1; RUN: llc -verify-machineinstrs < %s -mtriple=aarch64-none-linux-gnu | FileCheck %s --check-prefixes=SDAG,COMMON 2; RUN: llc -global-isel -global-isel-abort=1 -verify-machineinstrs < %s -mtriple=aarch64-none-linux-gnu | FileCheck %s --check-prefixes=GISEL,COMMON 3 4declare tailcc void @callee_stack0() 5declare tailcc void @callee_stack8([8 x i64], i64) 6declare tailcc void @callee_stack16([8 x i64], i64, i64) 7declare extern_weak tailcc void @callee_weak() 8 9define tailcc void @caller_to0_from0() nounwind { 10; COMMON-LABEL: caller_to0_from0: 11; COMMON-NEXT: // %bb. 12 13 tail call tailcc void @callee_stack0() 14 ret void 15 16; COMMON-NEXT: b callee_stack0 17} 18 19define tailcc void @caller_to0_from8([8 x i64], i64) #0 { 20; COMMON-LABEL: caller_to0_from8: 21 22 tail call tailcc void @callee_stack0() 23 ret void 24 25; COMMON: add sp, sp, #16 26; COMMON: .cfi_def_cfa_offset -16 27; COMMON-NEXT: b callee_stack0 28} 29 30define tailcc void @caller_to8_from0() "frame-pointer"="all" uwtable { 31; COMMON-LABEL: caller_to8_from0: 32 33; Key point is that the "42" should go #16 below incoming stack 34; pointer (we didn't have arg space to reuse). 35 tail call tailcc void @callee_stack8([8 x i64] undef, i64 42) 36 ret void 37 38; COMMON: str {{x[0-9]+}}, [x29, #16] 39; COMMON: ldp x29, x30, [sp], #16 40 ; If there is a sub here then the 42 will be briefly exposed to corruption 41 ; from an interrupt if the kernel does not honour a red-zone, and a larger 42 ; call could well overflow the red zone even if it is present. 43; COMMON-NOT: sub sp, 44; COMMON-NEXT: .cfi_def_cfa_offset 16 45; COMMON-NEXT: .cfi_restore w30 46; COMMON-NEXT: .cfi_restore w29 47; COMMON-NEXT: b callee_stack8 48} 49 50define tailcc void @caller_to8_from8([8 x i64], i64 %a) #0 { 51; COMMON-LABEL: caller_to8_from8: 52; COMMON-NOT: sub sp, 53 54; Key point is that the "%a" should go where at SP on entry. 55 tail call tailcc void @callee_stack8([8 x i64] undef, i64 42) 56 ret void 57 58; COMMON: str {{x[0-9]+}}, [sp] 59; COMMON-NEXT: b callee_stack8 60} 61 62define tailcc void @caller_to16_from8([8 x i64], i64 %a) #0 { 63; COMMON-LABEL: caller_to16_from8: 64; COMMON-NOT: sub sp, 65 66; Important point is that the call reuses the "dead" argument space 67; above %a on the stack. If it tries to go below incoming-SP then the 68; callee will not deallocate the space, even in tailcc. 69 tail call tailcc void @callee_stack16([8 x i64] undef, i64 42, i64 2) 70 71; COMMON: stp {{x[0-9]+}}, {{x[0-9]+}}, [sp] 72; COMMON-NEXT: b callee_stack16 73 ret void 74} 75 76 77define tailcc void @caller_to8_from24([8 x i64], i64 %a, i64 %b, i64 %c) #0 { 78; COMMON-LABEL: caller_to8_from24: 79; COMMON-NOT: sub sp, 80 81; Key point is that the "%a" should go where at #16 above SP on entry. 82 tail call tailcc void @callee_stack8([8 x i64] undef, i64 42) 83 ret void 84 85; COMMON: str {{x[0-9]+}}, [sp, #16]! 86; COMMON-NEXT: .cfi_def_cfa_offset -16 87; COMMON-NEXT: b callee_stack8 88} 89 90 91define tailcc void @caller_to16_from16([8 x i64], i64 %a, i64 %b) #0 { 92; COMMON-LABEL: caller_to16_from16: 93; COMMON-NOT: sub sp, 94 95; Here we want to make sure that both loads happen before the stores: 96; otherwise either %a or %b will be wrongly clobbered. 97 tail call tailcc void @callee_stack16([8 x i64] undef, i64 %b, i64 %a) 98 ret void 99 100; COMMON: ldp {{x[0-9]+}}, {{x[0-9]+}}, [sp] 101; COMMON: stp {{x[0-9]+}}, {{x[0-9]+}}, [sp] 102; COMMON-NEXT: b callee_stack16 103} 104 105define tailcc void @disable_tail_calls() nounwind "disable-tail-calls"="true" { 106; COMMON-LABEL: disable_tail_calls: 107; COMMON-NEXT: // %bb. 108 109 tail call tailcc void @callee_stack0() 110 ret void 111 112; COMMON: bl callee_stack0 113; COMMON: ret 114} 115 116; Weakly-referenced extern functions cannot be tail-called, as AAELF does 117; not define the behaviour of branch instructions to undefined weak symbols. 118define tailcc void @caller_weak() #0 { 119; COMMON-LABEL: caller_weak: 120; COMMON: bl callee_weak 121 tail call void @callee_weak() 122 ret void 123} 124 125declare { [2 x float] } @get_vec2() 126 127define { [3 x float] } @test_add_elem() #0 { 128; SDAG-LABEL: test_add_elem: 129; SDAG: bl get_vec2 130; SDAG: fmov s2, #1.0 131; SDAG: ret 132; GISEL-LABEL: test_add_elem: 133; GISEL: str x30, [sp, #-16]! 134; GISEL: bl get_vec2 135; GISEL: fmov s2, #1.0 136; GISEL: ldr x30, [sp], #16 137; GISEL: ret 138 139 %call = tail call { [2 x float] } @get_vec2() 140 %arr = extractvalue { [2 x float] } %call, 0 141 %arr.0 = extractvalue [2 x float] %arr, 0 142 %arr.1 = extractvalue [2 x float] %arr, 1 143 144 %res.0 = insertvalue { [3 x float] } undef, float %arr.0, 0, 0 145 %res.01 = insertvalue { [3 x float] } %res.0, float %arr.1, 0, 1 146 %res.012 = insertvalue { [3 x float] } %res.01, float 1.000000e+00, 0, 2 147 ret { [3 x float] } %res.012 148} 149 150declare double @get_double() 151define { double, [2 x double] } @test_mismatched_insert() #0 { 152; COMMON-LABEL: test_mismatched_insert: 153; COMMON: bl get_double 154; COMMON: bl get_double 155; COMMON: bl get_double 156; COMMON: ret 157 158 %val0 = call double @get_double() 159 %val1 = call double @get_double() 160 %val2 = tail call double @get_double() 161 162 %res.0 = insertvalue { double, [2 x double] } undef, double %val0, 0 163 %res.01 = insertvalue { double, [2 x double] } %res.0, double %val1, 1, 0 164 %res.012 = insertvalue { double, [2 x double] } %res.01, double %val2, 1, 1 165 166 ret { double, [2 x double] } %res.012 167} 168 169define void @fromC_totail() #0 { 170; COMMON-LABEL: fromC_totail: 171; COMMON: sub sp, sp, #32 172 173; COMMON-NOT: sub sp, 174; COMMON: mov w[[TMP:[0-9]+]], #42 175; COMMON: str x[[TMP]], [sp] 176; COMMON: bl callee_stack8 177 ; We must reset the stack to where it was before the call by undoing its extra stack pop. 178; COMMON: str x[[TMP]], [sp, #-16]! 179; COMMON: bl callee_stack8 180; COMMON: sub sp, sp, #16 181 182 call tailcc void @callee_stack8([8 x i64] undef, i64 42) 183 call tailcc void @callee_stack8([8 x i64] undef, i64 42) 184 ret void 185} 186 187define void @fromC_totail_noreservedframe(i32 %len) #0 { 188; COMMON-LABEL: fromC_totail_noreservedframe: 189; COMMON: stp x29, x30, [sp, #-32]! 190 191; COMMON: mov w[[TMP:[0-9]+]], #42 192 ; Note stack is subtracted here to allocate space for arg 193; COMMON: str x[[TMP]], [sp, #-16]! 194; COMMON: bl callee_stack8 195 ; And here. 196; COMMON: str x[[TMP]], [sp, #-16]! 197; COMMON: bl callee_stack8 198 ; But not restored here because callee_stack8 did that for us. 199; COMMON-NOT: sub sp, 200 201 ; Variable sized allocation prevents reserving frame at start of function so each call must allocate any stack space it needs. 202 %var = alloca i32, i32 %len 203 204 call tailcc void @callee_stack8([8 x i64] undef, i64 42) 205 call tailcc void @callee_stack8([8 x i64] undef, i64 42) 206 ret void 207} 208 209declare void @Ccallee_stack8([8 x i64], i64) 210 211define tailcc void @fromtail_toC() #0 { 212; COMMON-LABEL: fromtail_toC: 213; COMMON: sub sp, sp, #32 214 215; COMMON-NOT: sub sp, 216; COMMON: mov w[[TMP:[0-9]+]], #42 217; COMMON: str x[[TMP]], [sp] 218; COMMON: bl Ccallee_stack8 219 ; C callees will return with the stack exactly where we left it, so we mustn't try to fix anything. 220; COMMON-NOT: add sp, 221; COMMON-NOT: sub sp, 222; COMMON: str x[[TMP]], [sp]{{$}} 223; COMMON: bl Ccallee_stack8 224; COMMON-NOT: sub sp, 225 226 227 call void @Ccallee_stack8([8 x i64] undef, i64 42) 228 call void @Ccallee_stack8([8 x i64] undef, i64 42) 229 ret void 230} 231 232attributes #0 = { uwtable }