1 ; RUN: llc < %s -march=x86-64 -mattr=+sse4.2 | FileCheck %s 2 ; CHECK: psubw 3 ; CHECK-NEXT: pmullw 4 5 ; Widen a v5i16 to v8i16 to do a vector sub and multiple 6 7 define void @update(<5 x i16>* %dst, <5 x i16>* %src, i32 %n) nounwind { 8 entry: 9 %dst.addr = alloca <5 x i16>* ; <<5 x i16>**> [#uses=2] 10 %src.addr = alloca <5 x i16>* ; <<5 x i16>**> [#uses=2] 11 %n.addr = alloca i32 ; <i32*> [#uses=2] 12 %v = alloca <5 x i16>, align 16 ; <<5 x i16>*> [#uses=1] 13 %i = alloca i32, align 4 ; <i32*> [#uses=6] 14 store <5 x i16>* %dst, <5 x i16>** %dst.addr 15 store <5 x i16>* %src, <5 x i16>** %src.addr 16 store i32 %n, i32* %n.addr 17 store <5 x i16> < i16 1, i16 1, i16 1, i16 0, i16 0 >, <5 x i16>* %v 18 store i32 0, i32* %i 19 br label %forcond 20 21 forcond: ; preds = %forinc, %entry 22 %tmp = load i32, i32* %i ; <i32> [#uses=1] 23 %tmp1 = load i32, i32* %n.addr ; <i32> [#uses=1] 24 %cmp = icmp slt i32 %tmp, %tmp1 ; <i1> [#uses=1] 25 br i1 %cmp, label %forbody, label %afterfor 26 27 forbody: ; preds = %forcond 28 %tmp2 = load i32, i32* %i ; <i32> [#uses=1] 29 %tmp3 = load <5 x i16>*, <5 x i16>** %dst.addr ; <<5 x i16>*> [#uses=1] 30 %arrayidx = getelementptr <5 x i16>, <5 x i16>* %tmp3, i32 %tmp2 ; <<5 x i16>*> [#uses=1] 31 %tmp4 = load i32, i32* %i ; <i32> [#uses=1] 32 %tmp5 = load <5 x i16>*, <5 x i16>** %src.addr ; <<5 x i16>*> [#uses=1] 33 %arrayidx6 = getelementptr <5 x i16>, <5 x i16>* %tmp5, i32 %tmp4 ; <<5 x i16>*> [#uses=1] 34 %tmp7 = load <5 x i16>, <5 x i16>* %arrayidx6 ; <<5 x i16>> [#uses=1] 35 %sub = sub <5 x i16> %tmp7, < i16 271, i16 271, i16 271, i16 271, i16 271 > ; <<5 x i16>> [#uses=1] 36 %mul = mul <5 x i16> %sub, < i16 2, i16 4, i16 2, i16 2, i16 2 > ; <<5 x i16>> [#uses=1] 37 store <5 x i16> %mul, <5 x i16>* %arrayidx 38 br label %forinc 39 40 forinc: ; preds = %forbody 41 %tmp8 = load i32, i32* %i ; <i32> [#uses=1] 42 %inc = add i32 %tmp8, 1 ; <i32> [#uses=1] 43 store i32 %inc, i32* %i 44 br label %forcond 45 46 afterfor: ; preds = %forcond 47 ret void 48 } 49 50
