• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1; RUN: llc < %s -asm-verbose=false -verify-machineinstrs -disable-wasm-fallthrough-return-opt -wasm-disable-explicit-locals -wasm-keep-registers -mattr=+bulk-memory | FileCheck %s --check-prefixes CHECK,BULK-MEM
2; RUN: llc < %s -asm-verbose=false -verify-machineinstrs -disable-wasm-fallthrough-return-opt -wasm-disable-explicit-locals -wasm-keep-registers -mattr=-bulk-memory | FileCheck %s --check-prefixes CHECK,NO-BULK-MEM
3
4; Test that basic bulk memory codegen works correctly
5
6target datalayout = "e-m:e-p:32:32-i64:64-n32:64-S128"
7target triple = "wasm64-unknown-unknown"
8
9declare void @llvm.memcpy.p0i8.p0i8.i8(i8*, i8*, i8, i1)
10declare void @llvm.memcpy.p0i8.p0i8.i64(i8*, i8*, i64, i1)
11declare void @llvm.memcpy.p0i32.p0i32.i64(i32*, i32*, i64, i1)
12
13declare void @llvm.memmove.p0i8.p0i8.i8(i8*, i8*, i8, i1)
14declare void @llvm.memmove.p0i8.p0i8.i64(i8*, i8*, i64, i1)
15declare void @llvm.memmove.p0i32.p0i32.i64(i32*, i32*, i64, i1)
16
17declare void @llvm.memset.p0i8.i8(i8*, i8, i8, i1)
18declare void @llvm.memset.p0i8.i64(i8*, i8, i64, i1)
19declare void @llvm.memset.p0i32.i64(i32*, i8, i64, i1)
20
21; CHECK-LABEL: memcpy_i8:
22; NO-BULK-MEM-NOT: memory.copy
23; BULK-MEM-NEXT: .functype memcpy_i8 (i64, i64, i32) -> ()
24; BULK-MEM-NEXT: i64.extend_i32_u $push0=, $2
25; BULK-MEM-NEXT: memory.copy 0, 0, $0, $1, $pop0
26; BULK-MEM-NEXT: return
27define void @memcpy_i8(i8* %dest, i8* %src, i8 zeroext %len) {
28  call void @llvm.memcpy.p0i8.p0i8.i8(i8* %dest, i8* %src, i8 %len, i1 0)
29  ret void
30}
31
32; CHECK-LABEL: memmove_i8:
33; NO-BULK-MEM-NOT: memory.copy
34; BULK-MEM-NEXT: .functype memmove_i8 (i64, i64, i32) -> ()
35; BULK-MEM-NEXT: i64.extend_i32_u $push0=, $2
36; BULK-MEM-NEXT: memory.copy 0, 0, $0, $1, $pop0
37; BULK-MEM-NEXT: return
38define void @memmove_i8(i8* %dest, i8* %src, i8 zeroext %len) {
39  call void @llvm.memmove.p0i8.p0i8.i8(i8* %dest, i8* %src, i8 %len, i1 0)
40  ret void
41}
42
43; CHECK-LABEL: memset_i8:
44; NO-BULK-MEM-NOT: memory.fill
45; BULK-MEM-NEXT: .functype memset_i8 (i64, i32, i32) -> ()
46; BULK-MEM-NEXT: i64.extend_i32_u $push0=, $2
47; BULK-MEM-NEXT: memory.fill 0, $0, $1, $pop0
48; BULK-MEM-NEXT: return
49define void @memset_i8(i8* %dest, i8 %val, i8 zeroext %len) {
50  call void @llvm.memset.p0i8.i8(i8* %dest, i8 %val, i8 %len, i1 0)
51  ret void
52}
53
54; CHECK-LABEL: memcpy_i32:
55; NO-BULK-MEM-NOT: memory.copy
56; BULK-MEM-NEXT: .functype memcpy_i32 (i64, i64, i64) -> ()
57; BULK-MEM-NEXT: memory.copy 0, 0, $0, $1, $2
58; BULK-MEM-NEXT: return
59define void @memcpy_i32(i32* %dest, i32* %src, i64 %len) {
60  call void @llvm.memcpy.p0i32.p0i32.i64(i32* %dest, i32* %src, i64 %len, i1 0)
61  ret void
62}
63
64; CHECK-LABEL: memmove_i32:
65; NO-BULK-MEM-NOT: memory.copy
66; BULK-MEM-NEXT: .functype memmove_i32 (i64, i64, i64) -> ()
67; BULK-MEM-NEXT: memory.copy 0, 0, $0, $1, $2
68; BULK-MEM-NEXT: return
69define void @memmove_i32(i32* %dest, i32* %src, i64 %len) {
70  call void @llvm.memmove.p0i32.p0i32.i64(i32* %dest, i32* %src, i64 %len, i1 0)
71  ret void
72}
73
74; CHECK-LABEL: memset_i32:
75; NO-BULK-MEM-NOT: memory.fill
76; BULK-MEM-NEXT: .functype memset_i32 (i64, i32, i64) -> ()
77; BULK-MEM-NEXT: memory.fill 0, $0, $1, $2
78; BULK-MEM-NEXT: return
79define void @memset_i32(i32* %dest, i8 %val, i64 %len) {
80  call void @llvm.memset.p0i32.i64(i32* %dest, i8 %val, i64 %len, i1 0)
81  ret void
82}
83
84; CHECK-LABEL: memcpy_1:
85; CHECK-NEXT: .functype memcpy_1 (i64, i64) -> ()
86; CHECK-NEXT: i32.load8_u $push[[L0:[0-9]+]]=, 0($1)
87; CHECK-NEXT: i32.store8 0($0), $pop[[L0]]
88; CHECK-NEXT: return
89define void @memcpy_1(i8* %dest, i8* %src) {
90  call void @llvm.memcpy.p0i8.p0i8.i64(i8* %dest, i8* %src, i64 1, i1 0)
91  ret void
92}
93
94; CHECK-LABEL: memmove_1:
95; CHECK-NEXT: .functype memmove_1 (i64, i64) -> ()
96; CHECK-NEXT: i32.load8_u $push[[L0:[0-9]+]]=, 0($1)
97; CHECK-NEXT: i32.store8 0($0), $pop[[L0]]
98; CHECK-NEXT: return
99define void @memmove_1(i8* %dest, i8* %src) {
100  call void @llvm.memmove.p0i8.p0i8.i64(i8* %dest, i8* %src, i64 1, i1 0)
101  ret void
102}
103
104; CHECK-LABEL: memset_1:
105; NO-BULK-MEM-NOT: memory.fill
106; BULK-MEM-NEXT: .functype memset_1 (i64, i32) -> ()
107; BULK-MEM-NEXT: i32.store8 0($0), $1
108; BULK-MEM-NEXT: return
109define void @memset_1(i8* %dest, i8 %val) {
110  call void @llvm.memset.p0i8.i64(i8* %dest, i8 %val, i64 1, i1 0)
111  ret void
112}
113
114; CHECK-LABEL: memcpy_1024:
115; NO-BULK-MEM-NOT: memory.copy
116; BULK-MEM-NEXT: .functype memcpy_1024 (i64, i64) -> ()
117; BULK-MEM-NEXT: i64.const $push[[L0:[0-9]+]]=, 1024
118; BULK-MEM-NEXT: memory.copy 0, 0, $0, $1, $pop[[L0]]
119; BULK-MEM-NEXT: return
120define void @memcpy_1024(i8* %dest, i8* %src) {
121  call void @llvm.memcpy.p0i8.p0i8.i64(i8* %dest, i8* %src, i64 1024, i1 0)
122  ret void
123}
124
125; CHECK-LABEL: memmove_1024:
126; NO-BULK-MEM-NOT: memory.copy
127; BULK-MEM-NEXT: .functype memmove_1024 (i64, i64) -> ()
128; BULK-MEM-NEXT: i64.const $push[[L0:[0-9]+]]=, 1024
129; BULK-MEM-NEXT: memory.copy 0, 0, $0, $1, $pop[[L0]]
130; BULK-MEM-NEXT: return
131define void @memmove_1024(i8* %dest, i8* %src) {
132  call void @llvm.memmove.p0i8.p0i8.i64(i8* %dest, i8* %src, i64 1024, i1 0)
133  ret void
134}
135
136; CHECK-LABEL: memset_1024:
137; NO-BULK-MEM-NOT: memory.fill
138; BULK-MEM-NEXT: .functype memset_1024 (i64, i32) -> ()
139; BULK-MEM-NEXT: i64.const $push[[L0:[0-9]+]]=, 1024
140; BULK-MEM-NEXT: memory.fill 0, $0, $1, $pop[[L0]]
141; BULK-MEM-NEXT: return
142define void @memset_1024(i8* %dest, i8 %val) {
143  call void @llvm.memset.p0i8.i64(i8* %dest, i8 %val, i64 1024, i1 0)
144  ret void
145}
146
147; The following tests check that frame index elimination works for
148; bulk memory instructions. The stack pointer is bumped by 112 instead
149; of 100 because the stack pointer in WebAssembly is currently always
150; 16-byte aligned, even in leaf functions, although it is not written
151; back to the global in this case.
152
153; TODO: Change TransientStackAlignment to 1 to avoid this extra
154; arithmetic. This will require forcing the use of StackAlignment in
155; PrologEpilogEmitter.cpp when
156; WebAssemblyFrameLowering::needsSPWriteback would be true.
157
158; CHECK-LABEL: memcpy_alloca_src:
159; NO-BULK-MEM-NOT: memory.copy
160; BULK-MEM-NEXT: .functype memcpy_alloca_src (i64) -> ()
161; BULK-MEM-NEXT: global.get $push[[L0:[0-9]+]]=, __stack_pointer
162; BULK-MEM-NEXT: i64.const $push[[L1:[0-9]+]]=, 112
163; BULK-MEM-NEXT: i64.sub $push[[L2:[0-9]+]]=, $pop[[L0]], $pop[[L1]]
164; BULK-MEM-NEXT: i64.const $push[[L3:[0-9]+]]=, 12
165; BULK-MEM-NEXT: i64.add $push[[L4:[0-9]+]]=, $pop[[L2]], $pop[[L3]]
166; BULK-MEM-NEXT: i64.const $push[[L5:[0-9]+]]=, 100
167; BULK-MEM-NEXT: memory.copy 0, 0, $0, $pop[[L4]], $pop[[L5]]
168; BULK-MEM-NEXT: return
169define void @memcpy_alloca_src(i8* %dst) {
170  %a = alloca [100 x i8]
171  %p = bitcast [100 x i8]* %a to i8*
172  call void @llvm.memcpy.p0i8.p0i8.i64(i8* %dst, i8* %p, i64 100, i1 false)
173  ret void
174}
175
176; CHECK-LABEL: memcpy_alloca_dst:
177; NO-BULK-MEM-NOT: memory.copy
178; BULK-MEM-NEXT: .functype memcpy_alloca_dst (i64) -> ()
179; BULK-MEM-NEXT: global.get $push[[L0:[0-9]+]]=, __stack_pointer
180; BULK-MEM-NEXT: i64.const $push[[L1:[0-9]+]]=, 112
181; BULK-MEM-NEXT: i64.sub $push[[L2:[0-9]+]]=, $pop[[L0]], $pop[[L1]]
182; BULK-MEM-NEXT: i64.const $push[[L3:[0-9]+]]=, 12
183; BULK-MEM-NEXT: i64.add $push[[L4:[0-9]+]]=, $pop[[L2]], $pop[[L3]]
184; BULK-MEM-NEXT: i64.const $push[[L5:[0-9]+]]=, 100
185; BULK-MEM-NEXT: memory.copy 0, 0, $pop[[L4]], $0, $pop[[L5]]
186; BULK-MEM-NEXT: return
187define void @memcpy_alloca_dst(i8* %src) {
188  %a = alloca [100 x i8]
189  %p = bitcast [100 x i8]* %a to i8*
190  call void @llvm.memcpy.p0i8.p0i8.i64(i8* %p, i8* %src, i64 100, i1 false)
191  ret void
192}
193
194; CHECK-LABEL: memset_alloca:
195; NO-BULK-MEM-NOT: memory.fill
196; BULK-MEM-NEXT: .functype memset_alloca (i32) -> ()
197; BULK-MEM-NEXT: global.get $push[[L0:[0-9]+]]=, __stack_pointer
198; BULK-MEM-NEXT: i64.const $push[[L1:[0-9]+]]=, 112
199; BULK-MEM-NEXT: i64.sub $push[[L2:[0-9]+]]=, $pop[[L0]], $pop[[L1]]
200; BULK-MEM-NEXT: i64.const $push[[L3:[0-9]+]]=, 12
201; BULK-MEM-NEXT: i64.add $push[[L4:[0-9]+]]=, $pop[[L2]], $pop[[L3]]
202; BULK-MEM-NEXT: i64.const $push[[L5:[0-9]+]]=, 100
203; BULK-MEM-NEXT: memory.fill 0, $pop[[L4]], $0, $pop[[L5]]
204; BULK-MEM-NEXT: return
205define void @memset_alloca(i8 %val) {
206  %a = alloca [100 x i8]
207  %p = bitcast [100 x i8]* %a to i8*
208  call void @llvm.memset.p0i8.i64(i8* %p, i8 %val, i64 100, i1 false)
209  ret void
210}
211