40 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@float_array = common global [10000 x float] zeroinitializer, align 167@unsigned_array = common global [10000 x i32] zeroinitializer, align 168 9; If we need to scalarize the fptoui and then use inserts to build up the10; vector again, then there is certainly no value in going 256-bit wide.11; CHECK-NOT: fptoui <4 x float>12 13define void @convert(i32 %N) {14entry:15 %0 = icmp eq i32 %N, 016 br i1 %0, label %for.end, label %for.body.preheader17 18for.body.preheader: ; preds = %entry19 br label %for.body20 21for.body: ; preds = %for.body.preheader, %for.body22 %indvars.iv = phi i64 [ %indvars.iv.next, %for.body ], [ 0, %for.body.preheader ]23 %arrayidx = getelementptr inbounds [10000 x float], ptr @float_array, i64 0, i64 %indvars.iv24 %1 = load float, ptr %arrayidx, align 425 %conv = fptoui float %1 to i3226 %arrayidx2 = getelementptr inbounds [10000 x i32], ptr @unsigned_array, i64 0, i64 %indvars.iv27 store i32 %conv, ptr %arrayidx2, align 428 %indvars.iv.next = add nuw nsw i64 %indvars.iv, 129 %lftr.wideiv = trunc i64 %indvars.iv.next to i3230 %exitcond = icmp eq i32 %lftr.wideiv, %N31 br i1 %exitcond, label %for.end.loopexit, label %for.body32 33for.end.loopexit: ; preds = %for.body34 br label %for.end35 36for.end: ; preds = %for.end.loopexit, %entry37 ret void38}39 40