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
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
|
; RUN: llc -amdgpu-scalarize-global-loads=false -march=amdgcn -mcpu=verde -verify-machineinstrs < %s | FileCheck -check-prefixes=GCN,SI,FUNC %s
; RUN: llc -amdgpu-scalarize-global-loads=false -march=amdgcn -mcpu=tonga -mattr=-flat-for-global -verify-machineinstrs < %s | FileCheck -check-prefixes=GCN,VI,FUNC %s
; RUN: llc -amdgpu-scalarize-global-loads=false -march=r600 -mcpu=redwood < %s | FileCheck -check-prefixes=EG,FUNC %s
; mul24 and mad24 are affected
; FUNC-LABEL: {{^}}test_mul_v2i32:
; EG: MULLO_INT {{\*? *}}T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
; EG: MULLO_INT {{\*? *}}T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
; GCN: v_mul_lo_i32 v{{[0-9]+, v[0-9]+, v[0-9]+}}
; GCN: v_mul_lo_i32 v{{[0-9]+, v[0-9]+, v[0-9]+}}
define amdgpu_kernel void @test_mul_v2i32(<2 x i32> addrspace(1)* %out, <2 x i32> addrspace(1)* %in) {
%b_ptr = getelementptr <2 x i32>, <2 x i32> addrspace(1)* %in, i32 1
%a = load <2 x i32>, <2 x i32> addrspace(1) * %in
%b = load <2 x i32>, <2 x i32> addrspace(1) * %b_ptr
%result = mul <2 x i32> %a, %b
store <2 x i32> %result, <2 x i32> addrspace(1)* %out
ret void
}
; FUNC-LABEL: {{^}}v_mul_v4i32:
; EG: MULLO_INT {{\*? *}}T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
; EG: MULLO_INT {{\*? *}}T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
; EG: MULLO_INT {{\*? *}}T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
; EG: MULLO_INT {{\*? *}}T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
; GCN: v_mul_lo_i32 v{{[0-9]+, v[0-9]+, v[0-9]+}}
; GCN: v_mul_lo_i32 v{{[0-9]+, v[0-9]+, v[0-9]+}}
; GCN: v_mul_lo_i32 v{{[0-9]+, v[0-9]+, v[0-9]+}}
; GCN: v_mul_lo_i32 v{{[0-9]+, v[0-9]+, v[0-9]+}}
define amdgpu_kernel void @v_mul_v4i32(<4 x i32> addrspace(1)* %out, <4 x i32> addrspace(1)* %in) {
%b_ptr = getelementptr <4 x i32>, <4 x i32> addrspace(1)* %in, i32 1
%a = load <4 x i32>, <4 x i32> addrspace(1) * %in
%b = load <4 x i32>, <4 x i32> addrspace(1) * %b_ptr
%result = mul <4 x i32> %a, %b
store <4 x i32> %result, <4 x i32> addrspace(1)* %out
ret void
}
; FUNC-LABEL: {{^}}s_trunc_i64_mul_to_i32:
; GCN: s_load_dword
; GCN: s_load_dword
; GCN: s_mul_i32
; GCN: buffer_store_dword
define amdgpu_kernel void @s_trunc_i64_mul_to_i32(i32 addrspace(1)* %out, i64 %a, i64 %b) {
%mul = mul i64 %b, %a
%trunc = trunc i64 %mul to i32
store i32 %trunc, i32 addrspace(1)* %out, align 8
ret void
}
; FUNC-LABEL: {{^}}v_trunc_i64_mul_to_i32:
; GCN: s_load_dword
; GCN: s_load_dword
; GCN: v_mul_lo_i32
; GCN: buffer_store_dword
define amdgpu_kernel void @v_trunc_i64_mul_to_i32(i32 addrspace(1)* %out, i64 addrspace(1)* %aptr, i64 addrspace(1)* %bptr) nounwind {
%a = load i64, i64 addrspace(1)* %aptr, align 8
%b = load i64, i64 addrspace(1)* %bptr, align 8
%mul = mul i64 %b, %a
%trunc = trunc i64 %mul to i32
store i32 %trunc, i32 addrspace(1)* %out, align 8
ret void
}
; This 64-bit multiply should just use MUL_HI and MUL_LO, since the top
; 32-bits of both arguments are sign bits.
; FUNC-LABEL: {{^}}mul64_sext_c:
; EG-DAG: MULLO_INT
; EG-DAG: MULHI_INT
; GCN-DAG: s_mul_i32
; GCN-DAG: v_mul_hi_i32
define amdgpu_kernel void @mul64_sext_c(i64 addrspace(1)* %out, i32 %in) {
entry:
%0 = sext i32 %in to i64
%1 = mul i64 %0, 80
store i64 %1, i64 addrspace(1)* %out
ret void
}
; FUNC-LABEL: {{^}}v_mul64_sext_c:
; EG-DAG: MULLO_INT
; EG-DAG: MULHI_INT
; GCN-DAG: v_mul_lo_i32
; GCN-DAG: v_mul_hi_i32
; GCN: s_endpgm
define amdgpu_kernel void @v_mul64_sext_c(i64 addrspace(1)* %out, i32 addrspace(1)* %in) {
%val = load i32, i32 addrspace(1)* %in, align 4
%ext = sext i32 %val to i64
%mul = mul i64 %ext, 80
store i64 %mul, i64 addrspace(1)* %out, align 8
ret void
}
; FUNC-LABEL: {{^}}v_mul64_sext_inline_imm:
; GCN-DAG: v_mul_lo_i32 v{{[0-9]+}}, v{{[0-9]+}}, 9
; GCN-DAG: v_mul_hi_i32 v{{[0-9]+}}, v{{[0-9]+}}, 9
; GCN: s_endpgm
define amdgpu_kernel void @v_mul64_sext_inline_imm(i64 addrspace(1)* %out, i32 addrspace(1)* %in) {
%val = load i32, i32 addrspace(1)* %in, align 4
%ext = sext i32 %val to i64
%mul = mul i64 %ext, 9
store i64 %mul, i64 addrspace(1)* %out, align 8
ret void
}
; FUNC-LABEL: {{^}}s_mul_i32:
; GCN: s_load_dword [[SRC0:s[0-9]+]],
; GCN: s_load_dword [[SRC1:s[0-9]+]],
; GCN: s_mul_i32 [[SRESULT:s[0-9]+]], [[SRC0]], [[SRC1]]
; GCN: v_mov_b32_e32 [[VRESULT:v[0-9]+]], [[SRESULT]]
; GCN: buffer_store_dword [[VRESULT]],
; GCN: s_endpgm
define amdgpu_kernel void @s_mul_i32(i32 addrspace(1)* %out, i32 %a, i32 %b) nounwind {
%mul = mul i32 %a, %b
store i32 %mul, i32 addrspace(1)* %out, align 4
ret void
}
; FUNC-LABEL: {{^}}v_mul_i32:
; GCN: v_mul_lo_i32 v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}
define amdgpu_kernel void @v_mul_i32(i32 addrspace(1)* %out, i32 addrspace(1)* %in) {
%b_ptr = getelementptr i32, i32 addrspace(1)* %in, i32 1
%a = load i32, i32 addrspace(1)* %in
%b = load i32, i32 addrspace(1)* %b_ptr
%result = mul i32 %a, %b
store i32 %result, i32 addrspace(1)* %out
ret void
}
; A standard 64-bit multiply. The expansion should be around 6 instructions.
; It would be difficult to match the expansion correctly without writing
; a really complicated list of FileCheck expressions. I don't want
; to confuse people who may 'break' this test with a correct optimization,
; so this test just uses FUNC-LABEL to make sure the compiler does not
; crash with a 'failed to select' error.
; FUNC-LABEL: {{^}}s_mul_i64:
define amdgpu_kernel void @s_mul_i64(i64 addrspace(1)* %out, i64 %a, i64 %b) nounwind {
%mul = mul i64 %a, %b
store i64 %mul, i64 addrspace(1)* %out, align 8
ret void
}
; FUNC-LABEL: {{^}}v_mul_i64:
; GCN: v_mul_lo_i32
define amdgpu_kernel void @v_mul_i64(i64 addrspace(1)* %out, i64 addrspace(1)* %aptr, i64 addrspace(1)* %bptr) {
%a = load i64, i64 addrspace(1)* %aptr, align 8
%b = load i64, i64 addrspace(1)* %bptr, align 8
%mul = mul i64 %a, %b
store i64 %mul, i64 addrspace(1)* %out, align 8
ret void
}
; FUNC-LABEL: {{^}}mul32_in_branch:
; GCN: s_mul_i32
define amdgpu_kernel void @mul32_in_branch(i32 addrspace(1)* %out, i32 addrspace(1)* %in, i32 %a, i32 %b, i32 %c) {
entry:
%0 = icmp eq i32 %a, 0
br i1 %0, label %if, label %else
if:
%1 = load i32, i32 addrspace(1)* %in
br label %endif
else:
%2 = mul i32 %a, %b
br label %endif
endif:
%3 = phi i32 [%1, %if], [%2, %else]
store i32 %3, i32 addrspace(1)* %out
ret void
}
; FUNC-LABEL: {{^}}mul64_in_branch:
; GCN-DAG: s_mul_i32
; GCN-DAG: v_mul_hi_u32
; GCN: s_endpgm
define amdgpu_kernel void @mul64_in_branch(i64 addrspace(1)* %out, i64 addrspace(1)* %in, i64 %a, i64 %b, i64 %c) {
entry:
%0 = icmp eq i64 %a, 0
br i1 %0, label %if, label %else
if:
%1 = load i64, i64 addrspace(1)* %in
br label %endif
else:
%2 = mul i64 %a, %b
br label %endif
endif:
%3 = phi i64 [%1, %if], [%2, %else]
store i64 %3, i64 addrspace(1)* %out
ret void
}
; FIXME: Load dwordx4
; FUNC-LABEL: {{^}}s_mul_i128:
; GCN: s_load_dwordx2
; GCN: s_load_dwordx2
; GCN: s_load_dwordx2
; GCN: s_load_dwordx2
; SI: v_mul_hi_u32
; SI: v_mul_hi_u32
; SI: s_mul_i32
; SI: v_mul_hi_u32
; SI: s_mul_i32
; SI-DAG: s_mul_i32
; SI-DAG: v_mul_hi_u32
; SI-DAG: v_mul_hi_u32
; SI-DAG: s_mul_i32
; SI-DAG: s_mul_i32
; SI-DAG: v_mul_hi_u32
; SI: s_mul_i32
; SI: s_mul_i32
; SI: s_mul_i32
; SI: s_mul_i32
; SI: s_mul_i32
; VI: s_mul_i32
; VI: v_mul_hi_u32
; VI: v_mad_u64_u32
; VI: s_mul_i32
; VI: v_mul_hi_u32
; VI: v_mad_u64_u32
; VI: v_mad_u64_u32
; GCN: buffer_store_dwordx4
define amdgpu_kernel void @s_mul_i128(i128 addrspace(1)* %out, i128 %a, i128 %b) nounwind #0 {
%mul = mul i128 %a, %b
store i128 %mul, i128 addrspace(1)* %out
ret void
}
; FUNC-LABEL: {{^}}v_mul_i128:
; GCN: {{buffer|flat}}_load_dwordx4
; GCN: {{buffer|flat}}_load_dwordx4
; GCN-DAG: v_mul_lo_i32
; GCN-DAG: v_mul_hi_u32
; GCN-DAG: v_mul_hi_u32
; GCN-DAG: v_mul_lo_i32
; GCN-DAG: v_mul_hi_u32
; GCN-DAG: v_mul_hi_u32
; GCN-DAG: v_mul_lo_i32
; GCN-DAG: v_mul_lo_i32
; GCN-DAG: v_add_i32_e32
; SI-DAG: v_mul_hi_u32
; SI-DAG: v_mul_lo_i32
; SI-DAG: v_mul_hi_u32
; SI-DAG: v_mul_lo_i32
; SI-DAG: v_mul_lo_i32
; SI-DAG: v_mul_lo_i32
; SI-DAG: v_mul_lo_i32
; SI-DAG: v_mul_lo_i32
; VI-DAG: v_mad_u64_u32
; VI: v_mad_u64_u32
; VI: v_mad_u64_u32
; GCN: {{buffer|flat}}_store_dwordx4
define amdgpu_kernel void @v_mul_i128(i128 addrspace(1)* %out, i128 addrspace(1)* %aptr, i128 addrspace(1)* %bptr) #0 {
%tid = call i32 @llvm.r600.read.tidig.x()
%gep.a = getelementptr inbounds i128, i128 addrspace(1)* %aptr, i32 %tid
%gep.b = getelementptr inbounds i128, i128 addrspace(1)* %bptr, i32 %tid
%gep.out = getelementptr inbounds i128, i128 addrspace(1)* %bptr, i32 %tid
%a = load i128, i128 addrspace(1)* %gep.a
%b = load i128, i128 addrspace(1)* %gep.b
%mul = mul i128 %a, %b
store i128 %mul, i128 addrspace(1)* %gep.out
ret void
}
declare i32 @llvm.r600.read.tidig.x() #1
attributes #0 = { nounwind }
attributes #1 = { nounwind readnone}
|
