llvm.amdgcn.wwm.ll
3.34 KB
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
; NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
; RUN: llc -global-isel -march=amdgcn -mcpu=hawaii -stop-after=instruction-select -verify-machineinstrs < %s | FileCheck -check-prefix=GCN %s
define amdgpu_ps float @wwm_f32(float %val) {
; GCN-LABEL: name: wwm_f32
; GCN: bb.1 (%ir-block.0):
; GCN: liveins: $vgpr0
; GCN: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GCN: [[WWM:%[0-9]+]]:vgpr_32 = WWM [[COPY]], implicit $exec
; GCN: $vgpr0 = COPY [[WWM]]
; GCN: SI_RETURN_TO_EPILOG implicit $vgpr0
%ret = call float @llvm.amdgcn.wwm.f32(float %val)
ret float %ret
}
define amdgpu_ps float @wwm_v2f16(float %arg) {
; GCN-LABEL: name: wwm_v2f16
; GCN: bb.1 (%ir-block.0):
; GCN: liveins: $vgpr0
; GCN: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GCN: [[WWM:%[0-9]+]]:vgpr_32 = WWM [[COPY]], implicit $exec
; GCN: $vgpr0 = COPY [[WWM]]
; GCN: SI_RETURN_TO_EPILOG implicit $vgpr0
%val = bitcast float %arg to <2 x half>
%ret = call <2 x half> @llvm.amdgcn.wwm.v2f16(<2 x half> %val)
%bc = bitcast <2 x half> %ret to float
ret float %bc
}
define amdgpu_ps <2 x float> @wwm_f64(double %val) {
; GCN-LABEL: name: wwm_f64
; GCN: bb.1 (%ir-block.0):
; GCN: liveins: $vgpr0, $vgpr1
; GCN: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GCN: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
; GCN: [[REG_SEQUENCE:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1
; GCN: [[WWM:%[0-9]+]]:vreg_64 = WWM [[REG_SEQUENCE]], implicit $exec
; GCN: [[COPY2:%[0-9]+]]:vgpr_32 = COPY [[WWM]].sub0
; GCN: [[COPY3:%[0-9]+]]:vgpr_32 = COPY [[WWM]].sub1
; GCN: $vgpr0 = COPY [[COPY2]]
; GCN: $vgpr1 = COPY [[COPY3]]
; GCN: SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1
%ret = call double @llvm.amdgcn.wwm.f64(double %val)
%bitcast = bitcast double %ret to <2 x float>
ret <2 x float> %bitcast
}
; TODO
; define amdgpu_ps float @wwm_i1_vcc(float %val) {
; %vcc = fcmp oeq float %val, 0.0
; %ret = call i1 @llvm.amdgcn.wwm.i1(i1 %vcc)
; %select = select i1 %ret, float 1.0, float 0.0
; ret float %select
; }
define amdgpu_ps <3 x float> @wwm_v3f32(<3 x float> %val) {
; GCN-LABEL: name: wwm_v3f32
; GCN: bb.1 (%ir-block.0):
; GCN: liveins: $vgpr0, $vgpr1, $vgpr2
; GCN: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GCN: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
; GCN: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr2
; GCN: [[REG_SEQUENCE:%[0-9]+]]:vreg_96 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2
; GCN: [[WWM:%[0-9]+]]:vreg_96 = WWM [[REG_SEQUENCE]], implicit $exec
; GCN: [[COPY3:%[0-9]+]]:vgpr_32 = COPY [[WWM]].sub0
; GCN: [[COPY4:%[0-9]+]]:vgpr_32 = COPY [[WWM]].sub1
; GCN: [[COPY5:%[0-9]+]]:vgpr_32 = COPY [[WWM]].sub2
; GCN: $vgpr0 = COPY [[COPY3]]
; GCN: $vgpr1 = COPY [[COPY4]]
; GCN: $vgpr2 = COPY [[COPY5]]
; GCN: SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2
%ret = call <3 x float> @llvm.amdgcn.wwm.v3f32(<3 x float> %val)
ret <3 x float> %ret
}
declare i1 @llvm.amdgcn.wwm.i1(i1) #0
declare float @llvm.amdgcn.wwm.f32(float) #0
declare <2 x half> @llvm.amdgcn.wwm.v2f16(<2 x half>) #0
declare <3 x float> @llvm.amdgcn.wwm.v3f32(<3 x float>) #0
declare double @llvm.amdgcn.wwm.f64(double) #0
attributes #0 = { nounwind readnone speculatable }