Home | History | Annotate | Download | only in X86
      1 ; RUN: llc < %s -march=x86 -mattr=+sse2 | FileCheck %s
      2 
      3 ; When loading the shift amount from memory, avoid generating the splat.
      4 
      5 define void @shift5a(<4 x i32> %val, <4 x i32>* %dst, i32* %pamt) nounwind {
      6 entry:
      7 ; CHECK-LABEL: shift5a:
      8 ; CHECK: movd
      9 ; CHECK: pslld
     10   %amt = load i32, i32* %pamt 
     11   %tmp0 = insertelement <4 x i32> undef, i32 %amt, i32 0
     12   %shamt = shufflevector <4 x i32> %tmp0, <4 x i32> undef, <4 x i32> zeroinitializer 
     13   %shl = shl <4 x i32> %val, %shamt
     14   store <4 x i32> %shl, <4 x i32>* %dst
     15   ret void
     16 }
     17 
     18 
     19 define void @shift5b(<4 x i32> %val, <4 x i32>* %dst, i32* %pamt) nounwind {
     20 entry:
     21 ; CHECK-LABEL: shift5b:
     22 ; CHECK: movd
     23 ; CHECK: psrad
     24   %amt = load i32, i32* %pamt 
     25   %tmp0 = insertelement <4 x i32> undef, i32 %amt, i32 0
     26   %shamt = shufflevector <4 x i32> %tmp0, <4 x i32> undef, <4 x i32> zeroinitializer 
     27   %shr = ashr <4 x i32> %val, %shamt
     28   store <4 x i32> %shr, <4 x i32>* %dst
     29   ret void
     30 }
     31 
     32 
     33 define void @shift5c(<4 x i32> %val, <4 x i32>* %dst, i32 %amt) nounwind {
     34 entry:
     35 ; CHECK-LABEL: shift5c:
     36 ; CHECK: movd
     37 ; CHECK: pslld
     38   %tmp0 = insertelement <4 x i32> undef, i32 %amt, i32 0
     39   %shamt = shufflevector <4 x i32> %tmp0, <4 x i32> undef, <4 x i32> zeroinitializer
     40   %shl = shl <4 x i32> %val, %shamt
     41   store <4 x i32> %shl, <4 x i32>* %dst
     42   ret void
     43 }
     44 
     45 
     46 define void @shift5d(<4 x i32> %val, <4 x i32>* %dst, i32 %amt) nounwind {
     47 entry:
     48 ; CHECK-LABEL: shift5d:
     49 ; CHECK: movd
     50 ; CHECK: psrad
     51   %tmp0 = insertelement <4 x i32> undef, i32 %amt, i32 0
     52   %shamt = shufflevector <4 x i32> %tmp0, <4 x i32> undef, <4 x i32> zeroinitializer
     53   %shr = ashr <4 x i32> %val, %shamt
     54   store <4 x i32> %shr, <4 x i32>* %dst
     55   ret void
     56 }
     57