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
|
From 7946cf61b738bcad220710feb6208912e6a105e6 Mon Sep 17 00:00:00 2001
From: Geoff Berry <gberry@codeaurora.org>
Date: Thu, 21 Sep 2017 21:10:06 +0000
Subject: [PATCH] [AArch64] Fix bug in store of vector 0 DAGCombine.
Summary:
Avoid using XZR/WZR directly as operands to split stores of zero
vectors. Doing so can lead to the XZR/WZR being used by an instruction
that doesn't allow it (e.g. add).
Fixes bug 34674.
Reviewers: t.p.northover, efriedma, MatzeB
Subscribers: aemerson, rengolin, javed.absar, mcrosier, eraman, llvm-commits, kristof.beyls
Differential Revision: https://reviews.llvm.org/D38146
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@313916 91177308-0d34-0410-b5e6-96231b3b80d8
---
lib/Target/AArch64/AArch64ISelLowering.cpp | 19 ++++++++++++++-----
test/CodeGen/AArch64/arm64-memset-inline.ll | 4 ++--
test/CodeGen/AArch64/fastcc.ll | 12 +++++++++---
test/CodeGen/AArch64/ldst-opt.ll | 15 ++++++++++++++-
4 files changed, 39 insertions(+), 11 deletions(-)
diff --git a/lib/Target/AArch64/AArch64ISelLowering.cpp b/lib/Target/AArch64/AArch64ISelLowering.cpp
index cfee36d3477..ff9bf2a7daf 100644
--- a/lib/Target/AArch64/AArch64ISelLowering.cpp
+++ b/lib/Target/AArch64/AArch64ISelLowering.cpp
@@ -9448,11 +9448,20 @@ static SDValue replaceZeroVectorStore(SelectionDAG &DAG, StoreSDNode &St) {
return SDValue();
}
- // Use WZR/XZR here to prevent DAGCombiner::MergeConsecutiveStores from
- // undoing this transformation.
- SDValue SplatVal = VT.getVectorElementType().getSizeInBits() == 32
- ? DAG.getRegister(AArch64::WZR, MVT::i32)
- : DAG.getRegister(AArch64::XZR, MVT::i64);
+ // Use a CopyFromReg WZR/XZR here to prevent
+ // DAGCombiner::MergeConsecutiveStores from undoing this transformation.
+ SDLoc DL(&St);
+ unsigned ZeroReg;
+ EVT ZeroVT;
+ if (VT.getVectorElementType().getSizeInBits() == 32) {
+ ZeroReg = AArch64::WZR;
+ ZeroVT = MVT::i32;
+ } else {
+ ZeroReg = AArch64::XZR;
+ ZeroVT = MVT::i64;
+ }
+ SDValue SplatVal =
+ DAG.getCopyFromReg(DAG.getEntryNode(), DL, ZeroReg, ZeroVT);
return splitStoreSplat(DAG, St, SplatVal, NumVecElts);
}
diff --git a/test/CodeGen/AArch64/arm64-memset-inline.ll b/test/CodeGen/AArch64/arm64-memset-inline.ll
index 384aaa8541d..8c872cc6150 100644
--- a/test/CodeGen/AArch64/arm64-memset-inline.ll
+++ b/test/CodeGen/AArch64/arm64-memset-inline.ll
@@ -12,9 +12,9 @@ entry:
define void @t2() nounwind ssp {
entry:
; CHECK-LABEL: t2:
+; CHECK: stp xzr, xzr, [sp, #16]
; CHECK: strh wzr, [sp, #32]
-; CHECK: stp xzr, xzr, [sp, #8]
-; CHECK: str xzr, [sp, #24]
+; CHECK: str xzr, [sp, #8]
%buf = alloca [26 x i8], align 1
%0 = getelementptr inbounds [26 x i8], [26 x i8]* %buf, i32 0, i32 0
call void @llvm.memset.p0i8.i32(i8* %0, i8 0, i32 26, i32 1, i1 false)
diff --git a/test/CodeGen/AArch64/fastcc.ll b/test/CodeGen/AArch64/fastcc.ll
index fcc852263b4..3ea6df5be49 100644
--- a/test/CodeGen/AArch64/fastcc.ll
+++ b/test/CodeGen/AArch64/fastcc.ll
@@ -21,9 +21,11 @@ define fastcc void @func_stack0() {
call fastcc void @func_stack8([8 x i32] undef, i32 42)
; CHECK: bl func_stack8
; CHECK-NOT: sub sp, sp,
+; CHECK-NOT: [sp, #{{[-0-9]+}}]!
+; CHECK-NOT: [sp], #{{[-0-9]+}}
; CHECK-TAIL: bl func_stack8
-; CHECK-TAIL: sub sp, sp, #16
+; CHECK-TAIL: stp xzr, xzr, [sp, #-16]!
call fastcc void @func_stack32([8 x i32] undef, i128 0, i128 9)
@@ -72,10 +74,12 @@ define fastcc void @func_stack8([8 x i32], i32 %stacked) {
call fastcc void @func_stack8([8 x i32] undef, i32 42)
; CHECK: bl func_stack8
; CHECK-NOT: sub sp, sp,
+; CHECK-NOT: [sp, #{{[-0-9]+}}]!
+; CHECK-NOT: [sp], #{{[-0-9]+}}
; CHECK-TAIL: bl func_stack8
-; CHECK-TAIL: sub sp, sp, #16
+; CHECK-TAIL: stp xzr, xzr, [sp, #-16]!
call fastcc void @func_stack32([8 x i32] undef, i128 0, i128 9)
@@ -116,9 +120,11 @@ define fastcc void @func_stack32([8 x i32], i128 %stacked0, i128 %stacked1) {
call fastcc void @func_stack8([8 x i32] undef, i32 42)
; CHECK: bl func_stack8
; CHECK-NOT: sub sp, sp,
+; CHECK-NOT: [sp, #{{[-0-9]+}}]!
+; CHECK-NOT: [sp], #{{[-0-9]+}}
; CHECK-TAIL: bl func_stack8
-; CHECK-TAIL: sub sp, sp, #16
+; CHECK-TAIL: stp xzr, xzr, [sp, #-16]!
call fastcc void @func_stack32([8 x i32] undef, i128 0, i128 9)
diff --git a/test/CodeGen/AArch64/ldst-opt.ll b/test/CodeGen/AArch64/ldst-opt.ll
index 9307b6a3e47..e416dcb0f16 100644
--- a/test/CodeGen/AArch64/ldst-opt.ll
+++ b/test/CodeGen/AArch64/ldst-opt.ll
@@ -1667,4 +1667,17 @@ entry:
ret void
}
-
+; Check for bug 34674 where invalid add of xzr was being generated.
+; CHECK-LABEL: bug34674:
+; CHECK: // %entry
+; CHECK-NEXT: mov [[ZREG:x[0-9]+]], xzr
+; CHECK-DAG: stp [[ZREG]], [[ZREG]], [x0]
+; CHECK-DAG: add x{{[0-9]+}}, [[ZREG]], #1
+define i64 @bug34674(<2 x i64>* %p) {
+entry:
+ store <2 x i64> zeroinitializer, <2 x i64>* %p
+ %p2 = bitcast <2 x i64>* %p to i64*
+ %ld = load i64, i64* %p2
+ %add = add i64 %ld, 1
+ ret i64 %add
+}
--
2.14.1
|
