50 lines · plain
1; RUN: llc -mtriple=arm64-apple-ios7.0 -mcpu=cyclone %s -o - | FileCheck %s2; RUN: llc -mtriple=aarch64_be-linux-gnu -mcpu=cyclone %s -o - | FileCheck --check-prefix=CHECK-BE %s3 4define i128 @test_128bitmul(i128 %lhs, i128 %rhs) {5; CHECK-LABEL: test_128bitmul:6; CHECK: umulh [[HI:x[0-9]+]], x0, x27; CHECK: madd [[TEMP1:x[0-9]+]], x0, x3, [[HI]]8; CHECK-DAG: madd x1, x1, x2, [[TEMP1]]9; CHECK-DAG: mul x0, x0, x210; CHECK-NEXT: ret11 12; CHECK-BE-LABEL: test_128bitmul:13; CHECK-BE: umulh [[HI:x[0-9]+]], x1, x314; CHECK-BE: madd [[TEMP1:x[0-9]+]], x1, x2, [[HI]]15; CHECK-BE-DAG: madd x0, x0, x3, [[TEMP1]]16; CHECK-BE-DAG: mul x1, x1, x317; CHECK-BE-NEXT: ret18 19 %prod = mul i128 %lhs, %rhs20 ret i128 %prod21}22 23; The machine combiner should create madd instructions when24; optimizing for size because that's smaller than mul + add.25 26define i128 @test_128bitmul_optsize(i128 %lhs, i128 %rhs) optsize {27; CHECK-LABEL: test_128bitmul_optsize:28; CHECK: umulh [[HI:x[0-9]+]], x0, x229; CHECK-NEXT: madd [[TEMP1:x[0-9]+]], x0, x3, [[HI]]30; CHECK-DAG: madd x1, x1, x2, [[TEMP1]]31; CHECK-DAG: mul x0, x0, x232; CHECK-NEXT: ret33 34 %prod = mul i128 %lhs, %rhs35 ret i128 %prod36}37 38define i128 @test_128bitmul_minsize(i128 %lhs, i128 %rhs) minsize {39; CHECK-LABEL: test_128bitmul_minsize:40; CHECK: umulh [[HI:x[0-9]+]], x0, x241; CHECK-NEXT: madd [[TEMP1:x[0-9]+]], x0, x3, [[HI]]42; CHECK-DAG: madd x1, x1, x2, [[TEMP1]]43; CHECK-DAG: mul x0, x0, x244; CHECK-NEXT: ret45 46 %prod = mul i128 %lhs, %rhs47 ret i128 %prod48}49 50