partword.ll
7.03 KB
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
; RUN: opt -S %s -atomic-expand | FileCheck %s
;; Verify the cmpxchg and atomicrmw expansions where sub-word-size
;; instructions are not available.
;;; NOTE: this test is mostly target-independent -- any target which
;;; doesn't support cmpxchg of sub-word sizes would do.
target datalayout = "E-m:e-i64:64-n32:64-S128"
target triple = "sparcv9-unknown-unknown"
; CHECK-LABEL: @test_cmpxchg_i8(
; CHECK: fence seq_cst
; CHECK: %0 = ptrtoint i8* %arg to i64
; CHECK: %1 = and i64 %0, -4
; CHECK: %AlignedAddr = inttoptr i64 %1 to i32*
; CHECK: %PtrLSB = and i64 %0, 3
; CHECK: %2 = xor i64 %PtrLSB, 3
; CHECK: %3 = shl i64 %2, 3
; CHECK: %ShiftAmt = trunc i64 %3 to i32
; CHECK: %Mask = shl i32 255, %ShiftAmt
; CHECK: %Inv_Mask = xor i32 %Mask, -1
; CHECK: %4 = zext i8 %new to i32
; CHECK: %5 = shl i32 %4, %ShiftAmt
; CHECK: %6 = zext i8 %old to i32
; CHECK: %7 = shl i32 %6, %ShiftAmt
; CHECK: %8 = load i32, i32* %AlignedAddr
; CHECK: %9 = and i32 %8, %Inv_Mask
; CHECK: br label %partword.cmpxchg.loop
; CHECK:partword.cmpxchg.loop:
; CHECK: %10 = phi i32 [ %9, %entry ], [ %16, %partword.cmpxchg.failure ]
; CHECK: %11 = or i32 %10, %5
; CHECK: %12 = or i32 %10, %7
; CHECK: %13 = cmpxchg i32* %AlignedAddr, i32 %12, i32 %11 monotonic monotonic
; CHECK: %14 = extractvalue { i32, i1 } %13, 0
; CHECK: %15 = extractvalue { i32, i1 } %13, 1
; CHECK: br i1 %15, label %partword.cmpxchg.end, label %partword.cmpxchg.failure
; CHECK:partword.cmpxchg.failure:
; CHECK: %16 = and i32 %14, %Inv_Mask
; CHECK: %17 = icmp ne i32 %10, %16
; CHECK: br i1 %17, label %partword.cmpxchg.loop, label %partword.cmpxchg.end
; CHECK:partword.cmpxchg.end:
; CHECK: %18 = lshr i32 %14, %ShiftAmt
; CHECK: %19 = trunc i32 %18 to i8
; CHECK: %20 = insertvalue { i8, i1 } undef, i8 %19, 0
; CHECK: %21 = insertvalue { i8, i1 } %20, i1 %15, 1
; CHECK: fence seq_cst
; CHECK: %ret = extractvalue { i8, i1 } %21, 0
; CHECK: ret i8 %ret
define i8 @test_cmpxchg_i8(i8* %arg, i8 %old, i8 %new) {
entry:
%ret_succ = cmpxchg i8* %arg, i8 %old, i8 %new seq_cst monotonic
%ret = extractvalue { i8, i1 } %ret_succ, 0
ret i8 %ret
}
; CHECK-LABEL: @test_cmpxchg_i16(
; CHECK: fence seq_cst
; CHECK: %0 = ptrtoint i16* %arg to i64
; CHECK: %1 = and i64 %0, -4
; CHECK: %AlignedAddr = inttoptr i64 %1 to i32*
; CHECK: %PtrLSB = and i64 %0, 3
; CHECK: %2 = xor i64 %PtrLSB, 2
; CHECK: %3 = shl i64 %2, 3
; CHECK: %ShiftAmt = trunc i64 %3 to i32
; CHECK: %Mask = shl i32 65535, %ShiftAmt
; CHECK: %Inv_Mask = xor i32 %Mask, -1
; CHECK: %4 = zext i16 %new to i32
; CHECK: %5 = shl i32 %4, %ShiftAmt
; CHECK: %6 = zext i16 %old to i32
; CHECK: %7 = shl i32 %6, %ShiftAmt
; CHECK: %8 = load i32, i32* %AlignedAddr
; CHECK: %9 = and i32 %8, %Inv_Mask
; CHECK: br label %partword.cmpxchg.loop
; CHECK:partword.cmpxchg.loop:
; CHECK: %10 = phi i32 [ %9, %entry ], [ %16, %partword.cmpxchg.failure ]
; CHECK: %11 = or i32 %10, %5
; CHECK: %12 = or i32 %10, %7
; CHECK: %13 = cmpxchg i32* %AlignedAddr, i32 %12, i32 %11 monotonic monotonic
; CHECK: %14 = extractvalue { i32, i1 } %13, 0
; CHECK: %15 = extractvalue { i32, i1 } %13, 1
; CHECK: br i1 %15, label %partword.cmpxchg.end, label %partword.cmpxchg.failure
; CHECK:partword.cmpxchg.failure:
; CHECK: %16 = and i32 %14, %Inv_Mask
; CHECK: %17 = icmp ne i32 %10, %16
; CHECK: br i1 %17, label %partword.cmpxchg.loop, label %partword.cmpxchg.end
; CHECK:partword.cmpxchg.end:
; CHECK: %18 = lshr i32 %14, %ShiftAmt
; CHECK: %19 = trunc i32 %18 to i16
; CHECK: %20 = insertvalue { i16, i1 } undef, i16 %19, 0
; CHECK: %21 = insertvalue { i16, i1 } %20, i1 %15, 1
; CHECK: fence seq_cst
; CHECK: %ret = extractvalue { i16, i1 } %21, 0
; CHECK: ret i16 %ret
define i16 @test_cmpxchg_i16(i16* %arg, i16 %old, i16 %new) {
entry:
%ret_succ = cmpxchg i16* %arg, i16 %old, i16 %new seq_cst monotonic
%ret = extractvalue { i16, i1 } %ret_succ, 0
ret i16 %ret
}
; CHECK-LABEL: @test_add_i16(
; CHECK: fence seq_cst
; CHECK: %0 = ptrtoint i16* %arg to i64
; CHECK: %1 = and i64 %0, -4
; CHECK: %AlignedAddr = inttoptr i64 %1 to i32*
; CHECK: %PtrLSB = and i64 %0, 3
; CHECK: %2 = xor i64 %PtrLSB, 2
; CHECK: %3 = shl i64 %2, 3
; CHECK: %ShiftAmt = trunc i64 %3 to i32
; CHECK: %Mask = shl i32 65535, %ShiftAmt
; CHECK: %Inv_Mask = xor i32 %Mask, -1
; CHECK: %4 = zext i16 %val to i32
; CHECK: %ValOperand_Shifted = shl i32 %4, %ShiftAmt
; CHECK: %5 = load i32, i32* %AlignedAddr, align 4
; CHECK: br label %atomicrmw.start
; CHECK:atomicrmw.start:
; CHECK: %loaded = phi i32 [ %5, %entry ], [ %newloaded, %atomicrmw.start ]
; CHECK: %new = add i32 %loaded, %ValOperand_Shifted
; CHECK: %6 = and i32 %new, %Mask
; CHECK: %7 = and i32 %loaded, %Inv_Mask
; CHECK: %8 = or i32 %7, %6
; CHECK: %9 = cmpxchg i32* %AlignedAddr, i32 %loaded, i32 %8 monotonic monotonic
; CHECK: %success = extractvalue { i32, i1 } %9, 1
; CHECK: %newloaded = extractvalue { i32, i1 } %9, 0
; CHECK: br i1 %success, label %atomicrmw.end, label %atomicrmw.start
; CHECK:atomicrmw.end:
; CHECK: %10 = lshr i32 %newloaded, %ShiftAmt
; CHECK: %11 = trunc i32 %10 to i16
; CHECK: fence seq_cst
; CHECK: ret i16 %11
define i16 @test_add_i16(i16* %arg, i16 %val) {
entry:
%ret = atomicrmw add i16* %arg, i16 %val seq_cst
ret i16 %ret
}
; CHECK-LABEL: @test_xor_i16(
; (I'm going to just assert on the bits that differ from add, above.)
; CHECK:atomicrmw.start:
; CHECK: %new = xor i32 %loaded, %ValOperand_Shifted
; CHECK: %6 = cmpxchg i32* %AlignedAddr, i32 %loaded, i32 %new monotonic monotonic
; CHECK:atomicrmw.end:
define i16 @test_xor_i16(i16* %arg, i16 %val) {
entry:
%ret = atomicrmw xor i16* %arg, i16 %val seq_cst
ret i16 %ret
}
; CHECK-LABEL: @test_or_i16(
; (I'm going to just assert on the bits that differ from add, above.)
; CHECK:atomicrmw.start:
; CHECK: %new = or i32 %loaded, %ValOperand_Shifted
; CHECK: %6 = cmpxchg i32* %AlignedAddr, i32 %loaded, i32 %new monotonic monotonic
; CHECK:atomicrmw.end:
define i16 @test_or_i16(i16* %arg, i16 %val) {
entry:
%ret = atomicrmw or i16* %arg, i16 %val seq_cst
ret i16 %ret
}
; CHECK-LABEL: @test_and_i16(
; (I'm going to just assert on the bits that differ from add, above.)
; CHECK: %AndOperand = or i32 %Inv_Mask, %ValOperand_Shifted
; CHECK:atomicrmw.start:
; CHECK: %new = and i32 %loaded, %AndOperand
; CHECK: %6 = cmpxchg i32* %AlignedAddr, i32 %loaded, i32 %new monotonic monotonic
; CHECK:atomicrmw.end:
define i16 @test_and_i16(i16* %arg, i16 %val) {
entry:
%ret = atomicrmw and i16* %arg, i16 %val seq_cst
ret i16 %ret
}
; CHECK-LABEL: @test_min_i16(
; CHECK:atomicrmw.start:
; CHECK: %6 = lshr i32 %loaded, %ShiftAmt
; CHECK: %7 = trunc i32 %6 to i16
; CHECK: %8 = icmp sle i16 %7, %val
; CHECK: %new = select i1 %8, i16 %7, i16 %val
; CHECK: %9 = zext i16 %new to i32
; CHECK: %10 = shl i32 %9, %ShiftAmt
; CHECK: %11 = and i32 %loaded, %Inv_Mask
; CHECK: %12 = or i32 %11, %10
; CHECK: %13 = cmpxchg i32* %AlignedAddr, i32 %loaded, i32 %12 monotonic monotonic
; CHECK:atomicrmw.end:
define i16 @test_min_i16(i16* %arg, i16 %val) {
entry:
%ret = atomicrmw min i16* %arg, i16 %val seq_cst
ret i16 %ret
}