41 lines · plain
1; RUN: opt < %s -mcpu=core-avx2 -passes=loop-vectorize -S | FileCheck %s2 3target datalayout = "e-m:o-i64:64-f80:128-n8:16:32:64-S128"4target triple = "x86_64-apple-macosx"5 6@n = global i32 10000, align 47@double_array = common global [10000 x double] zeroinitializer, align 168@unsigned_array = common global [10000 x i32] zeroinitializer, align 169 10; If we need to scalarize the fptoui and then use inserts to build up the11; vector again, then there is certainly no value in going 256-bit wide.12; CHECK-NOT: fptoui <2 x double>13 14define void @convert() {15entry:16 %0 = load i32, ptr @n, align 417 %cmp4 = icmp eq i32 %0, 018 br i1 %cmp4, label %for.end, label %for.body.preheader19 20for.body.preheader: ; preds = %entry21 br label %for.body22 23for.body: ; preds = %for.body.preheader, %for.body24 %indvars.iv = phi i64 [ %indvars.iv.next, %for.body ], [ 0, %for.body.preheader ]25 %arrayidx = getelementptr inbounds [10000 x double], ptr @double_array, i64 0, i64 %indvars.iv26 %1 = load double, ptr %arrayidx, align 827 %conv = fptoui double %1 to i3228 %arrayidx2 = getelementptr inbounds [10000 x i32], ptr @unsigned_array, i64 0, i64 %indvars.iv29 store i32 %conv, ptr %arrayidx2, align 430 %indvars.iv.next = add nuw nsw i64 %indvars.iv, 131 %2 = trunc i64 %indvars.iv.next to i3232 %cmp = icmp ult i32 %2, %033 br i1 %cmp, label %for.body, label %for.end.loopexit34 35for.end.loopexit: ; preds = %for.body36 br label %for.end37 38for.end: ; preds = %for.end.loopexit, %entry39 ret void40}41