1; RUN: llc < %s -asm-verbose=false -disable-wasm-fallthrough-return-opt | FileCheck %s 2 3target datalayout = "e-m:e-p:32:32-i64:64-n32:64-S128" 4target triple = "wasm32-unknown-unknown" 5 6declare void @ext_func(i64* %ptr) 7declare void @ext_func_i32(i32* %ptr) 8 9; CHECK-LABEL: alloca32: 10; Check that there is an extra local for the stack pointer. 11; CHECK: .local i32{{$}} 12define void @alloca32() noredzone { 13 ; CHECK: i32.const $push[[L4:.+]]=, 0{{$}} 14 ; CHECK: i32.const $push[[L1:.+]]=, 0{{$}} 15 ; CHECK-NEXT: i32.load $push[[L2:.+]]=, __stack_pointer($pop[[L1]]) 16 ; CHECK-NEXT: i32.const $push[[L3:.+]]=, 16 17 ; CHECK-NEXT: i32.sub $push[[L8:.+]]=, $pop[[L2]], $pop[[L3]] 18 ; CHECK-NEXT: i32.store $push[[L10:.+]]=, __stack_pointer($pop[[L4]]), $pop[[L8]]{{$}} 19 ; CHECK-NEXT: tee_local $push[[L9:.+]]=, $[[SP:.+]]=, $pop[[L10]]{{$}} 20 %retval = alloca i32 21 ; CHECK: i32.const $push[[L0:.+]]=, 0 22 ; CHECK: i32.store {{.*}}=, 12($pop[[L9]]), $pop[[L0]] 23 store i32 0, i32* %retval 24 ; CHECK: i32.const $push[[L6:.+]]=, 0 25 ; CHECK-NEXT: i32.const $push[[L5:.+]]=, 16 26 ; CHECK-NEXT: i32.add $push[[L7:.+]]=, $[[SP]], $pop[[L5]] 27 ; CHECK-NEXT: i32.store $drop=, __stack_pointer($pop[[L6]]), $pop[[L7]] 28 ret void 29} 30 31; CHECK-LABEL: alloca3264: 32; CHECK: .local i32{{$}} 33define void @alloca3264() { 34 ; CHECK: i32.const $push[[L2:.+]]=, 0{{$}} 35 ; CHECK-NEXT: i32.load $push[[L3:.+]]=, __stack_pointer($pop[[L2]]) 36 ; CHECK-NEXT: i32.const $push[[L4:.+]]=, 16 37 ; CHECK-NEXT: i32.sub $push[[L6:.+]]=, $pop[[L3]], $pop[[L4]] 38 ; CHECK-NEXT: tee_local $push[[L5:.+]]=, $[[SP:.+]]=, $pop[[L6]] 39 %r1 = alloca i32 40 %r2 = alloca double 41 ; CHECK-NEXT: i32.const $push[[L0:.+]]=, 0 42 ; CHECK-NEXT: i32.store $drop=, 12($pop[[L5]]), $pop[[L0]] 43 store i32 0, i32* %r1 44 ; CHECK-NEXT: i64.const $push[[L1:.+]]=, 0 45 ; CHECK-NEXT: i64.store $drop=, 0($[[SP]]), $pop[[L1]] 46 store double 0.0, double* %r2 47 ; CHECK-NEXT: return 48 ret void 49} 50 51; CHECK-LABEL: allocarray: 52; CHECK: .local i32{{$}} 53define void @allocarray() { 54 ; CHECK: i32.const $push[[L7:.+]]=, 0{{$}} 55 ; CHECK: i32.const $push[[L4:.+]]=, 0{{$}} 56 ; CHECK-NEXT: i32.load $push[[L5:.+]]=, __stack_pointer($pop[[L4]]) 57 ; CHECK-NEXT: i32.const $push[[L6:.+]]=, 144{{$}} 58 ; CHECK-NEXT: i32.sub $push[[L11:.+]]=, $pop[[L5]], $pop[[L6]] 59 ; CHECK-NEXT: i32.store ${{.+}}=, __stack_pointer($pop[[L7]]), $pop[[L11]] 60 %r = alloca [33 x i32] 61 62 ; CHECK: i32.const $push{{.+}}=, 24 63 ; CHECK-NEXT: i32.add $push[[L3:.+]]=, $[[SP]], $pop{{.+}} 64 ; CHECK-NEXT: i32.const $push[[L1:.+]]=, 1{{$}} 65 ; CHECK-NEXT: i32.store $push[[L0:.+]]=, 0($pop[[L3]]), $pop[[L1]]{{$}} 66 ; CHECK-NEXT: i32.store $drop=, 12(${{.+}}), $pop[[L0]]{{$}} 67 %p = getelementptr [33 x i32], [33 x i32]* %r, i32 0, i32 0 68 store i32 1, i32* %p 69 %p2 = getelementptr [33 x i32], [33 x i32]* %r, i32 0, i32 3 70 store i32 1, i32* %p2 71 72 ; CHECK: i32.const $push[[L10:.+]]=, 0{{$}} 73 ; CHECK-NEXT: i32.const $push[[L8:.+]]=, 144 74 ; CHECK-NEXT: i32.add $push[[L19:.+]]=, $[[SP]], $pop[[L8]] 75 ; CHECK-NEXT: i32.store $drop=, __stack_pointer($pop[[L10]]), $pop[[L9]] 76 ret void 77} 78 79; CHECK-LABEL: non_mem_use 80define void @non_mem_use(i8** %addr) { 81 ; CHECK: i32.const $push[[L1:.+]]=, 48 82 ; CHECK-NEXT: i32.sub $push[[L11:.+]]=, {{.+}}, $pop[[L1]] 83 ; CHECK-NEXT: i32.store $[[SP:.+]]=, {{.+}}, $pop[[L11]] 84 %buf = alloca [27 x i8], align 16 85 %r = alloca i64 86 %r2 = alloca i64 87 ; %r is at SP+8 88 ; CHECK: tee_local $push[[L12:.+]]=, $[[SP:.+]]=, $pop{{.+}} 89 ; CHECK: i32.const $push[[OFF:.+]]=, 8 90 ; CHECK-NEXT: i32.add $push[[ARG1:.+]]=, $pop[[L12]], $pop[[OFF]] 91 ; CHECK-NEXT: call ext_func@FUNCTION, $pop[[ARG1]] 92 call void @ext_func(i64* %r) 93 ; %r2 is at SP+0, no add needed 94 ; CHECK-NEXT: call ext_func@FUNCTION, $[[SP]] 95 call void @ext_func(i64* %r2) 96 ; Use as a value, but in a store 97 ; %buf is at SP+16 98 ; CHECK: i32.const $push[[OFF:.+]]=, 16 99 ; CHECK-NEXT: i32.add $push[[VAL:.+]]=, $[[SP]], $pop[[OFF]] 100 ; CHECK-NEXT: i32.store {{.*}}=, 0($0), $pop[[VAL]] 101 %gep = getelementptr inbounds [27 x i8], [27 x i8]* %buf, i32 0, i32 0 102 store i8* %gep, i8** %addr 103 ret void 104} 105 106; CHECK-LABEL: allocarray_inbounds: 107; CHECK: .local i32{{$}} 108define void @allocarray_inbounds() { 109 ; CHECK: i32.const $push[[L6:.+]]=, 0{{$}} 110 ; CHECK: i32.const $push[[L3:.+]]=, 0{{$}} 111 ; CHECK-NEXT: i32.load $push[[L4:.+]]=, __stack_pointer($pop[[L3]]) 112 ; CHECK-NEXT: i32.const $push[[L5:.+]]=, 32{{$}} 113 ; CHECK-NEXT: i32.sub $push[[L10:.+]]=, $pop[[L4]], $pop[[L5]] 114 ; CHECK-NEXT: i32.store ${{.+}}=, __stack_pointer($pop[[L6]]), $pop[[L10]]{{$}} 115 %r = alloca [5 x i32] 116 ; CHECK: i32.const $push[[L3:.+]]=, 1 117 ; CHECK-DAG: i32.store $push{{.*}}=, 24(${{.+}}), $pop[[L3]] 118 %p = getelementptr inbounds [5 x i32], [5 x i32]* %r, i32 0, i32 0 119 store i32 1, i32* %p 120 ; This store should have both the GEP and the FI folded into it. 121 ; CHECK-DAG: i32.store {{.*}}=, 12(${{.+}}), $pop 122 %p2 = getelementptr inbounds [5 x i32], [5 x i32]* %r, i32 0, i32 3 123 store i32 1, i32* %p2 124 call void @ext_func(i64* null); 125 ; CHECK: call ext_func 126 ; CHECK: i32.const $push[[L6:.+]]=, 0{{$}} 127 ; CHECK-NEXT: i32.const $push[[L5:.+]]=, 32{{$}} 128 ; CHECK-NEXT: i32.add $push[[L7:.+]]=, ${{.+}}, $pop[[L5]] 129 ; CHECK-NEXT: i32.store $drop=, __stack_pointer($pop[[L6]]), $pop[[L7]] 130 ret void 131} 132 133; CHECK-LABEL: dynamic_alloca: 134define void @dynamic_alloca(i32 %alloc) { 135 ; CHECK: i32.const $push[[L7:.+]]=, 0{{$}} 136 ; CHECK: i32.const $push[[L1:.+]]=, 0{{$}} 137 ; CHECK-NEXT: i32.load $push[[L13:.+]]=, __stack_pointer($pop[[L1]]) 138 ; CHECK-NEXT: tee_local $push[[L12:.+]]=, [[SP:.+]], $pop[[L13]]{{$}} 139 ; Target independent codegen bumps the stack pointer. 140 ; CHECK: i32.sub 141 ; Check that SP is written back to memory after decrement 142 ; CHECK: i32.store $drop=, __stack_pointer($pop{{.+}}), 143 %r = alloca i32, i32 %alloc 144 ; Target-independent codegen also calculates the store addr 145 ; CHECK: call ext_func_i32@FUNCTION 146 call void @ext_func_i32(i32* %r) 147 ; CHECK: i32.const $push[[L3:.+]]=, 0{{$}} 148 ; CHECK: i32.store $drop=, __stack_pointer($pop[[L3]]), $pop{{.+}} 149 ret void 150} 151 152; CHECK-LABEL: dynamic_alloca_redzone: 153define void @dynamic_alloca_redzone(i32 %alloc) { 154 ; CHECK: i32.const $push[[L8:.+]]=, 0{{$}} 155 ; CHECK-NEXT: i32.load $push[[L13:.+]]=, __stack_pointer($pop[[L1]]) 156 ; CHECK-NEXT: tee_local $push[[L12:.+]]=, [[SP:.+]], $pop[[L13]]{{$}} 157 ; CHECK-NEXT: copy_local [[FP:.+]]=, $pop[[L12]]{{$}} 158 ; Target independent codegen bumps the stack pointer 159 ; CHECK: i32.sub 160 %r = alloca i32, i32 %alloc 161 ; CHECK-NEXT: tee_local $push[[L8:.+]]=, $0=, $pop 162 ; CHECK-NEXT: copy_local $drop=, $pop[[L8]]{{$}} 163 ; CHECK-NEXT: i32.const $push[[L6:.+]]=, 0{{$}} 164 ; CHECK-NEXT: i32.store $drop=, 0($0), $pop[[L6]]{{$}} 165 store i32 0, i32* %r 166 ; CHECK-NEXT: return 167 ret void 168} 169 170; CHECK-LABEL: dynamic_static_alloca: 171define void @dynamic_static_alloca(i32 %alloc) noredzone { 172 ; Decrement SP in the prolog by the static amount and writeback to memory. 173 ; CHECK: i32.const $push[[L7:.+]]=, 0{{$}} 174 ; CHECK: i32.const $push[[L8:.+]]=, 0{{$}} 175 ; CHECK: i32.const $push[[L9:.+]]=, 0{{$}} 176 ; CHECK-NEXT: i32.load $push[[L10:.+]]=, __stack_pointer($pop[[L9]]) 177 ; CHECK-NEXT: i32.const $push[[L11:.+]]=, 16 178 ; CHECK-NEXT: i32.sub $push[[L20:.+]]=, $pop[[L10]], $pop[[L11]] 179 ; CHECK-NEXT: tee_local $push[[L19:.+]]=, $[[FP:.+]]=, $pop[[L20]] 180 ; CHECK: i32.store $push[[L0:.+]]=, __stack_pointer($pop{{.+}}), $pop{{.+}} 181 ; Decrement SP in the body by the dynamic amount. 182 ; CHECK: i32.sub 183 ; Writeback to memory. 184 ; CHECK: i32.store $drop=, __stack_pointer($pop{{.+}}), $pop{{.+}} 185 %r1 = alloca i32 186 %r = alloca i32, i32 %alloc 187 store i32 0, i32* %r 188 ; CHEC: i32.store $drop=, 0($pop{{.+}}), $pop{{.+}} 189 ret void 190} 191 192; The use of the alloca in a phi causes a CopyToReg DAG node to be generated, 193; which has to have special handling because CopyToReg can't have a FI operand 194; CHECK-LABEL: copytoreg_fi: 195define void @copytoreg_fi(i1 %cond, i32* %b) { 196entry: 197 ; CHECK: i32.const $push[[L1:.+]]=, 16 198 ; CHECK-NEXT: i32.sub $push[[L3:.+]]=, {{.+}}, $pop[[L1]] 199 %addr = alloca i32 200 ; CHECK: i32.const $push[[OFF:.+]]=, 12 201 ; CHECK-NEXT: i32.add $push[[ADDR:.+]]=, $pop[[L3]], $pop[[OFF]] 202 ; CHECK-NEXT: copy_local [[COPY:.+]]=, $pop[[ADDR]] 203 br label %body 204body: 205 %a = phi i32* [%addr, %entry], [%b, %body] 206 store i32 1, i32* %a 207 ; CHECK: i32.store {{.*}}, 0([[COPY]]), 208 br i1 %cond, label %body, label %exit 209exit: 210 ret void 211} 212 213declare void @use_i8_star(i8*) 214declare i8* @llvm.frameaddress(i32) 215 216; Test __builtin_frame_address(0). 217; CHECK-LABEL: frameaddress_0: 218; CHECK: i32.const $push[[L0:.+]]=, 0{{$}} 219; CHECK-NEXT: i32.load $push[[L3:.+]]=, __stack_pointer($pop[[L0]]) 220; CHECK-NEXT: copy_local $push[[L4:.+]]=, $pop[[L3]]{{$}} 221; CHECK-NEXT: tee_local $push[[L2:.+]]=, $[[FP:.+]]=, $pop[[L4]]{{$}} 222; CHECK-NEXT: call use_i8_star@FUNCTION, $pop[[L2]] 223; CHECK-NEXT: i32.const $push[[L1:.+]]=, 0{{$}} 224; CHECK-NEXT: i32.store $drop=, __stack_pointer($pop[[L1]]), $[[FP]] 225define void @frameaddress_0() { 226 %t = call i8* @llvm.frameaddress(i32 0) 227 call void @use_i8_star(i8* %t) 228 ret void 229} 230 231; Test __builtin_frame_address(1). 232 233; CHECK-LABEL: frameaddress_1: 234; CHECK-NEXT: i32.const $push0=, 0{{$}} 235; CHECK-NEXT: call use_i8_star@FUNCTION, $pop0{{$}} 236; CHECK-NEXT: return{{$}} 237define void @frameaddress_1() { 238 %t = call i8* @llvm.frameaddress(i32 1) 239 call void @use_i8_star(i8* %t) 240 ret void 241} 242 243; Test a stack address passed to an inline asm. 244; CHECK-LABEL: inline_asm: 245; CHECK: __stack_pointer 246; CHECK: #APP 247; CHECK-NEXT: # %{{[0-9]+}}{{$}} 248; CHECK-NEXT: #NO_APP 249define void @inline_asm() { 250 %tmp = alloca i8 251 call void asm sideeffect "# %0", "r"(i8* %tmp) 252 ret void 253} 254 255; TODO: test over-aligned alloca 256