1; RUN: opt -passes=loop-vectorize -force-vector-width=2 -force-vector-interleave=1 -S < %s | FileCheck %s 2 3target 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" 4 5@A = common global [1024 x float] zeroinitializer, align 16 6@B = common global [1024 x float] zeroinitializer, align 16 7 8; Make sure we can vectorize in the presence of hoistable conditional loads. 9; CHECK-LABEL: @hoist_cond_load( 10; CHECK: load <2 x float> 11 12define void @hoist_cond_load() { 13entry: 14 br label %for.body 15for.body: 16 %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %if.end9 ] 17 %arrayidx = getelementptr inbounds [1024 x float], ptr @A, i64 0, i64 %indvars.iv 18 %arrayidx2 = getelementptr inbounds [1024 x float], ptr @B, i64 0, i64 %indvars.iv 19 %0 = load float, ptr %arrayidx2, align 4 20 %cmp3 = fcmp oeq float %0, 0.000000e+00 21 br i1 %cmp3, label %if.end9, label %if.else 22 23if.else: 24 %1 = load float, ptr %arrayidx, align 4 25 br label %if.end9 26 27if.end9: 28 %tmp.0 = phi float [ %1, %if.else ], [ 0.000000e+00, %for.body ] 29 store float %tmp.0, ptr %arrayidx, align 4 30 %indvars.iv.next = add i64 %indvars.iv, 1 31 %lftr.wideiv = trunc i64 %indvars.iv.next to i32 32 %exitcond = icmp ne i32 %lftr.wideiv, 1024 33 br i1 %exitcond, label %for.body, label %for.end 34 35for.end: 36 ret void 37} 38 39; However, we can't hoist loads whose address we have not seen unconditionally 40; accessed. One wide load is fine, but not the second. 41; CHECK-LABEL: @dont_hoist_cond_load( 42; CHECK: load <2 x float> 43; CHECK-NOT: load <2 x float> 44 45define void @dont_hoist_cond_load(ptr %a) { 46entry: 47 br label %for.body 48for.body: 49 %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %if.end9 ] 50 %arrayidx = getelementptr inbounds [1024 x float], ptr %a, i64 0, i64 %indvars.iv 51 %arrayidx2 = getelementptr inbounds [1024 x float], ptr @B, i64 0, i64 %indvars.iv 52 %0 = load float, ptr %arrayidx2, align 4 53 %cmp3 = fcmp oeq float %0, 0.000000e+00 54 br i1 %cmp3, label %if.end9, label %if.else 55 56if.else: 57 %1 = load float, ptr %arrayidx, align 4 58 br label %if.end9 59 60if.end9: 61 %tmp.0 = phi float [ %1, %if.else ], [ 0.000000e+00, %for.body ] 62 store float %tmp.0, ptr %arrayidx2, align 4 63 %indvars.iv.next = add i64 %indvars.iv, 1 64 %lftr.wideiv = trunc i64 %indvars.iv.next to i32 65 %exitcond = icmp ne i32 %lftr.wideiv, 1024 66 br i1 %exitcond, label %for.body, label %for.end 67 68for.end: 69 ret void 70} 71