1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py 2; RUN: opt < %s -vector-library=MASSV -force-vector-interleave=1 \ 3; RUN: -vectorizer-maximize-bandwidth -passes='default<O2>,inject-tli-mappings,loop-vectorize' \ 4; RUN: -mtriple=powerpc64le-unknown-linux -S -mcpu=pwr9 2>&1 | FileCheck %s 5 6define dso_local double @test(ptr %Arr) { 7; CHECK-LABEL: @test( 8; CHECK-NEXT: entry: 9; CHECK-NEXT: br label [[VECTOR_BODY:%.*]] 10; CHECK: vector.body: 11; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[ENTRY:%.*]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] 12; CHECK-NEXT: [[VEC_PHI:%.*]] = phi <2 x double> [ zeroinitializer, [[ENTRY]] ], [ [[TMP3:%.*]], [[VECTOR_BODY]] ] 13; CHECK-NEXT: [[TMP0:%.*]] = getelementptr inbounds nuw float, ptr [[ARR:%.*]], i64 [[INDEX]] 14; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <2 x float>, ptr [[TMP0]], align 4 15; CHECK-NEXT: [[TMP1:%.*]] = fpext <2 x float> [[WIDE_LOAD]] to <2 x double> 16; CHECK-NEXT: [[TMP2:%.*]] = tail call fast <2 x double> @__sind2(<2 x double> [[TMP1]]) 17; CHECK-NEXT: [[TMP3]] = fadd fast <2 x double> [[TMP2]], [[VEC_PHI]] 18; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 2 19; CHECK-NEXT: [[TMP4:%.*]] = icmp eq i64 [[INDEX_NEXT]], 128 20; CHECK-NEXT: br i1 [[TMP4]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]] 21; CHECK: middle.block: 22; CHECK-NEXT: [[DOTLCSSA:%.*]] = phi <2 x double> [ [[TMP3]], [[VECTOR_BODY]] ] 23; CHECK-NEXT: [[TMP5:%.*]] = tail call fast double @llvm.vector.reduce.fadd.v2f64(double 0.000000e+00, <2 x double> [[DOTLCSSA]]) 24; CHECK-NEXT: ret double [[TMP5]] 25; 26entry: 27 br label %for.cond 28 29for.cond: 30 %Sum.0 = phi double [ 0.000000e+00, %entry ], [ %add, %for.inc ] 31 %i.0 = phi i32 [ 0, %entry ], [ %inc, %for.inc ] 32 %cmp = icmp slt i32 %i.0, 128 33 br i1 %cmp, label %for.body, label %for.cond.cleanup 34 35for.cond.cleanup: 36 br label %for.end 37 38for.body: 39 %idxprom = sext i32 %i.0 to i64 40 %arrayidx = getelementptr inbounds float, ptr %Arr, i64 %idxprom 41 %0 = load float, ptr %arrayidx, align 4 42 %conv = fpext float %0 to double 43 %1 = call fast double @llvm.sin.f64(double %conv) 44 %add = fadd fast double %Sum.0, %1 45 br label %for.inc 46 47for.inc: 48 %inc = add nsw i32 %i.0, 1 49 br label %for.cond 50 51for.end: 52 ret double %Sum.0 53} 54 55declare double @llvm.sin.f64(double) 56