Skip to content

Commit e13e808

Browse files
dc03dc03
committed
[SROA] Limit the number of allowed slices when trying to split allocas
This patch adds a hidden CLI option "--sroa-max-alloca-slices", which is an integer that controls the maximum number of alloca slices SROA can consider before bailing out. This is useful because it may not be profitable to split memcpys into (possibly tens of) thousands of loads/stores. This also prevents an issue with exponential compile time explosion in passes like DSE and MemCpyOpt caused by excessive alloca splitting. Fixes rust-lang/rust#88580. Differential Revision: https://reviews.llvm.org/D159354
1 parent f807494 commit e13e808

File tree

2 files changed

+185
-0
lines changed

2 files changed

+185
-0
lines changed

‎llvm/lib/Transforms/Scalar/SROA.cpp

Lines changed: 11 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -121,6 +121,14 @@ static cl::opt<bool> SROAStrictInbounds("sroa-strict-inbounds", cl::init(false),
121121
/// Disable running mem2reg during SROA in order to test or debug SROA.
122122
static cl::opt<bool> SROASkipMem2Reg("sroa-skip-mem2reg", cl::init(false),
123123
cl::Hidden);
124+
125+
/// The maximum number of alloca slices allowed when splitting.
126+
static cl::opt<int>
127+
SROAMaxAllocaSlices("sroa-max-alloca-slices", cl::init(1024),
128+
cl::desc("Maximum number of alloca slices allowed "
129+
"after which splitting is not attempted"),
130+
cl::Hidden);
131+
124132
namespace {
125133

126134
/// Calculate the fragment of a variable to use when slicing a store
@@ -4961,6 +4969,9 @@ SROAPass::runOnAlloca(AllocaInst &AI) {
49614969
if (AS.isEscaped())
49624970
return {Changed, CFGChanged};
49634971

4972+
if (std::distance(AS.begin(), AS.end()) > SROAMaxAllocaSlices)
4973+
return {Changed, CFGChanged};
4974+
49644975
// Delete all the dead users of this alloca before splitting and rewriting it.
49654976
for (Instruction *DeadUser : AS.getDeadUsers()) {
49664977
// Free up everything used by this instruction.

‎llvm/test/Transforms/SROA/huge-size.ll

Lines changed: 174 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,174 @@
1+
; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 3
2+
; RUN: opt -passes='loop-unroll,sroa' --sroa-max-alloca-slices=16 -S -o - < %s | FileCheck %s
3+
4+
; (Very) Reduced from this Rust code:
5+
;
6+
; extern "C" {
7+
; fn _use(x: [[[Option::<usize>; 5]; 5]; 5]) -> bool;
8+
; }
9+
; fn main() {
10+
; let s = [[[Option::<usize>::None; 5]; 5]; 5];
11+
; unsafe {
12+
; _use(s);
13+
; }
14+
; }
15+
16+
define void @huge_size() {
17+
; CHECK-LABEL: define void @huge_size() {
18+
; CHECK-NEXT: start:
19+
; CHECK-NEXT: [[ARRAY:%.*]] = alloca [5 x [5 x [5 x { i64, i64 }]]], align 8
20+
; CHECK-NEXT: [[ARRAY_SUB_1:%.*]] = alloca [5 x [5 x { i64, i64 }]], align 8
21+
; CHECK-NEXT: br label [[LOOP_1:%.*]]
22+
; CHECK: loop.1:
23+
; CHECK-NEXT: br label [[LOOP_2:%.*]]
24+
; CHECK: loop.2:
25+
; CHECK-NEXT: store i64 0, ptr [[ARRAY_SUB_1]], align 1
26+
; CHECK-NEXT: [[ARRAY_SUB_2_SROA_6_0_ARRAY_SUB_1_SROA_IDX:%.*]] = getelementptr inbounds i8, ptr [[ARRAY_SUB_1]], i64 8
27+
; CHECK-NEXT: store i64 undef, ptr [[ARRAY_SUB_2_SROA_6_0_ARRAY_SUB_1_SROA_IDX]], align 1
28+
; CHECK-NEXT: [[ARRAY_SUB_2_SROA_69_0_ARRAY_SUB_1_SROA_IDX:%.*]] = getelementptr inbounds i8, ptr [[ARRAY_SUB_1]], i64 16
29+
; CHECK-NEXT: store i64 0, ptr [[ARRAY_SUB_2_SROA_69_0_ARRAY_SUB_1_SROA_IDX]], align 1
30+
; CHECK-NEXT: [[ARRAY_SUB_2_SROA_7_0_ARRAY_SUB_1_SROA_IDX:%.*]] = getelementptr inbounds i8, ptr [[ARRAY_SUB_1]], i64 24
31+
; CHECK-NEXT: store i64 undef, ptr [[ARRAY_SUB_2_SROA_7_0_ARRAY_SUB_1_SROA_IDX]], align 1
32+
; CHECK-NEXT: [[ARRAY_SUB_2_SROA_718_0_ARRAY_SUB_1_SROA_IDX:%.*]] = getelementptr inbounds i8, ptr [[ARRAY_SUB_1]], i64 32
33+
; CHECK-NEXT: store i64 0, ptr [[ARRAY_SUB_2_SROA_718_0_ARRAY_SUB_1_SROA_IDX]], align 1
34+
; CHECK-NEXT: [[ARRAY_SUB_2_SROA_8_0_ARRAY_SUB_1_SROA_IDX:%.*]] = getelementptr inbounds i8, ptr [[ARRAY_SUB_1]], i64 40
35+
; CHECK-NEXT: store i64 undef, ptr [[ARRAY_SUB_2_SROA_8_0_ARRAY_SUB_1_SROA_IDX]], align 1
36+
; CHECK-NEXT: [[ARRAY_SUB_2_SROA_827_0_ARRAY_SUB_1_SROA_IDX:%.*]] = getelementptr inbounds i8, ptr [[ARRAY_SUB_1]], i64 48
37+
; CHECK-NEXT: store i64 0, ptr [[ARRAY_SUB_2_SROA_827_0_ARRAY_SUB_1_SROA_IDX]], align 1
38+
; CHECK-NEXT: [[ARRAY_SUB_2_SROA_9_0_ARRAY_SUB_1_SROA_IDX:%.*]] = getelementptr inbounds i8, ptr [[ARRAY_SUB_1]], i64 56
39+
; CHECK-NEXT: store i64 undef, ptr [[ARRAY_SUB_2_SROA_9_0_ARRAY_SUB_1_SROA_IDX]], align 1
40+
; CHECK-NEXT: [[ARRAY_SUB_2_SROA_936_0_ARRAY_SUB_1_SROA_IDX:%.*]] = getelementptr inbounds i8, ptr [[ARRAY_SUB_1]], i64 64
41+
; CHECK-NEXT: store i64 0, ptr [[ARRAY_SUB_2_SROA_936_0_ARRAY_SUB_1_SROA_IDX]], align 1
42+
; CHECK-NEXT: [[ARRAY_SUB_2_SROA_10_0_ARRAY_SUB_1_SROA_IDX:%.*]] = getelementptr inbounds i8, ptr [[ARRAY_SUB_1]], i64 72
43+
; CHECK-NEXT: store i64 undef, ptr [[ARRAY_SUB_2_SROA_10_0_ARRAY_SUB_1_SROA_IDX]], align 1
44+
; CHECK-NEXT: [[GEP_2_1:%.*]] = getelementptr [5 x [5 x { i64, i64 }]], ptr [[ARRAY_SUB_1]], i64 0, i64 1
45+
; CHECK-NEXT: store i64 0, ptr [[GEP_2_1]], align 1
46+
; CHECK-NEXT: [[ARRAY_SUB_2_SROA_6_0_GEP_2_1_SROA_IDX:%.*]] = getelementptr inbounds i8, ptr [[GEP_2_1]], i64 8
47+
; CHECK-NEXT: store i64 undef, ptr [[ARRAY_SUB_2_SROA_6_0_GEP_2_1_SROA_IDX]], align 1
48+
; CHECK-NEXT: [[ARRAY_SUB_2_SROA_69_0_GEP_2_1_SROA_IDX:%.*]] = getelementptr inbounds i8, ptr [[GEP_2_1]], i64 16
49+
; CHECK-NEXT: store i64 0, ptr [[ARRAY_SUB_2_SROA_69_0_GEP_2_1_SROA_IDX]], align 1
50+
; CHECK-NEXT: [[ARRAY_SUB_2_SROA_7_0_GEP_2_1_SROA_IDX:%.*]] = getelementptr inbounds i8, ptr [[GEP_2_1]], i64 24
51+
; CHECK-NEXT: store i64 undef, ptr [[ARRAY_SUB_2_SROA_7_0_GEP_2_1_SROA_IDX]], align 1
52+
; CHECK-NEXT: [[ARRAY_SUB_2_SROA_718_0_GEP_2_1_SROA_IDX:%.*]] = getelementptr inbounds i8, ptr [[GEP_2_1]], i64 32
53+
; CHECK-NEXT: store i64 0, ptr [[ARRAY_SUB_2_SROA_718_0_GEP_2_1_SROA_IDX]], align 1
54+
; CHECK-NEXT: [[ARRAY_SUB_2_SROA_8_0_GEP_2_1_SROA_IDX:%.*]] = getelementptr inbounds i8, ptr [[GEP_2_1]], i64 40
55+
; CHECK-NEXT: store i64 undef, ptr [[ARRAY_SUB_2_SROA_8_0_GEP_2_1_SROA_IDX]], align 1
56+
; CHECK-NEXT: [[ARRAY_SUB_2_SROA_827_0_GEP_2_1_SROA_IDX:%.*]] = getelementptr inbounds i8, ptr [[GEP_2_1]], i64 48
57+
; CHECK-NEXT: store i64 0, ptr [[ARRAY_SUB_2_SROA_827_0_GEP_2_1_SROA_IDX]], align 1
58+
; CHECK-NEXT: [[ARRAY_SUB_2_SROA_9_0_GEP_2_1_SROA_IDX:%.*]] = getelementptr inbounds i8, ptr [[GEP_2_1]], i64 56
59+
; CHECK-NEXT: store i64 undef, ptr [[ARRAY_SUB_2_SROA_9_0_GEP_2_1_SROA_IDX]], align 1
60+
; CHECK-NEXT: [[ARRAY_SUB_2_SROA_936_0_GEP_2_1_SROA_IDX:%.*]] = getelementptr inbounds i8, ptr [[GEP_2_1]], i64 64
61+
; CHECK-NEXT: store i64 0, ptr [[ARRAY_SUB_2_SROA_936_0_GEP_2_1_SROA_IDX]], align 1
62+
; CHECK-NEXT: [[ARRAY_SUB_2_SROA_10_0_GEP_2_1_SROA_IDX:%.*]] = getelementptr inbounds i8, ptr [[GEP_2_1]], i64 72
63+
; CHECK-NEXT: store i64 undef, ptr [[ARRAY_SUB_2_SROA_10_0_GEP_2_1_SROA_IDX]], align 1
64+
; CHECK-NEXT: [[GEP_2_2:%.*]] = getelementptr [5 x [5 x { i64, i64 }]], ptr [[ARRAY_SUB_1]], i64 0, i64 2
65+
; CHECK-NEXT: store i64 0, ptr [[GEP_2_2]], align 1
66+
; CHECK-NEXT: [[ARRAY_SUB_2_SROA_6_0_GEP_2_2_SROA_IDX:%.*]] = getelementptr inbounds i8, ptr [[GEP_2_2]], i64 8
67+
; CHECK-NEXT: store i64 undef, ptr [[ARRAY_SUB_2_SROA_6_0_GEP_2_2_SROA_IDX]], align 1
68+
; CHECK-NEXT: [[ARRAY_SUB_2_SROA_69_0_GEP_2_2_SROA_IDX:%.*]] = getelementptr inbounds i8, ptr [[GEP_2_2]], i64 16
69+
; CHECK-NEXT: store i64 0, ptr [[ARRAY_SUB_2_SROA_69_0_GEP_2_2_SROA_IDX]], align 1
70+
; CHECK-NEXT: [[ARRAY_SUB_2_SROA_7_0_GEP_2_2_SROA_IDX:%.*]] = getelementptr inbounds i8, ptr [[GEP_2_2]], i64 24
71+
; CHECK-NEXT: store i64 undef, ptr [[ARRAY_SUB_2_SROA_7_0_GEP_2_2_SROA_IDX]], align 1
72+
; CHECK-NEXT: [[ARRAY_SUB_2_SROA_718_0_GEP_2_2_SROA_IDX:%.*]] = getelementptr inbounds i8, ptr [[GEP_2_2]], i64 32
73+
; CHECK-NEXT: store i64 0, ptr [[ARRAY_SUB_2_SROA_718_0_GEP_2_2_SROA_IDX]], align 1
74+
; CHECK-NEXT: [[ARRAY_SUB_2_SROA_8_0_GEP_2_2_SROA_IDX:%.*]] = getelementptr inbounds i8, ptr [[GEP_2_2]], i64 40
75+
; CHECK-NEXT: store i64 undef, ptr [[ARRAY_SUB_2_SROA_8_0_GEP_2_2_SROA_IDX]], align 1
76+
; CHECK-NEXT: [[ARRAY_SUB_2_SROA_827_0_GEP_2_2_SROA_IDX:%.*]] = getelementptr inbounds i8, ptr [[GEP_2_2]], i64 48
77+
; CHECK-NEXT: store i64 0, ptr [[ARRAY_SUB_2_SROA_827_0_GEP_2_2_SROA_IDX]], align 1
78+
; CHECK-NEXT: [[ARRAY_SUB_2_SROA_9_0_GEP_2_2_SROA_IDX:%.*]] = getelementptr inbounds i8, ptr [[GEP_2_2]], i64 56
79+
; CHECK-NEXT: store i64 undef, ptr [[ARRAY_SUB_2_SROA_9_0_GEP_2_2_SROA_IDX]], align 1
80+
; CHECK-NEXT: [[ARRAY_SUB_2_SROA_936_0_GEP_2_2_SROA_IDX:%.*]] = getelementptr inbounds i8, ptr [[GEP_2_2]], i64 64
81+
; CHECK-NEXT: store i64 0, ptr [[ARRAY_SUB_2_SROA_936_0_GEP_2_2_SROA_IDX]], align 1
82+
; CHECK-NEXT: [[ARRAY_SUB_2_SROA_10_0_GEP_2_2_SROA_IDX:%.*]] = getelementptr inbounds i8, ptr [[GEP_2_2]], i64 72
83+
; CHECK-NEXT: store i64 undef, ptr [[ARRAY_SUB_2_SROA_10_0_GEP_2_2_SROA_IDX]], align 1
84+
; CHECK-NEXT: [[GEP_2_3:%.*]] = getelementptr [5 x [5 x { i64, i64 }]], ptr [[ARRAY_SUB_1]], i64 0, i64 3
85+
; CHECK-NEXT: store i64 0, ptr [[GEP_2_3]], align 1
86+
; CHECK-NEXT: [[ARRAY_SUB_2_SROA_6_0_GEP_2_3_SROA_IDX:%.*]] = getelementptr inbounds i8, ptr [[GEP_2_3]], i64 8
87+
; CHECK-NEXT: store i64 undef, ptr [[ARRAY_SUB_2_SROA_6_0_GEP_2_3_SROA_IDX]], align 1
88+
; CHECK-NEXT: [[ARRAY_SUB_2_SROA_69_0_GEP_2_3_SROA_IDX:%.*]] = getelementptr inbounds i8, ptr [[GEP_2_3]], i64 16
89+
; CHECK-NEXT: store i64 0, ptr [[ARRAY_SUB_2_SROA_69_0_GEP_2_3_SROA_IDX]], align 1
90+
; CHECK-NEXT: [[ARRAY_SUB_2_SROA_7_0_GEP_2_3_SROA_IDX:%.*]] = getelementptr inbounds i8, ptr [[GEP_2_3]], i64 24
91+
; CHECK-NEXT: store i64 undef, ptr [[ARRAY_SUB_2_SROA_7_0_GEP_2_3_SROA_IDX]], align 1
92+
; CHECK-NEXT: [[ARRAY_SUB_2_SROA_718_0_GEP_2_3_SROA_IDX:%.*]] = getelementptr inbounds i8, ptr [[GEP_2_3]], i64 32
93+
; CHECK-NEXT: store i64 0, ptr [[ARRAY_SUB_2_SROA_718_0_GEP_2_3_SROA_IDX]], align 1
94+
; CHECK-NEXT: [[ARRAY_SUB_2_SROA_8_0_GEP_2_3_SROA_IDX:%.*]] = getelementptr inbounds i8, ptr [[GEP_2_3]], i64 40
95+
; CHECK-NEXT: store i64 undef, ptr [[ARRAY_SUB_2_SROA_8_0_GEP_2_3_SROA_IDX]], align 1
96+
; CHECK-NEXT: [[ARRAY_SUB_2_SROA_827_0_GEP_2_3_SROA_IDX:%.*]] = getelementptr inbounds i8, ptr [[GEP_2_3]], i64 48
97+
; CHECK-NEXT: store i64 0, ptr [[ARRAY_SUB_2_SROA_827_0_GEP_2_3_SROA_IDX]], align 1
98+
; CHECK-NEXT: [[ARRAY_SUB_2_SROA_9_0_GEP_2_3_SROA_IDX:%.*]] = getelementptr inbounds i8, ptr [[GEP_2_3]], i64 56
99+
; CHECK-NEXT: store i64 undef, ptr [[ARRAY_SUB_2_SROA_9_0_GEP_2_3_SROA_IDX]], align 1
100+
; CHECK-NEXT: [[ARRAY_SUB_2_SROA_936_0_GEP_2_3_SROA_IDX:%.*]] = getelementptr inbounds i8, ptr [[GEP_2_3]], i64 64
101+
; CHECK-NEXT: store i64 0, ptr [[ARRAY_SUB_2_SROA_936_0_GEP_2_3_SROA_IDX]], align 1
102+
; CHECK-NEXT: [[ARRAY_SUB_2_SROA_10_0_GEP_2_3_SROA_IDX:%.*]] = getelementptr inbounds i8, ptr [[GEP_2_3]], i64 72
103+
; CHECK-NEXT: store i64 undef, ptr [[ARRAY_SUB_2_SROA_10_0_GEP_2_3_SROA_IDX]], align 1
104+
; CHECK-NEXT: [[GEP_2_4:%.*]] = getelementptr [5 x [5 x { i64, i64 }]], ptr [[ARRAY_SUB_1]], i64 0, i64 4
105+
; CHECK-NEXT: store i64 0, ptr [[GEP_2_4]], align 1
106+
; CHECK-NEXT: [[ARRAY_SUB_2_SROA_6_0_GEP_2_4_SROA_IDX:%.*]] = getelementptr inbounds i8, ptr [[GEP_2_4]], i64 8
107+
; CHECK-NEXT: store i64 undef, ptr [[ARRAY_SUB_2_SROA_6_0_GEP_2_4_SROA_IDX]], align 1
108+
; CHECK-NEXT: [[ARRAY_SUB_2_SROA_69_0_GEP_2_4_SROA_IDX:%.*]] = getelementptr inbounds i8, ptr [[GEP_2_4]], i64 16
109+
; CHECK-NEXT: store i64 0, ptr [[ARRAY_SUB_2_SROA_69_0_GEP_2_4_SROA_IDX]], align 1
110+
; CHECK-NEXT: [[ARRAY_SUB_2_SROA_7_0_GEP_2_4_SROA_IDX:%.*]] = getelementptr inbounds i8, ptr [[GEP_2_4]], i64 24
111+
; CHECK-NEXT: store i64 undef, ptr [[ARRAY_SUB_2_SROA_7_0_GEP_2_4_SROA_IDX]], align 1
112+
; CHECK-NEXT: [[ARRAY_SUB_2_SROA_718_0_GEP_2_4_SROA_IDX:%.*]] = getelementptr inbounds i8, ptr [[GEP_2_4]], i64 32
113+
; CHECK-NEXT: store i64 0, ptr [[ARRAY_SUB_2_SROA_718_0_GEP_2_4_SROA_IDX]], align 1
114+
; CHECK-NEXT: [[ARRAY_SUB_2_SROA_8_0_GEP_2_4_SROA_IDX:%.*]] = getelementptr inbounds i8, ptr [[GEP_2_4]], i64 40
115+
; CHECK-NEXT: store i64 undef, ptr [[ARRAY_SUB_2_SROA_8_0_GEP_2_4_SROA_IDX]], align 1
116+
; CHECK-NEXT: [[ARRAY_SUB_2_SROA_827_0_GEP_2_4_SROA_IDX:%.*]] = getelementptr inbounds i8, ptr [[GEP_2_4]], i64 48
117+
; CHECK-NEXT: store i64 0, ptr [[ARRAY_SUB_2_SROA_827_0_GEP_2_4_SROA_IDX]], align 1
118+
; CHECK-NEXT: [[ARRAY_SUB_2_SROA_9_0_GEP_2_4_SROA_IDX:%.*]] = getelementptr inbounds i8, ptr [[GEP_2_4]], i64 56
119+
; CHECK-NEXT: store i64 undef, ptr [[ARRAY_SUB_2_SROA_9_0_GEP_2_4_SROA_IDX]], align 1
120+
; CHECK-NEXT: [[ARRAY_SUB_2_SROA_936_0_GEP_2_4_SROA_IDX:%.*]] = getelementptr inbounds i8, ptr [[GEP_2_4]], i64 64
121+
; CHECK-NEXT: store i64 0, ptr [[ARRAY_SUB_2_SROA_936_0_GEP_2_4_SROA_IDX]], align 1
122+
; CHECK-NEXT: [[ARRAY_SUB_2_SROA_10_0_GEP_2_4_SROA_IDX:%.*]] = getelementptr inbounds i8, ptr [[GEP_2_4]], i64 72
123+
; CHECK-NEXT: store i64 undef, ptr [[ARRAY_SUB_2_SROA_10_0_GEP_2_4_SROA_IDX]], align 1
124+
; CHECK-NEXT: br label [[LOOP_3:%.*]]
125+
; CHECK: loop.3:
126+
; CHECK-NEXT: call void @llvm.memcpy.p0.p0.i64(ptr [[ARRAY]], ptr [[ARRAY_SUB_1]], i64 1600, i1 false)
127+
; CHECK-NEXT: [[GEP_3_1:%.*]] = getelementptr [5 x [5 x [5 x { i64, i64 }]]], ptr [[ARRAY]], i64 0, i64 1
128+
; CHECK-NEXT: call void @llvm.memcpy.p0.p0.i64(ptr [[GEP_3_1]], ptr [[ARRAY_SUB_1]], i64 1600, i1 false)
129+
; CHECK-NEXT: [[GEP_3_2:%.*]] = getelementptr [5 x [5 x [5 x { i64, i64 }]]], ptr [[ARRAY]], i64 0, i64 2
130+
; CHECK-NEXT: call void @llvm.memcpy.p0.p0.i64(ptr [[GEP_3_2]], ptr [[ARRAY_SUB_1]], i64 1600, i1 false)
131+
; CHECK-NEXT: [[GEP_3_3:%.*]] = getelementptr [5 x [5 x [5 x { i64, i64 }]]], ptr [[ARRAY]], i64 0, i64 3
132+
; CHECK-NEXT: call void @llvm.memcpy.p0.p0.i64(ptr [[GEP_3_3]], ptr [[ARRAY_SUB_1]], i64 1600, i1 false)
133+
; CHECK-NEXT: [[GEP_3_4:%.*]] = getelementptr [5 x [5 x [5 x { i64, i64 }]]], ptr [[ARRAY]], i64 0, i64 4
134+
; CHECK-NEXT: call void @llvm.memcpy.p0.p0.i64(ptr [[GEP_3_4]], ptr [[ARRAY_SUB_1]], i64 1600, i1 false)
135+
; CHECK-NEXT: ret void
136+
;
137+
start:
138+
%array = alloca [5 x [5 x [5 x { i64, i64 }]]], align 8
139+
%array.sub.1 = alloca [5 x [5 x { i64, i64 }]], align 8
140+
%array.sub.2 = alloca [5 x { i64, i64 }], align 8
141+
br label %loop.1
142+
143+
; Set up %array.sub.2
144+
loop.1:
145+
%ind.1 = phi i64 [ 0, %start ], [ %ind.1.next, %loop.1 ]
146+
%gep.1 = getelementptr [5 x { i64, i64 }], ptr %array.sub.2, i64 0, i64 %ind.1
147+
store i64 0, ptr %gep.1, align 8
148+
%ind.1.next = add i64 %ind.1, 1
149+
%loop1.cond = icmp ult i64 %ind.1.next, 5
150+
br i1 %loop1.cond, label %loop.1, label %loop.2
151+
152+
; Set up %array.sub.1
153+
loop.2:
154+
%ind.2 = phi i64 [ 0, %loop.1 ], [ %ind.2.next, %loop.2 ]
155+
%gep.2 = getelementptr [5 x [5 x { i64, i64 }]], ptr %array.sub.1, i64 0, i64 %ind.2
156+
call void @llvm.memcpy.p0.p0.i64(ptr %gep.2, ptr %array.sub.2, i64 160, i1 false)
157+
%ind.2.next = add i64 %ind.2, 1
158+
%loop.2.cond = icmp ult i64 %ind.2.next, 5
159+
br i1 %loop.2.cond, label %loop.2, label %loop.3
160+
161+
; Set up %array
162+
loop.3:
163+
%ind.3 = phi i64 [ 0, %loop.2 ], [ %ind.3.next, %loop.3 ]
164+
%gep.3 = getelementptr [5 x [5 x [5 x { i64, i64 }]]], ptr %array, i64 0, i64 %ind.3
165+
call void @llvm.memcpy.p0.p0.i64(ptr %gep.3, ptr %array.sub.1, i64 1600, i1 false)
166+
%ind.3.next = add i64 %ind.3, 1
167+
%loop.3.cond = icmp ult i64 %ind.3.next, 5
168+
br i1 %loop.3.cond, label %loop.3, label %exit
169+
170+
exit:
171+
ret void
172+
}
173+
174+
declare void @llvm.memcpy.p0.p0.i64(ptr, ptr, i64, i1)

0 commit comments

Comments
 (0)