1; RUN: llc -mtriple=thumbv6m-eabi -disable-fp-elim=false %s -o - | FileCheck %s 2 3; struct S { int x[128]; } s; 4; int f(int *, int, int, int, struct S); 5; int g(int *, int, int, int, int, int); 6; int h(int *, int *, int *); 7; int u(int *, int *, int *, struct S, struct S); 8 9%struct.S = type { [128 x i32] } 10%struct.__va_list = type { i8* } 11 12@s = common dso_local global %struct.S zeroinitializer, align 4 13 14declare void @llvm.va_start(i8*) 15declare dso_local i32 @g(i32*, i32, i32, i32, i32, i32) local_unnamed_addr 16declare dso_local i32 @f(i32*, i32, i32, i32, %struct.S* byval align 4) local_unnamed_addr 17declare dso_local i32 @h(i32*, i32*, i32*) local_unnamed_addr 18declare dso_local i32 @u(i32*, i32*, i32*, %struct.S* byval align 4, %struct.S* byval align 4) local_unnamed_addr 19 20; 21; Test access to arguments, passed on stack (including varargs) 22; 23 24; Usual case, access via SP 25; int test_args_sp(int a, int b, int c, int d, int e) { 26; int v[4]; 27; return g(v, a, b, c, d, e); 28; } 29define dso_local i32 @test_args_sp(i32 %a, i32 %b, i32 %c, i32 %d, i32 %e) local_unnamed_addr { 30entry: 31 %v = alloca [4 x i32], align 4 32 %0 = bitcast [4 x i32]* %v to i8* 33 %arraydecay = getelementptr inbounds [4 x i32], [4 x i32]* %v, i32 0, i32 0 34 %call = call i32 @g(i32* nonnull %arraydecay, i32 %a, i32 %b, i32 %c, i32 %d, i32 %e) 35 ret i32 %call 36} 37; CHECK-LABEL: test_args_sp 38; Load `e` 39; CHECK: ldr r0, [sp, #40] 40; CHECK-NEXT: mov r5, sp 41; CHECK-NEXT: str r3, [r5] 42; Pass `e` on stack 43; CHECK-NEXT: str r0, [r5, #4] 44; CHECK: bl g 45 46; int test_varargs_sp(int a, ...) { 47; int v[4]; 48; __builtin_va_list ap; 49; __builtin_va_start(ap, a); 50; return g(v, a, 0, 0, 0, 0); 51; } 52define dso_local i32 @test_varargs_sp(i32 %a, ...) local_unnamed_addr { 53entry: 54 %v = alloca [4 x i32], align 4 55 %ap = alloca %struct.__va_list, align 4 56 %0 = bitcast [4 x i32]* %v to i8* 57 %1 = bitcast %struct.__va_list* %ap to i8* 58 call void @llvm.va_start(i8* nonnull %1) 59 %arraydecay = getelementptr inbounds [4 x i32], [4 x i32]* %v, i32 0, i32 0 60 %call = call i32 @g(i32* nonnull %arraydecay, i32 %a, i32 0, i32 0, i32 0, i32 0) 61 ret i32 %call 62} 63; CHECK-LABEL: test_varargs_sp 64; Three incoming varargs in registers 65; CHECK: sub sp, #12 66; CHECK: sub sp, #28 67; Incoming arguments area is accessed via SP 68; CHECK: add r0, sp, #36 69; CHECK: stm r0!, {r1, r2, r3} 70 71; Re-aligned stack, access via FP 72; int test_args_realign(int a, int b, int c, int d, int e) { 73; __attribute__((aligned(16))) int v[4]; 74; return g(v, a, b, c, d, e); 75; } 76; Function Attrs: nounwind 77define dso_local i32 @test_args_realign(i32 %a, i32 %b, i32 %c, i32 %d, i32 %e) local_unnamed_addr { 78entry: 79 %v = alloca [4 x i32], align 16 80 %0 = bitcast [4 x i32]* %v to i8* 81 %arraydecay = getelementptr inbounds [4 x i32], [4 x i32]* %v, i32 0, i32 0 82 %call = call i32 @g(i32* nonnull %arraydecay, i32 %a, i32 %b, i32 %c, i32 %d, i32 %e) 83 ret i32 %call 84} 85; CHECK-LABEL: test_args_realign 86; Setup frame pointer 87; CHECK: add r7, sp, #8 88; Align stack 89; CHECK: mov r4, sp 90; CHECK-NEXT: lsrs r4, r4, #4 91; CHECK-NEXT: lsls r4, r4, #4 92; CHECK-NEXT: mov sp, r4 93; Load `e` via FP 94; CHECK: ldr r0, [r7, #8] 95; CHECK-NEXT: mov r5, sp 96; CHECK-NEXT: str r3, [r5] 97; Pass `e` as argument 98; CHECK-NEXT: str r0, [r5, #4] 99; CHECK: bl g 100 101; int test_varargs_realign(int a, ...) { 102; __attribute__((aligned(16))) int v[4]; 103; __builtin_va_list ap; 104; __builtin_va_start(ap, a); 105; return g(v, a, 0, 0, 0, 0); 106; } 107define dso_local i32 @test_varargs_realign(i32 %a, ...) local_unnamed_addr { 108entry: 109 %v = alloca [4 x i32], align 16 110 %ap = alloca %struct.__va_list, align 4 111 %0 = bitcast [4 x i32]* %v to i8* 112 %1 = bitcast %struct.__va_list* %ap to i8* 113 call void @llvm.va_start(i8* nonnull %1) 114 %arraydecay = getelementptr inbounds [4 x i32], [4 x i32]* %v, i32 0, i32 0 115 %call = call i32 @g(i32* nonnull %arraydecay, i32 %a, i32 0, i32 0, i32 0, i32 0) 116 ret i32 %call 117} 118; CHECK-LABEL: test_varargs_realign 119; Three incoming register varargs 120; CHECK: sub sp, #12 121; Setup frame pointer 122; CHECK: add r7, sp, #8 123; Align stack 124; CHECK: mov r4, sp 125; CHECK-NEXT: lsrs r4, r4, #4 126; CHECK-NEXT: lsls r4, r4, #4 127; CHECK-NEXT: mov sp, r4 128; Incoming register varargs stored via FP 129; CHECK: str r3, [r7, #16] 130; CHECK-NEXT: str r2, [r7, #12] 131; CHECK-NEXT: str r1, [r7, #8] 132 133; VLAs present, access via FP 134; int test_args_vla(int a, int b, int c, int d, int e) { 135; int v[a]; 136; return g(v, a, b, c, d, e); 137; } 138define dso_local i32 @test_args_vla(i32 %a, i32 %b, i32 %c, i32 %d, i32 %e) local_unnamed_addr { 139entry: 140 %vla = alloca i32, i32 %a, align 4 141 %call = call i32 @g(i32* nonnull %vla, i32 %a, i32 %b, i32 %c, i32 %d, i32 %e) 142 ret i32 %call 143} 144; CHECK-LABEL: test_args_vla 145; Setup frame pointer 146; CHECK: add r7, sp, #12 147; Allocate outgoing stack arguments space 148; CHECK: sub sp, #4 149; Load `e` via FP 150; CHECK: ldr r5, [r7, #8] 151; CHECK-NEXT: mov r0, sp 152; Pass `d` and `e` as arguments 153; CHECK-NEXT: stm r0!, {r3, r5} 154; CHECK: bl g 155 156; int test_varargs_vla(int a, ...) { 157; int v[a]; 158; __builtin_va_list ap; 159; __builtin_va_start(ap, a); 160; return g(v, a, 0, 0, 0, 0); 161; } 162define dso_local i32 @test_varargs_vla(i32 %a, ...) local_unnamed_addr { 163entry: 164 %ap = alloca %struct.__va_list, align 4 165 %vla = alloca i32, i32 %a, align 4 166 %0 = bitcast %struct.__va_list* %ap to i8* 167 call void @llvm.va_start(i8* nonnull %0) 168 %call = call i32 @g(i32* nonnull %vla, i32 %a, i32 0, i32 0, i32 0, i32 0) 169 ret i32 %call 170} 171; CHECK-LABEL: test_varargs_vla 172; Three incoming register varargs 173; CHECK: sub sp, #12 174; Setup frame pointer 175; CHECK: add r7, sp, #8 176; Register varargs stored via FP 177; CHECK: str r3, [r7, #16] 178; CHECK-NEXT: str r2, [r7, #12] 179; CHECK-NEXT: str r1, [r7, #8] 180 181; Moving SP, access via SP 182; int test_args_moving_sp(int a, int b, int c, int d, int e) { 183; int v[4]; 184; return f(v, a, b + c + d, e, s) + h(v, v+1, v+2); 185; } 186define dso_local i32 @test_args_moving_sp(i32 %a, i32 %b, i32 %c, i32 %d, i32 %e) local_unnamed_addr { 187entry: 188 %v = alloca [4 x i32], align 4 189 %0 = bitcast [4 x i32]* %v to i8* 190 %arraydecay = getelementptr inbounds [4 x i32], [4 x i32]* %v, i32 0, i32 0 191 %add = add nsw i32 %c, %b 192 %add1 = add nsw i32 %add, %d 193 %call = call i32 @f(i32* nonnull %arraydecay, i32 %a, i32 %add1, i32 %e, %struct.S* byval nonnull align 4 @s) 194 %add.ptr = getelementptr inbounds [4 x i32], [4 x i32]* %v, i32 0, i32 1 195 %add.ptr5 = getelementptr inbounds [4 x i32], [4 x i32]* %v, i32 0, i32 2 196 %call6 = call i32 @h(i32* nonnull %arraydecay, i32* nonnull %add.ptr, i32* nonnull %add.ptr5) 197 %add7 = add nsw i32 %call6, %call 198 ret i32 %add7 199} 200; CHECK-LABEL: test_args_moving_sp 201; 20 bytes callee-saved area 202; CHECK: push {r4, r5, r6, r7, lr} 203; 20 bytes locals 204; CHECK: sub sp, #20 205; Allocate outgoing arguments space 206; CHECK: sub sp, #508 207; CHECK: sub sp, #4 208; Load `e` via SP, 552 = 512 + 20 + 20 209; CHECK: ldr r3, [sp, #552] 210; CHECK: bl f 211; Stack restored before next call 212; CHECK-NEXT: add sp, #508 213; CHECK-NEXT: add sp, #4 214; CHECK: bl h 215 216; int test_varargs_moving_sp(int a, ...) { 217; int v[4]; 218; __builtin_va_list ap; 219; __builtin_va_start(ap, a); 220; return f(v, a, 0, 0, s) + h(v, v+1, v+2); 221; } 222define dso_local i32 @test_varargs_moving_sp(i32 %a, ...) local_unnamed_addr { 223entry: 224 %v = alloca [4 x i32], align 4 225 %ap = alloca %struct.__va_list, align 4 226 %0 = bitcast [4 x i32]* %v to i8* 227 %1 = bitcast %struct.__va_list* %ap to i8* 228 call void @llvm.va_start(i8* nonnull %1) 229 %arraydecay = getelementptr inbounds [4 x i32], [4 x i32]* %v, i32 0, i32 0 230 %call = call i32 @f(i32* nonnull %arraydecay, i32 %a, i32 0, i32 0, %struct.S* byval nonnull align 4 @s) 231 %add.ptr = getelementptr inbounds [4 x i32], [4 x i32]* %v, i32 0, i32 1 232 %add.ptr5 = getelementptr inbounds [4 x i32], [4 x i32]* %v, i32 0, i32 2 233 %call6 = call i32 @h(i32* nonnull %arraydecay, i32* nonnull %add.ptr, i32* nonnull %add.ptr5) 234 %add = add nsw i32 %call6, %call 235 ret i32 %add 236} 237; CHECK-LABEL: test_varargs_moving_sp 238; Three incoming register varargs 239; CHECK: sub sp, #12 240; 16 bytes callee-saves 241; CHECK: push {r4, r5, r7, lr} 242; 20 bytes locals 243; CHECK: sub sp, #20 244; Incoming varargs stored via SP, 36 = 20 + 16 245; CHECK: add r0, sp, #36 246; CHECK-NEXT: stm r0!, {r1, r2, r3} 247 248; 249; Access to locals 250; 251 252; Usual case, access via SP. 253; int test_local(int n) { 254; int v[4]; 255; int x, y, z; 256; h(&x, &y, &z); 257; return g(v, x, y, z, 0, 0); 258; } 259define dso_local i32 @test_local(i32 %n) local_unnamed_addr { 260entry: 261 %v = alloca [4 x i32], align 4 262 %x = alloca i32, align 4 263 %y = alloca i32, align 4 264 %z = alloca i32, align 4 265 %0 = bitcast [4 x i32]* %v to i8* 266 %1 = bitcast i32* %x to i8* 267 %2 = bitcast i32* %y to i8* 268 %3 = bitcast i32* %z to i8* 269 %call = call i32 @h(i32* nonnull %x, i32* nonnull %y, i32* nonnull %z) 270 %arraydecay = getelementptr inbounds [4 x i32], [4 x i32]* %v, i32 0, i32 0 271 %4 = load i32, i32* %x, align 4 272 %5 = load i32, i32* %y, align 4 273 %6 = load i32, i32* %z, align 4 274 %call1 = call i32 @g(i32* nonnull %arraydecay, i32 %4, i32 %5, i32 %6, i32 0, i32 0) 275 ret i32 %call1 276} 277; CHECK-LABEL: test_local 278; Arguments to `h` relative to SP 279; CHECK: add r0, sp, #20 280; CHECK-NEXT: add r1, sp, #16 281; CHECK-NEXT: add r2, sp, #12 282; CHECK-NEXT: bl h 283; Load `x`, `y`, and `z` via SP 284; CHECK: ldr r1, [sp, #20] 285; CHECK-NEXT: ldr r2, [sp, #16] 286; CHECK-NEXT: ldr r3, [sp, #12] 287; CHECK: bl g 288 289; Re-aligned stack, access via SP. 290; int test_local_realign(int n) { 291; __attribute__((aligned(16))) int v[4]; 292; int x, y, z; 293; h(&x, &y, &z); 294; return g(v, x, y, z, 0, 0); 295; } 296define dso_local i32 @test_local_realign(i32 %n) local_unnamed_addr { 297entry: 298 %v = alloca [4 x i32], align 16 299 %x = alloca i32, align 4 300 %y = alloca i32, align 4 301 %z = alloca i32, align 4 302 %0 = bitcast [4 x i32]* %v to i8* 303 %1 = bitcast i32* %x to i8* 304 %2 = bitcast i32* %y to i8* 305 %3 = bitcast i32* %z to i8* 306 %call = call i32 @h(i32* nonnull %x, i32* nonnull %y, i32* nonnull %z) 307 %arraydecay = getelementptr inbounds [4 x i32], [4 x i32]* %v, i32 0, i32 0 308 %4 = load i32, i32* %x, align 4 309 %5 = load i32, i32* %y, align 4 310 %6 = load i32, i32* %z, align 4 311 %call1 = call i32 @g(i32* nonnull %arraydecay, i32 %4, i32 %5, i32 %6, i32 0, i32 0) 312 ret i32 %call1 313} 314; CHECK-LABEL: test_local_realign 315; Setup frame pointer 316; CHECK: add r7, sp, #8 317; Re-align stack 318; CHECK: mov r4, sp 319; CHECK-NEXT: lsrs r4, r4, #4 320; CHECK-NEXT: lsls r4, r4, #4 321; CHECK-NEXT: mov sp, r4 322; Arguments to `h` computed relative to SP 323; CHECK: add r0, sp, #28 324; CHECK-NEXT: add r1, sp, #24 325; CHECK-NEXT: add r2, sp, #20 326; CHECK-NEXT: bl h 327; Load `x`, `y`, and `z` via SP for passing to `g` 328; CHECK: ldr r1, [sp, #28] 329; CHECK-NEXT: ldr r2, [sp, #24] 330; CHECK-NEXT: ldr r3, [sp, #20] 331; CHECK: bl g 332 333; VLAs, access via BP. 334; int test_local_vla(int n) { 335; int v[n]; 336; int x, y, z; 337; h(&x, &y, &z); 338; return g(v, x, y, z, 0, 0); 339; } 340define dso_local i32 @test_local_vla(i32 %n) local_unnamed_addr { 341entry: 342 %x = alloca i32, align 4 343 %y = alloca i32, align 4 344 %z = alloca i32, align 4 345 %vla = alloca i32, i32 %n, align 4 346 %0 = bitcast i32* %x to i8* 347 %1 = bitcast i32* %y to i8* 348 %2 = bitcast i32* %z to i8* 349 %call = call i32 @h(i32* nonnull %x, i32* nonnull %y, i32* nonnull %z) 350 %3 = load i32, i32* %x, align 4 351 %4 = load i32, i32* %y, align 4 352 %5 = load i32, i32* %z, align 4 353 %call1 = call i32 @g(i32* nonnull %vla, i32 %3, i32 %4, i32 %5, i32 0, i32 0) 354 ret i32 %call1 355} 356; CHECK-LABEL: test_local_vla 357; Setup frame pointer 358; CHECK: add r7, sp, #12 359; Setup base pointer 360; CHECK: mov r6, sp 361; CHECK: mov r5, r6 362; Arguments to `h` compute relative to BP 363; CHECK: adds r0, r6, #7 364; CHECK-NEXT: adds r0, #1 365; CHECK-NEXT: adds r1, r6, #4 366; CHECK-NEXT: mov r2, r6 367; CHECK-NEXT: bl h 368; Load `x`, `y`, `z` via BP (r5 should still have the value of r6 from the move 369; above) 370; CHECK: ldr r3, [r5] 371; CHECK-NEXT: ldr r2, [r5, #4] 372; CHECK-NEXT: ldr r1, [r5, #8] 373; CHECK: bl g 374 375; Moving SP, access via SP. 376; int test_local_moving_sp(int n) { 377; int v[4]; 378; int x, y, z; 379; return u(v, &x, &y, s, s) + u(v, &y, &z, s, s); 380; } 381define dso_local i32 @test_local_moving_sp(i32 %n) local_unnamed_addr { 382entry: 383 %v = alloca [4 x i32], align 4 384 %x = alloca i32, align 4 385 %y = alloca i32, align 4 386 %z = alloca i32, align 4 387 %0 = bitcast [4 x i32]* %v to i8* 388 %1 = bitcast i32* %x to i8* 389 %2 = bitcast i32* %y to i8* 390 %3 = bitcast i32* %z to i8* 391 %arraydecay = getelementptr inbounds [4 x i32], [4 x i32]* %v, i32 0, i32 0 392 %call = call i32 @u(i32* nonnull %arraydecay, i32* nonnull %x, i32* nonnull %y, %struct.S* byval nonnull align 4 @s, %struct.S* byval nonnull align 4 @s) 393 %call2 = call i32 @u(i32* nonnull %arraydecay, i32* nonnull %y, i32* nonnull %z, %struct.S* byval nonnull align 4 @s, %struct.S* byval nonnull align 4 @s) 394 %add = add nsw i32 %call2, %call 395 ret i32 %add 396} 397; CHECK-LABEL: test_local_moving_sp 398; Locals area 399; CHECK: sub sp, #36 400; Outoging arguments 401; CHECK: sub sp, #508 402; CHECK-NEXT: sub sp, #508 403; CHECK-NEXT: sub sp, #8 404; Argument addresses computed relative to SP 405; CHECK: add r4, sp, #1020 406; CHECK-NEXT: adds r4, #24 407; CHECK: add r1, sp, #1020 408; CHECK-NEXT: adds r1, #20 409; CHECK: add r5, sp, #1020 410; CHECK-NEXT: adds r5, #16 411; CHECK: bl u 412; Stack restored before next call 413; CHECK: add sp, #508 414; CHECK-NEXT: add sp, #508 415; CHECK-NEXT: add sp, #8 416; CHECK: bl u 417