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
|
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc -mtriple=amdgcn -mcpu=gfx1100 -mattr=+real-true16 -verify-machineinstrs < %s | FileCheck %s --check-prefixes=GFX11,SDAG-GFX11,SDAG-GFX11-TRUE16
; RUN: llc -mtriple=amdgcn -mcpu=gfx1100 -mattr=-real-true16 -verify-machineinstrs < %s | FileCheck %s --check-prefixes=GFX11,SDAG-GFX11,SDAG-GFX11-FAKE16
; RUN: llc -global-isel -mtriple=amdgcn -mcpu=gfx1100 -mattr=+real-true16 -verify-machineinstrs < %s | FileCheck %s --check-prefixes=GFX11,GISEL-GFX11,GISEL-GFX11-TRUE16
; RUN: llc -global-isel -mtriple=amdgcn -mcpu=gfx1100 -mattr=-real-true16 -verify-machineinstrs < %s | FileCheck %s --check-prefixes=GFX11,GISEL-GFX11,GISEL-GFX11-FAKE16
declare half @llvm.amdgcn.fdot2.f16.f16(<2 x half> %a, <2 x half> %b, half %c)
define amdgpu_kernel void @test_llvm_amdgcn_fdot2_f16_f16(
; SDAG-GFX11-TRUE16-LABEL: test_llvm_amdgcn_fdot2_f16_f16:
; SDAG-GFX11-TRUE16: ; %bb.0: ; %entry
; SDAG-GFX11-TRUE16-NEXT: s_load_b256 s[0:7], s[4:5], 0x24
; SDAG-GFX11-TRUE16-NEXT: v_mov_b32_e32 v1, 0
; SDAG-GFX11-TRUE16-NEXT: s_waitcnt lgkmcnt(0)
; SDAG-GFX11-TRUE16-NEXT: global_load_d16_b16 v0, v1, s[6:7]
; SDAG-GFX11-TRUE16-NEXT: s_load_b32 s2, s[2:3], 0x0
; SDAG-GFX11-TRUE16-NEXT: s_load_b32 s3, s[4:5], 0x0
; SDAG-GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
; SDAG-GFX11-TRUE16-NEXT: v_dot2_f16_f16 v0.l, s2, s3, v0.l
; SDAG-GFX11-TRUE16-NEXT: global_store_b16 v1, v0, s[0:1]
; SDAG-GFX11-TRUE16-NEXT: s_endpgm
;
; SDAG-GFX11-FAKE16-LABEL: test_llvm_amdgcn_fdot2_f16_f16:
; SDAG-GFX11-FAKE16: ; %bb.0: ; %entry
; SDAG-GFX11-FAKE16-NEXT: s_load_b256 s[0:7], s[4:5], 0x24
; SDAG-GFX11-FAKE16-NEXT: v_mov_b32_e32 v0, 0
; SDAG-GFX11-FAKE16-NEXT: s_waitcnt lgkmcnt(0)
; SDAG-GFX11-FAKE16-NEXT: global_load_u16 v1, v0, s[6:7]
; SDAG-GFX11-FAKE16-NEXT: s_load_b32 s2, s[2:3], 0x0
; SDAG-GFX11-FAKE16-NEXT: s_load_b32 s3, s[4:5], 0x0
; SDAG-GFX11-FAKE16-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
; SDAG-GFX11-FAKE16-NEXT: v_dot2_f16_f16 v1, s2, s3, v1
; SDAG-GFX11-FAKE16-NEXT: global_store_b16 v0, v1, s[0:1]
; SDAG-GFX11-FAKE16-NEXT: s_endpgm
;
; GISEL-GFX11-TRUE16-LABEL: test_llvm_amdgcn_fdot2_f16_f16:
; GISEL-GFX11-TRUE16: ; %bb.0: ; %entry
; GISEL-GFX11-TRUE16-NEXT: s_load_b256 s[0:7], s[4:5], 0x24
; GISEL-GFX11-TRUE16-NEXT: v_mov_b32_e32 v1, 0
; GISEL-GFX11-TRUE16-NEXT: s_waitcnt lgkmcnt(0)
; GISEL-GFX11-TRUE16-NEXT: global_load_d16_b16 v0, v1, s[6:7]
; GISEL-GFX11-TRUE16-NEXT: s_load_b32 s2, s[2:3], 0x0
; GISEL-GFX11-TRUE16-NEXT: s_load_b32 s3, s[4:5], 0x0
; GISEL-GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
; GISEL-GFX11-TRUE16-NEXT: v_dot2_f16_f16 v0.l, s2, s3, v0.l
; GISEL-GFX11-TRUE16-NEXT: global_store_b16 v1, v0, s[0:1]
; GISEL-GFX11-TRUE16-NEXT: s_endpgm
;
; GISEL-GFX11-FAKE16-LABEL: test_llvm_amdgcn_fdot2_f16_f16:
; GISEL-GFX11-FAKE16: ; %bb.0: ; %entry
; GISEL-GFX11-FAKE16-NEXT: s_load_b256 s[0:7], s[4:5], 0x24
; GISEL-GFX11-FAKE16-NEXT: v_mov_b32_e32 v0, 0
; GISEL-GFX11-FAKE16-NEXT: s_waitcnt lgkmcnt(0)
; GISEL-GFX11-FAKE16-NEXT: global_load_u16 v1, v0, s[6:7]
; GISEL-GFX11-FAKE16-NEXT: s_load_b32 s2, s[2:3], 0x0
; GISEL-GFX11-FAKE16-NEXT: s_load_b32 s3, s[4:5], 0x0
; GISEL-GFX11-FAKE16-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
; GISEL-GFX11-FAKE16-NEXT: v_dot2_f16_f16 v1, s2, s3, v1
; GISEL-GFX11-FAKE16-NEXT: global_store_b16 v0, v1, s[0:1]
; GISEL-GFX11-FAKE16-NEXT: s_endpgm
ptr addrspace(1) %r,
ptr addrspace(1) %a,
ptr addrspace(1) %b,
ptr addrspace(1) %c) {
entry:
%a.val = load <2 x half>, ptr addrspace(1) %a
%b.val = load <2 x half>, ptr addrspace(1) %b
%c.val = load half, ptr addrspace(1) %c
%r.val = call half @llvm.amdgcn.fdot2.f16.f16(<2 x half> %a.val, <2 x half> %b.val, half %c.val)
store half %r.val, ptr addrspace(1) %r
ret void
}
define amdgpu_kernel void @test_llvm_amdgcn_fdot2_f16_f16_dpp(
; SDAG-GFX11-TRUE16-LABEL: test_llvm_amdgcn_fdot2_f16_f16_dpp:
; SDAG-GFX11-TRUE16: ; %bb.0: ; %entry
; SDAG-GFX11-TRUE16-NEXT: s_load_b128 s[0:3], s[4:5], 0x24
; SDAG-GFX11-TRUE16-NEXT: s_waitcnt lgkmcnt(0)
; SDAG-GFX11-TRUE16-NEXT: scratch_load_b32 v1, off, s1
; SDAG-GFX11-TRUE16-NEXT: scratch_load_b32 v2, off, s2
; SDAG-GFX11-TRUE16-NEXT: scratch_load_d16_b16 v0, off, s3
; SDAG-GFX11-TRUE16-NEXT: s_waitcnt vmcnt(2)
; SDAG-GFX11-TRUE16-NEXT: v_mov_b32_dpp v1, v1 quad_perm:[1,0,0,0] row_mask:0xf bank_mask:0xf bound_ctrl:1
; SDAG-GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0)
; SDAG-GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1)
; SDAG-GFX11-TRUE16-NEXT: v_dot2_f16_f16 v0.l, v1, v2, v0.l
; SDAG-GFX11-TRUE16-NEXT: scratch_store_b16 off, v0, s0
; SDAG-GFX11-TRUE16-NEXT: s_endpgm
;
; SDAG-GFX11-FAKE16-LABEL: test_llvm_amdgcn_fdot2_f16_f16_dpp:
; SDAG-GFX11-FAKE16: ; %bb.0: ; %entry
; SDAG-GFX11-FAKE16-NEXT: s_load_b128 s[0:3], s[4:5], 0x24
; SDAG-GFX11-FAKE16-NEXT: s_waitcnt lgkmcnt(0)
; SDAG-GFX11-FAKE16-NEXT: scratch_load_b32 v0, off, s2
; SDAG-GFX11-FAKE16-NEXT: scratch_load_u16 v1, off, s3
; SDAG-GFX11-FAKE16-NEXT: scratch_load_b32 v2, off, s1
; SDAG-GFX11-FAKE16-NEXT: s_waitcnt vmcnt(0)
; SDAG-GFX11-FAKE16-NEXT: v_dot2_f16_f16_e64_dpp v0, v2, v0, v1 quad_perm:[1,0,0,0] row_mask:0xf bank_mask:0xf bound_ctrl:1
; SDAG-GFX11-FAKE16-NEXT: scratch_store_b16 off, v0, s0
; SDAG-GFX11-FAKE16-NEXT: s_endpgm
;
; GISEL-GFX11-TRUE16-LABEL: test_llvm_amdgcn_fdot2_f16_f16_dpp:
; GISEL-GFX11-TRUE16: ; %bb.0: ; %entry
; GISEL-GFX11-TRUE16-NEXT: s_load_b128 s[0:3], s[4:5], 0x24
; GISEL-GFX11-TRUE16-NEXT: s_waitcnt lgkmcnt(0)
; GISEL-GFX11-TRUE16-NEXT: scratch_load_b32 v1, off, s1
; GISEL-GFX11-TRUE16-NEXT: scratch_load_b32 v2, off, s2
; GISEL-GFX11-TRUE16-NEXT: scratch_load_d16_b16 v0, off, s3
; GISEL-GFX11-TRUE16-NEXT: s_waitcnt vmcnt(2)
; GISEL-GFX11-TRUE16-NEXT: v_mov_b32_dpp v1, v1 quad_perm:[1,0,0,0] row_mask:0xf bank_mask:0xf bound_ctrl:1
; GISEL-GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0)
; GISEL-GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1)
; GISEL-GFX11-TRUE16-NEXT: v_dot2_f16_f16 v0.l, v1, v2, v0.l
; GISEL-GFX11-TRUE16-NEXT: scratch_store_b16 off, v0, s0
; GISEL-GFX11-TRUE16-NEXT: s_endpgm
;
; GISEL-GFX11-FAKE16-LABEL: test_llvm_amdgcn_fdot2_f16_f16_dpp:
; GISEL-GFX11-FAKE16: ; %bb.0: ; %entry
; GISEL-GFX11-FAKE16-NEXT: s_load_b128 s[0:3], s[4:5], 0x24
; GISEL-GFX11-FAKE16-NEXT: s_waitcnt lgkmcnt(0)
; GISEL-GFX11-FAKE16-NEXT: scratch_load_b32 v0, off, s1
; GISEL-GFX11-FAKE16-NEXT: scratch_load_b32 v1, off, s2
; GISEL-GFX11-FAKE16-NEXT: scratch_load_u16 v2, off, s3
; GISEL-GFX11-FAKE16-NEXT: s_waitcnt vmcnt(0)
; GISEL-GFX11-FAKE16-NEXT: v_dot2_f16_f16_e64_dpp v0, v0, v1, v2 quad_perm:[1,0,0,0] row_mask:0xf bank_mask:0xf bound_ctrl:1
; GISEL-GFX11-FAKE16-NEXT: scratch_store_b16 off, v0, s0
; GISEL-GFX11-FAKE16-NEXT: s_endpgm
ptr addrspace(5) %r,
ptr addrspace(5) %a,
ptr addrspace(5) %b,
ptr addrspace(5) %c) {
entry:
%a.val = load <2 x half>, ptr addrspace(5) %a
%b.val = load <2 x half>, ptr addrspace(5) %b
%c.val = load half, ptr addrspace(5) %c
%a.val.i32 = bitcast <2 x half> %a.val to i32
%dpp = call i32 @llvm.amdgcn.update.dpp.i32(i32 %a.val.i32, i32 %a.val.i32, i32 1, i32 15, i32 15, i1 1)
%a.val.dpp.v2half = bitcast i32 %dpp to <2 x half>
%r.val = call half @llvm.amdgcn.fdot2.f16.f16(<2 x half> %a.val.dpp.v2half, <2 x half> %b.val, half %c.val)
store half %r.val, ptr addrspace(5) %r
ret void
}
declare i32 @llvm.amdgcn.update.dpp.i32(i32, i32, i32, i32, i32, i1)
;; NOTE: These prefixes are unused and the list is autogenerated. Do not add tests below this line:
; GFX11: {{.*}}
; GISEL-GFX11: {{.*}}
; SDAG-GFX11: {{.*}}
|
