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
|
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc < %s -mtriple=amdgcn-amd-mesa3d -mcpu=fiji -verify-machineinstrs | FileCheck -check-prefix=VI %s
; RUN: llc < %s -mtriple=amdgcn-amd-mesa3d -mcpu=gfx900 -verify-machineinstrs | FileCheck -check-prefix=GFX9 %s
; RUN: llc < %s -mtriple=amdgcn-amd-mesa3d -mcpu=gfx1010 -verify-machineinstrs | FileCheck -check-prefix=GFX10 %s
; RUN: llc < %s -mtriple=amdgcn-amd-mesa3d -mcpu=gfx1100 -amdgpu-enable-delay-alu=0 -verify-machineinstrs | FileCheck -check-prefix=GFX10 %s
; ===================================================================================
; V_ADD3_U32
; ===================================================================================
define amdgpu_ps float @add3(i32 %a, i32 %b, i32 %c) {
; VI-LABEL: add3:
; VI: ; %bb.0:
; VI-NEXT: v_add_u32_e32 v0, vcc, v0, v1
; VI-NEXT: v_add_u32_e32 v0, vcc, v0, v2
; VI-NEXT: ; return to shader part epilog
;
; GFX9-LABEL: add3:
; GFX9: ; %bb.0:
; GFX9-NEXT: v_add3_u32 v0, v0, v1, v2
; GFX9-NEXT: ; return to shader part epilog
;
; GFX10-LABEL: add3:
; GFX10: ; %bb.0:
; GFX10-NEXT: v_add3_u32 v0, v0, v1, v2
; GFX10-NEXT: ; return to shader part epilog
%x = add i32 %a, %b
%result = add i32 %x, %c
%bc = bitcast i32 %result to float
ret float %bc
}
; V_MAD_U32_U24 is given higher priority.
define amdgpu_ps float @mad_no_add3(i32 %a, i32 %b, i32 %c, i32 %d, i32 %e) {
; VI-LABEL: mad_no_add3:
; VI: ; %bb.0:
; VI-NEXT: v_mad_u32_u24 v0, v0, v1, v4
; VI-NEXT: v_mad_u32_u24 v0, v2, v3, v0
; VI-NEXT: ; return to shader part epilog
;
; GFX9-LABEL: mad_no_add3:
; GFX9: ; %bb.0:
; GFX9-NEXT: v_mad_u32_u24 v0, v0, v1, v4
; GFX9-NEXT: v_mad_u32_u24 v0, v2, v3, v0
; GFX9-NEXT: ; return to shader part epilog
;
; GFX10-LABEL: mad_no_add3:
; GFX10: ; %bb.0:
; GFX10-NEXT: v_mad_u32_u24 v0, v0, v1, v4
; GFX10-NEXT: v_mad_u32_u24 v0, v2, v3, v0
; GFX10-NEXT: ; return to shader part epilog
%a0 = shl i32 %a, 8
%a1 = lshr i32 %a0, 8
%b0 = shl i32 %b, 8
%b1 = lshr i32 %b0, 8
%mul1 = mul i32 %a1, %b1
%c0 = shl i32 %c, 8
%c1 = lshr i32 %c0, 8
%d0 = shl i32 %d, 8
%d1 = lshr i32 %d0, 8
%mul2 = mul i32 %c1, %d1
%add0 = add i32 %e, %mul1
%add1 = add i32 %mul2, %add0
%bc = bitcast i32 %add1 to float
ret float %bc
}
; ThreeOp instruction variant not used due to Constant Bus Limitations
; TODO: with reassociation it is possible to replace a v_add_u32_e32 with a s_add_i32
define amdgpu_ps float @add3_vgpr_b(i32 inreg %a, i32 %b, i32 inreg %c) {
; VI-LABEL: add3_vgpr_b:
; VI: ; %bb.0:
; VI-NEXT: s_add_i32 s3, s3, s2
; VI-NEXT: v_add_u32_e32 v0, vcc, s3, v0
; VI-NEXT: ; return to shader part epilog
;
; GFX9-LABEL: add3_vgpr_b:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_add_i32 s3, s3, s2
; GFX9-NEXT: v_add_u32_e32 v0, s3, v0
; GFX9-NEXT: ; return to shader part epilog
;
; GFX10-LABEL: add3_vgpr_b:
; GFX10: ; %bb.0:
; GFX10-NEXT: v_add3_u32 v0, s3, s2, v0
; GFX10-NEXT: ; return to shader part epilog
%x = add i32 %a, %b
%result = add i32 %x, %c
%bc = bitcast i32 %result to float
ret float %bc
}
define amdgpu_ps float @add3_vgpr_all2(i32 %a, i32 %b, i32 %c) {
; VI-LABEL: add3_vgpr_all2:
; VI: ; %bb.0:
; VI-NEXT: v_add_u32_e32 v1, vcc, v1, v2
; VI-NEXT: v_add_u32_e32 v0, vcc, v0, v1
; VI-NEXT: ; return to shader part epilog
;
; GFX9-LABEL: add3_vgpr_all2:
; GFX9: ; %bb.0:
; GFX9-NEXT: v_add3_u32 v0, v1, v2, v0
; GFX9-NEXT: ; return to shader part epilog
;
; GFX10-LABEL: add3_vgpr_all2:
; GFX10: ; %bb.0:
; GFX10-NEXT: v_add3_u32 v0, v1, v2, v0
; GFX10-NEXT: ; return to shader part epilog
%x = add i32 %b, %c
%result = add i32 %a, %x
%bc = bitcast i32 %result to float
ret float %bc
}
define amdgpu_ps float @add3_vgpr_bc(i32 inreg %a, i32 %b, i32 %c) {
; VI-LABEL: add3_vgpr_bc:
; VI: ; %bb.0:
; VI-NEXT: v_add_u32_e32 v0, vcc, s2, v0
; VI-NEXT: v_add_u32_e32 v0, vcc, v0, v1
; VI-NEXT: ; return to shader part epilog
;
; GFX9-LABEL: add3_vgpr_bc:
; GFX9: ; %bb.0:
; GFX9-NEXT: v_add3_u32 v0, s2, v0, v1
; GFX9-NEXT: ; return to shader part epilog
;
; GFX10-LABEL: add3_vgpr_bc:
; GFX10: ; %bb.0:
; GFX10-NEXT: v_add3_u32 v0, s2, v0, v1
; GFX10-NEXT: ; return to shader part epilog
%x = add i32 %a, %b
%result = add i32 %x, %c
%bc = bitcast i32 %result to float
ret float %bc
}
define amdgpu_ps float @add3_vgpr_const(i32 %a, i32 %b) {
; VI-LABEL: add3_vgpr_const:
; VI: ; %bb.0:
; VI-NEXT: v_add_u32_e32 v0, vcc, v0, v1
; VI-NEXT: v_add_u32_e32 v0, vcc, 16, v0
; VI-NEXT: ; return to shader part epilog
;
; GFX9-LABEL: add3_vgpr_const:
; GFX9: ; %bb.0:
; GFX9-NEXT: v_add3_u32 v0, v0, v1, 16
; GFX9-NEXT: ; return to shader part epilog
;
; GFX10-LABEL: add3_vgpr_const:
; GFX10: ; %bb.0:
; GFX10-NEXT: v_add3_u32 v0, v0, v1, 16
; GFX10-NEXT: ; return to shader part epilog
%x = add i32 %a, %b
%result = add i32 %x, 16
%bc = bitcast i32 %result to float
ret float %bc
}
define amdgpu_ps <2 x float> @add3_multiuse_outer(i32 %a, i32 %b, i32 %c, i32 %x) {
; VI-LABEL: add3_multiuse_outer:
; VI: ; %bb.0:
; VI-NEXT: v_add_u32_e32 v0, vcc, v0, v1
; VI-NEXT: v_add_u32_e32 v0, vcc, v0, v2
; VI-NEXT: v_mul_lo_u32 v1, v0, v3
; VI-NEXT: ; return to shader part epilog
;
; GFX9-LABEL: add3_multiuse_outer:
; GFX9: ; %bb.0:
; GFX9-NEXT: v_add3_u32 v0, v0, v1, v2
; GFX9-NEXT: v_mul_lo_u32 v1, v0, v3
; GFX9-NEXT: ; return to shader part epilog
;
; GFX10-LABEL: add3_multiuse_outer:
; GFX10: ; %bb.0:
; GFX10-NEXT: v_add3_u32 v0, v0, v1, v2
; GFX10-NEXT: v_mul_lo_u32 v1, v0, v3
; GFX10-NEXT: ; return to shader part epilog
%inner = add i32 %a, %b
%outer = add i32 %inner, %c
%x1 = mul i32 %outer, %x
%r1 = insertelement <2 x i32> undef, i32 %outer, i32 0
%r0 = insertelement <2 x i32> %r1, i32 %x1, i32 1
%bc = bitcast <2 x i32> %r0 to <2 x float>
ret <2 x float> %bc
}
define amdgpu_ps <2 x float> @add3_multiuse_inner(i32 %a, i32 %b, i32 %c) {
; VI-LABEL: add3_multiuse_inner:
; VI: ; %bb.0:
; VI-NEXT: v_add_u32_e32 v0, vcc, v0, v1
; VI-NEXT: v_add_u32_e32 v1, vcc, v0, v2
; VI-NEXT: ; return to shader part epilog
;
; GFX9-LABEL: add3_multiuse_inner:
; GFX9: ; %bb.0:
; GFX9-NEXT: v_add_u32_e32 v0, v0, v1
; GFX9-NEXT: v_add_u32_e32 v1, v0, v2
; GFX9-NEXT: ; return to shader part epilog
;
; GFX10-LABEL: add3_multiuse_inner:
; GFX10: ; %bb.0:
; GFX10-NEXT: v_add_nc_u32_e32 v0, v0, v1
; GFX10-NEXT: v_add_nc_u32_e32 v1, v0, v2
; GFX10-NEXT: ; return to shader part epilog
%inner = add i32 %a, %b
%outer = add i32 %inner, %c
%r1 = insertelement <2 x i32> undef, i32 %inner, i32 0
%r0 = insertelement <2 x i32> %r1, i32 %outer, i32 1
%bc = bitcast <2 x i32> %r0 to <2 x float>
ret <2 x float> %bc
}
; A case where uniform values end up in VGPRs -- we could use v_add3_u32 here,
; but we don't.
define amdgpu_ps float @add3_uniform_vgpr(float inreg %a, float inreg %b, float inreg %c) {
; VI-LABEL: add3_uniform_vgpr:
; VI: ; %bb.0:
; VI-NEXT: v_add_f32_e64 v0, s2, 1.0
; VI-NEXT: v_add_f32_e64 v1, s3, 2.0
; VI-NEXT: v_mov_b32_e32 v2, 0x40400000
; VI-NEXT: v_add_f32_e32 v2, s4, v2
; VI-NEXT: v_add_u32_e32 v0, vcc, v0, v1
; VI-NEXT: v_add_u32_e32 v0, vcc, v0, v2
; VI-NEXT: ; return to shader part epilog
;
; GFX9-LABEL: add3_uniform_vgpr:
; GFX9: ; %bb.0:
; GFX9-NEXT: v_add_f32_e64 v0, s2, 1.0
; GFX9-NEXT: v_add_f32_e64 v1, s3, 2.0
; GFX9-NEXT: v_mov_b32_e32 v2, 0x40400000
; GFX9-NEXT: v_add_f32_e32 v2, s4, v2
; GFX9-NEXT: v_add_u32_e32 v0, v0, v1
; GFX9-NEXT: v_add_u32_e32 v0, v0, v2
; GFX9-NEXT: ; return to shader part epilog
;
; GFX10-LABEL: add3_uniform_vgpr:
; GFX10: ; %bb.0:
; GFX10-NEXT: v_add_f32_e64 v0, s2, 1.0
; GFX10-NEXT: v_add_f32_e64 v1, s3, 2.0
; GFX10-NEXT: v_add_f32_e64 v2, 0x40400000, s4
; GFX10-NEXT: v_add_nc_u32_e32 v0, v0, v1
; GFX10-NEXT: v_add_nc_u32_e32 v0, v0, v2
; GFX10-NEXT: ; return to shader part epilog
%a1 = fadd float %a, 1.0
%b2 = fadd float %b, 2.0
%c3 = fadd float %c, 3.0
%bc.a = bitcast float %a1 to i32
%bc.b = bitcast float %b2 to i32
%bc.c = bitcast float %c3 to i32
%x = add i32 %bc.a, %bc.b
%result = add i32 %x, %bc.c
%bc = bitcast i32 %result to float
ret float %bc
}
|