub1, implicit $exec
183  ; UNPACKED-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:sreg_64_xexec = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
184  ; UNPACKED-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE1]], [[COPY9]], implicit $exec
185  ; UNPACKED-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sreg_32_xm0 = V_READFIRSTLANE_B32 [[COPY10]].sub0, implicit $exec
186  ; UNPACKED-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sreg_32_xm0 = V_READFIRSTLANE_B32 [[COPY10]].sub1, implicit $exec
187  ; UNPACKED-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:sreg_64_xexec = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
188  ; UNPACKED-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[COPY10]], implicit $exec
189  ; UNPACKED-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
190  ; UNPACKED-NEXT:   [[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
191  ; UNPACKED-NEXT:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sreg_32_xm0 = V_READFIRSTLANE_B32 [[COPY6]], implicit $exec
192  ; UNPACKED-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]], [[COPY6]], implicit $exec
193  ; UNPACKED-NEXT:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U32_e64_]], [[S_AND_B64_]], implicit-def $scc
194  ; UNPACKED-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
195  ; UNPACKED-NEXT:   [[REG_SEQUENCE4:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY7]], %subreg.sub0, [[COPY8]], %subreg.sub1
196  ; UNPACKED-NEXT:   [[BUFFER_LOAD_FORMAT_D16_XYZW_gfx80_BOTHEN:%[0-9]+]]:vreg_128 = BUFFER_LOAD_FORMAT_D16_XYZW_gfx80_BOTHEN [[REG_SEQUENCE4]], [[REG_SEQUENCE3]], [[V_READFIRSTLANE_B32_4]], 0, 0, 0, 0, implicit $exec :: (dereferenceable load (<4 x s16>), align 1, addrspace 4)
197  ; UNPACKED-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
198  ; UNPACKED-NEXT:   SI_WATERFALL_LOOP %bb.2, implicit $exec
199  ; UNPACKED-NEXT: {{  $}}
200  ; UNPACKED-NEXT: bb.3:
201  ; UNPACKED-NEXT:   successors: %bb.4(0x80000000)
202  ; UNPACKED-NEXT: {{  $}}
203  ; UNPACKED-NEXT:   $exec = S_MOV_B64_term [[S_MOV_B64_]]
204  ; UNPACKED-NEXT: {{  $}}
205  ; UNPACKED-NEXT: bb.4:
206  ; UNPACKED-NEXT:   [[COPY11:%[0-9]+]]:vgpr_32 = COPY [[BUFFER_LOAD_FORMAT_D16_XYZW_gfx80_BOTHEN]].sub0
207  ; UNPACKED-NEXT:   [[COPY12:%[0-9]+]]:vgpr_32 = COPY [[BUFFER_LOAD_FORMAT_D16_XYZW_gfx80_BOTHEN]].sub1
208  ; UNPACKED-NEXT:   [[COPY13:%[0-9]+]]:vgpr_32 = COPY [[BUFFER_LOAD_FORMAT_D16_XYZW_gfx80_BOTHEN]].sub2
209  ; UNPACKED-NEXT:   [[COPY14:%[0-9]+]]:vgpr_32 = COPY [[BUFFER_LOAD_FORMAT_D16_XYZW_gfx80_BOTHEN]].sub3
210  ; UNPACKED-NEXT:   [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 65535
211  ; UNPACKED-NEXT:   [[COPY15:%[0-9]+]]:vgpr_32 = COPY [[S_MOV_B32_]]
212  ; UNPACKED-NEXT:   [[V_AND_B32_e64_:%[0-9]+]]:vgpr_32 = V_AND_B32_e64 [[COPY11]], [[COPY15]], implicit $exec
213  ; UNPACKED-NEXT:   [[COPY16:%[0-9]+]]:vgpr_32 = COPY [[S_MOV_B32_]]
214  ; UNPACKED-NEXT:   [[V_AND_B32_e64_1:%[0-9]+]]:vgpr_32 = V_AND_B32_e64 [[COPY12]], [[COPY16]], implicit $exec
215  ; UNPACKED-NEXT:   [[S_MOV_B32_1:%[0-9]+]]:sreg_32 = S_MOV_B32 16
216  ; UNPACKED-NEXT:   [[COPY17:%[0-9]+]]:vgpr_32 = COPY [[S_MOV_B32_1]]
217  ; UNPACKED-NEXT:   [[V_LSHLREV_B32_e64_:%[0-9]+]]:vgpr_32 = V_LSHLREV_B32_e64 [[COPY17]], [[V_AND_B32_e64_1]], implicit $exec
218  ; UNPACKED-NEXT:   [[V_OR_B32_e64_:%[0-9]+]]:vgpr_32 = V_OR_B32_e64 [[V_AND_B32_e64_]], [[V_LSHLREV_B32_e64_]], implicit $exec
219  ; UNPACKED-NEXT:   [[COPY18:%[0-9]+]]:vgpr_32 = COPY [[S_MOV_B32_]]
220  ; UNPACKED-NEXT:   [[V_AND_B32_e64_2:%[0-9]+]]:vgpr_32 = V_AND_B32_e64 [[COPY13]], [[COPY18]], implicit $exec
221  ; UNPACKED-NEXT:   [[COPY19:%[0-9]+]]:vgpr_32 = COPY [[S_MOV_B32_]]
222  ; UNPACKED-NEXT:   [[V_AND_B32_e64_3:%[0-9]+]]:vgpr_32 = V_AND_B32_e64 [[COPY14]], [[COPY19]], implicit $exec
223  ; UNPACKED-NEXT:   [[COPY20:%[0-9]+]]:vgpr_32 = COPY [[S_MOV_B32_1]]
224  ; UNPACKED-NEXT:   [[V_LSHLREV_B32_e64_1:%[0-9]+]]:vgpr_32 = V_LSHLREV_B32_e64 [[COPY20]], [[V_AND_B32_e64_3]], implicit $exec
225  ; UNPACKED-NEXT:   [[V_OR_B32_e64_1:%[0-9]+]]:vgpr_32 = V_OR_B32_e64 [[V_AND_B32_e64_2]], [[V_LSHLREV_B32_e64_1]], implicit $exec
226  ; UNPACKED-NEXT:   $vgpr0 = COPY [[V_OR_B32_e64_]]
227  ; UNPACKED-NEXT:   $vgpr1 = COPY [[V_OR_B32_e64_1]]
228  ; UNPACKED-NEXT:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1
229  ; PACKED-LABEL: name: struct_buffer_load_format_v4f16__vpr_rsrc__sgpr_vindex__sgpr_voffset__vgpr_soffset
230  ; PACKED: bb.1 (%ir-block.0):
231  ; PACKED-NEXT:   successors: %bb.2(0x80000000)
232  ; PACKED-NEXT:   liveins: $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4
233  ; PACKED-NEXT: {{  $}}
234  ; PACKED-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
235  ; PACKED-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
236  ; PACKED-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr2
237  ; PACKED-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr3
238  ; PACKED-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3
239  ; PACKED-NEXT:   [[COPY4:%[0-9]+]]:sreg_32 = COPY $sgpr2
240  ; PACKED-NEXT:   [[COPY5:%[0-9]+]]:sreg_32 = COPY $sgpr3
241  ; PACKED-NEXT:   [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr4
242  ; PACKED-NEXT:   [[COPY7:%[0-9]+]]:vgpr_32 = COPY [[COPY4]]
243  ; PACKED-NEXT:   [[COPY8:%[0-9]+]]:vgpr_32 = COPY [[COPY5]]
244  ; PACKED-NEXT:   [[COPY9:%[0-9]+]]:vreg_64 = COPY [[REG_SEQUENCE]].sub0_sub1
245  ; PACKED-NEXT:   [[COPY10:%[0-9]+]]:vreg_64 = COPY [[REG_SEQUENCE]].sub2_sub3
246  ; PACKED-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
247  ; PACKED-NEXT: {{  $}}
248  ; PACKED-NEXT: bb.2:
249  ; PACKED-NEXT:   successors: %bb.3(0x40000000), %bb.2(0x40000000)
250  ; PACKED-NEXT: {{  $}}
251  ; PACKED-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sreg_32_xm0 = V_READFIRSTLANE_B32 [[COPY9]].sub0, implicit $exec
252  ; PACKED-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sreg_32_xm0 = V_READFIRSTLANE_B32 [[COPY9]].sub1, implicit $exec
253  ; PACKED-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:sreg_64_xexec = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
254  ; PACKED-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE1]], [[COPY9]], implicit $exec
255  ; PACKED-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sreg_32_xm0 = V_READFIRSTLANE_B32 [[COPY10]].sub0, implicit $exec
256  ; PACKED-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sreg_32_xm0 = V_READFIRSTLANE_B32 [[COPY10]].sub1, implicit $exec
257  ; PACKED-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:sreg_64_xexec = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
258  ; PACKED-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[COPY10]], implicit $exec
259  ; PACKED-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
260  ; PACKED-NEXT:   [[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
261  ; PACKED-NEXT:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sreg_32_xm0 = V_READFIRSTLANE_B32 [[COPY6]], implicit $exec
262  ; PACKED-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]], [[COPY6]], implicit $exec
263  ; PACKED-NEXT:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U32_e64_]], [[S_AND_B64_]], implicit-def $scc
264  ; PACKED-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
265  ; PACKED-NEXT:   [[REG_SEQUENCE4:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY7]], %subreg.sub0, [[COPY8]], %subreg.sub1
266  ; PACKED-NEXT:   [[BUFFER_LOAD_FORMAT_D16_XYZW_BOTHEN:%[0-9]+]]:vreg_64 = BUFFER_LOAD_FORMAT_D16_XYZW_BOTHEN [[REG_SEQUENCE4]], [[REG_SEQUENCE3]], [[V_READFIRSTLANE_B32_4]], 0, 0, 0, 0, implicit $exec :: (dereferenceable load (<4 x s16>), align 1, addrspace 4)
267  ; PACKED-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
268  ; PACKED-NEXT:   SI_WATERFALL_LOOP %bb.2, implicit $exec
269  ; PACKED-NEXT: {{  $}}
270  ; PACKED-NEXT: bb.3:
271  ; PACKED-NEXT:   successors: %bb.4(0x80000000)
272  ; PACKED-NEXT: {{  $}}
273  ; PACKED-NEXT:   $exec = S_MOV_B64_term [[S_MOV_B64_]]
274  ; PACKED-NEXT: {{  $}}
275  ; PACKED-NEXT: bb.4:
276  ; PACKED-NEXT:   [[COPY11:%[0-9]+]]:vgpr_32 = COPY [[BUFFER_LOAD_FORMAT_D16_XYZW_BOTHEN]].sub0
277  ; PACKED-NEXT:   [[COPY12:%[0-9]+]]:vgpr_32 = COPY [[BUFFER_LOAD_FORMAT_D16_XYZW_BOTHEN]].sub1
278  ; PACKED-NEXT:   $vgpr0 = COPY [[COPY11]]
279  ; PACKED-NEXT:   $vgpr1 = COPY [[COPY12]]
280  ; PACKED-NEXT:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1
281  %val = call <4 x half> @llvm.amdgcn.struct.buffer.load.format.v4f16(<4 x i32> %rsrc, i32 %vindex, i32 %voffset, i32 %soffset, i32 0)
282  ret <4 x half> %val
283}
284
285define amdgpu_ps half @struct_buffer_load_format_f16__sgpr_rsrc__vgpr_vindex__vgpr_voffset__sgpr_soffset_voffsset_add_4095(<4 x i32> inreg %rsrc, i32 %vindex, i32 %voffset.base, i32 inreg %soffset) {
286  ; UNPACKED-LABEL: name: struct_buffer_load_format_f16__sgpr_rsrc__vgpr_vindex__vgpr_voffset__sgpr_soffset_voffsset_add_4095
287  ; UNPACKED: bb.1 (%ir-block.0):
288  ; UNPACKED-NEXT:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $vgpr0, $vgpr1
289  ; UNPACKED-NEXT: {{  $}}
290  ; UNPACKED-NEXT:   [[COPY:%[0-9]+]]:sreg_32 = COPY $sgpr2
291  ; UNPACKED-NEXT:   [[COPY1:%[0-9]+]]:sreg_32 = COPY $sgpr3
292  ; UNPACKED-NEXT:   [[COPY2:%[0-9]+]]:sreg_32 = COPY $sgpr4
293  ; UNPACKED-NEXT:   [[COPY3:%[0-9]+]]:sreg_32 = COPY $sgpr5
294  ; UNPACKED-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3
295  ; UNPACKED-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0
296  ; UNPACKED-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr1
297  ; UNPACKED-NEXT:   [[COPY6:%[0-9]+]]:sreg_32 = COPY $sgpr6
298  ; UNPACKED-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY4]], %subreg.sub0, [[COPY5]], %subreg.sub1
299  ; UNPACKED-NEXT:   [[BUFFER_LOAD_FORMAT_D16_X_gfx80_BOTHEN:%[0-9]+]]:vgpr_32 = BUFFER_LOAD_FORMAT_D16_X_gfx80_BOTHEN [[REG_SEQUENCE1]], [[REG_SEQUENCE]], [[COPY6]], 4095, 0, 0, 0, implicit $exec :: (dereferenceable load (s16), align 1, addrspace 4)
300  ; UNPACKED-NEXT:   $vgpr0 = COPY [[BUFFER_LOAD_FORMAT_D16_X_gfx80_BOTHEN]]
301  ; UNPACKED-NEXT:   SI_RETURN_TO_EPILOG implicit $vgpr0
302  ; PACKED-LABEL: name: struct_buffer_load_format_f16__sgpr_rsrc__vgpr_vindex__vgpr_voffset__sgpr_soffset_voffsset_add_4095
303  ; PACKED: bb.1 (%ir-block.0):
304  ; PACKED-NEXT:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $vgpr0, $vgpr1
305  ; PACKED-NEXT: {{  $}}
306  ; PACKED-NEXT:   [[COPY:%[0-9]+]]:sreg_32 = COPY $sgpr2
307  ; PACKED-NEXT:   [[COPY1:%[0-9]+]]:sreg_32 = COPY $sgpr3
308  ; PACKED-NEXT:   [[COPY2:%[0-9]+]]:sreg_32 = COPY $sgpr4
309  ; PACKED-NEXT:   [[COPY3:%[0-9]+]]:sreg_32 = COPY $sgpr5
310  ; PACKED-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3
311  ; PACKED-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0
312  ; PACKED-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr1
313  ; PACKED-NEXT:   [[COPY6:%[0-9]+]]:sreg_32 = COPY $sgpr6
314  ; PACKED-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY4]], %subreg.sub0, [[COPY5]], %subreg.sub1
315  ; PACKED-NEXT:   [[BUFFER_LOAD_FORMAT_D16_X_BOTHEN:%[0-9]+]]:vgpr_32 = BUFFER_LOAD_FORMAT_D16_X_BOTHEN [[REG_SEQUENCE1]], [[REG_SEQUENCE]], [[COPY6]], 4095, 0, 0, 0, implicit $exec :: (dereferenceable load (s16), align 1, addrspace 4)
316  ; PACKED-NEXT:   $vgpr0 = COPY [[BUFFER_LOAD_FORMAT_D16_X_BOTHEN]]
317  ; PACKED-NEXT:   SI_RETURN_TO_EPILOG implicit $vgpr0
318  %voffset = add i32 %voffset.base, 4095
319  %val = call half @llvm.amdgcn.struct.buffer.load.format.f16(<4 x i32> %rsrc, i32 %vindex, i32 %voffset, i32 %soffset, i32 0)
320  ret half %val
321}
322
323define amdgpu_ps half @struct_buffer_load_format_i16__sgpr_rsrc__vgpr_vindex__vgpr_voffset__sgpr_soffset(<4 x i32> inreg %rsrc, i32 %vindex, i32 %voffset, i32 inreg %soffset) {
324  ; UNPACKED-LABEL: name: struct_buffer_load_format_i16__sgpr_rsrc__vgpr_vindex__vgpr_voffset__sgpr_soffset
325  ; UNPACKED: bb.1 (%ir-block.0):
326  ; UNPACKED-NEXT:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $vgpr0, $vgpr1
327  ; UNPACKED-NEXT: {{  $}}
328  ; UNPACKED-NEXT:   [[COPY:%[0-9]+]]:sreg_32 = COPY $sgpr2
329  ; UNPACKED-NEXT:   [[COPY1:%[0-9]+]]:sreg_32 = COPY $sgpr3
330  ; UNPACKED-NEXT:   [[COPY2:%[0-9]+]]:sreg_32 = COPY $sgpr4
331  ; UNPACKED-NEXT:   [[COPY3:%[0-9]+]]:sreg_32 = COPY $sgpr5
332  ; UNPACKED-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3
333  ; UNPACKED-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0
334  ; UNPACKED-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr1
335  ; UNPACKED-NEXT:   [[COPY6:%[0-9]+]]:sreg_32 = COPY $sgpr6
336  ; UNPACKED-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY4]], %subreg.sub0, [[COPY5]], %subreg.sub1
337  ; UNPACKED-NEXT:   [[BUFFER_LOAD_FORMAT_D16_X_gfx80_BOTHEN:%[0-9]+]]:vgpr_32 = BUFFER_LOAD_FORMAT_D16_X_gfx80_BOTHEN [[REG_SEQUENCE1]], [[REG_SEQUENCE]], [[COPY6]], 0, 0, 0, 0, implicit $exec :: (dereferenceable load (s16), align 1, addrspace 4)
338  ; UNPACKED-NEXT:   $vgpr0 = COPY [[BUFFER_LOAD_FORMAT_D16_X_gfx80_BOTHEN]]
339  ; UNPACKED-NEXT:   SI_RETURN_TO_EPILOG implicit $vgpr0
340  ; PACKED-LABEL: name: struct_buffer_load_format_i16__sgpr_rsrc__vgpr_vindex__vgpr_voffset__sgpr_soffset
341  ; PACKED: bb.1 (%ir-block.0):
342  ; PACKED-NEXT:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $vgpr0, $vgpr1
343  ; PACKED-NEXT: {{  $}}
344  ; PACKED-NEXT:   [[COPY:%[0-9]+]]:sreg_32 = COPY $sgpr2
345  ; PACKED-NEXT:   [[COPY1:%[0-9]+]]:sreg_32 = COPY $sgpr3
346  ; PACKED-NEXT:   [[COPY2:%[0-9]+]]:sreg_32 = COPY $sgpr4
347  ; PACKED-NEXT:   [[COPY3:%[0-9]+]]:sreg_32 = COPY $sgpr5
348  ; PACKED-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3
349  ; PACKED-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0
350  ; PACKED-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr1
351  ; PACKED-NEXT:   [[COPY6:%[0-9]+]]:sreg_32 = COPY $sgpr6
352  ; PACKED-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY4]], %subreg.sub0, [[COPY5]], %subreg.sub1
353  ; PACKED-NEXT:   [[BUFFER_LOAD_FORMAT_D16_X_BOTHEN:%[0-9]+]]:vgpr_32 = BUFFER_LOAD_FORMAT_D16_X_BOTHEN [[REG_SEQUENCE1]], [[REG_SEQUENCE]], [[COPY6]], 0, 0, 0, 0, implicit $exec :: (dereferenceable load (s16), align 1, addrspace 4)
354  ; PACKED-NEXT:   $vgpr0 = COPY [[BUFFER_LOAD_FORMAT_D16_X_BOTHEN]]
355  ; PACKED-NEXT:   SI_RETURN_TO_EPILOG implicit $vgpr0
356  %val = call i16 @llvm.amdgcn.struct.buffer.load.format.i16(<4 x i32> %rsrc, i32 %vindex, i32 %voffset, i32 %soffset, i32 0)
357  %fval = bitcast i16 %val to half
358  ret half %fval
359}
360
361declare half @llvm.amdgcn.struct.buffer.load.format.f16(<4 x i32>, i32, i32, i32, i32 immarg) #0
362declare <2 x half> @llvm.amdgcn.struct.buffer.load.format.v2f16(<4 x i32>, i32, i32, i32, i32 immarg) #0
363declare <3 x half> @llvm.amdgcn.struct.buffer.load.format.v3f16(<4 x i32>, i32, i32, i32, i32 immarg) #0
364declare <4 x half> @llvm.amdgcn.struct.buffer.load.format.v4f16(<4 x i32>, i32, i32, i32, i32 immarg) #0
365declare i16 @llvm.amdgcn.struct.buffer.load.format.i16(<4 x i32>, i32, i32, i32, i32 immarg) #0
366
367attributes #0 = { nounwind readonly }
368