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
|
; RUN: llc -verify-machineinstrs -O3 -mcpu=pwr7 < %s | FileCheck %s
; RUN: llc -verify-machineinstrs -O3 -print-after=codegenprepare -mcpu=pwr7 < %s >%t 2>&1 && FileCheck --check-prefix=CHECK-UseAA <%t %s
target datalayout = "E-m:e-i64:64-n32:64"
target triple = "powerpc64-unknown-linux-gnu"
; Following test cases test enabling SeparateConstOffsetFromGEP pass in the PPC
; backend. If useAA() returns true, it will lower a GEP with multiple indices
; into GEPs with a single index, otherwise it will lower it into a
; "ptrtoint+arithmetics+inttoptr" form.
%struct = type { i32, i32, i32, i32, [20 x i32] }
; Check that when two complex GEPs are used in two basic blocks, LLVM can
; elimilate the common subexpression for the second use.
define void @test_GEP_CSE(ptr %string, ptr %adj, i32 %lib, i64 %idxprom) {
%liberties = getelementptr [240 x %struct], ptr %string, i64 1, i64 %idxprom, i32 3
%1 = load i32, ptr %liberties, align 4
%cmp = icmp eq i32 %1, %lib
br i1 %cmp, label %if.then, label %if.end
if.then: ; preds = %entry
%origin = getelementptr [240 x %struct], ptr %string, i64 1, i64 %idxprom, i32 2
%2 = load i32, ptr %origin, align 4
store i32 %2, ptr %adj, align 4
br label %if.end
if.end: ; preds = %if.then, %entry
ret void
}
; CHECK-UseAA-LABEL: @test_GEP_CSE(
; CHECK-UseAA: [[IDX:%[a-zA-Z0-9]+]] = mul i64 %idxprom, 96
; CHECK-UseAA: [[PTR1:%[a-zA-Z0-9]+]] = getelementptr i8, ptr %string, i64 [[IDX]]
; CHECK-UseAA: getelementptr i8, ptr [[PTR1]], i64 23052
; CHECK-UseAA: if.then:
; CHECK-UseAA: getelementptr i8, ptr [[PTR1]], i64 23048
%class.my = type { i32, [128 x i32], i32, [256 x %struct.pt]}
%struct.pt = type { ptr, i32, i32 }
%struct.point = type { i32, i32 }
; Check when a GEP is used across two basic block, LLVM can sink the address
; calculation and code gen can generate a better addressing mode for the second
; use.
define void @test_GEP_across_BB(ptr %this, i64 %idx) {
%1 = getelementptr %class.my, ptr %this, i64 0, i32 3, i64 %idx, i32 1
%2 = load i32, ptr %1, align 4
%3 = getelementptr %class.my, ptr %this, i64 0, i32 3, i64 %idx, i32 2
%4 = load i32, ptr %3, align 4
%5 = icmp eq i32 %2, %4
br i1 %5, label %if.true, label %exit
if.true:
%6 = shl i32 %4, 1
store i32 %6, ptr %3, align 4
br label %exit
exit:
%7 = add nsw i32 %4, 1
store i32 %7, ptr %1, align 4
ret void
}
; CHECK-LABEL: test_GEP_across_BB:
; CHECK-NOT: lwzu
; CHECK: blr
; CHECK-UseAA-LABEL: test_GEP_across_BB(
; CHECK-UseAA: [[PTR0:%[a-zA-Z0-9]+]] = getelementptr
; CHECK-UseAA: getelementptr i8, ptr [[PTR0]], i64 528
; CHECK-UseAA: getelementptr i8, ptr [[PTR0]], i64 532
; CHECK-UseAA: if.true:
; CHECK-UseAA: {{%sunk[a-zA-Z0-9]+}} = getelementptr i8, ptr [[PTR0]], i64 532
; CHECK-UseAA: exit:
; CHECK-UseAA: {{%sunk[a-zA-Z0-9]+}} = getelementptr i8, ptr [[PTR0]], i64 528
%struct.S = type { float, double }
@struct_array = global [1024 x %struct.S] zeroinitializer, align 16
; The following two test cases check we can extract constant from indices of
; struct type.
; The constant offsets are from indices "i64 %idxprom" and "i32 1". As the
; alloca size of %struct.S is 16, and "i32 1" is the 2rd element whose field
; offset is 8, the total constant offset is (5 * 16 + 8) = 88.
define ptr @test-struct_1(i32 %i) {
entry:
%add = add nsw i32 %i, 5
%idxprom = sext i32 %add to i64
%p = getelementptr [1024 x %struct.S], ptr @struct_array, i64 0, i64 %idxprom, i32 1
ret ptr %p
}
; CHECK-UseAA-LABEL: @test-struct_1(
; CHECK-UseAA: getelementptr i8, ptr %{{[a-zA-Z0-9]+}}, i64 88
%struct3 = type { i64, i32 }
%struct2 = type { %struct3, i32 }
%struct1 = type { i64, %struct2 }
%struct0 = type { i32, i32, ptr, [100 x %struct1] }
; The constant offsets are from indices "i32 3", "i64 %arrayidx" and "i32 1".
; "i32 3" is the 4th element whose field offset is 16. The alloca size of
; %struct1 is 32. "i32 1" is the 2rd element whose field offset is 8. So the
; total constant offset is 16 + (-2 * 32) + 8 = -40
define ptr @test-struct_2(ptr %ptr, i64 %idx) {
entry:
%arrayidx = add nsw i64 %idx, -2
%ptr2 = getelementptr %struct0, ptr %ptr, i64 0, i32 3, i64 %arrayidx, i32 1
ret ptr %ptr2
}
; CHECK-UseAA-LABEL: @test-struct_2(
; CHECK-UseAA: getelementptr i8, ptr %{{[a-zA-Z0-9]+}}, i64 -40
; Test that when a index is added from two constant, SeparateConstOffsetFromGEP
; pass does not generate incorrect result.
define void @test_const_add(ptr %in) {
%inc = add nsw i32 2, 1
%idxprom = sext i32 %inc to i64
%arrayidx = getelementptr [3 x i32], ptr %in, i64 %idxprom, i64 2
store i32 0, ptr %arrayidx, align 4
ret void
}
; CHECK-LABEL: test_const_add:
; CHECK: li [[REG:[0-9]+]], 0
; CHECK: stw [[REG]], 44(3)
; CHECK: blr
|
