1; RUN: llc < %s -mtriple=arm64-apple-darwin -enable-misched=0 -mcpu=cyclone | FileCheck %s 2 3; Trivial patchpoint codegen 4; 5define i64 @trivial_patchpoint_codegen(i64 %p1, i64 %p2, i64 %p3, i64 %p4) { 6entry: 7; CHECK-LABEL: trivial_patchpoint_codegen: 8; CHECK: movz x16, #0xdead, lsl #32 9; CHECK-NEXT: movk x16, #0xbeef, lsl #16 10; CHECK-NEXT: movk x16, #0xcafe 11; CHECK-NEXT: blr x16 12; CHECK: movz x16, #0xdead, lsl #32 13; CHECK-NEXT: movk x16, #0xbeef, lsl #16 14; CHECK-NEXT: movk x16, #0xcaff 15; CHECK-NEXT: blr x16 16; CHECK: ret 17 %resolveCall2 = inttoptr i64 244837814094590 to i8* 18 %result = tail call i64 (i64, i32, i8*, i32, ...)* @llvm.experimental.patchpoint.i64(i64 2, i32 20, i8* %resolveCall2, i32 4, i64 %p1, i64 %p2, i64 %p3, i64 %p4) 19 %resolveCall3 = inttoptr i64 244837814094591 to i8* 20 tail call void (i64, i32, i8*, i32, ...)* @llvm.experimental.patchpoint.void(i64 3, i32 20, i8* %resolveCall3, i32 2, i64 %p1, i64 %result) 21 ret i64 %result 22} 23 24; Caller frame metadata with stackmaps. This should not be optimized 25; as a leaf function. 26; 27; CHECK-LABEL: caller_meta_leaf 28; CHECK: mov x29, sp 29; CHECK-NEXT: sub sp, sp, #32 30; CHECK: Ltmp 31; CHECK: mov sp, x29 32; CHECK: ret 33 34define void @caller_meta_leaf() { 35entry: 36 %metadata = alloca i64, i32 3, align 8 37 store i64 11, i64* %metadata 38 store i64 12, i64* %metadata 39 store i64 13, i64* %metadata 40 call void (i64, i32, ...)* @llvm.experimental.stackmap(i64 4, i32 0, i64* %metadata) 41 ret void 42} 43 44; Test the webkit_jscc calling convention. 45; One argument will be passed in register, the other will be pushed on the stack. 46; Return value in x0. 47define void @jscall_patchpoint_codegen(i64 %p1, i64 %p2, i64 %p3, i64 %p4) { 48entry: 49; CHECK-LABEL: jscall_patchpoint_codegen: 50; CHECK: Ltmp 51; CHECK: str x{{.+}}, [sp] 52; CHECK-NEXT: mov x0, x{{.+}} 53; CHECK: Ltmp 54; CHECK-NEXT: movz x16, #0xffff, lsl #32 55; CHECK-NEXT: movk x16, #0xdead, lsl #16 56; CHECK-NEXT: movk x16, #0xbeef 57; CHECK-NEXT: blr x16 58 %resolveCall2 = inttoptr i64 281474417671919 to i8* 59 %result = tail call webkit_jscc i64 (i64, i32, i8*, i32, ...)* @llvm.experimental.patchpoint.i64(i64 5, i32 20, i8* %resolveCall2, i32 2, i64 %p4, i64 %p2) 60 %resolveCall3 = inttoptr i64 244837814038255 to i8* 61 tail call webkit_jscc void (i64, i32, i8*, i32, ...)* @llvm.experimental.patchpoint.void(i64 6, i32 20, i8* %resolveCall3, i32 2, i64 %p4, i64 %result) 62 ret void 63} 64 65; Test if the arguments are properly aligned and that we don't store undef arguments. 66define i64 @jscall_patchpoint_codegen2(i64 %callee) { 67entry: 68; CHECK-LABEL: jscall_patchpoint_codegen2: 69; CHECK: Ltmp 70; CHECK: orr w{{.+}}, wzr, #0x6 71; CHECK-NEXT: str x{{.+}}, [sp, #24] 72; CHECK-NEXT: orr w{{.+}}, wzr, #0x4 73; CHECK-NEXT: str w{{.+}}, [sp, #16] 74; CHECK-NEXT: orr w{{.+}}, wzr, #0x2 75; CHECK-NEXT: str x{{.+}}, [sp] 76; CHECK: Ltmp 77; CHECK-NEXT: movz x16, #0xffff, lsl #32 78; CHECK-NEXT: movk x16, #0xdead, lsl #16 79; CHECK-NEXT: movk x16, #0xbeef 80; CHECK-NEXT: blr x16 81 %call = inttoptr i64 281474417671919 to i8* 82 %result = call webkit_jscc i64 (i64, i32, i8*, i32, ...)* @llvm.experimental.patchpoint.i64(i64 7, i32 20, i8* %call, i32 6, i64 %callee, i64 2, i64 undef, i32 4, i32 undef, i64 6) 83 ret i64 %result 84} 85 86; Test if the arguments are properly aligned and that we don't store undef arguments. 87define i64 @jscall_patchpoint_codegen3(i64 %callee) { 88entry: 89; CHECK-LABEL: jscall_patchpoint_codegen3: 90; CHECK: Ltmp 91; CHECK: movz w{{.+}}, #0xa 92; CHECK-NEXT: str x{{.+}}, [sp, #48] 93; CHECK-NEXT: orr w{{.+}}, wzr, #0x8 94; CHECK-NEXT: str w{{.+}}, [sp, #36] 95; CHECK-NEXT: orr w{{.+}}, wzr, #0x6 96; CHECK-NEXT: str x{{.+}}, [sp, #24] 97; CHECK-NEXT: orr w{{.+}}, wzr, #0x4 98; CHECK-NEXT: str w{{.+}}, [sp, #16] 99; CHECK-NEXT: orr w{{.+}}, wzr, #0x2 100; CHECK-NEXT: str x{{.+}}, [sp] 101; CHECK: Ltmp 102; CHECK-NEXT: movz x16, #0xffff, lsl #32 103; CHECK-NEXT: movk x16, #0xdead, lsl #16 104; CHECK-NEXT: movk x16, #0xbeef 105; CHECK-NEXT: blr x16 106 %call = inttoptr i64 281474417671919 to i8* 107 %result = call webkit_jscc i64 (i64, i32, i8*, i32, ...)* @llvm.experimental.patchpoint.i64(i64 7, i32 20, i8* %call, i32 10, i64 %callee, i64 2, i64 undef, i32 4, i32 undef, i64 6, i32 undef, i32 8, i32 undef, i64 10) 108 ret i64 %result 109} 110 111; Test patchpoints reusing the same TargetConstant. 112; <rdar:15390785> Assertion failed: (CI.getNumArgOperands() >= NumArgs + 4) 113; There is no way to verify this, since it depends on memory allocation. 114; But I think it's useful to include as a working example. 115define i64 @testLowerConstant(i64 %arg, i64 %tmp2, i64 %tmp10, i64* %tmp33, i64 %tmp79) { 116entry: 117 %tmp80 = add i64 %tmp79, -16 118 %tmp81 = inttoptr i64 %tmp80 to i64* 119 %tmp82 = load i64* %tmp81, align 8 120 tail call void (i64, i32, ...)* @llvm.experimental.stackmap(i64 14, i32 8, i64 %arg, i64 %tmp2, i64 %tmp10, i64 %tmp82) 121 tail call void (i64, i32, i8*, i32, ...)* @llvm.experimental.patchpoint.void(i64 15, i32 32, i8* null, i32 3, i64 %arg, i64 %tmp10, i64 %tmp82) 122 %tmp83 = load i64* %tmp33, align 8 123 %tmp84 = add i64 %tmp83, -24 124 %tmp85 = inttoptr i64 %tmp84 to i64* 125 %tmp86 = load i64* %tmp85, align 8 126 tail call void (i64, i32, ...)* @llvm.experimental.stackmap(i64 17, i32 8, i64 %arg, i64 %tmp10, i64 %tmp86) 127 tail call void (i64, i32, i8*, i32, ...)* @llvm.experimental.patchpoint.void(i64 18, i32 32, i8* null, i32 3, i64 %arg, i64 %tmp10, i64 %tmp86) 128 ret i64 10 129} 130 131; Test small patchpoints that don't emit calls. 132define void @small_patchpoint_codegen(i64 %p1, i64 %p2, i64 %p3, i64 %p4) { 133entry: 134; CHECK-LABEL: small_patchpoint_codegen: 135; CHECK: Ltmp 136; CHECK: nop 137; CHECK-NEXT: nop 138; CHECK-NEXT: nop 139; CHECK-NEXT: nop 140; CHECK-NEXT: nop 141; CHECK-NEXT: ldp 142; CHECK-NEXT: ret 143 %result = tail call i64 (i64, i32, i8*, i32, ...)* @llvm.experimental.patchpoint.i64(i64 5, i32 20, i8* null, i32 2, i64 %p1, i64 %p2) 144 ret void 145} 146 147; Test that scratch registers are spilled around patchpoints 148; CHECK: InlineAsm End 149; CHECK-NEXT: mov x{{[0-9]+}}, x16 150; CHECK-NEXT: mov x{{[0-9]+}}, x17 151; CHECK-NEXT: Ltmp 152; CHECK-NEXT: nop 153define void @clobberScratch(i32* %p) { 154 %v = load i32* %p 155 tail call void asm sideeffect "nop", "~{x0},~{x1},~{x2},~{x3},~{x4},~{x5},~{x6},~{x7},~{x8},~{x9},~{x10},~{x11},~{x12},~{x13},~{x14},~{x15},~{x18},~{x19},~{x20},~{x21},~{x22},~{x23},~{x24},~{x25},~{x26},~{x27},~{x28},~{x29},~{x30},~{x31}"() nounwind 156 tail call void (i64, i32, i8*, i32, ...)* @llvm.experimental.patchpoint.void(i64 5, i32 20, i8* null, i32 0, i32* %p, i32 %v) 157 store i32 %v, i32* %p 158 ret void 159} 160 161declare void @llvm.experimental.stackmap(i64, i32, ...) 162declare void @llvm.experimental.patchpoint.void(i64, i32, i8*, i32, ...) 163declare i64 @llvm.experimental.patchpoint.i64(i64, i32, i8*, i32, ...) 164 165; CHECK-LABEL: test_i16: 166; CHECK: ldrh [[BREG:w[0-9]+]], [sp] 167; CHECK: add w0, w0, [[BREG]] 168define webkit_jscc i16 @test_i16(i16 zeroext %a, i16 zeroext %b) { 169 %sum = add i16 %a, %b 170 ret i16 %sum 171} 172