1; RUN: opt < %s -mcpu=core-avx2 -loop-vectorize -S | llc -mcpu=core-avx2 | FileCheck %s 2 3target datalayout = "e-m:o-i64:64-f80:128-n8:16:32:64-S128" 4target triple = "x86_64-apple-macosx" 5 6@float_array = common global [10000 x float] zeroinitializer, align 16 7@unsigned_array = common global [10000 x i32] zeroinitializer, align 16 8 9; If we need to scalarize the fptoui and then use inserts to build up the 10; vector again, then there is certainly no value in going 256-bit wide. 11; CHECK-NOT: vinserti128 12 13define void @convert(i32 %N) { 14entry: 15 %0 = icmp eq i32 %N, 0 16 br i1 %0, label %for.end, label %for.body.preheader 17 18for.body.preheader: ; preds = %entry 19 br label %for.body 20 21for.body: ; preds = %for.body.preheader, %for.body 22 %indvars.iv = phi i64 [ %indvars.iv.next, %for.body ], [ 0, %for.body.preheader ] 23 %arrayidx = getelementptr inbounds [10000 x float], [10000 x float]* @float_array, i64 0, i64 %indvars.iv 24 %1 = load float, float* %arrayidx, align 4 25 %conv = fptoui float %1 to i32 26 %arrayidx2 = getelementptr inbounds [10000 x i32], [10000 x i32]* @unsigned_array, i64 0, i64 %indvars.iv 27 store i32 %conv, i32* %arrayidx2, align 4 28 %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1 29 %lftr.wideiv = trunc i64 %indvars.iv.next to i32 30 %exitcond = icmp eq i32 %lftr.wideiv, %N 31 br i1 %exitcond, label %for.end.loopexit, label %for.body 32 33for.end.loopexit: ; preds = %for.body 34 br label %for.end 35 36for.end: ; preds = %for.end.loopexit, %entry 37 ret void 38} 39 40