182 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py2; RUN: opt < %s -passes=memcpyopt -S -verify-memoryssa | FileCheck %s3 4; Test whether memcpy-memcpy dependence is optimized across5; basic blocks (conditional branches and invokes).6; TODO: This is not supported yet.7 8%struct.s = type { i32, i32 }9 10@s_foo = private unnamed_addr constant %struct.s { i32 1, i32 2 }, align 411@s_baz = private unnamed_addr constant %struct.s { i32 1, i32 2 }, align 412@i = external constant ptr13 14declare void @qux()15declare void @llvm.memcpy.p0.p0.i64(ptr nocapture writeonly, ptr nocapture readonly, i64, i1)16declare void @__cxa_throw(ptr, ptr, ptr)17declare i32 @__gxx_personality_v0(...)18declare ptr @__cxa_begin_catch(ptr)19 20; A simple partial redundancy. Test that the second memcpy is optimized21; to copy directly from the original source rather than from the temporary.22 23define void @wobble(ptr noalias %dst, ptr %src, i1 %some_condition) {24; CHECK-LABEL: @wobble(25; CHECK-NEXT: bb:26; CHECK-NEXT: [[TEMP:%.*]] = alloca i8, i32 64, align 127; CHECK-NEXT: call void @llvm.memcpy.p0.p0.i64(ptr align 8 [[TEMP]], ptr nonnull align 8 [[SRC:%.*]], i64 64, i1 false)28; CHECK-NEXT: br i1 [[SOME_CONDITION:%.*]], label [[MORE:%.*]], label [[OUT:%.*]]29; CHECK: out:30; CHECK-NEXT: call void @qux()31; CHECK-NEXT: unreachable32; CHECK: more:33; CHECK-NEXT: call void @llvm.memcpy.p0.p0.i64(ptr align 8 [[DST:%.*]], ptr align 8 [[SRC]], i64 64, i1 false)34; CHECK-NEXT: ret void35;36bb:37 %temp = alloca i8, i32 6438 call void @llvm.memcpy.p0.p0.i64(ptr align 8 %temp, ptr nonnull align 8%src, i64 64, i1 false)39 br i1 %some_condition, label %more, label %out40 41out:42 call void @qux()43 unreachable44 45more:46 call void @llvm.memcpy.p0.p0.i64(ptr align 8 %dst, ptr align 8 %temp, i64 64, i1 false)47 ret void48}49 50; A CFG triangle with a partial redundancy targeting an alloca. Test that the51; memcpy inside the triangle is optimized to copy directly from the original52; source rather than from the temporary.53 54define i32 @foo(i1 %t3) {55; CHECK-LABEL: @foo(56; CHECK-NEXT: bb:57; CHECK-NEXT: [[S:%.*]] = alloca [[STRUCT_S:%.*]], align 458; CHECK-NEXT: [[T:%.*]] = alloca [[STRUCT_S]], align 459; CHECK-NEXT: call void @llvm.memcpy.p0.p0.i64(ptr align 4 [[S]], ptr align 4 @s_foo, i64 8, i1 false)60; CHECK-NEXT: br i1 [[T3:%.*]], label [[BB4:%.*]], label [[BB7:%.*]]61; CHECK: bb4:62; CHECK-NEXT: call void @llvm.memcpy.p0.p0.i64(ptr align 4 [[T]], ptr align 4 @s_foo, i64 8, i1 false)63; CHECK-NEXT: br label [[BB7]]64; CHECK: bb7:65; CHECK-NEXT: [[T9:%.*]] = load i32, ptr [[T]], align 466; CHECK-NEXT: [[T10:%.*]] = getelementptr [[STRUCT_S]], ptr [[T]], i32 0, i32 167; CHECK-NEXT: [[T11:%.*]] = load i32, ptr [[T10]], align 468; CHECK-NEXT: [[T12:%.*]] = add i32 [[T9]], [[T11]]69; CHECK-NEXT: ret i32 [[T12]]70;71bb:72 %s = alloca %struct.s, align 473 %t = alloca %struct.s, align 474 call void @llvm.memcpy.p0.p0.i64(ptr align 4 %s, ptr align 4 @s_foo, i64 8, i1 false)75 br i1 %t3, label %bb4, label %bb776 77bb4: ; preds = %bb78 call void @llvm.memcpy.p0.p0.i64(ptr align 4 %t, ptr align 4 %s, i64 8, i1 false)79 br label %bb780 81bb7: ; preds = %bb4, %bb82 %t9 = load i32, ptr %t, align 483 %t10 = getelementptr %struct.s, ptr %t, i32 0, i32 184 %t11 = load i32, ptr %t10, align 485 %t12 = add i32 %t9, %t1186 ret i32 %t1287}88 89; A CFG diamond with an invoke on one side, and a partially redundant memcpy90; into an alloca on the other. Test that the memcpy inside the diamond is91; optimized to copy ; directly from the original source rather than from the92; temporary. This more complex test represents a relatively common usage93; pattern.94 95define i32 @baz(i1 %t5) personality ptr @__gxx_personality_v0 {96; CHECK-LABEL: @baz(97; CHECK-NEXT: bb:98; CHECK-NEXT: [[S:%.*]] = alloca [[STRUCT_S:%.*]], align 499; CHECK-NEXT: [[T:%.*]] = alloca [[STRUCT_S]], align 4100; CHECK-NEXT: call void @llvm.memcpy.p0.p0.i64(ptr align 4 [[S]], ptr align 4 @s_baz, i64 8, i1 false)101; CHECK-NEXT: br i1 [[T5:%.*]], label [[BB6:%.*]], label [[BB22:%.*]]102; CHECK: bb6:103; CHECK-NEXT: invoke void @__cxa_throw(ptr null, ptr @i, ptr null)104; CHECK-NEXT: to label [[BB25:%.*]] unwind label [[BB9:%.*]]105; CHECK: bb9:106; CHECK-NEXT: [[T10:%.*]] = landingpad { ptr, i32 }107; CHECK-NEXT: catch ptr null108; CHECK-NEXT: br label [[BB13:%.*]]109; CHECK: bb13:110; CHECK-NEXT: [[T15:%.*]] = call ptr @__cxa_begin_catch(ptr null)111; CHECK-NEXT: br label [[BB23:%.*]]112; CHECK: bb22:113; CHECK-NEXT: call void @llvm.memcpy.p0.p0.i64(ptr align 4 [[T]], ptr align 4 @s_baz, i64 8, i1 false)114; CHECK-NEXT: br label [[BB23]]115; CHECK: bb23:116; CHECK-NEXT: [[T18:%.*]] = load i32, ptr [[T]], align 4117; CHECK-NEXT: [[T19:%.*]] = getelementptr inbounds [[STRUCT_S]], ptr [[T]], i32 0, i32 1118; CHECK-NEXT: [[T20:%.*]] = load i32, ptr [[T19]], align 4119; CHECK-NEXT: [[T21:%.*]] = add nsw i32 [[T18]], [[T20]]120; CHECK-NEXT: ret i32 [[T21]]121; CHECK: bb25:122; CHECK-NEXT: unreachable123;124bb:125 %s = alloca %struct.s, align 4126 %t = alloca %struct.s, align 4127 call void @llvm.memcpy.p0.p0.i64(ptr align 4 %s, ptr align 4 @s_baz, i64 8, i1 false)128 br i1 %t5, label %bb6, label %bb22129 130bb6: ; preds = %bb131 invoke void @__cxa_throw(ptr null, ptr @i, ptr null)132 to label %bb25 unwind label %bb9133 134bb9: ; preds = %bb6135 %t10 = landingpad { ptr, i32 }136 catch ptr null137 br label %bb13138 139bb13: ; preds = %bb9140 %t15 = call ptr @__cxa_begin_catch(ptr null)141 br label %bb23142 143bb22: ; preds = %bb144 call void @llvm.memcpy.p0.p0.i64(ptr align 4 %t, ptr align 4 %s, i64 8, i1 false)145 br label %bb23146 147bb23: ; preds = %bb22, %bb13148 %t18 = load i32, ptr %t, align 4149 %t19 = getelementptr inbounds %struct.s, ptr %t, i32 0, i32 1150 %t20 = load i32, ptr %t19, align 4151 %t21 = add nsw i32 %t18, %t20152 ret i32 %t21153 154bb25: ; preds = %bb6155 unreachable156}157 158define void @memphi_with_unrelated_clobber(i1 %cond, ptr %arg, ptr noalias %a, ptr noalias %b, ptr noalias %c) {159; CHECK-LABEL: @memphi_with_unrelated_clobber(160; CHECK-NEXT: entry:161; CHECK-NEXT: call void @llvm.memcpy.p0.p0.i64(ptr [[A:%.*]], ptr [[B:%.*]], i64 16, i1 false)162; CHECK-NEXT: br i1 [[COND:%.*]], label [[THEN:%.*]], label [[EXIT:%.*]]163; CHECK: then:164; CHECK-NEXT: store i64 0, ptr [[ARG:%.*]], align 4165; CHECK-NEXT: br label [[EXIT]]166; CHECK: exit:167; CHECK-NEXT: call void @llvm.memcpy.p0.p0.i64(ptr [[C:%.*]], ptr [[B]], i64 16, i1 false)168; CHECK-NEXT: ret void169;170entry:171 call void @llvm.memcpy.p0.p0.i64(ptr %a, ptr %b, i64 16, i1 false)172 br i1 %cond, label %then, label %exit173 174then:175 store i64 0, ptr %arg176 br label %exit177 178exit:179 call void @llvm.memcpy.p0.p0.i64(ptr %c, ptr %a, i64 16, i1 false)180 ret void181}182