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