53 lines · plain
1; RUN: opt < %s -passes=loop-vectorize -mattr=avx,+slow-unaligned-mem-32 -S | FileCheck %s --check-prefix=SLOWMEM32 --check-prefix=CHECK2; RUN: opt < %s -passes=loop-vectorize -mattr=avx,-slow-unaligned-mem-32 -S | FileCheck %s --check-prefix=FASTMEM32 --check-prefix=CHECK3 4target datalayout = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v64:64:64-v128:128:128-a0:0:64-s0:64:64-f80:128:128-n8:16:32:64-S128"5target triple = "x86_64-apple-macosx10.8.0"6 7; CHECK-LABEL: @read_mod_write_single_ptr(8; CHECK: load <8 x float>9; CHECK: ret void10define void @read_mod_write_single_ptr(ptr nocapture %a, i32 %n) nounwind uwtable ssp {11 %1 = icmp sgt i32 %n, 012 br i1 %1, label %.lr.ph, label %._crit_edge13 14.lr.ph: ; preds = %0, %.lr.ph15 %indvars.iv = phi i64 [ %indvars.iv.next, %.lr.ph ], [ 0, %0 ]16 %2 = getelementptr inbounds float, ptr %a, i64 %indvars.iv17 %3 = load float, ptr %2, align 418 %4 = fmul float %3, 3.000000e+0019 store float %4, ptr %2, align 420 %indvars.iv.next = add i64 %indvars.iv, 121 %lftr.wideiv = trunc i64 %indvars.iv.next to i3222 %exitcond = icmp eq i32 %lftr.wideiv, %n23 br i1 %exitcond, label %._crit_edge, label %.lr.ph24 25._crit_edge: ; preds = %.lr.ph, %026 ret void27}28 29 30; CHECK-LABEL: @read_mod_i64(31; SLOWMEM32: load <2 x i64>32; FASTMEM32: load <4 x i64>33; CHECK: ret void34define void @read_mod_i64(ptr nocapture %a, i32 %n) nounwind uwtable ssp {35 %1 = icmp sgt i32 %n, 036 br i1 %1, label %.lr.ph, label %._crit_edge37 38.lr.ph: ; preds = %0, %.lr.ph39 %indvars.iv = phi i64 [ %indvars.iv.next, %.lr.ph ], [ 0, %0 ]40 %2 = getelementptr inbounds i64, ptr %a, i64 %indvars.iv41 %3 = load i64, ptr %2, align 442 %4 = add i64 %3, 343 store i64 %4, ptr %2, align 444 %indvars.iv.next = add i64 %indvars.iv, 145 %lftr.wideiv = trunc i64 %indvars.iv.next to i3246 %exitcond = icmp eq i32 %lftr.wideiv, %n47 br i1 %exitcond, label %._crit_edge, label %.lr.ph48 49._crit_edge: ; preds = %.lr.ph, %050 ret void51}52 53