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 %s
3
4; Natural mapping
5define amdgpu_ps float @raw_buffer_load_format_f32__sgpr_rsrc__vgpr_voffset__sgpr_soffset(<4 x i32> inreg %rsrc, i32 %voffset, i32 inreg %soffset) {
6  ; CHECK-LABEL: name: raw_buffer_load_format_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:   [[BUFFER_LOAD_FORMAT_X_OFFEN:%[0-9]+]]:vgpr_32 = BUFFER_LOAD_FORMAT_X_OFFEN [[COPY4]], [[REG_SEQUENCE]], [[COPY5]], 0, 0, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4)
18  ; CHECK-NEXT:   $vgpr0 = COPY [[BUFFER_LOAD_FORMAT_X_OFFEN]]
19  ; CHECK-NEXT:   SI_RETURN_TO_EPILOG implicit $vgpr0
20  %val = call float @llvm.amdgcn.raw.buffer.load.format.f32(<4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 0)
21  ret float %val
22}
23
24define amdgpu_ps <2 x float> @raw_buffer_load_format_v2f32__sgpr_rsrc__vgpr_voffset__sgpr_soffset(<4 x i32> inreg %rsrc, i32 %voffset, i32 inreg %soffset) {
25  ; CHECK-LABEL: name: raw_buffer_load_format_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:   [[BUFFER_LOAD_FORMAT_XY_OFFEN:%[0-9]+]]:vreg_64 = BUFFER_LOAD_FORMAT_XY_OFFEN [[COPY4]], [[REG_SEQUENCE]], [[COPY5]], 0, 0, 0, 0, implicit $exec :: (dereferenceable load (<2 x s32>), align 1, addrspace 4)
37  ; CHECK-NEXT:   [[COPY6:%[0-9]+]]:vgpr_32 = COPY [[BUFFER_LOAD_FORMAT_XY_OFFEN]].sub0
38  ; CHECK-NEXT:   [[COPY7:%[0-9]+]]:vgpr_32 = COPY [[BUFFER_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.buffer.load.format.v2f32(<4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 0)
43  ret <2 x float> %val
44}
45
46define amdgpu_ps <3 x float> @raw_buffer_load_format_v3f32__sgpr_rsrc__vgpr_voffset__sgpr_soffset(<4 x i32> inreg %rsrc, i32 %voffset, i32 inreg %soffset) {
47  ; CHECK-LABEL: name: raw_buffer_load_format_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:   [[BUFFER_LOAD_FORMAT_XYZ_OFFEN:%[0-9]+]]:vreg_96 = BUFFER_LOAD_FORMAT_XYZ_OFFEN [[COPY4]], [[REG_SEQUENCE]], [[COPY5]], 0, 0, 0, 0, implicit $exec :: (dereferenceable load (<3 x s32>), align 1, addrspace 4)
59  ; CHECK-NEXT:   [[COPY6:%[0-9]+]]:vgpr_32 = COPY [[BUFFER_LOAD_FORMAT_XYZ_OFFEN]].sub0
60  ; CHECK-NEXT:   [[COPY7:%[0-9]+]]:vgpr_32 = COPY [[BUFFER_LOAD_FORMAT_XYZ_OFFEN]].sub1
61  ; CHECK-NEXT:   [[COPY8:%[0-9]+]]:vgpr_32 = COPY [[BUFFER_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.buffer.load.format.v3f32(<4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 0)
67  ret <3 x float> %val
68}
69
70define amdgpu_ps <4 x float> @raw_buffer_load_format_v4f32__sgpr_rsrc__vgpr_voffset__sgpr_soffset(<4 x i32> inreg %rsrc, i32 %voffset, i32 inreg %soffset) {
71  ; CHECK-LABEL: name: raw_buffer_load_format_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:   [[BUFFER_LOAD_FORMAT_XYZW_OFFEN:%[0-9]+]]:vreg_128 = BUFFER_LOAD_FORMAT_XYZW_OFFEN [[COPY4]], [[REG_SEQUENCE]], [[COPY5]], 0, 0, 0, 0, implicit $exec :: (dereferenceable load (<4 x s32>), align 1, addrspace 4)
83  ; CHECK-NEXT:   [[COPY6:%[0-9]+]]:vgpr_32 = COPY [[BUFFER_LOAD_FORMAT_XYZW_OFFEN]].sub0
84  ; CHECK-NEXT:   [[COPY7:%[0-9]+]]:vgpr_32 = COPY [[BUFFER_LOAD_FORMAT_XYZW_OFFEN]].sub1
85  ; CHECK-NEXT:   [[COPY8:%[0-9]+]]:vgpr_32 = COPY [[BUFFER_LOAD_FORMAT_XYZW_OFFEN]].sub2
86  ; CHECK-NEXT:   [[COPY9:%[0-9]+]]:vgpr_32 = COPY [[BUFFER_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.buffer.load.format.v4f32(<4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 0)
93  ret <4 x float> %val
94}
95
96; Waterfall for rsrc and soffset, copy for voffset
97define amdgpu_ps float @raw_buffer_load_format_f32__vgpr_rsrc__sgpr_voffset__vgpr_soffset(<4 x i32> %rsrc, i32 inreg %voffset, i32 %soffset) {
98  ; CHECK-LABEL: name: raw_buffer_load_format_f32__vgpr_rsrc__sgpr_voffset__vgpr_soffset
99  ; CHECK: bb.1 (%ir-block.0):
100  ; CHECK-NEXT:   successors: %bb.2(0x80000000)
101  ; CHECK-NEXT:   liveins: $sgpr2, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4
102  ; CHECK-NEXT: {{  $}}
103  ; CHECK-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
104  ; CHECK-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
105  ; CHECK-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr2
106  ; CHECK-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr3
107  ; CHECK-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3
108  ; CHECK-NEXT:   [[COPY4:%[0-9]+]]:sreg_32 = COPY $sgpr2
109  ; CHECK-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr4
110  ; CHECK-NEXT:   [[COPY6:%[0-9]+]]:vgpr_32 = COPY [[COPY4]]
111  ; CHECK-NEXT:   [[COPY7:%[0-9]+]]:vreg_64 = COPY [[REG_SEQUENCE]].sub0_sub1
112  ; CHECK-NEXT:   [[COPY8:%[0-9]+]]:vreg_64 = COPY [[REG_SEQUENCE]].sub2_sub3
113  ; CHECK-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
114  ; CHECK-NEXT: {{  $}}
115  ; CHECK-NEXT: bb.2:
116  ; CHECK-NEXT:   successors: %bb.3(0x40000000), %bb.2(0x40000000)
117  ; CHECK-NEXT: {{  $}}
118  ; CHECK-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sreg_32_xm0 = V_READFIRSTLANE_B32 [[COPY7]].sub0, implicit $exec
119  ; CHECK-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sreg_32_xm0 = V_READFIRSTLANE_B32 [[COPY7]].sub1, implicit $exec
120  ; CHECK-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:sreg_64_xexec = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
121  ; CHECK-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE1]], [[COPY7]], implicit $exec
122  ; CHECK-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sreg_32_xm0 = V_READFIRSTLANE_B32 [[COPY8]].sub0, implicit $exec
123  ; CHECK-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sreg_32_xm0 = V_READFIRSTLANE_B32 [[COPY8]].sub1, implicit $exec
124  ; CHECK-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:sreg_64_xexec = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
125  ; CHECK-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[COPY8]], implicit $exec
126  ; CHECK-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
127  ; CHECK-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
128  ; CHECK-NEXT:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sreg_32_xm0 = V_READFIRSTLANE_B32 [[COPY5]], implicit $exec
129  ; CHECK-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]], [[COPY5]], implicit $exec
130  ; CHECK-NEXT:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U32_e64_]], [[S_AND_B64_]], implicit-def $scc
131  ; CHECK-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_1]], implicit-def $exec, implicit-def $scc, implicit $exec
132  ; CHECK-NEXT:   [[BUFFER_LOAD_FORMAT_X_OFFEN:%[0-9]+]]:vgpr_32 = BUFFER_LOAD_FORMAT_X_OFFEN [[COPY6]], [[REG_SEQUENCE3]], [[V_READFIRSTLANE_B32_4]], 0, 0, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4)
133  ; CHECK-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
134  ; CHECK-NEXT:   SI_WATERFALL_LOOP %bb.2, implicit $exec
135  ; CHECK-NEXT: {{  $}}
136  ; CHECK-NEXT: bb.3:
137  ; CHECK-NEXT:   successors: %bb.4(0x80000000)
138  ; CHECK-NEXT: {{  $}}
139  ; CHECK-NEXT:   $exec = S_MOV_B64_term [[S_MOV_B64_]]
140  ; CHECK-NEXT: {{  $}}
141  ; CHECK-NEXT: bb.4:
142  ; CHECK-NEXT:   $vgpr0 = COPY [[BUFFER_LOAD_FORMAT_X_OFFEN]]
143  ; CHECK-NEXT:   SI_RETURN_TO_EPILOG implicit $vgpr0
144  %val = call float @llvm.amdgcn.raw.buffer.load.format.f32(<4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 0)
145  ret float %val
146}
147
148define amdgpu_ps <4 x float> @raw_buffer_load_format_v4f32__sgpr_rsrc__vgpr_voffset__sgpr_soffset_voffset_add_4095(<4 x i32> inreg %rsrc, i32 %voffset.base, i32 inreg %soffset) {
149  ; CHECK-LABEL: name: raw_buffer_load_format_v4f32__sgpr_rsrc__vgpr_voffset__sgpr_soffset_voffset_add_4095
150  ; CHECK: bb.1 (%ir-block.0):
151  ; CHECK-NEXT:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $vgpr0
152  ; CHECK-NEXT: {{  $}}
153  ; CHECK-NEXT:   [[COPY:%[0-9]+]]:sreg_32 = COPY $sgpr2
154  ; CHECK-NEXT:   [[COPY1:%[0-9]+]]:sreg_32 = COPY $sgpr3
155  ; CHECK-NEXT:   [[COPY2:%[0-9]+]]:sreg_32 = COPY $sgpr4
156  ; CHECK-NEXT:   [[COPY3:%[0-9]+]]:sreg_32 = COPY $sgpr5
157  ; CHECK-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3
158  ; CHECK-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0
159  ; CHECK-NEXT:   [[COPY5:%[0-9]+]]:sreg_32 = COPY $sgpr6
160  ; CHECK-NEXT:   [[BUFFER_LOAD_FORMAT_XYZW_OFFEN:%[0-9]+]]:vreg_128 = BUFFER_LOAD_FORMAT_XYZW_OFFEN [[COPY4]], [[REG_SEQUENCE]], [[COPY5]], 4095, 0, 0, 0, implicit $exec :: (dereferenceable load (<4 x s32>), align 1, addrspace 4)
161  ; CHECK-NEXT:   [[COPY6:%[0-9]+]]:vgpr_32 = COPY [[BUFFER_LOAD_FORMAT_XYZW_OFFEN]].sub0
162  ; CHECK-NEXT:   [[COPY7:%[0-9]+]]:vgpr_32 = COPY [[BUFFER_LOAD_FORMAT_XYZW_OFFEN]].sub1
163  ; CHECK-NEXT:   [[COPY8:%[0-9]+]]:vgpr_32 = COPY [[BUFFER_LOAD_FORMAT_XYZW_OFFEN]].sub2
164  ; CHECK-NEXT:   [[COPY9:%[0-9]+]]:vgpr_32 = COPY [[BUFFER_LOAD_FORMAT_XYZW_OFFEN]].sub3
165  ; CHECK-NEXT:   $vgpr0 = COPY [[COPY6]]
166  ; CHECK-NEXT:   $vgpr1 = COPY [[COPY7]]
167  ; CHECK-NEXT:   $vgpr2 = COPY [[COPY8]]
168  ; CHECK-NEXT:   $vgpr3 = COPY [[COPY9]]
169  ; CHECK-NEXT:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
170  %voffset = add i32 %voffset.base, 4095
171  %val = call <4 x float> @llvm.amdgcn.raw.buffer.load.format.v4f32(<4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 0)
172  ret <4 x float> %val
173}
174
175declare float @llvm.amdgcn.raw.buffer.load.format.f32(<4 x i32>, i32, i32, i32 immarg) #0
176declare <2 x float> @llvm.amdgcn.raw.buffer.load.format.v2f32(<4 x i32>, i32, i32, i32 immarg) #0
177declare <3 x float> @llvm.amdgcn.raw.buffer.load.format.v3f32(<4 x i32>, i32, i32, i32 immarg) #0
178declare <4 x float> @llvm.amdgcn.raw.buffer.load.format.v4f32(<4 x i32>, i32, i32, i32 immarg) #0
179
180attributes #0 = { nounwind readonly }
181