1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
|
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc -march=amdgcn -mcpu=gfx1100 -verify-machineinstrs < %s | FileCheck -check-prefixes=GFX11 %s
; RUN: llc -march=amdgcn -mcpu=gfx1200 -verify-machineinstrs < %s | FileCheck -check-prefixes=GFX12 %s
; RUN: llc -global-isel -march=amdgcn -mcpu=gfx1200 -verify-machineinstrs < %s | FileCheck -check-prefixes=GFX12-GISEL %s
define amdgpu_cs void @mixed_vmem_types(i32 inreg %globalTable, i32 inreg %perShaderTable, i32 inreg %descTable0, i32 inreg %descTable1, <3 x i32> inreg %WorkgroupId, i32 inreg %MultiDispatchInfo, <3 x i32> %LocalInvocationId) #0 {
; GFX11-LABEL: mixed_vmem_types:
; GFX11: ; %bb.0: ; %.entry
; GFX11-NEXT: s_getpc_b64 s[4:5]
; GFX11-NEXT: s_mov_b32 s0, s3
; GFX11-NEXT: s_mov_b32 s3, s5
; GFX11-NEXT: s_mov_b32 s1, s5
; GFX11-NEXT: s_load_b256 s[20:27], s[2:3], 0x40
; GFX11-NEXT: s_load_b512 s[4:19], s[0:1], 0x0
; GFX11-NEXT: s_load_b512 s[36:51], s[2:3], 0x0
; GFX11-NEXT: v_mov_b32_e32 v0, 0xbc00bc00
; GFX11-NEXT: s_waitcnt lgkmcnt(0)
; GFX11-NEXT: buffer_load_b32 v1, off, s[20:23], 0
; GFX11-NEXT: buffer_load_b32 v2, off, s[16:19], 0
; GFX11-NEXT: image_sample_lz v3, v0, s[8:15], s[4:7] dmask:0x1 dim:SQ_RSRC_IMG_2D a16
; GFX11-NEXT: buffer_load_b32 v4, off, s[40:43], 0
; GFX11-NEXT: image_sample_lz v0, v0, s[44:51], s[36:39] dmask:0x1 dim:SQ_RSRC_IMG_2D a16
; GFX11-NEXT: s_waitcnt vmcnt(4)
; GFX11-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0xac0, v1
; GFX11-NEXT: s_waitcnt vmcnt(3)
; GFX11-NEXT: v_cmp_eq_u32_e64 s0, 0xac0, v2
; GFX11-NEXT: s_waitcnt vmcnt(2)
; GFX11-NEXT: v_cmp_eq_f32_e64 s1, 1.0, v3
; GFX11-NEXT: s_waitcnt vmcnt(1)
; GFX11-NEXT: v_cmp_eq_u32_e64 s2, 0xac0, v4
; GFX11-NEXT: s_and_b32 s0, s0, vcc_lo
; GFX11-NEXT: s_waitcnt vmcnt(0)
; GFX11-NEXT: v_cmp_eq_f32_e32 vcc_lo, 0, v0
; GFX11-NEXT: s_and_b32 s0, s0, s1
; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(SALU_CYCLE_1)
; GFX11-NEXT: s_and_b32 s0, s0, s2
; GFX11-NEXT: s_and_b32 s0, s0, vcc_lo
; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
; GFX11-NEXT: v_cndmask_b32_e64 v0, 0, 1, s0
; GFX11-NEXT: buffer_store_b32 v0, off, s[24:27], 0
; GFX11-NEXT: s_endpgm
;
; GFX12-LABEL: mixed_vmem_types:
; GFX12: ; %bb.0: ; %.entry
; GFX12-NEXT: s_getpc_b64 s[4:5]
; GFX12-NEXT: s_mov_b32 s0, s3
; GFX12-NEXT: s_sext_i32_i16 s5, s5
; GFX12-NEXT: v_mov_b32_e32 v0, 0xbc00bc00
; GFX12-NEXT: s_mov_b32 s3, s5
; GFX12-NEXT: s_mov_b32 s1, s5
; GFX12-NEXT: s_load_b256 s[20:27], s[2:3], 0x40
; GFX12-NEXT: s_load_b512 s[4:19], s[0:1], 0x0
; GFX12-NEXT: s_load_b512 s[36:51], s[2:3], 0x0
; GFX12-NEXT: s_wait_kmcnt 0x0
; GFX12-NEXT: buffer_load_b32 v1, off, s[20:23], null
; GFX12-NEXT: buffer_load_b32 v2, off, s[16:19], null
; GFX12-NEXT: image_sample_lz v3, v0, s[8:15], s[4:7] dmask:0x1 dim:SQ_RSRC_IMG_2D a16
; GFX12-NEXT: buffer_load_b32 v4, off, s[40:43], null
; GFX12-NEXT: image_sample_lz v0, v0, s[44:51], s[36:39] dmask:0x1 dim:SQ_RSRC_IMG_2D a16
; GFX12-NEXT: s_wait_loadcnt 0x2
; GFX12-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0xac0, v1
; GFX12-NEXT: s_wait_loadcnt 0x1
; GFX12-NEXT: v_cmp_eq_u32_e64 s0, 0xac0, v2
; GFX12-NEXT: s_wait_samplecnt 0x1
; GFX12-NEXT: v_cmp_eq_f32_e64 s1, 1.0, v3
; GFX12-NEXT: s_wait_loadcnt 0x0
; GFX12-NEXT: v_cmp_eq_u32_e64 s2, 0xac0, v4
; GFX12-NEXT: s_and_b32 s0, s0, vcc_lo
; GFX12-NEXT: s_wait_samplecnt 0x0
; GFX12-NEXT: v_cmp_eq_f32_e32 vcc_lo, 0, v0
; GFX12-NEXT: s_and_b32 s0, s0, s1
; GFX12-NEXT: s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(SALU_CYCLE_1)
; GFX12-NEXT: s_and_b32 s0, s0, s2
; GFX12-NEXT: s_and_b32 s0, s0, vcc_lo
; GFX12-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
; GFX12-NEXT: v_cndmask_b32_e64 v0, 0, 1, s0
; GFX12-NEXT: buffer_store_b32 v0, off, s[24:27], null
; GFX12-NEXT: s_endpgm
;
; GFX12-GISEL-LABEL: mixed_vmem_types:
; GFX12-GISEL: ; %bb.0: ; %.entry
; GFX12-GISEL-NEXT: s_getpc_b64 s[20:21]
; GFX12-GISEL-NEXT: s_mov_b32 s0, s3
; GFX12-GISEL-NEXT: s_sext_i32_i16 s21, s21
; GFX12-GISEL-NEXT: v_mov_b32_e32 v0, 0xbc00bc00
; GFX12-GISEL-NEXT: s_mov_b32 s1, s21
; GFX12-GISEL-NEXT: s_mov_b32 s3, s21
; GFX12-GISEL-NEXT: s_load_b512 s[4:19], s[0:1], 0x0
; GFX12-GISEL-NEXT: s_clause 0x1
; GFX12-GISEL-NEXT: s_load_b256 s[20:27], s[2:3], 0x40
; GFX12-GISEL-NEXT: s_load_b512 s[36:51], s[2:3], 0x0
; GFX12-GISEL-NEXT: s_wait_kmcnt 0x0
; GFX12-GISEL-NEXT: image_sample_lz v1, v0, s[8:15], s[4:7] dmask:0x1 dim:SQ_RSRC_IMG_2D a16
; GFX12-GISEL-NEXT: buffer_load_b32 v2, off, s[16:19], null
; GFX12-GISEL-NEXT: buffer_load_b32 v3, off, s[20:23], null
; GFX12-GISEL-NEXT: buffer_load_b32 v4, off, s[40:43], null
; GFX12-GISEL-NEXT: image_sample_lz v0, v0, s[44:51], s[36:39] dmask:0x1 dim:SQ_RSRC_IMG_2D a16
; GFX12-GISEL-NEXT: s_wait_loadcnt 0x2
; GFX12-GISEL-NEXT: v_cmp_eq_u32_e64 s0, 0xac0, v2
; GFX12-GISEL-NEXT: s_wait_samplecnt 0x1
; GFX12-GISEL-NEXT: v_cmp_eq_f32_e32 vcc_lo, 1.0, v1
; GFX12-GISEL-NEXT: s_wait_loadcnt 0x1
; GFX12-GISEL-NEXT: v_cmp_eq_u32_e64 s1, 0xac0, v3
; GFX12-GISEL-NEXT: s_wait_loadcnt 0x0
; GFX12-GISEL-NEXT: v_cmp_eq_u32_e64 s2, 0xac0, v4
; GFX12-GISEL-NEXT: s_and_b32 s0, s0, vcc_lo
; GFX12-GISEL-NEXT: s_wait_samplecnt 0x0
; GFX12-GISEL-NEXT: v_cmp_eq_f32_e32 vcc_lo, 0, v0
; GFX12-GISEL-NEXT: s_and_b32 s0, s0, s1
; GFX12-GISEL-NEXT: s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(SALU_CYCLE_1)
; GFX12-GISEL-NEXT: s_and_b32 s0, s0, s2
; GFX12-GISEL-NEXT: s_and_b32 s0, s0, vcc_lo
; GFX12-GISEL-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
; GFX12-GISEL-NEXT: v_cndmask_b32_e64 v0, 0, 1, s0
; GFX12-GISEL-NEXT: buffer_store_b32 v0, off, s[24:27], null
; GFX12-GISEL-NEXT: s_endpgm
.entry:
%i = call i64 @llvm.amdgcn.s.getpc()
%extelt.offset = lshr i64 %i, 32
%.i1 = trunc i64 %extelt.offset to i32
%.upto0 = insertelement <2 x i32> poison, i32 %descTable1, i64 0
%i1 = insertelement <2 x i32> %.upto0, i32 %.i1, i64 1
%i2 = bitcast <2 x i32> %i1 to i64
%i3 = inttoptr i64 %i2 to ptr addrspace(4)
%.upto03 = insertelement <2 x i32> poison, i32 %descTable0, i64 0
%i4 = insertelement <2 x i32> %.upto03, i32 %.i1, i64 1
%i5 = bitcast <2 x i32> %i4 to i64
%i6 = inttoptr i64 %i5 to ptr addrspace(4)
%i7 = getelementptr i8, ptr addrspace(4) %i6, i64 80
%i8 = load <4 x i32>, ptr addrspace(4) %i7, align 16
%i9 = getelementptr i8, ptr addrspace(4) %i3, i64 48
%i10 = load <4 x i32>, ptr addrspace(4) %i9, align 16
%i11 = getelementptr i8, ptr addrspace(4) %i6, i64 64
%i12 = load <4 x i32>, ptr addrspace(4) %i11, align 16
%i13 = getelementptr i8, ptr addrspace(4) %i6, i64 16
%i14 = load <4 x i32>, ptr addrspace(4) %i13, align 16
%i15 = getelementptr i8, ptr addrspace(4) %i6, i64 32
%i16 = load <8 x i32>, ptr addrspace(4) %i15, align 32
%i17 = load <4 x i32>, ptr addrspace(4) %i6, align 16
%i18 = call float @llvm.amdgcn.image.sample.lz.2d.f32.f16.v8i32.v4i32(i32 1, half 0xHBC00, half 0xHBC00, <8 x i32> %i16, <4 x i32> %i17, i1 false, i32 0, i32 0)
%i19 = fcmp oeq float %i18, 0.000000e+00
%i20 = call i32 @llvm.amdgcn.raw.buffer.load.i32(<4 x i32> %i14, i32 0, i32 0, i32 0)
%.not = icmp eq i32 %i20, 2752
%i21 = call i32 @llvm.amdgcn.raw.buffer.load.i32(<4 x i32> %i12, i32 0, i32 0, i32 0)
%.not1 = icmp eq i32 %i21, 2752
%i22 = getelementptr i8, ptr addrspace(4) %i3, i64 16
%i23 = load <8 x i32>, ptr addrspace(4) %i22, align 32
%i24 = load <4 x i32>, ptr addrspace(4) %i3, align 16
%i25 = call float @llvm.amdgcn.image.sample.lz.2d.f32.f16.v8i32.v4i32(i32 1, half 0xHBC00, half 0xHBC00, <8 x i32> %i23, <4 x i32> %i24, i1 false, i32 0, i32 0)
%i26 = fcmp oeq float %i25, 1.000000e+00
%i27 = call i32 @llvm.amdgcn.raw.buffer.load.i32(<4 x i32> %i10, i32 0, i32 0, i32 0)
%.not2 = icmp eq i32 %i27, 2752
%i28 = select i1 %.not2, i1 %i26, i1 false
%i29 = select i1 %i28, i1 %.not1, i1 false
%i30 = select i1 %i29, i1 %.not, i1 false
%narrow2 = select i1 %i30, i1 %i19, i1 false
%.4 = zext i1 %narrow2 to i32
call void @llvm.amdgcn.raw.buffer.store.i32(i32 %.4, <4 x i32> %i8, i32 0, i32 0, i32 0)
ret void
}
|
