annotate test/CodeGen/AMDGPU/madmk.ll @ 147:c2174574ed3a

LLVM 10
author Shinji KONO <kono@ie.u-ryukyu.ac.jp>
date Wed, 14 Aug 2019 16:55:33 +0900
parents 803732b1fca8
children
Ignore whitespace changes - Everywhere: Within whitespace: At end of lines:
rev   line source
95
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
1 ; RUN: llc -march=amdgcn -verify-machineinstrs < %s | FileCheck -check-prefix=GCN -check-prefix=SI %s
121
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
2 ; XUN: llc -march=amdgcn -mcpu=tonga -mattr=-flat-for-global -verify-machineinstrs < %s | FileCheck -check-prefix=GCN -check-prefix=VI %s
95
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
3
120
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
4 ; FIXME: None of these trigger madmk emission anymore. It is still
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
5 ; possible, but requires the correct registers to be used which is
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
6 ; hard to trigger.
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
7
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
8 declare i32 @llvm.amdgcn.workitem.id.x() nounwind readnone
95
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
9 declare float @llvm.fabs.f32(float) nounwind readnone
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
10
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
11 ; GCN-LABEL: {{^}}madmk_f32:
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
12 ; GCN-DAG: buffer_load_dword [[VA:v[0-9]+]], {{v\[[0-9]+:[0-9]+\]}}, {{s\[[0-9]+:[0-9]+\]}}, 0 addr64{{$}}
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
13 ; GCN-DAG: buffer_load_dword [[VB:v[0-9]+]], {{v\[[0-9]+:[0-9]+\]}}, {{s\[[0-9]+:[0-9]+\]}}, 0 addr64 offset:4
120
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
14 ; GCN: v_mac_f32_e32 [[VB]], 0x41200000, [[VA]]
121
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
15 define amdgpu_kernel void @madmk_f32(float addrspace(1)* noalias %out, float addrspace(1)* noalias %in) nounwind {
120
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
16 %tid = tail call i32 @llvm.amdgcn.workitem.id.x() nounwind readnone
95
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
17 %gep.0 = getelementptr float, float addrspace(1)* %in, i32 %tid
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
18 %gep.1 = getelementptr float, float addrspace(1)* %gep.0, i32 1
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
19 %out.gep = getelementptr float, float addrspace(1)* %out, i32 %tid
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
20
120
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
21 %a = load volatile float, float addrspace(1)* %gep.0, align 4
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
22 %b = load volatile float, float addrspace(1)* %gep.1, align 4
95
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
23
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
24 %mul = fmul float %a, 10.0
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
25 %madmk = fadd float %mul, %b
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
26 store float %madmk, float addrspace(1)* %out.gep, align 4
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
27 ret void
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
28 }
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
29
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
30 ; GCN-LABEL: {{^}}madmk_2_use_f32:
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
31 ; GCN-DAG: buffer_load_dword [[VA:v[0-9]+]], {{v\[[0-9]+:[0-9]+\]}}, {{s\[[0-9]+:[0-9]+\]}}, 0 addr64{{$}}
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
32 ; GCN-DAG: buffer_load_dword [[VB:v[0-9]+]], {{v\[[0-9]+:[0-9]+\]}}, {{s\[[0-9]+:[0-9]+\]}}, 0 addr64 offset:4
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
33 ; GCN-DAG: buffer_load_dword [[VC:v[0-9]+]], {{v\[[0-9]+:[0-9]+\]}}, {{s\[[0-9]+:[0-9]+\]}}, 0 addr64 offset:8
147
c2174574ed3a LLVM 10
Shinji KONO <kono@ie.u-ryukyu.ac.jp>
parents: 121
diff changeset
34 ; GCN-DAG: s_mov_b32 [[SK:s[0-9]+]], 0x41200000
c2174574ed3a LLVM 10
Shinji KONO <kono@ie.u-ryukyu.ac.jp>
parents: 121
diff changeset
35 ; GCN-DAG: v_mac_f32_e32 [[VB]], [[SK]], [[VA]]
c2174574ed3a LLVM 10
Shinji KONO <kono@ie.u-ryukyu.ac.jp>
parents: 121
diff changeset
36 ; GCN-DAG: v_mac_f32_e32 [[VC]], [[SK]], [[VA]]
95
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
37 ; GCN: s_endpgm
121
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
38 define amdgpu_kernel void @madmk_2_use_f32(float addrspace(1)* noalias %out, float addrspace(1)* noalias %in) nounwind {
120
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
39 %tid = tail call i32 @llvm.amdgcn.workitem.id.x() nounwind readnone
95
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
40
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
41 %in.gep.0 = getelementptr float, float addrspace(1)* %in, i32 %tid
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
42 %in.gep.1 = getelementptr float, float addrspace(1)* %in.gep.0, i32 1
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
43 %in.gep.2 = getelementptr float, float addrspace(1)* %in.gep.0, i32 2
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
44
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
45 %out.gep.0 = getelementptr float, float addrspace(1)* %out, i32 %tid
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
46 %out.gep.1 = getelementptr float, float addrspace(1)* %in.gep.0, i32 1
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
47
120
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
48 %a = load volatile float, float addrspace(1)* %in.gep.0, align 4
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
49 %b = load volatile float, float addrspace(1)* %in.gep.1, align 4
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
50 %c = load volatile float, float addrspace(1)* %in.gep.2, align 4
95
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
51
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
52 %mul0 = fmul float %a, 10.0
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
53 %mul1 = fmul float %a, 10.0
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
54 %madmk0 = fadd float %mul0, %b
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
55 %madmk1 = fadd float %mul1, %c
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
56
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
57 store float %madmk0, float addrspace(1)* %out.gep.0, align 4
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
58 store float %madmk1, float addrspace(1)* %out.gep.1, align 4
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
59 ret void
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
60 }
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
61
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
62 ; We don't get any benefit if the constant is an inline immediate.
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
63 ; GCN-LABEL: {{^}}madmk_inline_imm_f32:
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
64 ; GCN-DAG: buffer_load_dword [[VA:v[0-9]+]], {{v\[[0-9]+:[0-9]+\]}}, {{s\[[0-9]+:[0-9]+\]}}, 0 addr64{{$}}
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
65 ; GCN-DAG: buffer_load_dword [[VB:v[0-9]+]], {{v\[[0-9]+:[0-9]+\]}}, {{s\[[0-9]+:[0-9]+\]}}, 0 addr64 offset:4
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
66 ; GCN: v_mac_f32_e32 [[VB]], 4.0, [[VA]]
121
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
67 define amdgpu_kernel void @madmk_inline_imm_f32(float addrspace(1)* noalias %out, float addrspace(1)* noalias %in) nounwind {
120
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
68 %tid = tail call i32 @llvm.amdgcn.workitem.id.x() nounwind readnone
95
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
69 %gep.0 = getelementptr float, float addrspace(1)* %in, i32 %tid
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
70 %gep.1 = getelementptr float, float addrspace(1)* %gep.0, i32 1
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
71 %out.gep = getelementptr float, float addrspace(1)* %out, i32 %tid
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
72
120
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
73 %a = load volatile float, float addrspace(1)* %gep.0, align 4
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
74 %b = load volatile float, float addrspace(1)* %gep.1, align 4
95
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
75
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
76 %mul = fmul float %a, 4.0
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
77 %madmk = fadd float %mul, %b
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
78 store float %madmk, float addrspace(1)* %out.gep, align 4
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
79 ret void
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
80 }
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
81
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
82 ; GCN-LABEL: {{^}}s_s_madmk_f32:
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
83 ; GCN-NOT: v_madmk_f32
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
84 ; GCN: v_mac_f32_e32
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
85 ; GCN: s_endpgm
147
c2174574ed3a LLVM 10
Shinji KONO <kono@ie.u-ryukyu.ac.jp>
parents: 121
diff changeset
86 define amdgpu_kernel void @s_s_madmk_f32(float addrspace(1)* noalias %out, [8 x i32], float %a, [8 x i32], float %b) nounwind {
120
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
87 %tid = tail call i32 @llvm.amdgcn.workitem.id.x() nounwind readnone
95
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
88 %out.gep = getelementptr float, float addrspace(1)* %out, i32 %tid
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
89
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
90 %mul = fmul float %a, 10.0
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
91 %madmk = fadd float %mul, %b
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
92 store float %madmk, float addrspace(1)* %out.gep, align 4
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
93 ret void
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
94 }
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
95
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
96 ; GCN-LABEL: {{^}}v_s_madmk_f32:
147
c2174574ed3a LLVM 10
Shinji KONO <kono@ie.u-ryukyu.ac.jp>
parents: 121
diff changeset
97 ; GCN: s_load_dword [[SREG:s[0-9]+]]
c2174574ed3a LLVM 10
Shinji KONO <kono@ie.u-ryukyu.ac.jp>
parents: 121
diff changeset
98 ; GCN: buffer_load_dword [[VREG1:v[0-9]+]]
c2174574ed3a LLVM 10
Shinji KONO <kono@ie.u-ryukyu.ac.jp>
parents: 121
diff changeset
99 ; GCN: v_mov_b32_e32 [[VREG2:v[0-9]+]], [[SREG]]
c2174574ed3a LLVM 10
Shinji KONO <kono@ie.u-ryukyu.ac.jp>
parents: 121
diff changeset
100 ; GCN: v_mac_f32_e32 [[VREG2]], 0x41200000, [[VREG1]]
95
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
101 ; GCN: s_endpgm
121
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
102 define amdgpu_kernel void @v_s_madmk_f32(float addrspace(1)* noalias %out, float addrspace(1)* noalias %in, float %b) nounwind {
120
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
103 %tid = tail call i32 @llvm.amdgcn.workitem.id.x() nounwind readnone
95
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
104 %gep.0 = getelementptr float, float addrspace(1)* %in, i32 %tid
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
105 %out.gep = getelementptr float, float addrspace(1)* %out, i32 %tid
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
106 %a = load float, float addrspace(1)* %gep.0, align 4
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
107
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
108 %mul = fmul float %a, 10.0
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
109 %madmk = fadd float %mul, %b
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
110 store float %madmk, float addrspace(1)* %out.gep, align 4
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
111 ret void
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
112 }
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
113
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
114 ; GCN-LABEL: {{^}}scalar_vector_madmk_f32:
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
115 ; GCN-NOT: v_madmk_f32
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
116 ; GCN: v_mac_f32_e32
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
117 ; GCN: s_endpgm
121
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
118 define amdgpu_kernel void @scalar_vector_madmk_f32(float addrspace(1)* noalias %out, float addrspace(1)* noalias %in, float %a) nounwind {
120
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
119 %tid = tail call i32 @llvm.amdgcn.workitem.id.x() nounwind readnone
95
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
120 %gep.0 = getelementptr float, float addrspace(1)* %in, i32 %tid
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
121 %out.gep = getelementptr float, float addrspace(1)* %out, i32 %tid
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
122 %b = load float, float addrspace(1)* %gep.0, align 4
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
123
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
124 %mul = fmul float %a, 10.0
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
125 %madmk = fadd float %mul, %b
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
126 store float %madmk, float addrspace(1)* %out.gep, align 4
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
127 ret void
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
128 }
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
129
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
130 ; GCN-LABEL: {{^}}no_madmk_src0_modifier_f32:
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
131 ; GCN-DAG: buffer_load_dword [[VA:v[0-9]+]], {{v\[[0-9]+:[0-9]+\]}}, {{s\[[0-9]+:[0-9]+\]}}, 0 addr64{{$}}
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
132 ; GCN-DAG: buffer_load_dword [[VB:v[0-9]+]], {{v\[[0-9]+:[0-9]+\]}}, {{s\[[0-9]+:[0-9]+\]}}, 0 addr64 offset:4
147
c2174574ed3a LLVM 10
Shinji KONO <kono@ie.u-ryukyu.ac.jp>
parents: 121
diff changeset
133 ; GCN-DAG: s_mov_b32 [[SK:s[0-9]+]], 0x41200000
c2174574ed3a LLVM 10
Shinji KONO <kono@ie.u-ryukyu.ac.jp>
parents: 121
diff changeset
134 ; GCN: v_mad_f32 {{v[0-9]+}}, |[[VA]]|, [[SK]], [[VB]]
121
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
135 define amdgpu_kernel void @no_madmk_src0_modifier_f32(float addrspace(1)* noalias %out, float addrspace(1)* noalias %in) nounwind {
120
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
136 %tid = tail call i32 @llvm.amdgcn.workitem.id.x() nounwind readnone
95
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
137 %gep.0 = getelementptr float, float addrspace(1)* %in, i32 %tid
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
138 %gep.1 = getelementptr float, float addrspace(1)* %gep.0, i32 1
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
139 %out.gep = getelementptr float, float addrspace(1)* %out, i32 %tid
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
140
120
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
141 %a = load volatile float, float addrspace(1)* %gep.0, align 4
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
142 %b = load volatile float, float addrspace(1)* %gep.1, align 4
95
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
143
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
144 %a.fabs = call float @llvm.fabs.f32(float %a) nounwind readnone
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
145
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
146 %mul = fmul float %a.fabs, 10.0
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
147 %madmk = fadd float %mul, %b
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
148 store float %madmk, float addrspace(1)* %out.gep, align 4
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
149 ret void
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
150 }
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
151
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
152 ; GCN-LABEL: {{^}}no_madmk_src2_modifier_f32:
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
153 ; GCN-DAG: buffer_load_dword [[VA:v[0-9]+]], {{v\[[0-9]+:[0-9]+\]}}, {{s\[[0-9]+:[0-9]+\]}}, 0 addr64{{$}}
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
154 ; GCN-DAG: buffer_load_dword [[VB:v[0-9]+]], {{v\[[0-9]+:[0-9]+\]}}, {{s\[[0-9]+:[0-9]+\]}}, 0 addr64 offset:4
147
c2174574ed3a LLVM 10
Shinji KONO <kono@ie.u-ryukyu.ac.jp>
parents: 121
diff changeset
155 ; GCN: v_mad_f32 {{v[0-9]+}}, {{v[0-9]+}}, {{[sv][0-9]+}}, |{{v[0-9]+}}|
121
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
156 define amdgpu_kernel void @no_madmk_src2_modifier_f32(float addrspace(1)* noalias %out, float addrspace(1)* noalias %in) nounwind {
120
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
157 %tid = tail call i32 @llvm.amdgcn.workitem.id.x() nounwind readnone
95
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
158 %gep.0 = getelementptr float, float addrspace(1)* %in, i32 %tid
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
159 %gep.1 = getelementptr float, float addrspace(1)* %gep.0, i32 1
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
160 %out.gep = getelementptr float, float addrspace(1)* %out, i32 %tid
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
161
120
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
162 %a = load volatile float, float addrspace(1)* %gep.0, align 4
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
163 %b = load volatile float, float addrspace(1)* %gep.1, align 4
95
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
164
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
165 %b.fabs = call float @llvm.fabs.f32(float %b) nounwind readnone
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
166
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
167 %mul = fmul float %a, 10.0
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
168 %madmk = fadd float %mul, %b.fabs
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
169 store float %madmk, float addrspace(1)* %out.gep, align 4
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
170 ret void
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
171 }
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
172
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
173 ; GCN-LABEL: {{^}}madmk_add_inline_imm_f32:
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
174 ; GCN: buffer_load_dword [[A:v[0-9]+]]
147
c2174574ed3a LLVM 10
Shinji KONO <kono@ie.u-ryukyu.ac.jp>
parents: 121
diff changeset
175 ; GCN: s_mov_b32 [[SK:s[0-9]+]], 0x41200000
c2174574ed3a LLVM 10
Shinji KONO <kono@ie.u-ryukyu.ac.jp>
parents: 121
diff changeset
176 ; GCN: v_mad_f32 {{v[0-9]+}}, [[A]], [[SK]], 2.0
121
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
177 define amdgpu_kernel void @madmk_add_inline_imm_f32(float addrspace(1)* noalias %out, float addrspace(1)* noalias %in) nounwind {
120
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
178 %tid = tail call i32 @llvm.amdgcn.workitem.id.x() nounwind readnone
95
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
179 %gep.0 = getelementptr float, float addrspace(1)* %in, i32 %tid
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
180 %out.gep = getelementptr float, float addrspace(1)* %out, i32 %tid
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
181
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
182 %a = load float, float addrspace(1)* %gep.0, align 4
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
183
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
184 %mul = fmul float %a, 10.0
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
185 %madmk = fadd float %mul, 2.0
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
186 store float %madmk, float addrspace(1)* %out.gep, align 4
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
187 ret void
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
188 }
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
189
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
190 ; SI-LABEL: {{^}}kill_madmk_verifier_error:
147
c2174574ed3a LLVM 10
Shinji KONO <kono@ie.u-ryukyu.ac.jp>
parents: 121
diff changeset
191 ; SI: s_or_b64
95
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
192 ; SI: s_xor_b64
120
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
193 ; SI: v_mac_f32_e32 {{v[0-9]+}}, 0x472aee8c, {{v[0-9]+}}
121
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
194 define amdgpu_kernel void @kill_madmk_verifier_error() nounwind {
95
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
195 bb:
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
196 br label %bb2
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
197
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
198 bb1: ; preds = %bb2
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
199 ret void
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
200
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
201 bb2: ; preds = %bb6, %bb
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
202 %tmp = phi float [ undef, %bb ], [ %tmp8, %bb6 ]
120
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
203 %tid = call i32 @llvm.amdgcn.mbcnt.lo(i32 -1, i32 0) #1
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
204 %f_tid = bitcast i32 %tid to float
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
205 %tmp3 = fsub float %f_tid, %tmp
95
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
206 %tmp5 = fcmp oeq float %tmp3, 1.000000e+04
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
207 br i1 %tmp5, label %bb1, label %bb6
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
208
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
209 bb6: ; preds = %bb2
147
c2174574ed3a LLVM 10
Shinji KONO <kono@ie.u-ryukyu.ac.jp>
parents: 121
diff changeset
210 %tmp7 = fmul float %tmp, 0x40E55DD180000000
c2174574ed3a LLVM 10
Shinji KONO <kono@ie.u-ryukyu.ac.jp>
parents: 121
diff changeset
211 %tmp8 = fadd float %tmp7, %tmp
95
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
212 br label %bb2
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
213 }
120
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
214
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
215 declare i32 @llvm.amdgcn.mbcnt.lo(i32, i32) #1
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
216
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
217 attributes #1 = { nounwind readnone }