78 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py2; RUN: opt -S -mtriple=x86_64 -mcpu=skylake-avx512 -passes=slp-vectorizer -pass-remarks-output=%t < %s | FileCheck %s3; RUN: FileCheck --input-file=%t --check-prefix=YAML %s4 5define i32 @test(ptr noalias %p, ptr noalias %addr) {6; CHECK-LABEL: @test(7; CHECK-NEXT: entry:8; CHECK-NEXT: [[OFF0_1:%.*]] = getelementptr inbounds i32, ptr [[ADDR:%.*]], i32 19; CHECK-NEXT: [[TMP0:%.*]] = call <15 x i32> @llvm.masked.load.v15i32.p0(ptr align 8 [[OFF0_1]], <15 x i1> <i1 true, i1 false, i1 true, i1 false, i1 true, i1 false, i1 true, i1 false, i1 true, i1 false, i1 true, i1 false, i1 true, i1 false, i1 true>, <15 x i32> poison)10; CHECK-NEXT: [[TMP3:%.*]] = shufflevector <15 x i32> [[TMP0]], <15 x i32> poison, <8 x i32> <i32 0, i32 2, i32 4, i32 6, i32 8, i32 10, i32 12, i32 14>11; CHECK-NEXT: [[TMP4:%.*]] = insertelement <8 x ptr> poison, ptr [[P:%.*]], i32 012; CHECK-NEXT: [[TMP5:%.*]] = shufflevector <8 x ptr> [[TMP4]], <8 x ptr> poison, <8 x i32> zeroinitializer13; CHECK-NEXT: [[TMP6:%.*]] = getelementptr i32, <8 x ptr> [[TMP5]], <8 x i32> [[TMP3]]14; CHECK-NEXT: [[TMP7:%.*]] = call <8 x i32> @llvm.masked.gather.v8i32.v8p0(<8 x ptr> align 4 [[TMP6]], <8 x i1> splat (i1 true), <8 x i32> poison)15; CHECK-NEXT: [[TMP8:%.*]] = call i32 @llvm.vector.reduce.add.v8i32(<8 x i32> [[TMP7]])16; CHECK-NEXT: ret i32 [[TMP8]]17;18; YAML: --- !Passed19 ; YAML-NEXT: Pass: slp-vectorizer20 ; YAML-NEXT: Name: VectorizedHorizontalReduction21 ; YAML-NEXT: Function: test22 ; YAML-NEXT: Args:23 ; YAML-NEXT: - String: 'Vectorized horizontal reduction with cost '24 ; YAML-NEXT: - Cost: '-10'25 ; YAML-NEXT: - String: ' and with tree size '26 ; YAML-NEXT: - TreeSize: '5'27entry:28 %off0.1 = getelementptr inbounds i32, ptr %addr, i32 129 %idx0 = load i32, ptr %off0.1, align 830 %gep0 = getelementptr inbounds i32, ptr %p, i32 %idx031 %ld0 = load i32, ptr %gep0, align 432 33 %off1.3 = getelementptr inbounds i32, ptr %addr, i32 334 %idx1 = load i32, ptr %off1.3, align 835 %gep1 = getelementptr inbounds i32, ptr %p, i32 %idx136 %ld1 = load i32, ptr %gep1, align 437 38 %off2.5 = getelementptr inbounds i32, ptr %addr, i32 539 %idx2 = load i32, ptr %off2.5, align 840 %gep2 = getelementptr inbounds i32, ptr %p, i32 %idx241 %ld2 = load i32, ptr %gep2, align 442 43 %off3.7 = getelementptr inbounds i32, ptr %addr, i32 744 %idx3 = load i32, ptr %off3.7, align 845 %gep3 = getelementptr inbounds i32, ptr %p, i32 %idx346 %ld3 = load i32, ptr %gep3, align 447 48 %off4.9 = getelementptr inbounds i32, ptr %addr, i32 949 %idx4 = load i32, ptr %off4.9, align 850 %gep4 = getelementptr inbounds i32, ptr %p, i32 %idx451 %ld4 = load i32, ptr %gep4, align 452 53 %off5.11 = getelementptr inbounds i32, ptr %addr, i32 1154 %idx5 = load i32, ptr %off5.11, align 855 %gep5 = getelementptr inbounds i32, ptr %p, i32 %idx556 %ld5 = load i32, ptr %gep5, align 457 58 %off6.13 = getelementptr inbounds i32, ptr %addr, i32 1359 %idx6 = load i32, ptr %off6.13, align 860 %gep6 = getelementptr inbounds i32, ptr %p, i32 %idx661 %ld6 = load i32, ptr %gep6, align 462 63 %off7.15 = getelementptr inbounds i32, ptr %addr, i32 1564 %idx7 = load i32, ptr %off7.15, align 865 %gep7 = getelementptr inbounds i32, ptr %p, i32 %idx766 %ld7 = load i32, ptr %gep7, align 467 68 %add0 = add nsw i32 %ld1, %ld069 %add1 = add nsw i32 %add0, %ld270 %add2 = add nsw i32 %add1, %ld371 %add3 = add nsw i32 %add2, %ld472 %add4 = add nsw i32 %add3, %ld573 %add5 = add nsw i32 %add4, %ld674 %add6 = add nsw i32 %add5, %ld775 76 ret i32 %add677}78