Home | History | Annotate | Download | only in AMDGPU
      1 ; RUN: llc -march=amdgcn -verify-machineinstrs < %s | FileCheck -check-prefix=SI -check-prefix=GCN -check-prefix=FUNC %s
      2 ; RUN: llc -march=amdgcn -mcpu=tonga -verify-machineinstrs < %s | FileCheck -check-prefix=VI -check-prefix=GCN -check-prefix=FUNC %s
      3 
      4 declare i64 @llvm.ctpop.i64(i64) nounwind readnone
      5 declare <2 x i64> @llvm.ctpop.v2i64(<2 x i64>) nounwind readnone
      6 declare <4 x i64> @llvm.ctpop.v4i64(<4 x i64>) nounwind readnone
      7 declare <8 x i64> @llvm.ctpop.v8i64(<8 x i64>) nounwind readnone
      8 declare <16 x i64> @llvm.ctpop.v16i64(<16 x i64>) nounwind readnone
      9 
     10 declare i65 @llvm.ctpop.i65(i65) nounwind readnone
     11 declare i128 @llvm.ctpop.i128(i128) nounwind readnone
     12 
     13 ; FUNC-LABEL: {{^}}s_ctpop_i64:
     14 ; SI: s_load_dwordx2 [[SVAL:s\[[0-9]+:[0-9]+\]]], s{{\[[0-9]+:[0-9]+\]}}, 0xb
     15 ; VI: s_load_dwordx2 [[SVAL:s\[[0-9]+:[0-9]+\]]], s{{\[[0-9]+:[0-9]+\]}}, 0x2c
     16 ; GCN: s_bcnt1_i32_b64 [[SRESULT:s[0-9]+]], [[SVAL]]
     17 ; GCN: v_mov_b32_e32 [[VRESULT:v[0-9]+]], [[SRESULT]]
     18 ; GCN: buffer_store_dword [[VRESULT]],
     19 ; GCN: s_endpgm
     20 define void @s_ctpop_i64(i32 addrspace(1)* noalias %out, i64 %val) nounwind {
     21   %ctpop = call i64 @llvm.ctpop.i64(i64 %val) nounwind readnone
     22   %truncctpop = trunc i64 %ctpop to i32
     23   store i32 %truncctpop, i32 addrspace(1)* %out, align 4
     24   ret void
     25 }
     26 
     27 ; FUNC-LABEL: {{^}}v_ctpop_i64:
     28 ; GCN: buffer_load_dwordx2 v{{\[}}[[LOVAL:[0-9]+]]:[[HIVAL:[0-9]+]]{{\]}},
     29 ; GCN: v_bcnt_u32_b32_e64 [[MIDRESULT:v[0-9]+]], v[[LOVAL]], 0
     30 ; SI-NEXT: v_bcnt_u32_b32_e32 [[RESULT:v[0-9]+]], v[[HIVAL]], [[MIDRESULT]]
     31 ; VI-NEXT: v_bcnt_u32_b32_e64 [[RESULT:v[0-9]+]], v[[HIVAL]], [[MIDRESULT]]
     32 ; GCN: buffer_store_dword [[RESULT]],
     33 ; GCN: s_endpgm
     34 define void @v_ctpop_i64(i32 addrspace(1)* noalias %out, i64 addrspace(1)* noalias %in) nounwind {
     35   %val = load i64, i64 addrspace(1)* %in, align 8
     36   %ctpop = call i64 @llvm.ctpop.i64(i64 %val) nounwind readnone
     37   %truncctpop = trunc i64 %ctpop to i32
     38   store i32 %truncctpop, i32 addrspace(1)* %out, align 4
     39   ret void
     40 }
     41 
     42 ; FIXME: or 0 should be replaxed with copy
     43 ; FUNC-LABEL: {{^}}v_ctpop_i64_user:
     44 ; GCN: buffer_load_dwordx2 v{{\[}}[[LOVAL:[0-9]+]]:[[HIVAL:[0-9]+]]{{\]}},
     45 ; GCN: v_bcnt_u32_b32_e64 [[MIDRESULT:v[0-9]+]], v[[LOVAL]], 0
     46 ; SI-NEXT: v_bcnt_u32_b32_e32 [[RESULT:v[0-9]+]], v[[HIVAL]], [[MIDRESULT]]
     47 ; VI-NEXT: v_bcnt_u32_b32_e64 [[RESULT:v[0-9]+]], v[[HIVAL]], [[MIDRESULT]]
     48 ; GCN-DAG: v_or_b32_e32 v[[RESULT_LO:[0-9]+]], s{{[0-9]+}}, [[RESULT]]
     49 ; GCN-DAG: v_or_b32_e64 v[[RESULT_HI:[0-9]+]], 0, s{{[0-9]+}}
     50 ; GCN: buffer_store_dwordx2 v{{\[}}[[RESULT_LO]]:[[RESULT_HI]]{{\]}}
     51 ; GCN: s_endpgm
     52 define void @v_ctpop_i64_user(i64 addrspace(1)* noalias %out, i64 addrspace(1)* noalias %in, i64 %s.val) nounwind {
     53   %val = load i64, i64 addrspace(1)* %in, align 8
     54   %ctpop = call i64 @llvm.ctpop.i64(i64 %val) nounwind readnone
     55   %or = or i64 %ctpop, %s.val
     56   store i64 %or, i64 addrspace(1)* %out
     57   ret void
     58 }
     59 
     60 ; FUNC-LABEL: {{^}}s_ctpop_v2i64:
     61 ; GCN: s_bcnt1_i32_b64
     62 ; GCN: s_bcnt1_i32_b64
     63 ; GCN: s_endpgm
     64 define void @s_ctpop_v2i64(<2 x i32> addrspace(1)* noalias %out, <2 x i64> %val) nounwind {
     65   %ctpop = call <2 x i64> @llvm.ctpop.v2i64(<2 x i64> %val) nounwind readnone
     66   %truncctpop = trunc <2 x i64> %ctpop to <2 x i32>
     67   store <2 x i32> %truncctpop, <2 x i32> addrspace(1)* %out, align 8
     68   ret void
     69 }
     70 
     71 ; FUNC-LABEL: {{^}}s_ctpop_v4i64:
     72 ; GCN: s_bcnt1_i32_b64
     73 ; GCN: s_bcnt1_i32_b64
     74 ; GCN: s_bcnt1_i32_b64
     75 ; GCN: s_bcnt1_i32_b64
     76 ; GCN: s_endpgm
     77 define void @s_ctpop_v4i64(<4 x i32> addrspace(1)* noalias %out, <4 x i64> %val) nounwind {
     78   %ctpop = call <4 x i64> @llvm.ctpop.v4i64(<4 x i64> %val) nounwind readnone
     79   %truncctpop = trunc <4 x i64> %ctpop to <4 x i32>
     80   store <4 x i32> %truncctpop, <4 x i32> addrspace(1)* %out, align 16
     81   ret void
     82 }
     83 
     84 ; FUNC-LABEL: {{^}}v_ctpop_v2i64:
     85 ; GCN: v_bcnt_u32_b32
     86 ; GCN: v_bcnt_u32_b32
     87 ; GCN: v_bcnt_u32_b32
     88 ; GCN: v_bcnt_u32_b32
     89 ; GCN: s_endpgm
     90 define void @v_ctpop_v2i64(<2 x i32> addrspace(1)* noalias %out, <2 x i64> addrspace(1)* noalias %in) nounwind {
     91   %val = load <2 x i64>, <2 x i64> addrspace(1)* %in, align 16
     92   %ctpop = call <2 x i64> @llvm.ctpop.v2i64(<2 x i64> %val) nounwind readnone
     93   %truncctpop = trunc <2 x i64> %ctpop to <2 x i32>
     94   store <2 x i32> %truncctpop, <2 x i32> addrspace(1)* %out, align 8
     95   ret void
     96 }
     97 
     98 ; FUNC-LABEL: {{^}}v_ctpop_v4i64:
     99 ; GCN: v_bcnt_u32_b32
    100 ; GCN: v_bcnt_u32_b32
    101 ; GCN: v_bcnt_u32_b32
    102 ; GCN: v_bcnt_u32_b32
    103 ; GCN: v_bcnt_u32_b32
    104 ; GCN: v_bcnt_u32_b32
    105 ; GCN: v_bcnt_u32_b32
    106 ; GCN: v_bcnt_u32_b32
    107 ; GCN: s_endpgm
    108 define void @v_ctpop_v4i64(<4 x i32> addrspace(1)* noalias %out, <4 x i64> addrspace(1)* noalias %in) nounwind {
    109   %val = load <4 x i64>, <4 x i64> addrspace(1)* %in, align 32
    110   %ctpop = call <4 x i64> @llvm.ctpop.v4i64(<4 x i64> %val) nounwind readnone
    111   %truncctpop = trunc <4 x i64> %ctpop to <4 x i32>
    112   store <4 x i32> %truncctpop, <4 x i32> addrspace(1)* %out, align 16
    113   ret void
    114 }
    115 
    116 ; FUNC-LABEL: {{^}}ctpop_i64_in_br:
    117 ; SI-DAG: s_load_dwordx2 s{{\[}}[[LOVAL:[0-9]+]]:[[HIVAL:[0-9]+]]{{\]}}, s[{{[0-9]+:[0-9]+}}], 0xd
    118 ; VI-DAG: s_load_dwordx2 s{{\[}}[[LOVAL:[0-9]+]]:[[HIVAL:[0-9]+]]{{\]}}, s[{{[0-9]+:[0-9]+}}], 0x34
    119 ; GCN-DAG: s_bcnt1_i32_b64 [[RESULT:s[0-9]+]], {{s\[}}[[LOVAL]]:[[HIVAL]]{{\]}}
    120 ; GCN-DAG: s_mov_b32 [[ZERO:s[0-9]+]], 0
    121 ; GCN-DAG: v_mov_b32_e32 v[[VLO:[0-9]+]], [[RESULT]]
    122 ; GCN-DAG: v_mov_b32_e32 v[[VHI:[0-9]+]], [[ZERO]]
    123 ; GCN: buffer_store_dwordx2 {{v\[}}[[VLO]]:[[VHI]]{{\]}}
    124 ; GCN: s_endpgm
    125 define void @ctpop_i64_in_br(i64 addrspace(1)* %out, i64 addrspace(1)* %in, i64 %ctpop_arg, i32 %cond) {
    126 entry:
    127   %tmp0 = icmp eq i32 %cond, 0
    128   br i1 %tmp0, label %if, label %else
    129 
    130 if:
    131   %tmp2 = call i64 @llvm.ctpop.i64(i64 %ctpop_arg)
    132   br label %endif
    133 
    134 else:
    135   %tmp3 = getelementptr i64, i64 addrspace(1)* %in, i32 1
    136   %tmp4 = load i64, i64 addrspace(1)* %tmp3
    137   br label %endif
    138 
    139 endif:
    140   %tmp5 = phi i64 [%tmp2, %if], [%tmp4, %else]
    141   store i64 %tmp5, i64 addrspace(1)* %out
    142   ret void
    143 }
    144 
    145 ; FUNC-LABEL: {{^}}s_ctpop_i128:
    146 ; GCN: s_bcnt1_i32_b64 [[SRESULT0:s[0-9]+]],
    147 ; GCN: s_bcnt1_i32_b64 [[SRESULT1:s[0-9]+]],
    148 ; GCN: s_add_i32 s{{[0-9]+}}, [[SRESULT1]], [[SRESULT0]]
    149 ; GCN: s_endpgm
    150 define void @s_ctpop_i128(i32 addrspace(1)* noalias %out, i128 %val) nounwind {
    151   %ctpop = call i128 @llvm.ctpop.i128(i128 %val) nounwind readnone
    152   %truncctpop = trunc i128 %ctpop to i32
    153   store i32 %truncctpop, i32 addrspace(1)* %out, align 4
    154   ret void
    155 }
    156 
    157 ; FUNC-LABEL: {{^}}s_ctpop_i65:
    158 ; GCN: s_and_b32
    159 ; GCN: s_bcnt1_i32_b64 [[REG0:s[0-9]+]],
    160 ; GCN: s_bcnt1_i32_b64 [[REG1:s[0-9]+]],
    161 ; GCN: s_add_i32 {{s[0-9]+}}, [[REG0]], [[REG1]]
    162 ; GCN: s_endpgm
    163 define void @s_ctpop_i65(i32 addrspace(1)* noalias %out, i65 %val) nounwind {
    164   %ctpop = call i65 @llvm.ctpop.i65(i65 %val) nounwind readnone
    165   %truncctpop = trunc i65 %ctpop to i32
    166   store i32 %truncctpop, i32 addrspace(1)* %out, align 4
    167   ret void
    168 }
    169 
    170 ; FIXME: Should not have extra add
    171 
    172 ; FUNC-LABEL: {{^}}v_ctpop_i128:
    173 ; GCN: buffer_load_dwordx4 v{{\[}}[[VAL0:[0-9]+]]:[[VAL3:[0-9]+]]{{\]}}, off, s{{\[[0-9]+:[0-9]+\]}}, 0{{$}}
    174 
    175 ; GCN-DAG: v_bcnt_u32_b32_e64 [[MIDRESULT0:v[0-9]+]], v{{[0-9]+}}, 0
    176 ; GCN-DAG: v_bcnt_u32_b32{{_e32|_e64}} [[MIDRESULT1:v[0-9]+]], v[[VAL3]], [[MIDRESULT0]]
    177 
    178 ; GCN-DAG: v_bcnt_u32_b32_e64 [[MIDRESULT2:v[0-9]+]], v[[VAL0]], 0
    179 ; GCN-DAG: v_bcnt_u32_b32{{_e32|_e64}} [[MIDRESULT3:v[0-9]+]], v{{[0-9]+}}, [[MIDRESULT2]]
    180 
    181 ; GCN: v_add_i32_e32 [[RESULT:v[0-9]+]], vcc, [[MIDRESULT1]], [[MIDRESULT2]]
    182 
    183 ; GCN: buffer_store_dword [[RESULT]],
    184 ; GCN: s_endpgm
    185 define void @v_ctpop_i128(i32 addrspace(1)* noalias %out, i128 addrspace(1)* noalias %in) nounwind {
    186   %val = load i128, i128 addrspace(1)* %in, align 8
    187   %ctpop = call i128 @llvm.ctpop.i128(i128 %val) nounwind readnone
    188   %truncctpop = trunc i128 %ctpop to i32
    189   store i32 %truncctpop, i32 addrspace(1)* %out, align 4
    190   ret void
    191 }
    192