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
|
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc < %s -mtriple=x86_64-pc-linux-gnu -mattr=+zu | FileCheck %s --check-prefixes=CHECK,ZU
; RUN: llc < %s -mtriple=x86_64-pc-linux-gnu | FileCheck %s --check-prefixes=CHECK,NOZU
; Test generation of 16b imulzu when -mattr=+zu is specified.
; The mulzu_* tests check for basic generation, which is limited to cases where a
; zero-extend of the result can be folded into imulzu.
; The remaining tests are modifications of selected test/CodeGen/X86/imul.ll tests with
; 16b multiplies, to check that common strength reductions in ISel are still performed
; when -mattr=+zu is in effect.
;
; FIXME: several cases from imul.ll covering DAG combines, in particular those using LEA,
; are not ported as X86's IsDesirableToPromoteOp has no way to accurately identify when
; promotion will permit a better sequence than an unpromoted imulzu.
; These cases should be added when they are implemented.
define i32 @mulzu_16_32(i16 %A) {
; ZU-LABEL: mulzu_16_32:
; ZU: # %bb.0:
; ZU-NEXT: imulzuw $1234, %di, %ax # imm = 0x4D2
; ZU-NEXT: retq
;
; NOZU-LABEL: mulzu_16_32:
; NOZU: # %bb.0:
; NOZU-NEXT: imull $1234, %edi, %eax # imm = 0x4D2
; NOZU-NEXT: movzwl %ax, %eax
; NOZU-NEXT: retq
%mul = mul i16 %A, 1234
%r = zext i16 %mul to i32
ret i32 %r
}
define i64 @mulzu_16_64(i16 %A) {
; ZU-LABEL: mulzu_16_64:
; ZU: # %bb.0:
; ZU-NEXT: imulzuw $1234, %di, %ax # imm = 0x4D2
; ZU-NEXT: retq
;
; NOZU-LABEL: mulzu_16_64:
; NOZU: # %bb.0:
; NOZU-NEXT: imull $1234, %edi, %eax # imm = 0x4D2
; NOZU-NEXT: movzwl %ax, %eax
; NOZU-NEXT: retq
%mul = mul i16 %A, 1234
%r = zext i16 %mul to i64
ret i64 %r
}
define i32 @mulzu_16_32_mem(ptr %P) {
; ZU-LABEL: mulzu_16_32_mem:
; ZU: # %bb.0:
; ZU-NEXT: imulzuw $1234, (%rdi), %ax # imm = 0x4D2
; ZU-NEXT: retq
;
; NOZU-LABEL: mulzu_16_32_mem:
; NOZU: # %bb.0:
; NOZU-NEXT: movzwl (%rdi), %eax
; NOZU-NEXT: imull $1234, %eax, %eax # imm = 0x4D2
; NOZU-NEXT: movzwl %ax, %eax
; NOZU-NEXT: retq
%gep = getelementptr i16, ptr %P, i64 0
%A = load i16, ptr %gep
%mul = mul i16 %A, 1234
%r = zext i16 %mul to i32
ret i32 %r
}
define i64 @mulzu_16_64_mem(ptr %P) {
; ZU-LABEL: mulzu_16_64_mem:
; ZU: # %bb.0:
; ZU-NEXT: imulzuw $1234, (%rdi), %ax # imm = 0x4D2
; ZU-NEXT: retq
;
; NOZU-LABEL: mulzu_16_64_mem:
; NOZU: # %bb.0:
; NOZU-NEXT: movzwl (%rdi), %eax
; NOZU-NEXT: imull $1234, %eax, %eax # imm = 0x4D2
; NOZU-NEXT: movzwl %ax, %eax
; NOZU-NEXT: retq
%gep = getelementptr i16, ptr %P, i64 0
%A = load i16, ptr %gep
%mul = mul i16 %A, 1234
%r = zext i16 %mul to i64
ret i64 %r
}
; The following mulzu cases check that imulzu is not
; generated in the absence of a single zext user. The ZU/NOZU
; cases should match.
define void @mulzu_16_store(i16 %A, ptr %R) {
; CHECK-LABEL: mulzu_16_store:
; CHECK: # %bb.0:
; CHECK-NEXT: imull $1234, %edi, %eax # imm = 0x4D2
; CHECK-NEXT: movw %ax, (%rsi)
; CHECK-NEXT: retq
%gep = getelementptr i16, ptr %R, i64 0
%mul = mul i16 %A, 1234
store i16 %mul, ptr %gep
ret void
}
define i32 @mulzu_16_store_32(i16 %A, ptr %R) {
; CHECK-LABEL: mulzu_16_store_32:
; CHECK: # %bb.0:
; CHECK-NEXT: imull $1234, %edi, %eax # imm = 0x4D2
; CHECK-NEXT: movw %ax, (%rsi)
; CHECK-NEXT: movzwl %ax, %eax
; CHECK-NEXT: retq
%gep = getelementptr i16, ptr %R, i64 0
%mul = mul i16 %A, 1234
store i16 %mul, ptr %gep
%r = zext i16 %mul to i32
ret i32 %r
}
define i64 @mulzu_16_store_64(i16 %A, ptr %R) {
; CHECK-LABEL: mulzu_16_store_64:
; CHECK: # %bb.0:
; CHECK-NEXT: imull $1234, %edi, %eax # imm = 0x4D2
; CHECK-NEXT: movw %ax, (%rsi)
; CHECK-NEXT: movzwl %ax, %eax
; CHECK-NEXT: retq
%gep = getelementptr i16, ptr %R, i64 0
%mul = mul i16 %A, 1234
store i16 %mul, ptr %gep
%r = zext i16 %mul to i64
ret i64 %r
}
define i32 @mulzu_sext_16_32(i16 %A) {
; CHECK-LABEL: mulzu_sext_16_32:
; CHECK: # %bb.0:
; CHECK-NEXT: imull $1234, %edi, %eax # imm = 0x4D2
; CHECK-NEXT: cwtl
; CHECK-NEXT: retq
%mul = mul i16 %A, 1234
%r = sext i16 %mul to i32
ret i32 %r
}
define i64 @mulzu_sext_16_64(i16 %A) {
; CHECK-LABEL: mulzu_sext_16_64:
; CHECK: # %bb.0:
; CHECK-NEXT: imull $1234, %edi, %eax # imm = 0x4D2
; CHECK-NEXT: movswq %ax, %rax
; CHECK-NEXT: retq
%mul = mul i16 %A, 1234
%r = sext i16 %mul to i64
ret i64 %r
}
; Tests ported from test/CodeGen/X86/imul.ll follow from this point.
; The generated code, which strength-reduces multiplies by certain
; constants, should be unaffected by enabling zu.
define i16 @mul4_16(i16 %A) {
;
; CHECK-LABEL: mul4_16:
; CHECK: # %bb.0:
; CHECK-NEXT: # kill: def $edi killed $edi def $rdi
; CHECK-NEXT: leal (,%rdi,4), %eax
; CHECK-NEXT: # kill: def $ax killed $ax killed $eax
; CHECK-NEXT: retq
%mul = mul i16 %A, 4
ret i16 %mul
}
define i16 @mul4096_16(i16 %A) {
;
; CHECK-LABEL: mul4096_16:
; CHECK: # %bb.0:
; CHECK-NEXT: movl %edi, %eax
; CHECK-NEXT: shll $12, %eax
; CHECK-NEXT: # kill: def $ax killed $ax killed $eax
; CHECK-NEXT: retq
%mul = mul i16 %A, 4096
ret i16 %mul
}
define i16 @mulmin4096_16(i16 %A) {
;
; CHECK-LABEL: mulmin4096_16:
; CHECK: # %bb.0:
; CHECK-NEXT: movl %edi, %eax
; CHECK-NEXT: shll $12, %eax
; CHECK-NEXT: negl %eax
; CHECK-NEXT: # kill: def $ax killed $ax killed $eax
; CHECK-NEXT: retq
%mul = mul i16 %A, -4096
ret i16 %mul
}
define i16 @mul4_16_minsize(i16 %A) minsize {
;
; CHECK-LABEL: mul4_16_minsize:
; CHECK: # %bb.0:
; CHECK-NEXT: # kill: def $edi killed $edi def $rdi
; CHECK-NEXT: leal (,%rdi,4), %eax
; CHECK-NEXT: # kill: def $ax killed $ax killed $eax
; CHECK-NEXT: retq
%mul = mul i16 %A, 4
ret i16 %mul
}
define i16 @mul0_16(i16 %A) {
;
; CHECK-LABEL: mul0_16:
; CHECK: # %bb.0:
; CHECK-NEXT: xorl %eax, %eax
; CHECK-NEXT: retq
%mul = mul i16 %A, 0
ret i16 %mul
}
define i16 @mul4294967295_16(i16 %A) {
;
; CHECK-LABEL: mul4294967295_16:
; CHECK: # %bb.0:
; CHECK-NEXT: movl %edi, %eax
; CHECK-NEXT: negl %eax
; CHECK-NEXT: # kill: def $ax killed $ax killed $eax
; CHECK-NEXT: retq
%mul = mul i16 %A, 4294967295
ret i16 %mul
}
|
