1cee313d2SEric Christopher; RUN: opt < %s -loop-vectorize -force-vector-interleave=1 -force-vector-width=4 -dce -instcombine -S | FileCheck %s 2cee313d2SEric Christopher 3cee313d2SEric Christophertarget 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" 4cee313d2SEric Christopher 5cee313d2SEric Christopher;CHECK-LABEL: @read_only_func( 6cee313d2SEric Christopher;CHECK: load <4 x i32> 7cee313d2SEric Christopher;CHECK: ret i32 8cee313d2SEric Christopherdefine i32 @read_only_func(i32* nocapture %A, i32* nocapture %B, i32 %n) nounwind uwtable readonly ssp { 9cee313d2SEric Christopher %1 = icmp sgt i32 %n, 0 10cee313d2SEric Christopher br i1 %1, label %.lr.ph, label %._crit_edge 11cee313d2SEric Christopher 12cee313d2SEric Christopher.lr.ph: ; preds = %0, %.lr.ph 13cee313d2SEric Christopher %indvars.iv = phi i64 [ %indvars.iv.next, %.lr.ph ], [ 0, %0 ] 14cee313d2SEric Christopher %sum.02 = phi i32 [ %9, %.lr.ph ], [ 0, %0 ] 15cee313d2SEric Christopher %2 = getelementptr inbounds i32, i32* %A, i64 %indvars.iv 16cee313d2SEric Christopher %3 = load i32, i32* %2, align 4 17cee313d2SEric Christopher %4 = add nsw i64 %indvars.iv, 13 18cee313d2SEric Christopher %5 = getelementptr inbounds i32, i32* %B, i64 %4 19cee313d2SEric Christopher %6 = load i32, i32* %5, align 4 20cee313d2SEric Christopher %7 = shl i32 %6, 1 21cee313d2SEric Christopher %8 = add i32 %3, %sum.02 22cee313d2SEric Christopher %9 = add i32 %8, %7 23cee313d2SEric Christopher %indvars.iv.next = add i64 %indvars.iv, 1 24cee313d2SEric Christopher %lftr.wideiv = trunc i64 %indvars.iv.next to i32 25cee313d2SEric Christopher %exitcond = icmp eq i32 %lftr.wideiv, %n 26cee313d2SEric Christopher br i1 %exitcond, label %._crit_edge, label %.lr.ph 27cee313d2SEric Christopher 28cee313d2SEric Christopher._crit_edge: ; preds = %.lr.ph, %0 29cee313d2SEric Christopher %sum.0.lcssa = phi i32 [ 0, %0 ], [ %9, %.lr.ph ] 30cee313d2SEric Christopher ret i32 %sum.0.lcssa 31cee313d2SEric Christopher} 32*1e21181aSMatt Arsenault 33*1e21181aSMatt Arsenault; Ensure that volatile loads are not vectorized in a read-only loop. 34*1e21181aSMatt Arsenault 35*1e21181aSMatt Arsenault;CHECK-LABEL: @read_only_func_volatile( 36*1e21181aSMatt Arsenault;CHECK-NOT: load <4 x i32> 37*1e21181aSMatt Arsenault;CHECK: ret i32 38*1e21181aSMatt Arsenaultdefine i32 @read_only_func_volatile(i32* nocapture %A, i32* nocapture %B, i32 %n) nounwind uwtable readonly ssp { 39*1e21181aSMatt Arsenault %1 = icmp sgt i32 %n, 0 40*1e21181aSMatt Arsenault br i1 %1, label %.lr.ph, label %._crit_edge 41*1e21181aSMatt Arsenault 42*1e21181aSMatt Arsenault.lr.ph: ; preds = %0, %.lr.ph 43*1e21181aSMatt Arsenault %indvars.iv = phi i64 [ %indvars.iv.next, %.lr.ph ], [ 0, %0 ] 44*1e21181aSMatt Arsenault %sum.02 = phi i32 [ %9, %.lr.ph ], [ 0, %0 ] 45*1e21181aSMatt Arsenault %2 = getelementptr inbounds i32, i32* %A, i64 %indvars.iv 46*1e21181aSMatt Arsenault %3 = load volatile i32, i32* %2, align 4 47*1e21181aSMatt Arsenault %4 = add nsw i64 %indvars.iv, 13 48*1e21181aSMatt Arsenault %5 = getelementptr inbounds i32, i32* %B, i64 %4 49*1e21181aSMatt Arsenault %6 = load i32, i32* %5, align 4 50*1e21181aSMatt Arsenault %7 = shl i32 %6, 1 51*1e21181aSMatt Arsenault %8 = add i32 %3, %sum.02 52*1e21181aSMatt Arsenault %9 = add i32 %8, %7 53*1e21181aSMatt Arsenault %indvars.iv.next = add i64 %indvars.iv, 1 54*1e21181aSMatt Arsenault %lftr.wideiv = trunc i64 %indvars.iv.next to i32 55*1e21181aSMatt Arsenault %exitcond = icmp eq i32 %lftr.wideiv, %n 56*1e21181aSMatt Arsenault br i1 %exitcond, label %._crit_edge, label %.lr.ph 57*1e21181aSMatt Arsenault 58*1e21181aSMatt Arsenault._crit_edge: ; preds = %.lr.ph, %0 59*1e21181aSMatt Arsenault %sum.0.lcssa = phi i32 [ 0, %0 ], [ %9, %.lr.ph ] 60*1e21181aSMatt Arsenault ret i32 %sum.0.lcssa 61*1e21181aSMatt Arsenault} 62