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
161
162
163
164
165
166
167
168
169
170
|
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 5
; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx942 -o - %s | FileCheck -check-prefix=GCN %s
define amdgpu_kernel void @MFMAExpInterleave(ptr addrspace(1) %out0, ptr addrspace(1) %out1, float %in0, <4 x float> %in1) {
; GCN-LABEL: MFMAExpInterleave:
; GCN: ; %bb.0:
; GCN-NEXT: s_load_dword s6, s[4:5], 0x10
; GCN-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x20
; GCN-NEXT: v_mov_b32_e32 v1, 0x3fb8aa3b
; GCN-NEXT: v_mov_b32_e32 v0, 1.0
; GCN-NEXT: s_mov_b32 s7, 0x42b17218
; GCN-NEXT: s_waitcnt lgkmcnt(0)
; GCN-NEXT: v_mul_f32_e32 v2, s6, v1
; GCN-NEXT: v_rndne_f32_e32 v3, v2
; GCN-NEXT: v_sub_f32_e32 v4, v2, v3
; GCN-NEXT: v_fma_f32 v1, s6, v1, -v2
; GCN-NEXT: v_mov_b32_e32 v2, 0x32a5705f
; GCN-NEXT: v_accvgpr_write_b32 a0, s0
; GCN-NEXT: v_fmac_f32_e32 v1, s6, v2
; GCN-NEXT: v_accvgpr_write_b32 a1, s1
; GCN-NEXT: v_accvgpr_write_b32 a2, s2
; GCN-NEXT: v_accvgpr_write_b32 a3, s3
; GCN-NEXT: v_add_f32_e32 v1, v4, v1
; GCN-NEXT: v_cvt_i32_f32_e32 v2, v3
; GCN-NEXT: v_mfma_f32_4x4x1_16b_f32 a[0:3], v0, v0, a[0:3]
; GCN-NEXT: v_exp_f32_e32 v1, v1
; GCN-NEXT: s_mov_b32 s0, 0x3fb8aa3b
; GCN-NEXT: v_mfma_f32_4x4x1_16b_f32 a[0:3], v0, v0, a[0:3]
; GCN-NEXT: ; iglp_opt mask(0x00000003)
; GCN-NEXT: v_ldexp_f32 v1, v1, v2
; GCN-NEXT: v_mov_b32_e32 v2, 0xc2ce8ed0
; GCN-NEXT: v_cmp_nlt_f32_e32 vcc, s6, v2
; GCN-NEXT: v_mov_b32_e32 v2, 0x42b17218
; GCN-NEXT: s_nop 0
; GCN-NEXT: v_cndmask_b32_e32 v1, 0, v1, vcc
; GCN-NEXT: v_cmp_ngt_f32_e32 vcc, s6, v2
; GCN-NEXT: v_mov_b32_e32 v2, 0x7f800000
; GCN-NEXT: s_mov_b32 s6, 0xc2ce8ed0
; GCN-NEXT: v_cndmask_b32_e32 v1, v2, v1, vcc
; GCN-NEXT: v_mul_f32_e32 v3, 0x3fb8aa3b, v1
; GCN-NEXT: v_fma_f32 v4, v1, s0, -v3
; GCN-NEXT: v_rndne_f32_e32 v5, v3
; GCN-NEXT: v_fmac_f32_e32 v4, 0x32a5705f, v1
; GCN-NEXT: v_sub_f32_e32 v3, v3, v5
; GCN-NEXT: v_add_f32_e32 v3, v3, v4
; GCN-NEXT: v_exp_f32_e32 v3, v3
; GCN-NEXT: v_cvt_i32_f32_e32 v4, v5
; GCN-NEXT: v_cmp_ngt_f32_e32 vcc, s6, v1
; GCN-NEXT: v_mfma_f32_4x4x1_16b_f32 a[0:3], v0, v0, a[0:3]
; GCN-NEXT: v_ldexp_f32 v3, v3, v4
; GCN-NEXT: v_cndmask_b32_e32 v3, 0, v3, vcc
; GCN-NEXT: v_cmp_nlt_f32_e32 vcc, s7, v1
; GCN-NEXT: s_nop 1
; GCN-NEXT: v_cndmask_b32_e32 v1, v2, v3, vcc
; GCN-NEXT: v_mul_f32_e32 v3, 0x3fb8aa3b, v1
; GCN-NEXT: v_fma_f32 v4, v1, s0, -v3
; GCN-NEXT: v_rndne_f32_e32 v5, v3
; GCN-NEXT: v_fmac_f32_e32 v4, 0x32a5705f, v1
; GCN-NEXT: v_sub_f32_e32 v3, v3, v5
; GCN-NEXT: v_add_f32_e32 v3, v3, v4
; GCN-NEXT: v_exp_f32_e32 v3, v3
; GCN-NEXT: v_cvt_i32_f32_e32 v4, v5
; GCN-NEXT: v_cmp_ngt_f32_e32 vcc, s6, v1
; GCN-NEXT: v_mfma_f32_4x4x1_16b_f32 a[0:3], v0, v0, a[0:3]
; GCN-NEXT: v_ldexp_f32 v3, v3, v4
; GCN-NEXT: v_cndmask_b32_e32 v3, 0, v3, vcc
; GCN-NEXT: v_cmp_nlt_f32_e32 vcc, s7, v1
; GCN-NEXT: s_nop 1
; GCN-NEXT: v_cndmask_b32_e32 v1, v2, v3, vcc
; GCN-NEXT: v_mul_f32_e32 v3, 0x3fb8aa3b, v1
; GCN-NEXT: v_fma_f32 v4, v1, s0, -v3
; GCN-NEXT: v_rndne_f32_e32 v5, v3
; GCN-NEXT: v_fmac_f32_e32 v4, 0x32a5705f, v1
; GCN-NEXT: v_sub_f32_e32 v3, v3, v5
; GCN-NEXT: v_add_f32_e32 v3, v3, v4
; GCN-NEXT: v_exp_f32_e32 v3, v3
; GCN-NEXT: v_cvt_i32_f32_e32 v4, v5
; GCN-NEXT: v_cmp_ngt_f32_e32 vcc, s6, v1
; GCN-NEXT: v_mfma_f32_4x4x1_16b_f32 a[0:3], v0, v0, a[0:3]
; GCN-NEXT: v_ldexp_f32 v3, v3, v4
; GCN-NEXT: v_cndmask_b32_e32 v3, 0, v3, vcc
; GCN-NEXT: v_cmp_nlt_f32_e32 vcc, s7, v1
; GCN-NEXT: s_nop 1
; GCN-NEXT: v_cndmask_b32_e32 v1, v2, v3, vcc
; GCN-NEXT: v_mul_f32_e32 v3, 0x3fb8aa3b, v1
; GCN-NEXT: v_fma_f32 v4, v1, s0, -v3
; GCN-NEXT: v_rndne_f32_e32 v5, v3
; GCN-NEXT: v_fmac_f32_e32 v4, 0x32a5705f, v1
; GCN-NEXT: v_sub_f32_e32 v3, v3, v5
; GCN-NEXT: v_add_f32_e32 v3, v3, v4
; GCN-NEXT: v_exp_f32_e32 v3, v3
; GCN-NEXT: v_cvt_i32_f32_e32 v4, v5
; GCN-NEXT: v_cmp_ngt_f32_e32 vcc, s6, v1
; GCN-NEXT: v_mfma_f32_4x4x1_16b_f32 a[0:3], v0, v0, a[0:3]
; GCN-NEXT: v_ldexp_f32 v3, v3, v4
; GCN-NEXT: v_cndmask_b32_e32 v3, 0, v3, vcc
; GCN-NEXT: v_cmp_nlt_f32_e32 vcc, s7, v1
; GCN-NEXT: s_nop 1
; GCN-NEXT: v_cndmask_b32_e32 v1, v2, v3, vcc
; GCN-NEXT: v_mul_f32_e32 v3, 0x3fb8aa3b, v1
; GCN-NEXT: v_fma_f32 v4, v1, s0, -v3
; GCN-NEXT: v_rndne_f32_e32 v5, v3
; GCN-NEXT: v_fmac_f32_e32 v4, 0x32a5705f, v1
; GCN-NEXT: v_sub_f32_e32 v3, v3, v5
; GCN-NEXT: v_add_f32_e32 v3, v3, v4
; GCN-NEXT: v_exp_f32_e32 v3, v3
; GCN-NEXT: v_cvt_i32_f32_e32 v4, v5
; GCN-NEXT: v_cmp_ngt_f32_e32 vcc, s6, v1
; GCN-NEXT: v_mfma_f32_4x4x1_16b_f32 a[0:3], v0, v0, a[0:3]
; GCN-NEXT: v_ldexp_f32 v3, v3, v4
; GCN-NEXT: v_cndmask_b32_e32 v3, 0, v3, vcc
; GCN-NEXT: v_cmp_nlt_f32_e32 vcc, s7, v1
; GCN-NEXT: s_nop 1
; GCN-NEXT: v_cndmask_b32_e32 v1, v2, v3, vcc
; GCN-NEXT: v_mul_f32_e32 v3, 0x3fb8aa3b, v1
; GCN-NEXT: v_fma_f32 v4, v1, s0, -v3
; GCN-NEXT: v_rndne_f32_e32 v5, v3
; GCN-NEXT: v_fmac_f32_e32 v4, 0x32a5705f, v1
; GCN-NEXT: v_sub_f32_e32 v3, v3, v5
; GCN-NEXT: v_add_f32_e32 v3, v3, v4
; GCN-NEXT: v_exp_f32_e32 v3, v3
; GCN-NEXT: v_cvt_i32_f32_e32 v4, v5
; GCN-NEXT: v_mfma_f32_4x4x1_16b_f32 a[0:3], v0, v0, a[0:3]
; GCN-NEXT: v_cmp_ngt_f32_e32 vcc, s6, v1
; GCN-NEXT: v_ldexp_f32 v0, v3, v4
; GCN-NEXT: s_nop 0
; GCN-NEXT: v_cndmask_b32_e32 v0, 0, v0, vcc
; GCN-NEXT: v_cmp_nlt_f32_e32 vcc, s7, v1
; GCN-NEXT: s_nop 1
; GCN-NEXT: v_cndmask_b32_e32 v0, v2, v0, vcc
; GCN-NEXT: v_mul_f32_e32 v1, 0x3fb8aa3b, v0
; GCN-NEXT: v_fma_f32 v3, v0, s0, -v1
; GCN-NEXT: v_rndne_f32_e32 v4, v1
; GCN-NEXT: v_fmac_f32_e32 v3, 0x32a5705f, v0
; GCN-NEXT: v_sub_f32_e32 v1, v1, v4
; GCN-NEXT: v_add_f32_e32 v1, v1, v3
; GCN-NEXT: v_exp_f32_e32 v1, v1
; GCN-NEXT: v_cvt_i32_f32_e32 v3, v4
; GCN-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x0
; GCN-NEXT: v_cmp_ngt_f32_e32 vcc, s6, v0
; GCN-NEXT: v_mov_b32_e32 v4, 0
; GCN-NEXT: v_ldexp_f32 v1, v1, v3
; GCN-NEXT: v_cndmask_b32_e32 v1, 0, v1, vcc
; GCN-NEXT: v_cmp_nlt_f32_e32 vcc, s7, v0
; GCN-NEXT: s_waitcnt lgkmcnt(0)
; GCN-NEXT: global_store_dwordx4 v4, a[0:3], s[0:1]
; GCN-NEXT: v_cndmask_b32_e32 v0, v2, v1, vcc
; GCN-NEXT: global_store_dword v4, v0, s[2:3]
; GCN-NEXT: s_endpgm
%mai0 = tail call <4 x float> @llvm.amdgcn.mfma.f32.4x4x1f32(float 1.0, float 1.0, <4 x float> %in1, i32 0, i32 0, i32 0)
%mai1 = tail call <4 x float> @llvm.amdgcn.mfma.f32.4x4x1f32(float 1.0, float 1.0, <4 x float> %mai0, i32 0, i32 0, i32 0)
%mai2 = tail call <4 x float> @llvm.amdgcn.mfma.f32.4x4x1f32(float 1.0, float 1.0, <4 x float> %mai1, i32 0, i32 0, i32 0)
%mai3 = tail call <4 x float> @llvm.amdgcn.mfma.f32.4x4x1f32(float 1.0, float 1.0, <4 x float> %mai2, i32 0, i32 0, i32 0)
%mai4 = tail call <4 x float> @llvm.amdgcn.mfma.f32.4x4x1f32(float 1.0, float 1.0, <4 x float> %mai3, i32 0, i32 0, i32 0)
%mai5 = tail call <4 x float> @llvm.amdgcn.mfma.f32.4x4x1f32(float 1.0, float 1.0, <4 x float> %mai4, i32 0, i32 0, i32 0)
%mai6 = tail call <4 x float> @llvm.amdgcn.mfma.f32.4x4x1f32(float 1.0, float 1.0, <4 x float> %mai5, i32 0, i32 0, i32 0)
%mai7 = tail call <4 x float> @llvm.amdgcn.mfma.f32.4x4x1f32(float 1.0, float 1.0, <4 x float> %mai6, i32 0, i32 0, i32 0)
%exp0 = call float @llvm.exp.f32(float %in0)
%exp1 = call float @llvm.exp.f32(float %exp0)
%exp2 = call float @llvm.exp.f32(float %exp1)
%exp3 = call float @llvm.exp.f32(float %exp2)
%exp4 = call float @llvm.exp.f32(float %exp3)
%exp5 = call float @llvm.exp.f32(float %exp4)
%exp6 = call float @llvm.exp.f32(float %exp5)
%exp7 = call float @llvm.exp.f32(float %exp6)
store <4 x float> %mai7, ptr addrspace(1) %out0
store float %exp7, ptr addrspace(1) %out1
tail call void @llvm.amdgcn.iglp.opt(i32 3)
ret void
}
|
