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
|
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc -mtriple=amdgcn < %s | FileCheck %s
; RUN: llc -mtriple=amdgcn -early-live-intervals < %s | FileCheck %s
; We may have subregister live ranges that are undefined on some paths. The
; verifier should not complain about this.
define amdgpu_kernel void @func() #0 {
; CHECK-LABEL: func:
; CHECK: ; %bb.0: ; %B0
; CHECK-NEXT: s_cmp_lg_u32 s8, 0
; CHECK-NEXT: s_mov_b32 s0, 0
; CHECK-NEXT: s_cbranch_scc1 .LBB0_2
; CHECK-NEXT: ; %bb.1: ; %B30.1
; CHECK-NEXT: s_mov_b32 s0, 0x7fc00000
; CHECK-NEXT: .LBB0_2: ; %B30.2
; CHECK-NEXT: v_mov_b32_e32 v0, s0
; CHECK-NEXT: s_mov_b32 m0, -1
; CHECK-NEXT: ds_write_b32 v0, v0
; CHECK-NEXT: s_endpgm
B0:
%id = call i32 @llvm.amdgcn.workgroup.id.x()
%cmp = icmp eq i32 %id, 0
br i1 %cmp, label %B1, label %B2
B1:
br label %B2
B2:
%v0 = phi <4 x float> [ zeroinitializer, %B1 ], [ <float 0.0, float 0.0, float 0.0, float poison>, %B0 ]
br i1 %cmp, label %B30.1, label %B30.2
B30.1:
%sub = fsub <4 x float> %v0, splat (float 0x7FF8000000000000)
br label %B30.2
B30.2:
%v3 = phi <4 x float> [ %sub, %B30.1 ], [ %v0, %B2 ]
%ve0 = extractelement <4 x float> %v3, i32 0
store float %ve0, ptr addrspace(3) poison, align 4
ret void
}
; FIXME: Extra undef subregister copy should be removed before
; overwritten with defined copy
define amdgpu_ps float @valley_partially_undef_copy() #0 {
; CHECK-LABEL: valley_partially_undef_copy:
; CHECK: ; %bb.0: ; %bb
; CHECK-NEXT: s_mov_b32 s3, 0xf000
; CHECK-NEXT: s_mov_b32 s2, -1
; CHECK-NEXT: buffer_load_dword v1, off, s[0:3], 0 glc
; CHECK-NEXT: s_waitcnt vmcnt(0)
; CHECK-NEXT: buffer_load_dword v0, off, s[0:3], 0 glc
; CHECK-NEXT: s_waitcnt vmcnt(0)
; CHECK-NEXT: v_mov_b32_e32 v2, 0x7fc00000
; CHECK-NEXT: buffer_store_dwordx4 v[0:3], off, s[0:3], 0
; CHECK-NEXT: buffer_store_dword v2, off, s[0:3], 0
; CHECK-NEXT: v_cmp_ne_u32_e32 vcc, 0, v1
; CHECK-NEXT: s_waitcnt expcnt(1)
; CHECK-NEXT: v_cndmask_b32_e64 v1, 0, 1, vcc
; CHECK-NEXT: v_cmp_ne_u32_e64 s[0:1], 1, v1
; CHECK-NEXT: .LBB1_1: ; %bb9
; CHECK-NEXT: ; =>This Inner Loop Header: Depth=1
; CHECK-NEXT: s_and_b64 vcc, exec, s[0:1]
; CHECK-NEXT: s_cbranch_vccnz .LBB1_1
; CHECK-NEXT: ; %bb.2: ; %bb11
; CHECK-NEXT: s_mov_b32 s3, 0xf000
; CHECK-NEXT: s_mov_b32 s2, -1
; CHECK-NEXT: buffer_store_dwordx4 v[0:3], off, s[0:3], 0
; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0)
; CHECK-NEXT: ; return to shader part epilog
bb:
%tmp = load volatile i32, ptr addrspace(1) poison, align 4
%tmp1 = load volatile i32, ptr addrspace(1) poison, align 4
%tmp2 = insertelement <4 x i32> poison, i32 %tmp1, i32 0
%tmp3 = bitcast i32 %tmp1 to float
%tmp4 = call <4 x float> @llvm.amdgcn.image.sample.2d.v4f32.f32(i32 15, float %tmp3, float %tmp3, <8 x i32> poison, <4 x i32> poison, i1 0, i32 0, i32 0)
%tmp5 = extractelement <4 x float> %tmp4, i32 0
%tmp6 = fmul float %tmp5, 0x7FF8000000000000
%tmp7 = fadd float %tmp6, %tmp6
%tmp8 = insertelement <4 x i32> %tmp2, i32 %tmp, i32 1
store <4 x i32> %tmp8, ptr addrspace(1) poison, align 16
store float %tmp7, ptr addrspace(1) poison, align 4
br label %bb9
bb9: ; preds = %bb9, %bb
%tmp10 = icmp eq i32 %tmp, 0
br i1 %tmp10, label %bb9, label %bb11
bb11: ; preds = %bb9
store <4 x i32> %tmp2, ptr addrspace(1) poison, align 16
ret float poison
}
; FIXME: Should be able to remove the undef copies
define amdgpu_kernel void @partially_undef_copy() #0 {
; CHECK-LABEL: partially_undef_copy:
; CHECK: ; %bb.0:
; CHECK-NEXT: ;;#ASMSTART
; CHECK-NEXT: v_mov_b32_e32 v5, 5
; CHECK-NEXT: ;;#ASMEND
; CHECK-NEXT: ;;#ASMSTART
; CHECK-NEXT: v_mov_b32_e32 v6, 6
; CHECK-NEXT: ;;#ASMEND
; CHECK-NEXT: v_mov_b32_e32 v0, v5
; CHECK-NEXT: v_mov_b32_e32 v1, v6
; CHECK-NEXT: v_mov_b32_e32 v2, v7
; CHECK-NEXT: v_mov_b32_e32 v3, v8
; CHECK-NEXT: s_mov_b32 s3, 0xf000
; CHECK-NEXT: s_mov_b32 s2, -1
; CHECK-NEXT: v_mov_b32_e32 v0, v6
; CHECK-NEXT: buffer_store_dwordx4 v[0:3], off, s[0:3], 0
; CHECK-NEXT: s_waitcnt vmcnt(0)
; CHECK-NEXT: ;;#ASMSTART
; CHECK-NEXT: v_nop
; CHECK-NEXT: ;;#ASMEND
; CHECK-NEXT: s_endpgm
%tmp0 = call i32 asm sideeffect "v_mov_b32_e32 v5, 5", "={v5}"()
%tmp1 = call i32 asm sideeffect "v_mov_b32_e32 v6, 6", "={v6}"()
%partially.undef.0 = insertelement <4 x i32> poison, i32 %tmp0, i32 0
%partially.undef.1 = insertelement <4 x i32> %partially.undef.0, i32 %tmp1, i32 0
store volatile <4 x i32> %partially.undef.1, ptr addrspace(1) poison, align 16
tail call void asm sideeffect "v_nop", "v={v[5:8]}"(<4 x i32> %partially.undef.0)
ret void
}
declare <4 x float> @llvm.amdgcn.image.sample.2d.v4f32.f32(i32, float, float, <8 x i32>, <4 x i32>, i1, i32, i32) #1
attributes #0 = { nounwind }
attributes #1 = { nounwind readonly }
|
