40 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 22; RUN: opt -mtriple=arm64-apple-ios -S -passes=slp-vectorizer < %s | FileCheck %s3 4; fshl instruction cost model is an overestimate causing this test to vectorize when it is not beneficial to do so.5define i64 @fshl(i64 %or1, i64 %or2, i64 %or3 ) {6; CHECK-LABEL: define i64 @fshl7; CHECK-SAME: (i64 [[OR1:%.*]], i64 [[OR2:%.*]], i64 [[OR3:%.*]]) {8; CHECK-NEXT: entry:9; CHECK-NEXT: [[OR4:%.*]] = tail call i64 @llvm.fshl.i64(i64 [[OR2]], i64 0, i64 1)10; CHECK-NEXT: [[XOR1:%.*]] = xor i64 [[OR4]], 011; CHECK-NEXT: [[OR5:%.*]] = tail call i64 @llvm.fshl.i64(i64 [[OR3]], i64 0, i64 2)12; CHECK-NEXT: [[XOR2:%.*]] = xor i64 [[OR5]], [[OR1]]13; CHECK-NEXT: [[ADD1:%.*]] = add i64 [[XOR1]], [[OR1]]14; CHECK-NEXT: [[ADD2:%.*]] = add i64 0, [[XOR2]]15; CHECK-NEXT: [[OR6:%.*]] = tail call i64 @llvm.fshl.i64(i64 [[OR1]], i64 [[OR2]], i64 17)16; CHECK-NEXT: [[XOR3:%.*]] = xor i64 [[OR6]], [[ADD1]]17; CHECK-NEXT: [[OR7:%.*]] = tail call i64 @llvm.fshl.i64(i64 0, i64 0, i64 21)18; CHECK-NEXT: [[XOR4:%.*]] = xor i64 [[OR7]], [[ADD2]]19; CHECK-NEXT: [[ADD3:%.*]] = or i64 [[XOR3]], [[ADD2]]20; CHECK-NEXT: [[XOR5:%.*]] = xor i64 [[ADD3]], [[XOR4]]21; CHECK-NEXT: ret i64 [[XOR5]]22;23entry:24 %or4 = tail call i64 @llvm.fshl.i64(i64 %or2, i64 0, i64 1)25 %xor1 = xor i64 %or4, 026 %or5 = tail call i64 @llvm.fshl.i64(i64 %or3, i64 0, i64 2)27 %xor2 = xor i64 %or5, %or128 %add1 = add i64 %xor1, %or129 %add2 = add i64 0, %xor230 %or6 = tail call i64 @llvm.fshl.i64(i64 %or1, i64 %or2, i64 17)31 %xor3 = xor i64 %or6, %add132 %or7 = tail call i64 @llvm.fshl.i64(i64 0, i64 0, i64 21)33 %xor4 = xor i64 %or7, %add234 %add3 = or i64 %xor3, %add235 %xor5 = xor i64 %add3, %xor436 ret i64 %xor537}38 39declare i64 @llvm.fshl.i64(i64, i64, i64)40