Home | History | Annotate | Download | only in AMDGPU
      1 ; RUN: llc -march=amdgcn -mcpu=SI -verify-machineinstrs < %s | FileCheck -check-prefix=SI -check-prefix=BOTH %s
      2 ; RUN: llc -march=amdgcn -mcpu=tonga -verify-machineinstrs < %s | FileCheck -check-prefix=VI -check-prefix=BOTH %s
      3 
      4 ; BOTH-LABEL: {{^}}s_rotr_i64:
      5 ; BOTH-DAG: s_sub_i32
      6 ; BOTH-DAG: s_lshr_b64
      7 ; BOTH-DAG: s_lshl_b64
      8 ; BOTH: s_or_b64
      9 define void @s_rotr_i64(i64 addrspace(1)* %in, i64 %x, i64 %y) {
     10 entry:
     11   %tmp0 = sub i64 64, %y
     12   %tmp1 = shl i64 %x, %tmp0
     13   %tmp2 = lshr i64 %x, %y
     14   %tmp3 = or i64 %tmp1, %tmp2
     15   store i64 %tmp3, i64 addrspace(1)* %in
     16   ret void
     17 }
     18 
     19 ; BOTH-LABEL: {{^}}v_rotr_i64:
     20 ; BOTH-DAG: v_sub_i32
     21 ; SI-DAG: v_lshr_b64
     22 ; SI-DAG: v_lshl_b64
     23 ; VI-DAG: v_lshrrev_b64
     24 ; VI-DAG: v_lshlrev_b64
     25 ; BOTH: v_or_b32
     26 ; BOTH: v_or_b32
     27 define void @v_rotr_i64(i64 addrspace(1)* %in, i64 addrspace(1)* %xptr, i64 addrspace(1)* %yptr) {
     28 entry:
     29   %x = load i64, i64 addrspace(1)* %xptr, align 8
     30   %y = load i64, i64 addrspace(1)* %yptr, align 8
     31   %tmp0 = sub i64 64, %y
     32   %tmp1 = shl i64 %x, %tmp0
     33   %tmp2 = lshr i64 %x, %y
     34   %tmp3 = or i64 %tmp1, %tmp2
     35   store i64 %tmp3, i64 addrspace(1)* %in
     36   ret void
     37 }
     38 
     39 ; BOTH-LABEL: {{^}}s_rotr_v2i64:
     40 define void @s_rotr_v2i64(<2 x i64> addrspace(1)* %in, <2 x i64> %x, <2 x i64> %y) {
     41 entry:
     42   %tmp0 = sub <2 x i64> <i64 64, i64 64>, %y
     43   %tmp1 = shl <2 x i64> %x, %tmp0
     44   %tmp2 = lshr <2 x i64> %x, %y
     45   %tmp3 = or <2 x i64> %tmp1, %tmp2
     46   store <2 x i64> %tmp3, <2 x i64> addrspace(1)* %in
     47   ret void
     48 }
     49 
     50 ; BOTH-LABEL: {{^}}v_rotr_v2i64:
     51 define void @v_rotr_v2i64(<2 x i64> addrspace(1)* %in, <2 x i64> addrspace(1)* %xptr, <2 x i64> addrspace(1)* %yptr) {
     52 entry:
     53   %x = load <2 x i64>, <2 x i64> addrspace(1)* %xptr, align 8
     54   %y = load <2 x i64>, <2 x i64> addrspace(1)* %yptr, align 8
     55   %tmp0 = sub <2 x i64> <i64 64, i64 64>, %y
     56   %tmp1 = shl <2 x i64> %x, %tmp0
     57   %tmp2 = lshr <2 x i64> %x, %y
     58   %tmp3 = or <2 x i64> %tmp1, %tmp2
     59   store <2 x i64> %tmp3, <2 x i64> addrspace(1)* %in
     60   ret void
     61 }
     62