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
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
|
; XUN: llc -mtriple=amdgcn-amd-amdhsa -denormal-fp-math-f32=preserve-sign -verify-machineinstrs < %s | FileCheck -check-prefixes=GCN,SI %s
; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=tonga -denormal-fp-math-f32=preserve-sign -mattr=-flat-for-global -verify-machineinstrs < %s | FileCheck -check-prefixes=GCN,VI,SIVI,VI-DENORM %s
; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=tonga -denormal-fp-math=preserve-sign -denormal-fp-math-f32=preserve-sign -mattr=-flat-for-global -verify-machineinstrs < %s | FileCheck -check-prefixes=GCN,VI,SIVI,VI-FLUSH %s
; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1010 -denormal-fp-math-f32=preserve-sign -mattr=-flat-for-global -verify-machineinstrs < %s | FileCheck -check-prefixes=GCN,GFX10,GFX8_10,GFX10-DENORM %s
; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1010 -denormal-fp-math=preserve-sign -denormal-fp-math-f32=preserve-sign -mattr=-flat-for-global -verify-machineinstrs < %s | FileCheck -check-prefixes=GCN,GFX10,GFX8_10,GFX10-FLUSH %s
; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1100 -denormal-fp-math-f32=preserve-sign -mattr=-flat-for-global -verify-machineinstrs < %s | FileCheck -check-prefixes=GCN,GFX10,GFX8_10,GFX11-DENORM %s
; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1100 -denormal-fp-math=preserve-sign -denormal-fp-math-f32=preserve-sign -mattr=-flat-for-global -verify-machineinstrs < %s | FileCheck -check-prefixes=GCN,GFX10,GFX8_10,GFX10-FLUSH %s
; Make sure (fmul (fadd x, x), c) -> (fmul x, (fmul 2.0, c)) doesn't
; make add an instruction if the fadd has more than one use.
declare half @llvm.fabs.f16(half) #1
declare float @llvm.fabs.f32(float) #1
; GCN-LABEL: {{^}}multiple_fadd_use_test_f32:
; SI: v_max_legacy_f32_e64 [[A16:v[0-9]+]],
; SI: v_add_f32_e32 [[A17:v[0-9]+]], [[A16]], [[A16]]
; SI: v_mul_f32_e32 [[A18:v[0-9]+]], [[A17]], [[A17]]
; SI: v_mad_f32 [[A20:v[0-9]+]], -[[A18]], [[A17]], 1.0
; SI: buffer_store_dword [[A20]]
; GFX8_10: v_add_f32_e64 v{{[0-9]+}}, s{{[0-9]+}}, -1.0
; GFX8_10: v_add_f32_e64 v{{[0-9]+}}, s{{[0-9]+}}, -1.0
; GFX8_10: v_cmp_gt_f32_e64 {{vcc|vcc_lo}}, |v{{[0-9]+}}|, |v{{[0-9]+}}|
; GFX8_10: v_cndmask_b32_e32
; GFX8_10: v_add_f32_e64 v{{[0-9]+}}, |v{{[0-9]+}}|, |v{{[0-9]+}}|
; GFX8_10: v_mul_f32_e32 v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}
; VI: v_mad_f32 v{{[0-9]+}}, -v{{[0-9]+}}, v{{[0-9]+}}, 1.0
; GFX10: v_fma_f32 v{{[0-9]+}}, -v{{[0-9]+}}, v{{[0-9]+}}, 1.0
define amdgpu_kernel void @multiple_fadd_use_test_f32(ptr addrspace(1) %out, float %x, float %y, float %z) #0 {
%a11 = fadd float %y, -1.0
%a12 = call float @llvm.fabs.f32(float %a11)
%a13 = fadd float %x, -1.0
%a14 = call float @llvm.fabs.f32(float %a13)
%a15 = fcmp ogt float %a12, %a14
%a16 = select i1 %a15, float %a12, float %a14
%a17 = fmul float %a16, 2.0
%a18 = fmul float %a17, %a17
%a19 = fmul float %a18, %a17
%a20 = fsub float 1.0, %a19
store float %a20, ptr addrspace(1) %out
ret void
}
; GCN-LABEL: {{^}}multiple_use_fadd_fmac_f32:
; GCN-DAG: v_add_f32_e64 [[MUL2:v[0-9]+]], [[X:s[0-9]+]], s{{[0-9]+}}
; SIVI-DAG: v_mac_f32_e64 [[MAD:v[0-9]+]], [[X]], 2.0
; GFX10-DAG: v_fma_f32 [[MAD:v[0-9]+]], [[X]], 2.0, s{{[0-9]+}}
; GCN-DAG: buffer_store_{{dword|b32}} [[MUL2]]
; GCN-DAG: buffer_store_{{dword|b32}} [[MAD]]
; GCN: s_endpgm
define amdgpu_kernel void @multiple_use_fadd_fmac_f32(ptr addrspace(1) %out, float %x, [8 x i32], float %y) #0 {
%out.gep.1 = getelementptr float, ptr addrspace(1) %out, i32 1
%mul2 = fmul fast float %x, 2.0
%mad = fadd fast float %mul2, %y
store volatile float %mul2, ptr addrspace(1) %out
store volatile float %mad, ptr addrspace(1) %out.gep.1
ret void
}
; GCN-LABEL: {{^}}multiple_use_fadd_fmad_f32:
; GCN-DAG: v_add_f32_e64 [[MUL2:v[0-9]+]], |[[X:s[0-9]+]]|, |s{{[0-9]+}}|
; SIVI-DAG: v_mad_f32 [[MAD:v[0-9]+]], |[[X]]|, 2.0, v{{[0-9]+}}
; GFX10-DAG: v_fma_f32 [[MAD:v[0-9]+]], |[[X]]|, 2.0, s{{[0-9]+}}
; GCN-DAG: buffer_store_{{dword|b32}} [[MUL2]]
; GCN-DAG: buffer_store_{{dword|b32}} [[MAD]]
; GCN: s_endpgm
define amdgpu_kernel void @multiple_use_fadd_fmad_f32(ptr addrspace(1) %out, float %x, float %y) #0 {
%out.gep.1 = getelementptr float, ptr addrspace(1) %out, i32 1
%x.abs = call float @llvm.fabs.f32(float %x)
%mul2 = fmul fast float %x.abs, 2.0
%mad = fadd fast float %mul2, %y
store volatile float %mul2, ptr addrspace(1) %out
store volatile float %mad, ptr addrspace(1) %out.gep.1
ret void
}
; GCN-LABEL: {{^}}multiple_use_fadd_multi_fmad_f32:
; SIVI: v_mad_f32 {{v[0-9]+}}, |[[X:s[0-9]+]]|, 2.0, v{{[0-9]+}}
; SIVI: v_mad_f32 {{v[0-9]+}}, |[[X]]|, 2.0, v{{[0-9]+}}
; GFX10: v_fma_f32 {{v[0-9]+}}, |[[X:s[0-9]+]]|, 2.0, {{s[0-9]+}}
; GFX10: v_fma_f32 {{v[0-9]+}}, |[[X]]|, 2.0, {{s[0-9]+}}
define amdgpu_kernel void @multiple_use_fadd_multi_fmad_f32(ptr addrspace(1) %out, float %x, float %y, float %z) #0 {
%out.gep.1 = getelementptr float, ptr addrspace(1) %out, i32 1
%x.abs = call float @llvm.fabs.f32(float %x)
%mul2 = fmul fast float %x.abs, 2.0
%mad0 = fadd fast float %mul2, %y
%mad1 = fadd fast float %mul2, %z
store volatile float %mad0, ptr addrspace(1) %out
store volatile float %mad1, ptr addrspace(1) %out.gep.1
ret void
}
; GCN-LABEL: {{^}}fmul_x2_xn2_f32:
; GCN: v_mul_f32_e64 [[TMP0:v[0-9]+]], [[X:s[0-9]+]], -4.0
; GCN: v_mul_f32_e32 [[RESULT:v[0-9]+]], [[X]], [[TMP0]]
; GCN: buffer_store_{{dword|b32}} [[RESULT]]
define amdgpu_kernel void @fmul_x2_xn2_f32(ptr addrspace(1) %out, float %x, float %y) #0 {
%out.gep.1 = getelementptr float, ptr addrspace(1) %out, i32 1
%mul2 = fmul fast float %x, 2.0
%muln2 = fmul fast float %x, -2.0
%mul = fmul fast float %mul2, %muln2
store volatile float %mul, ptr addrspace(1) %out
ret void
}
; GCN-LABEL: {{^}}fmul_x2_xn3_f32:
; SIVI: v_mov_b32_e32 [[K:v[0-9]+]], 0xc0c00000
; SIVI: v_mul_f32_e32 [[TMP0:v[0-9]+]], [[X:s[0-9]+]], [[K]]
; GFX10: v_mul_f32_e64 [[TMP0:v[0-9]+]], 0xc0c00000, [[X:s[0-9]+]]
; GCN: v_mul_f32_e32 [[RESULT:v[0-9]+]], [[X]], [[TMP0]]
; GCN: buffer_store_{{dword|b32}} [[RESULT]]
define amdgpu_kernel void @fmul_x2_xn3_f32(ptr addrspace(1) %out, float %x, float %y) #0 {
%out.gep.1 = getelementptr float, ptr addrspace(1) %out, i32 1
%mul2 = fmul fast float %x, 2.0
%muln2 = fmul fast float %x, -3.0
%mul = fmul fast float %mul2, %muln2
store volatile float %mul, ptr addrspace(1) %out
ret void
}
; GCN-LABEL: {{^}}multiple_fadd_use_test_f16:
; GFX8_10: v_add_f16_e64 v{{[0-9]+}}, s{{[0-9]+}}, -1.0
; GFX8_10: v_add_f16_e64 v{{[0-9]+}}, s{{[0-9]+}}, -1.0
; GFX8_10: v_cmp_gt_f16_e64 {{vcc|vcc_lo}}, |v{{[0-9]+}}|, |v{{[0-9]+}}|
; GFX8_10: v_cndmask_b32_e32
; GFX8_10: v_add_f16_e64 v{{[0-9]+}}, |v{{[0-9]+}}|, |v{{[0-9]+}}|
; GFX8_10: v_mul_f16_e32 v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}
; VI-FLUSH: v_mad_f16 v{{[0-9]+}}, -v{{[0-9]+}}, v{{[0-9]+}}, 1.0
; VI-DENORM: v_fma_f16 v{{[0-9]+}}, -v{{[0-9]+}}, v{{[0-9]+}}, 1.0
; GFX10-DENORM: v_fma_f16 v{{[0-9]+}}, -v{{[0-9]+}}, v{{[0-9]+}}, 1.0
; GFX11-DENORM: v_fma_f16 v{{[0-9]+}}, -v{{[0-9]+}}, v{{[0-9]+}}, 1.0
; GFX10-FLUSH: v_sub_f16_e32 v{{[0-9]+}}, 1.0, v{{[0-9]+}}
define amdgpu_kernel void @multiple_fadd_use_test_f16(ptr addrspace(1) %out, i16 zeroext %x.arg, i16 zeroext %y.arg, i16 zeroext %z.arg) #0 {
%x = bitcast i16 %x.arg to half
%y = bitcast i16 %y.arg to half
%z = bitcast i16 %z.arg to half
%a11 = fadd half %y, -1.0
%a12 = call half @llvm.fabs.f16(half %a11)
%a13 = fadd half %x, -1.0
%a14 = call half @llvm.fabs.f16(half %a13)
%a15 = fcmp ogt half %a12, %a14
%a16 = select i1 %a15, half %a12, half %a14
%a17 = fmul half %a16, 2.0
%a18 = fmul half %a17, %a17
%a19 = fmul half %a18, %a17
%a20 = fsub half 1.0, %a19
store half %a20, ptr addrspace(1) %out
ret void
}
; GCN-LABEL: {{^}}multiple_use_fadd_fmac_f16:
; GCN-DAG: v_add_f16_e64 [[MUL2:v[0-9]+]], [[X:s[0-9]+]], s{{[0-9]+}}
; VI-FLUSH-DAG: v_mac_f16_e64 [[MAD:v[0-9]+]], [[X]], 2.0
; VI-DENORM-DAG: v_fma_f16 [[MAD:v[0-9]+]], [[X]], 2.0, v{{[0-9]+}}
; GFX10-FLUSH-DAG: v_add_f16_e32 [[MAD:v[0-9]+]], s{{[0-9]+}}, [[MUL2]]
; GFX10-DENORM-DAG: v_fma_f16 [[MAD:v[0-9]+]], [[X]], 2.0, s{{[0-9]+}}
; GFX11-DENORM-DAG: v_fma_f16 [[MAD:v[0-9]+]], [[X]], 2.0, s{{[0-9]+}}
; GCN-DAG: buffer_store_{{short|b16}} [[MUL2]]
; GCN-DAG: buffer_store_{{short|b16}} [[MAD]]
; GCN: s_endpgm
define amdgpu_kernel void @multiple_use_fadd_fmac_f16(ptr addrspace(1) %out, i16 zeroext %x.arg, i16 zeroext %y.arg) #0 {
%x = bitcast i16 %x.arg to half
%y = bitcast i16 %y.arg to half
%out.gep.1 = getelementptr half, ptr addrspace(1) %out, i32 1
%mul2 = fmul fast half %x, 2.0
%mad = fadd fast half %mul2, %y
store volatile half %mul2, ptr addrspace(1) %out
store volatile half %mad, ptr addrspace(1) %out.gep.1
ret void
}
; GCN-LABEL: {{^}}multiple_use_fadd_fmad_f16:
; GCN-DAG: v_add_f16_e64 [[MUL2:v[0-9]+]], |[[X:s[0-9]+]]|, |s{{[0-9]+}}|
; VI-FLUSH-DAG: v_mad_f16 [[MAD:v[0-9]+]], |[[X]]|, 2.0, v{{[0-9]+}}
; VI-DENORM-DAG: v_fma_f16 [[MAD:v[0-9]+]], |[[X]]|, 2.0, v{{[0-9]+}}
; GFX10-FLUSH-DAG: v_add_f16_e32 [[MAD:v[0-9]+]], s{{[0-9]+}}, [[MUL2]]
; GFX10-DENORM-DAG: v_fma_f16 [[MAD:v[0-9]+]], |[[X]]|, 2.0, s{{[0-9]+}}
; GFX11-DENORM-DAG: v_fma_f16 [[MAD:v[0-9]+]], |[[X]]|, 2.0, s{{[0-9]+}}
; GCN-DAG: buffer_store_{{short|b16}} [[MUL2]]
; GCN-DAG: buffer_store_{{short|b16}} [[MAD]]
; GCN: s_endpgm
define amdgpu_kernel void @multiple_use_fadd_fmad_f16(ptr addrspace(1) %out, i16 zeroext %x.arg, i16 zeroext %y.arg) #0 {
%x = bitcast i16 %x.arg to half
%y = bitcast i16 %y.arg to half
%out.gep.1 = getelementptr half, ptr addrspace(1) %out, i32 1
%x.abs = call half @llvm.fabs.f16(half %x)
%mul2 = fmul fast half %x.abs, 2.0
%mad = fadd fast half %mul2, %y
store volatile half %mul2, ptr addrspace(1) %out
store volatile half %mad, ptr addrspace(1) %out.gep.1
ret void
}
; GCN-LABEL: {{^}}multiple_use_fadd_multi_fmad_f16:
; VI-FLUSH: v_mad_f16 {{v[0-9]+}}, |[[X:s[0-9]+]]|, 2.0, v{{[0-9]+}}
; VI-FLUSH: v_mad_f16 {{v[0-9]+}}, |[[X]]|, 2.0, v{{[0-9]+}}
; VI-DENORM: v_fma_f16 {{v[0-9]+}}, |[[X:s[0-9]+]]|, 2.0, v{{[0-9]+}}
; VI-DENORM: v_fma_f16 {{v[0-9]+}}, |[[X]]|, 2.0, v{{[0-9]+}}
; GFX10-FLUSH: v_add_f16_e64 [[MUL2:v[0-9]+]], |[[X:s[0-9]+]]|, |{{s[0-9]+}}|
; GFX10-FLUSH: v_add_f16_e32 {{v[0-9]+}}, {{s[0-9]+}}, [[MUL2]]
; GFX10-FLUSH: v_add_f16_e32 {{v[0-9]+}}, {{s[0-9]+}}, [[MUL2]]
; GFX10-DENORM: v_fma_f16 {{v[0-9]+}}, |[[X:s[0-9]+]]|, 2.0, s{{[0-9]+}}
; GFX11-DENORM: v_fma_f16 {{v[0-9]+}}, |[[X:s[0-9]+]]|, 2.0, s{{[0-9]+}}
; GFX10-DENORM: v_fma_f16 {{v[0-9]+}}, |[[X]]|, 2.0, s{{[0-9]+}}
; GFX11-DENORM: v_fma_f16 {{v[0-9]+}}, |[[X]]|, 2.0, s{{[0-9]+}}
define amdgpu_kernel void @multiple_use_fadd_multi_fmad_f16(ptr addrspace(1) %out, i16 zeroext %x.arg, i16 zeroext %y.arg, i16 zeroext %z.arg) #0 {
%x = bitcast i16 %x.arg to half
%y = bitcast i16 %y.arg to half
%z = bitcast i16 %z.arg to half
%out.gep.1 = getelementptr half, ptr addrspace(1) %out, i32 1
%x.abs = call half @llvm.fabs.f16(half %x)
%mul2 = fmul fast half %x.abs, 2.0
%mad0 = fadd fast half %mul2, %y
%mad1 = fadd fast half %mul2, %z
store volatile half %mad0, ptr addrspace(1) %out
store volatile half %mad1, ptr addrspace(1) %out.gep.1
ret void
}
; GCN-LABEL: {{^}}fmul_x2_xn2_f16:
; GCN: v_mul_f16_e64 [[TMP0:v[0-9]+]], [[X:s[0-9]+]], -4.0
; GCN: v_mul_f16_e32 [[RESULT:v[0-9]+]], [[X]], [[TMP0]]
; GCN: buffer_store_{{short|b16}} [[RESULT]]
define amdgpu_kernel void @fmul_x2_xn2_f16(ptr addrspace(1) %out, i16 zeroext %x.arg, i16 zeroext %y.arg) #0 {
%x = bitcast i16 %x.arg to half
%y = bitcast i16 %y.arg to half
%out.gep.1 = getelementptr half, ptr addrspace(1) %out, i32 1
%mul2 = fmul fast half %x, 2.0
%muln2 = fmul fast half %x, -2.0
%mul = fmul fast half %mul2, %muln2
store volatile half %mul, ptr addrspace(1) %out
ret void
}
; GCN-LABEL: {{^}}fmul_x2_xn3_f16:
; SIVI: v_mov_b32_e32 [[K:v[0-9]+]], 0xc600
; SIVI: v_mul_f16_e32 [[TMP0:v[0-9]+]], [[X:s[0-9]+]], [[K]]
; GFX10: v_mul_f16_e64 [[TMP0:v[0-9]+]], 0xc600, [[X:s[0-9]+]]
; GCN: v_mul_f16_e32 [[RESULT:v[0-9]+]], [[X]], [[TMP0]]
; GCN: buffer_store_{{short|b16}} [[RESULT]]
define amdgpu_kernel void @fmul_x2_xn3_f16(ptr addrspace(1) %out, i16 zeroext %x.arg, i16 zeroext %y.arg) #0 {
%x = bitcast i16 %x.arg to half
%y = bitcast i16 %y.arg to half
%out.gep.1 = getelementptr half, ptr addrspace(1) %out, i32 1
%mul2 = fmul fast half %x, 2.0
%muln2 = fmul fast half %x, -3.0
%mul = fmul fast half %mul2, %muln2
store volatile half %mul, ptr addrspace(1) %out
ret void
}
attributes #0 = { nounwind "unsafe-fp-math"="true" }
attributes #1 = { nounwind readnone }
|
