Index: lib/Target/X86/X86InstrAVX512.td
===================================================================
--- lib/Target/X86/X86InstrAVX512.td
+++ lib/Target/X86/X86InstrAVX512.td
@@ -8014,8 +8014,8 @@
 
 let Predicates = [HasAVX512] in
   defm VCVTPH2PSZ : avx512_cvtph2ps<v16f32_info, v16i16x_info, f256mem, loadv4i64,
-                                    WriteCvtPH2PSY>,
-                    avx512_cvtph2ps_sae<v16f32_info, v16i16x_info, WriteCvtPH2PSY>,
+                                    WriteCvtPH2PSZ>,
+                    avx512_cvtph2ps_sae<v16f32_info, v16i16x_info, WriteCvtPH2PSZ>,
                     EVEX, EVEX_V512, EVEX_CD8<32, CD8VH>;
 
 let Predicates = [HasVLX] in {
@@ -8068,8 +8068,8 @@
 
 let Predicates = [HasAVX512] in {
   defm VCVTPS2PHZ : avx512_cvtps2ph<v16i16x_info, v16f32_info, f256mem,
-                                    WriteCvtPS2PHY, WriteCvtPS2PHYSt>,
-                    avx512_cvtps2ph_sae<v16i16x_info, v16f32_info, WriteCvtPS2PH>,
+                                    WriteCvtPS2PHZ, WriteCvtPS2PHZSt>,
+                    avx512_cvtps2ph_sae<v16i16x_info, v16f32_info, WriteCvtPS2PHZ>,
                                         EVEX, EVEX_V512, EVEX_CD8<32, CD8VH>;
   let Predicates = [HasVLX] in {
     defm VCVTPS2PHZ256 : avx512_cvtps2ph<v8i16x_info, v8f32x_info, f128mem,
Index: lib/Target/X86/X86SchedBroadwell.td
===================================================================
--- lib/Target/X86/X86SchedBroadwell.td
+++ lib/Target/X86/X86SchedBroadwell.td
@@ -189,25 +189,31 @@
 defm : BWWriteResPair<WriteFAdd,    [BWPort1],  3, [1], 1, 5>; // Floating point add/sub.
 defm : BWWriteResPair<WriteFAddX,   [BWPort1],  3, [1], 1, 5>; // Floating point add/sub (XMM).
 defm : BWWriteResPair<WriteFAddY,   [BWPort1],  3, [1], 1, 6>; // Floating point add/sub (YMM/ZMM).
+defm : X86WriteResPairUnsupported<WriteFAddZ>;
 defm : BWWriteResPair<WriteFAdd64,  [BWPort1],  3, [1], 1, 5>; // Floating point double add/sub.
 defm : BWWriteResPair<WriteFAdd64X, [BWPort1],  3, [1], 1, 5>; // Floating point double add/sub (XMM).
 defm : BWWriteResPair<WriteFAdd64Y, [BWPort1],  3, [1], 1, 6>; // Floating point double add/sub (YMM/ZMM).
+defm : X86WriteResPairUnsupported<WriteFAdd64Z>;
 
 defm : BWWriteResPair<WriteFCmp,    [BWPort1],  3, [1], 1, 5>; // Floating point compare.
 defm : BWWriteResPair<WriteFCmpX,   [BWPort1],  3, [1], 1, 5>; // Floating point compare (XMM).
 defm : BWWriteResPair<WriteFCmpY,   [BWPort1],  3, [1], 1, 6>; // Floating point compare (YMM/ZMM).
+defm : X86WriteResPairUnsupported<WriteFCmpZ>;
 defm : BWWriteResPair<WriteFCmp64,  [BWPort1],  3, [1], 1, 5>; // Floating point double compare.
 defm : BWWriteResPair<WriteFCmp64X, [BWPort1],  3, [1], 1, 5>; // Floating point double compare (XMM).
 defm : BWWriteResPair<WriteFCmp64Y, [BWPort1],  3, [1], 1, 6>; // Floating point double compare (YMM/ZMM).
+defm : X86WriteResPairUnsupported<WriteFCmp64Z>;
 
 defm : BWWriteResPair<WriteFCom,    [BWPort1],  3>; // Floating point compare to flags.
 
 defm : BWWriteResPair<WriteFMul,    [BWPort01], 3, [1], 1, 5>; // Floating point multiplication.
 defm : BWWriteResPair<WriteFMulX,   [BWPort01], 3, [1], 1, 5>; // Floating point multiplication (XMM).
 defm : BWWriteResPair<WriteFMulY,   [BWPort01], 3, [1], 1, 6>; // Floating point multiplication (YMM/ZMM).
+defm : X86WriteResPairUnsupported<WriteFMulZ>;
 defm : BWWriteResPair<WriteFMul64,  [BWPort01], 3, [1], 1, 5>; // Floating point double multiplication.
 defm : BWWriteResPair<WriteFMul64X, [BWPort01], 3, [1], 1, 5>; // Floating point double multiplication (XMM).
 defm : BWWriteResPair<WriteFMul64Y, [BWPort01], 3, [1], 1, 6>; // Floating point double multiplication (YMM/ZMM).
+defm : X86WriteResPairUnsupported<WriteFMul64Z>;
 
 //defm : BWWriteResPair<WriteFDiv,     [BWPort0,BWFPDivider], 11, [1,3], 1, 5>; // Floating point division.
 defm : BWWriteResPair<WriteFDivX,    [BWPort0,BWFPDivider], 11, [1,5], 1, 5>; // Floating point division (XMM).
@@ -233,34 +239,45 @@
 defm : BWWriteResPair<WriteFRcp,   [BWPort0],  5, [1], 1, 5>; // Floating point reciprocal estimate.
 defm : BWWriteResPair<WriteFRcpX,  [BWPort0],  5, [1], 1, 5>; // Floating point reciprocal estimate (XMM).
 defm : BWWriteResPair<WriteFRcpY,  [BWPort0,BWPort015], 11, [2,1], 3, 6>; // Floating point reciprocal estimate (YMM/ZMM).
+defm : X86WriteResPairUnsupported<WriteFRcpZ>;
 
 defm : BWWriteResPair<WriteFRsqrt, [BWPort0],  5, [1], 1, 5>; // Floating point reciprocal square root estimate.
 defm : BWWriteResPair<WriteFRsqrtX,[BWPort0],  5, [1], 1, 5>; // Floating point reciprocal square root estimate (XMM).
 defm : BWWriteResPair<WriteFRsqrtY,[BWPort0,BWPort015], 11, [2,1], 3, 6>; // Floating point reciprocal square root estimate (YMM/ZMM).
+defm : X86WriteResPairUnsupported<WriteFRsqrtZ>;
 
 defm : BWWriteResPair<WriteFMA,    [BWPort01], 5, [1], 1, 5>; // Fused Multiply Add.
 defm : BWWriteResPair<WriteFMAX,   [BWPort01], 5, [1], 1, 5>; // Fused Multiply Add (XMM).
 defm : BWWriteResPair<WriteFMAY,   [BWPort01], 5, [1], 1, 6>; // Fused Multiply Add (YMM/ZMM).
+defm : X86WriteResPairUnsupported<WriteFMAZ>;
 defm : BWWriteResPair<WriteDPPD,   [BWPort0,BWPort1,BWPort5],  9, [1,1,1], 3, 5>; // Floating point double dot product.
 defm : BWWriteResPair<WriteDPPS,   [BWPort0,BWPort1,BWPort5], 14, [2,1,1], 4, 5>; // Floating point single dot product.
 defm : BWWriteResPair<WriteDPPSY,  [BWPort0,BWPort1,BWPort5], 14, [2,1,1], 4, 6>; // Floating point single dot product (YMM).
+defm : X86WriteResPairUnsupported<WriteDPPSZ>;
 defm : BWWriteResPair<WriteFSign,     [BWPort5], 1>; // Floating point fabs/fchs.
 defm : X86WriteRes<WriteFRnd,            [BWPort23],  6, [1],   1>; // Floating point rounding.
 defm : X86WriteRes<WriteFRndY,           [BWPort23],  6, [1],   1>; // Floating point rounding (YMM/ZMM).
+defm : X86WriteResPairUnsupported<WriteFRndZ>;
 defm : X86WriteRes<WriteFRndLd,  [BWPort1,BWPort23], 11, [2,1], 3>;
 defm : X86WriteRes<WriteFRndYLd, [BWPort1,BWPort23], 12, [2,1], 3>;
 defm : BWWriteResPair<WriteFLogic,    [BWPort5], 1, [1], 1, 5>; // Floating point and/or/xor logicals.
 defm : BWWriteResPair<WriteFLogicY,   [BWPort5], 1, [1], 1, 6>; // Floating point and/or/xor logicals (YMM/ZMM).
+defm : X86WriteResPairUnsupported<WriteFLogicZ>;
 defm : BWWriteResPair<WriteFTest,     [BWPort0], 1, [1], 1, 5>; // Floating point TEST instructions.
 defm : BWWriteResPair<WriteFTestY,    [BWPort0], 1, [1], 1, 6>; // Floating point TEST instructions (YMM/ZMM).
+defm : X86WriteResPairUnsupported<WriteFTestZ>;
 defm : BWWriteResPair<WriteFShuffle,  [BWPort5], 1, [1], 1, 5>; // Floating point vector shuffles.
 defm : BWWriteResPair<WriteFShuffleY, [BWPort5], 1, [1], 1, 6>; // Floating point vector shuffles (YMM/ZMM).
+defm : X86WriteResPairUnsupported<WriteFShuffleZ>;
 defm : BWWriteResPair<WriteFVarShuffle,  [BWPort5], 1, [1], 1, 5>; // Floating point vector variable shuffles.
 defm : BWWriteResPair<WriteFVarShuffleY, [BWPort5], 1, [1], 1, 6>; // Floating point vector variable shuffles.
+defm : X86WriteResPairUnsupported<WriteFVarShuffleZ>;
 defm : BWWriteResPair<WriteFBlend,  [BWPort015], 1, [1], 1, 5>; // Floating point vector blends.
 defm : BWWriteResPair<WriteFBlendY, [BWPort015], 1, [1], 1, 6>; // Floating point vector blends.
+defm : X86WriteResPairUnsupported<WriteFBlendZ>;
 defm : BWWriteResPair<WriteFVarBlend,  [BWPort5], 2, [2], 2, 5>; // Fp vector variable blends.
 defm : BWWriteResPair<WriteFVarBlendY, [BWPort5], 2, [2], 2, 6>; // Fp vector variable blends.
+defm : X86WriteResPairUnsupported<WriteFVarBlendZ>;
 
 // FMA Scheduling helper class.
 // class FMASC { X86FoldableSchedWrite Sched = WriteFAdd; }
@@ -291,31 +308,42 @@
 defm : BWWriteResPair<WriteVecALU,   [BWPort15], 1, [1], 1, 5>; // Vector integer ALU op, no logicals.
 defm : BWWriteResPair<WriteVecALUX,  [BWPort15], 1, [1], 1, 5>; // Vector integer ALU op, no logicals.
 defm : BWWriteResPair<WriteVecALUY,  [BWPort15], 1, [1], 1, 6>; // Vector integer ALU op, no logicals (YMM/ZMM).
+defm : X86WriteResPairUnsupported<WriteVecALUZ>;
 defm : BWWriteResPair<WriteVecLogic, [BWPort015], 1, [1], 1, 5>; // Vector integer and/or/xor.
 defm : BWWriteResPair<WriteVecLogicX,[BWPort015], 1, [1], 1, 5>; // Vector integer and/or/xor.
 defm : BWWriteResPair<WriteVecLogicY,[BWPort015], 1, [1], 1, 6>; // Vector integer and/or/xor (YMM/ZMM).
+defm : X86WriteResPairUnsupported<WriteVecLogicZ>;
 defm : BWWriteResPair<WriteVecTest,  [BWPort0,BWPort5], 2, [1,1], 2, 5>; // Vector integer TEST instructions.
 defm : BWWriteResPair<WriteVecTestY, [BWPort0,BWPort5], 4, [1,1], 2, 6>; // Vector integer TEST instructions (YMM/ZMM).
+defm : X86WriteResPairUnsupported<WriteVecTestZ>;
 defm : BWWriteResPair<WriteVecIMul,  [BWPort0],  5, [1], 1, 5>; // Vector integer multiply.
 defm : BWWriteResPair<WriteVecIMulX, [BWPort0],  5, [1], 1, 5>; // Vector integer multiply.
 defm : BWWriteResPair<WriteVecIMulY, [BWPort0],  5, [1], 1, 6>; // Vector integer multiply.
+defm : X86WriteResPairUnsupported<WriteVecIMulZ>;
 defm : BWWriteResPair<WritePMULLD,   [BWPort0], 10, [2], 2, 5>; // Vector PMULLD.
 defm : BWWriteResPair<WritePMULLDY,  [BWPort0], 10, [2], 2, 6>; // Vector PMULLD (YMM/ZMM).
+defm : X86WriteResPairUnsupported<WritePMULLDZ>;
 defm : BWWriteResPair<WriteShuffle,  [BWPort5], 1, [1], 1, 5>; // Vector shuffles.
 defm : BWWriteResPair<WriteShuffleX, [BWPort5], 1, [1], 1, 5>; // Vector shuffles.
 defm : BWWriteResPair<WriteShuffleY, [BWPort5], 1, [1], 1, 6>; // Vector shuffles (YMM/ZMM).
+defm : X86WriteResPairUnsupported<WriteShuffleZ>;
 defm : BWWriteResPair<WriteVarShuffle, [BWPort5], 1, [1], 1, 5>; // Vector variable shuffles.
 defm : BWWriteResPair<WriteVarShuffleX,[BWPort5], 1, [1], 1, 5>; // Vector variable shuffles.
 defm : BWWriteResPair<WriteVarShuffleY,[BWPort5], 1, [1], 1, 6>; // Vector variable shuffles (YMM/ZMM).
+defm : X86WriteResPairUnsupported<WriteVarShuffleZ>;
 defm : BWWriteResPair<WriteBlend,  [BWPort5], 1, [1], 1, 5>; // Vector blends.
 defm : BWWriteResPair<WriteBlendY, [BWPort5], 1, [1], 1, 6>; // Vector blends (YMM/ZMM).
+defm : X86WriteResPairUnsupported<WriteBlendZ>;
 defm : BWWriteResPair<WriteVarBlend,  [BWPort5], 2, [2], 2, 5>; // Vector variable blends.
 defm : BWWriteResPair<WriteVarBlendY, [BWPort5], 2, [2], 2, 6>; // Vector variable blends (YMM/ZMM).
+defm : X86WriteResPairUnsupported<WriteVarBlendZ>;
 defm : BWWriteResPair<WriteMPSAD,  [BWPort0, BWPort5], 7, [1, 2], 3, 5>; // Vector MPSAD.
 defm : BWWriteResPair<WriteMPSADY, [BWPort0, BWPort5], 7, [1, 2], 3, 6>; // Vector MPSAD.
+defm : X86WriteResPairUnsupported<WriteMPSADZ>;
 defm : BWWriteResPair<WritePSADBW,   [BWPort0], 5, [1], 1, 5>; // Vector PSADBW.
 defm : BWWriteResPair<WritePSADBWX,  [BWPort0], 5, [1], 1, 5>; // Vector PSADBW.
 defm : BWWriteResPair<WritePSADBWY,  [BWPort0], 5, [1], 1, 6>; // Vector PSADBW (YMM/ZMM).
+defm : X86WriteResPairUnsupported<WritePSADBWZ>;
 defm : BWWriteResPair<WritePHMINPOS, [BWPort0], 5>; // Vector PHMINPOS.
 
 // Vector integer shifts.
@@ -323,12 +351,15 @@
 defm : BWWriteResPair<WriteVecShiftX,    [BWPort0,BWPort5],  2, [1,1], 2, 5>;
 defm : X86WriteRes<WriteVecShiftY,       [BWPort0,BWPort5],  4, [1,1], 2>;
 defm : X86WriteRes<WriteVecShiftYLd,     [BWPort0,BWPort23], 7, [1,1], 2>;
+defm : X86WriteResPairUnsupported<WriteVecShiftZ>;
 
 defm : BWWriteResPair<WriteVecShiftImm,  [BWPort0],  1, [1], 1, 5>;
 defm : BWWriteResPair<WriteVecShiftImmX, [BWPort0],  1, [1], 1, 5>; // Vector integer immediate shifts (XMM).
 defm : BWWriteResPair<WriteVecShiftImmY, [BWPort0],  1, [1], 1, 6>; // Vector integer immediate shifts (YMM/ZMM).
+defm : X86WriteResPairUnsupported<WriteVecShiftImmZ>;
 defm : BWWriteResPair<WriteVarVecShift,  [BWPort0, BWPort5], 3, [2,1], 3, 5>; // Variable vector shifts.
 defm : BWWriteResPair<WriteVarVecShiftY, [BWPort0, BWPort5], 3, [2,1], 3, 6>; // Variable vector shifts (YMM/ZMM).
+defm : X86WriteResPairUnsupported<WriteVarVecShiftZ>;
 
 // Vector insert/extract operations.
 def : WriteRes<WriteVecInsert, [BWPort5]> {
@@ -354,33 +385,43 @@
 defm : BWWriteResPair<WriteCvtSS2I,   [BWPort1], 3>;
 defm : BWWriteResPair<WriteCvtPS2I,   [BWPort1], 3>;
 defm : BWWriteResPair<WriteCvtPS2IY,  [BWPort1], 3>;
+defm : X86WriteResPairUnsupported<WriteCvtPS2IZ>;
 defm : BWWriteResPair<WriteCvtSD2I,   [BWPort1], 3>;
 defm : BWWriteResPair<WriteCvtPD2I,   [BWPort1], 3>;
 defm : BWWriteResPair<WriteCvtPD2IY,  [BWPort1], 3>;
+defm : X86WriteResPairUnsupported<WriteCvtPD2IZ>;
 
 defm : BWWriteResPair<WriteCvtI2SS,   [BWPort1], 4>;
 defm : BWWriteResPair<WriteCvtI2PS,   [BWPort1], 4>;
 defm : BWWriteResPair<WriteCvtI2PSY,  [BWPort1], 4>;
+defm : X86WriteResPairUnsupported<WriteCvtI2PSZ>;
 defm : BWWriteResPair<WriteCvtI2SD,   [BWPort1], 4>;
 defm : BWWriteResPair<WriteCvtI2PD,   [BWPort1], 4>;
 defm : BWWriteResPair<WriteCvtI2PDY,  [BWPort1], 4>;
+defm : X86WriteResPairUnsupported<WriteCvtI2PDZ>;
 
 defm : BWWriteResPair<WriteCvtSS2SD,  [BWPort1], 3>;
 defm : BWWriteResPair<WriteCvtPS2PD,  [BWPort1], 3>;
 defm : BWWriteResPair<WriteCvtPS2PDY, [BWPort1], 3>;
+defm : X86WriteResPairUnsupported<WriteCvtPS2PDZ>;
 defm : BWWriteResPair<WriteCvtSD2SS,  [BWPort1], 3>;
 defm : BWWriteResPair<WriteCvtPD2PS,  [BWPort1], 3>;
 defm : BWWriteResPair<WriteCvtPD2PSY, [BWPort1], 3>;
+defm : X86WriteResPairUnsupported<WriteCvtPD2PSZ>;
 
 defm : X86WriteRes<WriteCvtPH2PS,     [BWPort0,BWPort5], 2, [1,1], 2>;
 defm : X86WriteRes<WriteCvtPH2PSY,    [BWPort0,BWPort5], 2, [1,1], 2>;
+defm : X86WriteResUnsupported<WriteCvtPH2PSZ>;
 defm : X86WriteRes<WriteCvtPH2PSLd,  [BWPort0,BWPort23], 6, [1,1], 2>;
 defm : X86WriteRes<WriteCvtPH2PSYLd, [BWPort0,BWPort23], 6, [1,1], 2>;
+defm : X86WriteResUnsupported<WriteCvtPH2PSZLd>;
 
 defm : X86WriteRes<WriteCvtPS2PH,    [BWPort1,BWPort5], 4, [1,1], 2>;
 defm : X86WriteRes<WriteCvtPS2PHY,   [BWPort1,BWPort5], 6, [1,1], 2>;
+defm : X86WriteResUnsupported<WriteCvtPS2PHZ>;
 defm : X86WriteRes<WriteCvtPS2PHSt,  [BWPort1,BWPort4,BWPort237], 5, [1,1,1], 3>;
 defm : X86WriteRes<WriteCvtPS2PHYSt, [BWPort1,BWPort4,BWPort237], 7, [1,1,1], 3>;
+defm : X86WriteResUnsupported<WriteCvtPS2PHZSt>;
 
 // Strings instructions.
 
Index: lib/Target/X86/X86SchedHaswell.td
===================================================================
--- lib/Target/X86/X86SchedHaswell.td
+++ lib/Target/X86/X86SchedHaswell.td
@@ -185,25 +185,31 @@
 defm : HWWriteResPair<WriteFAdd,    [HWPort1],  3, [1], 1, 5>;
 defm : HWWriteResPair<WriteFAddX,   [HWPort1],  3, [1], 1, 6>;
 defm : HWWriteResPair<WriteFAddY,   [HWPort1],  3, [1], 1, 7>;
+defm : HWWriteResPair<WriteFAddZ,   [HWPort1],  3, [1], 1, 7>; // Unsupported = 1
 defm : HWWriteResPair<WriteFAdd64,  [HWPort1],  3, [1], 1, 5>;
 defm : HWWriteResPair<WriteFAdd64X, [HWPort1],  3, [1], 1, 6>;
 defm : HWWriteResPair<WriteFAdd64Y, [HWPort1],  3, [1], 1, 7>;
+defm : HWWriteResPair<WriteFAdd64Z, [HWPort1],  3, [1], 1, 7>; // Unsupported = 1
 
 defm : HWWriteResPair<WriteFCmp,    [HWPort1],  3, [1], 1, 5>;
 defm : HWWriteResPair<WriteFCmpX,   [HWPort1],  3, [1], 1, 6>;
 defm : HWWriteResPair<WriteFCmpY,   [HWPort1],  3, [1], 1, 7>;
+defm : HWWriteResPair<WriteFCmpZ,   [HWPort1],  3, [1], 1, 7>; // Unsupported = 1
 defm : HWWriteResPair<WriteFCmp64,  [HWPort1],  3, [1], 1, 5>;
 defm : HWWriteResPair<WriteFCmp64X, [HWPort1],  3, [1], 1, 6>;
 defm : HWWriteResPair<WriteFCmp64Y, [HWPort1],  3, [1], 1, 7>;
+defm : HWWriteResPair<WriteFCmp64Z, [HWPort1],  3, [1], 1, 7>; // Unsupported = 1
 
 defm : HWWriteResPair<WriteFCom,    [HWPort1],  3>;
 
 defm : HWWriteResPair<WriteFMul,    [HWPort01],  5, [1], 1, 5>;
 defm : HWWriteResPair<WriteFMulX,   [HWPort01],  5, [1], 1, 6>;
 defm : HWWriteResPair<WriteFMulY,   [HWPort01],  5, [1], 1, 7>;
+defm : HWWriteResPair<WriteFMulZ,   [HWPort01],  5, [1], 1, 7>; // Unsupported = 1
 defm : HWWriteResPair<WriteFMul64,  [HWPort01],  5, [1], 1, 5>;
 defm : HWWriteResPair<WriteFMul64X, [HWPort01],  5, [1], 1, 6>;
 defm : HWWriteResPair<WriteFMul64Y, [HWPort01],  5, [1], 1, 7>;
+defm : HWWriteResPair<WriteFMul64Z, [HWPort01],  5, [1], 1, 7>; // Unsupported = 1
 
 defm : HWWriteResPair<WriteFDiv,    [HWPort0,HWFPDivider], 13, [1,7], 1, 5>;
 defm : HWWriteResPair<WriteFDivX,   [HWPort0,HWFPDivider], 13, [1,7], 1, 6>;
@@ -217,10 +223,12 @@
 defm : HWWriteResPair<WriteFRcp,   [HWPort0],  5, [1], 1, 5>;
 defm : HWWriteResPair<WriteFRcpX,  [HWPort0],  5, [1], 1, 6>;
 defm : HWWriteResPair<WriteFRcpY,  [HWPort0,HWPort015], 11, [2,1], 3, 7>;
+defm : HWWriteResPair<WriteFRcpZ,  [HWPort0,HWPort015], 11, [2,1], 3, 7>; // Unsupported = 1
 
 defm : HWWriteResPair<WriteFRsqrt, [HWPort0],  5, [1], 1, 5>;
 defm : HWWriteResPair<WriteFRsqrtX,[HWPort0],  5, [1], 1, 6>;
 defm : HWWriteResPair<WriteFRsqrtY,[HWPort0,HWPort015], 11, [2,1], 3, 7>;
+defm : HWWriteResPair<WriteFRsqrtZ,[HWPort0,HWPort015], 11, [2,1], 3, 7>; // Unsupported = 1
 
 defm : HWWriteResPair<WriteFSqrt,    [HWPort0,HWFPDivider], 11, [1,7], 1, 5>;
 defm : HWWriteResPair<WriteFSqrtX,   [HWPort0,HWFPDivider], 11, [1,7], 1, 6>;
@@ -235,60 +243,80 @@
 defm : HWWriteResPair<WriteFMA,   [HWPort01], 5, [1], 1, 5>;
 defm : HWWriteResPair<WriteFMAX,  [HWPort01], 5, [1], 1, 6>;
 defm : HWWriteResPair<WriteFMAY,  [HWPort01], 5, [1], 1, 7>;
+defm : HWWriteResPair<WriteFMAZ,  [HWPort01], 5, [1], 1, 7>; // Unsupported = 1
 defm : HWWriteResPair<WriteDPPD,  [HWPort0,HWPort1,HWPort5],  9, [1,1,1], 3, 6>;
 defm : HWWriteResPair<WriteDPPS,  [HWPort0,HWPort1,HWPort5], 14, [2,1,1], 4, 6>;
 defm : HWWriteResPair<WriteDPPSY, [HWPort0,HWPort1,HWPort5], 14, [2,1,1], 4, 7>;
+defm : HWWriteResPair<WriteDPPSZ, [HWPort0,HWPort1,HWPort5], 14, [2,1,1], 4, 7>; // Unsupported = 1
 defm : HWWriteResPair<WriteFSign,  [HWPort0], 1>;
 defm : X86WriteRes<WriteFRnd,            [HWPort23],  6, [1],   1>;
 defm : X86WriteRes<WriteFRndY,           [HWPort23],  6, [1],   1>;
+defm : X86WriteRes<WriteFRndZ,           [HWPort23],  6, [1],   1>; // Unsupported = 1
 defm : X86WriteRes<WriteFRndLd,  [HWPort1,HWPort23], 12, [2,1], 3>;
 defm : X86WriteRes<WriteFRndYLd, [HWPort1,HWPort23], 13, [2,1], 3>;
+defm : X86WriteRes<WriteFRndZLd, [HWPort1,HWPort23], 13, [2,1], 3>; // Unsupported = 1
 defm : HWWriteResPair<WriteFLogic,  [HWPort5], 1, [1], 1, 6>;
 defm : HWWriteResPair<WriteFLogicY, [HWPort5], 1, [1], 1, 7>;
+defm : HWWriteResPair<WriteFLogicZ, [HWPort5], 1, [1], 1, 7>; // Unsupported = 1
 defm : HWWriteResPair<WriteFTest,   [HWPort0], 1, [1], 1, 6>;
 defm : HWWriteResPair<WriteFTestY,  [HWPort0], 1, [1], 1, 7>;
+defm : HWWriteResPair<WriteFTestZ,  [HWPort0], 1, [1], 1, 7>; // Unsupported = 1
 defm : HWWriteResPair<WriteFShuffle,  [HWPort5], 1, [1], 1, 6>;
 defm : HWWriteResPair<WriteFShuffleY, [HWPort5], 1, [1], 1, 7>;
+defm : HWWriteResPair<WriteFShuffleZ, [HWPort5], 1, [1], 1, 7>; // Unsupported = 1
 defm : HWWriteResPair<WriteFVarShuffle,  [HWPort5], 1, [1], 1, 6>;
 defm : HWWriteResPair<WriteFVarShuffleY, [HWPort5], 1, [1], 1, 7>;
+defm : HWWriteResPair<WriteFVarShuffleZ, [HWPort5], 1, [1], 1, 7>; // Unsupported = 1
 defm : HWWriteResPair<WriteFBlend,  [HWPort015], 1, [1], 1, 6>;
 defm : HWWriteResPair<WriteFBlendY, [HWPort015], 1, [1], 1, 7>;
+defm : HWWriteResPair<WriteFBlendZ, [HWPort015], 1, [1], 1, 7>; // Unsupported = 1
 defm : HWWriteResPair<WriteFShuffle256, [HWPort5], 3, [1], 1, 7>;
 defm : HWWriteResPair<WriteFVarShuffle256, [HWPort5], 3, [1], 1, 7>;
 defm : HWWriteResPair<WriteFVarBlend,  [HWPort5], 2, [2], 2, 6>;
 defm : HWWriteResPair<WriteFVarBlendY, [HWPort5], 2, [2], 2, 7>;
+defm : HWWriteResPair<WriteFVarBlendZ, [HWPort5], 2, [2], 2, 7>; // Unsupported = 1
 
 // Conversion between integer and float.
 defm : HWWriteResPair<WriteCvtSD2I,   [HWPort1], 3>;
 defm : HWWriteResPair<WriteCvtPD2I,   [HWPort1], 3>;
 defm : HWWriteResPair<WriteCvtPD2IY,  [HWPort1], 3>;
+defm : HWWriteResPair<WriteCvtPD2IZ,  [HWPort1], 3>; // Unsupported = 1
 defm : HWWriteResPair<WriteCvtSS2I,   [HWPort1], 3>;
 defm : HWWriteResPair<WriteCvtPS2I,   [HWPort1], 3>;
 defm : HWWriteResPair<WriteCvtPS2IY,  [HWPort1], 3>;
+defm : HWWriteResPair<WriteCvtPS2IZ,  [HWPort1], 3>; // Unsupported = 1
 
 defm : HWWriteResPair<WriteCvtI2SD,   [HWPort1], 4>;
 defm : HWWriteResPair<WriteCvtI2PD,   [HWPort1], 4>;
 defm : HWWriteResPair<WriteCvtI2PDY,  [HWPort1], 4>;
+defm : HWWriteResPair<WriteCvtI2PDZ,  [HWPort1], 4>; // Unsupported = 1
 defm : HWWriteResPair<WriteCvtI2SS,   [HWPort1], 4>;
 defm : HWWriteResPair<WriteCvtI2PS,   [HWPort1], 4>;
 defm : HWWriteResPair<WriteCvtI2PSY,  [HWPort1], 4>;
+defm : HWWriteResPair<WriteCvtI2PSZ,  [HWPort1], 4>; // Unsupported = 1
 
 defm : HWWriteResPair<WriteCvtSS2SD,  [HWPort1], 3>;
 defm : HWWriteResPair<WriteCvtPS2PD,  [HWPort1], 3>;
 defm : HWWriteResPair<WriteCvtPS2PDY, [HWPort1], 3>;
+defm : HWWriteResPair<WriteCvtPS2PDZ, [HWPort1], 3>; // Unsupported = 1
 defm : HWWriteResPair<WriteCvtSD2SS,  [HWPort1], 3>;
 defm : HWWriteResPair<WriteCvtPD2PS,  [HWPort1], 3>;
 defm : HWWriteResPair<WriteCvtPD2PSY, [HWPort1], 3>;
+defm : HWWriteResPair<WriteCvtPD2PSZ, [HWPort1], 3>; // Unsupported = 1
 
 defm : X86WriteRes<WriteCvtPH2PS,     [HWPort0,HWPort5], 2, [1,1], 2>;
 defm : X86WriteRes<WriteCvtPH2PSY,    [HWPort0,HWPort5], 2, [1,1], 2>;
+defm : X86WriteRes<WriteCvtPH2PSZ,    [HWPort0,HWPort5], 2, [1,1], 2>; // Unsupported = 1
 defm : X86WriteRes<WriteCvtPH2PSLd,  [HWPort0,HWPort23], 6, [1,1], 2>;
 defm : X86WriteRes<WriteCvtPH2PSYLd, [HWPort0,HWPort23], 7, [1,1], 2>;
+defm : X86WriteRes<WriteCvtPH2PSZLd, [HWPort0,HWPort23], 7, [1,1], 2>; // Unsupported = 1
 
 defm : X86WriteRes<WriteCvtPS2PH,    [HWPort1,HWPort5], 4, [1,1], 2>;
 defm : X86WriteRes<WriteCvtPS2PHY,   [HWPort1,HWPort5], 6, [1,1], 2>;
+defm : X86WriteRes<WriteCvtPS2PHZ,   [HWPort1,HWPort5], 6, [1,1], 2>; // Unsupported = 1
 defm : X86WriteRes<WriteCvtPS2PHSt,  [HWPort1,HWPort4,HWPort5,HWPort237], 5, [1,1,1,1], 4>;
 defm : X86WriteRes<WriteCvtPS2PHYSt, [HWPort1,HWPort4,HWPort5,HWPort237], 7, [1,1,1,1], 4>;
+defm : X86WriteRes<WriteCvtPS2PHZSt, [HWPort1,HWPort4,HWPort5,HWPort237], 7, [1,1,1,1], 4>; // Unsupported = 1
 
 // Vector integer operations.
 defm : X86WriteRes<WriteVecLoad,         [HWPort23], 5, [1], 1>;
@@ -314,46 +342,61 @@
 defm : HWWriteResPair<WriteVecLogic, [HWPort015], 1, [1], 1, 5>;
 defm : HWWriteResPair<WriteVecLogicX,[HWPort015], 1, [1], 1, 6>;
 defm : HWWriteResPair<WriteVecLogicY,[HWPort015], 1, [1], 1, 7>;
+defm : HWWriteResPair<WriteVecLogicZ,[HWPort015], 1, [1], 1, 7>; // Unsupported = 1
 defm : HWWriteResPair<WriteVecTest,  [HWPort0,HWPort5], 2, [1,1], 2, 6>;
 defm : HWWriteResPair<WriteVecTestY, [HWPort0,HWPort5], 4, [1,1], 2, 7>;
+defm : HWWriteResPair<WriteVecTestZ, [HWPort0,HWPort5], 4, [1,1], 2, 7>; // Unsupported = 1
 defm : HWWriteResPair<WriteVecALU,   [HWPort15],  1, [1], 1, 5>;
 defm : HWWriteResPair<WriteVecALUX,  [HWPort15],  1, [1], 1, 6>;
 defm : HWWriteResPair<WriteVecALUY,  [HWPort15],  1, [1], 1, 7>;
+defm : HWWriteResPair<WriteVecALUZ,  [HWPort15],  1, [1], 1, 7>; // Unsupported = 1
 defm : HWWriteResPair<WriteVecIMul,  [HWPort0],  5, [1], 1, 5>;
 defm : HWWriteResPair<WriteVecIMulX, [HWPort0],  5, [1], 1, 6>;
 defm : HWWriteResPair<WriteVecIMulY, [HWPort0],  5, [1], 1, 7>;
+defm : HWWriteResPair<WriteVecIMulZ, [HWPort0],  5, [1], 1, 7>; // Unsupported = 1
 defm : HWWriteResPair<WritePMULLD,   [HWPort0], 10, [2], 2, 6>;
 defm : HWWriteResPair<WritePMULLDY,  [HWPort0], 10, [2], 2, 7>;
+defm : HWWriteResPair<WritePMULLDZ,  [HWPort0], 10, [2], 2, 7>; // Unsupported = 1
 defm : HWWriteResPair<WriteShuffle,  [HWPort5],  1, [1], 1, 5>;
 defm : HWWriteResPair<WriteShuffleX, [HWPort5],  1, [1], 1, 6>;
 defm : HWWriteResPair<WriteShuffleY, [HWPort5],  1, [1], 1, 7>;
+defm : HWWriteResPair<WriteShuffleZ, [HWPort5],  1, [1], 1, 7>; // Unsupported = 1
 defm : HWWriteResPair<WriteVarShuffle, [HWPort5], 1, [1], 1, 5>;
 defm : HWWriteResPair<WriteVarShuffleX,[HWPort5], 1, [1], 1, 6>;
 defm : HWWriteResPair<WriteVarShuffleY,[HWPort5], 1, [1], 1, 7>;
+defm : HWWriteResPair<WriteVarShuffleZ,[HWPort5], 1, [1], 1, 7>; // Unsupported = 1
 defm : HWWriteResPair<WriteBlend,  [HWPort5], 1, [1], 1, 6>;
 defm : HWWriteResPair<WriteBlendY, [HWPort5], 1, [1], 1, 7>;
+defm : HWWriteResPair<WriteBlendZ, [HWPort5], 1, [1], 1, 7>; // Unsupported = 1
 defm : HWWriteResPair<WriteShuffle256, [HWPort5], 3, [1], 1, 7>;
 defm : HWWriteResPair<WriteVarShuffle256, [HWPort5], 3, [1], 1, 7>;
 defm : HWWriteResPair<WriteVarBlend,  [HWPort5], 2, [2], 2, 6>;
 defm : HWWriteResPair<WriteVarBlendY, [HWPort5], 2, [2], 2, 7>;
+defm : HWWriteResPair<WriteVarBlendZ, [HWPort5], 2, [2], 2, 7>; // Unsupported = 1
 defm : HWWriteResPair<WriteMPSAD,  [HWPort0, HWPort5], 7, [1, 2], 3, 6>;
 defm : HWWriteResPair<WriteMPSADY, [HWPort0, HWPort5], 7, [1, 2], 3, 7>;
+defm : HWWriteResPair<WriteMPSADZ, [HWPort0, HWPort5], 7, [1, 2], 3, 7>; // Unsupported = 1
 defm : HWWriteResPair<WritePSADBW,  [HWPort0], 5, [1], 1, 5>;
 defm : HWWriteResPair<WritePSADBWX, [HWPort0], 5, [1], 1, 6>;
 defm : HWWriteResPair<WritePSADBWY, [HWPort0], 5, [1], 1, 7>;
+defm : HWWriteResPair<WritePSADBWZ, [HWPort0], 5, [1], 1, 7>; // Unsupported = 1
 defm : HWWriteResPair<WritePHMINPOS, [HWPort0],  5, [1], 1, 6>;
 
 // Vector integer shifts.
 defm : HWWriteResPair<WriteVecShift,     [HWPort0], 1, [1], 1, 5>;
 defm : HWWriteResPair<WriteVecShiftX,    [HWPort0,HWPort5],  2, [1,1], 2, 6>;
 defm : X86WriteRes<WriteVecShiftY,       [HWPort0,HWPort5],  4, [1,1], 2>;
+defm : X86WriteRes<WriteVecShiftZ,       [HWPort0,HWPort5],  4, [1,1], 2>; // Unsupported = 1
 defm : X86WriteRes<WriteVecShiftYLd,     [HWPort0,HWPort23], 8, [1,1], 2>;
+defm : X86WriteRes<WriteVecShiftZLd,     [HWPort0,HWPort23], 8, [1,1], 2>; // Unsupported = 1
 
 defm : HWWriteResPair<WriteVecShiftImm,  [HWPort0], 1, [1], 1, 5>;
 defm : HWWriteResPair<WriteVecShiftImmX, [HWPort0], 1, [1], 1, 6>;
 defm : HWWriteResPair<WriteVecShiftImmY, [HWPort0], 1, [1], 1, 7>;
+defm : HWWriteResPair<WriteVecShiftImmZ, [HWPort0], 1, [1], 1, 7>; // Unsupported = 1
 defm : HWWriteResPair<WriteVarVecShift,  [HWPort0, HWPort5], 3, [2,1], 3, 6>;
 defm : HWWriteResPair<WriteVarVecShiftY, [HWPort0, HWPort5], 3, [2,1], 3, 7>;
+defm : HWWriteResPair<WriteVarVecShiftZ, [HWPort0, HWPort5], 3, [2,1], 3, 7>; // Unsupported = 1
 
 // Vector insert/extract operations.
 def : WriteRes<WriteVecInsert, [HWPort5]> {
Index: lib/Target/X86/X86SchedSandyBridge.td
===================================================================
--- lib/Target/X86/X86SchedSandyBridge.td
+++ lib/Target/X86/X86SchedSandyBridge.td
@@ -176,25 +176,31 @@
 defm : SBWriteResPair<WriteFAdd,    [SBPort1],  3, [1], 1, 6>;
 defm : SBWriteResPair<WriteFAddX,   [SBPort1],  3, [1], 1, 6>;
 defm : SBWriteResPair<WriteFAddY,   [SBPort1],  3, [1], 1, 7>;
+defm : SBWriteResPair<WriteFAddZ,   [SBPort1],  3, [1], 1, 7>; // Unsupported = 1
 defm : SBWriteResPair<WriteFAdd64,  [SBPort1],  3, [1], 1, 6>;
 defm : SBWriteResPair<WriteFAdd64X, [SBPort1],  3, [1], 1, 6>;
 defm : SBWriteResPair<WriteFAdd64Y, [SBPort1],  3, [1], 1, 7>;
+defm : SBWriteResPair<WriteFAdd64Z, [SBPort1],  3, [1], 1, 7>; // Unsupported = 1
 
 defm : SBWriteResPair<WriteFCmp,    [SBPort1],  3, [1], 1, 6>;
 defm : SBWriteResPair<WriteFCmpX,   [SBPort1],  3, [1], 1, 6>;
 defm : SBWriteResPair<WriteFCmpY,   [SBPort1],  3, [1], 1, 7>;
+defm : SBWriteResPair<WriteFCmpZ,   [SBPort1],  3, [1], 1, 7>; // Unsupported = 1
 defm : SBWriteResPair<WriteFCmp64,  [SBPort1],  3, [1], 1, 6>;
 defm : SBWriteResPair<WriteFCmp64X, [SBPort1],  3, [1], 1, 6>;
 defm : SBWriteResPair<WriteFCmp64Y, [SBPort1],  3, [1], 1, 7>;
+defm : SBWriteResPair<WriteFCmp64Z, [SBPort1],  3, [1], 1, 7>; // Unsupported = 1
 
 defm : SBWriteResPair<WriteFCom,    [SBPort1],  3>;
 
 defm : SBWriteResPair<WriteFMul,    [SBPort0],  5, [1], 1, 6>;
 defm : SBWriteResPair<WriteFMulX,   [SBPort0],  5, [1], 1, 6>;
 defm : SBWriteResPair<WriteFMulY,   [SBPort0],  5, [1], 1, 7>;
+defm : SBWriteResPair<WriteFMulZ,   [SBPort0],  5, [1], 1, 7>; // Unsupported = 1
 defm : SBWriteResPair<WriteFMul64,  [SBPort0],  5, [1], 1, 6>;
 defm : SBWriteResPair<WriteFMul64X, [SBPort0],  5, [1], 1, 6>;
 defm : SBWriteResPair<WriteFMul64Y, [SBPort0],  5, [1], 1, 7>;
+defm : SBWriteResPair<WriteFMul64Z, [SBPort0],  5, [1], 1, 7>; // Unsupported = 1
 
 defm : SBWriteResPair<WriteFDiv,    [SBPort0,SBFPDivider], 14, [1,14], 1, 6>;
 defm : SBWriteResPair<WriteFDivX,   [SBPort0,SBFPDivider], 14, [1,14], 1, 6>;
@@ -208,10 +214,12 @@
 defm : SBWriteResPair<WriteFRcp,   [SBPort0],  5, [1], 1, 6>;
 defm : SBWriteResPair<WriteFRcpX,  [SBPort0],  5, [1], 1, 6>;
 defm : SBWriteResPair<WriteFRcpY,  [SBPort0,SBPort05],  7, [2,1], 3, 7>;
+defm : SBWriteResPair<WriteFRcpZ,  [SBPort0,SBPort05],  7, [2,1], 3, 7>; // Unsupported = 1
 
 defm : SBWriteResPair<WriteFRsqrt, [SBPort0],  5, [1], 1, 6>;
 defm : SBWriteResPair<WriteFRsqrtX,[SBPort0],  5, [1], 1, 6>;
 defm : SBWriteResPair<WriteFRsqrtY,[SBPort0,SBPort05],  7, [2,1], 3, 7>;
+defm : SBWriteResPair<WriteFRsqrtZ,[SBPort0,SBPort05],  7, [2,1], 3, 7>; // Unsupported = 1
 
 defm : SBWriteResPair<WriteFSqrt,    [SBPort0,SBFPDivider], 14, [1,14], 1, 6>;
 defm : SBWriteResPair<WriteFSqrtX,   [SBPort0,SBFPDivider], 14, [1,14], 1, 6>;
@@ -226,58 +234,78 @@
 defm : SBWriteResPair<WriteDPPD,   [SBPort0,SBPort1,SBPort5],  9, [1,1,1], 3, 6>;
 defm : SBWriteResPair<WriteDPPS,   [SBPort0,SBPort1,SBPort5], 12, [1,2,1], 4, 6>;
 defm : SBWriteResPair<WriteDPPSY,  [SBPort0,SBPort1,SBPort5], 12, [1,2,1], 4, 7>;
+defm : SBWriteResPair<WriteDPPSZ,  [SBPort0,SBPort1,SBPort5], 12, [1,2,1], 4, 7>; // Unsupported = 1
 defm : SBWriteResPair<WriteFSign,    [SBPort5], 1>;
 defm : SBWriteResPair<WriteFRnd,     [SBPort1], 3, [1], 1, 6>;
 defm : SBWriteResPair<WriteFRndY,    [SBPort1], 3, [1], 1, 7>;
+defm : SBWriteResPair<WriteFRndZ,    [SBPort1], 3, [1], 1, 7>; // Unsupported = 1
 defm : SBWriteResPair<WriteFLogic,   [SBPort5], 1, [1], 1, 6>;
 defm : SBWriteResPair<WriteFLogicY,  [SBPort5], 1, [1], 1, 7>;
+defm : SBWriteResPair<WriteFLogicZ,  [SBPort5], 1, [1], 1, 7>; // Unsupported = 1
 defm : SBWriteResPair<WriteFTest,    [SBPort0], 1, [1], 1, 6>;
 defm : SBWriteResPair<WriteFTestY,   [SBPort0], 1, [1], 1, 7>;
+defm : SBWriteResPair<WriteFTestZ,   [SBPort0], 1, [1], 1, 7>; // Unsupported = 1
 defm : SBWriteResPair<WriteFShuffle, [SBPort5], 1, [1], 1, 6>;
 defm : SBWriteResPair<WriteFShuffleY,[SBPort5], 1, [1], 1, 7>;
+defm : SBWriteResPair<WriteFShuffleZ,[SBPort5], 1, [1], 1, 7>; // Unsupported = 1
 defm : SBWriteResPair<WriteFVarShuffle, [SBPort5], 1, [1], 1, 6>;
 defm : SBWriteResPair<WriteFVarShuffleY,[SBPort5], 1, [1], 1, 7>;
+defm : SBWriteResPair<WriteFVarShuffleZ,[SBPort5], 1, [1], 1, 7>; // Unsupported = 1
 defm : SBWriteResPair<WriteFBlend,    [SBPort05], 1, [1], 1, 6>;
 defm : SBWriteResPair<WriteFBlendY,   [SBPort05], 1, [1], 1, 7>;
+defm : SBWriteResPair<WriteFBlendZ,   [SBPort05], 1, [1], 1, 7>; // Unsupported = 1
 defm : SBWriteResPair<WriteFVarBlend, [SBPort05], 2, [2], 2, 6>;
 defm : SBWriteResPair<WriteFVarBlendY,[SBPort05], 2, [2], 2, 7>;
+defm : SBWriteResPair<WriteFVarBlendZ,[SBPort05], 2, [2], 2, 7>; // Unsupported = 1
 
 // Conversion between integer and float.
 defm : SBWriteResPair<WriteCvtSS2I,   [SBPort0,SBPort1], 5, [1,1], 2>;
 defm : SBWriteResPair<WriteCvtPS2I,           [SBPort1], 3, [1], 1, 6>;
 defm : SBWriteResPair<WriteCvtPS2IY,          [SBPort1], 3, [1], 1, 7>;
+defm : SBWriteResPair<WriteCvtPS2IZ,          [SBPort1], 3, [1], 1, 7>; // Unsupported = 1
 defm : SBWriteResPair<WriteCvtSD2I,   [SBPort0,SBPort1], 5, [1,1], 2>;
 defm : SBWriteResPair<WriteCvtPD2I,   [SBPort1,SBPort5], 4, [1,1], 2, 6>;
 defm : X86WriteRes<WriteCvtPD2IY,     [SBPort1,SBPort5], 4, [1,1], 2>;
+defm : X86WriteRes<WriteCvtPD2IZ,     [SBPort1,SBPort5], 4, [1,1], 2>; // Unsupported = 1
 defm : X86WriteRes<WriteCvtPD2IYLd,   [SBPort1,SBPort5,SBPort23], 11, [1,1,1], 3>;
+defm : X86WriteRes<WriteCvtPD2IZLd,   [SBPort1,SBPort5,SBPort23], 11, [1,1,1], 3>; // Unsupported = 1
 
 defm : X86WriteRes<WriteCvtI2SS,      [SBPort1,SBPort5],  5, [1,2], 3>;
 defm : X86WriteRes<WriteCvtI2SSLd,    [SBPort1,SBPort5,SBPort23], 10, [1,1,1], 3>;
 defm : SBWriteResPair<WriteCvtI2PS,           [SBPort1],  3, [1], 1, 6>;
 defm : SBWriteResPair<WriteCvtI2PSY,          [SBPort1],  3, [1], 1, 7>;
+defm : SBWriteResPair<WriteCvtI2PSZ,          [SBPort1],  3, [1], 1, 7>; // Unsupported = 1
 defm : X86WriteRes<WriteCvtI2SD,      [SBPort1,SBPort5],  4, [1,1], 2>;
 defm : X86WriteRes<WriteCvtI2PD,      [SBPort1,SBPort5],  4, [1,1], 2>;
 defm : X86WriteRes<WriteCvtI2PDY,     [SBPort1,SBPort5],  4, [1,1], 2>;
+defm : X86WriteRes<WriteCvtI2PDZ,     [SBPort1,SBPort5],  4, [1,1], 2>; // Unsupported = 1
 defm : X86WriteRes<WriteCvtI2SDLd,   [SBPort1,SBPort23],  9, [1,1], 2>;
 defm : X86WriteRes<WriteCvtI2PDLd,   [SBPort1,SBPort5,SBPort23], 10, [1,1,1], 3>;
 defm : X86WriteRes<WriteCvtI2PDYLd,  [SBPort1,SBPort5,SBPort23], 10, [1,1,1], 3>;
+defm : X86WriteRes<WriteCvtI2PDZLd,  [SBPort1,SBPort5,SBPort23], 10, [1,1,1], 3>; // Unsupported = 1
 
 defm : SBWriteResPair<WriteCvtSS2SD,  [SBPort0], 1, [1], 1, 6>;
 defm : X86WriteRes<WriteCvtPS2PD,     [SBPort0,SBPort5], 2, [1,1], 2>;
 defm : X86WriteRes<WriteCvtPS2PDY,    [SBPort0,SBPort5], 2, [1,1], 2>;
+defm : X86WriteRes<WriteCvtPS2PDZ,    [SBPort0,SBPort5], 2, [1,1], 2>; // Unsupported = 1
 defm : X86WriteRes<WriteCvtPS2PDLd,  [SBPort0,SBPort23], 7, [1,1], 2>;
 defm : X86WriteRes<WriteCvtPS2PDYLd, [SBPort0,SBPort23], 7, [1,1], 2>;
+defm : X86WriteRes<WriteCvtPS2PDZLd, [SBPort0,SBPort23], 7, [1,1], 2>; // Unsupported = 1
 defm : SBWriteResPair<WriteCvtSD2SS,  [SBPort1,SBPort5], 4, [1,1], 2, 6>;
 defm : SBWriteResPair<WriteCvtPD2PS,  [SBPort1,SBPort5], 4, [1,1], 2, 6>;
 defm : SBWriteResPair<WriteCvtPD2PSY, [SBPort1,SBPort5], 4, [1,1], 2, 7>;
+defm : SBWriteResPair<WriteCvtPD2PSZ, [SBPort1,SBPort5], 4, [1,1], 2, 7>; // Unsupported = 1
 
 defm : SBWriteResPair<WriteCvtPH2PS,  [SBPort1], 3>;
 defm : SBWriteResPair<WriteCvtPH2PSY, [SBPort1], 3>;
+defm : SBWriteResPair<WriteCvtPH2PSZ, [SBPort1], 3>; // Unsupported = 1
 
 defm : X86WriteRes<WriteCvtPS2PH,    [SBPort1], 3, [1], 1>;
 defm : X86WriteRes<WriteCvtPS2PHY,   [SBPort1], 3, [1], 1>;
+defm : X86WriteRes<WriteCvtPS2PHZ,   [SBPort1], 3, [1], 1>; // Unsupported = 1
 defm : X86WriteRes<WriteCvtPS2PHSt,  [SBPort1, SBPort23, SBPort4], 4, [1,1,1], 1>;
 defm : X86WriteRes<WriteCvtPS2PHYSt, [SBPort1, SBPort23, SBPort4], 4, [1,1,1], 1>;
+defm : X86WriteRes<WriteCvtPS2PHZSt, [SBPort1, SBPort23, SBPort4], 4, [1,1,1], 1>; // Unsupported = 1
 
 // Vector integer operations.
 defm : X86WriteRes<WriteVecLoad,         [SBPort23], 5, [1], 1>;
@@ -303,42 +331,56 @@
 defm : SBWriteResPair<WriteVecLogic, [SBPort015], 1, [1], 1, 5>;
 defm : SBWriteResPair<WriteVecLogicX,[SBPort015], 1, [1], 1, 6>;
 defm : SBWriteResPair<WriteVecLogicY,[SBPort015], 1, [1], 1, 7>;
+defm : SBWriteResPair<WriteVecLogicZ,[SBPort015], 1, [1], 1, 7>; // Unsupported = 1
 defm : SBWriteResPair<WriteVecTest,  [SBPort0,SBPort5], 2, [1,1], 2, 6>;
 defm : SBWriteResPair<WriteVecTestY, [SBPort0,SBPort5], 2, [1,1], 2, 7>;
+defm : SBWriteResPair<WriteVecTestZ, [SBPort0,SBPort5], 2, [1,1], 2, 7>; // Unsupported = 1
 defm : SBWriteResPair<WriteVecALU,   [SBPort1],  3, [1], 1, 5>;
 defm : SBWriteResPair<WriteVecALUX,  [SBPort15], 1, [1], 1, 6>;
 defm : SBWriteResPair<WriteVecALUY,  [SBPort15], 1, [1], 1, 7>;
+defm : SBWriteResPair<WriteVecALUZ,  [SBPort15], 1, [1], 1, 7>; // Unsupported = 1
 defm : SBWriteResPair<WriteVecIMul,  [SBPort0], 5, [1], 1, 5>;
 defm : SBWriteResPair<WriteVecIMulX, [SBPort0], 5, [1], 1, 6>;
 defm : SBWriteResPair<WriteVecIMulY, [SBPort0], 5, [1], 1, 7>;
+defm : SBWriteResPair<WriteVecIMulZ, [SBPort0], 5, [1], 1, 7>; // Unsupported = 1
 defm : SBWriteResPair<WritePMULLD,   [SBPort0], 5, [1], 1, 6>;
 defm : SBWriteResPair<WritePMULLDY,  [SBPort0], 5, [1], 1, 7>; // TODO this is probably wrong for 256/512-bit for the "generic" model
+defm : SBWriteResPair<WritePMULLDZ,  [SBPort0], 5, [1], 1, 7>;  // Unsupported = 1
 defm : SBWriteResPair<WriteShuffle,  [SBPort5], 1, [1], 1, 5>;
 defm : SBWriteResPair<WriteShuffleX, [SBPort15], 1, [1], 1, 6>;
 defm : SBWriteResPair<WriteShuffleY, [SBPort5], 1, [1], 1, 7>;
+defm : SBWriteResPair<WriteShuffleZ, [SBPort5], 1, [1], 1, 7>; // Unsupported = 1
 defm : SBWriteResPair<WriteVarShuffle,  [SBPort15], 1, [1], 1, 5>;
 defm : SBWriteResPair<WriteVarShuffleX, [SBPort15], 1, [1], 1, 6>;
 defm : SBWriteResPair<WriteVarShuffleY, [SBPort15], 1, [1], 1, 7>;
+defm : SBWriteResPair<WriteVarShuffleZ, [SBPort15], 1, [1], 1, 7>; // Unsupported = 1
 defm : SBWriteResPair<WriteBlend,   [SBPort15], 1, [1], 1, 6>;
 defm : SBWriteResPair<WriteBlendY,  [SBPort15], 1, [1], 1, 7>;
+defm : SBWriteResPair<WriteBlendZ,  [SBPort15], 1, [1], 1, 7>; // Unsupported = 1
 defm : SBWriteResPair<WriteVarBlend, [SBPort15], 2, [2], 2, 6>;
 defm : SBWriteResPair<WriteVarBlendY,[SBPort15], 2, [2], 2, 7>;
+defm : SBWriteResPair<WriteVarBlendZ,[SBPort15], 2, [2], 2, 7>; // Unsupported = 1
 defm : SBWriteResPair<WriteMPSAD,  [SBPort0, SBPort15], 7, [1,2], 3, 6>;
 defm : SBWriteResPair<WriteMPSADY, [SBPort0, SBPort15], 7, [1,2], 3, 7>;
+defm : SBWriteResPair<WriteMPSADZ, [SBPort0, SBPort15], 7, [1,2], 3, 7>; // Unsupported = 1
 defm : SBWriteResPair<WritePSADBW,  [SBPort0], 5, [1], 1, 5>;
 defm : SBWriteResPair<WritePSADBWX, [SBPort0], 5, [1], 1, 6>;
 defm : SBWriteResPair<WritePSADBWY, [SBPort0], 5, [1], 1, 7>;
+defm : SBWriteResPair<WritePSADBWZ, [SBPort0], 5, [1], 1, 7>; // Unsupported = 1
 defm : SBWriteResPair<WritePHMINPOS,  [SBPort0], 5, [1], 1, 6>;
 
 // Vector integer shifts.
 defm : SBWriteResPair<WriteVecShift,     [SBPort5], 1, [1], 1, 5>;
 defm : SBWriteResPair<WriteVecShiftX,    [SBPort0,SBPort15], 2, [1,1], 2, 6>;
 defm : SBWriteResPair<WriteVecShiftY,    [SBPort0,SBPort15], 4, [1,1], 2, 7>;
+defm : SBWriteResPair<WriteVecShiftZ,    [SBPort0,SBPort15], 4, [1,1], 2, 7>; // Unsupported = 1
 defm : SBWriteResPair<WriteVecShiftImm,  [SBPort5], 1, [1], 1, 5>;
 defm : SBWriteResPair<WriteVecShiftImmX, [SBPort0], 1, [1], 1, 6>;
 defm : SBWriteResPair<WriteVecShiftImmY, [SBPort0], 1, [1], 1, 7>;
+defm : SBWriteResPair<WriteVecShiftImmZ, [SBPort0], 1, [1], 1, 7>; // Unsupported = 1
 defm : SBWriteResPair<WriteVarVecShift,  [SBPort0], 1, [1], 1, 6>;
 defm : SBWriteResPair<WriteVarVecShiftY, [SBPort0], 1, [1], 1, 7>;
+defm : SBWriteResPair<WriteVarVecShiftZ, [SBPort0], 1, [1], 1, 7>; // Unsupported = 1
 
 // Vector insert/extract operations.
 def : WriteRes<WriteVecInsert, [SBPort5,SBPort15]> {
@@ -365,9 +407,11 @@
 
 defm : SBWriteResPair<WriteFHAdd,  [SBPort1,SBPort5], 5, [1,2], 3, 6>;
 defm : SBWriteResPair<WriteFHAddY, [SBPort1,SBPort5], 5, [1,2], 3, 7>;
+defm : SBWriteResPair<WriteFHAddZ, [SBPort1,SBPort5], 5, [1,2], 3, 7>; // Unsupported = 1
 defm : SBWriteResPair<WritePHAdd,  [SBPort15], 3, [3], 3, 5>;
 defm : SBWriteResPair<WritePHAddX, [SBPort15], 3, [3], 3, 6>;
 defm : SBWriteResPair<WritePHAddY, [SBPort15], 3, [3], 3, 7>;
+defm : SBWriteResPair<WritePHAddZ, [SBPort15], 3, [3], 3, 7>; // Unsupported = 1
 
 ////////////////////////////////////////////////////////////////////////////////
 // String instructions.
@@ -484,6 +528,7 @@
 defm : SBWriteResPair<WriteFMA,  [SBPort01],  5>;
 defm : SBWriteResPair<WriteFMAX, [SBPort01],  5>;
 defm : SBWriteResPair<WriteFMAY, [SBPort01],  5>;
+defm : SBWriteResPair<WriteFMAZ, [SBPort01],  5>;  // Unsupported = 1
 
 // Remaining SNB instrs.
 
Index: lib/Target/X86/X86SchedSkylakeClient.td
===================================================================
--- lib/Target/X86/X86SchedSkylakeClient.td
+++ lib/Target/X86/X86SchedSkylakeClient.td
@@ -183,76 +183,93 @@
 defm : X86WriteRes<WriteEMMS,          [SKLPort05,SKLPort0156], 10, [9,1], 10>;
 
 defm : SKLWriteResPair<WriteFAdd,     [SKLPort01],  4, [1], 1, 5>; // Floating point add/sub.
-defm : SKLWriteResPair<WriteFAddX,    [SKLPort01],  4, [1], 1, 6>; // Floating point add/sub (XMM).
-defm : SKLWriteResPair<WriteFAddY,    [SKLPort01],  4, [1], 1, 7>; // Floating point add/sub (YMM/ZMM).
+defm : SKLWriteResPair<WriteFAddX,    [SKLPort01],  4, [1], 1, 6>;
+defm : SKLWriteResPair<WriteFAddY,    [SKLPort01],  4, [1], 1, 7>;
+defm : X86WriteResPairUnsupported<WriteFAddZ>;
 defm : SKLWriteResPair<WriteFAdd64,   [SKLPort01],  4, [1], 1, 5>; // Floating point double add/sub.
-defm : SKLWriteResPair<WriteFAdd64X,  [SKLPort01],  4, [1], 1, 6>; // Floating point double add/sub (XMM).
-defm : SKLWriteResPair<WriteFAdd64Y,  [SKLPort01],  4, [1], 1, 7>; // Floating point double add/sub (YMM/ZMM).
+defm : SKLWriteResPair<WriteFAdd64X,  [SKLPort01],  4, [1], 1, 6>;
+defm : SKLWriteResPair<WriteFAdd64Y,  [SKLPort01],  4, [1], 1, 7>;
+defm : X86WriteResPairUnsupported<WriteFAdd64Z>;
 
 defm : SKLWriteResPair<WriteFCmp,     [SKLPort01],  4, [1], 1, 5>; // Floating point compare.
-defm : SKLWriteResPair<WriteFCmpX,    [SKLPort01],  4, [1], 1, 6>; // Floating point compare (XMM).
-defm : SKLWriteResPair<WriteFCmpY,    [SKLPort01],  4, [1], 1, 7>; // Floating point compare (YMM/ZMM).
+defm : SKLWriteResPair<WriteFCmpX,    [SKLPort01],  4, [1], 1, 6>;
+defm : SKLWriteResPair<WriteFCmpY,    [SKLPort01],  4, [1], 1, 7>;
+defm : X86WriteResPairUnsupported<WriteFCmpZ>;
 defm : SKLWriteResPair<WriteFCmp64,   [SKLPort01],  4, [1], 1, 5>; // Floating point double compare.
-defm : SKLWriteResPair<WriteFCmp64X,  [SKLPort01],  4, [1], 1, 6>; // Floating point double compare (XMM).
-defm : SKLWriteResPair<WriteFCmp64Y,  [SKLPort01],  4, [1], 1, 7>; // Floating point double compare (YMM/ZMM).
+defm : SKLWriteResPair<WriteFCmp64X,  [SKLPort01],  4, [1], 1, 6>;
+defm : SKLWriteResPair<WriteFCmp64Y,  [SKLPort01],  4, [1], 1, 7>;
+defm : X86WriteResPairUnsupported<WriteFCmp64Z>;
 
 defm : SKLWriteResPair<WriteFCom,      [SKLPort0],  2>; // Floating point compare to flags.
 
 defm : SKLWriteResPair<WriteFMul,     [SKLPort01],  4, [1], 1, 5>; // Floating point multiplication.
-defm : SKLWriteResPair<WriteFMulX,    [SKLPort01],  4, [1], 1, 6>; // Floating point multiplication (XMM).
-defm : SKLWriteResPair<WriteFMulY,    [SKLPort01],  4, [1], 1, 7>; // Floating point multiplication (YMM/ZMM).
+defm : SKLWriteResPair<WriteFMulX,    [SKLPort01],  4, [1], 1, 6>;
+defm : SKLWriteResPair<WriteFMulY,    [SKLPort01],  4, [1], 1, 7>;
+defm : X86WriteResPairUnsupported<WriteFMulZ>;
 defm : SKLWriteResPair<WriteFMul64,   [SKLPort01],  4, [1], 1, 5>; // Floating point double multiplication.
-defm : SKLWriteResPair<WriteFMul64X,  [SKLPort01],  4, [1], 1, 6>; // Floating point double multiplication (XMM).
-defm : SKLWriteResPair<WriteFMul64Y,  [SKLPort01],  4, [1], 1, 7>; // Floating point double multiplication (YMM/ZMM).
+defm : SKLWriteResPair<WriteFMul64X,  [SKLPort01],  4, [1], 1, 6>;
+defm : SKLWriteResPair<WriteFMul64Y,  [SKLPort01],  4, [1], 1, 7>;
+defm : X86WriteResPairUnsupported<WriteFMul64Z>;
 
 defm : SKLWriteResPair<WriteFDiv,     [SKLPort0,SKLFPDivider], 11, [1,3], 1, 5>; // Floating point division.
-//defm : SKLWriteResPair<WriteFDivX,    [SKLPort0,SKLFPDivider], 11, [1,3], 1, 6>; // Floating point division (XMM).
-defm : SKLWriteResPair<WriteFDivY,    [SKLPort0,SKLFPDivider], 11, [1,5], 1, 7>; // Floating point division (YMM).
+//defm : SKLWriteResPair<WriteFDivX,    [SKLPort0,SKLFPDivider], 11, [1,3], 1, 6>;
+defm : SKLWriteResPair<WriteFDivY,    [SKLPort0,SKLFPDivider], 11, [1,5], 1, 7>;
 defm : X86WriteResPairUnsupported<WriteFDivZ>;
 //defm : SKLWriteResPair<WriteFDiv64,   [SKLPort0,SKLFPDivider], 14, [1,3], 1, 5>; // Floating point double division.
-//defm : SKLWriteResPair<WriteFDiv64X,  [SKLPort0,SKLFPDivider], 14, [1,3], 1, 6>; // Floating point double division (XMM).
-//defm : SKLWriteResPair<WriteFDiv64Y,  [SKLPort0,SKLFPDivider], 14, [1,5], 1, 7>; // Floating point double division (YMM).
+//defm : SKLWriteResPair<WriteFDiv64X,  [SKLPort0,SKLFPDivider], 14, [1,3], 1, 6>;
+//defm : SKLWriteResPair<WriteFDiv64Y,  [SKLPort0,SKLFPDivider], 14, [1,5], 1, 7>;
 defm : X86WriteResPairUnsupported<WriteFDiv64Z>;
 
 defm : SKLWriteResPair<WriteFSqrt,    [SKLPort0,SKLFPDivider], 12, [1,3], 1, 5>; // Floating point square root.
-defm : SKLWriteResPair<WriteFSqrtX,   [SKLPort0,SKLFPDivider], 12, [1,3], 1, 6>; // Floating point square root (XMM).
-defm : SKLWriteResPair<WriteFSqrtY,   [SKLPort0,SKLFPDivider], 12, [1,6], 1, 7>; // Floating point square root (YMM).
+defm : SKLWriteResPair<WriteFSqrtX,   [SKLPort0,SKLFPDivider], 12, [1,3], 1, 6>;
+defm : SKLWriteResPair<WriteFSqrtY,   [SKLPort0,SKLFPDivider], 12, [1,6], 1, 7>;
 defm : X86WriteResPairUnsupported<WriteFSqrtZ>;
 defm : SKLWriteResPair<WriteFSqrt64,  [SKLPort0,SKLFPDivider], 18, [1,6], 1, 5>; // Floating point double square root.
-defm : SKLWriteResPair<WriteFSqrt64X, [SKLPort0,SKLFPDivider], 18, [1,6], 1, 6>; // Floating point double square root (XMM).
-defm : SKLWriteResPair<WriteFSqrt64Y, [SKLPort0,SKLFPDivider], 18, [1,12],1, 7>; // Floating point double square root (YMM).
+defm : SKLWriteResPair<WriteFSqrt64X, [SKLPort0,SKLFPDivider], 18, [1,6], 1, 6>;
+defm : SKLWriteResPair<WriteFSqrt64Y, [SKLPort0,SKLFPDivider], 18, [1,12],1, 7>;
 defm : X86WriteResPairUnsupported<WriteFSqrt64Z>;
 defm : SKLWriteResPair<WriteFSqrt80,  [SKLPort0,SKLFPDivider], 21, [1,7]>; // Floating point long double square root.
 
 defm : SKLWriteResPair<WriteFRcp,   [SKLPort0], 4, [1], 1, 5>; // Floating point reciprocal estimate.
-defm : SKLWriteResPair<WriteFRcpX,  [SKLPort0], 4, [1], 1, 6>; // Floating point reciprocal estimate (XMM).
-defm : SKLWriteResPair<WriteFRcpY,  [SKLPort0], 4, [1], 1, 7>; // Floating point reciprocal estimate (YMM/ZMM).
+defm : SKLWriteResPair<WriteFRcpX,  [SKLPort0], 4, [1], 1, 6>;
+defm : SKLWriteResPair<WriteFRcpY,  [SKLPort0], 4, [1], 1, 7>;
+defm : X86WriteResPairUnsupported<WriteFRcpZ>;
 
 defm : SKLWriteResPair<WriteFRsqrt, [SKLPort0], 4, [1], 1, 5>; // Floating point reciprocal square root estimate.
-defm : SKLWriteResPair<WriteFRsqrtX,[SKLPort0], 4, [1], 1, 6>; // Floating point reciprocal square root estimate (XMM).
-defm : SKLWriteResPair<WriteFRsqrtY,[SKLPort0], 4, [1], 1, 7>; // Floating point reciprocal square root estimate (YMM/ZMM).
+defm : SKLWriteResPair<WriteFRsqrtX,[SKLPort0], 4, [1], 1, 6>;
+defm : SKLWriteResPair<WriteFRsqrtY,[SKLPort0], 4, [1], 1, 7>;
+defm : X86WriteResPairUnsupported<WriteFRsqrtZ>;
 
 defm : SKLWriteResPair<WriteFMA,    [SKLPort01], 4, [1], 1, 5>; // Fused Multiply Add.
-defm : SKLWriteResPair<WriteFMAX,   [SKLPort01], 4, [1], 1, 6>; // Fused Multiply Add (XMM).
-defm : SKLWriteResPair<WriteFMAY,   [SKLPort01], 4, [1], 1, 7>; // Fused Multiply Add (YMM/ZMM).
+defm : SKLWriteResPair<WriteFMAX,   [SKLPort01], 4, [1], 1, 6>;
+defm : SKLWriteResPair<WriteFMAY,   [SKLPort01], 4, [1], 1, 7>;
+defm : X86WriteResPairUnsupported<WriteFMAZ>;
 defm : SKLWriteResPair<WriteDPPD,   [SKLPort5,SKLPort01],  9, [1,2], 3, 6>; // Floating point double dot product.
-defm : SKLWriteResPair<WriteDPPS,   [SKLPort5,SKLPort01], 13, [1,3], 4, 6>; // Floating point single dot product.
-defm : SKLWriteResPair<WriteDPPSY,  [SKLPort5,SKLPort01], 13, [1,3], 4, 7>; // Floating point single dot product (YMM).
+defm : SKLWriteResPair<WriteDPPS,   [SKLPort5,SKLPort01], 13, [1,3], 4, 6>;
+defm : SKLWriteResPair<WriteDPPSY,  [SKLPort5,SKLPort01], 13, [1,3], 4, 7>;
+defm : X86WriteResPairUnsupported<WriteDPPSZ>;
 defm : SKLWriteResPair<WriteFSign,   [SKLPort0], 1>; // Floating point fabs/fchs.
 defm : SKLWriteResPair<WriteFRnd,     [SKLPort01], 8, [2], 2, 6>; // Floating point rounding.
-defm : SKLWriteResPair<WriteFRndY,    [SKLPort01], 8, [2], 2, 7>; // Floating point rounding (YMM/ZMM).
+defm : SKLWriteResPair<WriteFRndY,    [SKLPort01], 8, [2], 2, 7>;
+defm : X86WriteResPairUnsupported<WriteFRndZ>;
 defm : SKLWriteResPair<WriteFLogic,  [SKLPort015], 1, [1], 1, 6>; // Floating point and/or/xor logicals.
-defm : SKLWriteResPair<WriteFLogicY, [SKLPort015], 1, [1], 1, 7>; // Floating point and/or/xor logicals (YMM/ZMM).
+defm : SKLWriteResPair<WriteFLogicY, [SKLPort015], 1, [1], 1, 7>;
+defm : X86WriteResPairUnsupported<WriteFLogicZ>;
 defm : SKLWriteResPair<WriteFTest,   [SKLPort0], 2, [1], 1, 6>; // Floating point TEST instructions.
-defm : SKLWriteResPair<WriteFTestY,  [SKLPort0], 2, [1], 1, 7>; // Floating point TEST instructions (YMM/ZMM).
+defm : SKLWriteResPair<WriteFTestY,  [SKLPort0], 2, [1], 1, 7>;
+defm : X86WriteResPairUnsupported<WriteFTestZ>;
 defm : SKLWriteResPair<WriteFShuffle,  [SKLPort5], 1, [1], 1, 6>; // Floating point vector shuffles.
-defm : SKLWriteResPair<WriteFShuffleY, [SKLPort5], 1, [1], 1, 7>; // Floating point vector shuffles (YMM/ZMM).
+defm : SKLWriteResPair<WriteFShuffleY, [SKLPort5], 1, [1], 1, 7>;
+defm : X86WriteResPairUnsupported<WriteFShuffleZ>;
 defm : SKLWriteResPair<WriteFVarShuffle,  [SKLPort5], 1, [1], 1, 6>; // Floating point vector shuffles.
-defm : SKLWriteResPair<WriteFVarShuffleY, [SKLPort5], 1, [1], 1, 7>; // Floating point vector shuffles.
+defm : SKLWriteResPair<WriteFVarShuffleY, [SKLPort5], 1, [1], 1, 7>;
+defm : X86WriteResPairUnsupported<WriteFVarShuffleZ>;
 defm : SKLWriteResPair<WriteFBlend,  [SKLPort015], 1, [1], 1, 6>; // Floating point vector blends.
-defm : SKLWriteResPair<WriteFBlendY, [SKLPort015], 1, [1], 1, 7>; // Floating point vector blends.
+defm : SKLWriteResPair<WriteFBlendY, [SKLPort015], 1, [1], 1, 7>;
+defm : X86WriteResPairUnsupported<WriteFBlendZ>;
 defm : SKLWriteResPair<WriteFVarBlend, [SKLPort015], 2, [2], 2, 6>; // Fp vector variable blends.
-defm : SKLWriteResPair<WriteFVarBlendY,[SKLPort015], 2, [2], 2, 7>; // Fp vector variable blends.
+defm : SKLWriteResPair<WriteFVarBlendY,[SKLPort015], 2, [2], 2, 7>;
+defm : X86WriteResPairUnsupported<WriteFVarBlendZ>;
 
 // FMA Scheduling helper class.
 // class FMASC { X86FoldableSchedWrite Sched = WriteFAdd; }
@@ -279,33 +296,44 @@
 defm : X86WriteRes<WriteVecMoveFromGpr,  [SKLPort5], 1, [1], 1>;
 
 defm : SKLWriteResPair<WriteVecALU,   [SKLPort05], 1, [1], 1, 5>; // Vector integer ALU op, no logicals.
-defm : SKLWriteResPair<WriteVecALUX,  [SKLPort01], 1, [1], 1, 6>; // Vector integer ALU op, no logicals (XMM).
-defm : SKLWriteResPair<WriteVecALUY,  [SKLPort01], 1, [1], 1, 7>; // Vector integer ALU op, no logicals (YMM/ZMM).
+defm : SKLWriteResPair<WriteVecALUX,  [SKLPort01], 1, [1], 1, 6>;
+defm : SKLWriteResPair<WriteVecALUY,  [SKLPort01], 1, [1], 1, 7>;
+defm : X86WriteResPairUnsupported<WriteVecALUZ>;
 defm : SKLWriteResPair<WriteVecLogic, [SKLPort05],  1, [1], 1, 5>; // Vector integer and/or/xor.
-defm : SKLWriteResPair<WriteVecLogicX,[SKLPort015], 1, [1], 1, 6>; // Vector integer and/or/xor (XMM).
-defm : SKLWriteResPair<WriteVecLogicY,[SKLPort015], 1, [1], 1, 7>; // Vector integer and/or/xor (YMM/ZMM).
+defm : SKLWriteResPair<WriteVecLogicX,[SKLPort015], 1, [1], 1, 6>;
+defm : SKLWriteResPair<WriteVecLogicY,[SKLPort015], 1, [1], 1, 7>;
+defm : X86WriteResPairUnsupported<WriteVecLogicZ>;
 defm : SKLWriteResPair<WriteVecTest,  [SKLPort0,SKLPort5], 3, [1,1], 2, 6>; // Vector integer TEST instructions.
-defm : SKLWriteResPair<WriteVecTestY, [SKLPort0,SKLPort5], 3, [1,1], 2, 7>; // Vector integer TEST instructions (YMM/ZMM).
+defm : SKLWriteResPair<WriteVecTestY, [SKLPort0,SKLPort5], 3, [1,1], 2, 7>;
+defm : X86WriteResPairUnsupported<WriteVecTestZ>;
 defm : SKLWriteResPair<WriteVecIMul,  [SKLPort0] ,  4, [1], 1, 5>; // Vector integer multiply.
-defm : SKLWriteResPair<WriteVecIMulX, [SKLPort01],  4, [1], 1, 6>; // Vector integer multiply (XMM).
-defm : SKLWriteResPair<WriteVecIMulY, [SKLPort01],  4, [1], 1, 7>; // Vector integer multiply (YMM/ZMM).
+defm : SKLWriteResPair<WriteVecIMulX, [SKLPort01],  4, [1], 1, 6>;
+defm : SKLWriteResPair<WriteVecIMulY, [SKLPort01],  4, [1], 1, 7>;
+defm : X86WriteResPairUnsupported<WriteVecIMulZ>;
 defm : SKLWriteResPair<WritePMULLD,   [SKLPort01], 10, [2], 2, 6>; // Vector PMULLD.
-defm : SKLWriteResPair<WritePMULLDY,  [SKLPort01], 10, [2], 2, 7>; // Vector PMULLD (YMM/ZMM).
+defm : SKLWriteResPair<WritePMULLDY,  [SKLPort01], 10, [2], 2, 7>;
+defm : X86WriteResPairUnsupported<WritePMULLDZ>;
 defm : SKLWriteResPair<WriteShuffle,  [SKLPort5], 1, [1], 1, 5>; // Vector shuffles.
-defm : SKLWriteResPair<WriteShuffleX, [SKLPort5], 1, [1], 1, 6>; // Vector shuffles (XMM).
-defm : SKLWriteResPair<WriteShuffleY, [SKLPort5], 1, [1], 1, 7>; // Vector shuffles (YMM/ZMM).
+defm : SKLWriteResPair<WriteShuffleX, [SKLPort5], 1, [1], 1, 6>;
+defm : SKLWriteResPair<WriteShuffleY, [SKLPort5], 1, [1], 1, 7>;
+defm : X86WriteResPairUnsupported<WriteShuffleZ>;
 defm : SKLWriteResPair<WriteVarShuffle,  [SKLPort5], 1, [1], 1, 5>; // Vector shuffles.
-defm : SKLWriteResPair<WriteVarShuffleX, [SKLPort5], 1, [1], 1, 6>; // Vector shuffles (XMM).
-defm : SKLWriteResPair<WriteVarShuffleY, [SKLPort5], 1, [1], 1, 7>; // Vector shuffles (YMM/ZMM).
+defm : SKLWriteResPair<WriteVarShuffleX, [SKLPort5], 1, [1], 1, 6>;
+defm : SKLWriteResPair<WriteVarShuffleY, [SKLPort5], 1, [1], 1, 7>;
+defm : X86WriteResPairUnsupported<WriteVarShuffleZ>;
 defm : SKLWriteResPair<WriteBlend,  [SKLPort5], 1, [1], 1, 6>; // Vector blends.
-defm : SKLWriteResPair<WriteBlendY, [SKLPort5], 1, [1], 1, 7>; // Vector blends (YMM/ZMM).
+defm : SKLWriteResPair<WriteBlendY, [SKLPort5], 1, [1], 1, 7>;
+defm : X86WriteResPairUnsupported<WriteBlendZ>;
 defm : SKLWriteResPair<WriteVarBlend,  [SKLPort015], 2, [2], 2, 6>; // Vector variable blends.
-defm : SKLWriteResPair<WriteVarBlendY, [SKLPort015], 2, [2], 2, 6>; // Vector variable blends (YMM/ZMM).
+defm : SKLWriteResPair<WriteVarBlendY, [SKLPort015], 2, [2], 2, 6>;
+defm : X86WriteResPairUnsupported<WriteVarBlendZ>;
 defm : SKLWriteResPair<WriteMPSAD,  [SKLPort5], 4, [2], 2, 6>; // Vector MPSAD.
-defm : SKLWriteResPair<WriteMPSADY, [SKLPort5], 4, [2], 2, 7>; // Vector MPSAD (YMM/ZMM).
+defm : SKLWriteResPair<WriteMPSADY, [SKLPort5], 4, [2], 2, 7>;
+defm : X86WriteResPairUnsupported<WriteMPSADZ>;
 defm : SKLWriteResPair<WritePSADBW,  [SKLPort5], 3, [1], 1, 5>; // Vector PSADBW.
-defm : SKLWriteResPair<WritePSADBWX, [SKLPort5], 3, [1], 1, 6>; // Vector PSADBW (XMM).
-defm : SKLWriteResPair<WritePSADBWY, [SKLPort5], 3, [1], 1, 7>; // Vector PSADBW (YMM/ZMM).
+defm : SKLWriteResPair<WritePSADBWX, [SKLPort5], 3, [1], 1, 6>;
+defm : SKLWriteResPair<WritePSADBWY, [SKLPort5], 3, [1], 1, 7>;
+defm : X86WriteResPairUnsupported<WritePSADBWZ>;
 defm : SKLWriteResPair<WritePHMINPOS, [SKLPort01], 4, [1], 1, 6>; // Vector PHMINPOS.
 
 // Vector integer shifts.
@@ -314,12 +342,15 @@
 defm : X86WriteRes<WriteVecShiftY,        [SKLPort5,SKLPort01],  4, [1,1], 2>;
 defm : X86WriteRes<WriteVecShiftXLd,      [SKLPort01,SKLPort23], 7, [1,1], 2>;
 defm : X86WriteRes<WriteVecShiftYLd,      [SKLPort01,SKLPort23], 8, [1,1], 2>;
+defm : X86WriteResPairUnsupported<WriteVecShiftZ>;
 
-defm : SKLWriteResPair<WriteVecShiftImm,  [SKLPort0],  1, [1], 1, 5>;
-defm : SKLWriteResPair<WriteVecShiftImmX, [SKLPort01], 1, [1], 1, 6>; // Vector integer immediate shifts (XMM).
-defm : SKLWriteResPair<WriteVecShiftImmY, [SKLPort01], 1, [1], 1, 7>; // Vector integer immediate shifts (YMM/ZMM).
+defm : SKLWriteResPair<WriteVecShiftImm,  [SKLPort0],  1, [1], 1, 5>; // Vector integer immediate shifts.
+defm : SKLWriteResPair<WriteVecShiftImmX, [SKLPort01], 1, [1], 1, 6>;
+defm : SKLWriteResPair<WriteVecShiftImmY, [SKLPort01], 1, [1], 1, 7>;
+defm : X86WriteResPairUnsupported<WriteVecShiftImmZ>;
 defm : SKLWriteResPair<WriteVarVecShift,  [SKLPort01], 1, [1], 1, 6>; // Variable vector shifts.
-defm : SKLWriteResPair<WriteVarVecShiftY, [SKLPort01], 1, [1], 1, 7>; // Variable vector shifts (YMM/ZMM).
+defm : SKLWriteResPair<WriteVarVecShiftY, [SKLPort01], 1, [1], 1, 7>;
+defm : X86WriteResPairUnsupported<WriteVarVecShiftZ>;
 
 // Vector insert/extract operations.
 def : WriteRes<WriteVecInsert, [SKLPort5]> {
@@ -346,33 +377,43 @@
 defm : SKLWriteResPair<WriteCvtSS2I,   [SKLPort1], 3>;
 defm : SKLWriteResPair<WriteCvtPS2I,   [SKLPort1], 3>;
 defm : SKLWriteResPair<WriteCvtPS2IY,  [SKLPort1], 3>;
+defm : X86WriteResPairUnsupported<WriteCvtPS2IZ>;
 defm : SKLWriteResPair<WriteCvtSD2I,   [SKLPort1], 3>;
 defm : SKLWriteResPair<WriteCvtPD2I,   [SKLPort1], 3>;
 defm : SKLWriteResPair<WriteCvtPD2IY,  [SKLPort1], 3>;
+defm : X86WriteResPairUnsupported<WriteCvtPD2IZ>;
 
 defm : SKLWriteResPair<WriteCvtI2SS,   [SKLPort1], 4>;
 defm : SKLWriteResPair<WriteCvtI2PS,   [SKLPort1], 4>;
 defm : SKLWriteResPair<WriteCvtI2PSY,  [SKLPort1], 4>;
+defm : X86WriteResPairUnsupported<WriteCvtI2PSZ>;
 defm : SKLWriteResPair<WriteCvtI2SD,   [SKLPort1], 4>;
 defm : SKLWriteResPair<WriteCvtI2PD,   [SKLPort1], 4>;
 defm : SKLWriteResPair<WriteCvtI2PDY,  [SKLPort1], 4>;
+defm : X86WriteResPairUnsupported<WriteCvtI2PDZ>;
 
 defm : SKLWriteResPair<WriteCvtSS2SD,  [SKLPort1], 3>;
 defm : SKLWriteResPair<WriteCvtPS2PD,  [SKLPort1], 3>;
 defm : SKLWriteResPair<WriteCvtPS2PDY, [SKLPort1], 3>;
+defm : X86WriteResPairUnsupported<WriteCvtPS2PDZ>;
 defm : SKLWriteResPair<WriteCvtSD2SS,  [SKLPort1], 3>;
 defm : SKLWriteResPair<WriteCvtPD2PS,  [SKLPort1], 3>;
 defm : SKLWriteResPair<WriteCvtPD2PSY, [SKLPort1], 3>;
+defm : X86WriteResPairUnsupported<WriteCvtPD2PSZ>;
 
 defm : X86WriteRes<WriteCvtPH2PS,    [SKLPort5,SKLPort015],  5, [1,1], 2>;
 defm : X86WriteRes<WriteCvtPH2PSY,    [SKLPort5,SKLPort01],  7, [1,1], 2>;
+defm : X86WriteResUnsupported<WriteCvtPH2PSZ>;
 defm : X86WriteRes<WriteCvtPH2PSLd,  [SKLPort23,SKLPort01],  9, [1,1], 2>;
 defm : X86WriteRes<WriteCvtPH2PSYLd, [SKLPort23,SKLPort01], 10, [1,1], 2>;
+defm : X86WriteResUnsupported<WriteCvtPH2PSZLd>;
 
 defm : X86WriteRes<WriteCvtPS2PH,                       [SKLPort5,SKLPort015], 5, [1,1], 2>;
 defm : X86WriteRes<WriteCvtPS2PHY,                       [SKLPort5,SKLPort01], 7, [1,1], 2>;
+defm : X86WriteResUnsupported<WriteCvtPS2PHZ>;
 defm : X86WriteRes<WriteCvtPS2PHSt,  [SKLPort4,SKLPort5,SKLPort237,SKLPort01], 6, [1,1,1,1], 4>;
 defm : X86WriteRes<WriteCvtPS2PHYSt, [SKLPort4,SKLPort5,SKLPort237,SKLPort01], 8, [1,1,1,1], 4>;
+defm : X86WriteResUnsupported<WriteCvtPS2PHZSt>;
 
 // Strings instructions.
 
Index: lib/Target/X86/X86SchedSkylakeServer.td
===================================================================
--- lib/Target/X86/X86SchedSkylakeServer.td
+++ lib/Target/X86/X86SchedSkylakeServer.td
@@ -182,77 +182,94 @@
 defm : X86WriteRes<WriteFMoveY,        [SKXPort015], 1, [1], 1>;
 defm : X86WriteRes<WriteEMMS,          [SKXPort05,SKXPort0156], 10, [9,1], 10>;
 
-defm : SKXWriteResPair<WriteFAdd,     [SKXPort015],  4, [1], 1, 5>; // Floating point add/sub.
-defm : SKXWriteResPair<WriteFAddX,    [SKXPort015],  4, [1], 1, 6>; // Floating point add/sub (XMM).
-defm : SKXWriteResPair<WriteFAddY,    [SKXPort015],  4, [1], 1, 7>; // Floating point add/sub (YMM/ZMM).
-defm : SKXWriteResPair<WriteFAdd64,   [SKXPort015],  4, [1], 1, 5>; // Floating point double add/sub.
-defm : SKXWriteResPair<WriteFAdd64X,  [SKXPort015],  4, [1], 1, 6>; // Floating point double add/sub (XMM).
-defm : SKXWriteResPair<WriteFAdd64Y,  [SKXPort015],  4, [1], 1, 7>; // Floating point double add/sub (YMM/ZMM).
-
-defm : SKXWriteResPair<WriteFCmp,     [SKXPort015],  4, [1], 1, 5>; // Floating point compare.
-defm : SKXWriteResPair<WriteFCmpX,    [SKXPort015],  4, [1], 1, 6>; // Floating point compare (XMM).
-defm : SKXWriteResPair<WriteFCmpY,    [SKXPort015],  4, [1], 1, 7>; // Floating point compare (YMM/ZMM).
-defm : SKXWriteResPair<WriteFCmp64,   [SKXPort015],  4, [1], 1, 5>; // Floating point double compare.
-defm : SKXWriteResPair<WriteFCmp64X,  [SKXPort015],  4, [1], 1, 6>; // Floating point double compare (XMM).
-defm : SKXWriteResPair<WriteFCmp64Y,  [SKXPort015],  4, [1], 1, 7>; // Floating point double compare (YMM/ZMM).
+defm : SKXWriteResPair<WriteFAdd,      [SKXPort01],  4, [1], 1, 5>; // Floating point add/sub.
+defm : SKXWriteResPair<WriteFAddX,     [SKXPort01],  4, [1], 1, 6>;
+defm : SKXWriteResPair<WriteFAddY,     [SKXPort01],  4, [1], 1, 7>;
+defm : SKXWriteResPair<WriteFAddZ,     [SKXPort05],  4, [1], 1, 7>;
+defm : SKXWriteResPair<WriteFAdd64,    [SKXPort01],  4, [1], 1, 5>; // Floating point double add/sub.
+defm : SKXWriteResPair<WriteFAdd64X,   [SKXPort01],  4, [1], 1, 6>;
+defm : SKXWriteResPair<WriteFAdd64Y,   [SKXPort01],  4, [1], 1, 7>;
+defm : SKXWriteResPair<WriteFAdd64Z,   [SKXPort05],  4, [1], 1, 7>;
+
+defm : SKXWriteResPair<WriteFCmp,      [SKXPort01],  4, [1], 1, 5>; // Floating point compare.
+defm : SKXWriteResPair<WriteFCmpX,     [SKXPort01],  4, [1], 1, 6>;
+defm : SKXWriteResPair<WriteFCmpY,     [SKXPort01],  4, [1], 1, 7>;
+defm : SKXWriteResPair<WriteFCmpZ,     [SKXPort05],  4, [1], 1, 7>;
+defm : SKXWriteResPair<WriteFCmp64,    [SKXPort01],  4, [1], 1, 5>; // Floating point double compare.
+defm : SKXWriteResPair<WriteFCmp64X,   [SKXPort01],  4, [1], 1, 6>;
+defm : SKXWriteResPair<WriteFCmp64Y,   [SKXPort01],  4, [1], 1, 7>;
+defm : SKXWriteResPair<WriteFCmp64Z,   [SKXPort05],  4, [1], 1, 7>;
 
 defm : SKXWriteResPair<WriteFCom,       [SKXPort0],  2>; // Floating point compare to flags.
 
-defm : SKXWriteResPair<WriteFMul,     [SKXPort015],  4, [1], 1, 5>; // Floating point multiplication.
-defm : SKXWriteResPair<WriteFMulX,    [SKXPort015],  4, [1], 1, 6>; // Floating point multiplication (XMM).
-defm : SKXWriteResPair<WriteFMulY,    [SKXPort015],  4, [1], 1, 7>; // Floating point multiplication (YMM/ZMM).
-defm : SKXWriteResPair<WriteFMul64,   [SKXPort015],  4, [1], 1, 5>; // Floating point double multiplication.
-defm : SKXWriteResPair<WriteFMul64X,  [SKXPort015],  4, [1], 1, 6>; // Floating point double multiplication (XMM).
-defm : SKXWriteResPair<WriteFMul64Y,  [SKXPort015],  4, [1], 1, 7>; // Floating point double multiplication (YMM/ZMM).
+defm : SKXWriteResPair<WriteFMul,      [SKXPort01],  4, [1], 1, 5>; // Floating point multiplication.
+defm : SKXWriteResPair<WriteFMulX,     [SKXPort01],  4, [1], 1, 6>;
+defm : SKXWriteResPair<WriteFMulY,     [SKXPort01],  4, [1], 1, 7>;
+defm : SKXWriteResPair<WriteFMulZ,     [SKXPort05],  4, [1], 1, 7>;
+defm : SKXWriteResPair<WriteFMul64,    [SKXPort01],  4, [1], 1, 5>; // Floating point double multiplication.
+defm : SKXWriteResPair<WriteFMul64X,   [SKXPort01],  4, [1], 1, 6>;
+defm : SKXWriteResPair<WriteFMul64Y,   [SKXPort01],  4, [1], 1, 7>;
+defm : SKXWriteResPair<WriteFMul64Z,   [SKXPort05],  4, [1], 1, 7>;
 
 defm : SKXWriteResPair<WriteFDiv,     [SKXPort0,SKXFPDivider], 11, [1,3], 1, 5>; // 10-14 cycles. // Floating point division.
-//defm : SKXWriteResPair<WriteFDivX,    [SKXPort0,SKXFPDivider], 11, [1,3], 1, 6>; // 10-14 cycles. // Floating point division (XMM).
-defm : SKXWriteResPair<WriteFDivY,    [SKXPort0,SKXFPDivider], 11, [1,5], 1, 7>; // 10-14 cycles. // Floating point division (YMM).
-defm : SKXWriteResPair<WriteFDivZ,    [SKXPort0,SKXPort5,SKXFPDivider], 18, [2,1,10], 3, 7>; // 10-14 cycles. // Floating point division (ZMM).
+//defm : SKXWriteResPair<WriteFDivX,    [SKXPort0,SKXFPDivider], 11, [1,3], 1, 6>; // 10-14 cycles.
+defm : SKXWriteResPair<WriteFDivY,    [SKXPort0,SKXFPDivider], 11, [1,5], 1, 7>; // 10-14 cycles.
+defm : SKXWriteResPair<WriteFDivZ,    [SKXPort0,SKXPort5,SKXFPDivider], 18, [2,1,10], 3, 7>; // 10-14 cycles.
 //defm : SKXWriteResPair<WriteFDiv64,   [SKXPort0,SKXFPDivider], 14, [1,3], 1, 5>; // 10-14 cycles. // Floating point division.
-//defm : SKXWriteResPair<WriteFDiv64X,  [SKXPort0,SKXFPDivider], 14, [1,3], 1, 6>; // 10-14 cycles. // Floating point division (XMM).
-//defm : SKXWriteResPair<WriteFDiv64Y,  [SKXPort0,SKXFPDivider], 14, [1,5], 1, 7>; // 10-14 cycles. // Floating point division (YMM).
-defm : SKXWriteResPair<WriteFDiv64Z,  [SKXPort0,SKXPort5,SKXFPDivider], 23, [2,1,16], 3, 7>; // 10-14 cycles. // Floating point division (ZMM).
+//defm : SKXWriteResPair<WriteFDiv64X,  [SKXPort0,SKXFPDivider], 14, [1,3], 1, 6>; // 10-14 cycles.
+//defm : SKXWriteResPair<WriteFDiv64Y,  [SKXPort0,SKXFPDivider], 14, [1,5], 1, 7>; // 10-14 cycles.
+defm : SKXWriteResPair<WriteFDiv64Z,  [SKXPort0,SKXPort5,SKXFPDivider], 23, [2,1,16], 3, 7>; // 10-14 cycles.
 
 defm : SKXWriteResPair<WriteFSqrt,    [SKXPort0,SKXFPDivider], 12, [1,3], 1, 5>; // Floating point square root.
-defm : SKXWriteResPair<WriteFSqrtX,   [SKXPort0,SKXFPDivider], 12, [1,3], 1, 6>; // Floating point square root (XMM).
-defm : SKXWriteResPair<WriteFSqrtY,   [SKXPort0,SKXFPDivider], 12, [1,6], 1, 7>; // Floating point square root (YMM).
-defm : SKXWriteResPair<WriteFSqrtZ,   [SKXPort0,SKXPort5,SKXFPDivider], 20, [2,1,12], 3, 7>; // Floating point square root (ZMM).
+defm : SKXWriteResPair<WriteFSqrtX,   [SKXPort0,SKXFPDivider], 12, [1,3], 1, 6>;
+defm : SKXWriteResPair<WriteFSqrtY,   [SKXPort0,SKXFPDivider], 12, [1,6], 1, 7>;
+defm : SKXWriteResPair<WriteFSqrtZ,   [SKXPort0,SKXPort5,SKXFPDivider], 20, [2,1,12], 3, 7>;
 defm : SKXWriteResPair<WriteFSqrt64,  [SKXPort0,SKXFPDivider], 18, [1,6], 1, 5>; // Floating point double square root.
-defm : SKXWriteResPair<WriteFSqrt64X, [SKXPort0,SKXFPDivider], 18, [1,6], 1, 6>; // Floating point double square root (XMM).
-defm : SKXWriteResPair<WriteFSqrt64Y, [SKXPort0,SKXFPDivider], 18, [1,12],1, 7>; // Floating point double square root (YMM).
-defm : SKXWriteResPair<WriteFSqrt64Z, [SKXPort0,SKXPort5,SKXFPDivider], 32, [2,1,24], 3, 7>; // Floating point double square root (ZMM).
+defm : SKXWriteResPair<WriteFSqrt64X, [SKXPort0,SKXFPDivider], 18, [1,6], 1, 6>;
+defm : SKXWriteResPair<WriteFSqrt64Y, [SKXPort0,SKXFPDivider], 18, [1,12],1, 7>;
+defm : SKXWriteResPair<WriteFSqrt64Z, [SKXPort0,SKXPort5,SKXFPDivider], 32, [2,1,24], 3, 7>;
 defm : SKXWriteResPair<WriteFSqrt80,  [SKXPort0,SKXFPDivider], 21, [1,7]>; // Floating point long double square root.
 
 defm : SKXWriteResPair<WriteFRcp,   [SKXPort0],  4, [1], 1, 5>; // Floating point reciprocal estimate.
-defm : SKXWriteResPair<WriteFRcpX,  [SKXPort0],  4, [1], 1, 6>; // Floating point reciprocal estimate (XMM).
-defm : SKXWriteResPair<WriteFRcpY,  [SKXPort0],  4, [1], 1, 7>; // Floating point reciprocal estimate (YMM/ZMM).
+defm : SKXWriteResPair<WriteFRcpX,  [SKXPort0],  4, [1], 1, 6>;
+defm : SKXWriteResPair<WriteFRcpY,  [SKXPort0],  4, [1], 1, 7>;
+defm : SKXWriteResPair<WriteFRcpZ,  [SKXPort0,SKXPort5],  4, [2,1], 3, 7>;
 
 defm : SKXWriteResPair<WriteFRsqrt, [SKXPort0],  4, [1], 1, 5>; // Floating point reciprocal square root estimate.
-defm : SKXWriteResPair<WriteFRsqrtX,[SKXPort0],  4, [1], 1, 6>; // Floating point reciprocal square root estimate (XMM).
-defm : SKXWriteResPair<WriteFRsqrtY,[SKXPort0],  4, [1], 1, 7>; // Floating point reciprocal square root estimate (YMM/ZMM).
-
-defm : SKXWriteResPair<WriteFMA,  [SKXPort015],  4, [1], 1, 5>; // Fused Multiply Add.
-defm : SKXWriteResPair<WriteFMAX, [SKXPort015],  4, [1], 1, 6>; // Fused Multiply Add (XMM).
-defm : SKXWriteResPair<WriteFMAY, [SKXPort015],  4, [1], 1, 7>; // Fused Multiply Add (YMM/ZMM).
+defm : SKXWriteResPair<WriteFRsqrtX,[SKXPort0],  4, [1], 1, 6>;
+defm : SKXWriteResPair<WriteFRsqrtY,[SKXPort0],  4, [1], 1, 7>;
+defm : SKXWriteResPair<WriteFRsqrtZ,[SKXPort0,SKXPort5],  9, [2,1], 3, 7>;
+
+defm : SKXWriteResPair<WriteFMA,  [SKXPort01],  4, [1], 1, 5>; // Fused Multiply Add.
+defm : SKXWriteResPair<WriteFMAX, [SKXPort01],  4, [1], 1, 6>;
+defm : SKXWriteResPair<WriteFMAY, [SKXPort01],  4, [1], 1, 7>;
+defm : SKXWriteResPair<WriteFMAZ, [SKXPort05],  4, [1], 1, 7>;
 defm : SKXWriteResPair<WriteDPPD, [SKXPort5,SKXPort015],  9, [1,2], 3, 6>; // Floating point double dot product.
-defm : SKXWriteResPair<WriteDPPS, [SKXPort5,SKXPort015], 13, [1,3], 4, 6>; // Floating point single dot product.
-defm : SKXWriteResPair<WriteDPPSY,[SKXPort5,SKXPort015], 13, [1,3], 4, 7>; // Floating point single dot product (YMM).
+defm : SKXWriteResPair<WriteDPPS, [SKXPort5,SKXPort015], 13, [1,3], 4, 6>;
+defm : SKXWriteResPair<WriteDPPSY,[SKXPort5,SKXPort015], 13, [1,3], 4, 7>;
+defm : SKXWriteResPair<WriteDPPSZ,[SKXPort5,SKXPort015], 13, [1,3], 4, 7>;
 defm : SKXWriteResPair<WriteFSign,  [SKXPort0],  1>; // Floating point fabs/fchs.
-defm : SKXWriteResPair<WriteFRnd,   [SKXPort015], 8, [2], 2, 6>; // Floating point rounding.
-defm : SKXWriteResPair<WriteFRndY,  [SKXPort015], 8, [2], 2, 7>; // Floating point rounding (YMM/ZMM).
+defm : SKXWriteResPair<WriteFRnd,   [SKXPort01], 8, [2], 2, 6>; // Floating point rounding.
+defm : SKXWriteResPair<WriteFRndY,  [SKXPort01], 8, [2], 2, 7>;
+defm : SKXWriteResPair<WriteFRndZ,  [SKXPort05], 8, [2], 2, 7>;
 defm : SKXWriteResPair<WriteFLogic, [SKXPort015], 1, [1], 1, 6>; // Floating point and/or/xor logicals.
-defm : SKXWriteResPair<WriteFLogicY, [SKXPort015], 1, [1], 1, 7>; // Floating point and/or/xor logicals (YMM/ZMM).
+defm : SKXWriteResPair<WriteFLogicY, [SKXPort015], 1, [1], 1, 7>;
+defm : SKXWriteResPair<WriteFLogicZ, [SKXPort05], 1, [1], 1, 7>;
 defm : SKXWriteResPair<WriteFTest,  [SKXPort0], 2, [1], 1, 6>; // Floating point TEST instructions.
-defm : SKXWriteResPair<WriteFTestY, [SKXPort0], 2, [1], 1, 7>; // Floating point TEST instructions (YMM/ZMM).
+defm : SKXWriteResPair<WriteFTestY, [SKXPort0], 2, [1], 1, 7>;
+defm : SKXWriteResPair<WriteFTestZ, [SKXPort0], 2, [1], 1, 7>;
 defm : SKXWriteResPair<WriteFShuffle,  [SKXPort5], 1, [1], 1, 6>; // Floating point vector shuffles.
-defm : SKXWriteResPair<WriteFShuffleY, [SKXPort5], 1, [1], 1, 7>; // Floating point vector shuffles (YMM/ZMM).
+defm : SKXWriteResPair<WriteFShuffleY, [SKXPort5], 1, [1], 1, 7>;
+defm : SKXWriteResPair<WriteFShuffleZ, [SKXPort5], 1, [1], 1, 7>;
 defm : SKXWriteResPair<WriteFVarShuffle,  [SKXPort5], 1, [1], 1, 6>; // Floating point vector variable shuffles.
-defm : SKXWriteResPair<WriteFVarShuffleY, [SKXPort5], 1, [1], 1, 7>; // Floating point vector variable shuffles.
+defm : SKXWriteResPair<WriteFVarShuffleY, [SKXPort5], 1, [1], 1, 7>;
+defm : SKXWriteResPair<WriteFVarShuffleZ, [SKXPort5], 1, [1], 1, 7>;
 defm : SKXWriteResPair<WriteFBlend, [SKXPort015], 1, [1], 1, 6>; // Floating point vector blends.
-defm : SKXWriteResPair<WriteFBlendY,[SKXPort015], 1, [1], 1, 7>; // Floating point vector blends.
+defm : SKXWriteResPair<WriteFBlendY,[SKXPort015], 1, [1], 1, 7>;
+defm : SKXWriteResPair<WriteFBlendZ,[SKXPort015], 1, [1], 1, 7>;
 defm : SKXWriteResPair<WriteFVarBlend, [SKXPort015], 2, [2], 2, 6>; // Fp vector variable blends.
-defm : SKXWriteResPair<WriteFVarBlendY,[SKXPort015], 2, [2], 2, 7>; // Fp vector variable blends.
+defm : SKXWriteResPair<WriteFVarBlendY,[SKXPort015], 2, [2], 2, 7>;
+defm : SKXWriteResPair<WriteFVarBlendZ,[SKXPort015], 2, [2], 2, 7>;
 
 // FMA Scheduling helper class.
 // class FMASC { X86FoldableSchedWrite Sched = WriteFAdd; }
@@ -279,47 +296,62 @@
 defm : X86WriteRes<WriteVecMoveFromGpr,  [SKXPort5], 1, [1], 1>;
 
 defm : SKXWriteResPair<WriteVecALU,   [SKXPort05], 1, [1], 1, 5>; // Vector integer ALU op, no logicals.
-defm : SKXWriteResPair<WriteVecALUX,  [SKXPort01], 1, [1], 1, 6>; // Vector integer ALU op, no logicals (XMM).
-defm : SKXWriteResPair<WriteVecALUY,  [SKXPort01], 1, [1], 1, 7>; // Vector integer ALU op, no logicals (YMM/ZMM).
+defm : SKXWriteResPair<WriteVecALUX,  [SKXPort01], 1, [1], 1, 6>;
+defm : SKXWriteResPair<WriteVecALUY,  [SKXPort01], 1, [1], 1, 7>;
+defm : SKXWriteResPair<WriteVecALUZ,  [SKXPort0], 1, [1], 1, 7>;
 defm : SKXWriteResPair<WriteVecLogic, [SKXPort05],  1, [1], 1, 5>; // Vector integer and/or/xor.
-defm : SKXWriteResPair<WriteVecLogicX,[SKXPort015], 1, [1], 1, 6>; // Vector integer and/or/xor (XMM).
-defm : SKXWriteResPair<WriteVecLogicY,[SKXPort015], 1, [1], 1, 7>; // Vector integer and/or/xor (YMM/ZMM).
+defm : SKXWriteResPair<WriteVecLogicX,[SKXPort015], 1, [1], 1, 6>;
+defm : SKXWriteResPair<WriteVecLogicY,[SKXPort015], 1, [1], 1, 7>;
+defm : SKXWriteResPair<WriteVecLogicZ,[SKXPort05], 1, [1], 1, 7>;
 defm : SKXWriteResPair<WriteVecTest,  [SKXPort0,SKXPort5], 3, [1,1], 2, 6>; // Vector integer TEST instructions.
-defm : SKXWriteResPair<WriteVecTestY, [SKXPort0,SKXPort5], 3, [1,1], 2, 7>; // Vector integer TEST instructions (YMM/ZMM).
+defm : SKXWriteResPair<WriteVecTestY, [SKXPort0,SKXPort5], 3, [1,1], 2, 7>;
+defm : SKXWriteResPair<WriteVecTestZ, [SKXPort0,SKXPort5], 3, [1,1], 2, 7>;
 defm : SKXWriteResPair<WriteVecIMul,  [SKXPort0],    4, [1], 1, 5>; // Vector integer multiply.
-defm : SKXWriteResPair<WriteVecIMulX, [SKXPort015],  4, [1], 1, 6>; // Vector integer multiply (XMM).
-defm : SKXWriteResPair<WriteVecIMulY, [SKXPort015],  4, [1], 1, 7>; // Vector integer multiply (YMM/ZMM).
-defm : SKXWriteResPair<WritePMULLD,   [SKXPort015], 10, [2], 2, 6>; // Vector PMULLD.
-defm : SKXWriteResPair<WritePMULLDY,  [SKXPort015], 10, [2], 2, 7>; // Vector PMULLD (YMM/ZMM).
+defm : SKXWriteResPair<WriteVecIMulX, [SKXPort01],  4, [1], 1, 6>;
+defm : SKXWriteResPair<WriteVecIMulY, [SKXPort01],  4, [1], 1, 7>;
+defm : SKXWriteResPair<WriteVecIMulZ, [SKXPort05],  4, [1], 1, 7>;
+defm : SKXWriteResPair<WritePMULLD,   [SKXPort01], 10, [2], 2, 6>; // Vector PMULLD.
+defm : SKXWriteResPair<WritePMULLDY,  [SKXPort01], 10, [2], 2, 7>;
+defm : SKXWriteResPair<WritePMULLDZ,  [SKXPort05], 10, [2], 2, 7>;
 defm : SKXWriteResPair<WriteShuffle,  [SKXPort5], 1, [1], 1, 5>; // Vector shuffles.
-defm : SKXWriteResPair<WriteShuffleX, [SKXPort5], 1, [1], 1, 6>; // Vector shuffles (XMM).
-defm : SKXWriteResPair<WriteShuffleY, [SKXPort5], 1, [1], 1, 7>; // Vector shuffles (YMM/ZMM).
+defm : SKXWriteResPair<WriteShuffleX, [SKXPort5], 1, [1], 1, 6>;
+defm : SKXWriteResPair<WriteShuffleY, [SKXPort5], 1, [1], 1, 7>;
+defm : SKXWriteResPair<WriteShuffleZ, [SKXPort5], 1, [1], 1, 7>;
 defm : SKXWriteResPair<WriteVarShuffle,  [SKXPort5], 1, [1], 1, 5>; // Vector variable shuffles.
-defm : SKXWriteResPair<WriteVarShuffleX, [SKXPort5], 1, [1], 1, 6>; // Vector variable shuffles (XMM).
-defm : SKXWriteResPair<WriteVarShuffleY, [SKXPort5], 1, [1], 1, 7>; // Vector variable shuffles (YMM/ZMM).
+defm : SKXWriteResPair<WriteVarShuffleX, [SKXPort5], 1, [1], 1, 6>;
+defm : SKXWriteResPair<WriteVarShuffleY, [SKXPort5], 1, [1], 1, 7>;
+defm : SKXWriteResPair<WriteVarShuffleZ, [SKXPort5], 1, [1], 1, 7>;
 defm : SKXWriteResPair<WriteBlend, [SKXPort5], 1, [1], 1, 6>; // Vector blends.
-defm : SKXWriteResPair<WriteBlendY,[SKXPort5], 1, [1], 1, 7>; // Vector blends (YMM/ZMM).
+defm : SKXWriteResPair<WriteBlendY,[SKXPort5], 1, [1], 1, 7>;
+defm : SKXWriteResPair<WriteBlendZ,[SKXPort5], 1, [1], 1, 7>;
 defm : SKXWriteResPair<WriteVarBlend, [SKXPort015], 2, [2], 2, 6>; // Vector variable blends.
-defm : SKXWriteResPair<WriteVarBlendY,[SKXPort015], 2, [2], 2, 6>; // Vector variable blends (YMM/ZMM).
+defm : SKXWriteResPair<WriteVarBlendY,[SKXPort015], 2, [2], 2, 6>;
+defm : SKXWriteResPair<WriteVarBlendZ,[SKXPort05],  2, [1], 1, 6>;
 defm : SKXWriteResPair<WriteMPSAD,   [SKXPort5], 4, [2], 2, 6>; // Vector MPSAD.
-defm : SKXWriteResPair<WriteMPSADY,  [SKXPort5], 4, [2], 2, 7>; // Vector MPSAD.
+defm : SKXWriteResPair<WriteMPSADY,  [SKXPort5], 4, [2], 2, 7>;
+defm : SKXWriteResPair<WriteMPSADZ,  [SKXPort5], 4, [2], 2, 7>;
 defm : SKXWriteResPair<WritePSADBW,  [SKXPort5], 3, [1], 1, 5>; // Vector PSADBW.
-defm : SKXWriteResPair<WritePSADBWX, [SKXPort5], 3, [1], 1, 6>; // Vector PSADBW.
-defm : SKXWriteResPair<WritePSADBWY, [SKXPort5], 3, [1], 1, 7>; // Vector PSADBW.
-defm : SKXWriteResPair<WritePHMINPOS, [SKXPort015], 4, [1], 1, 6>; // Vector PHMINPOS.
+defm : SKXWriteResPair<WritePSADBWX, [SKXPort5], 3, [1], 1, 6>;
+defm : SKXWriteResPair<WritePSADBWY, [SKXPort5], 3, [1], 1, 7>;
+defm : SKXWriteResPair<WritePSADBWZ, [SKXPort5], 3, [1], 1, 7>;
+defm : SKXWriteResPair<WritePHMINPOS, [SKXPort0], 4, [1], 1, 6>; // Vector PHMINPOS.
 
 // Vector integer shifts.
 defm : SKXWriteResPair<WriteVecShift, [SKXPort0], 1, [1], 1, 5>;
 defm : X86WriteRes<WriteVecShiftX,    [SKXPort5,SKXPort01],  2, [1,1], 2>;
 defm : X86WriteRes<WriteVecShiftY,    [SKXPort5,SKXPort01],  4, [1,1], 2>;
+defm : X86WriteRes<WriteVecShiftZ,    [SKXPort5,SKXPort0],   4, [1,1], 2>;
 defm : X86WriteRes<WriteVecShiftXLd,  [SKXPort01,SKXPort23], 7, [1,1], 2>;
 defm : X86WriteRes<WriteVecShiftYLd,  [SKXPort01,SKXPort23], 8, [1,1], 2>;
+defm : X86WriteRes<WriteVecShiftZLd,  [SKXPort0,SKXPort23],  8, [1,1], 2>;
 
 defm : SKXWriteResPair<WriteVecShiftImm,  [SKXPort0],  1, [1], 1, 5>;
-defm : SKXWriteResPair<WriteVecShiftImmX, [SKXPort01], 1, [1], 1, 6>; // Vector integer immediate shifts (XMM).
-defm : SKXWriteResPair<WriteVecShiftImmY, [SKXPort01], 1, [1], 1, 7>; // Vector integer immediate shifts (YMM/ZMM).
+defm : SKXWriteResPair<WriteVecShiftImmX, [SKXPort01], 1, [1], 1, 6>; // Vector integer immediate shifts.
+defm : SKXWriteResPair<WriteVecShiftImmY, [SKXPort01], 1, [1], 1, 7>;
+defm : SKXWriteResPair<WriteVecShiftImmZ, [SKXPort0], 1, [1], 1, 7>;
 defm : SKXWriteResPair<WriteVarVecShift,  [SKXPort01], 1, [1], 1, 6>; // Variable vector shifts.
-defm : SKXWriteResPair<WriteVarVecShiftY, [SKXPort01], 1, [1], 1, 7>; // Variable vector shifts (YMM/ZMM).
+defm : SKXWriteResPair<WriteVarVecShiftY, [SKXPort01], 1, [1], 1, 7>;
+defm : SKXWriteResPair<WriteVarVecShiftZ, [SKXPort0], 1, [1], 1, 7>;
 
 // Vector insert/extract operations.
 def : WriteRes<WriteVecInsert, [SKXPort5]> {
@@ -343,36 +375,46 @@
 }
 
 // Conversion between integer and float.
-defm : SKXWriteResPair<WriteCvtSS2I,   [SKXPort0,SKXPort015], 6, [1,1], 2>;
-defm : SKXWriteResPair<WriteCvtPS2I,   [SKXPort1], 3>;
-defm : SKXWriteResPair<WriteCvtPS2IY,  [SKXPort1], 3>;
-defm : SKXWriteResPair<WriteCvtSD2I,   [SKXPort0,SKXPort015], 6, [1,1], 2>;
-defm : SKXWriteResPair<WriteCvtPD2I,   [SKXPort1], 3>;
-defm : SKXWriteResPair<WriteCvtPD2IY,  [SKXPort1], 3>;
+defm : SKXWriteResPair<WriteCvtSS2I,   [SKXPort01], 6, [2], 2>; // Needs more work: DD vs DQ.
+defm : SKXWriteResPair<WriteCvtPS2I,   [SKXPort01], 3>;
+defm : SKXWriteResPair<WriteCvtPS2IY,  [SKXPort01], 3>;
+defm : SKXWriteResPair<WriteCvtPS2IZ,  [SKXPort05], 3>;
+defm : SKXWriteResPair<WriteCvtSD2I,   [SKXPort01], 6, [2], 2>;
+defm : SKXWriteResPair<WriteCvtPD2I,   [SKXPort01], 3>;
+defm : SKXWriteResPair<WriteCvtPD2IY,  [SKXPort01], 3>;
+defm : SKXWriteResPair<WriteCvtPD2IZ,  [SKXPort05], 3>;
 
 defm : SKXWriteResPair<WriteCvtI2SS,   [SKXPort1], 4>;
-defm : SKXWriteResPair<WriteCvtI2PS,   [SKXPort1], 4>;
-defm : SKXWriteResPair<WriteCvtI2PSY,  [SKXPort1], 4>;
+defm : SKXWriteResPair<WriteCvtI2PS,   [SKXPort01], 4>;
+defm : SKXWriteResPair<WriteCvtI2PSY,  [SKXPort01], 4>;
+defm : SKXWriteResPair<WriteCvtI2PSZ,  [SKXPort05], 4>;  // Needs more work: DD vs DQ.
 defm : SKXWriteResPair<WriteCvtI2SD,   [SKXPort1], 4>;
-defm : SKXWriteResPair<WriteCvtI2PD,   [SKXPort0,SKXPort5], 5, [1,1], 2>;
-defm : SKXWriteResPair<WriteCvtI2PDY,  [SKXPort1], 4>;
+defm : SKXWriteResPair<WriteCvtI2PD,   [SKXPort01], 4>;
+defm : SKXWriteResPair<WriteCvtI2PDY,  [SKXPort01], 4>;
+defm : SKXWriteResPair<WriteCvtI2PDZ,  [SKXPort05], 4>;
 
 defm : SKXWriteResPair<WriteCvtSS2SD,  [SKXPort1], 3>;
 defm : SKXWriteResPair<WriteCvtPS2PD,  [SKXPort1], 3>;
-defm : SKXWriteResPair<WriteCvtPS2PDY, [SKXPort1], 3>;
+defm : SKXWriteResPair<WriteCvtPS2PDY, [SKXPort5,SKXPort01], 3, [1,1], 2>;
+defm : SKXWriteResPair<WriteCvtPS2PDZ, [SKXPort05], 3, [2], 2>;
 defm : SKXWriteResPair<WriteCvtSD2SS,  [SKXPort1], 3>;
 defm : SKXWriteResPair<WriteCvtPD2PS,  [SKXPort1], 3>;
-defm : SKXWriteResPair<WriteCvtPD2PSY, [SKXPort1], 3>;
-
-defm : X86WriteRes<WriteCvtPH2PS,     [SKXPort5,SKXPort015],  5, [1,1], 2>;
-defm : X86WriteRes<WriteCvtPH2PSY,    [SKXPort5,SKXPort015],  7, [1,1], 2>;
-defm : X86WriteRes<WriteCvtPH2PSLd,  [SKXPort23,SKXPort015],  9, [1,1], 2>;
-defm : X86WriteRes<WriteCvtPH2PSYLd, [SKXPort23,SKXPort015], 10, [1,1], 2>;
-
-defm : X86WriteRes<WriteCvtPS2PH,                        [SKXPort5,SKXPort015], 5, [1,1], 2>;
-defm : X86WriteRes<WriteCvtPS2PHY,                       [SKXPort5,SKXPort015], 7, [1,1], 2>;
-defm : X86WriteRes<WriteCvtPS2PHSt,  [SKXPort4,SKXPort5,SKXPort237,SKXPort015], 6, [1,1,1,1], 4>;
-defm : X86WriteRes<WriteCvtPS2PHYSt, [SKXPort4,SKXPort5,SKXPort237,SKXPort015], 8, [1,1,1,1], 4>;
+defm : SKXWriteResPair<WriteCvtPD2PSY, [SKXPort5,SKXPort01], 3, [1,1], 2>;
+defm : SKXWriteResPair<WriteCvtPD2PSZ, [SKXPort05], 3, [2], 2>;
+
+defm : X86WriteRes<WriteCvtPH2PS,     [SKXPort5,SKXPort01],  5, [1,1], 2>;
+defm : X86WriteRes<WriteCvtPH2PSY,    [SKXPort5,SKXPort01],  7, [1,1], 2>;
+defm : X86WriteRes<WriteCvtPH2PSZ,    [SKXPort5,SKXPort0],   7, [1,1], 2>;
+defm : X86WriteRes<WriteCvtPH2PSLd,  [SKXPort23,SKXPort01],  9, [1,1], 2>;
+defm : X86WriteRes<WriteCvtPH2PSYLd, [SKXPort23,SKXPort01], 10, [1,1], 2>;
+defm : X86WriteRes<WriteCvtPH2PSZLd, [SKXPort23,SKXPort05], 10, [1,1], 2>;
+
+defm : X86WriteRes<WriteCvtPS2PH,    [SKXPort5,SKXPort01], 5, [1,1], 2>;
+defm : X86WriteRes<WriteCvtPS2PHY,   [SKXPort5,SKXPort01], 7, [1,1], 2>;
+defm : X86WriteRes<WriteCvtPS2PHZ,   [SKXPort5,SKXPort05], 7, [1,1], 2>;
+defm : X86WriteRes<WriteCvtPS2PHSt,  [SKXPort4,SKXPort5,SKXPort237,SKXPort01], 6, [1,1,1,1], 4>;
+defm : X86WriteRes<WriteCvtPS2PHYSt, [SKXPort4,SKXPort5,SKXPort237,SKXPort01], 8, [1,1,1,1], 4>;
+defm : X86WriteRes<WriteCvtPS2PHZSt, [SKXPort4,SKXPort5,SKXPort237,SKXPort05], 8, [1,1,1,1], 4>;
 
 // Strings instructions.
 
@@ -564,7 +606,7 @@
   let NumMicroOps = 1;
   let ResourceCycles = [1];
 }
-def: InstRW<[SKXWriteResGroup7], (instrs CDQ, CQO, CLAC, STAC)>;
+def: InstRW<[SKXWriteResGroup7], (instrs CDQ, CQO, CLAC, STAC, LAHF, SAHF)>;
 def: InstRW<[SKXWriteResGroup7], (instregex "BT(16|32|64)ri8",
                                             "BT(16|32|64)rr",
                                             "BTC(16|32|64)ri8",
@@ -589,15 +631,15 @@
   let NumMicroOps = 1;
   let ResourceCycles = [1];
 }
-def: InstRW<[SKXWriteResGroup9], (instregex "VBLENDMPD(Z|Z128|Z256)rr",
-                                            "VBLENDMPS(Z|Z128|Z256)rr",
+def: InstRW<[SKXWriteResGroup9], (instregex "VBLENDMPD(Z128|Z256)rr",
+                                            "VBLENDMPS(Z128|Z256)rr",
                                             "VPADD(B|D|Q|W)(Y|Z|Z128|Z256)rr",
                                             "(V?)PADD(B|D|Q|W)rr",
                                             "VPBLENDD(Y?)rri",
-                                            "VPBLENDMB(Z|Z128|Z256)rr",
-                                            "VPBLENDMD(Z|Z128|Z256)rr",
-                                            "VPBLENDMQ(Z|Z128|Z256)rr",
-                                            "VPBLENDMW(Z|Z128|Z256)rr",
+                                            "VPBLENDMB(Z128|Z256)rr",
+                                            "VPBLENDMD(Z128|Z256)rr",
+                                            "VPBLENDMQ(Z128|Z256)rr",
+                                            "VPBLENDMW(Z128|Z256)rr",
                                             "VPSUB(B|D|Q|W)(Y|Z|Z128|Z256)rr",
                                             "(V?)PSUB(B|D|Q|W)rr",
                                             "VPTERNLOGD(Z|Z128|Z256)rri",
@@ -610,9 +652,7 @@
 }
 def: InstRW<[SKXWriteResGroup10], (instrs CBW, CWDE, CDQE,
                                           CMC, STC)>;
-def: InstRW<[SKXWriteResGroup10], (instrs LAHF, SAHF)>; // TODO: This doesn't match Agner's data
-def: InstRW<[SKXWriteResGroup10], (instregex "NOOP",
-                                             "SGDT64m",
+def: InstRW<[SKXWriteResGroup10], (instregex "SGDT64m",
                                              "SIDT64m",
                                              "SMSW16m",
                                              "STRm",
@@ -739,8 +779,7 @@
   let NumMicroOps = 1;
   let ResourceCycles = [1];
 }
-def: InstRW<[SKXWriteResGroup30], (instregex "KADD(B|D|Q|W)rr",
-                                             "KMOV(B|D|Q|W)rk",
+def: InstRW<[SKXWriteResGroup30], (instregex "KMOV(B|D|Q|W)rk",
                                              "KORTEST(B|D|Q|W)rr",
                                              "KTEST(B|D|Q|W)rr")>;
 
@@ -768,6 +807,7 @@
   let ResourceCycles = [1];
 }
 def: InstRW<[SKXWriteResGroup32], (instregex "(ADD|SUB|SUBR)_(FPrST0|FST0r|FrST0)",
+                                             "KADD(B|D|Q|W)rr",
                                              "KSHIFTL(B|D|Q|W)ri",
                                              "KSHIFTR(B|D|Q|W)ri",
                                              "KUNPCKBWrr",
@@ -907,26 +947,44 @@
 }
 def: InstRW<[SKXWriteResGroup49], (instregex "MUL_(FPrST0|FST0r|FrST0)")>;
 
-def SKXWriteResGroup50 : SchedWriteRes<[SKXPort015]> {
+def SKXWriteResGroup50 : SchedWriteRes<[SKXPort01]> {
   let Latency = 4;
   let NumMicroOps = 1;
   let ResourceCycles = [1];
 }
-def: InstRW<[SKXWriteResGroup50], (instregex "VCVTDQ2PS(Y|Z|Z128|Z256)rr",
+def: InstRW<[SKXWriteResGroup50], (instregex "VCVTDQ2PS(Y|Z128|Z256)rr",
                                              "(V?)CVTDQ2PSrr",
-                                             "VCVTPD2QQ(Z|Z128|Z256)rr",
-                                             "VCVTPD2UQQ(Z|Z128|Z256)rr",
-                                             "VCVTPS2DQ(Y|Z|Z128|Z256)rr",
+                                             "VCVTPD2QQ(Z128|Z256)rr",
+                                             "VCVTPD2UQQ(Z128|Z256)rr",
+                                             "VCVTPS2DQ(Y|Z128|Z256)rr",
                                              "(V?)CVTPS2DQrr",
-                                             "VCVTPS2UDQ(Z|Z128|Z256)rr",
-                                             "VCVTQQ2PD(Z|Z128|Z256)rr",
-                                             "VCVTTPD2QQ(Z|Z128|Z256)rr",
-                                             "VCVTTPD2UQQ(Z|Z128|Z256)rr",
-                                             "VCVTTPS2DQ(Y|Z|Z128|Z256)rr",
+                                             "VCVTPS2UDQ(Z128|Z256)rr",
+                                             "VCVTQQ2PD(Z128|Z256)rr",
+                                             "VCVTTPD2QQ(Z128|Z256)rr",
+                                             "VCVTTPD2UQQ(Z128|Z256)rr",
+                                             "VCVTTPS2DQ(Z128|Z256)rr",
                                              "(V?)CVTTPS2DQrr",
-                                             "VCVTTPS2UDQ(Z|Z128|Z256)rr",
-                                             "VCVTUDQ2PS(Z|Z128|Z256)rr",
-                                             "VCVTUQQ2PD(Z|Z128|Z256)rr")>;
+                                             "VCVTTPS2UDQ(Z128|Z256)rr",
+                                             "VCVTUDQ2PS(Z128|Z256)rr",
+                                             "VCVTUQQ2PD(Z128|Z256)rr")>;
+
+def SKXWriteResGroup50z : SchedWriteRes<[SKXPort05]> {
+  let Latency = 4;
+  let NumMicroOps = 1;
+  let ResourceCycles = [1];
+}
+def: InstRW<[SKXWriteResGroup50z], (instrs VCVTDQ2PSZrr,
+                                          VCVTPD2QQZrr,
+                                          VCVTPD2UQQZrr,
+                                          VCVTPS2DQZrr,
+                                          VCVTPS2UDQZrr,
+                                          VCVTQQ2PDZrr,
+                                          VCVTTPD2QQZrr,
+                                          VCVTTPD2UQQZrr,
+                                          VCVTTPS2DQZrr,
+                                          VCVTTPS2UDQZrr,
+                                          VCVTUDQ2PSZrr,
+                                          VCVTUQQ2PDZrr)>;
 
 def SKXWriteResGroup51 : SchedWriteRes<[SKXPort5]> {
   let Latency = 4;
@@ -1010,13 +1068,6 @@
                                              "MOVZX(16|32|64)rm8",
                                              "(V?)MOVDDUPrm")>;  // TODO: Should this be SKXWriteResGroup71?
 
-def SKXWriteResGroup59 : SchedWriteRes<[SKXPort015]> {
-  let Latency = 5;
-  let NumMicroOps = 2;
-  let ResourceCycles = [2];
-}
-def: InstRW<[SKXWriteResGroup59], (instregex "VCVTSD2SSZrr")>;
-
 def SKXWriteResGroup61 : SchedWriteRes<[SKXPort5,SKXPort015]> {
   let Latency = 5;
   let NumMicroOps = 2;
@@ -1036,6 +1087,7 @@
                                              "VCVTPS2UQQZ128rr",
                                              "VCVTQQ2PSZ128rr",
                                              "(V?)CVTSD2SSrr",
+                                             "VCVTSD2SSZrr",
                                              "(V?)CVTSI(64)?2SDrr",
                                              "VCVTSI2SSZrr",
                                              "(V?)CVTSI2SSrr",
@@ -1136,7 +1188,7 @@
                                              "VPBROADCASTDrm",
                                              "VPBROADCASTQrm")>;
 
-def SKXWriteResGroup72 : SchedWriteRes<[SKXPort0]> {
+def SKXWriteResGroup72 : SchedWriteRes<[SKXPort5]> {
   let Latency = 6;
   let NumMicroOps = 2;
   let ResourceCycles = [2];
@@ -1286,7 +1338,7 @@
                                              "VPBROADCASTDYrm",
                                              "VPBROADCASTQYrm")>;
 
-def SKXWriteResGroup90 : SchedWriteRes<[SKXPort0,SKXPort5]> {
+def SKXWriteResGroup90 : SchedWriteRes<[SKXPort01,SKXPort5]> {
   let Latency = 7;
   let NumMicroOps = 2;
   let ResourceCycles = [1,1];
@@ -1318,21 +1370,40 @@
   let NumMicroOps = 2;
   let ResourceCycles = [1,1];
 }
-def: InstRW<[SKXWriteResGroup93], (instregex "VCVTDQ2PD(Z|Z256)rr",
-                                             "VCVTPD2DQ(Y|Z|Z256)rr",
-                                             "VCVTPD2PS(Y|Z|Z256)rr",
-                                             "VCVTPD2UDQ(Z|Z256)rr",
-                                             "VCVTPS2PD(Y|Z|Z256)rr",
-                                             "VCVTPS2QQ(Z|Z256)rr",
-                                             "VCVTPS2UQQ(Z|Z256)rr",
-                                             "VCVTQQ2PS(Z|Z256)rr",
-                                             "VCVTTPD2DQ(Y|Z|Z256)rr",
-                                             "VCVTTPD2UDQ(Z|Z256)rr",
-                                             "VCVTTPS2QQ(Z|Z256)rr",
-                                             "VCVTTPS2UQQ(Z|Z256)rr",
-                                             "VCVTUDQ2PD(Z|Z256)rr",
-                                             "VCVTUQQ2PS(Z|Z256)rr")>;
-
+def: InstRW<[SKXWriteResGroup93], (instregex "VCVTDQ2PDZ256rr",
+                                             "VCVTPD2DQ(Y|Z256)rr",
+                                             "VCVTPD2PS(Y|Z256)rr",
+                                             "VCVTPD2UDQZ256rr",
+                                             "VCVTPS2PD(Y|Z256)rr",
+                                             "VCVTPS2QQZ256rr",
+                                             "VCVTPS2UQQZ256rr",
+                                             "VCVTQQ2PSZ256rr",
+                                             "VCVTTPD2DQ(Y|Z256)rr",
+                                             "VCVTTPD2UDQZ256rr",
+                                             "VCVTTPS2QQZ256rr",
+                                             "VCVTTPS2UQQZ256rr",
+                                             "VCVTUDQ2PDZ256rr",
+                                             "VCVTUQQ2PSZ256rr")>;
+
+def SKXWriteResGroup93z : SchedWriteRes<[SKXPort5,SKXPort05]> {
+  let Latency = 7;
+  let NumMicroOps = 2;
+  let ResourceCycles = [1,1];
+}
+def: InstRW<[SKXWriteResGroup93z], (instregex "VCVTDQ2PDZrr",
+                                              "VCVTPD2DQZrr",
+                                              "VCVTPD2PSZrr",
+                                              "VCVTPD2UDQZrr",
+                                              "VCVTPS2PDZrr",
+                                              "VCVTPS2QQZrr",
+                                              "VCVTPS2UQQZrr",
+                                              "VCVTQQ2PSZrr",
+                                              "VCVTTPD2DQZrr",
+                                              "VCVTTPD2UDQZrr",
+                                              "VCVTTPS2QQZrr",
+                                              "VCVTTPS2UQQZrr",
+                                              "VCVTUDQ2PDZrr",
+                                              "VCVTUQQ2PSZrr")>;
 
 def SKXWriteResGroup95 : SchedWriteRes<[SKXPort23,SKXPort015]> {
   let Latency = 7;
@@ -1405,8 +1476,8 @@
   let ResourceCycles = [1,1,1];
 }
 def: InstRW<[SKXWriteResGroup100], (instregex "VCVTSS2USI64Zrr",
-                                              "VCVTTSS2SI(64)?Zrr",
-                                              "(V?)CVTTSS2SI(64)?rr",
+                                              "(V?)CVTSS2SI64(Z?)rr",
+                                              "(V?)CVTTSS2SI64(Z?)rr",
                                               "VCVTTSS2USI64Zrr")>;
 
 def SKXWriteResGroup101 : SchedWriteRes<[SKXPort0,SKXPort23,SKXPort05]> {
@@ -1751,16 +1822,6 @@
 def: InstRW<[SKXWriteResGroup137], (instregex "MMX_CVT(T?)PS2PIirm",
                                               "(V?)CVTPS2PDrm")>;
 
-def SKXWriteResGroup138 : SchedWriteRes<[SKXPort0,SKXPort015]> {
-  let Latency = 9;
-  let NumMicroOps = 3;
-  let ResourceCycles = [2,1];
-}
-def: InstRW<[SKXWriteResGroup138], (instregex "VRCP14PDZr(b?)",
-                                              "VRCP14PSZr(b?)",
-                                              "VRSQRT14PDZr(b?)",
-                                              "VRSQRT14PSZr(b?)")>;
-
 def SKXWriteResGroup142 : SchedWriteRes<[SKXPort1,SKXPort5,SKXPort23]> {
   let Latency = 9;
   let NumMicroOps = 3;
@@ -2009,12 +2070,19 @@
 }
 def: InstRW<[SKXWriteResGroup171], (instrs LOOPE, LOOPNE)>;
 
-def SKXWriteResGroup174 : SchedWriteRes<[SKXPort015]> {
+def SKXWriteResGroup174 : SchedWriteRes<[SKXPort01]> {
+  let Latency = 12;
+  let NumMicroOps = 3;
+  let ResourceCycles = [3];
+}
+def: InstRW<[SKXWriteResGroup174], (instregex "VPMULLQ(Z128|Z256)rr")>;
+
+def SKXWriteResGroup174z : SchedWriteRes<[SKXPort05]> {
   let Latency = 12;
   let NumMicroOps = 3;
   let ResourceCycles = [3];
 }
-def: InstRW<[SKXWriteResGroup174], (instregex "VPMULLQ(Z|Z128|Z256)rr")>;
+def: InstRW<[SKXWriteResGroup174z], (instregex "VPMULLQZrr")>;
 
 def SKXWriteResGroup175 : SchedWriteRes<[SKXPort5,SKXPort23]> {
   let Latency = 12;
@@ -2143,16 +2211,6 @@
 }
 def: InstRW<[SKXWriteResGroup195], (instregex "RCL(8|16|32|64)mCL")>;
 
-def SKXWriteResGroup198 : SchedWriteRes<[SKXPort0,SKXPort23,SKXPort015]> {
-  let Latency = 16;
-  let NumMicroOps = 4;
-  let ResourceCycles = [2,1,1];
-}
-def: InstRW<[SKXWriteResGroup198], (instregex "VRCP14PDZm(b?)",
-                                              "VRCP14PSZm(b?)",
-                                              "VRSQRT14PDZm(b?)",
-                                              "VRSQRT14PSZm(b?)")>;
-
 def SKXWriteResGroup199 : SchedWriteRes<[SKXPort4,SKXPort23,SKXPort237,SKXPort06,SKXPort15,SKXPort0156]> {
   let Latency = 16;
   let NumMicroOps = 14;
Index: lib/Target/X86/X86Schedule.td
===================================================================
--- lib/Target/X86/X86Schedule.td
+++ lib/Target/X86/X86Schedule.td
@@ -62,7 +62,6 @@
   }
 }
 
-
 // Multiclass that wraps X86FoldableSchedWrite for each vector width.
 class X86SchedWriteWidths<X86FoldableSchedWrite sScl,
                           X86FoldableSchedWrite s128,
@@ -177,23 +176,29 @@
 
 defm WriteFAdd    : X86SchedWritePair; // Floating point add/sub.
 defm WriteFAddX   : X86SchedWritePair; // Floating point add/sub (XMM).
-defm WriteFAddY   : X86SchedWritePair; // Floating point add/sub (YMM/ZMM).
+defm WriteFAddY   : X86SchedWritePair; // Floating point add/sub (YMM).
+defm WriteFAddZ   : X86SchedWritePair; // Floating point add/sub (ZMM).
 defm WriteFAdd64  : X86SchedWritePair; // Floating point double add/sub.
 defm WriteFAdd64X : X86SchedWritePair; // Floating point double add/sub (XMM).
-defm WriteFAdd64Y : X86SchedWritePair; // Floating point double add/sub (YMM/ZMM).
+defm WriteFAdd64Y : X86SchedWritePair; // Floating point double add/sub (YMM).
+defm WriteFAdd64Z : X86SchedWritePair; // Floating point double add/sub (ZMM).
 defm WriteFCmp    : X86SchedWritePair; // Floating point compare.
 defm WriteFCmpX   : X86SchedWritePair; // Floating point compare (XMM).
-defm WriteFCmpY   : X86SchedWritePair; // Floating point compare (YMM/ZMM).
+defm WriteFCmpY   : X86SchedWritePair; // Floating point compare (YMM).
+defm WriteFCmpZ   : X86SchedWritePair; // Floating point compare (ZMM).
 defm WriteFCmp64  : X86SchedWritePair; // Floating point double compare.
 defm WriteFCmp64X : X86SchedWritePair; // Floating point double compare (XMM).
-defm WriteFCmp64Y : X86SchedWritePair; // Floating point double compare (YMM/ZMM).
+defm WriteFCmp64Y : X86SchedWritePair; // Floating point double compare (YMM).
+defm WriteFCmp64Z : X86SchedWritePair; // Floating point double compare (ZMM).
 defm WriteFCom    : X86SchedWritePair; // Floating point compare to flags.
 defm WriteFMul    : X86SchedWritePair; // Floating point multiplication.
 defm WriteFMulX   : X86SchedWritePair; // Floating point multiplication (XMM).
-defm WriteFMulY   : X86SchedWritePair; // Floating point multiplication (YMM/ZMM).
+defm WriteFMulY   : X86SchedWritePair; // Floating point multiplication (YMM).
+defm WriteFMulZ   : X86SchedWritePair; // Floating point multiplication (YMM).
 defm WriteFMul64  : X86SchedWritePair; // Floating point double multiplication.
 defm WriteFMul64X : X86SchedWritePair; // Floating point double multiplication (XMM).
-defm WriteFMul64Y : X86SchedWritePair; // Floating point double multiplication (YMM/ZMM).
+defm WriteFMul64Y : X86SchedWritePair; // Floating point double multiplication (YMM).
+defm WriteFMul64Z : X86SchedWritePair; // Floating point double multiplication (ZMM).
 defm WriteFDiv    : X86SchedWritePair; // Floating point division.
 defm WriteFDivX   : X86SchedWritePair; // Floating point division (XMM).
 defm WriteFDivY   : X86SchedWritePair; // Floating point division (YMM).
@@ -213,41 +218,54 @@
 defm WriteFSqrt80  : X86SchedWritePair; // Floating point long double square root.
 defm WriteFRcp   : X86SchedWritePair; // Floating point reciprocal estimate.
 defm WriteFRcpX  : X86SchedWritePair; // Floating point reciprocal estimate (XMM).
-defm WriteFRcpY  : X86SchedWritePair; // Floating point reciprocal estimate (YMM/ZMM).
+defm WriteFRcpY  : X86SchedWritePair; // Floating point reciprocal estimate (YMM).
+defm WriteFRcpZ  : X86SchedWritePair; // Floating point reciprocal estimate (ZMM).
 defm WriteFRsqrt : X86SchedWritePair; // Floating point reciprocal square root estimate.
 defm WriteFRsqrtX: X86SchedWritePair; // Floating point reciprocal square root estimate (XMM).
-defm WriteFRsqrtY: X86SchedWritePair; // Floating point reciprocal square root estimate (YMM/ZMM).
+defm WriteFRsqrtY: X86SchedWritePair; // Floating point reciprocal square root estimate (YMM).
+defm WriteFRsqrtZ: X86SchedWritePair; // Floating point reciprocal square root estimate (ZMM).
 defm WriteFMA    : X86SchedWritePair; // Fused Multiply Add.
 defm WriteFMAX   : X86SchedWritePair; // Fused Multiply Add (XMM).
-defm WriteFMAY   : X86SchedWritePair; // Fused Multiply Add (YMM/ZMM).
+defm WriteFMAY   : X86SchedWritePair; // Fused Multiply Add (YMM).
+defm WriteFMAZ   : X86SchedWritePair; // Fused Multiply Add (ZMM).
 defm WriteDPPD   : X86SchedWritePair; // Floating point double dot product.
 defm WriteDPPS   : X86SchedWritePair; // Floating point single dot product.
 defm WriteDPPSY  : X86SchedWritePair; // Floating point single dot product (YMM).
+defm WriteDPPSZ  : X86SchedWritePair; // Floating point single dot product (ZMM).
 defm WriteFSign  : X86SchedWritePair; // Floating point fabs/fchs.
 defm WriteFRnd   : X86SchedWritePair; // Floating point rounding.
-defm WriteFRndY  : X86SchedWritePair; // Floating point rounding (YMM/ZMM).
+defm WriteFRndY  : X86SchedWritePair; // Floating point rounding (YMM).
+defm WriteFRndZ  : X86SchedWritePair; // Floating point rounding (ZMM).
 defm WriteFLogic  : X86SchedWritePair; // Floating point and/or/xor logicals.
-defm WriteFLogicY : X86SchedWritePair; // Floating point and/or/xor logicals (YMM/ZMM).
+defm WriteFLogicY : X86SchedWritePair; // Floating point and/or/xor logicals (YMM).
+defm WriteFLogicZ : X86SchedWritePair; // Floating point and/or/xor logicals (ZMM).
 defm WriteFTest   : X86SchedWritePair; // Floating point TEST instructions.
-defm WriteFTestY  : X86SchedWritePair; // Floating point TEST instructions (YMM/ZMM).
+defm WriteFTestY  : X86SchedWritePair; // Floating point TEST instructions (YMM).
+defm WriteFTestZ  : X86SchedWritePair; // Floating point TEST instructions (ZMM).
 defm WriteFShuffle  : X86SchedWritePair; // Floating point vector shuffles.
-defm WriteFShuffleY : X86SchedWritePair; // Floating point vector shuffles (YMM/ZMM).
+defm WriteFShuffleY : X86SchedWritePair; // Floating point vector shuffles (YMM).
+defm WriteFShuffleZ : X86SchedWritePair; // Floating point vector shuffles (ZMM).
 defm WriteFVarShuffle  : X86SchedWritePair; // Floating point vector variable shuffles.
-defm WriteFVarShuffleY : X86SchedWritePair; // Floating point vector variable shuffles (YMM/ZMM).
+defm WriteFVarShuffleY : X86SchedWritePair; // Floating point vector variable shuffles (YMM).
+defm WriteFVarShuffleZ : X86SchedWritePair; // Floating point vector variable shuffles (ZMM).
 defm WriteFBlend  : X86SchedWritePair; // Floating point vector blends.
-defm WriteFBlendY : X86SchedWritePair; // Floating point vector blends (YMM/ZMM).
+defm WriteFBlendY : X86SchedWritePair; // Floating point vector blends (YMM).
+defm WriteFBlendZ : X86SchedWritePair; // Floating point vector blends (ZMM).
 defm WriteFVarBlend  : X86SchedWritePair; // Fp vector variable blends.
-defm WriteFVarBlendY : X86SchedWritePair; // Fp vector variable blends (YMM/ZMM).
+defm WriteFVarBlendY : X86SchedWritePair; // Fp vector variable blends (YMM).
+defm WriteFVarBlendZ : X86SchedWritePair; // Fp vector variable blends (YMZMM).
 
 // FMA Scheduling helper class.
 class FMASC { X86FoldableSchedWrite Sched = WriteFAdd; }
 
 // Horizontal Add/Sub (float and integer)
 defm WriteFHAdd  : X86SchedWritePair;
-defm WriteFHAddY : X86SchedWritePair; // YMM/ZMM.
+defm WriteFHAddY : X86SchedWritePair;
+defm WriteFHAddZ : X86SchedWritePair;
 defm WritePHAdd  : X86SchedWritePair;
-defm WritePHAddX : X86SchedWritePair; // XMM.
-defm WritePHAddY : X86SchedWritePair; // YMM/ZMM.
+defm WritePHAddX : X86SchedWritePair;
+defm WritePHAddY : X86SchedWritePair;
+defm WritePHAddZ : X86SchedWritePair;
 
 // Vector integer operations.
 def  WriteVecLoad         : SchedWrite;
@@ -272,38 +290,51 @@
 
 defm WriteVecALU    : X86SchedWritePair; // Vector integer ALU op, no logicals.
 defm WriteVecALUX   : X86SchedWritePair; // Vector integer ALU op, no logicals (XMM).
-defm WriteVecALUY   : X86SchedWritePair; // Vector integer ALU op, no logicals (YMM/ZMM).
+defm WriteVecALUY   : X86SchedWritePair; // Vector integer ALU op, no logicals (YMM).
+defm WriteVecALUZ   : X86SchedWritePair; // Vector integer ALU op, no logicals (ZMM).
 defm WriteVecLogic  : X86SchedWritePair; // Vector integer and/or/xor logicals.
 defm WriteVecLogicX : X86SchedWritePair; // Vector integer and/or/xor logicals (XMM).
-defm WriteVecLogicY : X86SchedWritePair; // Vector integer and/or/xor logicals (YMM/ZMM).
+defm WriteVecLogicY : X86SchedWritePair; // Vector integer and/or/xor logicals (YMM).
+defm WriteVecLogicZ : X86SchedWritePair; // Vector integer and/or/xor logicals (ZMM).
 defm WriteVecTest  : X86SchedWritePair; // Vector integer TEST instructions.
-defm WriteVecTestY : X86SchedWritePair; // Vector integer TEST instructions (YMM/ZMM).
+defm WriteVecTestY : X86SchedWritePair; // Vector integer TEST instructions (YMM).
+defm WriteVecTestZ : X86SchedWritePair; // Vector integer TEST instructions (ZMM).
 defm WriteVecShift  : X86SchedWritePair; // Vector integer shifts (default).
 defm WriteVecShiftX : X86SchedWritePair; // Vector integer shifts (XMM).
-defm WriteVecShiftY : X86SchedWritePair; // Vector integer shifts (YMM/ZMM).
+defm WriteVecShiftY : X86SchedWritePair; // Vector integer shifts (YMM).
+defm WriteVecShiftZ : X86SchedWritePair; // Vector integer shifts (ZMM).
 defm WriteVecShiftImm : X86SchedWritePair; // Vector integer immediate shifts (default).
 defm WriteVecShiftImmX: X86SchedWritePair; // Vector integer immediate shifts (XMM).
-defm WriteVecShiftImmY: X86SchedWritePair; // Vector integer immediate shifts (YMM/ZMM).
+defm WriteVecShiftImmY: X86SchedWritePair; // Vector integer immediate shifts (YMM).
+defm WriteVecShiftImmZ: X86SchedWritePair; // Vector integer immediate shifts (ZMM).
 defm WriteVecIMul  : X86SchedWritePair; // Vector integer multiply (default).
 defm WriteVecIMulX : X86SchedWritePair; // Vector integer multiply (XMM).
-defm WriteVecIMulY : X86SchedWritePair; // Vector integer multiply (YMM/ZMM).
+defm WriteVecIMulY : X86SchedWritePair; // Vector integer multiply (YMM).
+defm WriteVecIMulZ : X86SchedWritePair; // Vector integer multiply (ZMM).
 defm WritePMULLD   : X86SchedWritePair; // Vector PMULLD.
-defm WritePMULLDY   : X86SchedWritePair; // Vector PMULLD (YMM/ZMM).
+defm WritePMULLDY   : X86SchedWritePair; // Vector PMULLD (YMM).
+defm WritePMULLDZ   : X86SchedWritePair; // Vector PMULLD (ZMM).
 defm WriteShuffle  : X86SchedWritePair; // Vector shuffles.
 defm WriteShuffleX : X86SchedWritePair; // Vector shuffles (XMM).
-defm WriteShuffleY : X86SchedWritePair; // Vector shuffles (YMM/ZMM).
+defm WriteShuffleY : X86SchedWritePair; // Vector shuffles (YMM).
+defm WriteShuffleZ : X86SchedWritePair; // Vector shuffles (ZMM).
 defm WriteVarShuffle  : X86SchedWritePair; // Vector variable shuffles.
 defm WriteVarShuffleX : X86SchedWritePair; // Vector variable shuffles (XMM).
-defm WriteVarShuffleY : X86SchedWritePair; // Vector variable shuffles (YMM/ZMM).
+defm WriteVarShuffleY : X86SchedWritePair; // Vector variable shuffles (YMM).
+defm WriteVarShuffleZ : X86SchedWritePair; // Vector variable shuffles (ZMM).
 defm WriteBlend  : X86SchedWritePair; // Vector blends.
-defm WriteBlendY : X86SchedWritePair; // Vector blends (YMM/ZMM).
+defm WriteBlendY : X86SchedWritePair; // Vector blends (YMM).
+defm WriteBlendZ : X86SchedWritePair; // Vector blends (ZMM).
 defm WriteVarBlend  : X86SchedWritePair; // Vector variable blends.
-defm WriteVarBlendY : X86SchedWritePair; // Vector variable blends (YMM/ZMM).
+defm WriteVarBlendY : X86SchedWritePair; // Vector variable blends (YMM).
+defm WriteVarBlendZ : X86SchedWritePair; // Vector variable blends (ZMM).
 defm WritePSADBW  : X86SchedWritePair; // Vector PSADBW.
 defm WritePSADBWX : X86SchedWritePair; // Vector PSADBW (XMM).
-defm WritePSADBWY : X86SchedWritePair; // Vector PSADBW (YMM/ZMM).
+defm WritePSADBWY : X86SchedWritePair; // Vector PSADBW (YMM).
+defm WritePSADBWZ : X86SchedWritePair; // Vector PSADBW (ZMM).
 defm WriteMPSAD  : X86SchedWritePair; // Vector MPSAD.
-defm WriteMPSADY : X86SchedWritePair; // Vector MPSAD (YMM/ZMM).
+defm WriteMPSADY : X86SchedWritePair; // Vector MPSAD (YMM).
+defm WriteMPSADZ : X86SchedWritePair; // Vector MPSAD (ZMM).
 defm WritePHMINPOS : X86SchedWritePair; // Vector PHMINPOS.
 
 // Vector insert/extract operations.
@@ -320,35 +351,44 @@
 // Conversion between integer and float.
 defm WriteCvtSD2I  : X86SchedWritePair; // Double -> Integer.
 defm WriteCvtPD2I  : X86SchedWritePair; // Double -> Integer (XMM).
-defm WriteCvtPD2IY : X86SchedWritePair; // Double -> Integer (YMM/ZMM).
+defm WriteCvtPD2IY : X86SchedWritePair; // Double -> Integer (YMM).
+defm WriteCvtPD2IZ : X86SchedWritePair; // Double -> Integer (ZMM).
 
 defm WriteCvtSS2I  : X86SchedWritePair; // Float -> Integer.
 defm WriteCvtPS2I  : X86SchedWritePair; // Float -> Integer (XMM).
-defm WriteCvtPS2IY : X86SchedWritePair; // Float -> Integer (YMM/ZMM).
+defm WriteCvtPS2IY : X86SchedWritePair; // Float -> Integer (YMM).
+defm WriteCvtPS2IZ : X86SchedWritePair; // Float -> Integer (ZMM).
 
 defm WriteCvtI2SD  : X86SchedWritePair; // Integer -> Double.
 defm WriteCvtI2PD  : X86SchedWritePair; // Integer -> Double (XMM).
-defm WriteCvtI2PDY : X86SchedWritePair; // Integer -> Double (YMM/ZMM).
+defm WriteCvtI2PDY : X86SchedWritePair; // Integer -> Double (YMM).
+defm WriteCvtI2PDZ : X86SchedWritePair; // Integer -> Double (ZMM).
 
 defm WriteCvtI2SS  : X86SchedWritePair; // Integer -> Float.
 defm WriteCvtI2PS  : X86SchedWritePair; // Integer -> Float (XMM).
-defm WriteCvtI2PSY : X86SchedWritePair; // Integer -> Float (YMM/ZMM).
+defm WriteCvtI2PSY : X86SchedWritePair; // Integer -> Float (YMM).
+defm WriteCvtI2PSZ : X86SchedWritePair; // Integer -> Float (ZMM).
 
 defm WriteCvtSS2SD  : X86SchedWritePair; // Float -> Double size conversion.
 defm WriteCvtPS2PD  : X86SchedWritePair; // Float -> Double size conversion (XMM).
-defm WriteCvtPS2PDY : X86SchedWritePair; // Float -> Double size conversion (YMM/ZMM).
+defm WriteCvtPS2PDY : X86SchedWritePair; // Float -> Double size conversion (YMM).
+defm WriteCvtPS2PDZ : X86SchedWritePair; // Float -> Double size conversion (ZMM).
 
 defm WriteCvtSD2SS  : X86SchedWritePair; // Double -> Float size conversion.
 defm WriteCvtPD2PS  : X86SchedWritePair; // Double -> Float size conversion (XMM).
-defm WriteCvtPD2PSY : X86SchedWritePair; // Double -> Float size conversion (YMM/ZMM).
+defm WriteCvtPD2PSY : X86SchedWritePair; // Double -> Float size conversion (YMM).
+defm WriteCvtPD2PSZ : X86SchedWritePair; // Double -> Float size conversion (ZMM).
 
 defm WriteCvtPH2PS    : X86SchedWritePair; // Half -> Float size conversion.
-defm WriteCvtPH2PSY   : X86SchedWritePair; // Half -> Float size conversion (YMM/ZMM).
+defm WriteCvtPH2PSY   : X86SchedWritePair; // Half -> Float size conversion (YMM).
+defm WriteCvtPH2PSZ   : X86SchedWritePair; // Half -> Float size conversion (ZMM).
 
 def  WriteCvtPS2PH    : SchedWrite; // // Float -> Half size conversion.
-def  WriteCvtPS2PHY   : SchedWrite; // // Float -> Half size conversion (YMM/ZMM).
+def  WriteCvtPS2PHY   : SchedWrite; // // Float -> Half size conversion (YMM).
+def  WriteCvtPS2PHZ   : SchedWrite; // // Float -> Half size conversion (ZMM).
 def  WriteCvtPS2PHSt  : SchedWrite; // // Float -> Half + store size conversion.
-def  WriteCvtPS2PHYSt : SchedWrite; // // Float -> Half + store size conversion (YMM/ZMM).
+def  WriteCvtPS2PHYSt : SchedWrite; // // Float -> Half + store size conversion (YMM).
+def  WriteCvtPS2PHZSt : SchedWrite; // // Float -> Half + store size conversion (ZMM).
 
 // CRC32 instruction.
 defm WriteCRC32 : X86SchedWritePair;
@@ -387,7 +427,8 @@
 defm WriteShuffle256 : X86SchedWritePair; // 256-bit width vector shuffles.
 defm WriteVarShuffle256 : X86SchedWritePair; // 256-bit width vector variable shuffles.
 defm WriteVarVecShift  : X86SchedWritePair; // Variable vector shifts.
-defm WriteVarVecShiftY : X86SchedWritePair; // Variable vector shifts (YMM/ZMM).
+defm WriteVarVecShiftY : X86SchedWritePair; // Variable vector shifts (YMM).
+defm WriteVarVecShiftZ : X86SchedWritePair; // Variable vector shifts (ZMM).
 
 // Old microcoded instructions that nobody use.
 def WriteMicrocoded : SchedWrite;
@@ -441,25 +482,25 @@
 
 // Vector width wrappers.
 def SchedWriteFAdd
- : X86SchedWriteWidths<WriteFAdd, WriteFAddX, WriteFAddY, WriteFAddY>;
+ : X86SchedWriteWidths<WriteFAdd, WriteFAddX, WriteFAddY, WriteFAddZ>;
 def SchedWriteFAdd64
- : X86SchedWriteWidths<WriteFAdd64, WriteFAdd64X, WriteFAdd64Y, WriteFAdd64Y>;
+ : X86SchedWriteWidths<WriteFAdd64, WriteFAdd64X, WriteFAdd64Y, WriteFAdd64Z>;
 def SchedWriteFHAdd
- : X86SchedWriteWidths<WriteFHAdd, WriteFHAdd, WriteFHAddY, WriteFHAddY>;
+ : X86SchedWriteWidths<WriteFHAdd, WriteFHAdd, WriteFHAddY, WriteFHAddZ>;
 def SchedWriteFCmp
- : X86SchedWriteWidths<WriteFCmp, WriteFCmpX, WriteFCmpY, WriteFCmpY>;
+ : X86SchedWriteWidths<WriteFCmp, WriteFCmpX, WriteFCmpY, WriteFCmpZ>;
 def SchedWriteFCmp64
- : X86SchedWriteWidths<WriteFCmp64, WriteFCmp64X, WriteFCmp64Y, WriteFCmp64Y>;
+ : X86SchedWriteWidths<WriteFCmp64, WriteFCmp64X, WriteFCmp64Y, WriteFCmp64Z>;
 def SchedWriteFMul
- : X86SchedWriteWidths<WriteFMul, WriteFMulX, WriteFMulY, WriteFMulY>;
+ : X86SchedWriteWidths<WriteFMul, WriteFMulX, WriteFMulY, WriteFMulZ>;
 def SchedWriteFMul64
- : X86SchedWriteWidths<WriteFMul64, WriteFMul64X, WriteFMul64Y, WriteFMul64Y>;
+ : X86SchedWriteWidths<WriteFMul64, WriteFMul64X, WriteFMul64Y, WriteFMul64Z>;
 def SchedWriteFMA
- : X86SchedWriteWidths<WriteFMA, WriteFMAX, WriteFMAY, WriteFMAY>;
+ : X86SchedWriteWidths<WriteFMA, WriteFMAX, WriteFMAY, WriteFMAZ>;
 def SchedWriteDPPD
  : X86SchedWriteWidths<WriteDPPD, WriteDPPD, WriteDPPD, WriteDPPD>;
 def SchedWriteDPPS
- : X86SchedWriteWidths<WriteDPPS, WriteDPPS, WriteDPPSY, WriteDPPSY>;
+ : X86SchedWriteWidths<WriteDPPS, WriteDPPS, WriteDPPSY, WriteDPPSZ>;
 def SchedWriteFDiv
  : X86SchedWriteWidths<WriteFDiv, WriteFDivX, WriteFDivY, WriteFDivZ>;
 def SchedWriteFDiv64
@@ -471,90 +512,90 @@
  : X86SchedWriteWidths<WriteFSqrt64, WriteFSqrt64X,
                        WriteFSqrt64Y, WriteFSqrt64Z>;
 def SchedWriteFRcp
- : X86SchedWriteWidths<WriteFRcp, WriteFRcpX, WriteFRcpY, WriteFRcpY>;
+ : X86SchedWriteWidths<WriteFRcp, WriteFRcpX, WriteFRcpY, WriteFRcpZ>;
 def SchedWriteFRsqrt
- : X86SchedWriteWidths<WriteFRsqrt, WriteFRsqrtX, WriteFRsqrtY, WriteFRsqrtY>;
+ : X86SchedWriteWidths<WriteFRsqrt, WriteFRsqrtX, WriteFRsqrtY, WriteFRsqrtZ>;
 def SchedWriteFRnd
- : X86SchedWriteWidths<WriteFRnd, WriteFRnd, WriteFRndY, WriteFRndY>;
+ : X86SchedWriteWidths<WriteFRnd, WriteFRnd, WriteFRndY, WriteFRndZ>;
 def SchedWriteFLogic
- : X86SchedWriteWidths<WriteFLogic, WriteFLogic, WriteFLogicY, WriteFLogicY>;
+ : X86SchedWriteWidths<WriteFLogic, WriteFLogic, WriteFLogicY, WriteFLogicZ>;
 def SchedWriteFTest
- : X86SchedWriteWidths<WriteFTest, WriteFTest, WriteFTestY, WriteFTestY>;
+ : X86SchedWriteWidths<WriteFTest, WriteFTest, WriteFTestY, WriteFTestZ>;
 
 def SchedWriteFShuffle
  : X86SchedWriteWidths<WriteFShuffle, WriteFShuffle,
-                       WriteFShuffleY, WriteFShuffleY>;
+                       WriteFShuffleY, WriteFShuffleZ>;
 def SchedWriteFVarShuffle
  : X86SchedWriteWidths<WriteFVarShuffle, WriteFVarShuffle,
-                       WriteFVarShuffleY, WriteFVarShuffleY>;
+                       WriteFVarShuffleY, WriteFVarShuffleZ>;
 def SchedWriteFBlend
- : X86SchedWriteWidths<WriteFBlend, WriteFBlend, WriteFBlendY, WriteFBlendY>;
+ : X86SchedWriteWidths<WriteFBlend, WriteFBlend, WriteFBlendY, WriteFBlendZ>;
 def SchedWriteFVarBlend
  : X86SchedWriteWidths<WriteFVarBlend, WriteFVarBlend,
-                       WriteFVarBlendY, WriteFVarBlendY>;
+                       WriteFVarBlendY, WriteFVarBlendZ>;
 
 def SchedWriteCvtDQ2PD
  : X86SchedWriteWidths<WriteCvtI2SD, WriteCvtI2PD,
-                       WriteCvtI2PDY, WriteCvtI2PDY>;
+                       WriteCvtI2PDY, WriteCvtI2PDZ>;
 def SchedWriteCvtDQ2PS
  : X86SchedWriteWidths<WriteCvtI2SS, WriteCvtI2PS,
-                       WriteCvtI2PSY, WriteCvtI2PSY>;
+                       WriteCvtI2PSY, WriteCvtI2PSZ>;
 def SchedWriteCvtPD2DQ
  : X86SchedWriteWidths<WriteCvtSD2I, WriteCvtPD2I,
-                       WriteCvtPD2IY, WriteCvtPD2IY>;
+                       WriteCvtPD2IY, WriteCvtPD2IZ>;
 def SchedWriteCvtPS2DQ
  : X86SchedWriteWidths<WriteCvtSS2I, WriteCvtPS2I,
-                       WriteCvtPS2IY, WriteCvtPS2IY>;
+                       WriteCvtPS2IY, WriteCvtPS2IZ>;
 def SchedWriteCvtPS2PD
  : X86SchedWriteWidths<WriteCvtSS2SD, WriteCvtPS2PD,
-                       WriteCvtPS2PDY, WriteCvtPS2PDY>;
+                       WriteCvtPS2PDY, WriteCvtPS2PDZ>;
 def SchedWriteCvtPD2PS
  : X86SchedWriteWidths<WriteCvtSD2SS, WriteCvtPD2PS,
-                       WriteCvtPD2PSY, WriteCvtPD2PSY>;
+                       WriteCvtPD2PSY, WriteCvtPD2PSZ>;
 
 def SchedWriteVecALU
- : X86SchedWriteWidths<WriteVecALU, WriteVecALUX, WriteVecALUY, WriteVecALUY>;
+ : X86SchedWriteWidths<WriteVecALU, WriteVecALUX, WriteVecALUY, WriteVecALUZ>;
 def SchedWritePHAdd
- : X86SchedWriteWidths<WritePHAdd, WritePHAddX, WritePHAddY, WritePHAddY>;
+ : X86SchedWriteWidths<WritePHAdd, WritePHAddX, WritePHAddY, WritePHAddZ>;
 def SchedWriteVecLogic
  : X86SchedWriteWidths<WriteVecLogic, WriteVecLogicX,
-                       WriteVecLogicY, WriteVecLogicY>;
+                       WriteVecLogicY, WriteVecLogicZ>;
 def SchedWriteVecTest
  : X86SchedWriteWidths<WriteVecTest, WriteVecTest,
-                       WriteVecTestY, WriteVecTestY>;
+                       WriteVecTestY, WriteVecTestZ>;
 def SchedWriteVecShift
  : X86SchedWriteWidths<WriteVecShift, WriteVecShiftX,
-                       WriteVecShiftY, WriteVecShiftY>;
+                       WriteVecShiftY, WriteVecShiftZ>;
 def SchedWriteVecShiftImm
  : X86SchedWriteWidths<WriteVecShiftImm, WriteVecShiftImmX,
-                       WriteVecShiftImmY, WriteVecShiftImmY>;
+                       WriteVecShiftImmY, WriteVecShiftImmZ>;
 def SchedWriteVarVecShift
  : X86SchedWriteWidths<WriteVarVecShift, WriteVarVecShift,
-                       WriteVarVecShiftY, WriteVarVecShiftY>;
+                       WriteVarVecShiftY, WriteVarVecShiftZ>;
 def SchedWriteVecIMul
  : X86SchedWriteWidths<WriteVecIMul, WriteVecIMulX,
-                       WriteVecIMulY, WriteVecIMulY>;
+                       WriteVecIMulY, WriteVecIMulZ>;
 def SchedWritePMULLD
  : X86SchedWriteWidths<WritePMULLD, WritePMULLD,
-                       WritePMULLDY, WritePMULLDY>;
+                       WritePMULLDY, WritePMULLDZ>;
 def SchedWriteMPSAD
  : X86SchedWriteWidths<WriteMPSAD, WriteMPSAD,
-                       WriteMPSADY, WriteMPSADY>;
+                       WriteMPSADY, WriteMPSADZ>;
 def SchedWritePSADBW
  : X86SchedWriteWidths<WritePSADBW, WritePSADBWX,
-                       WritePSADBWY, WritePSADBWY>;
+                       WritePSADBWY, WritePSADBWZ>;
 
 def SchedWriteShuffle
  : X86SchedWriteWidths<WriteShuffle, WriteShuffleX,
-                       WriteShuffleY, WriteShuffleY>;
+                       WriteShuffleY, WriteShuffleZ>;
 def SchedWriteVarShuffle
  : X86SchedWriteWidths<WriteVarShuffle, WriteVarShuffleX,
-                       WriteVarShuffleY, WriteVarShuffleY>;
+                       WriteVarShuffleY, WriteVarShuffleZ>;
 def SchedWriteBlend
- : X86SchedWriteWidths<WriteBlend, WriteBlend, WriteBlendY, WriteBlendY>;
+ : X86SchedWriteWidths<WriteBlend, WriteBlend, WriteBlendY, WriteBlendZ>;
 def SchedWriteVarBlend
  : X86SchedWriteWidths<WriteVarBlend, WriteVarBlend,
-                       WriteVarBlendY, WriteVarBlendY>;
+                       WriteVarBlendY, WriteVarBlendZ>;
 
 // Vector size wrappers.
 def SchedWriteFAddSizes
Index: lib/Target/X86/X86ScheduleAtom.td
===================================================================
--- lib/Target/X86/X86ScheduleAtom.td
+++ lib/Target/X86/X86ScheduleAtom.td
@@ -210,28 +210,36 @@
 defm : AtomWriteResPair<WriteFAdd,           [AtomPort0],  [AtomPort0],  5,  5,  [5],  [5]>;
 defm : AtomWriteResPair<WriteFAddX,          [AtomPort0],  [AtomPort0],  5,  5,  [5],  [5]>;
 defm : AtomWriteResPair<WriteFAddY,          [AtomPort0],  [AtomPort0],  5,  5,  [5],  [5]>;
+defm : X86WriteResPairUnsupported<WriteFAddZ>;
 defm : AtomWriteResPair<WriteFAdd64,         [AtomPort0],  [AtomPort0],  5,  5,  [5],  [5]>;
 defm : AtomWriteResPair<WriteFAdd64X,       [AtomPort01], [AtomPort01],  6,  7,  [6],  [7]>;
 defm : AtomWriteResPair<WriteFAdd64Y,       [AtomPort01], [AtomPort01],  6,  7,  [6],  [7]>;
+defm : X86WriteResPairUnsupported<WriteFAdd64Z>;
 defm : AtomWriteResPair<WriteFCmp,           [AtomPort0],  [AtomPort0],  5,  5,  [5],  [5]>;
 defm : AtomWriteResPair<WriteFCmpX,          [AtomPort0],  [AtomPort0],  5,  5,  [5],  [5]>;
 defm : AtomWriteResPair<WriteFCmpY,          [AtomPort0],  [AtomPort0],  5,  5,  [5],  [5]>;
+defm : X86WriteResPairUnsupported<WriteFCmpZ>;
 defm : AtomWriteResPair<WriteFCmp64,         [AtomPort0],  [AtomPort0],  5,  5,  [5],  [5]>;
 defm : AtomWriteResPair<WriteFCmp64X,       [AtomPort01], [AtomPort01],  6,  7,  [6],  [7]>;
 defm : AtomWriteResPair<WriteFCmp64Y,       [AtomPort01], [AtomPort01],  6,  7,  [6],  [7]>;
+defm : X86WriteResPairUnsupported<WriteFCmp64Z>;
 defm : AtomWriteResPair<WriteFCom,           [AtomPort0],  [AtomPort0],  5,  5,  [5],  [5]>;
 defm : AtomWriteResPair<WriteFMul,           [AtomPort0],  [AtomPort0],  4,  4,  [4],  [4]>;
 defm : AtomWriteResPair<WriteFMulX,          [AtomPort0],  [AtomPort0],  5,  5,  [5],  [5]>;
 defm : AtomWriteResPair<WriteFMulY,          [AtomPort0],  [AtomPort0],  4,  4,  [4],  [4]>;
+defm : X86WriteResPairUnsupported<WriteFMulZ>;
 defm : AtomWriteResPair<WriteFMul64,         [AtomPort0],  [AtomPort0],  5,  5,  [5],  [5]>;
 defm : AtomWriteResPair<WriteFMul64X,       [AtomPort01], [AtomPort01],  9, 10,  [9], [10]>;
 defm : AtomWriteResPair<WriteFMul64Y,       [AtomPort01], [AtomPort01],  9, 10,  [9], [10]>;
+defm : X86WriteResPairUnsupported<WriteFMul64Z>;
 defm : AtomWriteResPair<WriteFRcp,           [AtomPort0],  [AtomPort0],  4,  4,  [4],  [4]>;
 defm : AtomWriteResPair<WriteFRcpX,         [AtomPort01], [AtomPort01],  9, 10,  [9], [10]>;
 defm : AtomWriteResPair<WriteFRcpY,         [AtomPort01], [AtomPort01],  9, 10,  [9], [10]>;
+defm : X86WriteResPairUnsupported<WriteFRcpZ>;
 defm : AtomWriteResPair<WriteFRsqrt,         [AtomPort0],  [AtomPort0],  4,  4,  [4],  [4]>;
 defm : AtomWriteResPair<WriteFRsqrtX,       [AtomPort01], [AtomPort01],  9, 10,  [9], [10]>;
 defm : AtomWriteResPair<WriteFRsqrtY,       [AtomPort01], [AtomPort01],  9, 10,  [9], [10]>;
+defm : X86WriteResPairUnsupported<WriteFRsqrtZ>;
 defm : AtomWriteResPair<WriteFDiv,          [AtomPort01], [AtomPort01], 34, 34, [34], [34]>;
 defm : AtomWriteResPair<WriteFDivX,         [AtomPort01], [AtomPort01], 70, 70, [70], [70]>;
 defm : AtomWriteResPair<WriteFDivY,         [AtomPort01], [AtomPort01], 70, 70, [70], [70]>;
@@ -252,24 +260,33 @@
 defm : AtomWriteResPair<WriteFSign,          [AtomPort1],  [AtomPort1]>;
 defm : AtomWriteResPair<WriteFRnd,           [AtomPort0],  [AtomPort0],  5,  5,  [5],  [5]>;
 defm : AtomWriteResPair<WriteFRndY,          [AtomPort0],  [AtomPort0],  5,  5,  [5],  [5]>;
+defm : X86WriteResPairUnsupported<WriteFRndZ>;
 defm : AtomWriteResPair<WriteFLogic,        [AtomPort01],  [AtomPort0]>;
 defm : X86WriteResPairUnsupported<WriteFLogicY>;
+defm : X86WriteResPairUnsupported<WriteFLogicZ>;
 defm : AtomWriteResPair<WriteFTest,         [AtomPort01],  [AtomPort0]>;
 defm : X86WriteResPairUnsupported<WriteFTestY>;
+defm : X86WriteResPairUnsupported<WriteFTestZ>;
 defm : AtomWriteResPair<WriteFShuffle,       [AtomPort0],  [AtomPort0]>;
 defm : X86WriteResPairUnsupported<WriteFShuffleY>;
+defm : X86WriteResPairUnsupported<WriteFShuffleZ>;
 defm : X86WriteResPairUnsupported<WriteFVarShuffle>;
 defm : X86WriteResPairUnsupported<WriteFVarShuffleY>;
+defm : X86WriteResPairUnsupported<WriteFVarShuffleZ>;
 defm : X86WriteResPairUnsupported<WriteFMA>;
 defm : X86WriteResPairUnsupported<WriteFMAX>;
 defm : X86WriteResPairUnsupported<WriteFMAY>;
+defm : X86WriteResPairUnsupported<WriteFMAZ>;
 defm : X86WriteResPairUnsupported<WriteDPPD>;
 defm : X86WriteResPairUnsupported<WriteDPPS>;
 defm : X86WriteResPairUnsupported<WriteDPPSY>;
+defm : X86WriteResPairUnsupported<WriteDPPSZ>;
 defm : X86WriteResPairUnsupported<WriteFBlend>;
 defm : X86WriteResPairUnsupported<WriteFBlendY>;
+defm : X86WriteResPairUnsupported<WriteFBlendZ>;
 defm : X86WriteResPairUnsupported<WriteFVarBlend>;
 defm : X86WriteResPairUnsupported<WriteFVarBlendY>;
+defm : X86WriteResPairUnsupported<WriteFVarBlendZ>;
 defm : X86WriteResPairUnsupported<WriteFShuffle256>;
 defm : X86WriteResPairUnsupported<WriteFVarShuffle256>;
 
@@ -280,30 +297,39 @@
 defm : AtomWriteResPair<WriteCvtSS2I,   [AtomPort01], [AtomPort01], 8, 9, [8], [9]>;
 defm : AtomWriteResPair<WriteCvtPS2I,   [AtomPort01], [AtomPort01], 6, 7, [6], [7]>;
 defm : AtomWriteResPair<WriteCvtPS2IY,  [AtomPort01], [AtomPort01], 6, 7, [6], [7]>;
+defm : X86WriteResPairUnsupported<WriteCvtPS2IZ>;
 defm : AtomWriteResPair<WriteCvtSD2I,   [AtomPort01], [AtomPort01], 8, 9, [8], [9]>;
 defm : AtomWriteResPair<WriteCvtPD2I,   [AtomPort01], [AtomPort01], 7, 8, [7], [8]>;
 defm : AtomWriteResPair<WriteCvtPD2IY,  [AtomPort01], [AtomPort01], 7, 8, [7], [8]>;
+defm : X86WriteResPairUnsupported<WriteCvtPD2IZ>;
 
 defm : AtomWriteResPair<WriteCvtI2SS,   [AtomPort01], [AtomPort01], 6, 7, [6], [7]>;
 defm : AtomWriteResPair<WriteCvtI2PS,   [AtomPort01], [AtomPort01], 6, 7, [6], [7]>;
 defm : AtomWriteResPair<WriteCvtI2PSY,  [AtomPort01], [AtomPort01], 6, 7, [6], [7]>;
+defm : X86WriteResPairUnsupported<WriteCvtI2PSZ>;
 defm : AtomWriteResPair<WriteCvtI2SD,   [AtomPort01], [AtomPort01], 6, 7, [6], [7]>;
 defm : AtomWriteResPair<WriteCvtI2PD,   [AtomPort01], [AtomPort01], 7, 8, [7], [8]>;
 defm : AtomWriteResPair<WriteCvtI2PDY,  [AtomPort01], [AtomPort01], 7, 8, [7], [8]>;
+defm : X86WriteResPairUnsupported<WriteCvtI2PDZ>;
 
 defm : AtomWriteResPair<WriteCvtSS2SD,  [AtomPort01], [AtomPort01], 6, 7, [6], [7]>;
 defm : AtomWriteResPair<WriteCvtPS2PD,  [AtomPort01], [AtomPort01], 7, 8, [7], [8]>;
 defm : AtomWriteResPair<WriteCvtPS2PDY, [AtomPort01], [AtomPort01], 7, 8, [7], [8]>;
+defm : X86WriteResPairUnsupported<WriteCvtPS2PDZ>;
 defm : AtomWriteResPair<WriteCvtSD2SS,  [AtomPort01], [AtomPort01], 6, 7, [6], [7]>;
 defm : AtomWriteResPair<WriteCvtPD2PS,  [AtomPort01], [AtomPort01], 7, 8, [7], [8]>;
 defm : AtomWriteResPair<WriteCvtPD2PSY, [AtomPort01], [AtomPort01], 7, 8, [7], [8]>;
+defm : X86WriteResPairUnsupported<WriteCvtPD2PSZ>;
 
 defm : X86WriteResPairUnsupported<WriteCvtPH2PS>;
 defm : X86WriteResPairUnsupported<WriteCvtPH2PSY>;
+defm : X86WriteResPairUnsupported<WriteCvtPH2PSZ>;
 defm : X86WriteResUnsupported<WriteCvtPS2PH>;
 defm : X86WriteResUnsupported<WriteCvtPS2PHSt>;
 defm : X86WriteResUnsupported<WriteCvtPS2PHY>;
+defm : X86WriteResUnsupported<WriteCvtPS2PHZ>;
 defm : X86WriteResUnsupported<WriteCvtPS2PHYSt>;
+defm : X86WriteResUnsupported<WriteCvtPS2PHZSt>;
 
 ////////////////////////////////////////////////////////////////////////////////
 // Vector integer operations.
@@ -334,42 +360,56 @@
 defm : AtomWriteResPair<WriteVecALU,       [AtomPort01],  [AtomPort0], 1, 1>;
 defm : AtomWriteResPair<WriteVecALUX,      [AtomPort01],  [AtomPort0], 1, 1>;
 defm : AtomWriteResPair<WriteVecALUY,      [AtomPort01],  [AtomPort0], 1, 1>;
+defm : X86WriteResPairUnsupported<WriteVecALUZ>;
 defm : AtomWriteResPair<WriteVecLogic,     [AtomPort01],  [AtomPort0], 1, 1>;
 defm : AtomWriteResPair<WriteVecLogicX,    [AtomPort01],  [AtomPort0], 1, 1>;
 defm : AtomWriteResPair<WriteVecLogicY,    [AtomPort01],  [AtomPort0], 1, 1>;
+defm : X86WriteResPairUnsupported<WriteVecLogicZ>;
 defm : AtomWriteResPair<WriteVecTest,      [AtomPort01],  [AtomPort0], 1, 1>;
 defm : AtomWriteResPair<WriteVecTestY,     [AtomPort01],  [AtomPort0], 1, 1>;
+defm : X86WriteResPairUnsupported<WriteVecTestZ>;
 defm : AtomWriteResPair<WriteVecShift,     [AtomPort01], [AtomPort01], 2, 3, [2], [3]>;
 defm : AtomWriteResPair<WriteVecShiftX,    [AtomPort01], [AtomPort01], 2, 3, [2], [3]>;
 defm : AtomWriteResPair<WriteVecShiftY,    [AtomPort01], [AtomPort01], 2, 3, [2], [3]>;
+defm : X86WriteResPairUnsupported<WriteVecShiftZ>;
 defm : AtomWriteResPair<WriteVecShiftImm,  [AtomPort01], [AtomPort01], 1, 1, [1], [1]>;
 defm : AtomWriteResPair<WriteVecShiftImmX, [AtomPort01], [AtomPort01], 1, 1, [1], [1]>;
 defm : AtomWriteResPair<WriteVecShiftImmY, [AtomPort01], [AtomPort01], 1, 1, [1], [1]>;
+defm : X86WriteResPairUnsupported<WriteVecShiftImmZ>;
 defm : AtomWriteResPair<WriteVecIMul,       [AtomPort0],  [AtomPort0], 4, 4, [4], [4]>;
 defm : AtomWriteResPair<WriteVecIMulX,      [AtomPort0],  [AtomPort0], 5, 5, [5], [5]>;
 defm : AtomWriteResPair<WriteVecIMulY,      [AtomPort0],  [AtomPort0], 5, 5, [5], [5]>;
+defm : X86WriteResPairUnsupported<WriteVecIMulZ>;
 defm : AtomWriteResPair<WritePMULLD,       [AtomPort01],  [AtomPort0], 1, 1>;
 defm : AtomWriteResPair<WritePMULLDY,      [AtomPort01],  [AtomPort0], 1, 1>;
+defm : X86WriteResPairUnsupported<WritePMULLDZ>;
 defm : AtomWriteResPair<WritePHMINPOS,      [AtomPort0],  [AtomPort0], 5, 5, [5], [5]>;
 defm : AtomWriteResPair<WriteMPSAD,        [AtomPort01],  [AtomPort0], 1, 1>;
 defm : AtomWriteResPair<WriteMPSADY,       [AtomPort01],  [AtomPort0], 1, 1>;
+defm : X86WriteResPairUnsupported<WriteMPSADZ>;
 defm : AtomWriteResPair<WritePSADBW,       [AtomPort01], [AtomPort01], 4, 4, [4], [4]>;
 defm : AtomWriteResPair<WritePSADBWX,       [AtomPort0],  [AtomPort0], 5, 5, [5], [5]>;
 defm : AtomWriteResPair<WritePSADBWY,       [AtomPort0],  [AtomPort0], 5, 5, [5], [5]>;
+defm : X86WriteResPairUnsupported<WritePSADBWZ>;
 defm : AtomWriteResPair<WriteShuffle,       [AtomPort0],  [AtomPort0], 1, 1>;
 defm : AtomWriteResPair<WriteShuffleX,      [AtomPort0],  [AtomPort0], 1, 1>;
 defm : AtomWriteResPair<WriteShuffleY,      [AtomPort0],  [AtomPort0], 1, 1>;
+defm : X86WriteResPairUnsupported<WriteShuffleZ>;
 defm : AtomWriteResPair<WriteVarShuffle,    [AtomPort0],  [AtomPort0], 1, 1>;
 defm : AtomWriteResPair<WriteVarShuffleX,  [AtomPort01], [AtomPort01], 4, 5, [4], [5]>;
 defm : AtomWriteResPair<WriteVarShuffleY,  [AtomPort01], [AtomPort01], 4, 5, [4], [5]>;
+defm : X86WriteResPairUnsupported<WriteVarShuffleZ>;
 defm : X86WriteResPairUnsupported<WriteBlend>;
 defm : X86WriteResPairUnsupported<WriteBlendY>;
+defm : X86WriteResPairUnsupported<WriteBlendZ>;
 defm : X86WriteResPairUnsupported<WriteVarBlend>;
 defm : X86WriteResPairUnsupported<WriteVarBlendY>;
+defm : X86WriteResPairUnsupported<WriteVarBlendZ>;
 defm : X86WriteResPairUnsupported<WriteShuffle256>;
 defm : X86WriteResPairUnsupported<WriteVarShuffle256>;
 defm : X86WriteResPairUnsupported<WriteVarVecShift>;
 defm : X86WriteResPairUnsupported<WriteVarVecShiftY>;
+defm : X86WriteResPairUnsupported<WriteVarVecShiftZ>;
 
 ////////////////////////////////////////////////////////////////////////////////
 // Vector insert/extract operations.
Index: lib/Target/X86/X86ScheduleBtVer2.td
===================================================================
--- lib/Target/X86/X86ScheduleBtVer2.td
+++ lib/Target/X86/X86ScheduleBtVer2.td
@@ -298,34 +298,44 @@
 defm : JWriteResFpuPair<WriteFAdd,         [JFPU0, JFPA],  3>;
 defm : JWriteResFpuPair<WriteFAddX,        [JFPU0, JFPA],  3>;
 defm : JWriteResYMMPair<WriteFAddY,        [JFPU0, JFPA],  3, [2,2], 2>;
+defm : X86WriteResPairUnsupported<WriteFAddZ>;
 defm : JWriteResFpuPair<WriteFAdd64,       [JFPU0, JFPA],  3>;
 defm : JWriteResFpuPair<WriteFAdd64X,      [JFPU0, JFPA],  3>;
 defm : JWriteResYMMPair<WriteFAdd64Y,      [JFPU0, JFPA],  3, [2,2], 2>;
+defm : X86WriteResPairUnsupported<WriteFAdd64Z>;
 defm : JWriteResFpuPair<WriteFCmp,         [JFPU0, JFPA],  2>;
 defm : JWriteResFpuPair<WriteFCmpX,        [JFPU0, JFPA],  2>;
 defm : JWriteResYMMPair<WriteFCmpY,        [JFPU0, JFPA],  2, [2,2], 2>;
+defm : X86WriteResPairUnsupported<WriteFCmpZ>;
 defm : JWriteResFpuPair<WriteFCmp64,       [JFPU0, JFPA],  2>;
 defm : JWriteResFpuPair<WriteFCmp64X,      [JFPU0, JFPA],  2>;
 defm : JWriteResYMMPair<WriteFCmp64Y,      [JFPU0, JFPA],  2, [2,2], 2>;
+defm : X86WriteResPairUnsupported<WriteFCmp64Z>;
 defm : JWriteResFpuPair<WriteFCom,  [JFPU0, JFPA, JALU0],  3>;
 defm : JWriteResFpuPair<WriteFMul,         [JFPU1, JFPM],  2>;
 defm : JWriteResFpuPair<WriteFMulX,        [JFPU1, JFPM],  2>;
 defm : JWriteResYMMPair<WriteFMulY,        [JFPU1, JFPM],  2, [2,2], 2>;
+defm : X86WriteResPairUnsupported<WriteFMulZ>;
 defm : JWriteResFpuPair<WriteFMul64,       [JFPU1, JFPM],  4, [1,2]>;
 defm : JWriteResFpuPair<WriteFMul64X,      [JFPU1, JFPM],  4, [1,2]>;
 defm : JWriteResYMMPair<WriteFMul64Y,      [JFPU1, JFPM],  4, [2,4], 2>;
+defm : X86WriteResPairUnsupported<WriteFMul64Z>;
 defm : X86WriteResPairUnsupported<WriteFMA>;
 defm : X86WriteResPairUnsupported<WriteFMAX>;
 defm : X86WriteResPairUnsupported<WriteFMAY>;
+defm : X86WriteResPairUnsupported<WriteFMAZ>;
 defm : JWriteResFpuPair<WriteDPPD,   [JFPU1, JFPM, JFPA],  9, [1, 3, 3],  3>;
 defm : JWriteResFpuPair<WriteDPPS,   [JFPU1, JFPM, JFPA], 11, [1, 3, 3],  5>;
 defm : JWriteResYMMPair<WriteDPPSY,  [JFPU1, JFPM, JFPA], 12, [2, 6, 6], 10>;
+defm : X86WriteResPairUnsupported<WriteDPPSZ>;
 defm : JWriteResFpuPair<WriteFRcp,         [JFPU1, JFPM],  2>;
 defm : JWriteResFpuPair<WriteFRcpX,        [JFPU1, JFPM],  2>;
 defm : JWriteResYMMPair<WriteFRcpY,        [JFPU1, JFPM],  2, [2,2], 2>;
+defm : X86WriteResPairUnsupported<WriteFRcpZ>;
 defm : JWriteResFpuPair<WriteFRsqrt,       [JFPU1, JFPM],  2>;
 defm : JWriteResFpuPair<WriteFRsqrtX,      [JFPU1, JFPM],  2>;
 defm : JWriteResYMMPair<WriteFRsqrtY,      [JFPU1, JFPM],  2, [2,2], 2>;
+defm : X86WriteResPairUnsupported<WriteFRsqrtZ>;
 defm : JWriteResFpuPair<WriteFDiv,         [JFPU1, JFPM], 19, [1, 19]>;
 defm : JWriteResFpuPair<WriteFDivX,        [JFPU1, JFPM], 19, [1, 19]>;
 defm : JWriteResYMMPair<WriteFDivY,        [JFPU1, JFPM], 38, [2, 38], 2>;
@@ -346,18 +356,25 @@
 defm : JWriteResFpuPair<WriteFSign,        [JFPU1, JFPM],  2>;
 defm : JWriteResFpuPair<WriteFRnd,         [JFPU1, JSTC],  3>;
 defm : JWriteResYMMPair<WriteFRndY,        [JFPU1, JSTC],  3, [2,2], 2>;
+defm : X86WriteResPairUnsupported<WriteFRndZ>;
 defm : JWriteResFpuPair<WriteFLogic,      [JFPU01, JFPX],  1>;
 defm : JWriteResYMMPair<WriteFLogicY,     [JFPU01, JFPX],  1, [2, 2], 2>;
+defm : X86WriteResPairUnsupported<WriteFLogicZ>;
 defm : JWriteResFpuPair<WriteFTest,       [JFPU0, JFPA, JALU0], 3>;
 defm : JWriteResYMMPair<WriteFTestY ,     [JFPU01, JFPX, JFPA, JALU0], 4, [2, 2, 2, 1], 3>;
+defm : X86WriteResPairUnsupported<WriteFTestZ>;
 defm : JWriteResFpuPair<WriteFShuffle,    [JFPU01, JFPX],  1>;
 defm : JWriteResYMMPair<WriteFShuffleY,   [JFPU01, JFPX],  1, [2, 2], 2>;
+defm : X86WriteResPairUnsupported<WriteFShuffleZ>;
 defm : JWriteResFpuPair<WriteFVarShuffle, [JFPU01, JFPX],  2, [1, 4], 3>;
 defm : JWriteResYMMPair<WriteFVarShuffleY,[JFPU01, JFPX],  3, [2, 6], 6>;
+defm : X86WriteResPairUnsupported<WriteFVarShuffleZ>;
 defm : JWriteResFpuPair<WriteFBlend,      [JFPU01, JFPX],  1>;
 defm : JWriteResYMMPair<WriteFBlendY,     [JFPU01, JFPX],  1, [2, 2], 2>;
+defm : X86WriteResPairUnsupported<WriteFBlendZ>;
 defm : JWriteResFpuPair<WriteFVarBlend,   [JFPU01, JFPX],  2, [1, 4], 3>;
 defm : JWriteResYMMPair<WriteFVarBlendY,  [JFPU01, JFPX],  3, [2, 6], 6>;
+defm : X86WriteResPairUnsupported<WriteFVarBlendZ>;
 defm : JWriteResFpuPair<WriteFShuffle256, [JFPU01, JFPX],  1>;
 defm : X86WriteResPairUnsupported<WriteFVarShuffle256>;
 
@@ -368,33 +385,42 @@
 defm : JWriteResFpuPair<WriteCvtSS2I,      [JFPU1, JSTC, JFPA, JALU0], 7, [1,1,1,1], 2>;
 defm : JWriteResFpuPair<WriteCvtPS2I,      [JFPU1, JSTC], 3, [1,1], 1>;
 defm : JWriteResYMMPair<WriteCvtPS2IY,     [JFPU1, JSTC], 3, [2,2], 2>;
+defm : X86WriteResPairUnsupported<WriteCvtPS2IZ>;
 defm : JWriteResFpuPair<WriteCvtSD2I,      [JFPU1, JSTC, JFPA, JALU0], 7, [1,1,1,1], 2>;
 defm : JWriteResFpuPair<WriteCvtPD2I,      [JFPU1, JSTC], 3, [1,1], 1>;
 defm : JWriteResYMMPair<WriteCvtPD2IY,     [JFPU1, JSTC, JFPX], 6, [2,2,4], 3>;
+defm : X86WriteResPairUnsupported<WriteCvtPD2IZ>;
 
 // FIXME: f+3 ST, LD+STC latency
 defm : JWriteResFpuPair<WriteCvtI2SS,      [JFPU1, JSTC], 9, [1,1], 2>;
 defm : JWriteResFpuPair<WriteCvtI2PS,      [JFPU1, JSTC], 3, [1,1], 1>;
 defm : JWriteResYMMPair<WriteCvtI2PSY,     [JFPU1, JSTC], 3, [2,2], 2>;
+defm : X86WriteResPairUnsupported<WriteCvtI2PSZ>;
 defm : JWriteResFpuPair<WriteCvtI2SD,      [JFPU1, JSTC], 9, [1,1], 2>;
 defm : JWriteResFpuPair<WriteCvtI2PD,      [JFPU1, JSTC], 3, [1,1], 1>;
 defm : JWriteResYMMPair<WriteCvtI2PDY,     [JFPU1, JSTC], 3, [2,2], 2>;
+defm : X86WriteResPairUnsupported<WriteCvtI2PDZ>;
 
 defm : JWriteResFpuPair<WriteCvtSS2SD,      [JFPU1, JSTC], 7, [1,2], 2>;
 defm : JWriteResFpuPair<WriteCvtPS2PD,      [JFPU1, JSTC], 2, [1,1], 1>;
 defm : JWriteResYMMPair<WriteCvtPS2PDY,     [JFPU1, JSTC], 2, [2,2], 2>;
+defm : X86WriteResPairUnsupported<WriteCvtPS2PDZ>;
 
 defm : JWriteResFpuPair<WriteCvtSD2SS,    [JFPU1, JSTC], 7, [1,2], 2>;
 defm : JWriteResFpuPair<WriteCvtPD2PS,    [JFPU1, JSTC], 3, [1,1], 1>;
 defm : JWriteResYMMPair<WriteCvtPD2PSY,   [JFPU1, JSTC, JFPX], 6, [2,2,4], 3>;
+defm : X86WriteResPairUnsupported<WriteCvtPD2PSZ>;
 
 defm : JWriteResFpuPair<WriteCvtPH2PS,     [JFPU1, JSTC], 3, [1,1], 1>;
 defm : JWriteResYMMPair<WriteCvtPH2PSY,    [JFPU1, JSTC], 3, [2,2], 2>;
+defm : X86WriteResPairUnsupported<WriteCvtPH2PSZ>;
 
 defm : X86WriteRes<WriteCvtPS2PH,                 [JFPU1, JSTC], 3, [1,1], 1>;
 defm : X86WriteRes<WriteCvtPS2PHY,          [JFPU1, JSTC, JFPX], 6, [2,2,2], 3>;
+defm : X86WriteResUnsupported<WriteCvtPS2PHZ>;
 defm : X86WriteRes<WriteCvtPS2PHSt,        [JFPU1, JSTC, JSAGU], 4, [1,1,1], 1>;
 defm : X86WriteRes<WriteCvtPS2PHYSt, [JFPU1, JSTC, JFPX, JSAGU], 7, [2,2,2,1], 3>;
+defm : X86WriteResUnsupported<WriteCvtPS2PHZSt>;
 
 ////////////////////////////////////////////////////////////////////////////////
 // Vector integer operations.
@@ -425,42 +451,56 @@
 defm : JWriteResFpuPair<WriteVecALU,      [JFPU01, JVALU], 1>;
 defm : JWriteResFpuPair<WriteVecALUX,     [JFPU01, JVALU], 1>;
 defm : JWriteResFpuPair<WriteVecALUY,     [JFPU01, JVALU], 1>;
+defm : X86WriteResPairUnsupported<WriteVecALUZ>;
 defm : JWriteResFpuPair<WriteVecShift,    [JFPU01, JVALU], 1>;
 defm : JWriteResFpuPair<WriteVecShiftX,   [JFPU01, JVALU], 1>;
 defm : JWriteResFpuPair<WriteVecShiftY,   [JFPU01, JVALU], 1>;
+defm : X86WriteResPairUnsupported<WriteVecShiftZ>;
 defm : JWriteResFpuPair<WriteVecShiftImm, [JFPU01, JVALU], 1>;
 defm : JWriteResFpuPair<WriteVecShiftImmX,[JFPU01, JVALU], 1>;
 defm : JWriteResFpuPair<WriteVecShiftImmY,[JFPU01, JVALU], 1>;
+defm : X86WriteResPairUnsupported<WriteVecShiftImmZ>;
 defm : JWriteResFpuPair<WriteVecIMul,     [JFPU0, JVIMUL], 2>;
 defm : JWriteResFpuPair<WriteVecIMulX,    [JFPU0, JVIMUL], 2>;
 defm : JWriteResFpuPair<WriteVecIMulY,    [JFPU0, JVIMUL], 2>;
+defm : X86WriteResPairUnsupported<WriteVecIMulZ>;
 defm : JWriteResFpuPair<WritePMULLD,      [JFPU0, JFPU01, JVIMUL, JVALU], 4, [2, 1, 2, 1], 3>;
 defm : JWriteResFpuPair<WritePMULLDY,     [JFPU0, JFPU01, JVIMUL, JVALU], 4, [2, 1, 2, 1], 3>;
+defm : X86WriteResPairUnsupported<WritePMULLDZ>;
 defm : JWriteResFpuPair<WriteMPSAD,       [JFPU0, JVIMUL], 3, [1, 2]>;
 defm : JWriteResFpuPair<WriteMPSADY,      [JFPU0, JVIMUL], 3, [1, 2]>;
+defm : X86WriteResPairUnsupported<WriteMPSADZ>;
 defm : JWriteResFpuPair<WritePSADBW,      [JFPU01, JVALU], 2>;
 defm : JWriteResFpuPair<WritePSADBWX,     [JFPU01, JVALU], 2>;
 defm : JWriteResFpuPair<WritePSADBWY,     [JFPU01, JVALU], 2>;
+defm : X86WriteResPairUnsupported<WritePSADBWZ>;
 defm : JWriteResFpuPair<WritePHMINPOS,    [JFPU0,  JVALU], 2>;
 defm : JWriteResFpuPair<WriteShuffle,     [JFPU01, JVALU], 1>;
 defm : JWriteResFpuPair<WriteShuffleX,    [JFPU01, JVALU], 1>;
 defm : JWriteResFpuPair<WriteShuffleY,    [JFPU01, JVALU], 1>;
+defm : X86WriteResPairUnsupported<WriteShuffleZ>;
 defm : JWriteResFpuPair<WriteVarShuffle,  [JFPU01, JVALU], 2, [1, 4], 3>;
 defm : JWriteResFpuPair<WriteVarShuffleX, [JFPU01, JVALU], 2, [1, 4], 3>;
 defm : JWriteResFpuPair<WriteVarShuffleY, [JFPU01, JVALU], 2, [1, 4], 3>;
+defm : X86WriteResPairUnsupported<WriteVarShuffleZ>;
 defm : JWriteResFpuPair<WriteBlend,       [JFPU01, JVALU], 1>;
 defm : JWriteResFpuPair<WriteBlendY,      [JFPU01, JVALU], 1>;
+defm : X86WriteResPairUnsupported<WriteBlendZ>;
 defm : JWriteResFpuPair<WriteVarBlend,    [JFPU01, JVALU], 2, [1, 4], 3>;
 defm : JWriteResFpuPair<WriteVarBlendY,   [JFPU01, JVALU], 2, [1, 4], 3>;
+defm : X86WriteResPairUnsupported<WriteVarBlendZ>;
 defm : JWriteResFpuPair<WriteVecLogic,    [JFPU01, JVALU], 1>;
 defm : JWriteResFpuPair<WriteVecLogicX,   [JFPU01, JVALU], 1>;
 defm : X86WriteResPairUnsupported<WriteVecLogicY>;
+defm : X86WriteResPairUnsupported<WriteVecLogicZ>;
 defm : JWriteResFpuPair<WriteVecTest,     [JFPU0, JFPA, JALU0], 3>;
-defm : JWriteResYMMPair<WriteVecTestY ,   [JFPU01, JFPX, JFPA, JALU0], 4, [2, 2, 2, 1], 3>;
+defm : JWriteResYMMPair<WriteVecTestY,    [JFPU01, JFPX, JFPA, JALU0], 4, [2, 2, 2, 1], 3>;
+defm : X86WriteResPairUnsupported<WriteVecTestZ>;
 defm : JWriteResFpuPair<WriteShuffle256,  [JFPU01, JVALU], 1>;
 defm : X86WriteResPairUnsupported<WriteVarShuffle256>;
 defm : X86WriteResPairUnsupported<WriteVarVecShift>;
 defm : X86WriteResPairUnsupported<WriteVarVecShiftY>;
+defm : X86WriteResPairUnsupported<WriteVarVecShiftZ>;
 
 ////////////////////////////////////////////////////////////////////////////////
 // Vector insert/extract operations.
@@ -555,7 +595,7 @@
   let Latency = 0;
 }
 
-// Certain instructions that use the same register for both source 
+// Certain instructions that use the same register for both source
 // operands do not have a real dependency on the previous contents of the
 // register, and thus, do not have to wait before completing. They can be
 // optimized out at register renaming stage.
Index: lib/Target/X86/X86ScheduleSLM.td
===================================================================
--- lib/Target/X86/X86ScheduleSLM.td
+++ lib/Target/X86/X86ScheduleSLM.td
@@ -159,22 +159,28 @@
 defm : SLMWriteResPair<WriteFAdd,     [SLM_FPC_RSV1], 3>;
 defm : SLMWriteResPair<WriteFAddX,    [SLM_FPC_RSV1], 3>;
 defm : SLMWriteResPair<WriteFAddY,    [SLM_FPC_RSV1], 3>;
+defm : X86WriteResPairUnsupported<WriteFAddZ>;
 defm : SLMWriteResPair<WriteFAdd64,   [SLM_FPC_RSV1], 3>;
 defm : SLMWriteResPair<WriteFAdd64X,  [SLM_FPC_RSV1], 3>;
 defm : SLMWriteResPair<WriteFAdd64Y,  [SLM_FPC_RSV1], 3>;
+defm : X86WriteResPairUnsupported<WriteFAdd64Z>;
 defm : SLMWriteResPair<WriteFCmp,     [SLM_FPC_RSV1], 3>;
 defm : SLMWriteResPair<WriteFCmpX,    [SLM_FPC_RSV1], 3>;
 defm : SLMWriteResPair<WriteFCmpY,    [SLM_FPC_RSV1], 3>;
+defm : X86WriteResPairUnsupported<WriteFCmpZ>;
 defm : SLMWriteResPair<WriteFCmp64,   [SLM_FPC_RSV1], 3>;
 defm : SLMWriteResPair<WriteFCmp64X,  [SLM_FPC_RSV1], 3>;
 defm : SLMWriteResPair<WriteFCmp64Y,  [SLM_FPC_RSV1], 3>;
+defm : X86WriteResPairUnsupported<WriteFCmp64Z>;
 defm : SLMWriteResPair<WriteFCom,     [SLM_FPC_RSV1], 3>;
 defm : SLMWriteResPair<WriteFMul,     [SLM_FPC_RSV0, SLMFPMultiplier], 5, [1,2]>;
 defm : SLMWriteResPair<WriteFMulX,    [SLM_FPC_RSV0, SLMFPMultiplier], 5, [1,2]>;
 defm : SLMWriteResPair<WriteFMulY,    [SLM_FPC_RSV0, SLMFPMultiplier], 5, [1,2]>;
+defm : X86WriteResPairUnsupported<WriteFMulZ>;
 defm : SLMWriteResPair<WriteFMul64,   [SLM_FPC_RSV0, SLMFPMultiplier], 5, [1,2]>;
 defm : SLMWriteResPair<WriteFMul64X,  [SLM_FPC_RSV0, SLMFPMultiplier], 5, [1,2]>;
 defm : SLMWriteResPair<WriteFMul64Y,  [SLM_FPC_RSV0, SLMFPMultiplier], 5, [1,2]>;
+defm : X86WriteResPairUnsupported<WriteFMul64Z>;
 defm : SLMWriteResPair<WriteFDiv,     [SLM_FPC_RSV0, SLMFPDivider], 19, [1,17]>;
 defm : SLMWriteResPair<WriteFDivX,    [SLM_FPC_RSV0, SLMFPDivider], 39, [1,39]>;
 defm : SLMWriteResPair<WriteFDivY,    [SLM_FPC_RSV0, SLMFPDivider], 39, [1,39]>;
@@ -186,9 +192,11 @@
 defm : SLMWriteResPair<WriteFRcp,     [SLM_FPC_RSV0], 5>;
 defm : SLMWriteResPair<WriteFRcpX,    [SLM_FPC_RSV0], 5>;
 defm : SLMWriteResPair<WriteFRcpY,    [SLM_FPC_RSV0], 5>;
+defm : X86WriteResPairUnsupported<WriteFRcpZ>;
 defm : SLMWriteResPair<WriteFRsqrt,   [SLM_FPC_RSV0], 5>;
 defm : SLMWriteResPair<WriteFRsqrtX,  [SLM_FPC_RSV0], 5>;
 defm : SLMWriteResPair<WriteFRsqrtY,  [SLM_FPC_RSV0], 5>;
+defm : X86WriteResPairUnsupported<WriteFRsqrtZ>;
 defm : SLMWriteResPair<WriteFSqrt,    [SLM_FPC_RSV0,SLMFPDivider], 20, [1,20], 1, 3>;
 defm : SLMWriteResPair<WriteFSqrtX,   [SLM_FPC_RSV0,SLMFPDivider], 41, [1,40], 1, 3>;
 defm : SLMWriteResPair<WriteFSqrtY,   [SLM_FPC_RSV0,SLMFPDivider], 41, [1,40], 1, 3>;
@@ -201,40 +209,52 @@
 defm : SLMWriteResPair<WriteDPPD,   [SLM_FPC_RSV1], 3>;
 defm : SLMWriteResPair<WriteDPPS,   [SLM_FPC_RSV1], 3>;
 defm : SLMWriteResPair<WriteDPPSY,  [SLM_FPC_RSV1], 3>;
+defm : X86WriteResPairUnsupported<WriteDPPSZ>;
 defm : SLMWriteResPair<WriteFSign,  [SLM_FPC_RSV01], 1>;
 defm : SLMWriteResPair<WriteFRnd,   [SLM_FPC_RSV1], 3>;
 defm : SLMWriteResPair<WriteFRndY,  [SLM_FPC_RSV1], 3>;
+defm : X86WriteResPairUnsupported<WriteFRndZ>;
 defm : SLMWriteResPair<WriteFLogic, [SLM_FPC_RSV01], 1>;
 defm : SLMWriteResPair<WriteFLogicY, [SLM_FPC_RSV01], 1>;
+defm : X86WriteResPairUnsupported<WriteFLogicZ>;
 defm : SLMWriteResPair<WriteFTest,  [SLM_FPC_RSV01], 1>;
 defm : SLMWriteResPair<WriteFTestY, [SLM_FPC_RSV01], 1>;
+defm : X86WriteResPairUnsupported<WriteFTestZ>;
 defm : SLMWriteResPair<WriteFShuffle,  [SLM_FPC_RSV0], 1>;
 defm : SLMWriteResPair<WriteFShuffleY, [SLM_FPC_RSV0], 1>;
+defm : X86WriteResPairUnsupported<WriteFShuffleZ>;
 defm : SLMWriteResPair<WriteFVarShuffle, [SLM_FPC_RSV0],  1>;
 defm : SLMWriteResPair<WriteFVarShuffleY,[SLM_FPC_RSV0],  1>;
+defm : X86WriteResPairUnsupported<WriteFVarShuffleZ>;
 defm : SLMWriteResPair<WriteFBlend,  [SLM_FPC_RSV0],  1>;
 
 // Conversion between integer and float.
 defm : SLMWriteResPair<WriteCvtSS2I,   [SLM_FPC_RSV01], 4>;
 defm : SLMWriteResPair<WriteCvtPS2I,   [SLM_FPC_RSV01], 4>;
 defm : SLMWriteResPair<WriteCvtPS2IY,  [SLM_FPC_RSV01], 4>;
+defm : X86WriteResPairUnsupported<WriteCvtPS2IZ>;
 defm : SLMWriteResPair<WriteCvtSD2I,   [SLM_FPC_RSV01], 4>;
 defm : SLMWriteResPair<WriteCvtPD2I,   [SLM_FPC_RSV01], 4>;
 defm : SLMWriteResPair<WriteCvtPD2IY,  [SLM_FPC_RSV01], 4>;
+defm : X86WriteResPairUnsupported<WriteCvtPD2IZ>;
 
 defm : SLMWriteResPair<WriteCvtI2SS,   [SLM_FPC_RSV01], 4>;
 defm : SLMWriteResPair<WriteCvtI2PS,   [SLM_FPC_RSV01], 4>;
 defm : SLMWriteResPair<WriteCvtI2PSY,  [SLM_FPC_RSV01], 4>;
+defm : X86WriteResPairUnsupported<WriteCvtI2PSZ>;
 defm : SLMWriteResPair<WriteCvtI2SD,   [SLM_FPC_RSV01], 4>;
 defm : SLMWriteResPair<WriteCvtI2PD,   [SLM_FPC_RSV01], 4>;
 defm : SLMWriteResPair<WriteCvtI2PDY,  [SLM_FPC_RSV01], 4>;
+defm : X86WriteResPairUnsupported<WriteCvtI2PDZ>;
 
 defm : SLMWriteResPair<WriteCvtSS2SD,  [SLM_FPC_RSV01], 4>;
 defm : SLMWriteResPair<WriteCvtPS2PD,  [SLM_FPC_RSV01], 4>;
 defm : SLMWriteResPair<WriteCvtPS2PDY, [SLM_FPC_RSV01], 4>;
+defm : X86WriteResPairUnsupported<WriteCvtPS2PDZ>;
 defm : SLMWriteResPair<WriteCvtSD2SS,  [SLM_FPC_RSV01], 4>;
 defm : SLMWriteResPair<WriteCvtPD2PS,  [SLM_FPC_RSV01], 4>;
 defm : SLMWriteResPair<WriteCvtPD2PSY, [SLM_FPC_RSV01], 4>;
+defm : X86WriteResPairUnsupported<WriteCvtPD2PSZ>;
 
 // Vector integer operations.
 def  : WriteRes<WriteVecLoad,         [SLM_MEC_RSV]> { let Latency = 3; }
@@ -260,37 +280,49 @@
 defm : SLMWriteResPair<WriteVecShift,    [SLM_FPC_RSV0],  1>;
 defm : SLMWriteResPair<WriteVecShiftX,   [SLM_FPC_RSV0],  1>;
 defm : SLMWriteResPair<WriteVecShiftY,   [SLM_FPC_RSV0],  1>;
+defm : X86WriteResPairUnsupported<WriteVecShiftZ>;
 defm : SLMWriteResPair<WriteVecShiftImm, [SLM_FPC_RSV0],  1>;
 defm : SLMWriteResPair<WriteVecShiftImmX,[SLM_FPC_RSV0],  1>;
 defm : SLMWriteResPair<WriteVecShiftImmY,[SLM_FPC_RSV0],  1>;
+defm : X86WriteResPairUnsupported<WriteVecShiftImmZ>;
 defm : SLMWriteResPair<WriteVecLogic, [SLM_FPC_RSV01], 1>;
 defm : SLMWriteResPair<WriteVecLogicX,[SLM_FPC_RSV01], 1>;
 defm : SLMWriteResPair<WriteVecLogicY,[SLM_FPC_RSV01], 1>;
+defm : X86WriteResPairUnsupported<WriteVecLogicZ>;
 defm : SLMWriteResPair<WriteVecTest,  [SLM_FPC_RSV01], 1>;
 defm : SLMWriteResPair<WriteVecTestY, [SLM_FPC_RSV01], 1>;
+defm : X86WriteResPairUnsupported<WriteVecTestZ>;
 defm : SLMWriteResPair<WriteVecALU,   [SLM_FPC_RSV01],  1>;
 defm : SLMWriteResPair<WriteVecALUX,  [SLM_FPC_RSV01],  1>;
 defm : SLMWriteResPair<WriteVecALUY,  [SLM_FPC_RSV01],  1>;
+defm : X86WriteResPairUnsupported<WriteVecALUZ>;
 defm : SLMWriteResPair<WriteVecIMul,  [SLM_FPC_RSV0],   4>;
 defm : SLMWriteResPair<WriteVecIMulX, [SLM_FPC_RSV0],   4>;
 defm : SLMWriteResPair<WriteVecIMulY, [SLM_FPC_RSV0],   4>;
+defm : X86WriteResPairUnsupported<WriteVecIMulZ>;
 // FIXME: The below is closer to correct, but caused some perf regressions.
 //defm : SLMWriteResPair<WritePMULLD,  [SLM_FPC_RSV0],   11, [11], 7>;
 defm : SLMWriteResPair<WritePMULLD,  [SLM_FPC_RSV0],   4>;
 defm : SLMWriteResPair<WritePMULLDY, [SLM_FPC_RSV0],   4>;
+defm : X86WriteResPairUnsupported<WritePMULLDZ>;
 defm : SLMWriteResPair<WriteShuffle,  [SLM_FPC_RSV0],  1>;
 defm : SLMWriteResPair<WriteShuffleY, [SLM_FPC_RSV0],  1>;
+defm : X86WriteResPairUnsupported<WriteShuffleZ>;
 defm : SLMWriteResPair<WriteShuffleX, [SLM_FPC_RSV0],  1>;
 defm : SLMWriteResPair<WriteVarShuffle,  [SLM_FPC_RSV0],  1>;
 defm : SLMWriteResPair<WriteVarShuffleX, [SLM_FPC_RSV0],  1>;
 defm : SLMWriteResPair<WriteVarShuffleY, [SLM_FPC_RSV0],  1>;
+defm : X86WriteResPairUnsupported<WriteVarShuffleZ>;
 defm : SLMWriteResPair<WriteBlend,  [SLM_FPC_RSV0],  1>;
 defm : SLMWriteResPair<WriteBlendY, [SLM_FPC_RSV0],  1>;
+defm : X86WriteResPairUnsupported<WriteBlendZ>;
 defm : SLMWriteResPair<WriteMPSAD,  [SLM_FPC_RSV0],  7>;
 defm : SLMWriteResPair<WriteMPSADY, [SLM_FPC_RSV0],  7>;
+defm : X86WriteResPairUnsupported<WriteMPSADZ>;
 defm : SLMWriteResPair<WritePSADBW,  [SLM_FPC_RSV0],  4>;
 defm : SLMWriteResPair<WritePSADBWX, [SLM_FPC_RSV0],  4>;
 defm : SLMWriteResPair<WritePSADBWY, [SLM_FPC_RSV0],  4>;
+defm : X86WriteResPairUnsupported<WritePSADBWZ>;
 defm : SLMWriteResPair<WritePHMINPOS,  [SLM_FPC_RSV0],   4>;
 
 // Vector insert/extract operations.
@@ -309,9 +341,11 @@
 
 defm : SLMWriteResPair<WriteFHAdd,   [SLM_FPC_RSV01], 3, [2]>;
 defm : SLMWriteResPair<WriteFHAddY,  [SLM_FPC_RSV01], 3, [2]>;
+defm : X86WriteResPairUnsupported<WriteFHAddZ>;
 defm : SLMWriteResPair<WritePHAdd,   [SLM_FPC_RSV01], 1>;
 defm : SLMWriteResPair<WritePHAddX,  [SLM_FPC_RSV01], 1>;
 defm : SLMWriteResPair<WritePHAddY,  [SLM_FPC_RSV01], 1>;
+defm : X86WriteResPairUnsupported<WritePHAddZ>;
 
 // String instructions.
 // Packed Compare Implicit Length Strings, Return Mask
@@ -407,25 +441,33 @@
 // scheduling resources anyway.
 def  : WriteRes<WriteIMulH, [SLM_FPC_RSV0]>;
 defm : X86WriteResPairUnsupported<WriteFBlendY>;
+defm : X86WriteResPairUnsupported<WriteFBlendZ>;
 defm : SLMWriteResPair<WriteVarBlend, [SLM_FPC_RSV0], 1>;
 defm : X86WriteResPairUnsupported<WriteVarBlendY>;
+defm : X86WriteResPairUnsupported<WriteVarBlendZ>;
 defm : SLMWriteResPair<WriteFVarBlend, [SLM_FPC_RSV0], 1>;
 defm : X86WriteResPairUnsupported<WriteFVarBlendY>;
+defm : X86WriteResPairUnsupported<WriteFVarBlendZ>;
 defm : X86WriteResPairUnsupported<WriteFShuffle256>;
 defm : X86WriteResPairUnsupported<WriteFVarShuffle256>;
 defm : X86WriteResPairUnsupported<WriteShuffle256>;
 defm : X86WriteResPairUnsupported<WriteVarShuffle256>;
 defm : SLMWriteResPair<WriteVarVecShift,  [SLM_FPC_RSV0],  1>;
 defm : X86WriteResPairUnsupported<WriteVarVecShiftY>;
+defm : X86WriteResPairUnsupported<WriteVarVecShiftZ>;
 defm : X86WriteResPairUnsupported<WriteFMA>;
 defm : X86WriteResPairUnsupported<WriteFMAX>;
 defm : X86WriteResPairUnsupported<WriteFMAY>;
+defm : X86WriteResPairUnsupported<WriteFMAZ>;
 
 defm : X86WriteResPairUnsupported<WriteCvtPH2PS>;
 defm : X86WriteResPairUnsupported<WriteCvtPH2PSY>;
+defm : X86WriteResPairUnsupported<WriteCvtPH2PSZ>;
 defm : X86WriteResUnsupported<WriteCvtPS2PH>;
 defm : X86WriteResUnsupported<WriteCvtPS2PHY>;
+defm : X86WriteResUnsupported<WriteCvtPS2PHZ>;
 defm : X86WriteResUnsupported<WriteCvtPS2PHSt>;
 defm : X86WriteResUnsupported<WriteCvtPS2PHYSt>;
+defm : X86WriteResUnsupported<WriteCvtPS2PHZSt>;
 
 } // SchedModel
Index: lib/Target/X86/X86ScheduleZnver1.td
===================================================================
--- lib/Target/X86/X86ScheduleZnver1.td
+++ lib/Target/X86/X86ScheduleZnver1.td
@@ -212,34 +212,45 @@
 defm : ZnWriteResFpuPair<WriteFAdd,      [ZnFPU0],  3>;
 defm : ZnWriteResFpuPair<WriteFAddX,     [ZnFPU0],  3>;
 defm : ZnWriteResFpuPair<WriteFAddY,     [ZnFPU0],  3>;
+defm : X86WriteResPairUnsupported<WriteFAddZ>;
 defm : ZnWriteResFpuPair<WriteFAdd64,    [ZnFPU0],  3>;
 defm : ZnWriteResFpuPair<WriteFAdd64X,   [ZnFPU0],  3>;
 defm : ZnWriteResFpuPair<WriteFAdd64Y,   [ZnFPU0],  3>;
+defm : X86WriteResPairUnsupported<WriteFAdd64Z>;
 defm : ZnWriteResFpuPair<WriteFCmp,      [ZnFPU0],  3>;
 defm : ZnWriteResFpuPair<WriteFCmpX,     [ZnFPU0],  3>;
 defm : ZnWriteResFpuPair<WriteFCmpY,     [ZnFPU0],  3>;
+defm : X86WriteResPairUnsupported<WriteFCmpZ>;
 defm : ZnWriteResFpuPair<WriteFCmp64,    [ZnFPU0],  3>;
 defm : ZnWriteResFpuPair<WriteFCmp64X,   [ZnFPU0],  3>;
 defm : ZnWriteResFpuPair<WriteFCmp64Y,   [ZnFPU0],  3>;
+defm : X86WriteResPairUnsupported<WriteFCmp64Z>;
 defm : ZnWriteResFpuPair<WriteFCom,      [ZnFPU0],  3>;
 defm : ZnWriteResFpuPair<WriteFBlend,    [ZnFPU01], 1>;
 defm : ZnWriteResFpuPair<WriteFBlendY,   [ZnFPU01], 1>;
+defm : X86WriteResPairUnsupported<WriteFBlendZ>;
 defm : ZnWriteResFpuPair<WriteFVarBlend, [ZnFPU01], 1>;
 defm : ZnWriteResFpuPair<WriteFVarBlendY,[ZnFPU01], 1>;
+defm : X86WriteResPairUnsupported<WriteFVarBlendZ>;
 defm : ZnWriteResFpuPair<WriteVarBlend,  [ZnFPU0],  1>;
 defm : ZnWriteResFpuPair<WriteVarBlendY, [ZnFPU0],  1>;
+defm : X86WriteResPairUnsupported<WriteVarBlendZ>;
 defm : ZnWriteResFpuPair<WriteCvtSS2I,   [ZnFPU3],  5>;
 defm : ZnWriteResFpuPair<WriteCvtPS2I,   [ZnFPU3],  5>;
 defm : ZnWriteResFpuPair<WriteCvtPS2IY,  [ZnFPU3],  5>;
+defm : X86WriteResPairUnsupported<WriteCvtPS2IZ>;
 defm : ZnWriteResFpuPair<WriteCvtSD2I,   [ZnFPU3],  5>;
 defm : ZnWriteResFpuPair<WriteCvtPD2I,   [ZnFPU3],  5>;
 defm : ZnWriteResFpuPair<WriteCvtPD2IY,  [ZnFPU3],  5>;
+defm : X86WriteResPairUnsupported<WriteCvtPD2IZ>;
 defm : ZnWriteResFpuPair<WriteCvtI2SS,   [ZnFPU3],  5>;
 defm : ZnWriteResFpuPair<WriteCvtI2PS,   [ZnFPU3],  5>;
 defm : ZnWriteResFpuPair<WriteCvtI2PSY,  [ZnFPU3],  5>;
+defm : X86WriteResPairUnsupported<WriteCvtI2PSZ>;
 defm : ZnWriteResFpuPair<WriteCvtI2SD,   [ZnFPU3],  5>;
 defm : ZnWriteResFpuPair<WriteCvtI2PD,   [ZnFPU3],  5>;
 defm : ZnWriteResFpuPair<WriteCvtI2PDY,  [ZnFPU3],  5>;
+defm : X86WriteResPairUnsupported<WriteCvtI2PDZ>;
 defm : ZnWriteResFpuPair<WriteFDiv,      [ZnFPU3], 15>;
 defm : ZnWriteResFpuPair<WriteFDivX,     [ZnFPU3], 15>;
 //defm : ZnWriteResFpuPair<WriteFDivY,     [ZnFPU3], 15>;
@@ -251,29 +262,39 @@
 defm : ZnWriteResFpuPair<WriteFSign,     [ZnFPU3],  2>;
 defm : ZnWriteResFpuPair<WriteFRnd,      [ZnFPU3],  4, [1], 1, 7, 1>; // FIXME: Should folds require 1 extra uops?
 defm : ZnWriteResFpuPair<WriteFRndY,     [ZnFPU3],  4, [1], 1, 7, 1>; // FIXME: Should folds require 1 extra uops?
+defm : X86WriteResPairUnsupported<WriteFRndZ>;
 defm : ZnWriteResFpuPair<WriteFLogic,    [ZnFPU],   1>;
 defm : ZnWriteResFpuPair<WriteFLogicY,   [ZnFPU],   1>;
+defm : X86WriteResPairUnsupported<WriteFLogicZ>;
 defm : ZnWriteResFpuPair<WriteFTest,     [ZnFPU],   1>;
 defm : ZnWriteResFpuPair<WriteFTestY,    [ZnFPU],   1>;
+defm : X86WriteResPairUnsupported<WriteFTestZ>;
 defm : ZnWriteResFpuPair<WriteFShuffle,  [ZnFPU12], 1>;
 defm : ZnWriteResFpuPair<WriteFShuffleY, [ZnFPU12], 1>;
+defm : X86WriteResPairUnsupported<WriteFShuffleZ>;
 defm : ZnWriteResFpuPair<WriteFVarShuffle, [ZnFPU12], 1>;
 defm : ZnWriteResFpuPair<WriteFVarShuffleY,[ZnFPU12], 1>;
+defm : X86WriteResPairUnsupported<WriteFVarShuffleZ>;
 defm : ZnWriteResFpuPair<WriteFMul,      [ZnFPU01], 3, [1], 1, 7, 1>;
 defm : ZnWriteResFpuPair<WriteFMulX,     [ZnFPU01], 3, [1], 1, 7, 1>;
 defm : ZnWriteResFpuPair<WriteFMulY,     [ZnFPU01], 4, [1], 1, 7, 1>;
+defm : X86WriteResPairUnsupported<WriteFMulZ>;
 defm : ZnWriteResFpuPair<WriteFMul64,    [ZnFPU01], 3, [1], 1, 7, 1>;
 defm : ZnWriteResFpuPair<WriteFMul64X,   [ZnFPU01], 3, [1], 1, 7, 1>;
 defm : ZnWriteResFpuPair<WriteFMul64Y,   [ZnFPU01], 4, [1], 1, 7, 1>;
+defm : X86WriteResPairUnsupported<WriteFMul64Z>;
 defm : ZnWriteResFpuPair<WriteFMA,       [ZnFPU03], 5>;
 defm : ZnWriteResFpuPair<WriteFMAX,      [ZnFPU03], 5>;
 defm : ZnWriteResFpuPair<WriteFMAY,      [ZnFPU03], 5>;
+defm : X86WriteResPairUnsupported<WriteFMAZ>;
 defm : ZnWriteResFpuPair<WriteFRcp,      [ZnFPU01], 5>;
 defm : ZnWriteResFpuPair<WriteFRcpX,     [ZnFPU01], 5>;
 defm : ZnWriteResFpuPair<WriteFRcpY,     [ZnFPU01], 5, [1], 1, 7, 2>;
+defm : X86WriteResPairUnsupported<WriteFRcpZ>;
 //defm : ZnWriteResFpuPair<WriteFRsqrt,    [ZnFPU02], 5>;
 defm : ZnWriteResFpuPair<WriteFRsqrtX,   [ZnFPU01], 5, [1], 1, 7, 1>;
 //defm : ZnWriteResFpuPair<WriteFRsqrtY,   [ZnFPU01], 5, [2], 2>;
+defm : X86WriteResPairUnsupported<WriteFRsqrtZ>;
 defm : ZnWriteResFpuPair<WriteFSqrt,     [ZnFPU3], 20, [20]>;
 defm : ZnWriteResFpuPair<WriteFSqrtX,    [ZnFPU3], 20, [20]>;
 defm : ZnWriteResFpuPair<WriteFSqrtY,    [ZnFPU3], 28, [28], 1, 7, 1>;
@@ -309,40 +330,52 @@
 defm : ZnWriteResFpuPair<WriteVecShift,   [ZnFPU],   1>;
 defm : ZnWriteResFpuPair<WriteVecShiftX,  [ZnFPU2],  1>;
 defm : ZnWriteResFpuPair<WriteVecShiftY,  [ZnFPU2],  2>;
+defm : X86WriteResPairUnsupported<WriteVecShiftZ>;
 defm : ZnWriteResFpuPair<WriteVecShiftImm,  [ZnFPU], 1>;
 defm : ZnWriteResFpuPair<WriteVecShiftImmX, [ZnFPU], 1>;
 defm : ZnWriteResFpuPair<WriteVecShiftImmY, [ZnFPU], 1>;
+defm : X86WriteResPairUnsupported<WriteVecShiftImmZ>;
 defm : ZnWriteResFpuPair<WriteVecLogic,   [ZnFPU],   1>;
 defm : ZnWriteResFpuPair<WriteVecLogicX,  [ZnFPU],   1>;
 defm : ZnWriteResFpuPair<WriteVecLogicY,  [ZnFPU],   1>;
+defm : X86WriteResPairUnsupported<WriteVecLogicZ>;
 defm : ZnWriteResFpuPair<WriteVecTest,    [ZnFPU12], 1, [2], 1, 7, 1>;
 defm : ZnWriteResFpuPair<WriteVecTestY,   [ZnFPU12], 1, [2], 1, 7, 1>;
+defm : X86WriteResPairUnsupported<WriteVecTestZ>;
 defm : ZnWriteResFpuPair<WriteVecALU,     [ZnFPU],   1>;
 defm : ZnWriteResFpuPair<WriteVecALUX,    [ZnFPU],   1>;
 defm : ZnWriteResFpuPair<WriteVecALUY,    [ZnFPU],   1>;
+defm : X86WriteResPairUnsupported<WriteVecALUZ>;
 defm : ZnWriteResFpuPair<WriteVecIMul,    [ZnFPU0],  4>;
 defm : ZnWriteResFpuPair<WriteVecIMulX,   [ZnFPU0],  4>;
 defm : ZnWriteResFpuPair<WriteVecIMulY,   [ZnFPU0],  4>;
+defm : X86WriteResPairUnsupported<WriteVecIMulZ>;
 defm : ZnWriteResFpuPair<WritePMULLD,     [ZnFPU0],  4, [1], 1, 7, 1>; // FIXME
 defm : ZnWriteResFpuPair<WritePMULLDY,    [ZnFPU0],  5, [2], 1, 7, 1>; // FIXME
+defm : X86WriteResPairUnsupported<WritePMULLDZ>;
 defm : ZnWriteResFpuPair<WriteShuffle,    [ZnFPU],   1>;
 defm : ZnWriteResFpuPair<WriteShuffleX,   [ZnFPU],   1>;
 defm : ZnWriteResFpuPair<WriteShuffleY,   [ZnFPU],   1>;
+defm : X86WriteResPairUnsupported<WriteShuffleZ>;
 defm : ZnWriteResFpuPair<WriteVarShuffle, [ZnFPU],   1>;
 defm : ZnWriteResFpuPair<WriteVarShuffleX,[ZnFPU],   1>;
 defm : ZnWriteResFpuPair<WriteVarShuffleY,[ZnFPU],   1>;
+defm : X86WriteResPairUnsupported<WriteVarShuffleZ>;
 defm : ZnWriteResFpuPair<WriteBlend,      [ZnFPU01], 1>;
 defm : ZnWriteResFpuPair<WriteBlendY,     [ZnFPU01], 1>;
+defm : X86WriteResPairUnsupported<WriteBlendZ>;
 defm : ZnWriteResFpuPair<WriteShuffle256, [ZnFPU],   2>;
 defm : ZnWriteResFpuPair<WriteVarShuffle256, [ZnFPU],   2>;
 defm : ZnWriteResFpuPair<WritePSADBW,     [ZnFPU0],  3>;
 defm : ZnWriteResFpuPair<WritePSADBWX,    [ZnFPU0],  3>;
 defm : ZnWriteResFpuPair<WritePSADBWY,    [ZnFPU0],  3>;
+defm : X86WriteResPairUnsupported<WritePSADBWZ>;
 defm : ZnWriteResFpuPair<WritePHMINPOS,   [ZnFPU0],  4>;
 
 // Vector Shift Operations
 defm : ZnWriteResFpuPair<WriteVarVecShift,  [ZnFPU12], 1>;
 defm : ZnWriteResFpuPair<WriteVarVecShiftY, [ZnFPU12], 1>;
+defm : X86WriteResPairUnsupported<WriteVarVecShiftZ>;
 
 // Vector insert/extract operations.
 defm : ZnWriteResFpuPair<WriteVecInsert,   [ZnFPU],   1>;
@@ -1144,9 +1177,10 @@
 // CVTPD2PS.
 // x,x.
 def : SchedAlias<WriteCvtPD2PS,  ZnWriteCVTPD2PSr>;
-
 // y,y.
 def : SchedAlias<WriteCvtPD2PSY, ZnWriteCVTPD2PSYr>;
+// z,z.
+defm : X86WriteResUnsupported<WriteCvtPD2PSZ>;
 
 def ZnWriteCVTPD2PSLd: SchedWriteRes<[ZnAGU,ZnFPU03]> {
   let Latency = 11;
@@ -1161,6 +1195,8 @@
   let Latency = 11;
 }
 def : SchedAlias<WriteCvtPD2PSYLd, ZnWriteCVTPD2PSYLd>;
+// z,m512
+defm : X86WriteResUnsupported<WriteCvtPD2PSZLd>;
 
 // CVTSD2SS.
 // x,x.
@@ -1185,12 +1221,14 @@
 }
 def : SchedAlias<WriteCvtPS2PDLd, ZnWriteCVTPS2PDLd>;
 def : SchedAlias<WriteCvtPS2PDYLd, ZnWriteCVTPS2PDLd>;
+defm : X86WriteResUnsupported<WriteCvtPS2PDZLd>;
 
 // y,x.
 def ZnWriteVCVTPS2PDY : SchedWriteRes<[ZnFPU3]> {
   let Latency = 3;
 }
 def : SchedAlias<WriteCvtPS2PDY, ZnWriteVCVTPS2PDY>;
+defm : X86WriteResUnsupported<WriteCvtPS2PDZ>;
 
 // CVTSS2SD.
 // x,x.
@@ -1288,17 +1326,21 @@
 // x,v,i.
 def : SchedAlias<WriteCvtPS2PH,    ZnWriteMicrocoded>;
 def : SchedAlias<WriteCvtPS2PHY,   ZnWriteMicrocoded>;
+defm : X86WriteResUnsupported<WriteCvtPS2PHZ>;
 // m,v,i.
 def : SchedAlias<WriteCvtPS2PHSt,  ZnWriteMicrocoded>;
 def : SchedAlias<WriteCvtPS2PHYSt, ZnWriteMicrocoded>;
+defm : X86WriteResUnsupported<WriteCvtPS2PHZSt>;
 
 // VCVTPH2PS.
 // v,x.
 def : SchedAlias<WriteCvtPH2PS,    ZnWriteMicrocoded>;
 def : SchedAlias<WriteCvtPH2PSY,   ZnWriteMicrocoded>;
+defm : X86WriteResUnsupported<WriteCvtPH2PSZ>;
 // v,m.
 def : SchedAlias<WriteCvtPH2PSLd,  ZnWriteMicrocoded>;
 def : SchedAlias<WriteCvtPH2PSYLd, ZnWriteMicrocoded>;
+defm : X86WriteResUnsupported<WriteCvtPH2PSZLd>;
 
 //-- SSE4A instructions --//
 // EXTRQ