1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py 2; RUN: opt -S -passes=iroutliner -ir-outlining-no-cost < %s | FileCheck %s 3 4; These functions are constructed slightly differently so that they require 5; different output blocks for the values used outside of the region. We are 6; checking that two output blocks are created with different values. 7 8define void @outline_outputs1() #0 { 9; CHECK-LABEL: @outline_outputs1( 10; CHECK-NEXT: entry: 11; CHECK-NEXT: [[DOTLOC:%.*]] = alloca i32, align 4 12; CHECK-NEXT: [[ADD_LOC:%.*]] = alloca i32, align 4 13; CHECK-NEXT: [[A:%.*]] = alloca i32, align 4 14; CHECK-NEXT: [[B:%.*]] = alloca i32, align 4 15; CHECK-NEXT: [[OUTPUT:%.*]] = alloca i32, align 4 16; CHECK-NEXT: [[RESULT:%.*]] = alloca i32, align 4 17; CHECK-NEXT: call void @llvm.lifetime.start.p0(i64 -1, ptr [[ADD_LOC]]) 18; CHECK-NEXT: call void @llvm.lifetime.start.p0(i64 -1, ptr [[DOTLOC]]) 19; CHECK-NEXT: call void @outlined_ir_func_0(ptr [[A]], ptr [[B]], ptr [[OUTPUT]], ptr [[ADD_LOC]], ptr [[DOTLOC]], i32 0) 20; CHECK-NEXT: [[ADD_RELOAD:%.*]] = load i32, ptr [[ADD_LOC]], align 4 21; CHECK-NEXT: [[DOTRELOAD:%.*]] = load i32, ptr [[DOTLOC]], align 4 22; CHECK-NEXT: call void @llvm.lifetime.end.p0(i64 -1, ptr [[ADD_LOC]]) 23; CHECK-NEXT: call void @llvm.lifetime.end.p0(i64 -1, ptr [[DOTLOC]]) 24; CHECK-NEXT: [[TMP0:%.*]] = load i32, ptr [[OUTPUT]], align 4 25; CHECK-NEXT: call void @outlined_ir_func_1(i32 [[DOTRELOAD]], i32 [[ADD_RELOAD]], ptr [[RESULT]]) 26; CHECK-NEXT: ret void 27; 28entry: 29 %a = alloca i32, align 4 30 %b = alloca i32, align 4 31 %output = alloca i32, align 4 32 %result = alloca i32, align 4 33 store i32 2, ptr %a, align 4 34 store i32 3, ptr %b, align 4 35 %0 = load i32, ptr %a, align 4 36 %1 = load i32, ptr %b, align 4 37 %add = add i32 %0, %1 38 %sub = sub i32 %0, %1 39 store i32 %add, ptr %output, align 4 40 %2 = load i32, ptr %output, align 4 41 %3 = load i32, ptr %output, align 4 42 %mul = mul i32 %2, %add 43 store i32 %mul, ptr %result, align 4 44 ret void 45} 46 47define void @outline_outputs2() #0 { 48; CHECK-LABEL: @outline_outputs2( 49; CHECK-NEXT: entry: 50; CHECK-NEXT: [[DOTLOC:%.*]] = alloca i32, align 4 51; CHECK-NEXT: [[SUB_LOC:%.*]] = alloca i32, align 4 52; CHECK-NEXT: [[A:%.*]] = alloca i32, align 4 53; CHECK-NEXT: [[B:%.*]] = alloca i32, align 4 54; CHECK-NEXT: [[OUTPUT:%.*]] = alloca i32, align 4 55; CHECK-NEXT: [[RESULT:%.*]] = alloca i32, align 4 56; CHECK-NEXT: call void @llvm.lifetime.start.p0(i64 -1, ptr [[SUB_LOC]]) 57; CHECK-NEXT: call void @llvm.lifetime.start.p0(i64 -1, ptr [[DOTLOC]]) 58; CHECK-NEXT: call void @outlined_ir_func_0(ptr [[A]], ptr [[B]], ptr [[OUTPUT]], ptr [[SUB_LOC]], ptr [[DOTLOC]], i32 1) 59; CHECK-NEXT: [[SUB_RELOAD:%.*]] = load i32, ptr [[SUB_LOC]], align 4 60; CHECK-NEXT: [[DOTRELOAD:%.*]] = load i32, ptr [[DOTLOC]], align 4 61; CHECK-NEXT: call void @llvm.lifetime.end.p0(i64 -1, ptr [[SUB_LOC]]) 62; CHECK-NEXT: call void @llvm.lifetime.end.p0(i64 -1, ptr [[DOTLOC]]) 63; CHECK-NEXT: call void @outlined_ir_func_1(i32 [[DOTRELOAD]], i32 [[SUB_RELOAD]], ptr [[RESULT]]) 64; CHECK-NEXT: ret void 65; 66entry: 67 %a = alloca i32, align 4 68 %b = alloca i32, align 4 69 %output = alloca i32, align 4 70 %result = alloca i32, align 4 71 store i32 2, ptr %a, align 4 72 store i32 3, ptr %b, align 4 73 %0 = load i32, ptr %a, align 4 74 %1 = load i32, ptr %b, align 4 75 %add = add i32 %0, %1 76 %sub = sub i32 %0, %1 77 store i32 %add, ptr %output, align 4 78 %2 = load i32, ptr %output, align 4 79 %mul = mul i32 %2, %sub 80 store i32 %mul, ptr %result, align 4 81 ret void 82} 83 84; CHECK: define internal void @outlined_ir_func_0(ptr [[ARG0:%.*]], ptr [[ARG1:%.*]], ptr [[ARG2:%.*]], ptr [[ARG3:%.*]], ptr [[ARG4:%.*]], i32 [[ARG5:%.*]]) #1 { 85; CHECK: entry_to_outline: 86; CHECK-NEXT: store i32 2, ptr [[ARG0]], align 4 87; CHECK-NEXT: store i32 3, ptr [[ARG1]], align 4 88; CHECK-NEXT: [[TMP0:%.*]] = load i32, ptr [[ARG0]], align 4 89; CHECK-NEXT: [[TMP1:%.*]] = load i32, ptr [[ARG1]], align 4 90; CHECK-NEXT: [[ADD:%.*]] = add i32 [[TMP0]], [[TMP1]] 91; CHECK-NEXT: [[SUB:%.*]] = sub i32 [[TMP0]], [[TMP1]] 92; CHECK-NEXT: store i32 [[ADD]], ptr [[ARG2]], align 4 93; CHECK-NEXT: [[TMP2:%.*]] = load i32, ptr [[ARG2]], align 4 94 95; CHECK: _after_outline.exitStub: 96; CHECK-NEXT: switch i32 [[ARG5]], label [[BLOCK:%.*]] [ 97; CHECK-NEXT: i32 0, label %[[BLOCK_0:.*]] 98; CHECK-NEXT: i32 1, label %[[BLOCK_1:.*]] 99 100; CHECK: [[BLOCK_0]]: 101; CHECK-NEXT: store i32 [[ADD]], ptr [[ARG3]], align 4 102; CHECK-NEXT: store i32 [[TMP2]], ptr [[ARG4]], align 4 103 104; CHECK: [[BLOCK_1]]: 105; CHECK-NEXT: store i32 [[SUB]], ptr [[ARG3]], align 4 106; CHECK-NEXT: store i32 [[TMP2]], ptr [[ARG4]], align 4 107