-; RUN: llvm-as < %s | llc -march=x86 -mattr=+sse42 -disable-mmx -o %t -f
-; RUN: grep psubw %t | count 1
-; RUN: grep pmullw %t | count 1
+; RUN: llc < %s -march=x86-64 -mattr=+sse4.2 | FileCheck %s
+; CHECK: psubw
+; CHECK-NEXT: pmullw
; Widen a v5i16 to v8i16 to do a vector sub and multiple
-target datalayout = "e-p:32:32:32-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:32:64-f32:32:32-f64:32:64-v64:64:64-v128:128:128-a0:0:64-f80:128:128"
-
define void @update(<5 x i16>* %dst, <5 x i16>* %src, i32 %n) nounwind {
entry:
%dst.addr = alloca <5 x i16>* ; <<5 x i16>**> [#uses=2]
br label %forcond
forcond: ; preds = %forinc, %entry
- %tmp = load i32* %i ; <i32> [#uses=1]
- %tmp1 = load i32* %n.addr ; <i32> [#uses=1]
+ %tmp = load i32, i32* %i ; <i32> [#uses=1]
+ %tmp1 = load i32, i32* %n.addr ; <i32> [#uses=1]
%cmp = icmp slt i32 %tmp, %tmp1 ; <i1> [#uses=1]
br i1 %cmp, label %forbody, label %afterfor
forbody: ; preds = %forcond
- %tmp2 = load i32* %i ; <i32> [#uses=1]
- %tmp3 = load <5 x i16>** %dst.addr ; <<5 x i16>*> [#uses=1]
- %arrayidx = getelementptr <5 x i16>* %tmp3, i32 %tmp2 ; <<5 x i16>*> [#uses=1]
- %tmp4 = load i32* %i ; <i32> [#uses=1]
- %tmp5 = load <5 x i16>** %src.addr ; <<5 x i16>*> [#uses=1]
- %arrayidx6 = getelementptr <5 x i16>* %tmp5, i32 %tmp4 ; <<5 x i16>*> [#uses=1]
- %tmp7 = load <5 x i16>* %arrayidx6 ; <<5 x i16>> [#uses=1]
+ %tmp2 = load i32, i32* %i ; <i32> [#uses=1]
+ %tmp3 = load <5 x i16>*, <5 x i16>** %dst.addr ; <<5 x i16>*> [#uses=1]
+ %arrayidx = getelementptr <5 x i16>, <5 x i16>* %tmp3, i32 %tmp2 ; <<5 x i16>*> [#uses=1]
+ %tmp4 = load i32, i32* %i ; <i32> [#uses=1]
+ %tmp5 = load <5 x i16>*, <5 x i16>** %src.addr ; <<5 x i16>*> [#uses=1]
+ %arrayidx6 = getelementptr <5 x i16>, <5 x i16>* %tmp5, i32 %tmp4 ; <<5 x i16>*> [#uses=1]
+ %tmp7 = load <5 x i16>, <5 x i16>* %arrayidx6 ; <<5 x i16>> [#uses=1]
%sub = sub <5 x i16> %tmp7, < i16 271, i16 271, i16 271, i16 271, i16 271 > ; <<5 x i16>> [#uses=1]
- %mul = mul <5 x i16> %sub, < i16 2, i16 2, i16 2, i16 2, i16 2 > ; <<5 x i16>> [#uses=1]
+ %mul = mul <5 x i16> %sub, < i16 2, i16 4, i16 2, i16 2, i16 2 > ; <<5 x i16>> [#uses=1]
store <5 x i16> %mul, <5 x i16>* %arrayidx
br label %forinc
forinc: ; preds = %forbody
- %tmp8 = load i32* %i ; <i32> [#uses=1]
+ %tmp8 = load i32, i32* %i ; <i32> [#uses=1]
%inc = add i32 %tmp8, 1 ; <i32> [#uses=1]
store i32 %inc, i32* %i
br label %forcond