1; Test basic memory profiler instrumentation. 2; 3; RUN: opt < %s -passes='function(memprof),memprof-module' -S | FileCheck --check-prefixes=CHECK,CHECK-S3 %s 4; RUN: opt < %s -passes='function(memprof),memprof-module' -memprof-mapping-scale=5 -S | FileCheck --check-prefixes=CHECK,CHECK-S5 %s 5 6; We need the requires since both memprof and memprof-module require reading module level metadata which is done once by the memprof-globals-md analysis 7; RUN: opt < %s -passes='function(memprof),module(memprof-module)' -S | FileCheck --check-prefixes=CHECK,CHECK-S3 %s 8; RUN: opt < %s -passes='function(memprof),module(memprof-module)' -memprof-mapping-scale=5 -S | FileCheck --check-prefixes=CHECK,CHECK-S5 %s 9 10target datalayout = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v64:64:64-v128:128:128-a0:0:64-s0:64:64-f80:128:128-n8:16:32:64" 11target triple = "x86_64-unknown-linux-gnu" 12 13; CHECK: @llvm.used = appending global [1 x ptr] [ptr @memprof.module_ctor] 14; CHECK: @llvm.global_ctors = appending global [1 x { i32, ptr, ptr }] [{ i32, ptr, ptr } { i32 1, ptr @memprof.module_ctor, ptr null }] 15 16define i32 @test_load(ptr %a) { 17entry: 18 %tmp1 = load i32, ptr %a, align 4 19 ret i32 %tmp1 20} 21; CHECK-LABEL: @test_load 22; CHECK: %[[SHADOW_OFFSET:[^ ]*]] = load i64, ptr @__memprof_shadow_memory_dynamic_address 23; CHECK-NEXT: %[[LOAD_ADDR:[^ ]*]] = ptrtoint ptr %a to i64 24; CHECK-NEXT: %[[MASKED_ADDR:[^ ]*]] = and i64 %[[LOAD_ADDR]], -64 25; CHECK-S3-NEXT: %[[SHIFTED_ADDR:[^ ]*]] = lshr i64 %[[MASKED_ADDR]], 3 26; CHECK-S5-NEXT: %[[SHIFTED_ADDR:[^ ]*]] = lshr i64 %[[MASKED_ADDR]], 5 27; CHECK-NEXT: add i64 %[[SHIFTED_ADDR]], %[[SHADOW_OFFSET]] 28; CHECK-NEXT: %[[LOAD_SHADOW_PTR:[^ ]*]] = inttoptr 29; CHECK-NEXT: %[[LOAD_SHADOW:[^ ]*]] = load i64, ptr %[[LOAD_SHADOW_PTR]] 30; CHECK-NEXT: %[[NEW_SHADOW:[^ ]*]] = add i64 %[[LOAD_SHADOW]], 1 31; CHECK-NEXT: store i64 %[[NEW_SHADOW]], ptr %[[LOAD_SHADOW_PTR]] 32; The actual load. 33; CHECK-NEXT: %tmp1 = load i32, ptr %a 34; CHECK-NEXT: ret i32 %tmp1 35 36define void @test_store(ptr %a) { 37entry: 38 store i32 42, ptr %a, align 4 39 ret void 40} 41; CHECK-LABEL: @test_store 42; CHECK: %[[SHADOW_OFFSET:[^ ]*]] = load i64, ptr @__memprof_shadow_memory_dynamic_address 43; CHECK-NEXT: %[[STORE_ADDR:[^ ]*]] = ptrtoint ptr %a to i64 44; CHECK-NEXT: %[[MASKED_ADDR:[^ ]*]] = and i64 %[[STORE_ADDR]], -64 45; CHECK-S3-NEXT: %[[SHIFTED_ADDR:[^ ]*]] = lshr i64 %[[MASKED_ADDR]], 3 46; CHECK-S5-NEXT: %[[SHIFTED_ADDR:[^ ]*]] = lshr i64 %[[MASKED_ADDR]], 5 47; CHECK-NEXT: add i64 %[[SHIFTED_ADDR]], %[[SHADOW_OFFSET]] 48; CHECK-NEXT: %[[STORE_SHADOW_PTR:[^ ]*]] = inttoptr 49; CHECK-NEXT: %[[STORE_SHADOW:[^ ]*]] = load i64, ptr %[[STORE_SHADOW_PTR]] 50; CHECK-NEXT: %[[NEW_SHADOW:[^ ]*]] = add i64 %[[STORE_SHADOW]], 1 51; CHECK-NEXT: store i64 %[[NEW_SHADOW]], ptr %[[STORE_SHADOW_PTR]] 52; The actual store. 53; CHECK-NEXT: store i32 42, ptr %a 54; CHECK-NEXT: ret void 55 56define void @FP80Test(ptr nocapture %a) nounwind uwtable { 57entry: 58 store x86_fp80 0xK3FFF8000000000000000, ptr %a, align 16 59 ret void 60} 61; CHECK-LABEL: @FP80Test 62; Exactly one shadow update for store access. 63; CHECK-NOT: store i64 64; CHECK: %[[NEW_ST_SHADOW:[^ ]*]] = add i64 %{{.*}}, 1 65; CHECK-NEXT: store i64 %[[NEW_ST_SHADOW]] 66; CHECK-NOT: store i64 67; The actual store. 68; CHECK: store x86_fp80 0xK3FFF8000000000000000, ptr %a 69; CHECK: ret void 70 71define void @i40test(ptr %a, ptr %b) nounwind uwtable { 72entry: 73 %t = load i40, ptr %a 74 store i40 %t, ptr %b, align 8 75 ret void 76} 77; CHECK-LABEL: @i40test 78; Exactly one shadow update for load access. 79; CHECK-NOT: store i64 80; CHECK: %[[NEW_LD_SHADOW:[^ ]*]] = add i64 %{{.*}}, 1 81; CHECK-NEXT: store i64 %[[NEW_LD_SHADOW]] 82; CHECK-NOT: store i64 83; The actual load. 84; CHECK: %t = load i40, ptr %a 85; Exactly one shadow update for store access. 86; CHECK-NOT: store i64 87; CHECK: %[[NEW_ST_SHADOW:[^ ]*]] = add i64 %{{.*}}, 1 88; CHECK-NEXT: store i64 %[[NEW_ST_SHADOW]] 89; CHECK-NOT: store i64 90; The actual store. 91; CHECK: store i40 %t, ptr %b 92; CHECK: ret void 93 94define void @i64test_align1(ptr %b) nounwind uwtable { 95 entry: 96 store i64 0, ptr %b, align 1 97 ret void 98} 99; CHECK-LABEL: @i64test 100; Exactly one shadow update for store access. 101; CHECK-NOT: store i64 102; CHECK: %[[NEW_ST_SHADOW:[^ ]*]] = add i64 %{{.*}}, 1 103; CHECK-NEXT: store i64 %[[NEW_ST_SHADOW]] 104; CHECK-NOT: store i64 105; The actual store. 106; CHECK: store i64 0, ptr %b 107; CHECK: ret void 108 109define void @i80test(ptr %a, ptr %b) nounwind uwtable { 110 entry: 111 %t = load i80, ptr %a 112 store i80 %t, ptr %b, align 8 113 ret void 114} 115; CHECK-LABEL: i80test 116; Exactly one shadow update for load access. 117; CHECK-NOT: store i64 118; CHECK: %[[NEW_LD_SHADOW:[^ ]*]] = add i64 %{{.*}}, 1 119; CHECK-NEXT: store i64 %[[NEW_LD_SHADOW]] 120; CHECK-NOT: store i64 121; The actual load. 122; CHECK: %t = load i80, ptr %a 123; Exactly one shadow update for store access. 124; CHECK-NOT: store i64 125; CHECK: %[[NEW_ST_SHADOW:[^ ]*]] = add i64 %{{.*}}, 1 126; CHECK-NEXT: store i64 %[[NEW_ST_SHADOW]] 127; CHECK-NOT: store i64 128; The actual store. 129; CHECK: store i80 %t, ptr %b 130; CHECK: ret void 131 132; memprof should not instrument functions with available_externally linkage. 133define available_externally i32 @f_available_externally(ptr %a) { 134entry: 135 %tmp1 = load i32, ptr %a 136 ret i32 %tmp1 137} 138; CHECK-LABEL: @f_available_externally 139; CHECK-NOT: __memprof_shadow_memory_dynamic_address 140; CHECK: ret i32 141 142declare void @llvm.memset.p0.i64(ptr nocapture, i8, i64, i1) nounwind 143declare void @llvm.memmove.p0.p0.i64(ptr nocapture, ptr nocapture readonly, i64, i1) nounwind 144declare void @llvm.memcpy.p0.p0.i64(ptr nocapture, ptr nocapture readonly, i64, i1) nounwind 145 146define void @memintr_test(ptr %a, ptr %b) nounwind uwtable { 147 entry: 148 tail call void @llvm.memset.p0.i64(ptr %a, i8 0, i64 100, i1 false) 149 tail call void @llvm.memmove.p0.p0.i64(ptr %a, ptr %b, i64 100, i1 false) 150 tail call void @llvm.memcpy.p0.p0.i64(ptr %a, ptr %b, i64 100, i1 false) 151 ret void 152} 153 154; CHECK-LABEL: memintr_test 155; CHECK: __memprof_memset 156; CHECK: __memprof_memmove 157; CHECK: __memprof_memcpy 158; CHECK: ret void 159 160declare void @llvm.memset.element.unordered.atomic.p0.i64(ptr nocapture writeonly, i8, i64, i32) nounwind 161declare void @llvm.memmove.element.unordered.atomic.p0.p0.i64(ptr nocapture writeonly, ptr nocapture readonly, i64, i32) nounwind 162declare void @llvm.memcpy.element.unordered.atomic.p0.p0.i64(ptr nocapture writeonly, ptr nocapture readonly, i64, i32) nounwind 163 164define void @memintr_element_atomic_test(ptr %a, ptr %b) nounwind uwtable { 165 ; This is a canary test to make sure that these don't get lowered into calls that don't 166 ; have the element-atomic property. Eventually, memprof will have to be enhanced to lower 167 ; these properly. 168 ; CHECK-LABEL: memintr_element_atomic_test 169 ; CHECK: tail call void @llvm.memset.element.unordered.atomic.p0.i64(ptr align 1 %a, i8 0, i64 100, i32 1) 170 ; CHECK: tail call void @llvm.memmove.element.unordered.atomic.p0.p0.i64(ptr align 1 %a, ptr align 1 %b, i64 100, i32 1) 171 ; CHECK: tail call void @llvm.memcpy.element.unordered.atomic.p0.p0.i64(ptr align 1 %a, ptr align 1 %b, i64 100, i32 1) 172 ; CHECK: ret void 173 tail call void @llvm.memset.element.unordered.atomic.p0.i64(ptr align 1 %a, i8 0, i64 100, i32 1) 174 tail call void @llvm.memmove.element.unordered.atomic.p0.p0.i64(ptr align 1 %a, ptr align 1 %b, i64 100, i32 1) 175 tail call void @llvm.memcpy.element.unordered.atomic.p0.p0.i64(ptr align 1 %a, ptr align 1 %b, i64 100, i32 1) 176 ret void 177} 178 179 180; CHECK: define internal void @memprof.module_ctor() 181; CHECK: call void @__memprof_init() 182