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
|
; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
; RUN: opt -passes=indvars -S %s | FileCheck %s
target datalayout = "e-m:o-i64:64-i128:128-n32:64-S128"
define float @ashr_expansion_valid(i64 %x, ptr %ptr) {
; CHECK-LABEL: @ashr_expansion_valid(
; CHECK-NEXT: entry:
; CHECK-NEXT: [[BOUND:%.*]] = ashr exact i64 [[X:%.*]], 4
; CHECK-NEXT: [[UMAX:%.*]] = call i64 @llvm.umax.i64(i64 [[BOUND]], i64 1)
; CHECK-NEXT: br label [[LOOP:%.*]]
; CHECK: loop:
; CHECK-NEXT: [[IV:%.*]] = phi i64 [ 0, [[ENTRY:%.*]] ], [ [[IV_NEXT:%.*]], [[LOOP]] ]
; CHECK-NEXT: [[RED:%.*]] = phi float [ 0.000000e+00, [[ENTRY]] ], [ [[RED_NEXT:%.*]], [[LOOP]] ]
; CHECK-NEXT: [[GEP:%.*]] = getelementptr float, ptr [[PTR:%.*]], i64 [[IV]]
; CHECK-NEXT: [[LV:%.*]] = load float, ptr [[GEP]], align 4
; CHECK-NEXT: [[RED_NEXT]] = fadd float [[LV]], [[RED]]
; CHECK-NEXT: [[IV_NEXT]] = add nuw i64 [[IV]], 1
; CHECK-NEXT: [[EXITCOND:%.*]] = icmp ne i64 [[IV_NEXT]], [[UMAX]]
; CHECK-NEXT: br i1 [[EXITCOND]], label [[LOOP]], label [[EXIT:%.*]]
; CHECK: exit:
; CHECK-NEXT: [[LCSSA_RED_NEXT:%.*]] = phi float [ [[RED_NEXT]], [[LOOP]] ]
; CHECK-NEXT: ret float [[LCSSA_RED_NEXT]]
;
entry:
%bound = ashr exact i64 %x, 4
br label %loop
loop:
%iv = phi i64 [ 0, %entry ], [ %iv.next, %loop ]
%red = phi float [ 0.0, %entry ], [ %red.next, %loop ]
%gep = getelementptr float, ptr %ptr, i64 %iv
%lv = load float, ptr %gep
%red.next = fadd float %lv, %red
%iv.next = add nuw i64 %iv, 1
%cond = icmp ult i64 %iv.next, %bound
br i1 %cond, label %loop, label %exit
exit: ; preds = %bb135
%lcssa.red.next = phi float [ %red.next, %loop ]
ret float %lcssa.red.next
}
; No explicit ashr, but a chain of operations that can be replaced by ashr.
define float @ashr_equivalent_expansion(i64 %x, ptr %ptr) {
; CHECK-LABEL: @ashr_equivalent_expansion(
; CHECK-NEXT: entry:
; CHECK-NEXT: [[ABS_X:%.*]] = call i64 @llvm.abs.i64(i64 [[X:%.*]], i1 false)
; CHECK-NEXT: [[DIV:%.*]] = udiv exact i64 [[ABS_X]], 16
; CHECK-NEXT: [[T0:%.*]] = call i64 @llvm.smax.i64(i64 [[X]], i64 -1)
; CHECK-NEXT: [[T1:%.*]] = call i64 @llvm.smin.i64(i64 [[T0]], i64 1)
; CHECK-NEXT: [[BOUND:%.*]] = mul nsw i64 [[DIV]], [[T1]]
; CHECK-NEXT: [[UMAX:%.*]] = call i64 @llvm.umax.i64(i64 [[BOUND]], i64 1)
; CHECK-NEXT: br label [[LOOP:%.*]]
; CHECK: loop:
; CHECK-NEXT: [[IV:%.*]] = phi i64 [ 0, [[ENTRY:%.*]] ], [ [[IV_NEXT:%.*]], [[LOOP]] ]
; CHECK-NEXT: [[RED:%.*]] = phi float [ 0.000000e+00, [[ENTRY]] ], [ [[RED_NEXT:%.*]], [[LOOP]] ]
; CHECK-NEXT: [[GEP:%.*]] = getelementptr float, ptr [[PTR:%.*]], i64 [[IV]]
; CHECK-NEXT: [[LV:%.*]] = load float, ptr [[GEP]], align 4
; CHECK-NEXT: [[RED_NEXT]] = fadd float [[LV]], [[RED]]
; CHECK-NEXT: [[IV_NEXT]] = add nuw i64 [[IV]], 1
; CHECK-NEXT: [[EXITCOND:%.*]] = icmp ne i64 [[IV_NEXT]], [[UMAX]]
; CHECK-NEXT: br i1 [[EXITCOND]], label [[LOOP]], label [[EXIT:%.*]]
; CHECK: exit:
; CHECK-NEXT: [[LCSSA_RED_NEXT:%.*]] = phi float [ [[RED_NEXT]], [[LOOP]] ]
; CHECK-NEXT: ret float [[LCSSA_RED_NEXT]]
;
entry:
%abs_x = call i64 @llvm.abs.i64(i64 %x, i1 false)
%div = udiv exact i64 %abs_x, 16
%t0 = call i64 @llvm.smax.i64(i64 %x, i64 -1)
%t1 = call i64 @llvm.smin.i64(i64 %t0, i64 1)
%bound = mul nsw i64 %div, %t1
br label %loop
loop:
%iv = phi i64 [ 0, %entry ], [ %iv.next, %loop ]
%red = phi float [ 0.0, %entry ], [ %red.next, %loop ]
%gep = getelementptr float, ptr %ptr, i64 %iv
%lv = load float, ptr %gep
%red.next = fadd float %lv, %red
%iv.next = add nuw i64 %iv, 1
%cond = icmp ult i64 %iv.next, %bound
br i1 %cond, label %loop, label %exit
exit: ; preds = %bb135
%lcssa.red.next = phi float [ %red.next, %loop ]
ret float %lcssa.red.next
}
; Chain of operations that *cannot* be replaced by ashr, because the udiv is
; missing exact.
define float @no_ashr_due_to_missing_exact_udiv(i64 %x, ptr %ptr) {
; CHECK-LABEL: @no_ashr_due_to_missing_exact_udiv(
; CHECK-NEXT: entry:
; CHECK-NEXT: [[ABS_X:%.*]] = call i64 @llvm.abs.i64(i64 [[X:%.*]], i1 false)
; CHECK-NEXT: [[DIV:%.*]] = udiv i64 [[ABS_X]], 16
; CHECK-NEXT: [[T0:%.*]] = call i64 @llvm.smax.i64(i64 [[X]], i64 -1)
; CHECK-NEXT: [[T1:%.*]] = call i64 @llvm.smin.i64(i64 [[T0]], i64 1)
; CHECK-NEXT: [[BOUND:%.*]] = mul nsw i64 [[DIV]], [[T1]]
; CHECK-NEXT: [[UMAX:%.*]] = call i64 @llvm.umax.i64(i64 [[BOUND]], i64 1)
; CHECK-NEXT: br label [[LOOP:%.*]]
; CHECK: loop:
; CHECK-NEXT: [[IV:%.*]] = phi i64 [ 0, [[ENTRY:%.*]] ], [ [[IV_NEXT:%.*]], [[LOOP]] ]
; CHECK-NEXT: [[RED:%.*]] = phi float [ 0.000000e+00, [[ENTRY]] ], [ [[RED_NEXT:%.*]], [[LOOP]] ]
; CHECK-NEXT: [[GEP:%.*]] = getelementptr float, ptr [[PTR:%.*]], i64 [[IV]]
; CHECK-NEXT: [[LV:%.*]] = load float, ptr [[GEP]], align 4
; CHECK-NEXT: [[RED_NEXT]] = fadd float [[LV]], [[RED]]
; CHECK-NEXT: [[IV_NEXT]] = add nuw i64 [[IV]], 1
; CHECK-NEXT: [[EXITCOND:%.*]] = icmp ne i64 [[IV_NEXT]], [[UMAX]]
; CHECK-NEXT: br i1 [[EXITCOND]], label [[LOOP]], label [[EXIT:%.*]]
; CHECK: exit:
; CHECK-NEXT: [[LCSSA_RED_NEXT:%.*]] = phi float [ [[RED_NEXT]], [[LOOP]] ]
; CHECK-NEXT: ret float [[LCSSA_RED_NEXT]]
;
entry:
%abs_x = call i64 @llvm.abs.i64(i64 %x, i1 false)
%div = udiv i64 %abs_x, 16
%t0 = call i64 @llvm.smax.i64(i64 %x, i64 -1)
%t1 = call i64 @llvm.smin.i64(i64 %t0, i64 1)
%bound = mul nsw i64 %div, %t1
br label %loop
loop:
%iv = phi i64 [ 0, %entry ], [ %iv.next, %loop ]
%red = phi float [ 0.0, %entry ], [ %red.next, %loop ]
%gep = getelementptr float, ptr %ptr, i64 %iv
%lv = load float, ptr %gep
%red.next = fadd float %lv, %red
%iv.next = add nuw i64 %iv, 1
%cond = icmp ult i64 %iv.next, %bound
br i1 %cond, label %loop, label %exit
exit: ; preds = %bb135
%lcssa.red.next = phi float [ %red.next, %loop ]
ret float %lcssa.red.next
}
; Chain of operations that *cannot* be replaced by ashr, because abs and
; signum have different operands.
define float @no_ashr_due_to_different_ops(i64 %x, i64 %y, ptr %ptr) {
; CHECK-LABEL: @no_ashr_due_to_different_ops(
; CHECK-NEXT: entry:
; CHECK-NEXT: [[ABS_X:%.*]] = call i64 @llvm.abs.i64(i64 [[X:%.*]], i1 false)
; CHECK-NEXT: [[DIV:%.*]] = udiv i64 [[ABS_X]], 16
; CHECK-NEXT: [[T0:%.*]] = call i64 @llvm.smax.i64(i64 [[Y:%.*]], i64 -1)
; CHECK-NEXT: [[T1:%.*]] = call i64 @llvm.smin.i64(i64 [[T0]], i64 1)
; CHECK-NEXT: [[BOUND:%.*]] = mul nsw i64 [[DIV]], [[T1]]
; CHECK-NEXT: [[UMAX:%.*]] = call i64 @llvm.umax.i64(i64 [[BOUND]], i64 1)
; CHECK-NEXT: br label [[LOOP:%.*]]
; CHECK: loop:
; CHECK-NEXT: [[IV:%.*]] = phi i64 [ 0, [[ENTRY:%.*]] ], [ [[IV_NEXT:%.*]], [[LOOP]] ]
; CHECK-NEXT: [[RED:%.*]] = phi float [ 0.000000e+00, [[ENTRY]] ], [ [[RED_NEXT:%.*]], [[LOOP]] ]
; CHECK-NEXT: [[GEP:%.*]] = getelementptr float, ptr [[PTR:%.*]], i64 [[IV]]
; CHECK-NEXT: [[LV:%.*]] = load float, ptr [[GEP]], align 4
; CHECK-NEXT: [[RED_NEXT]] = fadd float [[LV]], [[RED]]
; CHECK-NEXT: [[IV_NEXT]] = add nuw i64 [[IV]], 1
; CHECK-NEXT: [[EXITCOND:%.*]] = icmp ne i64 [[IV_NEXT]], [[UMAX]]
; CHECK-NEXT: br i1 [[EXITCOND]], label [[LOOP]], label [[EXIT:%.*]]
; CHECK: exit:
; CHECK-NEXT: [[LCSSA_RED_NEXT:%.*]] = phi float [ [[RED_NEXT]], [[LOOP]] ]
; CHECK-NEXT: ret float [[LCSSA_RED_NEXT]]
;
entry:
%abs_x = call i64 @llvm.abs.i64(i64 %x, i1 false)
%div = udiv i64 %abs_x, 16
%t0 = call i64 @llvm.smax.i64(i64 %y, i64 -1)
%t1 = call i64 @llvm.smin.i64(i64 %t0, i64 1)
%bound = mul nsw i64 %div, %t1
br label %loop
loop:
%iv = phi i64 [ 0, %entry ], [ %iv.next, %loop ]
%red = phi float [ 0.0, %entry ], [ %red.next, %loop ]
%gep = getelementptr float, ptr %ptr, i64 %iv
%lv = load float, ptr %gep
%red.next = fadd float %lv, %red
%iv.next = add nuw i64 %iv, 1
%cond = icmp ult i64 %iv.next, %bound
br i1 %cond, label %loop, label %exit
exit: ; preds = %bb135
%lcssa.red.next = phi float [ %red.next, %loop ]
ret float %lcssa.red.next
}
declare i64 @llvm.abs.i64(i64, i1)
declare i64 @llvm.smax.i64(i64, i64)
declare i64 @llvm.smin.i64(i64, i64)
|
