171 lines · plain
1// NOTE: Assertions have been autogenerated by utils/update_cc_test_checks.py UTC_ARGS: --version 52// RUN: %clang_cc1 -O0 -cl-std=CL1.2 -triple amdgcn-amd-amdhsa -emit-llvm %s -o - | FileCheck -check-prefixes=CL12 %s3// RUN: %clang_cc1 -O0 -cl-std=CL2.0 -triple amdgcn-amd-amdhsa -emit-llvm %s -o - | FileCheck -check-prefixes=CL20 %s4 5// CL12-LABEL: define dso_local void @func1(6// CL12-SAME: ptr addrspace(5) noundef [[X:%.*]]) #[[ATTR0:[0-9]+]] {7// CL12-NEXT: [[ENTRY:.*:]]8// CL12-NEXT: [[X_ADDR:%.*]] = alloca ptr addrspace(5), align 4, addrspace(5)9// CL12-NEXT: store ptr addrspace(5) [[X]], ptr addrspace(5) [[X_ADDR]], align 410// CL12-NEXT: [[TMP0:%.*]] = load ptr addrspace(5), ptr addrspace(5) [[X_ADDR]], align 411// CL12-NEXT: store i32 1, ptr addrspace(5) [[TMP0]], align 412// CL12-NEXT: ret void13//14// CL20-LABEL: define dso_local void @func1(15// CL20-SAME: ptr noundef [[X:%.*]]) #[[ATTR0:[0-9]+]] {16// CL20-NEXT: [[ENTRY:.*:]]17// CL20-NEXT: [[X_ADDR:%.*]] = alloca ptr, align 8, addrspace(5)18// CL20-NEXT: [[X_ADDR_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[X_ADDR]] to ptr19// CL20-NEXT: store ptr [[X]], ptr [[X_ADDR_ASCAST]], align 820// CL20-NEXT: [[TMP0:%.*]] = load ptr, ptr [[X_ADDR_ASCAST]], align 821// CL20-NEXT: store i32 1, ptr [[TMP0]], align 422// CL20-NEXT: ret void23//24void func1(int *x) {25 *x = 1;26}27 28// CL12-LABEL: define dso_local void @func2(29// CL12-SAME: ) #[[ATTR0]] {30// CL12-NEXT: [[ENTRY:.*:]]31// CL12-NEXT: [[LV1:%.*]] = alloca i32, align 4, addrspace(5)32// CL12-NEXT: [[LV2:%.*]] = alloca i32, align 4, addrspace(5)33// CL12-NEXT: [[LA:%.*]] = alloca [100 x i32], align 4, addrspace(5)34// CL12-NEXT: [[LP1:%.*]] = alloca ptr addrspace(5), align 4, addrspace(5)35// CL12-NEXT: [[LP2:%.*]] = alloca ptr addrspace(5), align 4, addrspace(5)36// CL12-NEXT: [[LVC:%.*]] = alloca i32, align 4, addrspace(5)37// CL12-NEXT: store i32 1, ptr addrspace(5) [[LV1]], align 438// CL12-NEXT: store i32 2, ptr addrspace(5) [[LV2]], align 439// CL12-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [100 x i32], ptr addrspace(5) [[LA]], i64 0, i64 040// CL12-NEXT: store i32 3, ptr addrspace(5) [[ARRAYIDX]], align 441// CL12-NEXT: store ptr addrspace(5) [[LV1]], ptr addrspace(5) [[LP1]], align 442// CL12-NEXT: [[ARRAYDECAY:%.*]] = getelementptr inbounds [100 x i32], ptr addrspace(5) [[LA]], i64 0, i64 043// CL12-NEXT: store ptr addrspace(5) [[ARRAYDECAY]], ptr addrspace(5) [[LP2]], align 444// CL12-NEXT: call void @func1(ptr addrspace(5) noundef [[LV1]]) #[[ATTR2:[0-9]+]]45// CL12-NEXT: store i32 4, ptr addrspace(5) [[LVC]], align 446// CL12-NEXT: store i32 4, ptr addrspace(5) [[LV1]], align 447// CL12-NEXT: ret void48//49// CL20-LABEL: define dso_local void @func2(50// CL20-SAME: ) #[[ATTR0]] {51// CL20-NEXT: [[ENTRY:.*:]]52// CL20-NEXT: [[LV1:%.*]] = alloca i32, align 4, addrspace(5)53// CL20-NEXT: [[LV2:%.*]] = alloca i32, align 4, addrspace(5)54// CL20-NEXT: [[LA:%.*]] = alloca [100 x i32], align 4, addrspace(5)55// CL20-NEXT: [[LP1:%.*]] = alloca ptr, align 8, addrspace(5)56// CL20-NEXT: [[LP2:%.*]] = alloca ptr, align 8, addrspace(5)57// CL20-NEXT: [[LVC:%.*]] = alloca i32, align 4, addrspace(5)58// CL20-NEXT: store i32 1, ptr addrspace(5) [[LV1]], align 459// CL20-NEXT: store i32 2, ptr addrspace(5) [[LV2]], align 460// CL20-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [100 x i32], ptr addrspace(5) [[LA]], i64 0, i64 061// CL20-NEXT: store i32 3, ptr addrspace(5) [[ARRAYIDX]], align 462// CL20-NEXT: [[LV1_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[LV1]] to ptr63// CL20-NEXT: store ptr [[LV1_ASCAST]], ptr addrspace(5) [[LP1]], align 864// CL20-NEXT: [[ARRAYDECAY:%.*]] = getelementptr inbounds [100 x i32], ptr addrspace(5) [[LA]], i64 0, i64 065// CL20-NEXT: [[ARRAYDECAY_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[ARRAYDECAY]] to ptr66// CL20-NEXT: store ptr [[ARRAYDECAY_ASCAST]], ptr addrspace(5) [[LP2]], align 867// CL20-NEXT: [[LV1_ASCAST1:%.*]] = addrspacecast ptr addrspace(5) [[LV1]] to ptr68// CL20-NEXT: call void @func1(ptr noundef [[LV1_ASCAST1]]) #[[ATTR2:[0-9]+]]69// CL20-NEXT: store i32 4, ptr addrspace(5) [[LVC]], align 470// CL20-NEXT: store i32 4, ptr addrspace(5) [[LV1]], align 471// CL20-NEXT: ret void72//73void func2(void) {74 int lv1;75 lv1 = 1;76 int lv2 = 2;77 78 int la[100];79 la[0] = 3;80 81 int *lp1 = &lv1;82 83 int *lp2 = la;84 85 func1(&lv1);86 87 const int lvc = 4;88 lv1 = lvc;89}90 91// CL12-LABEL: define dso_local void @func3(92// CL12-SAME: ) #[[ATTR0]] {93// CL12-NEXT: [[ENTRY:.*:]]94// CL12-NEXT: [[A:%.*]] = alloca [16 x [1 x float]], align 4, addrspace(5)95// CL12-NEXT: call void @llvm.memset.p5.i64(ptr addrspace(5) align 4 [[A]], i8 0, i64 64, i1 false)96// CL12-NEXT: ret void97//98// CL20-LABEL: define dso_local void @func3(99// CL20-SAME: ) #[[ATTR0]] {100// CL20-NEXT: [[ENTRY:.*:]]101// CL20-NEXT: [[A:%.*]] = alloca [16 x [1 x float]], align 4, addrspace(5)102// CL20-NEXT: call void @llvm.memset.p5.i64(ptr addrspace(5) align 4 [[A]], i8 0, i64 64, i1 false)103// CL20-NEXT: ret void104//105void func3(void) {106 float a[16][1] = {{0.}};107}108 109// CL12-LABEL: define dso_local void @wrong_store_type_private_pointer_alloca(110// CL12-SAME: ) #[[ATTR0]] {111// CL12-NEXT: [[ENTRY:.*:]]112// CL12-NEXT: [[VAR:%.*]] = alloca i64, align 8, addrspace(5)113// CL12-NEXT: [[ALLOCA_ADDR:%.*]] = alloca ptr addrspace(5), align 4, addrspace(5)114// CL12-NEXT: store i64 5, ptr addrspace(5) [[VAR]], align 8115// CL12-NEXT: store ptr addrspace(5) [[VAR]], ptr addrspace(5) [[ALLOCA_ADDR]], align 4116// CL12-NEXT: [[TMP0:%.*]] = load ptr addrspace(5), ptr addrspace(5) [[ALLOCA_ADDR]], align 4117// CL12-NEXT: store i64 8, ptr addrspace(5) [[TMP0]], align 8118// CL12-NEXT: ret void119//120// CL20-LABEL: define dso_local void @wrong_store_type_private_pointer_alloca(121// CL20-SAME: ) #[[ATTR0]] {122// CL20-NEXT: [[ENTRY:.*:]]123// CL20-NEXT: [[VAR:%.*]] = alloca i64, align 8, addrspace(5)124// CL20-NEXT: [[ALLOCA_ADDR:%.*]] = alloca ptr addrspace(5), align 4, addrspace(5)125// CL20-NEXT: store i64 5, ptr addrspace(5) [[VAR]], align 8126// CL20-NEXT: store ptr addrspace(5) [[VAR]], ptr addrspace(5) [[ALLOCA_ADDR]], align 4127// CL20-NEXT: [[TMP0:%.*]] = load ptr addrspace(5), ptr addrspace(5) [[ALLOCA_ADDR]], align 4128// CL20-NEXT: store i64 8, ptr addrspace(5) [[TMP0]], align 8129// CL20-NEXT: ret void130//131void wrong_store_type_private_pointer_alloca() {132 long var = 5;133 134 // This needs to write an addrspace(5) pointer to the temporary alloca,135 // which should be allocated with the correct size.136 __private long *alloca_addr = &var;137 *alloca_addr = 8;138}139 140// CL12-LABEL: define dso_local void @wrong_store_type_private_pointer_as_generic_alloca(141// CL12-SAME: ) #[[ATTR0]] {142// CL12-NEXT: [[ENTRY:.*:]]143// CL12-NEXT: [[VAR:%.*]] = alloca i64, align 8, addrspace(5)144// CL12-NEXT: [[ALLOCA_ADDR_AS_GENERIC:%.*]] = alloca ptr addrspace(5), align 4, addrspace(5)145// CL12-NEXT: store i64 5, ptr addrspace(5) [[VAR]], align 8146// CL12-NEXT: store ptr addrspace(5) [[VAR]], ptr addrspace(5) [[ALLOCA_ADDR_AS_GENERIC]], align 4147// CL12-NEXT: [[TMP0:%.*]] = load ptr addrspace(5), ptr addrspace(5) [[ALLOCA_ADDR_AS_GENERIC]], align 4148// CL12-NEXT: store i64 9, ptr addrspace(5) [[TMP0]], align 8149// CL12-NEXT: ret void150//151// CL20-LABEL: define dso_local void @wrong_store_type_private_pointer_as_generic_alloca(152// CL20-SAME: ) #[[ATTR0]] {153// CL20-NEXT: [[ENTRY:.*:]]154// CL20-NEXT: [[VAR:%.*]] = alloca i64, align 8, addrspace(5)155// CL20-NEXT: [[ALLOCA_ADDR_AS_GENERIC:%.*]] = alloca ptr, align 8, addrspace(5)156// CL20-NEXT: store i64 5, ptr addrspace(5) [[VAR]], align 8157// CL20-NEXT: [[VAR_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[VAR]] to ptr158// CL20-NEXT: store ptr [[VAR_ASCAST]], ptr addrspace(5) [[ALLOCA_ADDR_AS_GENERIC]], align 8159// CL20-NEXT: [[TMP0:%.*]] = load ptr, ptr addrspace(5) [[ALLOCA_ADDR_AS_GENERIC]], align 8160// CL20-NEXT: store i64 9, ptr [[TMP0]], align 8161// CL20-NEXT: ret void162//163void wrong_store_type_private_pointer_as_generic_alloca() {164 long var = 5;165 166 // This needs to write an addrspace(0) pointer to the temporary alloca in167 // CL2.0, which should be allocated with the correct size.168 long *alloca_addr_as_generic = &var;169 *alloca_addr_as_generic = 9;170}171