182 lines · plain
1; Test basic memory profiler instrumentation.2;3; RUN: opt < %s -passes='function(memprof),memprof-module' -S | FileCheck --check-prefixes=CHECK,CHECK-S3 %s4; RUN: opt < %s -passes='function(memprof),memprof-module' -memprof-mapping-scale=5 -S | FileCheck --check-prefixes=CHECK,CHECK-S5 %s5 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 analysis7; RUN: opt < %s -passes='function(memprof),module(memprof-module)' -S | FileCheck --check-prefixes=CHECK,CHECK-S3 %s8; RUN: opt < %s -passes='function(memprof),module(memprof-module)' -memprof-mapping-scale=5 -S | FileCheck --check-prefixes=CHECK,CHECK-S5 %s9 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 419 ret i32 %tmp120}21; CHECK-LABEL: @test_load22; CHECK: %[[SHADOW_OFFSET:[^ ]*]] = load i64, ptr @__memprof_shadow_memory_dynamic_address23; CHECK-NEXT: %[[LOAD_ADDR:[^ ]*]] = ptrtoint ptr %a to i6424; CHECK-NEXT: %[[MASKED_ADDR:[^ ]*]] = and i64 %[[LOAD_ADDR]], -6425; CHECK-S3-NEXT: %[[SHIFTED_ADDR:[^ ]*]] = lshr i64 %[[MASKED_ADDR]], 326; CHECK-S5-NEXT: %[[SHIFTED_ADDR:[^ ]*]] = lshr i64 %[[MASKED_ADDR]], 527; CHECK-NEXT: add i64 %[[SHIFTED_ADDR]], %[[SHADOW_OFFSET]]28; CHECK-NEXT: %[[LOAD_SHADOW_PTR:[^ ]*]] = inttoptr29; CHECK-NEXT: %[[LOAD_SHADOW:[^ ]*]] = load i64, ptr %[[LOAD_SHADOW_PTR]]30; CHECK-NEXT: %[[NEW_SHADOW:[^ ]*]] = add i64 %[[LOAD_SHADOW]], 131; CHECK-NEXT: store i64 %[[NEW_SHADOW]], ptr %[[LOAD_SHADOW_PTR]]32; The actual load.33; CHECK-NEXT: %tmp1 = load i32, ptr %a34; CHECK-NEXT: ret i32 %tmp135 36define void @test_store(ptr %a) {37entry:38 store i32 42, ptr %a, align 439 ret void40}41; CHECK-LABEL: @test_store42; CHECK: %[[SHADOW_OFFSET:[^ ]*]] = load i64, ptr @__memprof_shadow_memory_dynamic_address43; CHECK-NEXT: %[[STORE_ADDR:[^ ]*]] = ptrtoint ptr %a to i6444; CHECK-NEXT: %[[MASKED_ADDR:[^ ]*]] = and i64 %[[STORE_ADDR]], -6445; CHECK-S3-NEXT: %[[SHIFTED_ADDR:[^ ]*]] = lshr i64 %[[MASKED_ADDR]], 346; CHECK-S5-NEXT: %[[SHIFTED_ADDR:[^ ]*]] = lshr i64 %[[MASKED_ADDR]], 547; CHECK-NEXT: add i64 %[[SHIFTED_ADDR]], %[[SHADOW_OFFSET]]48; CHECK-NEXT: %[[STORE_SHADOW_PTR:[^ ]*]] = inttoptr49; CHECK-NEXT: %[[STORE_SHADOW:[^ ]*]] = load i64, ptr %[[STORE_SHADOW_PTR]]50; CHECK-NEXT: %[[NEW_SHADOW:[^ ]*]] = add i64 %[[STORE_SHADOW]], 151; CHECK-NEXT: store i64 %[[NEW_SHADOW]], ptr %[[STORE_SHADOW_PTR]]52; The actual store.53; CHECK-NEXT: store i32 42, ptr %a54; CHECK-NEXT: ret void55 56define void @FP80Test(ptr nocapture %a) nounwind uwtable {57entry:58 store x86_fp80 0xK3FFF8000000000000000, ptr %a, align 1659 ret void60}61; CHECK-LABEL: @FP80Test62; Exactly one shadow update for store access.63; CHECK-NOT: store i6464; CHECK: %[[NEW_ST_SHADOW:[^ ]*]] = add i64 %{{.*}}, 165; CHECK-NEXT: store i64 %[[NEW_ST_SHADOW]]66; CHECK-NOT: store i6467; The actual store.68; CHECK: store x86_fp80 0xK3FFF8000000000000000, ptr %a69; CHECK: ret void70 71define void @i40test(ptr %a, ptr %b) nounwind uwtable {72entry:73 %t = load i40, ptr %a74 store i40 %t, ptr %b, align 875 ret void76}77; CHECK-LABEL: @i40test78; Exactly one shadow update for load access.79; CHECK-NOT: store i6480; CHECK: %[[NEW_LD_SHADOW:[^ ]*]] = add i64 %{{.*}}, 181; CHECK-NEXT: store i64 %[[NEW_LD_SHADOW]]82; CHECK-NOT: store i6483; The actual load.84; CHECK: %t = load i40, ptr %a85; Exactly one shadow update for store access.86; CHECK-NOT: store i6487; CHECK: %[[NEW_ST_SHADOW:[^ ]*]] = add i64 %{{.*}}, 188; CHECK-NEXT: store i64 %[[NEW_ST_SHADOW]]89; CHECK-NOT: store i6490; The actual store.91; CHECK: store i40 %t, ptr %b92; CHECK: ret void93 94define void @i64test_align1(ptr %b) nounwind uwtable {95 entry:96 store i64 0, ptr %b, align 197 ret void98}99; CHECK-LABEL: @i64test100; Exactly one shadow update for store access.101; CHECK-NOT: store i64102; CHECK: %[[NEW_ST_SHADOW:[^ ]*]] = add i64 %{{.*}}, 1103; CHECK-NEXT: store i64 %[[NEW_ST_SHADOW]]104; CHECK-NOT: store i64105; The actual store.106; CHECK: store i64 0, ptr %b107; CHECK: ret void108 109define void @i80test(ptr %a, ptr %b) nounwind uwtable {110 entry:111 %t = load i80, ptr %a112 store i80 %t, ptr %b, align 8113 ret void114}115; CHECK-LABEL: i80test116; Exactly one shadow update for load access.117; CHECK-NOT: store i64118; CHECK: %[[NEW_LD_SHADOW:[^ ]*]] = add i64 %{{.*}}, 1119; CHECK-NEXT: store i64 %[[NEW_LD_SHADOW]]120; CHECK-NOT: store i64121; The actual load.122; CHECK: %t = load i80, ptr %a123; Exactly one shadow update for store access.124; CHECK-NOT: store i64125; CHECK: %[[NEW_ST_SHADOW:[^ ]*]] = add i64 %{{.*}}, 1126; CHECK-NEXT: store i64 %[[NEW_ST_SHADOW]]127; CHECK-NOT: store i64128; The actual store.129; CHECK: store i80 %t, ptr %b130; CHECK: ret void131 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 %a136 ret i32 %tmp1137}138; CHECK-LABEL: @f_available_externally139; CHECK-NOT: __memprof_shadow_memory_dynamic_address140; CHECK: ret i32141 142declare void @llvm.memset.p0.i64(ptr nocapture, i8, i64, i1) nounwind143declare void @llvm.memmove.p0.p0.i64(ptr nocapture, ptr nocapture readonly, i64, i1) nounwind144declare void @llvm.memcpy.p0.p0.i64(ptr nocapture, ptr nocapture readonly, i64, i1) nounwind145 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 void152}153 154; CHECK-LABEL: memintr_test155; CHECK: __memprof_memset156; CHECK: __memprof_memmove157; CHECK: __memprof_memcpy158; CHECK: ret void159 160declare void @llvm.memset.element.unordered.atomic.p0.i64(ptr nocapture writeonly, i8, i64, i32) nounwind161declare void @llvm.memmove.element.unordered.atomic.p0.p0.i64(ptr nocapture writeonly, ptr nocapture readonly, i64, i32) nounwind162declare void @llvm.memcpy.element.unordered.atomic.p0.p0.i64(ptr nocapture writeonly, ptr nocapture readonly, i64, i32) nounwind163 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't166 ; have the element-atomic property. Eventually, memprof will have to be enhanced to lower167 ; these properly.168 ; CHECK-LABEL: memintr_element_atomic_test169 ; 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 void173 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 void177}178 179 180; CHECK: define internal void @memprof.module_ctor()181; CHECK: call void @__memprof_init()182