1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py 2 3; Verify the call site info. If the call site info is not 4; in the valid state, an assert should be triggered. 5; RUN: llc < %s -debug-entry-values -stop-after=machineverifier 6 7; REQUIRES: asserts 8; RUN: llc -mcpu=haswell < %s -O2 2>&1 | FileCheck %s 9; For test: 10; 2 invariant loads, 1 for OBJC_SELECTOR_REFERENCES_ 11; and 1 for objc_msgSend from the GOT 12; For test_multi_def: 13; 2 invariant load (full multiply, both loads should be hoisted.) 14; For test_div_def: 15; 2 invariant load (full divide, both loads should be hoisted.) 1 additional instruction for a zeroing edx that gets hoisted and then rematerialized. 16 17target datalayout = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v64:64:64-v128:128:128-a0:0:64-s0:64:64-f80:128:128-n8:16:32:64-S128" 18target triple = "x86_64-apple-macosx10.7.2" 19 20@"\01L_OBJC_METH_VAR_NAME_" = internal global [4 x i8] c"foo\00", section "__TEXT,__objc_methname,cstring_literals", align 1 21@"\01L_OBJC_SELECTOR_REFERENCES_" = internal global i8* getelementptr inbounds ([4 x i8], [4 x i8]* @"\01L_OBJC_METH_VAR_NAME_", i64 0, i64 0), section "__DATA, __objc_selrefs, literal_pointers, no_dead_strip" 22@"\01L_OBJC_IMAGE_INFO" = internal constant [2 x i32] [i32 0, i32 16], section "__DATA, __objc_imageinfo, regular, no_dead_strip" 23@llvm.used = appending global [3 x i8*] [i8* getelementptr inbounds ([4 x i8], [4 x i8]* @"\01L_OBJC_METH_VAR_NAME_", i32 0, i32 0), i8* bitcast (i8** @"\01L_OBJC_SELECTOR_REFERENCES_" to i8*), i8* bitcast ([2 x i32]* @"\01L_OBJC_IMAGE_INFO" to i8*)], section "llvm.metadata" 24 25define void @test(i8* %x) uwtable ssp { 26; CHECK-LABEL: test: 27; CHECK: ## %bb.0: ## %entry 28; CHECK-NEXT: pushq %rbp 29; CHECK-NEXT: .cfi_def_cfa_offset 16 30; CHECK-NEXT: pushq %r15 31; CHECK-NEXT: .cfi_def_cfa_offset 24 32; CHECK-NEXT: pushq %r14 33; CHECK-NEXT: .cfi_def_cfa_offset 32 34; CHECK-NEXT: pushq %rbx 35; CHECK-NEXT: .cfi_def_cfa_offset 40 36; CHECK-NEXT: pushq %rax 37; CHECK-NEXT: .cfi_def_cfa_offset 48 38; CHECK-NEXT: .cfi_offset %rbx, -40 39; CHECK-NEXT: .cfi_offset %r14, -32 40; CHECK-NEXT: .cfi_offset %r15, -24 41; CHECK-NEXT: .cfi_offset %rbp, -16 42; CHECK-NEXT: movq %rdi, %rbx 43; CHECK-NEXT: movl $10000, %ebp ## imm = 0x2710 44; CHECK-NEXT: movq {{.*}}(%rip), %r14 45; CHECK-NEXT: movq _objc_msgSend@{{.*}}(%rip), %r15 46; CHECK-NEXT: .p2align 4, 0x90 47; CHECK-NEXT: LBB0_1: ## %for.body 48; CHECK-NEXT: ## =>This Inner Loop Header: Depth=1 49; CHECK-NEXT: movq %rbx, %rdi 50; CHECK-NEXT: movq %r14, %rsi 51; CHECK-NEXT: callq *%r15 52; CHECK-NEXT: decl %ebp 53; CHECK-NEXT: jne LBB0_1 54; CHECK-NEXT: ## %bb.2: ## %for.end 55; CHECK-NEXT: addq $8, %rsp 56; CHECK-NEXT: popq %rbx 57; CHECK-NEXT: popq %r14 58; CHECK-NEXT: popq %r15 59; CHECK-NEXT: popq %rbp 60; CHECK-NEXT: retq 61entry: 62 br label %for.body 63 64for.body: ; preds = %for.body, %entry 65 %i.01 = phi i32 [ 0, %entry ], [ %inc, %for.body ] 66 %0 = load i8*, i8** @"\01L_OBJC_SELECTOR_REFERENCES_", align 8, !invariant.load !0 67 %call = tail call i8* bitcast (i8* (i8*, i8*, ...)* @objc_msgSend to i8* (i8*, i8*)*)(i8* %x, i8* %0) 68 %inc = add i32 %i.01, 1 69 %exitcond = icmp eq i32 %inc, 10000 70 br i1 %exitcond, label %for.end, label %for.body 71 72for.end: ; preds = %for.body 73 ret void 74} 75 76define void @test_unordered(i8* %x) uwtable ssp { 77; CHECK-LABEL: test_unordered: 78; CHECK: ## %bb.0: ## %entry 79; CHECK-NEXT: pushq %rbp 80; CHECK-NEXT: .cfi_def_cfa_offset 16 81; CHECK-NEXT: pushq %r15 82; CHECK-NEXT: .cfi_def_cfa_offset 24 83; CHECK-NEXT: pushq %r14 84; CHECK-NEXT: .cfi_def_cfa_offset 32 85; CHECK-NEXT: pushq %rbx 86; CHECK-NEXT: .cfi_def_cfa_offset 40 87; CHECK-NEXT: pushq %rax 88; CHECK-NEXT: .cfi_def_cfa_offset 48 89; CHECK-NEXT: .cfi_offset %rbx, -40 90; CHECK-NEXT: .cfi_offset %r14, -32 91; CHECK-NEXT: .cfi_offset %r15, -24 92; CHECK-NEXT: .cfi_offset %rbp, -16 93; CHECK-NEXT: movq %rdi, %rbx 94; CHECK-NEXT: movl $10000, %ebp ## imm = 0x2710 95; CHECK-NEXT: movq {{.*}}(%rip), %r14 96; CHECK-NEXT: movq _objc_msgSend@{{.*}}(%rip), %r15 97; CHECK-NEXT: .p2align 4, 0x90 98; CHECK-NEXT: LBB1_1: ## %for.body 99; CHECK-NEXT: ## =>This Inner Loop Header: Depth=1 100; CHECK-NEXT: movq %rbx, %rdi 101; CHECK-NEXT: movq %r14, %rsi 102; CHECK-NEXT: callq *%r15 103; CHECK-NEXT: decl %ebp 104; CHECK-NEXT: jne LBB1_1 105; CHECK-NEXT: ## %bb.2: ## %for.end 106; CHECK-NEXT: addq $8, %rsp 107; CHECK-NEXT: popq %rbx 108; CHECK-NEXT: popq %r14 109; CHECK-NEXT: popq %r15 110; CHECK-NEXT: popq %rbp 111; CHECK-NEXT: retq 112entry: 113 br label %for.body 114 115for.body: ; preds = %for.body, %entry 116 %i.01 = phi i32 [ 0, %entry ], [ %inc, %for.body ] 117 %0 = load atomic i8*, i8** @"\01L_OBJC_SELECTOR_REFERENCES_" unordered, align 8, !invariant.load !0 118 %call = tail call i8* bitcast (i8* (i8*, i8*, ...)* @objc_msgSend to i8* (i8*, i8*)*)(i8* %x, i8* %0) 119 %inc = add i32 %i.01, 1 120 %exitcond = icmp eq i32 %inc, 10000 121 br i1 %exitcond, label %for.end, label %for.body 122 123for.end: ; preds = %for.body 124 ret void 125} 126 127define void @test_volatile(i8* %x) uwtable ssp { 128; CHECK-LABEL: test_volatile: 129; CHECK: ## %bb.0: ## %entry 130; CHECK-NEXT: pushq %rbp 131; CHECK-NEXT: .cfi_def_cfa_offset 16 132; CHECK-NEXT: pushq %r14 133; CHECK-NEXT: .cfi_def_cfa_offset 24 134; CHECK-NEXT: pushq %rbx 135; CHECK-NEXT: .cfi_def_cfa_offset 32 136; CHECK-NEXT: .cfi_offset %rbx, -32 137; CHECK-NEXT: .cfi_offset %r14, -24 138; CHECK-NEXT: .cfi_offset %rbp, -16 139; CHECK-NEXT: movq %rdi, %rbx 140; CHECK-NEXT: movl $10000, %ebp ## imm = 0x2710 141; CHECK-NEXT: movq _objc_msgSend@{{.*}}(%rip), %r14 142; CHECK-NEXT: .p2align 4, 0x90 143; CHECK-NEXT: LBB2_1: ## %for.body 144; CHECK-NEXT: ## =>This Inner Loop Header: Depth=1 145; CHECK-NEXT: movq {{.*}}(%rip), %rsi 146; CHECK-NEXT: movq %rbx, %rdi 147; CHECK-NEXT: callq *%r14 148; CHECK-NEXT: decl %ebp 149; CHECK-NEXT: jne LBB2_1 150; CHECK-NEXT: ## %bb.2: ## %for.end 151; CHECK-NEXT: popq %rbx 152; CHECK-NEXT: popq %r14 153; CHECK-NEXT: popq %rbp 154; CHECK-NEXT: retq 155entry: 156 br label %for.body 157 158for.body: ; preds = %for.body, %entry 159 %i.01 = phi i32 [ 0, %entry ], [ %inc, %for.body ] 160 %0 = load volatile i8*, i8** @"\01L_OBJC_SELECTOR_REFERENCES_", align 8, !invariant.load !0 161 %call = tail call i8* bitcast (i8* (i8*, i8*, ...)* @objc_msgSend to i8* (i8*, i8*)*)(i8* %x, i8* %0) 162 %inc = add i32 %i.01, 1 163 %exitcond = icmp eq i32 %inc, 10000 164 br i1 %exitcond, label %for.end, label %for.body 165 166for.end: ; preds = %for.body 167 ret void 168} 169 170define void @test_seq_cst(i8* %x) uwtable ssp { 171; CHECK-LABEL: test_seq_cst: 172; CHECK: ## %bb.0: ## %entry 173; CHECK-NEXT: pushq %rbp 174; CHECK-NEXT: .cfi_def_cfa_offset 16 175; CHECK-NEXT: pushq %r14 176; CHECK-NEXT: .cfi_def_cfa_offset 24 177; CHECK-NEXT: pushq %rbx 178; CHECK-NEXT: .cfi_def_cfa_offset 32 179; CHECK-NEXT: .cfi_offset %rbx, -32 180; CHECK-NEXT: .cfi_offset %r14, -24 181; CHECK-NEXT: .cfi_offset %rbp, -16 182; CHECK-NEXT: movq %rdi, %rbx 183; CHECK-NEXT: movl $10000, %ebp ## imm = 0x2710 184; CHECK-NEXT: movq _objc_msgSend@{{.*}}(%rip), %r14 185; CHECK-NEXT: .p2align 4, 0x90 186; CHECK-NEXT: LBB3_1: ## %for.body 187; CHECK-NEXT: ## =>This Inner Loop Header: Depth=1 188; CHECK-NEXT: movq {{.*}}(%rip), %rsi 189; CHECK-NEXT: movq %rbx, %rdi 190; CHECK-NEXT: callq *%r14 191; CHECK-NEXT: decl %ebp 192; CHECK-NEXT: jne LBB3_1 193; CHECK-NEXT: ## %bb.2: ## %for.end 194; CHECK-NEXT: popq %rbx 195; CHECK-NEXT: popq %r14 196; CHECK-NEXT: popq %rbp 197; CHECK-NEXT: retq 198entry: 199 br label %for.body 200 201for.body: ; preds = %for.body, %entry 202 %i.01 = phi i32 [ 0, %entry ], [ %inc, %for.body ] 203 %0 = load atomic i8*, i8** @"\01L_OBJC_SELECTOR_REFERENCES_" seq_cst, align 8, !invariant.load !0 204 %call = tail call i8* bitcast (i8* (i8*, i8*, ...)* @objc_msgSend to i8* (i8*, i8*)*)(i8* %x, i8* %0) 205 %inc = add i32 %i.01, 1 206 %exitcond = icmp eq i32 %inc, 10000 207 br i1 %exitcond, label %for.end, label %for.body 208 209for.end: ; preds = %for.body 210 ret void 211} 212 213declare i8* @objc_msgSend(i8*, i8*, ...) nonlazybind 214 215define void @test_multi_def(i64* dereferenceable(8) %x1, 216; CHECK-LABEL: test_multi_def: 217; CHECK: ## %bb.0: ## %entry 218; CHECK-NEXT: movq %rdx, %rax 219; CHECK-NEXT: xorl %r8d, %r8d 220; CHECK-NEXT: movq (%rdi), %rdx 221; CHECK-NEXT: movq (%rsi), %r9 222; CHECK-NEXT: .p2align 4, 0x90 223; CHECK-NEXT: LBB4_2: ## %for.body 224; CHECK-NEXT: ## =>This Inner Loop Header: Depth=1 225; CHECK-NEXT: mulxq %r9, %rsi, %rdi 226; CHECK-NEXT: addq %rsi, (%rax) 227; CHECK-NEXT: adcq %rdi, 8(%rax) 228; CHECK-NEXT: ## %bb.1: ## %for.check 229; CHECK-NEXT: ## in Loop: Header=BB4_2 Depth=1 230; CHECK-NEXT: incq %r8 231; CHECK-NEXT: addq $16, %rax 232; CHECK-NEXT: cmpq %rcx, %r8 233; CHECK-NEXT: jl LBB4_2 234; CHECK-NEXT: ## %bb.3: ## %exit 235; CHECK-NEXT: retq 236 i64* dereferenceable(8) %x2, 237 i128* %y, i64 %count) nounwind { 238entry: 239 br label %for.body 240 241for.check: 242 %inc = add nsw i64 %i, 1 243 %done = icmp sge i64 %inc, %count 244 br i1 %done, label %exit, label %for.body 245 246for.body: 247 %i = phi i64 [ 0, %entry ], [ %inc, %for.check ] 248 %x1_load = load i64, i64* %x1, align 8, !invariant.load !0 249 %x1_zext = zext i64 %x1_load to i128 250 %x2_load = load i64, i64* %x2, align 8, !invariant.load !0 251 %x2_zext = zext i64 %x2_load to i128 252 %x_prod = mul i128 %x1_zext, %x2_zext 253 %y_elem = getelementptr inbounds i128, i128* %y, i64 %i 254 %y_load = load i128, i128* %y_elem, align 8 255 %y_plus = add i128 %x_prod, %y_load 256 store i128 %y_plus, i128* %y_elem, align 8 257 br label %for.check 258 259exit: 260 ret void 261} 262 263define void @test_div_def(i32* dereferenceable(8) %x1, 264; CHECK-LABEL: test_div_def: 265; CHECK: ## %bb.0: ## %entry 266; CHECK-NEXT: movq %rdx, %r8 267; CHECK-NEXT: xorl %r9d, %r9d 268; CHECK-NEXT: movl (%rdi), %edi 269; CHECK-NEXT: movl (%rsi), %esi 270; CHECK-NEXT: .p2align 4, 0x90 271; CHECK-NEXT: LBB5_2: ## %for.body 272; CHECK-NEXT: ## =>This Inner Loop Header: Depth=1 273; CHECK-NEXT: movl %edi, %eax 274; CHECK-NEXT: xorl %edx, %edx 275; CHECK-NEXT: divl %esi 276; CHECK-NEXT: addl %eax, (%r8,%r9,4) 277; CHECK-NEXT: ## %bb.1: ## %for.check 278; CHECK-NEXT: ## in Loop: Header=BB5_2 Depth=1 279; CHECK-NEXT: incq %r9 280; CHECK-NEXT: cmpl %ecx, %r9d 281; CHECK-NEXT: jl LBB5_2 282; CHECK-NEXT: ## %bb.3: ## %exit 283; CHECK-NEXT: retq 284 i32* dereferenceable(8) %x2, 285 i32* %y, i32 %count) nounwind { 286entry: 287 br label %for.body 288 289for.check: 290 %inc = add nsw i32 %i, 1 291 %done = icmp sge i32 %inc, %count 292 br i1 %done, label %exit, label %for.body 293 294for.body: 295 %i = phi i32 [ 0, %entry ], [ %inc, %for.check ] 296 %x1_load = load i32, i32* %x1, align 8, !invariant.load !0 297 %x2_load = load i32, i32* %x2, align 8, !invariant.load !0 298 %x_quot = udiv i32 %x1_load, %x2_load 299 %y_elem = getelementptr inbounds i32, i32* %y, i32 %i 300 %y_load = load i32, i32* %y_elem, align 8 301 %y_plus = add i32 %x_quot, %y_load 302 store i32 %y_plus, i32* %y_elem, align 8 303 br label %for.check 304 305exit: 306 ret void 307} 308 309!0 = !{} 310