188 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py2; RUN: opt -S -passes=verify,iroutliner < %s | FileCheck %s3; RUN: opt -S -passes=verify,iroutliner -ir-outlining-no-cost < %s | FileCheck %s -check-prefix=NOCOST4 5; This test checks that we have different results from when the cost model6; is on versus when it is off. That is, if the number of instructions needed to7; handle the arguments is greater than the number of instructions being added,8; we do not outline.9 10define void @function1() #0 {11; CHECK-LABEL: @function1(12; CHECK-NEXT: entry:13; CHECK-NEXT: [[A:%.*]] = alloca i32, align 414; CHECK-NEXT: [[B:%.*]] = alloca i32, align 415; CHECK-NEXT: call void @outlined_ir_func_0(ptr [[A]], ptr [[B]])16; CHECK-NEXT: ret void17;18; NOCOST-LABEL: @function1(19; NOCOST-NEXT: entry:20; NOCOST-NEXT: [[A:%.*]] = alloca i32, align 421; NOCOST-NEXT: [[B:%.*]] = alloca i32, align 422; NOCOST-NEXT: call void @outlined_ir_func_0(ptr [[A]], ptr [[B]])23; NOCOST-NEXT: ret void24;25entry:26 %a = alloca i32, align 427 %b = alloca i32, align 428 %0 = load i32, ptr %a, align 429 %1 = load i32, ptr %b, align 430 %add = add i32 %0, %131 %mul = mul i32 %0, %132 %sub = sub i32 %0, %133 %div = sdiv i32 %0, %134 %add1 = add i32 %0, %135 %mul1 = mul i32 %0, %136 %sub1 = sub i32 %0, %137 %div1 = sdiv i32 %0, %138 %add2 = add i32 %0, %139 %mul2 = mul i32 %0, %140 %sub2 = sub i32 %0, %141 %div2 = sdiv i32 %0, %142 ret void43}44 45define void @function2() #0 {46; CHECK-LABEL: @function2(47; CHECK-NEXT: entry:48; CHECK-NEXT: [[A:%.*]] = alloca i32, align 449; CHECK-NEXT: [[B:%.*]] = alloca i32, align 450; CHECK-NEXT: call void @outlined_ir_func_0(ptr [[A]], ptr [[B]])51; CHECK-NEXT: ret void52;53; NOCOST-LABEL: @function2(54; NOCOST-NEXT: entry:55; NOCOST-NEXT: [[A:%.*]] = alloca i32, align 456; NOCOST-NEXT: [[B:%.*]] = alloca i32, align 457; NOCOST-NEXT: call void @outlined_ir_func_0(ptr [[A]], ptr [[B]])58; NOCOST-NEXT: ret void59;60entry:61 %a = alloca i32, align 462 %b = alloca i32, align 463 %0 = load i32, ptr %a, align 464 %1 = load i32, ptr %b, align 465 %add = add i32 %0, %166 %mul = mul i32 %0, %167 %sub = sub i32 %0, %168 %div = sdiv i32 %0, %169 %add1 = add i32 %0, %170 %mul1 = mul i32 %0, %171 %sub1 = sub i32 %0, %172 %div1 = sdiv i32 %0, %173 %add2 = add i32 %0, %174 %mul2 = mul i32 %0, %175 %sub2 = sub i32 %0, %176 %div2 = sdiv i32 %0, %177 ret void78}79 80define void @function3() #0 {81; CHECK-LABEL: @function3(82; CHECK-NEXT: entry:83; CHECK-NEXT: [[A:%.*]] = alloca i32, align 484; CHECK-NEXT: [[B:%.*]] = alloca i32, align 485; CHECK-NEXT: [[OUTPUT:%.*]] = alloca i32, align 486; CHECK-NEXT: [[RESULT:%.*]] = alloca i32, align 487; CHECK-NEXT: store i32 2, ptr [[A]], align 488; CHECK-NEXT: store i32 3, ptr [[B]], align 489; CHECK-NEXT: [[TMP0:%.*]] = load i32, ptr [[A]], align 490; CHECK-NEXT: [[TMP1:%.*]] = load i32, ptr [[B]], align 491; CHECK-NEXT: [[ADD:%.*]] = add i32 [[TMP0]], [[TMP1]]92; CHECK-NEXT: store i32 [[ADD]], ptr [[OUTPUT]], align 493; CHECK-NEXT: [[TMP2:%.*]] = load i32, ptr [[OUTPUT]], align 494; CHECK-NEXT: [[TMP3:%.*]] = load i32, ptr [[OUTPUT]], align 495; CHECK-NEXT: [[MUL:%.*]] = mul i32 [[TMP2]], [[ADD]]96; CHECK-NEXT: store i32 [[MUL]], ptr [[RESULT]], align 497; CHECK-NEXT: ret void98;99; NOCOST-LABEL: @function3(100; NOCOST-NEXT: entry:101; NOCOST-NEXT: [[DOTLOC:%.*]] = alloca i32, align 4102; NOCOST-NEXT: [[ADD_LOC:%.*]] = alloca i32, align 4103; NOCOST-NEXT: [[A:%.*]] = alloca i32, align 4104; NOCOST-NEXT: [[B:%.*]] = alloca i32, align 4105; NOCOST-NEXT: [[OUTPUT:%.*]] = alloca i32, align 4106; NOCOST-NEXT: [[RESULT:%.*]] = alloca i32, align 4107; NOCOST-NEXT: call void @llvm.lifetime.start.p0(ptr [[ADD_LOC]])108; NOCOST-NEXT: call void @llvm.lifetime.start.p0(ptr [[DOTLOC]])109; NOCOST-NEXT: call void @outlined_ir_func_1(ptr [[A]], ptr [[B]], ptr [[OUTPUT]], ptr [[ADD_LOC]], ptr [[DOTLOC]])110; NOCOST-NEXT: [[ADD_RELOAD:%.*]] = load i32, ptr [[ADD_LOC]], align 4111; NOCOST-NEXT: [[DOTRELOAD:%.*]] = load i32, ptr [[DOTLOC]], align 4112; NOCOST-NEXT: call void @llvm.lifetime.end.p0(ptr [[ADD_LOC]])113; NOCOST-NEXT: call void @llvm.lifetime.end.p0(ptr [[DOTLOC]])114; NOCOST-NEXT: [[TMP0:%.*]] = load i32, ptr [[OUTPUT]], align 4115; NOCOST-NEXT: call void @outlined_ir_func_2(i32 [[DOTRELOAD]], i32 [[ADD_RELOAD]], ptr [[RESULT]])116; NOCOST-NEXT: ret void117;118entry:119 %a = alloca i32, align 4120 %b = alloca i32, align 4121 %output = alloca i32, align 4122 %result = alloca i32, align 4123 store i32 2, ptr %a, align 4124 store i32 3, ptr %b, align 4125 %0 = load i32, ptr %a, align 4126 %1 = load i32, ptr %b, align 4127 %add = add i32 %0, %1128 store i32 %add, ptr %output, align 4129 %2 = load i32, ptr %output, align 4130 %3 = load i32, ptr %output, align 4131 %mul = mul i32 %2, %add132 store i32 %mul, ptr %result, align 4133 ret void134}135 136define void @function4() #0 {137; CHECK-LABEL: @function4(138; CHECK-NEXT: entry:139; CHECK-NEXT: [[A:%.*]] = alloca i32, align 4140; CHECK-NEXT: [[B:%.*]] = alloca i32, align 4141; CHECK-NEXT: [[OUTPUT:%.*]] = alloca i32, align 4142; CHECK-NEXT: [[RESULT:%.*]] = alloca i32, align 4143; CHECK-NEXT: store i32 2, ptr [[A]], align 4144; CHECK-NEXT: store i32 3, ptr [[B]], align 4145; CHECK-NEXT: [[TMP0:%.*]] = load i32, ptr [[A]], align 4146; CHECK-NEXT: [[TMP1:%.*]] = load i32, ptr [[B]], align 4147; CHECK-NEXT: [[ADD:%.*]] = add i32 [[TMP0]], [[TMP1]]148; CHECK-NEXT: store i32 [[ADD]], ptr [[OUTPUT]], align 4149; CHECK-NEXT: [[TMP2:%.*]] = load i32, ptr [[OUTPUT]], align 4150; CHECK-NEXT: [[MUL:%.*]] = mul i32 [[TMP2]], [[ADD]]151; CHECK-NEXT: store i32 [[MUL]], ptr [[RESULT]], align 4152; CHECK-NEXT: ret void153;154; NOCOST-LABEL: @function4(155; NOCOST-NEXT: entry:156; NOCOST-NEXT: [[DOTLOC:%.*]] = alloca i32, align 4157; NOCOST-NEXT: [[ADD_LOC:%.*]] = alloca i32, align 4158; NOCOST-NEXT: [[A:%.*]] = alloca i32, align 4159; NOCOST-NEXT: [[B:%.*]] = alloca i32, align 4160; NOCOST-NEXT: [[OUTPUT:%.*]] = alloca i32, align 4161; NOCOST-NEXT: [[RESULT:%.*]] = alloca i32, align 4162; NOCOST-NEXT: call void @llvm.lifetime.start.p0(ptr [[ADD_LOC]])163; NOCOST-NEXT: call void @llvm.lifetime.start.p0(ptr [[DOTLOC]])164; NOCOST-NEXT: call void @outlined_ir_func_1(ptr [[A]], ptr [[B]], ptr [[OUTPUT]], ptr [[ADD_LOC]], ptr [[DOTLOC]])165; NOCOST-NEXT: [[ADD_RELOAD:%.*]] = load i32, ptr [[ADD_LOC]], align 4166; NOCOST-NEXT: [[DOTRELOAD:%.*]] = load i32, ptr [[DOTLOC]], align 4167; NOCOST-NEXT: call void @llvm.lifetime.end.p0(ptr [[ADD_LOC]])168; NOCOST-NEXT: call void @llvm.lifetime.end.p0(ptr [[DOTLOC]])169; NOCOST-NEXT: call void @outlined_ir_func_2(i32 [[DOTRELOAD]], i32 [[ADD_RELOAD]], ptr [[RESULT]])170; NOCOST-NEXT: ret void171;172entry:173 %a = alloca i32, align 4174 %b = alloca i32, align 4175 %output = alloca i32, align 4176 %result = alloca i32, align 4177 store i32 2, ptr %a, align 4178 store i32 3, ptr %b, align 4179 %0 = load i32, ptr %a, align 4180 %1 = load i32, ptr %b, align 4181 %add = add i32 %0, %1182 store i32 %add, ptr %output, align 4183 %2 = load i32, ptr %output, align 4184 %mul = mul i32 %2, %add185 store i32 %mul, ptr %result, align 4186 ret void187}188