1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py 2; RUN: opt %s -loop-vectorize -instcombine -simplifycfg -mtriple=x86_64-unknown-linux-gnu -mattr=avx512vl,avx512dq,avx512bw -S | FileCheck %s 3 4@bytes = global [128 x i8] zeroinitializer, align 16 5 6; Make sure we end up with vector code for this loop. We used to try to create 7; a VF=64,UF=4 loop, but the scalar trip count is only 128 so 8; the vector loop was dead code leaving only a scalar remainder. 9define zeroext i8 @sum() { 10; CHECK-LABEL: @sum( 11; CHECK-NEXT: entry: 12; CHECK-NEXT: br label [[VECTOR_BODY:%.*]] 13; CHECK: vector.body: 14; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[ENTRY:%.*]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] 15; CHECK-NEXT: [[VEC_PHI:%.*]] = phi <64 x i8> [ zeroinitializer, [[ENTRY]] ], [ [[TMP4:%.*]], [[VECTOR_BODY]] ] 16; CHECK-NEXT: [[VEC_PHI1:%.*]] = phi <64 x i8> [ zeroinitializer, [[ENTRY]] ], [ [[TMP5:%.*]], [[VECTOR_BODY]] ] 17; CHECK-NEXT: [[TMP0:%.*]] = getelementptr inbounds [128 x i8], [128 x i8]* @bytes, i64 0, i64 [[INDEX]] 18; CHECK-NEXT: [[TMP1:%.*]] = bitcast i8* [[TMP0]] to <64 x i8>* 19; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <64 x i8>, <64 x i8>* [[TMP1]], align 16 20; CHECK-NEXT: [[TMP2:%.*]] = getelementptr inbounds i8, i8* [[TMP0]], i64 64 21; CHECK-NEXT: [[TMP3:%.*]] = bitcast i8* [[TMP2]] to <64 x i8>* 22; CHECK-NEXT: [[WIDE_LOAD2:%.*]] = load <64 x i8>, <64 x i8>* [[TMP3]], align 16 23; CHECK-NEXT: [[TMP4]] = add <64 x i8> [[WIDE_LOAD]], [[VEC_PHI]] 24; CHECK-NEXT: [[TMP5]] = add <64 x i8> [[WIDE_LOAD2]], [[VEC_PHI1]] 25; CHECK-NEXT: [[INDEX_NEXT]] = add i64 [[INDEX]], 128 26; CHECK-NEXT: [[TMP6:%.*]] = icmp eq i64 [[INDEX]], 0 27; CHECK-NEXT: br i1 [[TMP6]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop !0 28; CHECK: middle.block: 29; CHECK-NEXT: [[BIN_RDX:%.*]] = add <64 x i8> [[TMP5]], [[TMP4]] 30; CHECK-NEXT: [[TMP7:%.*]] = call i8 @llvm.vector.reduce.add.v64i8(<64 x i8> [[BIN_RDX]]) 31; CHECK-NEXT: ret i8 [[TMP7]] 32; 33entry: 34 br label %for.body 35 36for.body: ; preds = %for.body, %entry 37 %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ] 38 %r.010 = phi i8 [ 0, %entry ], [ %add, %for.body ] 39 %arrayidx = getelementptr inbounds [128 x i8], [128 x i8]* @bytes, i64 0, i64 %indvars.iv 40 %0 = load i8, i8* %arrayidx, align 1 41 %add = add i8 %0, %r.010 42 %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1 43 %exitcond = icmp eq i64 %indvars.iv.next, 128 44 br i1 %exitcond, label %for.end, label %for.body 45 46for.end: ; preds = %for.body 47 %add.lcssa = phi i8 [ %add, %for.body ] 48 ret i8 %add.lcssa 49} 50