1; RUN: llc < %s -march=x86 -mattr=+sse42 | FileCheck %s 2; CHECK: sarb 3; CHECK: sarb 4; CHECK: sarb 5; CHECK: sarb 6; CHECK: sarb 7; CHECK: sarb 8; CHECK: sarb 9; CHECK: sarb 10 11; v8i8 that is widen to v16i8 then split 12; FIXME: This is widen to v16i8 and split to 16 and we then rebuild the vector. 13; Unfortunately, we don't split the store so we don't get the code we want. 14 15define void @update(i64* %dst_i, i64* %src_i, i32 %n) nounwind { 16entry: 17 %dst_i.addr = alloca i64* ; <i64**> [#uses=2] 18 %src_i.addr = alloca i64* ; <i64**> [#uses=2] 19 %n.addr = alloca i32 ; <i32*> [#uses=2] 20 %i = alloca i32, align 4 ; <i32*> [#uses=8] 21 %dst = alloca <8 x i8>*, align 4 ; <<8 x i8>**> [#uses=2] 22 %src = alloca <8 x i8>*, align 4 ; <<8 x i8>**> [#uses=2] 23 store i64* %dst_i, i64** %dst_i.addr 24 store i64* %src_i, i64** %src_i.addr 25 store i32 %n, i32* %n.addr 26 store i32 0, i32* %i 27 br label %forcond 28 29forcond: ; preds = %forinc, %entry 30 %tmp = load i32* %i ; <i32> [#uses=1] 31 %tmp1 = load i32* %n.addr ; <i32> [#uses=1] 32 %cmp = icmp slt i32 %tmp, %tmp1 ; <i1> [#uses=1] 33 br i1 %cmp, label %forbody, label %afterfor 34 35forbody: ; preds = %forcond 36 %tmp2 = load i32* %i ; <i32> [#uses=1] 37 %tmp3 = load i64** %dst_i.addr ; <i64*> [#uses=1] 38 %arrayidx = getelementptr i64* %tmp3, i32 %tmp2 ; <i64*> [#uses=1] 39 %conv = bitcast i64* %arrayidx to <8 x i8>* ; <<8 x i8>*> [#uses=1] 40 store <8 x i8>* %conv, <8 x i8>** %dst 41 %tmp4 = load i32* %i ; <i32> [#uses=1] 42 %tmp5 = load i64** %src_i.addr ; <i64*> [#uses=1] 43 %arrayidx6 = getelementptr i64* %tmp5, i32 %tmp4 ; <i64*> [#uses=1] 44 %conv7 = bitcast i64* %arrayidx6 to <8 x i8>* ; <<8 x i8>*> [#uses=1] 45 store <8 x i8>* %conv7, <8 x i8>** %src 46 %tmp8 = load i32* %i ; <i32> [#uses=1] 47 %tmp9 = load <8 x i8>** %dst ; <<8 x i8>*> [#uses=1] 48 %arrayidx10 = getelementptr <8 x i8>* %tmp9, i32 %tmp8 ; <<8 x i8>*> [#uses=1] 49 %tmp11 = load i32* %i ; <i32> [#uses=1] 50 %tmp12 = load <8 x i8>** %src ; <<8 x i8>*> [#uses=1] 51 %arrayidx13 = getelementptr <8 x i8>* %tmp12, i32 %tmp11 ; <<8 x i8>*> [#uses=1] 52 %tmp14 = load <8 x i8>* %arrayidx13 ; <<8 x i8>> [#uses=1] 53 %add = add <8 x i8> %tmp14, < i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1 > ; <<8 x i8>> [#uses=1] 54 %shr = ashr <8 x i8> %add, < i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2 > ; <<8 x i8>> [#uses=1] 55 store <8 x i8> %shr, <8 x i8>* %arrayidx10 56 br label %forinc 57 58forinc: ; preds = %forbody 59 %tmp15 = load i32* %i ; <i32> [#uses=1] 60 %inc = add i32 %tmp15, 1 ; <i32> [#uses=1] 61 store i32 %inc, i32* %i 62 br label %forcond 63 64afterfor: ; preds = %forcond 65 ret void 66} 67 68