X-Git-Url: http://plrg.eecs.uci.edu/git/?p=oota-llvm.git;a=blobdiff_plain;f=lib%2FTarget%2FX86%2FX86InstrAVX512.td;h=62f28b79ecdb1ec6e07069243eb4b20ef7cec548;hp=ca124ed253c0f0dea56506513dc8cbbe65d22422;hb=0e0cc9459a89ac4f5e4cb5003d223a974c2d4d94;hpb=cc4e8d7e1d66cddd7daa443aea22e83964c6ae1d diff --git a/lib/Target/X86/X86InstrAVX512.td b/lib/Target/X86/X86InstrAVX512.td index ca124ed253c..62f28b79ecd 100644 --- a/lib/Target/X86/X86InstrAVX512.td +++ b/lib/Target/X86/X86InstrAVX512.td @@ -276,6 +276,22 @@ multiclass AVX512_maskable_3src O, Format F, X86VectorVTInfo _, OpcodeStr, AttSrcAsm, IntelSrcAsm, RHS, (vselect _.KRCWM:$mask, RHS, _.RC:$src1)>; +// Similar to AVX512_maskable_3rc but in this case the input VT for the tied +// operand differs from the output VT. This requires a bitconvert on +// the preserved vector going into the vselect. +multiclass AVX512_maskable_3src_cast O, Format F, X86VectorVTInfo OutVT, + X86VectorVTInfo InVT, + dag Outs, dag NonTiedIns, string OpcodeStr, + string AttSrcAsm, string IntelSrcAsm, + dag RHS> : + AVX512_maskable_common; + multiclass AVX512_maskable_3src_scalar O, Format F, X86VectorVTInfo _, dag Outs, dag NonTiedIns, string OpcodeStr, string AttSrcAsm, string IntelSrcAsm, @@ -1047,6 +1063,46 @@ defm VBROADCASTF32X8 : avx512_subvec_broadcast_rm<0x1b, "vbroadcastf32x8", EVEX_V512, EVEX_CD8<32, CD8VT8>; } +multiclass avx512_broadcast_32x2 opc, string OpcodeStr, + X86VectorVTInfo _Dst, X86VectorVTInfo _Src, + SDNode OpNode = X86SubVBroadcast> { + + defm r : AVX512_maskable, + T8PD, EVEX; + let mayLoad = 1 in + defm m : AVX512_maskable, + T8PD, EVEX, EVEX_CD8<_Src.EltSize, CD8VT2>; +} + +multiclass avx512_common_broadcast_32x2 opc, string OpcodeStr, + AVX512VLVectorVTInfo _> { + let Predicates = [HasDQI] in + defm Z : avx512_broadcast_32x2, + EVEX_V512; + let Predicates = [HasDQI, HasVLX] in + defm Z256 : avx512_broadcast_32x2, + EVEX_V256; +} + +multiclass avx512_common_broadcast_i32x2 opc, string OpcodeStr, + AVX512VLVectorVTInfo _> : + avx512_common_broadcast_32x2 { + + let Predicates = [HasDQI, HasVLX] in + defm Z128 : avx512_broadcast_32x2, EVEX_V128; +} + +defm VPBROADCASTI32X2 : avx512_common_broadcast_i32x2<0x59, "vbroadcasti32x2", + avx512vl_i32_info>; +defm VPBROADCASTF32X2 : avx512_common_broadcast_32x2<0x19, "vbroadcastf32x2", + avx512vl_f32_info>; + def : Pat<(v16f32 (X86VBroadcast (v16f32 VR512:$src))), (VBROADCASTSSZr (EXTRACT_SUBREG (v16f32 VR512:$src), sub_xmm))>; def : Pat<(v16f32 (X86VBroadcast (v8f32 VR256X:$src))), @@ -1073,114 +1129,178 @@ def : Pat<(v8f64 (X86VBroadcast FR64X:$src)), //===----------------------------------------------------------------------===// // AVX-512 BROADCAST MASK TO VECTOR REGISTER //--- - -multiclass avx512_mask_broadcast opc, string OpcodeStr, - RegisterClass KRC> { -let Predicates = [HasCDI] in -def Zrr : AVX512XS8I opc, string OpcodeStr, + X86VectorVTInfo _, RegisterClass KRC> { + def rr : AVX512XS8I, EVEX, EVEX_V512; - -let Predicates = [HasCDI, HasVLX] in { -def Z128rr : AVX512XS8I, EVEX, EVEX_V128; -def Z256rr : AVX512XS8I, EVEX, EVEX_V256; + [(set _.RC:$dst, (_.VT (X86VBroadcastm KRC:$src)))]>, EVEX; } + +multiclass avx512_mask_broadcast opc, string OpcodeStr, + AVX512VLVectorVTInfo VTInfo, RegisterClass KRC> { + let Predicates = [HasCDI] in + defm Z : avx512_mask_broadcastm, EVEX_V512; + let Predicates = [HasCDI, HasVLX] in { + defm Z256 : avx512_mask_broadcastm, EVEX_V256; + defm Z128 : avx512_mask_broadcastm, EVEX_V128; + } } -let Predicates = [HasCDI] in { defm VPBROADCASTMW2D : avx512_mask_broadcast<0x3A, "vpbroadcastmw2d", - VK16>; + avx512vl_i32_info, VK16>; defm VPBROADCASTMB2Q : avx512_mask_broadcast<0x2A, "vpbroadcastmb2q", - VK8>, VEX_W; -} + avx512vl_i64_info, VK8>, VEX_W; //===----------------------------------------------------------------------===// -// -- VPERM2I - 3 source operands form -- -multiclass avx512_perm_3src opc, string OpcodeStr, - SDNode OpNode, X86VectorVTInfo _> { +// -- VPERMI2 - 3 source operands form -- +multiclass avx512_perm_i opc, string OpcodeStr, + X86VectorVTInfo _, X86VectorVTInfo IdxVT> { let Constraints = "$src1 = $dst" in { - defm rr: AVX512_maskable_3src, EVEX_4V, + (_.VT (X86VPermi2X IdxVT.RC:$src1, _.RC:$src2, _.RC:$src3))>, EVEX_4V, AVX5128IBase; let mayLoad = 1 in - defm rm: AVX512_maskable_3src, EVEX_4V, AVX5128IBase; } } -multiclass avx512_perm_3src_mb opc, string OpcodeStr, - SDNode OpNode, X86VectorVTInfo _> { +multiclass avx512_perm_i_mb opc, string OpcodeStr, + X86VectorVTInfo _, X86VectorVTInfo IdxVT> { let mayLoad = 1, Constraints = "$src1 = $dst" in - defm rmb: AVX512_maskable_3src, AVX5128IBase, EVEX_4V, EVEX_B; } -multiclass avx512_perm_3src_sizes opc, string OpcodeStr, - SDNode OpNode, AVX512VLVectorVTInfo VTInfo> { - let Predicates = [HasAVX512] in - defm NAME: avx512_perm_3src, - avx512_perm_3src_mb, EVEX_V512; +multiclass avx512_perm_i_sizes opc, string OpcodeStr, + AVX512VLVectorVTInfo VTInfo, + AVX512VLVectorVTInfo ShuffleMask> { + defm NAME: avx512_perm_i, + avx512_perm_i_mb, EVEX_V512; let Predicates = [HasVLX] in { - defm NAME#128: avx512_perm_3src, - avx512_perm_3src_mb, - EVEX_V128; - defm NAME#256: avx512_perm_3src, - avx512_perm_3src_mb, - EVEX_V256; + defm NAME#128: avx512_perm_i, + avx512_perm_i_mb, EVEX_V128; + defm NAME#256: avx512_perm_i, + avx512_perm_i_mb, EVEX_V256; } } -multiclass avx512_perm_3src_sizes_w opc, string OpcodeStr, - SDNode OpNode, AVX512VLVectorVTInfo VTInfo> { + +multiclass avx512_perm_i_sizes_w opc, string OpcodeStr, + AVX512VLVectorVTInfo VTInfo, + AVX512VLVectorVTInfo Idx> { let Predicates = [HasBWI] in - defm NAME: avx512_perm_3src, - avx512_perm_3src_mb, - EVEX_V512; + defm NAME: avx512_perm_i, EVEX_V512; let Predicates = [HasBWI, HasVLX] in { - defm NAME#128: avx512_perm_3src, - avx512_perm_3src_mb, - EVEX_V128; - defm NAME#256: avx512_perm_3src, - avx512_perm_3src_mb, - EVEX_V256; - } -} -defm VPERMI2D : avx512_perm_3src_sizes<0x76, "vpermi2d", X86VPermiv3, - avx512vl_i32_info>, EVEX_CD8<32, CD8VF>; -defm VPERMI2Q : avx512_perm_3src_sizes<0x76, "vpermi2q", X86VPermiv3, - avx512vl_i64_info>, VEX_W, EVEX_CD8<64, CD8VF>; -defm VPERMI2PS : avx512_perm_3src_sizes<0x77, "vpermi2ps", X86VPermiv3, - avx512vl_f32_info>, EVEX_CD8<32, CD8VF>; -defm VPERMI2PD : avx512_perm_3src_sizes<0x77, "vpermi2pd", X86VPermiv3, - avx512vl_f64_info>, VEX_W, EVEX_CD8<64, CD8VF>; - -defm VPERMT2D : avx512_perm_3src_sizes<0x7E, "vpermt2d", X86VPermv3, - avx512vl_i32_info>, EVEX_CD8<32, CD8VF>; -defm VPERMT2Q : avx512_perm_3src_sizes<0x7E, "vpermt2q", X86VPermv3, - avx512vl_i64_info>, VEX_W, EVEX_CD8<64, CD8VF>; -defm VPERMT2PS : avx512_perm_3src_sizes<0x7F, "vpermt2ps", X86VPermv3, - avx512vl_f32_info>, EVEX_CD8<32, CD8VF>; -defm VPERMT2PD : avx512_perm_3src_sizes<0x7F, "vpermt2pd", X86VPermv3, - avx512vl_f64_info>, VEX_W, EVEX_CD8<64, CD8VF>; - -defm VPERMT2W : avx512_perm_3src_sizes_w<0x7D, "vpermt2w", X86VPermv3, - avx512vl_i16_info>, VEX_W, EVEX_CD8<16, CD8VF>; -defm VPERMI2W : avx512_perm_3src_sizes_w<0x75, "vpermi2w", X86VPermiv3, - avx512vl_i16_info>, VEX_W, EVEX_CD8<16, CD8VF>; + defm NAME#128: avx512_perm_i, EVEX_V128; + defm NAME#256: avx512_perm_i, EVEX_V256; + } +} + +defm VPERMI2D : avx512_perm_i_sizes<0x76, "vpermi2d", + avx512vl_i32_info, avx512vl_i32_info>, EVEX_CD8<32, CD8VF>; +defm VPERMI2Q : avx512_perm_i_sizes<0x76, "vpermi2q", + avx512vl_i64_info, avx512vl_i64_info>, VEX_W, EVEX_CD8<64, CD8VF>; +defm VPERMI2W : avx512_perm_i_sizes_w<0x75, "vpermi2w", + avx512vl_i16_info, avx512vl_i16_info>, VEX_W, EVEX_CD8<16, CD8VF>; +defm VPERMI2PS : avx512_perm_i_sizes<0x77, "vpermi2ps", + avx512vl_f32_info, avx512vl_i32_info>, EVEX_CD8<32, CD8VF>; +defm VPERMI2PD : avx512_perm_i_sizes<0x77, "vpermi2pd", + avx512vl_f64_info, avx512vl_i64_info>, VEX_W, EVEX_CD8<64, CD8VF>; + +// VPERMT2 +multiclass avx512_perm_t opc, string OpcodeStr, + X86VectorVTInfo _, X86VectorVTInfo IdxVT> { +let Constraints = "$src1 = $dst" in { + defm rr: AVX512_maskable_3src, EVEX_4V, + AVX5128IBase; + + let mayLoad = 1 in + defm rm: AVX512_maskable_3src, + EVEX_4V, AVX5128IBase; + } +} +multiclass avx512_perm_t_mb opc, string OpcodeStr, + X86VectorVTInfo _, X86VectorVTInfo IdxVT> { + let mayLoad = 1, Constraints = "$src1 = $dst" in + defm rmb: AVX512_maskable_3src, + AVX5128IBase, EVEX_4V, EVEX_B; +} + +multiclass avx512_perm_t_sizes opc, string OpcodeStr, + AVX512VLVectorVTInfo VTInfo, + AVX512VLVectorVTInfo ShuffleMask> { + defm NAME: avx512_perm_t, + avx512_perm_t_mb, EVEX_V512; + let Predicates = [HasVLX] in { + defm NAME#128: avx512_perm_t, + avx512_perm_t_mb, EVEX_V128; + defm NAME#256: avx512_perm_t, + avx512_perm_t_mb, EVEX_V256; + } +} + +multiclass avx512_perm_t_sizes_w opc, string OpcodeStr, + AVX512VLVectorVTInfo VTInfo, + AVX512VLVectorVTInfo Idx> { + let Predicates = [HasBWI] in + defm NAME: avx512_perm_t, EVEX_V512; + let Predicates = [HasBWI, HasVLX] in { + defm NAME#128: avx512_perm_t, EVEX_V128; + defm NAME#256: avx512_perm_t, EVEX_V256; + } +} + +defm VPERMT2D : avx512_perm_t_sizes<0x7E, "vpermt2d", + avx512vl_i32_info, avx512vl_i32_info>, EVEX_CD8<32, CD8VF>; +defm VPERMT2Q : avx512_perm_t_sizes<0x7E, "vpermt2q", + avx512vl_i64_info, avx512vl_i64_info>, VEX_W, EVEX_CD8<64, CD8VF>; +defm VPERMT2W : avx512_perm_t_sizes_w<0x7D, "vpermt2w", + avx512vl_i16_info, avx512vl_i16_info>, VEX_W, EVEX_CD8<16, CD8VF>; +defm VPERMT2PS : avx512_perm_t_sizes<0x7F, "vpermt2ps", + avx512vl_f32_info, avx512vl_i32_info>, EVEX_CD8<32, CD8VF>; +defm VPERMT2PD : avx512_perm_t_sizes<0x7F, "vpermt2pd", + avx512vl_f64_info, avx512vl_i64_info>, VEX_W, EVEX_CD8<64, CD8VF>; //===----------------------------------------------------------------------===// // AVX-512 - BLEND using mask @@ -2276,16 +2396,6 @@ defm KUNPCKBW : avx512_mask_unpck<"bw", VK16, v16i1, VK8, HasAVX512>, PD; defm KUNPCKWD : avx512_mask_unpck<"wd", VK32, v32i1, VK16, HasBWI>, PS; defm KUNPCKDQ : avx512_mask_unpck<"dq", VK64, v64i1, VK32, HasBWI>, PS, VEX_W; -multiclass avx512_mask_unpck_int { - let Predicates = [HasAVX512] in - def : Pat<(!cast("int_x86_avx512_"##IntName##"_bw") - (i16 GR16:$src1), (i16 GR16:$src2)), - (COPY_TO_REGCLASS (!cast(InstName##"BWrr") - (v16i1 (COPY_TO_REGCLASS GR16:$src1, VK16)), - (v16i1 (COPY_TO_REGCLASS GR16:$src2, VK16))), GR16)>; -} -defm : avx512_mask_unpck_int<"kunpck", "KUNPCK">; - // Mask bit testing multiclass avx512_mask_testop opc, string OpcodeStr, RegisterClass KRC, SDNode OpNode, Predicate prd> { @@ -2376,24 +2486,48 @@ def : Pat<(v16i1 (insert_subvector undef, (v8i1 VK8:$src), (iPTR 0))), def : Pat<(v8i1 (extract_subvector (v16i1 VK16:$src), (iPTR 8))), (v8i1 (COPY_TO_REGCLASS (KSHIFTRWri VK16:$src, (i8 8)), VK8))>; +def : Pat<(v16i1 (extract_subvector (v32i1 VK32:$src), (iPTR 0))), + (v16i1 (COPY_TO_REGCLASS VK32:$src, VK16))>; + def : Pat<(v32i1 (extract_subvector (v64i1 VK64:$src), (iPTR 0))), (v32i1 (COPY_TO_REGCLASS VK64:$src, VK32))>; def : Pat<(v32i1 (extract_subvector (v64i1 VK64:$src), (iPTR 32))), (v32i1 (COPY_TO_REGCLASS (KSHIFTRQri VK64:$src, (i8 32)), VK32))>; -let Predicates = [HasVLX] in { - def : Pat<(v8i1 (insert_subvector undef, (v4i1 VK4:$src), (iPTR 0))), - (v8i1 (COPY_TO_REGCLASS VK4:$src, VK8))>; - def : Pat<(v8i1 (insert_subvector undef, (v2i1 VK2:$src), (iPTR 0))), - (v8i1 (COPY_TO_REGCLASS VK2:$src, VK8))>; - def : Pat<(v4i1 (insert_subvector undef, (v2i1 VK2:$src), (iPTR 0))), - (v4i1 (COPY_TO_REGCLASS VK2:$src, VK4))>; - def : Pat<(v4i1 (extract_subvector (v8i1 VK8:$src), (iPTR 0))), - (v4i1 (COPY_TO_REGCLASS VK8:$src, VK4))>; - def : Pat<(v2i1 (extract_subvector (v8i1 VK8:$src), (iPTR 0))), - (v2i1 (COPY_TO_REGCLASS VK8:$src, VK2))>; -} +def : Pat<(v4i1 (extract_subvector (v8i1 VK8:$src), (iPTR 0))), + (v4i1 (COPY_TO_REGCLASS VK8:$src, VK4))>; +def : Pat<(v2i1 (extract_subvector (v8i1 VK8:$src), (iPTR 0))), + (v2i1 (COPY_TO_REGCLASS VK8:$src, VK2))>; + +def : Pat<(v4i1 (insert_subvector undef, (v2i1 VK2:$src), (iPTR 0))), + (v4i1 (COPY_TO_REGCLASS VK2:$src, VK4))>; + +def : Pat<(v8i1 (insert_subvector undef, (v4i1 VK4:$src), (iPTR 0))), + (v8i1 (COPY_TO_REGCLASS VK4:$src, VK8))>; +def : Pat<(v8i1 (insert_subvector undef, (v2i1 VK2:$src), (iPTR 0))), + (v8i1 (COPY_TO_REGCLASS VK2:$src, VK8))>; + +def : Pat<(v32i1 (insert_subvector undef, VK2:$src, (iPTR 0))), + (v32i1 (COPY_TO_REGCLASS VK2:$src, VK32))>; +def : Pat<(v32i1 (insert_subvector undef, VK4:$src, (iPTR 0))), + (v32i1 (COPY_TO_REGCLASS VK4:$src, VK32))>; +def : Pat<(v32i1 (insert_subvector undef, VK8:$src, (iPTR 0))), + (v32i1 (COPY_TO_REGCLASS VK8:$src, VK32))>; +def : Pat<(v32i1 (insert_subvector undef, VK16:$src, (iPTR 0))), + (v32i1 (COPY_TO_REGCLASS VK16:$src, VK32))>; + +def : Pat<(v64i1 (insert_subvector undef, VK2:$src, (iPTR 0))), + (v64i1 (COPY_TO_REGCLASS VK2:$src, VK64))>; +def : Pat<(v64i1 (insert_subvector undef, VK4:$src, (iPTR 0))), + (v64i1 (COPY_TO_REGCLASS VK4:$src, VK64))>; +def : Pat<(v64i1 (insert_subvector undef, VK8:$src, (iPTR 0))), + (v64i1 (COPY_TO_REGCLASS VK8:$src, VK64))>; +def : Pat<(v64i1 (insert_subvector undef, VK16:$src, (iPTR 0))), + (v64i1 (COPY_TO_REGCLASS VK16:$src, VK64))>; +def : Pat<(v64i1 (insert_subvector undef, VK32:$src, (iPTR 0))), + (v64i1 (COPY_TO_REGCLASS VK32:$src, VK64))>; + def : Pat<(v8i1 (X86vshli VK8:$src, (i8 imm:$imm))), (v8i1 (COPY_TO_REGCLASS @@ -2513,23 +2647,21 @@ multiclass avx512_load_vl opc, string OpcodeStr, multiclass avx512_store opc, string OpcodeStr, X86VectorVTInfo _, PatFrag st_frag, PatFrag mstore> { - let isCodeGenOnly = 1, ForceDisassemble = 1, hasSideEffects = 0 in { - def rr_alt : AVX512PI, EVEX; - let Constraints = "$src1 = $dst" in - def rrk_alt : AVX512PI, EVEX; + def rrk_REV : AVX512PI, EVEX, EVEX_K; - def rrkz_alt : AVX512PI, EVEX, EVEX_KZ; - } + let mayStore = 1 in { def mr : AVX512PI; -let Predicates = [HasAVX512, NoVLX] in { -def: Pat<(X86mstore addr:$ptr, VK8WM:$mask, (v8f32 VR256:$src)), - (VMOVUPSZmrk addr:$ptr, - (v16i1 (COPY_TO_REGCLASS VK8WM:$mask, VK16WM)), - (INSERT_SUBREG (v16f32 (IMPLICIT_DEF)), VR256:$src, sub_ymm))>; - -def: Pat<(v8f32 (masked_load addr:$ptr, VK8WM:$mask, undef)), - (v8f32 (EXTRACT_SUBREG (v16f32 (VMOVUPSZrmkz - (v16i1 (COPY_TO_REGCLASS VK8WM:$mask, VK16WM)), addr:$ptr)), sub_ymm))>; - -def: Pat<(v8f32 (masked_load addr:$ptr, VK8WM:$mask, (v8f32 VR256:$src0))), - (v8f32 (EXTRACT_SUBREG (v16f32 (VMOVUPSZrmk - (INSERT_SUBREG (v16f32 (IMPLICIT_DEF)), VR256:$src0, sub_ymm), - (v16i1 (COPY_TO_REGCLASS VK8WM:$mask, VK16WM)), addr:$ptr)), sub_ymm))>; -} - defm VMOVDQA32 : avx512_alignedload_vl<0x6F, "vmovdqa32", avx512vl_i32_info, HasAVX512>, avx512_alignedstore_vl<0x7F, "vmovdqa32", avx512vl_i32_info, @@ -2711,17 +2827,6 @@ def : Pat<(v16i32 (vselect VK16WM:$mask, (v16i32 immAllZerosV), (v16i32 VR512:$src))), (VMOVDQU32Zrrkz (KNOTWrr VK16WM:$mask), VR512:$src)>; } -// NoVLX patterns -let Predicates = [HasAVX512, NoVLX] in { -def: Pat<(X86mstore addr:$ptr, VK8WM:$mask, (v8i32 VR256:$src)), - (VMOVDQU32Zmrk addr:$ptr, - (v16i1 (COPY_TO_REGCLASS VK8WM:$mask, VK16WM)), - (INSERT_SUBREG (v16i32 (IMPLICIT_DEF)), VR256:$src, sub_ymm))>; - -def: Pat<(v8i32 (masked_load addr:$ptr, VK8WM:$mask, undef)), - (v8i32 (EXTRACT_SUBREG (v16i32 (VMOVDQU32Zrmkz - (v16i1 (COPY_TO_REGCLASS VK8WM:$mask, VK16WM)), addr:$ptr)), sub_ymm))>; -} // Move Int Doubleword to Packed Double Int // @@ -2774,13 +2879,13 @@ def VMOVDI2SSZrm : AVX512BI<0x6E, MRMSrcMem, (outs FR32X:$dst), (ins i32mem:$sr // def VMOVPDI2DIZrr : AVX512BI<0x7E, MRMDestReg, (outs GR32:$dst), (ins VR128X:$src), "vmovd\t{$src, $dst|$dst, $src}", - [(set GR32:$dst, (vector_extract (v4i32 VR128X:$src), + [(set GR32:$dst, (extractelt (v4i32 VR128X:$src), (iPTR 0)))], IIC_SSE_MOVD_ToGP>, EVEX, VEX_LIG; def VMOVPDI2DIZmr : AVX512BI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, VR128X:$src), "vmovd\t{$src, $dst|$dst, $src}", - [(store (i32 (vector_extract (v4i32 VR128X:$src), + [(store (i32 (extractelt (v4i32 VR128X:$src), (iPTR 0))), addr:$dst)], IIC_SSE_MOVDQ>, EVEX, VEX_LIG, EVEX_CD8<32, CD8VT1>; @@ -2801,6 +2906,11 @@ def VMOVPQIto64Zmr : I<0xD6, MRMDestMem, (outs), EVEX, PD, VEX_LIG, VEX_W, EVEX_CD8<64, CD8VT1>, Sched<[WriteStore]>, Requires<[HasAVX512, In64BitMode]>; +def VMOV64toPQIZrr_REV : AVX512BI<0xD6, MRMDestReg, (outs VR128X:$dst), + (ins VR128X:$src), + "vmovq.s\t{$src, $dst|$dst, $src}",[]>, + EVEX, VEX_W, VEX_LIG; + // Move Scalar Single to Double Int // let isCodeGenOnly = 1 in { @@ -2829,71 +2939,74 @@ def VMOVQI2PQIZrm : AVX512BI<0x6E, MRMSrcMem, (outs VR128X:$dst), // AVX-512 MOVSS, MOVSD //===----------------------------------------------------------------------===// -multiclass avx512_move_scalar { - let hasSideEffects = 0 in { - def rr : SI<0x10, MRMSrcReg, (outs VR128X:$dst), (ins VR128X:$src1, RC:$src2), - !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"), - [(set VR128X:$dst, (vt (OpNode VR128X:$src1, - (scalar_to_vector RC:$src2))))], - IIC_SSE_MOV_S_RR>, EVEX_4V, VEX_LIG; - let Constraints = "$src1 = $dst" in - def rrk : SI<0x10, MRMSrcReg, (outs VR128X:$dst), - (ins VR128X:$src1, VK1WM:$mask, RC:$src2, RC:$src3), - !strconcat(asm, - "\t{$src3, $src2, $dst {${mask}}|$dst {${mask}}, $src2, $src3}"), - [], IIC_SSE_MOV_S_RR>, EVEX_4V, VEX_LIG, EVEX_K; - def rm : SI<0x10, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src), - !strconcat(asm, "\t{$src, $dst|$dst, $src}"), - [(set RC:$dst, (mem_pat addr:$src))], IIC_SSE_MOV_S_RM>, - EVEX, VEX_LIG; +multiclass avx512_move_scalar { + defm rr_Int : AVX512_maskable_scalar<0x10, MRMSrcReg, _, (outs _.RC:$dst), + (ins _.RC:$src1, _.RC:$src2), + asm, "$src2, $src1","$src1, $src2", + (_.VT (OpNode (_.VT _.RC:$src1), + (_.VT _.RC:$src2))), + IIC_SSE_MOV_S_RR>, EVEX_4V; + let Constraints = "$src1 = $dst" , mayLoad = 1 in + defm rm_Int : AVX512_maskable_3src_scalar<0x10, MRMSrcMem, _, + (outs _.RC:$dst), + (ins _.ScalarMemOp:$src), + asm,"$src","$src", + (_.VT (OpNode (_.VT _.RC:$src1), + (_.VT (scalar_to_vector + (_.ScalarLdFrag addr:$src)))))>, EVEX; + let isCodeGenOnly = 1 in { + def rr : AVX512PI<0x10, MRMSrcReg, (outs _.RC:$dst), + (ins _.RC:$src1, _.FRC:$src2), + !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"), + [(set _.RC:$dst, (_.VT (OpNode _.RC:$src1, + (scalar_to_vector _.FRC:$src2))))], + _.ExeDomain,IIC_SSE_MOV_S_RR>, EVEX_4V; + let mayLoad = 1 in + def rm : AVX512PI<0x10, MRMSrcMem, (outs _.FRC:$dst), (ins _.ScalarMemOp:$src), + !strconcat(asm, "\t{$src, $dst|$dst, $src}"), + [(set _.FRC:$dst, (_.ScalarLdFrag addr:$src))], + _.ExeDomain, IIC_SSE_MOV_S_RM>, EVEX; + } let mayStore = 1 in { - def mr: SI<0x11, MRMDestMem, (outs), (ins x86memop:$dst, RC:$src), - !strconcat(asm, "\t{$src, $dst|$dst, $src}"), - [(store RC:$src, addr:$dst)], IIC_SSE_MOV_S_MR>, - EVEX, VEX_LIG; - def mrk: SI<0x11, MRMDestMem, (outs), (ins x86memop:$dst, VK1WM:$mask, RC:$src), - !strconcat(asm, "\t{$src, $dst {${mask}}|$dst {${mask}}, $src}"), - [], IIC_SSE_MOV_S_MR>, - EVEX, VEX_LIG, EVEX_K; + def mr: AVX512PI<0x11, MRMDestMem, (outs), (ins _.ScalarMemOp:$dst, _.FRC:$src), + !strconcat(asm, "\t{$src, $dst|$dst, $src}"), + [(store _.FRC:$src, addr:$dst)], _.ExeDomain, IIC_SSE_MOV_S_MR>, + EVEX; + def mrk: AVX512PI<0x11, MRMDestMem, (outs), + (ins _.ScalarMemOp:$dst, VK1WM:$mask, _.FRC:$src), + !strconcat(asm, "\t{$src, $dst {${mask}}|$dst {${mask}}, $src}"), + [], _.ExeDomain, IIC_SSE_MOV_S_MR>, EVEX, EVEX_K; } // mayStore - } //hasSideEffects = 0 } -let ExeDomain = SSEPackedSingle in -defm VMOVSSZ : avx512_move_scalar<"movss", FR32X, X86Movss, v4f32, f32mem, - loadf32>, XS, EVEX_CD8<32, CD8VT1>; +defm VMOVSSZ : avx512_move_scalar<"vmovss", X86Movss, f32x_info>, + VEX_LIG, XS, EVEX_CD8<32, CD8VT1>; -let ExeDomain = SSEPackedDouble in -defm VMOVSDZ : avx512_move_scalar<"movsd", FR64X, X86Movsd, v2f64, f64mem, - loadf64>, XD, VEX_W, EVEX_CD8<64, CD8VT1>; +defm VMOVSDZ : avx512_move_scalar<"vmovsd", X86Movsd, f64x_info>, + VEX_LIG, XD, VEX_W, EVEX_CD8<64, CD8VT1>; def : Pat<(f32 (X86select VK1WM:$mask, (f32 FR32X:$src1), (f32 FR32X:$src2))), - (COPY_TO_REGCLASS (VMOVSSZrrk (COPY_TO_REGCLASS FR32X:$src2, VR128X), - VK1WM:$mask, (f32 (IMPLICIT_DEF)), FR32X:$src1), FR32X)>; + (COPY_TO_REGCLASS (VMOVSSZrr_Intk (COPY_TO_REGCLASS FR32X:$src2, VR128X), + VK1WM:$mask, (v4f32 (IMPLICIT_DEF)),(COPY_TO_REGCLASS FR32X:$src1, VR128X)), FR32X)>; def : Pat<(f64 (X86select VK1WM:$mask, (f64 FR64X:$src1), (f64 FR64X:$src2))), - (COPY_TO_REGCLASS (VMOVSDZrrk (COPY_TO_REGCLASS FR64X:$src2, VR128X), - VK1WM:$mask, (f64 (IMPLICIT_DEF)), FR64X:$src1), FR64X)>; + (COPY_TO_REGCLASS (VMOVSDZrr_Intk (COPY_TO_REGCLASS FR64X:$src2, VR128X), + VK1WM:$mask, (v2f64 (IMPLICIT_DEF)), (COPY_TO_REGCLASS FR64X:$src1, VR128X)), FR64X)>; def : Pat<(int_x86_avx512_mask_store_ss addr:$dst, VR128X:$src, GR8:$mask), (VMOVSSZmrk addr:$dst, (i1 (COPY_TO_REGCLASS GR8:$mask, VK1WM)), (COPY_TO_REGCLASS VR128X:$src, FR32X))>; -// For the disassembler -let isCodeGenOnly = 1, ForceDisassemble = 1, hasSideEffects = 0 in { - def VMOVSSZrr_REV : SI<0x11, MRMDestReg, (outs VR128X:$dst), - (ins VR128X:$src1, FR32X:$src2), - "movss\t{$src2, $src1, $dst|$dst, $src1, $src2}", [], - IIC_SSE_MOV_S_RR>, - XS, EVEX_4V, VEX_LIG; - def VMOVSDZrr_REV : SI<0x11, MRMDestReg, (outs VR128X:$dst), - (ins VR128X:$src1, FR64X:$src2), - "movsd\t{$src2, $src1, $dst|$dst, $src1, $src2}", [], - IIC_SSE_MOV_S_RR>, - XD, EVEX_4V, VEX_LIG, VEX_W; -} +defm VMOVSSZrr_REV : AVX512_maskable_in_asm<0x11, MRMDestReg, f32x_info, + (outs VR128X:$dst), (ins VR128X:$src1, VR128X:$src2), + "vmovss.s", "$src2, $src1", "$src1, $src2", []>, + XS, EVEX_4V, VEX_LIG; + +defm VMOVSSDrr_REV : AVX512_maskable_in_asm<0x11, MRMDestReg, f64x_info, + (outs VR128X:$dst), (ins VR128X:$src1, VR128X:$src2), + "vmovsd.s", "$src2, $src1", "$src1, $src2", []>, + XD, EVEX_4V, VEX_LIG, VEX_W; let Predicates = [HasAVX512] in { let AddedComplexity = 15 in { @@ -2977,10 +3090,10 @@ let Predicates = [HasAVX512] in { (EXTRACT_SUBREG (v4i64 VR256X:$src), sub_xmm)), sub_xmm)>; // Extract and store. - def : Pat<(store (f32 (vector_extract (v4f32 VR128X:$src), (iPTR 0))), + def : Pat<(store (f32 (extractelt (v4f32 VR128X:$src), (iPTR 0))), addr:$dst), (VMOVSSZmr addr:$dst, (COPY_TO_REGCLASS (v4f32 VR128X:$src), FR32X))>; - def : Pat<(store (f64 (vector_extract (v2f64 VR128X:$src), (iPTR 0))), + def : Pat<(store (f64 (extractelt (v2f64 VR128X:$src), (iPTR 0))), addr:$dst), (VMOVSDZmr addr:$dst, (COPY_TO_REGCLASS (v2f64 VR128X:$src), FR64X))>; @@ -3044,7 +3157,7 @@ def VMOVZPQILo2PQIZrr : AVX512XSI<0x7E, MRMSrcReg, (outs VR128X:$dst), (v2i64 VR128X:$src))))], IIC_SSE_MOVQ_RR>, EVEX, VEX_W; -let AddedComplexity = 20 in +let AddedComplexity = 20 , isCodeGenOnly = 1 in def VMOVZPQILo2PQIZrm : AVX512XSI<0x7E, MRMSrcMem, (outs VR128X:$dst), (ins i128mem:$src), "vmovq\t{$src, $dst|$dst, $src}", @@ -4194,55 +4307,6 @@ multiclass avx512_pshufb_sizes opc, string OpcodeStr, SDNode OpNode> { defm VPSHUFB: avx512_pshufb_sizes<0x00, "vpshufb", X86pshufb>; -//===----------------------------------------------------------------------===// -// AVX-512 - MOVDDUP -//===----------------------------------------------------------------------===// - -multiclass avx512_movddup { -def rr : AVX512PDI<0x12, MRMSrcReg, (outs RC:$dst), (ins RC:$src), - !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"), - [(set RC:$dst, (VT (X86Movddup RC:$src)))]>, EVEX; -def rm : AVX512PDI<0x12, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src), - !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"), - [(set RC:$dst, - (VT (X86Movddup (memop_frag addr:$src))))]>, EVEX; -} - -defm VMOVDDUPZ : avx512_movddup<"vmovddup", VR512, v8f64, f512mem, loadv8f64>, - VEX_W, EVEX_V512, EVEX_CD8<64, CD8VF>; -def : Pat<(X86Movddup (v8f64 (scalar_to_vector (loadf64 addr:$src)))), - (VMOVDDUPZrm addr:$src)>; - -//===---------------------------------------------------------------------===// -// Replicate Single FP - MOVSHDUP and MOVSLDUP -//===---------------------------------------------------------------------===// -multiclass avx512_replicate_sfp op, SDNode OpNode, string OpcodeStr, - ValueType vt, RegisterClass RC, PatFrag mem_frag, - X86MemOperand x86memop> { - def rr : AVX512XSI, EVEX; - let mayLoad = 1 in - def rm : AVX512XSI, EVEX; -} - -defm VMOVSHDUPZ : avx512_replicate_sfp<0x16, X86Movshdup, "vmovshdup", - v16f32, VR512, loadv16f32, f512mem>, EVEX_V512, - EVEX_CD8<32, CD8VF>; -defm VMOVSLDUPZ : avx512_replicate_sfp<0x12, X86Movsldup, "vmovsldup", - v16f32, VR512, loadv16f32, f512mem>, EVEX_V512, - EVEX_CD8<32, CD8VF>; - -def : Pat<(v16i32 (X86Movshdup VR512:$src)), (VMOVSHDUPZrr VR512:$src)>; -def : Pat<(v16i32 (X86Movshdup (loadv16i32 addr:$src))), - (VMOVSHDUPZrm addr:$src)>; -def : Pat<(v16i32 (X86Movsldup VR512:$src)), (VMOVSLDUPZrr VR512:$src)>; -def : Pat<(v16i32 (X86Movsldup (loadv16i32 addr:$src))), - (VMOVSLDUPZrm addr:$src)>; - //===----------------------------------------------------------------------===// // Move Low to High and High to Low packed FP Instructions //===----------------------------------------------------------------------===// @@ -4269,6 +4333,115 @@ let Predicates = [HasAVX512] in { (VMOVHLPSZrr VR128X:$src1, VR128X:$src2)>; } +//===----------------------------------------------------------------------===// +// VMOVHPS/PD VMOVLPS Instructions +// All patterns was taken from SSS implementation. +//===----------------------------------------------------------------------===// +multiclass avx512_mov_hilo_packed opc, string OpcodeStr, SDNode OpNode, + X86VectorVTInfo _> { + let mayLoad = 1 in + def rm : AVX512, EVEX_4V; +} + +defm VMOVHPSZ128 : avx512_mov_hilo_packed<0x16, "vmovhps", X86Movlhps, + v4f32x_info>, EVEX_CD8<32, CD8VT2>, PS; +defm VMOVHPDZ128 : avx512_mov_hilo_packed<0x16, "vmovhpd", X86Movlhpd, + v2f64x_info>, EVEX_CD8<64, CD8VT1>, PD, VEX_W; +defm VMOVLPSZ128 : avx512_mov_hilo_packed<0x12, "vmovlps", X86Movlps, + v4f32x_info>, EVEX_CD8<32, CD8VT2>, PS; +defm VMOVLPDZ128 : avx512_mov_hilo_packed<0x12, "vmovlpd", X86Movlpd, + v2f64x_info>, EVEX_CD8<64, CD8VT1>, PD, VEX_W; + +let Predicates = [HasAVX512] in { + // VMOVHPS patterns + def : Pat<(X86Movlhps VR128X:$src1, + (bc_v4f32 (v2i64 (scalar_to_vector (loadi64 addr:$src2))))), + (VMOVHPSZ128rm VR128X:$src1, addr:$src2)>; + def : Pat<(X86Movlhps VR128X:$src1, + (bc_v4i32 (v2i64 (X86vzload addr:$src2)))), + (VMOVHPSZ128rm VR128X:$src1, addr:$src2)>; + // VMOVHPD patterns + def : Pat<(v2f64 (X86Unpckl VR128X:$src1, + (scalar_to_vector (loadf64 addr:$src2)))), + (VMOVHPDZ128rm VR128X:$src1, addr:$src2)>; + def : Pat<(v2f64 (X86Unpckl VR128X:$src1, + (bc_v2f64 (v2i64 (scalar_to_vector (loadi64 addr:$src2)))))), + (VMOVHPDZ128rm VR128X:$src1, addr:$src2)>; + // VMOVLPS patterns + def : Pat<(v4f32 (X86Movlps VR128X:$src1, (load addr:$src2))), + (VMOVLPSZ128rm VR128X:$src1, addr:$src2)>; + def : Pat<(v4i32 (X86Movlps VR128X:$src1, (load addr:$src2))), + (VMOVLPSZ128rm VR128X:$src1, addr:$src2)>; + // VMOVLPD patterns + def : Pat<(v2f64 (X86Movlpd VR128X:$src1, (load addr:$src2))), + (VMOVLPDZ128rm VR128X:$src1, addr:$src2)>; + def : Pat<(v2i64 (X86Movlpd VR128X:$src1, (load addr:$src2))), + (VMOVLPDZ128rm VR128X:$src1, addr:$src2)>; + def : Pat<(v2f64 (X86Movsd VR128X:$src1, + (v2f64 (scalar_to_vector (loadf64 addr:$src2))))), + (VMOVLPDZ128rm VR128X:$src1, addr:$src2)>; +} + +let mayStore = 1 in { +def VMOVHPSZ128mr : AVX512PSI<0x17, MRMDestMem, (outs), + (ins f64mem:$dst, VR128X:$src), + "vmovhps\t{$src, $dst|$dst, $src}", + [(store (f64 (vector_extract + (X86Unpckh (bc_v2f64 (v4f32 VR128X:$src)), + (bc_v2f64 (v4f32 VR128X:$src))), + (iPTR 0))), addr:$dst)], IIC_SSE_MOV_LH>, + EVEX, EVEX_CD8<32, CD8VT2>; +def VMOVHPDZ128mr : AVX512PDI<0x17, MRMDestMem, (outs), + (ins f64mem:$dst, VR128X:$src), + "vmovhpd\t{$src, $dst|$dst, $src}", + [(store (f64 (vector_extract + (v2f64 (X86Unpckh VR128X:$src, VR128X:$src)), + (iPTR 0))), addr:$dst)], IIC_SSE_MOV_LH>, + EVEX, EVEX_CD8<64, CD8VT1>, VEX_W; +def VMOVLPSZ128mr : AVX512PSI<0x13, MRMDestMem, (outs), + (ins f64mem:$dst, VR128X:$src), + "vmovlps\t{$src, $dst|$dst, $src}", + [(store (f64 (vector_extract (bc_v2f64 (v4f32 VR128X:$src)), + (iPTR 0))), addr:$dst)], + IIC_SSE_MOV_LH>, + EVEX, EVEX_CD8<32, CD8VT2>; +def VMOVLPDZ128mr : AVX512PDI<0x13, MRMDestMem, (outs), + (ins f64mem:$dst, VR128X:$src), + "vmovlpd\t{$src, $dst|$dst, $src}", + [(store (f64 (vector_extract (v2f64 VR128X:$src), + (iPTR 0))), addr:$dst)], + IIC_SSE_MOV_LH>, + EVEX, EVEX_CD8<64, CD8VT1>, VEX_W; +} +let Predicates = [HasAVX512] in { + // VMOVHPD patterns + def : Pat<(store (f64 (vector_extract + (v2f64 (X86VPermilpi VR128X:$src, (i8 1))), + (iPTR 0))), addr:$dst), + (VMOVHPDZ128mr addr:$dst, VR128X:$src)>; + // VMOVLPS patterns + def : Pat<(store (v4f32 (X86Movlps (load addr:$src1), VR128X:$src2)), + addr:$src1), + (VMOVLPSZ128mr addr:$src1, VR128X:$src2)>; + def : Pat<(store (v4i32 (X86Movlps + (bc_v4i32 (loadv2i64 addr:$src1)), VR128X:$src2)), addr:$src1), + (VMOVLPSZ128mr addr:$src1, VR128X:$src2)>; + // VMOVLPD patterns + def : Pat<(store (v2f64 (X86Movlpd (load addr:$src1), VR128X:$src2)), + addr:$src1), + (VMOVLPDZ128mr addr:$src1, VR128X:$src2)>; + def : Pat<(store (v2i64 (X86Movlpd (load addr:$src1), VR128X:$src2)), + addr:$src1), + (VMOVLPDZ128mr addr:$src1, VR128X:$src2)>; +} //===----------------------------------------------------------------------===// // FMA - Fused Multiply Operations // @@ -5299,7 +5472,7 @@ defm VCVTQQ2PS : avx512_cvtqq2ps<0x5B, "vcvtqq2ps", sint_to_fp, defm VCVTUQQ2PS : avx512_cvtqq2ps<0x7A, "vcvtuqq2ps", uint_to_fp, X86VUlongToFpRnd>, VEX_W, XD, EVEX_CD8<64, CD8VF>; -let Predicates = [NoVLX] in { +let Predicates = [HasAVX512, NoVLX] in { def : Pat<(v8i32 (fp_to_uint (v8f32 VR256X:$src1))), (EXTRACT_SUBREG (v16i32 (VCVTTPS2UDQZrr (v16f32 (SUBREG_TO_REG (i32 0), VR256X:$src1, sub_ymm)))), sub_ymm)>; @@ -5405,6 +5578,29 @@ let Predicates = [HasAVX512] in { EVEX, EVEX_V128, EVEX_CD8<32, CD8VH>; } } + +// Unordered/Ordered scalar fp compare with Sea and set EFLAGS +multiclass avx512_ord_cmp_sae opc, X86VectorVTInfo _, SDNode OpNode, + string OpcodeStr> { + def rb: AVX512, EVEX, EVEX_B, VEX_LIG, EVEX_V128, + Sched<[WriteFAdd]>; +} + +let Defs = [EFLAGS], Predicates = [HasAVX512] in { + defm VUCOMISSZ : avx512_ord_cmp_sae<0x2E, v4f32x_info, X86ucomiSae, "vucomiss">, + AVX512PSIi8Base, EVEX_CD8<32, CD8VT1>; + defm VUCOMISDZ : avx512_ord_cmp_sae<0x2E, v2f64x_info, X86ucomiSae, "vucomisd">, + AVX512PDIi8Base, VEX_W, EVEX_CD8<64, CD8VT1>; + defm VCOMISSZ : avx512_ord_cmp_sae<0x2F, v4f32x_info, X86comiSae, "vcomiss">, + AVX512PSIi8Base, EVEX_CD8<32, CD8VT1>; + defm VCOMISDZ : avx512_ord_cmp_sae<0x2F, v2f64x_info, X86comiSae, "vcomisd">, + AVX512PDIi8Base, VEX_W, EVEX_CD8<64, CD8VT1>; +} + let Defs = [EFLAGS], Predicates = [HasAVX512] in { defm VUCOMISSZ : sse12_ord_cmp<0x2E, FR32X, X86cmp, f32, f32mem, loadf32, "ucomiss">, PS, EVEX, VEX_LIG, @@ -5711,12 +5907,12 @@ multiclass avx512_sqrt_scalar opc, string OpcodeStr,X86VectorVTInfo _, EVEX_B, EVEX_RC; let isCodeGenOnly = 1 in { - def r : SI; let mayLoad = 1 in - def m : SI; } @@ -6036,6 +6232,23 @@ defm VPMOVWB : avx512_trunc_wb<0x30, "vpmovwb", X86vtrunc>; defm VPMOVSWB : avx512_trunc_sat_wb<0x20, "s", X86vtruncs>; defm VPMOVUSWB : avx512_trunc_sat_wb<0x10, "us", X86vtruncus>; +let Predicates = [HasAVX512, NoVLX] in { +def: Pat<(v8i16 (X86vtrunc (v8i32 VR256X:$src))), + (v8i16 (EXTRACT_SUBREG + (v16i16 (VPMOVDWZrr (v16i32 (SUBREG_TO_REG (i32 0), + VR256X:$src, sub_ymm)))), sub_xmm))>; +def: Pat<(v4i32 (X86vtrunc (v4i64 VR256X:$src))), + (v4i32 (EXTRACT_SUBREG + (v8i32 (VPMOVQDZrr (v8i64 (SUBREG_TO_REG (i32 0), + VR256X:$src, sub_ymm)))), sub_xmm))>; +} + +let Predicates = [HasBWI, NoVLX] in { +def: Pat<(v16i8 (X86vtrunc (v16i16 VR256X:$src))), + (v16i8 (EXTRACT_SUBREG (VPMOVWBZrr (v32i16 (SUBREG_TO_REG (i32 0), + VR256X:$src, sub_ymm))), sub_xmm))>; +} + multiclass avx512_extend_common opc, string OpcodeStr, X86VectorVTInfo DestInfo, X86VectorVTInfo SrcInfo, X86MemOperand x86memop, PatFrag LdFrag, SDNode OpNode>{ @@ -6890,13 +7103,13 @@ defm VDBPSADBW: avx512_common_3Op_rm_imm8<0x42, X86dbpsadbw, "vdbpsadbw" , multiclass avx512_unary_rm opc, string OpcodeStr, SDNode OpNode, X86VectorVTInfo _> { defm rr : AVX512_maskable, EVEX, AVX5128IBase; let mayLoad = 1 in defm rm : AVX512_maskable, EVEX, AVX5128IBase, EVEX_CD8<_.EltSize, CD8VF>; @@ -6907,7 +7120,7 @@ multiclass avx512_unary_rmb opc, string OpcodeStr, SDNode OpNode, avx512_unary_rm { let mayLoad = 1 in defm rmb : AVX512_maskable opc, string OpcodeStr, SDNode OpNode, multiclass avx512_unary_rm_vl_dq opc_d, bits<8> opc_q, string OpcodeStr, SDNode OpNode, Predicate prd> { - defm Q : avx512_unary_rmb_vl, VEX_W; - defm D : avx512_unary_rmb_vl; + defm D : avx512_unary_rmb_vl; } multiclass avx512_unary_rm_vl_bw opc_b, bits<8> opc_w, string OpcodeStr, SDNode OpNode, Predicate prd> { - defm W : avx512_unary_rm_vl; - defm B : avx512_unary_rm_vl; + defm W : avx512_unary_rm_vl; + defm B : avx512_unary_rm_vl; } multiclass avx512_unary_rm_vl_all opc_b, bits<8> opc_w, @@ -6986,6 +7200,65 @@ multiclass avx512_ctlz opc, string OpcodeStr, Predicate prd>{ defm VPLZCNT : avx512_ctlz<0x44, "vplzcnt", HasCDI>; defm VPCONFLICT : avx512_unary_rm_vl_dq<0xC4, 0xC4, "vpconflict", X86Conflict, HasCDI>; +//===---------------------------------------------------------------------===// +// Replicate Single FP - MOVSHDUP and MOVSLDUP +//===---------------------------------------------------------------------===// +multiclass avx512_replicate opc, string OpcodeStr, SDNode OpNode>{ + defm NAME: avx512_unary_rm_vl, XS; + let isCodeGenOnly = 1 in + defm NAME#_I: avx512_unary_rm_vl, XS; +} + +defm VMOVSHDUP : avx512_replicate<0x16, "vmovshdup", X86Movshdup>; +defm VMOVSLDUP : avx512_replicate<0x12, "vmovsldup", X86Movsldup>; + +//===----------------------------------------------------------------------===// +// AVX-512 - MOVDDUP +//===----------------------------------------------------------------------===// + +multiclass avx512_movddup_128 opc, string OpcodeStr, SDNode OpNode, + X86VectorVTInfo _> { + defm rr : AVX512_maskable, EVEX; + let mayLoad = 1 in + defm rm : AVX512_maskable, + EVEX, EVEX_CD8<_.EltSize, CD8VH>; +} + +multiclass avx512_movddup_common opc, string OpcodeStr, SDNode OpNode, + AVX512VLVectorVTInfo VTInfo> { + + defm Z : avx512_unary_rm, EVEX_V512; + + let Predicates = [HasAVX512, HasVLX] in { + defm Z256 : avx512_unary_rm, + EVEX_V256; + defm Z128 : avx512_movddup_128, + EVEX_V128; + } +} + +multiclass avx512_movddup opc, string OpcodeStr, SDNode OpNode>{ + defm NAME: avx512_movddup_common, XD, VEX_W; + let isCodeGenOnly = 1 in + defm NAME#_I: avx512_movddup_common; +} + +defm VMOVDDUP : avx512_movddup<0x12, "vmovddup", X86Movddup>; + +def : Pat<(X86Movddup (loadv2f64 addr:$src)), + (VMOVDDUPZ128rm addr:$src)>, Requires<[HasAVX512, HasVLX]>; +def : Pat<(v2f64 (X86VBroadcast (loadf64 addr:$src))), + (VMOVDDUPZ128rm addr:$src)>, Requires<[HasAVX512, HasVLX]>; + //===----------------------------------------------------------------------===// // AVX-512 - Unpack Instructions //===----------------------------------------------------------------------===// @@ -7048,6 +7321,11 @@ multiclass avx512_extract_elt_w { (X86pextrw (_.VT _.RC:$src1), imm:$src2))]>, EVEX, PD; + def rr_REV : AVX512Ii8<0x15, MRMDestReg, (outs GR32orGR64:$dst), + (ins _.RC:$src1, u8imm:$src2), + OpcodeStr#".s\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>, + EVEX, TAPD; + defm NAME : avx512_extract_elt_bw_m<0x15, OpcodeStr, X86pextrw, _>, TAPD; } } @@ -7175,32 +7453,34 @@ defm VPSRLDQ : avx512_shift_packed_all<0x73, X86vshrdq, MRM3r, MRM3m, "vpsrldq", multiclass avx512_psadbw_packed opc, SDNode OpNode, - string OpcodeStr, X86VectorVTInfo _src>{ + string OpcodeStr, X86VectorVTInfo _dst, + X86VectorVTInfo _src>{ def rr : AVX512BI; + [(set _dst.RC:$dst,(_dst.VT + (OpNode (_src.VT _src.RC:$src1), + (_src.VT _src.RC:$src2))))]>; let mayLoad = 1 in def rm : AVX512BI; } multiclass avx512_psadbw_packed_all opc, SDNode OpNode, string OpcodeStr, Predicate prd> { let Predicates = [prd] in - defm Z512 : avx512_psadbw_packed, - EVEX_V512; + defm Z512 : avx512_psadbw_packed, EVEX_V512; let Predicates = [prd, HasVLX] in { - defm Z256 : avx512_psadbw_packed, - EVEX_V256; - defm Z128 : avx512_psadbw_packed, - EVEX_V128; + defm Z256 : avx512_psadbw_packed, EVEX_V256; + defm Z128 : avx512_psadbw_packed, EVEX_V128; } }