1; RUN: llc < %s -mcpu=generic -mtriple=x86_64-pc-win32 | FileCheck %s 2 3; Verify that the var arg parameters which are passed in registers are stored 4; in home stack slots allocated by the caller and that AP is correctly 5; calculated. 6define void @average_va(i32 %count, ...) nounwind { 7entry: 8; CHECK: pushq 9; CHECK: movq %r9, 40(%rsp) 10; CHECK: movq %r8, 32(%rsp) 11; CHECK: movq %rdx, 24(%rsp) 12; CHECK: leaq 24(%rsp), %rax 13 14 %ap = alloca i8*, align 8 ; <i8**> [#uses=1] 15 %ap1 = bitcast i8** %ap to i8* ; <i8*> [#uses=1] 16 call void @llvm.va_start(i8* %ap1) 17 ret void 18} 19 20declare void @llvm.va_start(i8*) nounwind 21declare void @llvm.va_copy(i8*, i8*) nounwind 22 23; CHECK-LABEL: f5: 24; CHECK: pushq 25; CHECK: leaq 56(%rsp), 26define i8* @f5(i64 %a0, i64 %a1, i64 %a2, i64 %a3, i64 %a4, ...) nounwind { 27entry: 28 %ap = alloca i8*, align 8 29 %ap1 = bitcast i8** %ap to i8* 30 call void @llvm.va_start(i8* %ap1) 31 ret i8* %ap1 32} 33 34; CHECK-LABEL: f4: 35; CHECK: pushq 36; CHECK: leaq 48(%rsp), 37define i8* @f4(i64 %a0, i64 %a1, i64 %a2, i64 %a3, ...) nounwind { 38entry: 39 %ap = alloca i8*, align 8 40 %ap1 = bitcast i8** %ap to i8* 41 call void @llvm.va_start(i8* %ap1) 42 ret i8* %ap1 43} 44 45; CHECK-LABEL: f3: 46; CHECK: pushq 47; CHECK: leaq 40(%rsp), 48define i8* @f3(i64 %a0, i64 %a1, i64 %a2, ...) nounwind { 49entry: 50 %ap = alloca i8*, align 8 51 %ap1 = bitcast i8** %ap to i8* 52 call void @llvm.va_start(i8* %ap1) 53 ret i8* %ap1 54} 55 56; WinX86_64 uses char* for va_list. Verify that the correct amount of bytes 57; are copied using va_copy. 58 59; CHECK-LABEL: copy1: 60; CHECK: subq $16 61; CHECK: leaq 32(%rsp), [[REG_copy1:%[a-z]+]] 62; CHECK: movq [[REG_copy1]], 8(%rsp) 63; CHECK: movq [[REG_copy1]], (%rsp) 64; CHECK: addq $16 65; CHECK: ret 66define void @copy1(i64 %a0, ...) nounwind { 67entry: 68 %ap = alloca i8*, align 8 69 %cp = alloca i8*, align 8 70 %ap1 = bitcast i8** %ap to i8* 71 %cp1 = bitcast i8** %cp to i8* 72 call void @llvm.va_start(i8* %ap1) 73 call void @llvm.va_copy(i8* %cp1, i8* %ap1) 74 ret void 75} 76 77; CHECK-LABEL: copy4: 78; CHECK: subq $16 79; CHECK: leaq 56(%rsp), [[REG_copy4:%[a-z]+]] 80; CHECK: movq [[REG_copy4]], 8(%rsp) 81; CHECK: movq [[REG_copy4]], (%rsp) 82; CHECK: addq $16 83; CHECK: ret 84define void @copy4(i64 %a0, i64 %a1, i64 %a2, i64 %a3, ...) nounwind { 85entry: 86 %ap = alloca i8*, align 8 87 %cp = alloca i8*, align 8 88 %ap1 = bitcast i8** %ap to i8* 89 %cp1 = bitcast i8** %cp to i8* 90 call void @llvm.va_start(i8* %ap1) 91 call void @llvm.va_copy(i8* %cp1, i8* %ap1) 92 ret void 93} 94 95; CHECK-LABEL: arg4: 96; CHECK: pushq 97; va_start (optimized away as overwritten by va_arg) 98; va_arg: 99; CHECK: leaq 52(%rsp), [[REG_arg4_2:%[a-z]+]] 100; CHECK: movq [[REG_arg4_2]], (%rsp) 101; CHECK: movl 48(%rsp), %eax 102; CHECK: popq 103; CHECK: ret 104define i32 @arg4(i64 %a0, i64 %a1, i64 %a2, i64 %a3, ...) nounwind { 105entry: 106 %ap = alloca i8*, align 8 107 %ap1 = bitcast i8** %ap to i8* 108 call void @llvm.va_start(i8* %ap1) 109 %tmp = va_arg i8** %ap, i32 110 ret i32 %tmp 111} 112 113define void @sret_arg(i32* sret %agg.result, i8* nocapture readnone %format, ...) { 114entry: 115 %ap = alloca i8* 116 %ap_i8 = bitcast i8** %ap to i8* 117 call void @llvm.va_start(i8* %ap_i8) 118 %tmp = va_arg i8** %ap, i32 119 store i32 %tmp, i32* %agg.result 120 ret void 121} 122; CHECK-LABEL: sret_arg: 123; CHECK: pushq 124; CHECK-DAG: movq %r9, 40(%rsp) 125; CHECK-DAG: movq %r8, 32(%rsp) 126; CHECK: movl 32(%rsp), %[[tmp:[^ ]*]] 127; CHECK: movl %[[tmp]], (%[[sret:[^ ]*]]) 128; CHECK: movq %[[sret]], %rax 129; CHECK: popq 130; CHECK: retq 131