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
|
; RUN: llc -mtriple=amdgcn -mcpu=hawaii < %s | FileCheck -check-prefix=CI -check-prefix=GCN %s
; RUN: llc -mtriple=amdgcn -mcpu=tahiti < %s | FileCheck -check-prefix=SI -check-prefix=GCN %s
; RUN: llc -mtriple=amdgcn -mcpu=tonga -mattr=-flat-for-global < %s | FileCheck -check-prefix=CI -check-prefix=GCN %s
; GCN-LABEL: {{^}}fdiv_f64:
; GCN-DAG: buffer_load_dwordx2 [[NUM:v\[[0-9]+:[0-9]+\]]], off, {{s\[[0-9]+:[0-9]+\]}}, 0
; GCN-DAG: buffer_load_dwordx2 [[DEN:v\[[0-9]+:[0-9]+\]]], off, {{s\[[0-9]+:[0-9]+\]}}, 0 offset:8
; CI-DAG: v_div_scale_f64 [[SCALE0:v\[[0-9]+:[0-9]+\]]], {{s\[[0-9]+:[0-9]+\]}}, [[DEN]], [[DEN]], [[NUM]]
; CI-DAG: v_div_scale_f64 [[SCALE1:v\[[0-9]+:[0-9]+\]]], vcc, [[NUM]], [[DEN]], [[NUM]]
; Check for div_scale bug workaround on SI
; SI-DAG: v_div_scale_f64 [[SCALE0:v\[[0-9]+:[0-9]+\]]], {{s\[[0-9]+:[0-9]+\]}}, [[DEN]], [[DEN]], [[NUM]]
; SI-DAG: v_div_scale_f64 [[SCALE1:v\[[0-9]+:[0-9]+\]]], {{s\[[0-9]+:[0-9]+\]}}, [[NUM]], [[DEN]], [[NUM]]
; GCN-DAG: v_rcp_f64_e32 [[RCP_SCALE0:v\[[0-9]+:[0-9]+\]]], [[SCALE0]]
; SI-DAG: v_cmp_eq_u32_e32 vcc, {{v[0-9]+}}, {{v[0-9]+}}
; SI-DAG: v_cmp_eq_u32_e64 [[CMP0:s\[[0-9]+:[0-9]+\]]], {{v[0-9]+}}, {{v[0-9]+}}
; SI-DAG: s_xor_b64 vcc, [[CMP0]], vcc
; GCN-DAG: v_fma_f64 [[FMA0:v\[[0-9]+:[0-9]+\]]], -[[SCALE0]], [[RCP_SCALE0]], 1.0
; GCN-DAG: v_fma_f64 [[FMA1:v\[[0-9]+:[0-9]+\]]], [[RCP_SCALE0]], [[FMA0]], [[RCP_SCALE0]]
; GCN-DAG: v_fma_f64 [[FMA2:v\[[0-9]+:[0-9]+\]]], -[[SCALE0]], [[FMA1]], 1.0
; GCN-DAG: v_fma_f64 [[FMA3:v\[[0-9]+:[0-9]+\]]], [[FMA1]], [[FMA2]], [[FMA1]]
; GCN-DAG: v_mul_f64 [[MUL:v\[[0-9]+:[0-9]+\]]], [[SCALE1]], [[FMA3]]
; GCN-DAG: v_fma_f64 [[FMA4:v\[[0-9]+:[0-9]+\]]], -[[SCALE0]], [[MUL]], [[SCALE1]]
; GCN: v_div_fmas_f64 [[FMAS:v\[[0-9]+:[0-9]+\]]], [[FMA4]], [[FMA3]], [[MUL]]
; GCN: v_div_fixup_f64 [[RESULT:v\[[0-9]+:[0-9]+\]]], [[FMAS]], [[DEN]], [[NUM]]
; GCN: buffer_store_dwordx2 [[RESULT]]
; GCN: s_endpgm
define amdgpu_kernel void @fdiv_f64(ptr addrspace(1) %out, ptr addrspace(1) %in) #0 {
%gep.1 = getelementptr double, ptr addrspace(1) %in, i32 1
%num = load volatile double, ptr addrspace(1) %in
%den = load volatile double, ptr addrspace(1) %gep.1
%result = fdiv double %num, %den
store double %result, ptr addrspace(1) %out
ret void
}
; GCN-LABEL: {{^}}v_fdiv_f64_afn:
; GCN: v_rcp_f64_e32 v[4:5], v[2:3]
; GCN: v_fma_f64 v[6:7], -v[2:3], v[4:5], 1.0
; GCN: v_fma_f64 v[4:5], v[6:7], v[4:5], v[4:5]
; GCN: v_fma_f64 v[6:7], -v[2:3], v[4:5], 1.0
; GCN: v_fma_f64 v[4:5], v[6:7], v[4:5], v[4:5]
; GCN: v_mul_f64 v[6:7], v[0:1], v[4:5]
; GCN: v_fma_f64 v[0:1], -v[2:3], v[6:7], v[0:1]
; GCN: v_fma_f64 v[0:1], v[0:1], v[4:5], v[6:7]
; GCN: s_setpc_b64
define double @v_fdiv_f64_afn(double %x, double %y) #0 {
%result = fdiv afn double %x, %y
ret double %result
}
; GCN-LABEL: {{^}}v_rcp_f64_afn:
; GCN: v_rcp_f64_e32 v[2:3], v[0:1]
; GCN: v_fma_f64 v[4:5], -v[0:1], v[2:3], 1.0
; GCN: v_fma_f64 v[2:3], v[4:5], v[2:3], v[2:3]
; GCN: v_fma_f64 v[4:5], -v[0:1], v[2:3], 1.0
; GCN: v_fma_f64 v[2:3], v[4:5], v[2:3], v[2:3]
; GCN: v_fma_f64 v[0:1], -v[0:1], v[2:3], 1.0
; GCN: v_fma_f64 v[0:1], v[0:1], v[2:3], v[2:3]
; GCN: s_setpc_b64
define double @v_rcp_f64_afn(double %x) #0 {
%result = fdiv afn double 1.0, %x
ret double %result
}
; GCN-LABEL: {{^}}fdiv_f64_s_v:
define amdgpu_kernel void @fdiv_f64_s_v(ptr addrspace(1) %out, ptr addrspace(1) %in, double %num) #0 {
%den = load double, ptr addrspace(1) %in
%result = fdiv double %num, %den
store double %result, ptr addrspace(1) %out
ret void
}
; GCN-LABEL: {{^}}fdiv_f64_v_s:
define amdgpu_kernel void @fdiv_f64_v_s(ptr addrspace(1) %out, ptr addrspace(1) %in, double %den) #0 {
%num = load double, ptr addrspace(1) %in
%result = fdiv double %num, %den
store double %result, ptr addrspace(1) %out
ret void
}
; GCN-LABEL: {{^}}fdiv_f64_s_s:
define amdgpu_kernel void @fdiv_f64_s_s(ptr addrspace(1) %out, double %num, double %den) #0 {
%result = fdiv double %num, %den
store double %result, ptr addrspace(1) %out
ret void
}
; GCN-LABEL: {{^}}v_fdiv_v2f64:
define amdgpu_kernel void @v_fdiv_v2f64(ptr addrspace(1) %out, ptr addrspace(1) %in) #0 {
%gep.1 = getelementptr <2 x double>, ptr addrspace(1) %in, i32 1
%num = load <2 x double>, ptr addrspace(1) %in
%den = load <2 x double>, ptr addrspace(1) %gep.1
%result = fdiv <2 x double> %num, %den
store <2 x double> %result, ptr addrspace(1) %out
ret void
}
; GCN-LABEL: {{^}}s_fdiv_v2f64:
define amdgpu_kernel void @s_fdiv_v2f64(ptr addrspace(1) %out, <2 x double> %num, <2 x double> %den) {
%result = fdiv <2 x double> %num, %den
store <2 x double> %result, ptr addrspace(1) %out
ret void
}
; GCN-LABEL: {{^}}v_fdiv_v4f64:
define amdgpu_kernel void @v_fdiv_v4f64(ptr addrspace(1) %out, ptr addrspace(1) %in) #0 {
%gep.1 = getelementptr <4 x double>, ptr addrspace(1) %in, i32 1
%num = load <4 x double>, ptr addrspace(1) %in
%den = load <4 x double>, ptr addrspace(1) %gep.1
%result = fdiv <4 x double> %num, %den
store <4 x double> %result, ptr addrspace(1) %out
ret void
}
; GCN-LABEL: {{^}}s_fdiv_v4f64:
define amdgpu_kernel void @s_fdiv_v4f64(ptr addrspace(1) %out, <4 x double> %num, <4 x double> %den) #0 {
%result = fdiv <4 x double> %num, %den
store <4 x double> %result, ptr addrspace(1) %out
ret void
}
; GCN-LABEL: {{^}}div_fast_2_x_pat_f64:
; GCN: v_mul_f64 [[MUL:v\[[0-9]+:[0-9]+\]]], s{{\[[0-9]+:[0-9]+\]}}, 0.5
; GCN: buffer_store_dwordx2 [[MUL]]
define amdgpu_kernel void @div_fast_2_x_pat_f64(ptr addrspace(1) %out) #1 {
%x = load double, ptr addrspace(1) poison
%rcp = fdiv fast double %x, 2.0
store double %rcp, ptr addrspace(1) %out, align 4
ret void
}
; GCN-LABEL: {{^}}div_fast_k_x_pat_f64:
; GCN-DAG: v_mov_b32_e32 v[[K_LO:[0-9]+]], 0x9999999a
; GCN-DAG: v_mov_b32_e32 v[[K_HI:[0-9]+]], 0x3fb99999
; GCN: v_mul_f64 [[MUL:v\[[0-9]+:[0-9]+\]]], s{{\[[0-9]+:[0-9]+\]}}, v[[[K_LO]]:[[K_HI]]]
; GCN: buffer_store_dwordx2 [[MUL]]
define amdgpu_kernel void @div_fast_k_x_pat_f64(ptr addrspace(1) %out) #1 {
%x = load double, ptr addrspace(1) poison
%rcp = fdiv fast double %x, 10.0
store double %rcp, ptr addrspace(1) %out, align 4
ret void
}
; GCN-LABEL: {{^}}div_fast_neg_k_x_pat_f64:
; GCN-DAG: v_mov_b32_e32 v[[K_LO:[0-9]+]], 0x9999999a
; GCN-DAG: v_mov_b32_e32 v[[K_HI:[0-9]+]], 0xbfb99999
; GCN: v_mul_f64 [[MUL:v\[[0-9]+:[0-9]+\]]], s{{\[[0-9]+:[0-9]+\]}}, v[[[K_LO]]:[[K_HI]]]
; GCN: buffer_store_dwordx2 [[MUL]]
define amdgpu_kernel void @div_fast_neg_k_x_pat_f64(ptr addrspace(1) %out) #1 {
%x = load double, ptr addrspace(1) poison
%rcp = fdiv fast double %x, -10.0
store double %rcp, ptr addrspace(1) %out, align 4
ret void
}
attributes #0 = { nounwind }
attributes #1 = { nounwind "unsafe-fp-math"="true" }
|
