1 // RUN: %clang_cc1 -emit-llvm -triple arm64-- -fexperimental-max-bitint-width=1024 -o - %s -O0 | FileCheck %s --check-prefix=CHECK-O0 2 // RUN: %clang_cc1 -emit-llvm -disable-llvm-optzns -fexperimental-max-bitint-width=1024 -triple arm64-- -o - %s -O3 | FileCheck %s --check-prefix=CHECK-O3 3 4 struct large { 5 void* pointers[8]; 6 }; 7 8 void pass_large(struct large); 9 void pass_large_BitInt(_BitInt(129)); 10 11 // For arm64, we don't use byval to pass structs and _BitInt(>128) type, but instead we create 12 // temporary allocas. 13 // 14 // Make sure we generate the appropriate lifetime markers for the temporary 15 // allocas so that the optimizer can re-use stack slots if possible. 16 void example(void) { 17 struct large l = {0}; 18 pass_large(l); 19 pass_large(l); 20 } 21 // CHECK-O0-LABEL: define{{.*}} void @example( 22 // The alloca for the struct on the stack. 23 // CHECK-O0: %[[l:[0-9A-Za-z-]+]] = alloca %struct.large, align 8 24 // The alloca for the temporary stack space that we use to pass the argument. 25 // CHECK-O0-NEXT: %[[byvaltemp:[0-9A-Za-z-]+]] = alloca %struct.large, align 8 26 // Another one to pass the argument to the second function call. 27 // CHECK-O0-NEXT: %[[byvaltemp1:[0-9A-Za-z-]+]] = alloca %struct.large, align 8 28 // First, memset `l` to 0. 29 // CHECK-O0-NEXT: call void @llvm.memset.p0.i64(ptr align 8 %[[l]], i8 0, i64 64, i1 false) 30 // Then, memcpy `l` to the temporary stack space. 31 // CHECK-O0-NEXT: call void @llvm.memcpy.p0.p0.i64(ptr align 8 %[[byvaltemp]], ptr align 8 %[[l]], i64 64, i1 false) 32 // Finally, call using a pointer to the temporary stack space. 33 // CHECK-O0-NEXT: call void @pass_large(ptr noundef %[[byvaltemp]]) 34 // Now, do the same for the second call, using the second temporary alloca. 35 // CHECK-O0-NEXT: call void @llvm.memcpy.p0.p0.i64(ptr align 8 %[[byvaltemp1]], ptr align 8 %[[l]], i64 64, i1 false) 36 // CHECK-O0-NEXT: call void @pass_large(ptr noundef %[[byvaltemp1]]) 37 // CHECK-O0-NEXT: ret void 38 // 39 // At O3, we should have lifetime markers to help the optimizer re-use the temporary allocas. 40 // 41 // CHECK-O3-LABEL: define{{.*}} void @example( 42 // The alloca for the struct on the stack. 43 // CHECK-O3: %[[l:[0-9A-Za-z-]+]] = alloca %struct.large, align 8 44 // The alloca for the temporary stack space that we use to pass the argument. 45 // CHECK-O3-NEXT: %[[byvaltemp:[0-9A-Za-z-]+]] = alloca %struct.large, align 8 46 // Another one to pass the argument to the second function call. 47 // CHECK-O3-NEXT: %[[byvaltemp1:[0-9A-Za-z-]+]] = alloca %struct.large, align 8 48 // 49 // Mark the start of the lifetime for `l` 50 // CHECK-O3-NEXT: call void @llvm.lifetime.start.p0(i64 64, ptr %[[l]]) 51 // 52 // First, memset `l` to 0. 53 // CHECK-O3-NEXT: call void @llvm.memset.p0.i64(ptr align 8 %[[l]], i8 0, i64 64, i1 false) 54 // 55 // Lifetime of the first temporary starts here and ends right after the call. 56 // CHECK-O3-NEXT: call void @llvm.lifetime.start.p0(i64 64, ptr %[[byvaltemp]]) 57 // 58 // Then, memcpy `l` to the temporary stack space. 59 // CHECK-O3-NEXT: call void @llvm.memcpy.p0.p0.i64(ptr align 8 %[[byvaltemp]], ptr align 8 %[[l]], i64 64, i1 false) 60 // Finally, call using a pointer to the temporary stack space. 61 // CHECK-O3-NEXT: call void @pass_large(ptr noundef %[[byvaltemp]]) 62 // 63 // The lifetime of the temporary used to pass a pointer to the struct ends here. 64 // CHECK-O3-NEXT: call void @llvm.lifetime.end.p0(i64 64, ptr %[[byvaltemp]]) 65 // 66 // Now, do the same for the second call, using the second temporary alloca. 67 // CHECK-O3-NEXT: call void @llvm.lifetime.start.p0(i64 64, ptr %[[byvaltemp1]]) 68 // CHECK-O3-NEXT: call void @llvm.memcpy.p0.p0.i64(ptr align 8 %[[byvaltemp1]], ptr align 8 %[[l]], i64 64, i1 false) 69 // CHECK-O3-NEXT: call void @pass_large(ptr noundef %[[byvaltemp1]]) 70 // CHECK-O3-NEXT: call void @llvm.lifetime.end.p0(i64 64, ptr %[[byvaltemp1]]) 71 // 72 // Mark the end of the lifetime of `l`. 73 // CHECK-O3-NEXT: call void @llvm.lifetime.end.p0(i64 64, ptr %l) 74 // CHECK-O3-NEXT: ret void 75 76 void example_BitInt(void) { 77 _BitInt(129) l = {0}; 78 pass_large_BitInt(l); 79 pass_large_BitInt(l); 80 } 81 // CHECK-O0-LABEL: define dso_local void @example_BitInt( 82 // CHECK-O0-NEXT: entry: 83 // CHECK-O0-NEXT: [[L:%.*]] = alloca i256, align 16 84 // CHECK-O0-NEXT: [[INDIRECT_ARG_TEMP:%.*]] = alloca i256, align 16 85 // CHECK-O0-NEXT: [[INDIRECT_ARG_TEMP1:%.*]] = alloca i256, align 16 86 // CHECK-O0-NEXT: store i256 0, ptr [[L]], align 16 87 // CHECK-O0-NEXT: [[TMP0:%.*]] = load i256, ptr [[L]], align 16 88 // CHECK-O0-NEXT: [[LOADEDV:%.*]] = trunc i256 [[TMP0]] to i129 89 // CHECK-O0-NEXT: [[STOREDV:%.*]] = sext i129 [[LOADEDV]] to i256 90 // CHECK-O0-NEXT: store i256 [[STOREDV]], ptr [[INDIRECT_ARG_TEMP]], align 16 91 // CHECK-O0-NEXT: call void @pass_large_BitInt(ptr noundef [[INDIRECT_ARG_TEMP]]) 92 // CHECK-O0-NEXT: [[TMP1:%.*]] = load i256, ptr [[L]], align 16 93 // CHECK-O0-NEXT: [[LOADEDV1:%.*]] = trunc i256 [[TMP1]] to i129 94 // CHECK-O0-NEXT: [[STOREDV1:%.*]] = sext i129 [[LOADEDV1]] to i256 95 // CHECK-O0-NEXT: store i256 [[STOREDV1]], ptr [[INDIRECT_ARG_TEMP1]], align 16 96 // CHECK-O0-NEXT: call void @pass_large_BitInt(ptr noundef [[INDIRECT_ARG_TEMP1]]) 97 // CHECK-O0-NEXT: ret void 98 // 99 // CHECK-O3-LABEL: define dso_local void @example_BitInt( 100 // CHECK-O3-NEXT: entry: 101 // CHECK-O3-NEXT: [[L:%.*]] = alloca i256, align 16 102 // CHECK-O3-NEXT: [[INDIRECT_ARG_TEMP:%.*]] = alloca i256, align 16 103 // CHECK-O3-NEXT: [[INDIRECT_ARG_TEMP1:%.*]] = alloca i256, align 16 104 // CHECK-O3-NEXT: call void @llvm.lifetime.start.p0(i64 32, ptr [[L]]) 105 // CHECK-O3-NEXT: store i256 0, ptr [[L]], align 16, !tbaa [[TBAA6:![0-9]+]] 106 // CHECK-O3-NEXT: [[TMP0:%.*]] = load i256, ptr [[L]], align 16, !tbaa [[TBAA6]] 107 // CHECK-O3-NEXT: [[LOADEDV:%.*]] = trunc i256 [[TMP0]] to i129 108 // CHECK-O3-NEXT: call void @llvm.lifetime.start.p0(i64 32, ptr [[INDIRECT_ARG_TEMP]]) 109 // CHECK-O3-NEXT: [[STOREDV:%.*]] = sext i129 [[LOADEDV]] to i256 110 // CHECK-O3-NEXT: store i256 [[STOREDV]], ptr [[INDIRECT_ARG_TEMP]], align 16, !tbaa [[TBAA6]] 111 // CHECK-O3-NEXT: call void @pass_large_BitInt(ptr noundef [[INDIRECT_ARG_TEMP]]) 112 // CHECK-O3-NEXT: call void @llvm.lifetime.end.p0(i64 32, ptr [[INDIRECT_ARG_TEMP]]) 113 // CHECK-O3-NEXT: [[TMP1:%.*]] = load i256, ptr [[L]], align 16, !tbaa [[TBAA6]] 114 // CHECK-O3-NEXT: [[LOADEDV1:%.*]] = trunc i256 [[TMP1]] to i129 115 // CHECK-O3-NEXT: call void @llvm.lifetime.start.p0(i64 32, ptr [[INDIRECT_ARG_TEMP1]]) 116 // CHECK-O3-NEXT: [[STOREDV1:%.*]] = sext i129 [[LOADEDV1]] to i256 117 // CHECK-O3-NEXT: store i256 [[STOREDV1]], ptr [[INDIRECT_ARG_TEMP1]], align 16, !tbaa [[TBAA6]] 118 // CHECK-O3-NEXT: call void @pass_large_BitInt(ptr noundef [[INDIRECT_ARG_TEMP1]]) 119 // CHECK-O3-NEXT: call void @llvm.lifetime.end.p0(i64 32, ptr [[INDIRECT_ARG_TEMP1]]) 120 // CHECK-O3-NEXT: call void @llvm.lifetime.end.p0(i64 32, ptr [[L]]) 121