X-Git-Url: http://plrg.eecs.uci.edu/git/?p=oota-llvm.git;a=blobdiff_plain;f=lib%2FTarget%2FPowerPC%2FPPCInstrAltivec.td;h=2de852b48434748a8127b893e7355aadf156576a;hp=cc9cf0abfc1aaf83ead281d28e336f83b2c8ce9f;hb=b69d556c370b32dee9f64d8250e51aad33963cc2;hpb=a3acc2b6cf093571812e7e55d936cf188c695e23 diff --git a/lib/Target/PowerPC/PPCInstrAltivec.td b/lib/Target/PowerPC/PPCInstrAltivec.td index cc9cf0abfc1..2de852b4843 100644 --- a/lib/Target/PowerPC/PPCInstrAltivec.td +++ b/lib/Target/PowerPC/PPCInstrAltivec.td @@ -22,111 +22,160 @@ def vnot_ppc : PatFrag<(ops node:$in), def vpkuhum_shuffle : PatFrag<(ops node:$lhs, node:$rhs), (vector_shuffle node:$lhs, node:$rhs), [{ - return PPC::isVPKUHUMShuffleMask(cast(N), false); + return PPC::isVPKUHUMShuffleMask(cast(N), 0, *CurDAG); }]>; def vpkuwum_shuffle : PatFrag<(ops node:$lhs, node:$rhs), (vector_shuffle node:$lhs, node:$rhs), [{ - return PPC::isVPKUWUMShuffleMask(cast(N), false); + return PPC::isVPKUWUMShuffleMask(cast(N), 0, *CurDAG); }]>; def vpkuhum_unary_shuffle : PatFrag<(ops node:$lhs, node:$rhs), (vector_shuffle node:$lhs, node:$rhs), [{ - return PPC::isVPKUHUMShuffleMask(cast(N), true); + return PPC::isVPKUHUMShuffleMask(cast(N), 1, *CurDAG); }]>; def vpkuwum_unary_shuffle : PatFrag<(ops node:$lhs, node:$rhs), (vector_shuffle node:$lhs, node:$rhs), [{ - return PPC::isVPKUWUMShuffleMask(cast(N), true); + return PPC::isVPKUWUMShuffleMask(cast(N), 1, *CurDAG); }]>; +// These fragments are provided for little-endian, where the inputs must be +// swapped for correct semantics. +def vpkuhum_swapped_shuffle : PatFrag<(ops node:$lhs, node:$rhs), + (vector_shuffle node:$lhs, node:$rhs), [{ + return PPC::isVPKUHUMShuffleMask(cast(N), 2, *CurDAG); +}]>; +def vpkuwum_swapped_shuffle : PatFrag<(ops node:$lhs, node:$rhs), + (vector_shuffle node:$lhs, node:$rhs), [{ + return PPC::isVPKUWUMShuffleMask(cast(N), 2, *CurDAG); +}]>; def vmrglb_shuffle : PatFrag<(ops node:$lhs, node:$rhs), (vector_shuffle (v16i8 node:$lhs), node:$rhs), [{ - return PPC::isVMRGLShuffleMask(cast(N), 1, false); + return PPC::isVMRGLShuffleMask(cast(N), 1, 0, *CurDAG); }]>; def vmrglh_shuffle : PatFrag<(ops node:$lhs, node:$rhs), (vector_shuffle (v16i8 node:$lhs), node:$rhs), [{ - return PPC::isVMRGLShuffleMask(cast(N), 2, false); + return PPC::isVMRGLShuffleMask(cast(N), 2, 0, *CurDAG); }]>; def vmrglw_shuffle : PatFrag<(ops node:$lhs, node:$rhs), (vector_shuffle (v16i8 node:$lhs), node:$rhs), [{ - return PPC::isVMRGLShuffleMask(cast(N), 4, false); + return PPC::isVMRGLShuffleMask(cast(N), 4, 0, *CurDAG); }]>; def vmrghb_shuffle : PatFrag<(ops node:$lhs, node:$rhs), (vector_shuffle (v16i8 node:$lhs), node:$rhs), [{ - return PPC::isVMRGHShuffleMask(cast(N), 1, false); + return PPC::isVMRGHShuffleMask(cast(N), 1, 0, *CurDAG); }]>; def vmrghh_shuffle : PatFrag<(ops node:$lhs, node:$rhs), (vector_shuffle (v16i8 node:$lhs), node:$rhs), [{ - return PPC::isVMRGHShuffleMask(cast(N), 2, false); + return PPC::isVMRGHShuffleMask(cast(N), 2, 0, *CurDAG); }]>; def vmrghw_shuffle : PatFrag<(ops node:$lhs, node:$rhs), (vector_shuffle (v16i8 node:$lhs), node:$rhs), [{ - return PPC::isVMRGHShuffleMask(cast(N), 4, false); + return PPC::isVMRGHShuffleMask(cast(N), 4, 0, *CurDAG); }]>; def vmrglb_unary_shuffle : PatFrag<(ops node:$lhs, node:$rhs), (vector_shuffle (v16i8 node:$lhs), node:$rhs), [{ - return PPC::isVMRGLShuffleMask(cast(N), 1, true); + return PPC::isVMRGLShuffleMask(cast(N), 1, 1, *CurDAG); }]>; def vmrglh_unary_shuffle : PatFrag<(ops node:$lhs, node:$rhs), (vector_shuffle node:$lhs, node:$rhs), [{ - return PPC::isVMRGLShuffleMask(cast(N), 2, true); + return PPC::isVMRGLShuffleMask(cast(N), 2, 1, *CurDAG); }]>; def vmrglw_unary_shuffle : PatFrag<(ops node:$lhs, node:$rhs), (vector_shuffle node:$lhs, node:$rhs), [{ - return PPC::isVMRGLShuffleMask(cast(N), 4, true); + return PPC::isVMRGLShuffleMask(cast(N), 4, 1, *CurDAG); }]>; def vmrghb_unary_shuffle : PatFrag<(ops node:$lhs, node:$rhs), (vector_shuffle node:$lhs, node:$rhs), [{ - return PPC::isVMRGHShuffleMask(cast(N), 1, true); + return PPC::isVMRGHShuffleMask(cast(N), 1, 1, *CurDAG); }]>; def vmrghh_unary_shuffle : PatFrag<(ops node:$lhs, node:$rhs), (vector_shuffle node:$lhs, node:$rhs), [{ - return PPC::isVMRGHShuffleMask(cast(N), 2, true); + return PPC::isVMRGHShuffleMask(cast(N), 2, 1, *CurDAG); }]>; def vmrghw_unary_shuffle : PatFrag<(ops node:$lhs, node:$rhs), (vector_shuffle node:$lhs, node:$rhs), [{ - return PPC::isVMRGHShuffleMask(cast(N), 4, true); + return PPC::isVMRGHShuffleMask(cast(N), 4, 1, *CurDAG); +}]>; + + +// These fragments are provided for little-endian, where the inputs must be +// swapped for correct semantics. +def vmrglb_swapped_shuffle : PatFrag<(ops node:$lhs, node:$rhs), + (vector_shuffle (v16i8 node:$lhs), node:$rhs), [{ + return PPC::isVMRGLShuffleMask(cast(N), 1, 2, *CurDAG); +}]>; +def vmrglh_swapped_shuffle : PatFrag<(ops node:$lhs, node:$rhs), + (vector_shuffle node:$lhs, node:$rhs), [{ + return PPC::isVMRGLShuffleMask(cast(N), 2, 2, *CurDAG); +}]>; +def vmrglw_swapped_shuffle : PatFrag<(ops node:$lhs, node:$rhs), + (vector_shuffle node:$lhs, node:$rhs), [{ + return PPC::isVMRGLShuffleMask(cast(N), 4, 2, *CurDAG); +}]>; +def vmrghb_swapped_shuffle : PatFrag<(ops node:$lhs, node:$rhs), + (vector_shuffle node:$lhs, node:$rhs), [{ + return PPC::isVMRGHShuffleMask(cast(N), 1, 2, *CurDAG); +}]>; +def vmrghh_swapped_shuffle : PatFrag<(ops node:$lhs, node:$rhs), + (vector_shuffle node:$lhs, node:$rhs), [{ + return PPC::isVMRGHShuffleMask(cast(N), 2, 2, *CurDAG); +}]>; +def vmrghw_swapped_shuffle : PatFrag<(ops node:$lhs, node:$rhs), + (vector_shuffle node:$lhs, node:$rhs), [{ + return PPC::isVMRGHShuffleMask(cast(N), 4, 2, *CurDAG); }]>; def VSLDOI_get_imm : SDNodeXForm; def vsldoi_shuffle : PatFrag<(ops node:$lhs, node:$rhs), (vector_shuffle node:$lhs, node:$rhs), [{ - return PPC::isVSLDOIShuffleMask(N, false) != -1; + return PPC::isVSLDOIShuffleMask(N, 0, *CurDAG) != -1; }], VSLDOI_get_imm>; /// VSLDOI_unary* - These are used to match vsldoi(X,X), which is turned into /// vector_shuffle(X,undef,mask) by the dag combiner. def VSLDOI_unary_get_imm : SDNodeXForm; def vsldoi_unary_shuffle : PatFrag<(ops node:$lhs, node:$rhs), (vector_shuffle node:$lhs, node:$rhs), [{ - return PPC::isVSLDOIShuffleMask(N, true) != -1; + return PPC::isVSLDOIShuffleMask(N, 1, *CurDAG) != -1; }], VSLDOI_unary_get_imm>; +/// VSLDOI_swapped* - These fragments are provided for little-endian, where +/// the inputs must be swapped for correct semantics. +def VSLDOI_swapped_get_imm : SDNodeXForm; +def vsldoi_swapped_shuffle : PatFrag<(ops node:$lhs, node:$rhs), + (vector_shuffle node:$lhs, node:$rhs), [{ + return PPC::isVSLDOIShuffleMask(N, 2, *CurDAG) != -1; +}], VSLDOI_get_imm>; + + // VSPLT*_get_imm xform function: convert vector_shuffle mask to VSPLT* imm. def VSPLTB_get_imm : SDNodeXForm; def vspltb_shuffle : PatFrag<(ops node:$lhs, node:$rhs), (vector_shuffle node:$lhs, node:$rhs), [{ return PPC::isSplatShuffleMask(cast(N), 1); }], VSPLTB_get_imm>; def VSPLTH_get_imm : SDNodeXForm; def vsplth_shuffle : PatFrag<(ops node:$lhs, node:$rhs), (vector_shuffle node:$lhs, node:$rhs), [{ return PPC::isSplatShuffleMask(cast(N), 2); }], VSPLTH_get_imm>; def VSPLTW_get_imm : SDNodeXForm; def vspltw_shuffle : PatFrag<(ops node:$lhs, node:$rhs), (vector_shuffle node:$lhs, node:$rhs), [{ @@ -164,7 +213,7 @@ def vecspltisw : PatLeaf<(build_vector), [{ // VA1a_Int_Ty - A VAForm_1a intrinsic definition of specific type. class VA1a_Int_Ty xo, string opc, Intrinsic IntID, ValueType Ty> : VAForm_1a; // VA1a_Int_Ty2 - A VAForm_1a intrinsic definition where the type of the @@ -172,7 +221,7 @@ class VA1a_Int_Ty xo, string opc, Intrinsic IntID, ValueType Ty> class VA1a_Int_Ty2 xo, string opc, Intrinsic IntID, ValueType OutTy, ValueType InTy> : VAForm_1a; // VA1a_Int_Ty3 - A VAForm_1a intrinsic definition where there are two @@ -180,14 +229,14 @@ class VA1a_Int_Ty2 xo, string opc, Intrinsic IntID, ValueType OutTy, class VA1a_Int_Ty3 xo, string opc, Intrinsic IntID, ValueType OutTy, ValueType In1Ty, ValueType In2Ty> : VAForm_1a; // VX1_Int_Ty - A VXForm_1 intrinsic definition of specific type. class VX1_Int_Ty xo, string opc, Intrinsic IntID, ValueType Ty> : VXForm_1; // VX1_Int_Ty2 - A VXForm_1 intrinsic definition where the type of the @@ -195,7 +244,7 @@ class VX1_Int_Ty xo, string opc, Intrinsic IntID, ValueType Ty> class VX1_Int_Ty2 xo, string opc, Intrinsic IntID, ValueType OutTy, ValueType InTy> : VXForm_1; // VX1_Int_Ty3 - A VXForm_1 intrinsic definition where there are two @@ -203,13 +252,13 @@ class VX1_Int_Ty2 xo, string opc, Intrinsic IntID, ValueType OutTy, class VX1_Int_Ty3 xo, string opc, Intrinsic IntID, ValueType OutTy, ValueType In1Ty, ValueType In2Ty> : VXForm_1; // VX2_Int_SP - A VXForm_2 intrinsic definition of vector single-precision type. class VX2_Int_SP xo, string opc, Intrinsic IntID> : VXForm_2; // VX2_Int_Ty2 - A VXForm_2 intrinsic definition where the type of the @@ -217,118 +266,156 @@ class VX2_Int_SP xo, string opc, Intrinsic IntID> class VX2_Int_Ty2 xo, string opc, Intrinsic IntID, ValueType OutTy, ValueType InTy> : VXForm_2; +class VXBX_Int_Ty xo, string opc, Intrinsic IntID, ValueType Ty> + : VXForm_BX; + +class VXCR_Int_Ty xo, string opc, Intrinsic IntID, ValueType Ty> + : VXForm_CR; + //===----------------------------------------------------------------------===// // Instruction Definitions. -def HasAltivec : Predicate<"PPCSubTarget.hasAltivec()">; +def HasAltivec : Predicate<"PPCSubTarget->hasAltivec()">; let Predicates = [HasAltivec] in { +def DSS : DSS_Form<0, 822, (outs), (ins u5imm:$STRM), + "dss $STRM", IIC_LdStLoad /*FIXME*/, [(int_ppc_altivec_dss imm:$STRM)]>, + Deprecated { + let A = 0; + let B = 0; +} + +def DSSALL : DSS_Form<1, 822, (outs), (ins), + "dssall", IIC_LdStLoad /*FIXME*/, [(int_ppc_altivec_dssall)]>, + Deprecated { + let STRM = 0; + let A = 0; + let B = 0; +} + +def DST : DSS_Form<0, 342, (outs), (ins u5imm:$STRM, gprc:$rA, gprc:$rB), + "dst $rA, $rB, $STRM", IIC_LdStLoad /*FIXME*/, + [(int_ppc_altivec_dst i32:$rA, i32:$rB, imm:$STRM)]>, + Deprecated; + +def DSTT : DSS_Form<1, 342, (outs), (ins u5imm:$STRM, gprc:$rA, gprc:$rB), + "dstt $rA, $rB, $STRM", IIC_LdStLoad /*FIXME*/, + [(int_ppc_altivec_dstt i32:$rA, i32:$rB, imm:$STRM)]>, + Deprecated; + +def DSTST : DSS_Form<0, 374, (outs), (ins u5imm:$STRM, gprc:$rA, gprc:$rB), + "dstst $rA, $rB, $STRM", IIC_LdStLoad /*FIXME*/, + [(int_ppc_altivec_dstst i32:$rA, i32:$rB, imm:$STRM)]>, + Deprecated; + +def DSTSTT : DSS_Form<1, 374, (outs), (ins u5imm:$STRM, gprc:$rA, gprc:$rB), + "dststt $rA, $rB, $STRM", IIC_LdStLoad /*FIXME*/, + [(int_ppc_altivec_dststt i32:$rA, i32:$rB, imm:$STRM)]>, + Deprecated; + let isCodeGenOnly = 1 in { -def DSS : DSS_Form<822, (outs), - (ins u5imm:$ZERO0, u5imm:$STRM,u5imm:$ZERO1,u5imm:$ZERO2), - "dss $STRM", LdStLoad /*FIXME*/, []>; -def DSSALL : DSS_Form<822, (outs), - (ins u5imm:$ONE, u5imm:$ZERO0,u5imm:$ZERO1,u5imm:$ZERO2), - "dssall", LdStLoad /*FIXME*/, []>; -def DST : DSS_Form<342, (outs), - (ins u5imm:$ZERO, u5imm:$STRM, gprc:$rA, gprc:$rB), - "dst $rA, $rB, $STRM", LdStLoad /*FIXME*/, []>; -def DSTT : DSS_Form<342, (outs), - (ins u5imm:$ONE, u5imm:$STRM, gprc:$rA, gprc:$rB), - "dstt $rA, $rB, $STRM", LdStLoad /*FIXME*/, []>; -def DSTST : DSS_Form<374, (outs), - (ins u5imm:$ZERO, u5imm:$STRM, gprc:$rA, gprc:$rB), - "dstst $rA, $rB, $STRM", LdStLoad /*FIXME*/, []>; -def DSTSTT : DSS_Form<374, (outs), - (ins u5imm:$ONE, u5imm:$STRM, gprc:$rA, gprc:$rB), - "dststt $rA, $rB, $STRM", LdStLoad /*FIXME*/, []>; - -def DST64 : DSS_Form<342, (outs), - (ins u5imm:$ZERO, u5imm:$STRM, g8rc:$rA, gprc:$rB), - "dst $rA, $rB, $STRM", LdStLoad /*FIXME*/, []>; -def DSTT64 : DSS_Form<342, (outs), - (ins u5imm:$ONE, u5imm:$STRM, g8rc:$rA, gprc:$rB), - "dstt $rA, $rB, $STRM", LdStLoad /*FIXME*/, []>; -def DSTST64 : DSS_Form<374, (outs), - (ins u5imm:$ZERO, u5imm:$STRM, g8rc:$rA, gprc:$rB), - "dstst $rA, $rB, $STRM", LdStLoad /*FIXME*/, []>; -def DSTSTT64 : DSS_Form<374, (outs), - (ins u5imm:$ONE, u5imm:$STRM, g8rc:$rA, gprc:$rB), - "dststt $rA, $rB, $STRM", LdStLoad /*FIXME*/, []>; + // The very same instructions as above, but formally matching 64bit registers. + def DST64 : DSS_Form<0, 342, (outs), (ins u5imm:$STRM, g8rc:$rA, gprc:$rB), + "dst $rA, $rB, $STRM", IIC_LdStLoad /*FIXME*/, + [(int_ppc_altivec_dst i64:$rA, i32:$rB, imm:$STRM)]>, + Deprecated; + + def DSTT64 : DSS_Form<1, 342, (outs), (ins u5imm:$STRM, g8rc:$rA, gprc:$rB), + "dstt $rA, $rB, $STRM", IIC_LdStLoad /*FIXME*/, + [(int_ppc_altivec_dstt i64:$rA, i32:$rB, imm:$STRM)]>, + Deprecated; + + def DSTST64 : DSS_Form<0, 374, (outs), (ins u5imm:$STRM, g8rc:$rA, gprc:$rB), + "dstst $rA, $rB, $STRM", IIC_LdStLoad /*FIXME*/, + [(int_ppc_altivec_dstst i64:$rA, i32:$rB, + imm:$STRM)]>, + Deprecated; + + def DSTSTT64 : DSS_Form<1, 374, (outs), (ins u5imm:$STRM, g8rc:$rA, gprc:$rB), + "dststt $rA, $rB, $STRM", IIC_LdStLoad /*FIXME*/, + [(int_ppc_altivec_dststt i64:$rA, i32:$rB, + imm:$STRM)]>, + Deprecated; } def MFVSCR : VXForm_4<1540, (outs vrrc:$vD), (ins), - "mfvscr $vD", LdStStore, + "mfvscr $vD", IIC_LdStStore, [(set v8i16:$vD, (int_ppc_altivec_mfvscr))]>; def MTVSCR : VXForm_5<1604, (outs), (ins vrrc:$vB), - "mtvscr $vB", LdStLoad, + "mtvscr $vB", IIC_LdStLoad, [(int_ppc_altivec_mtvscr v4i32:$vB)]>; let canFoldAsLoad = 1, PPC970_Unit = 2 in { // Loads. def LVEBX: XForm_1<31, 7, (outs vrrc:$vD), (ins memrr:$src), - "lvebx $vD, $src", LdStLoad, + "lvebx $vD, $src", IIC_LdStLoad, [(set v16i8:$vD, (int_ppc_altivec_lvebx xoaddr:$src))]>; def LVEHX: XForm_1<31, 39, (outs vrrc:$vD), (ins memrr:$src), - "lvehx $vD, $src", LdStLoad, + "lvehx $vD, $src", IIC_LdStLoad, [(set v8i16:$vD, (int_ppc_altivec_lvehx xoaddr:$src))]>; def LVEWX: XForm_1<31, 71, (outs vrrc:$vD), (ins memrr:$src), - "lvewx $vD, $src", LdStLoad, + "lvewx $vD, $src", IIC_LdStLoad, [(set v4i32:$vD, (int_ppc_altivec_lvewx xoaddr:$src))]>; def LVX : XForm_1<31, 103, (outs vrrc:$vD), (ins memrr:$src), - "lvx $vD, $src", LdStLoad, + "lvx $vD, $src", IIC_LdStLoad, [(set v4i32:$vD, (int_ppc_altivec_lvx xoaddr:$src))]>; def LVXL : XForm_1<31, 359, (outs vrrc:$vD), (ins memrr:$src), - "lvxl $vD, $src", LdStLoad, + "lvxl $vD, $src", IIC_LdStLoad, [(set v4i32:$vD, (int_ppc_altivec_lvxl xoaddr:$src))]>; } def LVSL : XForm_1<31, 6, (outs vrrc:$vD), (ins memrr:$src), - "lvsl $vD, $src", LdStLoad, + "lvsl $vD, $src", IIC_LdStLoad, [(set v16i8:$vD, (int_ppc_altivec_lvsl xoaddr:$src))]>, PPC970_Unit_LSU; def LVSR : XForm_1<31, 38, (outs vrrc:$vD), (ins memrr:$src), - "lvsr $vD, $src", LdStLoad, + "lvsr $vD, $src", IIC_LdStLoad, [(set v16i8:$vD, (int_ppc_altivec_lvsr xoaddr:$src))]>, PPC970_Unit_LSU; let PPC970_Unit = 2 in { // Stores. def STVEBX: XForm_8<31, 135, (outs), (ins vrrc:$rS, memrr:$dst), - "stvebx $rS, $dst", LdStStore, + "stvebx $rS, $dst", IIC_LdStStore, [(int_ppc_altivec_stvebx v16i8:$rS, xoaddr:$dst)]>; def STVEHX: XForm_8<31, 167, (outs), (ins vrrc:$rS, memrr:$dst), - "stvehx $rS, $dst", LdStStore, + "stvehx $rS, $dst", IIC_LdStStore, [(int_ppc_altivec_stvehx v8i16:$rS, xoaddr:$dst)]>; def STVEWX: XForm_8<31, 199, (outs), (ins vrrc:$rS, memrr:$dst), - "stvewx $rS, $dst", LdStStore, + "stvewx $rS, $dst", IIC_LdStStore, [(int_ppc_altivec_stvewx v4i32:$rS, xoaddr:$dst)]>; def STVX : XForm_8<31, 231, (outs), (ins vrrc:$rS, memrr:$dst), - "stvx $rS, $dst", LdStStore, + "stvx $rS, $dst", IIC_LdStStore, [(int_ppc_altivec_stvx v4i32:$rS, xoaddr:$dst)]>; def STVXL : XForm_8<31, 487, (outs), (ins vrrc:$rS, memrr:$dst), - "stvxl $rS, $dst", LdStStore, + "stvxl $rS, $dst", IIC_LdStStore, [(int_ppc_altivec_stvxl v4i32:$rS, xoaddr:$dst)]>; } let PPC970_Unit = 5 in { // VALU Operations. // VA-Form instructions. 3-input AltiVec ops. +let isCommutable = 1 in { def VMADDFP : VAForm_1<46, (outs vrrc:$vD), (ins vrrc:$vA, vrrc:$vC, vrrc:$vB), - "vmaddfp $vD, $vA, $vC, $vB", VecFP, + "vmaddfp $vD, $vA, $vC, $vB", IIC_VecFP, [(set v4f32:$vD, (fma v4f32:$vA, v4f32:$vC, v4f32:$vB))]>; // FIXME: The fma+fneg pattern won't match because fneg is not legal. def VNMSUBFP: VAForm_1<47, (outs vrrc:$vD), (ins vrrc:$vA, vrrc:$vC, vrrc:$vB), - "vnmsubfp $vD, $vA, $vC, $vB", VecFP, + "vnmsubfp $vD, $vA, $vC, $vB", IIC_VecFP, [(set v4f32:$vD, (fneg (fma v4f32:$vA, v4f32:$vC, - (fneg v4f32:$vB))))]>; + (fneg v4f32:$vB))))]>; def VMHADDSHS : VA1a_Int_Ty<32, "vmhaddshs", int_ppc_altivec_vmhaddshs, v8i16>; def VMHRADDSHS : VA1a_Int_Ty<33, "vmhraddshs", int_ppc_altivec_vmhraddshs, v8i16>; def VMLADDUHM : VA1a_Int_Ty<34, "vmladduhm", int_ppc_altivec_vmladduhm, v8i16>; +} // isCommutable def VPERM : VA1a_Int_Ty3<43, "vperm", int_ppc_altivec_vperm, v4i32, v4i32, v16i8>; @@ -336,23 +423,24 @@ def VSEL : VA1a_Int_Ty<42, "vsel", int_ppc_altivec_vsel, v4i32>; // Shuffles. def VSLDOI : VAForm_2<44, (outs vrrc:$vD), (ins vrrc:$vA, vrrc:$vB, u5imm:$SH), - "vsldoi $vD, $vA, $vB, $SH", VecFP, + "vsldoi $vD, $vA, $vB, $SH", IIC_VecFP, [(set v16i8:$vD, (vsldoi_shuffle:$SH v16i8:$vA, v16i8:$vB))]>; // VX-Form instructions. AltiVec arithmetic ops. +let isCommutable = 1 in { def VADDFP : VXForm_1<10, (outs vrrc:$vD), (ins vrrc:$vA, vrrc:$vB), - "vaddfp $vD, $vA, $vB", VecFP, + "vaddfp $vD, $vA, $vB", IIC_VecFP, [(set v4f32:$vD, (fadd v4f32:$vA, v4f32:$vB))]>; def VADDUBM : VXForm_1<0, (outs vrrc:$vD), (ins vrrc:$vA, vrrc:$vB), - "vaddubm $vD, $vA, $vB", VecGeneral, + "vaddubm $vD, $vA, $vB", IIC_VecGeneral, [(set v16i8:$vD, (add v16i8:$vA, v16i8:$vB))]>; def VADDUHM : VXForm_1<64, (outs vrrc:$vD), (ins vrrc:$vA, vrrc:$vB), - "vadduhm $vD, $vA, $vB", VecGeneral, + "vadduhm $vD, $vA, $vB", IIC_VecGeneral, [(set v8i16:$vD, (add v8i16:$vA, v8i16:$vB))]>; def VADDUWM : VXForm_1<128, (outs vrrc:$vD), (ins vrrc:$vA, vrrc:$vB), - "vadduwm $vD, $vA, $vB", VecGeneral, + "vadduwm $vD, $vA, $vB", IIC_VecGeneral, [(set v4i32:$vD, (add v4i32:$vA, v4i32:$vB))]>; def VADDCUW : VX1_Int_Ty<384, "vaddcuw", int_ppc_altivec_vaddcuw, v4i32>; @@ -362,57 +450,59 @@ def VADDSWS : VX1_Int_Ty<896, "vaddsws", int_ppc_altivec_vaddsws, v4i32>; def VADDUBS : VX1_Int_Ty<512, "vaddubs", int_ppc_altivec_vaddubs, v16i8>; def VADDUHS : VX1_Int_Ty<576, "vadduhs", int_ppc_altivec_vadduhs, v8i16>; def VADDUWS : VX1_Int_Ty<640, "vadduws", int_ppc_altivec_vadduws, v4i32>; - - +} // isCommutable + +let isCommutable = 1 in def VAND : VXForm_1<1028, (outs vrrc:$vD), (ins vrrc:$vA, vrrc:$vB), - "vand $vD, $vA, $vB", VecFP, + "vand $vD, $vA, $vB", IIC_VecFP, [(set v4i32:$vD, (and v4i32:$vA, v4i32:$vB))]>; def VANDC : VXForm_1<1092, (outs vrrc:$vD), (ins vrrc:$vA, vrrc:$vB), - "vandc $vD, $vA, $vB", VecFP, + "vandc $vD, $vA, $vB", IIC_VecFP, [(set v4i32:$vD, (and v4i32:$vA, (vnot_ppc v4i32:$vB)))]>; def VCFSX : VXForm_1<842, (outs vrrc:$vD), (ins u5imm:$UIMM, vrrc:$vB), - "vcfsx $vD, $vB, $UIMM", VecFP, + "vcfsx $vD, $vB, $UIMM", IIC_VecFP, [(set v4f32:$vD, (int_ppc_altivec_vcfsx v4i32:$vB, imm:$UIMM))]>; def VCFUX : VXForm_1<778, (outs vrrc:$vD), (ins u5imm:$UIMM, vrrc:$vB), - "vcfux $vD, $vB, $UIMM", VecFP, + "vcfux $vD, $vB, $UIMM", IIC_VecFP, [(set v4f32:$vD, (int_ppc_altivec_vcfux v4i32:$vB, imm:$UIMM))]>; def VCTSXS : VXForm_1<970, (outs vrrc:$vD), (ins u5imm:$UIMM, vrrc:$vB), - "vctsxs $vD, $vB, $UIMM", VecFP, + "vctsxs $vD, $vB, $UIMM", IIC_VecFP, [(set v4i32:$vD, (int_ppc_altivec_vctsxs v4f32:$vB, imm:$UIMM))]>; def VCTUXS : VXForm_1<906, (outs vrrc:$vD), (ins u5imm:$UIMM, vrrc:$vB), - "vctuxs $vD, $vB, $UIMM", VecFP, + "vctuxs $vD, $vB, $UIMM", IIC_VecFP, [(set v4i32:$vD, (int_ppc_altivec_vctuxs v4f32:$vB, imm:$UIMM))]>; // Defines with the UIM field set to 0 for floating-point // to integer (fp_to_sint/fp_to_uint) conversions and integer // to floating-point (sint_to_fp/uint_to_fp) conversions. -let VA = 0 in { +let isCodeGenOnly = 1, VA = 0 in { def VCFSX_0 : VXForm_1<842, (outs vrrc:$vD), (ins vrrc:$vB), - "vcfsx $vD, $vB, 0", VecFP, + "vcfsx $vD, $vB, 0", IIC_VecFP, [(set v4f32:$vD, (int_ppc_altivec_vcfsx v4i32:$vB, 0))]>; def VCTUXS_0 : VXForm_1<906, (outs vrrc:$vD), (ins vrrc:$vB), - "vctuxs $vD, $vB, 0", VecFP, + "vctuxs $vD, $vB, 0", IIC_VecFP, [(set v4i32:$vD, (int_ppc_altivec_vctuxs v4f32:$vB, 0))]>; def VCFUX_0 : VXForm_1<778, (outs vrrc:$vD), (ins vrrc:$vB), - "vcfux $vD, $vB, 0", VecFP, + "vcfux $vD, $vB, 0", IIC_VecFP, [(set v4f32:$vD, (int_ppc_altivec_vcfux v4i32:$vB, 0))]>; def VCTSXS_0 : VXForm_1<970, (outs vrrc:$vD), (ins vrrc:$vB), - "vctsxs $vD, $vB, 0", VecFP, + "vctsxs $vD, $vB, 0", IIC_VecFP, [(set v4i32:$vD, (int_ppc_altivec_vctsxs v4f32:$vB, 0))]>; } def VEXPTEFP : VX2_Int_SP<394, "vexptefp", int_ppc_altivec_vexptefp>; def VLOGEFP : VX2_Int_SP<458, "vlogefp", int_ppc_altivec_vlogefp>; +let isCommutable = 1 in { def VAVGSB : VX1_Int_Ty<1282, "vavgsb", int_ppc_altivec_vavgsb, v16i8>; def VAVGSH : VX1_Int_Ty<1346, "vavgsh", int_ppc_altivec_vavgsh, v8i16>; def VAVGSW : VX1_Int_Ty<1410, "vavgsw", int_ppc_altivec_vavgsw, v4i32>; @@ -434,24 +524,25 @@ def VMINSW : VX1_Int_Ty< 898, "vminsw", int_ppc_altivec_vminsw, v4i32>; def VMINUB : VX1_Int_Ty< 514, "vminub", int_ppc_altivec_vminub, v16i8>; def VMINUH : VX1_Int_Ty< 578, "vminuh", int_ppc_altivec_vminuh, v8i16>; def VMINUW : VX1_Int_Ty< 642, "vminuw", int_ppc_altivec_vminuw, v4i32>; +} // isCommutable def VMRGHB : VXForm_1< 12, (outs vrrc:$vD), (ins vrrc:$vA, vrrc:$vB), - "vmrghb $vD, $vA, $vB", VecFP, + "vmrghb $vD, $vA, $vB", IIC_VecFP, [(set v16i8:$vD, (vmrghb_shuffle v16i8:$vA, v16i8:$vB))]>; def VMRGHH : VXForm_1< 76, (outs vrrc:$vD), (ins vrrc:$vA, vrrc:$vB), - "vmrghh $vD, $vA, $vB", VecFP, + "vmrghh $vD, $vA, $vB", IIC_VecFP, [(set v16i8:$vD, (vmrghh_shuffle v16i8:$vA, v16i8:$vB))]>; def VMRGHW : VXForm_1<140, (outs vrrc:$vD), (ins vrrc:$vA, vrrc:$vB), - "vmrghw $vD, $vA, $vB", VecFP, + "vmrghw $vD, $vA, $vB", IIC_VecFP, [(set v16i8:$vD, (vmrghw_shuffle v16i8:$vA, v16i8:$vB))]>; def VMRGLB : VXForm_1<268, (outs vrrc:$vD), (ins vrrc:$vA, vrrc:$vB), - "vmrglb $vD, $vA, $vB", VecFP, + "vmrglb $vD, $vA, $vB", IIC_VecFP, [(set v16i8:$vD, (vmrglb_shuffle v16i8:$vA, v16i8:$vB))]>; def VMRGLH : VXForm_1<332, (outs vrrc:$vD), (ins vrrc:$vA, vrrc:$vB), - "vmrglh $vD, $vA, $vB", VecFP, + "vmrglh $vD, $vA, $vB", IIC_VecFP, [(set v16i8:$vD, (vmrglh_shuffle v16i8:$vA, v16i8:$vB))]>; def VMRGLW : VXForm_1<396, (outs vrrc:$vD), (ins vrrc:$vA, vrrc:$vB), - "vmrglw $vD, $vA, $vB", VecFP, + "vmrglw $vD, $vA, $vB", IIC_VecFP, [(set v16i8:$vD, (vmrglw_shuffle v16i8:$vA, v16i8:$vB))]>; def VMSUMMBM : VA1a_Int_Ty3<37, "vmsummbm", int_ppc_altivec_vmsummbm, @@ -467,6 +558,7 @@ def VMSUMUHM : VA1a_Int_Ty3<38, "vmsumuhm", int_ppc_altivec_vmsumuhm, def VMSUMUHS : VA1a_Int_Ty3<39, "vmsumuhs", int_ppc_altivec_vmsumuhs, v4i32, v8i16, v4i32>; +let isCommutable = 1 in { def VMULESB : VX1_Int_Ty2<776, "vmulesb", int_ppc_altivec_vmulesb, v8i16, v16i8>; def VMULESH : VX1_Int_Ty2<840, "vmulesh", int_ppc_altivec_vmulesh, @@ -483,6 +575,7 @@ def VMULOUB : VX1_Int_Ty2< 8, "vmuloub", int_ppc_altivec_vmuloub, v8i16, v16i8>; def VMULOUH : VX1_Int_Ty2< 72, "vmulouh", int_ppc_altivec_vmulouh, v4i32, v8i16>; +} // isCommutable def VREFP : VX2_Int_SP<266, "vrefp", int_ppc_altivec_vrefp>; def VRFIM : VX2_Int_SP<714, "vrfim", int_ppc_altivec_vrfim>; @@ -494,16 +587,16 @@ def VRSQRTEFP : VX2_Int_SP<330, "vrsqrtefp", int_ppc_altivec_vrsqrtefp>; def VSUBCUW : VX1_Int_Ty<1408, "vsubcuw", int_ppc_altivec_vsubcuw, v4i32>; def VSUBFP : VXForm_1<74, (outs vrrc:$vD), (ins vrrc:$vA, vrrc:$vB), - "vsubfp $vD, $vA, $vB", VecGeneral, + "vsubfp $vD, $vA, $vB", IIC_VecGeneral, [(set v4f32:$vD, (fsub v4f32:$vA, v4f32:$vB))]>; def VSUBUBM : VXForm_1<1024, (outs vrrc:$vD), (ins vrrc:$vA, vrrc:$vB), - "vsububm $vD, $vA, $vB", VecGeneral, + "vsububm $vD, $vA, $vB", IIC_VecGeneral, [(set v16i8:$vD, (sub v16i8:$vA, v16i8:$vB))]>; def VSUBUHM : VXForm_1<1088, (outs vrrc:$vD), (ins vrrc:$vA, vrrc:$vB), - "vsubuhm $vD, $vA, $vB", VecGeneral, + "vsubuhm $vD, $vA, $vB", IIC_VecGeneral, [(set v8i16:$vD, (sub v8i16:$vA, v8i16:$vB))]>; def VSUBUWM : VXForm_1<1152, (outs vrrc:$vD), (ins vrrc:$vA, vrrc:$vB), - "vsubuwm $vD, $vA, $vB", VecGeneral, + "vsubuwm $vD, $vA, $vB", IIC_VecGeneral, [(set v4i32:$vD, (sub v4i32:$vA, v4i32:$vB))]>; def VSUBSBS : VX1_Int_Ty<1792, "vsubsbs" , int_ppc_altivec_vsubsbs, v16i8>; @@ -524,15 +617,17 @@ def VSUM4UBS: VX1_Int_Ty3<1544, "vsum4ubs", int_ppc_altivec_vsum4ubs, v4i32, v16i8, v4i32>; def VNOR : VXForm_1<1284, (outs vrrc:$vD), (ins vrrc:$vA, vrrc:$vB), - "vnor $vD, $vA, $vB", VecFP, + "vnor $vD, $vA, $vB", IIC_VecFP, [(set v4i32:$vD, (vnot_ppc (or v4i32:$vA, v4i32:$vB)))]>; +let isCommutable = 1 in { def VOR : VXForm_1<1156, (outs vrrc:$vD), (ins vrrc:$vA, vrrc:$vB), - "vor $vD, $vA, $vB", VecFP, + "vor $vD, $vA, $vB", IIC_VecFP, [(set v4i32:$vD, (or v4i32:$vA, v4i32:$vB))]>; def VXOR : VXForm_1<1220, (outs vrrc:$vD), (ins vrrc:$vA, vrrc:$vB), - "vxor $vD, $vA, $vB", VecFP, + "vxor $vD, $vA, $vB", IIC_VecFP, [(set v4i32:$vD, (xor v4i32:$vA, v4i32:$vB))]>; +} // isCommutable def VRLB : VX1_Int_Ty< 4, "vrlb", int_ppc_altivec_vrlb, v16i8>; def VRLH : VX1_Int_Ty< 68, "vrlh", int_ppc_altivec_vrlh, v8i16>; @@ -546,15 +641,15 @@ def VSLH : VX1_Int_Ty< 324, "vslh", int_ppc_altivec_vslh, v8i16>; def VSLW : VX1_Int_Ty< 388, "vslw", int_ppc_altivec_vslw, v4i32>; def VSPLTB : VXForm_1<524, (outs vrrc:$vD), (ins u5imm:$UIMM, vrrc:$vB), - "vspltb $vD, $vB, $UIMM", VecPerm, + "vspltb $vD, $vB, $UIMM", IIC_VecPerm, [(set v16i8:$vD, (vspltb_shuffle:$UIMM v16i8:$vB, (undef)))]>; def VSPLTH : VXForm_1<588, (outs vrrc:$vD), (ins u5imm:$UIMM, vrrc:$vB), - "vsplth $vD, $vB, $UIMM", VecPerm, + "vsplth $vD, $vB, $UIMM", IIC_VecPerm, [(set v16i8:$vD, (vsplth_shuffle:$UIMM v16i8:$vB, (undef)))]>; def VSPLTW : VXForm_1<652, (outs vrrc:$vD), (ins u5imm:$UIMM, vrrc:$vB), - "vspltw $vD, $vB, $UIMM", VecPerm, + "vspltw $vD, $vB, $UIMM", IIC_VecPerm, [(set v16i8:$vD, (vspltw_shuffle:$UIMM v16i8:$vB, (undef)))]>; @@ -570,13 +665,13 @@ def VSRW : VX1_Int_Ty< 644, "vsrw" , int_ppc_altivec_vsrw , v4i32>; def VSPLTISB : VXForm_3<780, (outs vrrc:$vD), (ins s5imm:$SIMM), - "vspltisb $vD, $SIMM", VecPerm, + "vspltisb $vD, $SIMM", IIC_VecPerm, [(set v16i8:$vD, (v16i8 vecspltisb:$SIMM))]>; def VSPLTISH : VXForm_3<844, (outs vrrc:$vD), (ins s5imm:$SIMM), - "vspltish $vD, $SIMM", VecPerm, + "vspltish $vD, $SIMM", IIC_VecPerm, [(set v8i16:$vD, (v8i16 vecspltish:$SIMM))]>; def VSPLTISW : VXForm_3<908, (outs vrrc:$vD), (ins s5imm:$SIMM), - "vspltisw $vD, $SIMM", VecPerm, + "vspltisw $vD, $SIMM", IIC_VecPerm, [(set v4i32:$vD, (v4i32 vecspltisw:$SIMM))]>; // Vector Pack. @@ -591,13 +686,13 @@ def VPKSWSS : VX1_Int_Ty2<462, "vpkswss", int_ppc_altivec_vpkswss, def VPKSWUS : VX1_Int_Ty2<334, "vpkswus", int_ppc_altivec_vpkswus, v8i16, v4i32>; def VPKUHUM : VXForm_1<14, (outs vrrc:$vD), (ins vrrc:$vA, vrrc:$vB), - "vpkuhum $vD, $vA, $vB", VecFP, + "vpkuhum $vD, $vA, $vB", IIC_VecFP, [(set v16i8:$vD, (vpkuhum_shuffle v16i8:$vA, v16i8:$vB))]>; def VPKUHUS : VX1_Int_Ty2<142, "vpkuhus", int_ppc_altivec_vpkuhus, v16i8, v8i16>; def VPKUWUM : VXForm_1<78, (outs vrrc:$vD), (ins vrrc:$vA, vrrc:$vB), - "vpkuwum $vD, $vA, $vB", VecFP, + "vpkuwum $vD, $vA, $vB", IIC_VecFP, [(set v16i8:$vD, (vpkuwum_shuffle v16i8:$vA, v16i8:$vB))]>; def VPKUWUS : VX1_Int_Ty2<206, "vpkuwus", int_ppc_altivec_vpkuwus, @@ -621,10 +716,12 @@ def VUPKLSH : VX2_Int_Ty2<718, "vupklsh", int_ppc_altivec_vupklsh, // Altivec Comparisons. class VCMP xo, string asmstr, ValueType Ty> - : VXRForm_1; class VCMPo xo, string asmstr, ValueType Ty> - : VXRForm_1 { let Defs = [CR6]; let RC = 1; @@ -663,46 +760,36 @@ def VCMPGTSW : VCMP <902, "vcmpgtsw $vD, $vA, $vB" , v4i32>; def VCMPGTSWo : VCMPo<902, "vcmpgtsw. $vD, $vA, $vB", v4i32>; def VCMPGTUW : VCMP <646, "vcmpgtuw $vD, $vA, $vB" , v4i32>; def VCMPGTUWo : VCMPo<646, "vcmpgtuw. $vD, $vA, $vB", v4i32>; - -let isCodeGenOnly = 1 in -def V_SET0 : VXForm_setzero<1220, (outs vrrc:$vD), (ins), - "vxor $vD, $vD, $vD", VecFP, + +let isCodeGenOnly = 1 in { +def V_SET0B : VXForm_setzero<1220, (outs vrrc:$vD), (ins), + "vxor $vD, $vD, $vD", IIC_VecFP, + [(set v16i8:$vD, (v16i8 immAllZerosV))]>; +def V_SET0H : VXForm_setzero<1220, (outs vrrc:$vD), (ins), + "vxor $vD, $vD, $vD", IIC_VecFP, + [(set v8i16:$vD, (v8i16 immAllZerosV))]>; +def V_SET0 : VXForm_setzero<1220, (outs vrrc:$vD), (ins), + "vxor $vD, $vD, $vD", IIC_VecFP, [(set v4i32:$vD, (v4i32 immAllZerosV))]>; + let IMM=-1 in { -def V_SETALLONES : VXForm_3<908, (outs vrrc:$vD), (ins), - "vspltisw $vD, -1", VecFP, +def V_SETALLONESB : VXForm_3<908, (outs vrrc:$vD), (ins), + "vspltisw $vD, -1", IIC_VecFP, + [(set v16i8:$vD, (v16i8 immAllOnesV))]>; +def V_SETALLONESH : VXForm_3<908, (outs vrrc:$vD), (ins), + "vspltisw $vD, -1", IIC_VecFP, + [(set v8i16:$vD, (v8i16 immAllOnesV))]>; +def V_SETALLONES : VXForm_3<908, (outs vrrc:$vD), (ins), + "vspltisw $vD, -1", IIC_VecFP, [(set v4i32:$vD, (v4i32 immAllOnesV))]>; } +} } // VALU Operations. //===----------------------------------------------------------------------===// // Additional Altivec Patterns // -// DS* intrinsics -def : Pat<(int_ppc_altivec_dssall), (DSSALL 1, 0, 0, 0)>; -def : Pat<(int_ppc_altivec_dss imm:$STRM), (DSS 0, imm:$STRM, 0, 0)>; - -// * 32-bit -def : Pat<(int_ppc_altivec_dst i32:$rA, i32:$rB, imm:$STRM), - (DST 0, imm:$STRM, $rA, $rB)>; -def : Pat<(int_ppc_altivec_dstt i32:$rA, i32:$rB, imm:$STRM), - (DSTT 1, imm:$STRM, $rA, $rB)>; -def : Pat<(int_ppc_altivec_dstst i32:$rA, i32:$rB, imm:$STRM), - (DSTST 0, imm:$STRM, $rA, $rB)>; -def : Pat<(int_ppc_altivec_dststt i32:$rA, i32:$rB, imm:$STRM), - (DSTSTT 1, imm:$STRM, $rA, $rB)>; - -// * 64-bit -def : Pat<(int_ppc_altivec_dst i64:$rA, i32:$rB, imm:$STRM), - (DST64 0, imm:$STRM, $rA, $rB)>; -def : Pat<(int_ppc_altivec_dstt i64:$rA, i32:$rB, imm:$STRM), - (DSTT64 1, imm:$STRM, $rA, $rB)>; -def : Pat<(int_ppc_altivec_dstst i64:$rA, i32:$rB, imm:$STRM), - (DSTST64 0, imm:$STRM, $rA, $rB)>; -def : Pat<(int_ppc_altivec_dststt i64:$rA, i32:$rB, imm:$STRM), - (DSTSTT64 1, imm:$STRM, $rA, $rB)>; - // Loads. def : Pat<(v4i32 (load xoaddr:$src)), (LVX xoaddr:$src)>; @@ -714,18 +801,27 @@ def : Pat<(store v4i32:$rS, xoaddr:$dst), def : Pat<(v16i8 (bitconvert (v8i16 VRRC:$src))), (v16i8 VRRC:$src)>; def : Pat<(v16i8 (bitconvert (v4i32 VRRC:$src))), (v16i8 VRRC:$src)>; def : Pat<(v16i8 (bitconvert (v4f32 VRRC:$src))), (v16i8 VRRC:$src)>; +def : Pat<(v16i8 (bitconvert (v2i64 VRRC:$src))), (v16i8 VRRC:$src)>; def : Pat<(v8i16 (bitconvert (v16i8 VRRC:$src))), (v8i16 VRRC:$src)>; def : Pat<(v8i16 (bitconvert (v4i32 VRRC:$src))), (v8i16 VRRC:$src)>; def : Pat<(v8i16 (bitconvert (v4f32 VRRC:$src))), (v8i16 VRRC:$src)>; +def : Pat<(v8i16 (bitconvert (v2i64 VRRC:$src))), (v8i16 VRRC:$src)>; def : Pat<(v4i32 (bitconvert (v16i8 VRRC:$src))), (v4i32 VRRC:$src)>; def : Pat<(v4i32 (bitconvert (v8i16 VRRC:$src))), (v4i32 VRRC:$src)>; def : Pat<(v4i32 (bitconvert (v4f32 VRRC:$src))), (v4i32 VRRC:$src)>; +def : Pat<(v4i32 (bitconvert (v2i64 VRRC:$src))), (v4i32 VRRC:$src)>; def : Pat<(v4f32 (bitconvert (v16i8 VRRC:$src))), (v4f32 VRRC:$src)>; def : Pat<(v4f32 (bitconvert (v8i16 VRRC:$src))), (v4f32 VRRC:$src)>; def : Pat<(v4f32 (bitconvert (v4i32 VRRC:$src))), (v4f32 VRRC:$src)>; +def : Pat<(v4f32 (bitconvert (v2i64 VRRC:$src))), (v4f32 VRRC:$src)>; + +def : Pat<(v2i64 (bitconvert (v16i8 VRRC:$src))), (v2i64 VRRC:$src)>; +def : Pat<(v2i64 (bitconvert (v8i16 VRRC:$src))), (v2i64 VRRC:$src)>; +def : Pat<(v2i64 (bitconvert (v4i32 VRRC:$src))), (v2i64 VRRC:$src)>; +def : Pat<(v2i64 (bitconvert (v4f32 VRRC:$src))), (v2i64 VRRC:$src)>; // Shuffles. @@ -737,6 +833,16 @@ def:Pat<(vpkuwum_unary_shuffle v16i8:$vA, undef), def:Pat<(vpkuhum_unary_shuffle v16i8:$vA, undef), (VPKUHUM $vA, $vA)>; +// Match vsldoi(y,x), vpkuwum(y,x), vpkuhum(y,x), i.e., swapped operands. +// These fragments are matched for little-endian, where the inputs must +// be swapped for correct semantics. +def:Pat<(vsldoi_swapped_shuffle:$in v16i8:$vA, v16i8:$vB), + (VSLDOI $vB, $vA, (VSLDOI_swapped_get_imm $in))>; +def:Pat<(vpkuwum_swapped_shuffle v16i8:$vA, v16i8:$vB), + (VPKUWUM $vB, $vA)>; +def:Pat<(vpkuhum_swapped_shuffle v16i8:$vA, v16i8:$vB), + (VPKUHUM $vB, $vA)>; + // Match vmrg*(x,x) def:Pat<(vmrglb_unary_shuffle v16i8:$vA, undef), (VMRGLB $vA, $vA)>; @@ -751,6 +857,22 @@ def:Pat<(vmrghh_unary_shuffle v16i8:$vA, undef), def:Pat<(vmrghw_unary_shuffle v16i8:$vA, undef), (VMRGHW $vA, $vA)>; +// Match vmrg*(y,x), i.e., swapped operands. These fragments +// are matched for little-endian, where the inputs must be +// swapped for correct semantics. +def:Pat<(vmrglb_swapped_shuffle v16i8:$vA, v16i8:$vB), + (VMRGLB $vB, $vA)>; +def:Pat<(vmrglh_swapped_shuffle v16i8:$vA, v16i8:$vB), + (VMRGLH $vB, $vA)>; +def:Pat<(vmrglw_swapped_shuffle v16i8:$vA, v16i8:$vB), + (VMRGLW $vB, $vA)>; +def:Pat<(vmrghb_swapped_shuffle v16i8:$vA, v16i8:$vB), + (VMRGHB $vB, $vA)>; +def:Pat<(vmrghh_swapped_shuffle v16i8:$vA, v16i8:$vB), + (VMRGHH $vB, $vA)>; +def:Pat<(vmrghw_swapped_shuffle v16i8:$vA, v16i8:$vB), + (VMRGHW $vB, $vA)>; + // Logical Operations def : Pat<(vnot_ppc v4i32:$vA), (VNOR $vA, $vA)>; @@ -826,3 +948,131 @@ def : Pat<(v4f32 (fnearbyint v4f32:$vA)), } // end HasAltivec +def HasP8Altivec : Predicate<"PPCSubTarget->hasP8Altivec()">; +def HasP8Crypto : Predicate<"PPCSubTarget->hasP8Crypto()">; +let Predicates = [HasP8Altivec] in { + +let isCommutable = 1 in { +def VMULESW : VX1_Int_Ty2<904, "vmulesw", int_ppc_altivec_vmulesw, + v2i64, v4i32>; +def VMULEUW : VX1_Int_Ty2<648, "vmuleuw", int_ppc_altivec_vmuleuw, + v2i64, v4i32>; +def VMULOSW : VX1_Int_Ty2<392, "vmulosw", int_ppc_altivec_vmulosw, + v2i64, v4i32>; +def VMULOUW : VX1_Int_Ty2<136, "vmulouw", int_ppc_altivec_vmulouw, + v2i64, v4i32>; +def VMULUWM : VX1_Int_Ty<137, "vmuluwm", int_ppc_altivec_vmuluwm, + v4i32>; +def VMAXSD : VX1_Int_Ty<450, "vmaxsd", int_ppc_altivec_vmaxsd, v2i64>; +def VMAXUD : VX1_Int_Ty<194, "vmaxud", int_ppc_altivec_vmaxud, v2i64>; +def VMINSD : VX1_Int_Ty<962, "vminsd", int_ppc_altivec_vminsd, v2i64>; +def VMIDUD : VX1_Int_Ty<706, "vminud", int_ppc_altivec_vminud, v2i64>; +} // isCommutable + +def VRLD : VX1_Int_Ty<196, "vrld", int_ppc_altivec_vrld, v2i64>; +def VSLD : VX1_Int_Ty<1476, "vsld", int_ppc_altivec_vsld, v2i64>; +def VSRD : VX1_Int_Ty<1732, "vsrd", int_ppc_altivec_vsrd, v2i64>; +def VSRAD : VX1_Int_Ty<964, "vsrad", int_ppc_altivec_vsrad, v2i64>; + + +// Vector Integer Arithmetic Instructions +let isCommutable = 1 in { +def VADDUDM : VXForm_1<192, (outs vrrc:$vD), (ins vrrc:$vA, vrrc:$vB), + "vaddudm $vD, $vA, $vB", IIC_VecGeneral, + [(set v2i64:$vD, (add v2i64:$vA, v2i64:$vB))]>; +} // isCommutable + +def VSUBUDM : VXForm_1<1216, (outs vrrc:$vD), (ins vrrc:$vA, vrrc:$vB), + "vsubudm $vD, $vA, $vB", IIC_VecGeneral, + [(set v2i64:$vD, (sub v2i64:$vA, v2i64:$vB))]>; + +// Count Leading Zeros +def VCLZB : VXForm_2<1794, (outs vrrc:$vD), (ins vrrc:$vB), + "vclzb $vD, $vB", IIC_VecGeneral, + [(set v16i8:$vD, (ctlz v16i8:$vB))]>; +def VCLZH : VXForm_2<1858, (outs vrrc:$vD), (ins vrrc:$vB), + "vclzh $vD, $vB", IIC_VecGeneral, + [(set v8i16:$vD, (ctlz v8i16:$vB))]>; +def VCLZW : VXForm_2<1922, (outs vrrc:$vD), (ins vrrc:$vB), + "vclzw $vD, $vB", IIC_VecGeneral, + [(set v4i32:$vD, (ctlz v4i32:$vB))]>; +def VCLZD : VXForm_2<1986, (outs vrrc:$vD), (ins vrrc:$vB), + "vclzd $vD, $vB", IIC_VecGeneral, + [(set v2i64:$vD, (ctlz v2i64:$vB))]>; + +// Population Count +def VPOPCNTB : VXForm_2<1795, (outs vrrc:$vD), (ins vrrc:$vB), + "vpopcntb $vD, $vB", IIC_VecGeneral, + [(set v16i8:$vD, (ctpop v16i8:$vB))]>; +def VPOPCNTH : VXForm_2<1859, (outs vrrc:$vD), (ins vrrc:$vB), + "vpopcnth $vD, $vB", IIC_VecGeneral, + [(set v8i16:$vD, (ctpop v8i16:$vB))]>; +def VPOPCNTW : VXForm_2<1923, (outs vrrc:$vD), (ins vrrc:$vB), + "vpopcntw $vD, $vB", IIC_VecGeneral, + [(set v4i32:$vD, (ctpop v4i32:$vB))]>; +def VPOPCNTD : VXForm_2<1987, (outs vrrc:$vD), (ins vrrc:$vB), + "vpopcntd $vD, $vB", IIC_VecGeneral, + [(set v2i64:$vD, (ctpop v2i64:$vB))]>; + +let isCommutable = 1 in { +// FIXME: Use AddedComplexity > 400 to ensure these patterns match before the +// VSX equivalents. We need to fix this up at some point. Two possible +// solutions for this problem: +// 1. Disable Altivec patterns that compete with VSX patterns using the +// !HasVSX predicate. This essentially favours VSX over Altivec, in +// hopes of reducing register pressure (larger register set using VSX +// instructions than VMX instructions) +// 2. Employ a more disciplined use of AddedComplexity, which would provide +// more fine-grained control than option 1. This would be beneficial +// if we find situations where Altivec is really preferred over VSX. +def VEQV : VXForm_1<1668, (outs vrrc:$vD), (ins vrrc:$vA, vrrc:$vB), + "veqv $vD, $vA, $vB", IIC_VecGeneral, + [(set v4i32:$vD, (vnot_ppc (xor v4i32:$vA, v4i32:$vB)))]>; +def VNAND : VXForm_1<1412, (outs vrrc:$vD), (ins vrrc:$vA, vrrc:$vB), + "vnand $vD, $vA, $vB", IIC_VecGeneral, + [(set v4i32:$vD, (vnot_ppc (and v4i32:$vA, v4i32:$vB)))]>; +} // isCommutable + +def VORC : VXForm_1<1348, (outs vrrc:$vD), (ins vrrc:$vA, vrrc:$vB), + "vorc $vD, $vA, $vB", IIC_VecGeneral, + [(set v4i32:$vD, (or v4i32:$vA, + (vnot_ppc v4i32:$vB)))]>; + +// i64 element comparisons. +def VCMPEQUD : VCMP <199, "vcmpequd $vD, $vA, $vB" , v2i64>; +def VCMPEQUDo : VCMPo<199, "vcmpequd. $vD, $vA, $vB", v2i64>; +def VCMPGTSD : VCMP <967, "vcmpgtsd $vD, $vA, $vB" , v2i64>; +def VCMPGTSDo : VCMPo<967, "vcmpgtsd. $vD, $vA, $vB", v2i64>; +def VCMPGTUD : VCMP <711, "vcmpgtud $vD, $vA, $vB" , v2i64>; +def VCMPGTUDo : VCMPo<711, "vcmpgtud. $vD, $vA, $vB", v2i64>; + +// The cryptography instructions that do not require Category:Vector.Crypto +def VPMSUMB : VX1_Int_Ty<1032, "vpmsumb", + int_ppc_altivec_crypto_vpmsumb, v16i8>; +def VPMSUMH : VX1_Int_Ty<1096, "vpmsumh", + int_ppc_altivec_crypto_vpmsumh, v8i16>; +def VPMSUMW : VX1_Int_Ty<1160, "vpmsumw", + int_ppc_altivec_crypto_vpmsumw, v4i32>; +def VPMSUMD : VX1_Int_Ty<1224, "vpmsumd", + int_ppc_altivec_crypto_vpmsumd, v2i64>; +def VPERMXOR : VA1a_Int_Ty<45, "vpermxor", + int_ppc_altivec_crypto_vpermxor, v16i8>; + +} // end HasP8Altivec + +// Crypto instructions (from builtins) +let Predicates = [HasP8Crypto] in { +def VSHASIGMAW : VXCR_Int_Ty<1666, "vshasigmaw", + int_ppc_altivec_crypto_vshasigmaw, v4i32>; +def VSHASIGMAD : VXCR_Int_Ty<1730, "vshasigmad", + int_ppc_altivec_crypto_vshasigmad, v2i64>; +def VCIPHER : VX1_Int_Ty<1288, "vcipher", int_ppc_altivec_crypto_vcipher, + v2i64>; +def VCIPHERLAST : VX1_Int_Ty<1289, "vcipherlast", + int_ppc_altivec_crypto_vcipherlast, v2i64>; +def VNCIPHER : VX1_Int_Ty<1352, "vncipher", + int_ppc_altivec_crypto_vncipher, v2i64>; +def VNCIPHERLAST : VX1_Int_Ty<1353, "vncipherlast", + int_ppc_altivec_crypto_vncipherlast, v2i64>; +def VSBOX : VXBX_Int_Ty<1480, "vsbox", int_ppc_altivec_crypto_vsbox, v2i64>; +} // HasP8Crypto