1; RUN: opt -force-vector-interleave=2 -force-vector-width=4 -passes=loop-vectorize -pass-remarks-analysis=loop-vectorize -disable-output < %s 2>&1 | FileCheck %s 2 3; CHECK: remark: mixed-precision.c:3:26: floating point conversion changes vector width. Mixed floating point precision requires an up/down cast that will negatively impact performance. 4define void @f(ptr noalias nocapture %X, i64 %N) { 5entry: 6 br label %for.body 7 8for.cond.cleanup: 9 ret void 10 11for.body: 12 %i = phi i64 [ %inc, %for.body ], [ 0, %entry ] 13 %arrayidx = getelementptr inbounds float, ptr %X, i64 %i 14 %0 = load float, ptr %arrayidx, align 4 15 %conv = fpext float %0 to double, !dbg !9 16 %mul = fmul double %conv, 0x3FD5555555555555 17 %conv3 = fptrunc double %mul to float 18 store float %conv3, ptr %arrayidx, align 4 19 %inc = add nuw i64 %i, 1 20 %exitcond.not = icmp eq i64 %inc, %N 21 br i1 %exitcond.not, label %for.cond.cleanup, label %for.body 22} 23 24; CHECK: remark: mixed-precision.c:8:8: floating point conversion changes vector width. Mixed floating point precision requires an up/down cast that will negatively impact performance. 25; CHECK: remark: mixed-precision.c:7:16: floating point conversion changes vector width. Mixed floating point precision requires an up/down cast that will negatively impact performance. 26; CHECK-NOT: remark: mixed-precision.c:7:16: floating point conversion changes vector width. Mixed floating point precision requires an up/down cast that will negatively impact performance. 27define void @g(ptr noalias nocapture %X, ptr noalias nocapture %Y, i64 %N) { 28entry: 29 %pi = alloca double 30 store double 0x400921FB54442D18, ptr %pi 31 %fac = load double, ptr %pi 32 br label %for.body 33 34for.body: 35 %i = phi i64 [ %inc, %for.body ], [ 0, %entry ] 36 %arrayidx = getelementptr inbounds float, ptr %X, i64 %i 37 %0 = load float, ptr %arrayidx, align 4 38 %conv = fpext float %0 to double, !dbg !10 39 %mul = fmul double %conv, %fac 40 %conv1 = fptrunc double %mul to float 41 store float %conv1, ptr %arrayidx, align 4 42 %arrayidx5 = getelementptr inbounds float, ptr %Y, i64 %i 43 %1 = load float, ptr %arrayidx5, align 4 44 %conv2 = fpext float %1 to double, !dbg !11 45 %mul2 = fmul double %conv2, %fac 46 %conv3 = fptrunc double %mul2 to float 47 store float %conv3, ptr %arrayidx5, align 4 48 %inc = add nuw nsw i64 %i, 1 49 %exitcond.not = icmp eq i64 %inc, %N 50 br i1 %exitcond.not, label %for.cond.cleanup, label %for.body 51 52for.cond.cleanup: 53 ret void 54} 55 56!llvm.dbg.cu = !{!0} 57!llvm.module.flags = !{!3, !4} 58 59!0 = distinct !DICompileUnit(language: DW_LANG_C99, file: !1, producer: "clang version 12.0.0", isOptimized: true, runtimeVersion: 0, emissionKind: NoDebug, enums: !2, splitDebugInlining: false, nameTableKind: None) 60!1 = !DIFile(filename: "mixed-precision.c", directory: "/tmp/mixed-precision.c") 61!2 = !{} 62!3 = !{i32 2, !"Debug Info Version", i32 3} 63!4 = !{i32 1, !"wchar_size", i32 4} 64!6 = distinct !DISubprogram(name: "f", scope: !1, file: !1, line: 1, type: !8, scopeLine: 1, flags: DIFlagPrototyped, spFlags: DISPFlagDefinition | DISPFlagOptimized, unit: !0, retainedNodes: !2) 65!7 = distinct !DISubprogram(name: "g", scope: !1, file: !1, line: 5, type: !8, scopeLine: 5, flags: DIFlagPrototyped, spFlags: DISPFlagDefinition | DISPFlagOptimized, unit: !0, retainedNodes: !2) 66!8 = !DISubroutineType(types: !2) 67!9 = !DILocation(line: 3, column: 26, scope: !6) 68!10 = !DILocation(line: 7, column: 16, scope: !7) 69!11 = !DILocation(line: 8, column: 8, scope: !7) 70