diff --git a/llvm/lib/Target/X86/X86SchedBroadwell.td b/llvm/lib/Target/X86/X86SchedBroadwell.td
index 999ecd3..ee29d63 100755
--- a/llvm/lib/Target/X86/X86SchedBroadwell.td
+++ b/llvm/lib/Target/X86/X86SchedBroadwell.td
@@ -206,9 +206,13 @@
 defm : BWWriteResPair<WriteVecIMul,  [BWPort0],   5>; // Vector integer multiply.
 defm : BWWriteResPair<WritePMULLD,   [BWPort0], 10, [2], 2, 5>; // PMULLD
 defm : BWWriteResPair<WriteShuffle,  [BWPort5], 1, [1], 1, 5>; // Vector shuffles.
+defm : BWWriteResPair<WriteShuffleY, [BWPort5], 1, [1], 1, 6>; // Vector shuffles (YMM/ZMM).
 defm : BWWriteResPair<WriteVarShuffle, [BWPort5], 1, [1], 1, 5>; // Vector variable shuffles.
-defm : BWWriteResPair<WriteBlend,  [BWPort5],  1>; // Vector blends.
+defm : BWWriteResPair<WriteVarShuffleY,[BWPort5], 1, [1], 1, 6>; // Vector variable shuffles (YMM/ZMM).
+defm : BWWriteResPair<WriteBlend,  [BWPort5], 1, [1], 1, 5>; // Vector blends.
+defm : BWWriteResPair<WriteBlendY, [BWPort5], 1, [1], 1, 6>; // Vector blends (YMM/ZMM).
 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 : BWWriteResPair<WriteMPSAD,  [BWPort0, BWPort5], 7, [1, 2], 3, 5>; // Vector MPSAD.
 defm : BWWriteResPair<WritePSADBW,   [BWPort0],   5>; // Vector PSADBW.
 defm : BWWriteResPair<WritePHMINPOS, [BWPort0],   5>; // Vector PHMINPOS.
@@ -1079,30 +1083,6 @@
                                             "FCOMP32m",
                                             "FCOMP64m")>;
 
-def BWWriteResGroup75 : SchedWriteRes<[BWPort5,BWPort23]> {
-  let Latency = 7;
-  let NumMicroOps = 2;
-  let ResourceCycles = [1,1];
-}
-def: InstRW<[BWWriteResGroup75], (instregex "VPACKSSDWYrm",
-                                            "VPACKSSWBYrm",
-                                            "VPACKUSDWYrm",
-                                            "VPACKUSWBYrm",
-                                            "VPALIGNRYrmi",
-                                            "VPBLENDWYrmi",
-                                            "VPSHUFBYrm",
-                                            "VPSHUFDYmi",
-                                            "VPSHUFHWYmi",
-                                            "VPSHUFLWYmi",
-                                            "VPUNPCKHBWYrm",
-                                            "VPUNPCKHDQYrm",
-                                            "VPUNPCKHQDQYrm",
-                                            "VPUNPCKHWDYrm",
-                                            "VPUNPCKLBWYrm",
-                                            "VPUNPCKLDQYrm",
-                                            "VPUNPCKLQDQYrm",
-                                            "VPUNPCKLWDYrm")>;
-
 def BWWriteResGroup76 : SchedWriteRes<[BWPort23,BWPort15]> {
   let Latency = 7;
   let NumMicroOps = 2;
@@ -1293,7 +1273,6 @@
 }
 def: InstRW<[BWWriteResGroup94], (instregex "VMASKMOVPDYrm",
                                             "VMASKMOVPSYrm",
-                                            "VPBLENDVBYrm",
                                             "VPMASKMOVDYrm",
                                             "VPMASKMOVQYrm")>;
 
diff --git a/llvm/lib/Target/X86/X86SchedHaswell.td b/llvm/lib/Target/X86/X86SchedHaswell.td
index 8473c65..aec362f 100644
--- a/llvm/lib/Target/X86/X86SchedHaswell.td
+++ b/llvm/lib/Target/X86/X86SchedHaswell.td
@@ -201,11 +201,15 @@
 defm : HWWriteResPair<WriteVecIMul,  [HWPort0],   5>;
 defm : HWWriteResPair<WritePMULLD,   [HWPort0], 10, [2], 2, 6>;
 defm : HWWriteResPair<WriteShuffle,  [HWPort5],  1, [1], 1, 5>;
-defm : HWWriteResPair<WriteVarShuffle,[HWPort5], 1, [1], 1, 5>;
+defm : HWWriteResPair<WriteShuffleY, [HWPort5],  1, [1], 1, 7>;
+defm : HWWriteResPair<WriteVarShuffle, [HWPort5], 1, [1], 1, 6>;
+defm : HWWriteResPair<WriteVarShuffleY,[HWPort5], 1, [1], 1, 7>;
 defm : HWWriteResPair<WriteBlend,  [HWPort5], 1, [1], 1, 6>;
+defm : HWWriteResPair<WriteBlendY, [HWPort5], 1, [1], 1, 7>;
 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<WriteVarVecShift,  [HWPort0, HWPort5], 2, [2, 1]>;
 defm : HWWriteResPair<WriteMPSAD,  [HWPort0, HWPort5], 7, [1, 2], 3, 6>;
 defm : HWWriteResPair<WritePSADBW, [HWPort0], 5>;
@@ -880,7 +884,6 @@
                                             "(V?)PACKUSDWrm",
                                             "(V?)PACKUSWBrm",
                                             "(V?)PALIGNRrmi",
-                                            "(V?)PSHUFBrm",
                                             "(V?)PSHUFDmi",
                                             "(V?)PSHUFHWmi",
                                             "(V?)PSHUFLWmi",
@@ -898,27 +901,16 @@
   let NumMicroOps = 2;
   let ResourceCycles = [1,1];
 }
-def: InstRW<[HWWriteResGroup13_1], (instregex "VPACKSSDWYrm",
-                                              "VPACKSSWBYrm",
-                                              "VPACKUSDWYrm",
-                                              "VPACKUSWBYrm",
-                                              "VPALIGNRYrmi",
-                                              "VPBLENDWYrmi",
-                                              "VPMOVSXBDYrm",
+def: InstRW<[HWWriteResGroup13_1], (instregex "VPMOVSXBDYrm",
                                               "VPMOVSXBQYrm",
-                                              "VPMOVSXWQYrm",
-                                              "VPSHUFBYrm",
-                                              "VPSHUFDYmi",
-                                              "VPSHUFHWYmi",
-                                              "VPSHUFLWYmi",
-                                              "VPUNPCKHBWYrm",
-                                              "VPUNPCKHDQYrm",
-                                              "VPUNPCKHQDQYrm",
-                                              "VPUNPCKHWDYrm",
-                                              "VPUNPCKLBWYrm",
-                                              "VPUNPCKLDQYrm",
-                                              "VPUNPCKLQDQYrm",
-                                              "VPUNPCKLWDYrm")>;
+                                              "VPMOVSXWQYrm")>;
+
+def HWWriteResGroup13_2 : SchedWriteRes<[HWPort5,HWPort23]> {
+  let Latency = 6;
+  let NumMicroOps = 2;
+  let ResourceCycles = [1,1];
+}
+def: InstRW<[HWWriteResGroup13_2], (instregex "MMX_PSHUFBrm")>;
 
 def HWWriteResGroup14 : SchedWriteRes<[HWPort6,HWPort23]> {
   let Latency = 6;
@@ -1244,7 +1236,6 @@
 }
 def: InstRW<[HWWriteResGroup36_1], (instregex "VMASKMOVPDYrm",
                                               "VMASKMOVPSYrm",
-                                              "VPBLENDVBYrm",
                                               "VPMASKMOVDYrm",
                                               "VPMASKMOVQYrm")>;
 
diff --git a/llvm/lib/Target/X86/X86SchedSandyBridge.td b/llvm/lib/Target/X86/X86SchedSandyBridge.td
index 0b0817c..8923eb1 100644
--- a/llvm/lib/Target/X86/X86SchedSandyBridge.td
+++ b/llvm/lib/Target/X86/X86SchedSandyBridge.td
@@ -181,9 +181,13 @@
 defm : SBWriteResPair<WriteVecIMul,  [SBPort0], 5>;
 defm : SBWriteResPair<WritePMULLD,   [SBPort0], 5, [1], 1, 6>; // TODO this is probably wrong for 256/512-bit for the "generic" model
 defm : SBWriteResPair<WriteShuffle,  [SBPort5], 1, [1], 1, 5>;
-defm : SBWriteResPair<WriteVarShuffle,  [SBPort15], 1, [1], 1, 5>;
+defm : SBWriteResPair<WriteShuffleY, [SBPort5], 1, [1], 1, 7>;
+defm : SBWriteResPair<WriteVarShuffle,  [SBPort15], 1, [1], 1, 6>;
+defm : SBWriteResPair<WriteVarShuffleY, [SBPort15], 1, [1], 1, 7>;
 defm : SBWriteResPair<WriteBlend,   [SBPort15], 1, [1], 1, 6>;
+defm : SBWriteResPair<WriteBlendY,  [SBPort15], 1, [1], 1, 7>;
 defm : SBWriteResPair<WriteVarBlend, [SBPort15], 2, [2], 2, 6>;
+defm : SBWriteResPair<WriteVarBlendY,[SBPort15], 2, [2], 2, 7>;
 defm : SBWriteResPair<WriteMPSAD, [SBPort0, SBPort15], 7, [1,2], 3, 6>;
 defm : SBWriteResPair<WritePSADBW,  [SBPort0], 5>;
 defm : SBWriteResPair<WritePHMINPOS,  [SBPort0], 5, [1], 1, 6>;
@@ -884,6 +888,7 @@
 }
 def: InstRW<[SBWriteResGroup51], (instregex "MMX_PABS(B|D|W)rm",
                                             "MMX_PALIGNRrmi",
+                                            "MMX_PSHUFBrm",
                                             "MMX_PSIGN(B|D|W)rm")>;
 
 def SBWriteResGroup52 : SchedWriteRes<[SBPort23,SBPort015]> {
@@ -995,7 +1000,6 @@
                                             "(V?)PMOVZXDQrm",
                                             "(V?)PMOVZXWDrm",
                                             "(V?)PMOVZXWQrm",
-                                            "(V?)PSHUFBrm",
                                             "(V?)PSHUFDmi",
                                             "(V?)PSHUFHWmi",
                                             "(V?)PSHUFLWmi",
diff --git a/llvm/lib/Target/X86/X86SchedSkylakeClient.td b/llvm/lib/Target/X86/X86SchedSkylakeClient.td
index b76369d..37ad168 100644
--- a/llvm/lib/Target/X86/X86SchedSkylakeClient.td
+++ b/llvm/lib/Target/X86/X86SchedSkylakeClient.td
@@ -202,9 +202,13 @@
 defm : SKLWriteResPair<WriteVecIMul,  [SKLPort0],   5>; // Vector integer multiply.
 defm : SKLWriteResPair<WritePMULLD,   [SKLPort01], 10, [2], 2, 6>;
 defm : SKLWriteResPair<WriteShuffle,  [SKLPort5], 1, [1], 1, 5>; // Vector shuffles.
-defm : SKLWriteResPair<WriteVarShuffle,  [SKLPort5], 1, [1], 1, 5>; // Vector shuffles.
+defm : SKLWriteResPair<WriteShuffleY, [SKLPort5], 1, [1], 1, 7>; // Vector shuffles (YMM/ZMM).
+defm : SKLWriteResPair<WriteVarShuffle,  [SKLPort5], 1, [1], 1, 6>; // Vector shuffles.
+defm : SKLWriteResPair<WriteVarShuffleY, [SKLPort5], 1, [1], 1, 7>; // Vector shuffles (YMM/ZMM).
 defm : SKLWriteResPair<WriteBlend,  [SKLPort5], 1, [1], 1, 6>; // Vector blends.
+defm : SKLWriteResPair<WriteBlendY, [SKLPort5], 1, [1], 1, 7>; // Vector blends (YMM/ZMM).
 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<WriteMPSAD,  [SKLPort5], 4, [2], 2, 6>; // Vector MPSAD.
 defm : SKLWriteResPair<WritePSADBW, [SKLPort5], 3>; // Vector PSADBW.
 defm : SKLWriteResPair<WritePHMINPOS, [SKLPort01], 4, [1], 1, 6>; // Vector PHMINPOS.
@@ -1268,7 +1272,6 @@
                                              "(V?)PALIGNRrmi",
                                              "VPBROADCASTBrm",
                                              "VPBROADCASTWrm",
-                                             "(V?)PSHUFBrm",
                                              "(V?)PSHUFDmi",
                                              "(V?)PSHUFHWmi",
                                              "(V?)PSHUFLWmi",
@@ -1281,6 +1284,13 @@
                                              "(V?)PUNPCKLQDQrm",
                                              "(V?)PUNPCKLWDrm")>;
 
+def SKLWriteResGroup88a : SchedWriteRes<[SKLPort5,SKLPort23]> {
+  let Latency = 6;
+  let NumMicroOps = 2;
+  let ResourceCycles = [1,1];
+}
+def: InstRW<[SKLWriteResGroup88a], (instregex "MMX_PSHUFBrm")>;
+
 def SKLWriteResGroup89 : SchedWriteRes<[SKLPort5,SKLPort01]> {
   let Latency = 7;
   let NumMicroOps = 2;
@@ -1496,29 +1506,11 @@
                                               "FCOM64m",
                                               "FCOMP32m",
                                               "FCOMP64m",
-                                              "VPACKSSDWYrm",
-                                              "VPACKSSWBYrm",
-                                              "VPACKUSDWYrm",
-                                              "VPACKUSWBYrm",
-                                              "VPALIGNRYrmi",
-                                              "VPBLENDWYrmi",
                                               "VPBROADCASTBYrm",
                                               "VPBROADCASTWYrm",
                                               "VPMOVSXBDYrm",
                                               "VPMOVSXBQYrm",
-                                              "VPMOVSXWQYrm",
-                                              "VPSHUFBYrm",
-                                              "VPSHUFDYmi",
-                                              "VPSHUFHWYmi",
-                                              "VPSHUFLWYmi",
-                                              "VPUNPCKHBWYrm",
-                                              "VPUNPCKHDQYrm",
-                                              "VPUNPCKHQDQYrm",
-                                              "VPUNPCKHWDYrm",
-                                              "VPUNPCKLBWYrm",
-                                              "VPUNPCKLDQYrm",
-                                              "VPUNPCKLQDQYrm",
-                                              "VPUNPCKLWDYrm")>;
+                                              "VPMOVSXWQYrm")>;
 
 def SKLWriteResGroup109 : SchedWriteRes<[SKLPort01,SKLPort23]> {
   let Latency = 8;
diff --git a/llvm/lib/Target/X86/X86SchedSkylakeServer.td b/llvm/lib/Target/X86/X86SchedSkylakeServer.td
index 667b519..4ad0364 100755
--- a/llvm/lib/Target/X86/X86SchedSkylakeServer.td
+++ b/llvm/lib/Target/X86/X86SchedSkylakeServer.td
@@ -202,9 +202,13 @@
 defm : SKXWriteResPair<WriteVecIMul,  [SKXPort0],   5>; // Vector integer multiply.
 defm : SKXWriteResPair<WritePMULLD,   [SKXPort015], 10, [2], 2, 6>; // Vector integer multiply.
 defm : SKXWriteResPair<WriteShuffle,  [SKXPort5], 1, [1], 1, 5>; // Vector shuffles.
-defm : SKXWriteResPair<WriteVarShuffle,  [SKXPort5], 1, [1], 1, 5>; // Vector variable shuffles.
+defm : SKXWriteResPair<WriteShuffleY, [SKXPort5], 1, [1], 1, 7>; // Vector shuffles (YMM/ZMM).
+defm : SKXWriteResPair<WriteVarShuffle,  [SKXPort5], 1, [1], 1, 6>; // Vector variable shuffles.
+defm : SKXWriteResPair<WriteVarShuffleY, [SKXPort5], 1, [1], 1, 7>; // Vector variable shuffles (YMM/ZMM).
 defm : SKXWriteResPair<WriteBlend, [SKXPort5], 1, [1], 1, 6>; // Vector blends.
+defm : SKXWriteResPair<WriteBlendY,[SKXPort5], 1, [1], 1, 7>; // Vector blends (YMM/ZMM).
 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<WriteMPSAD,  [SKXPort5], 4, [2], 2, 6>; // Vector MPSAD.
 defm : SKXWriteResPair<WritePSADBW, [SKXPort5], 3, [1], 1, 6>; // Vector PSADBW.
 defm : SKXWriteResPair<WritePHMINPOS, [SKXPort015], 4, [1], 1, 6>; // Vector PHMINPOS.
@@ -2167,8 +2171,6 @@
                                              "VPBROADCASTBrm",
                                              "VPBROADCASTWZ128m(b?)",
                                              "VPBROADCASTWrm",
-                                             "VPSHUFBZ128rm(b?)",
-                                             "(V?)PSHUFBrm",
                                              "VPSHUFDZ128m(b?)i",
                                              "(V?)PSHUFDmi",
                                              "VPSHUFHWZ128mi(b?)",
@@ -2194,6 +2196,13 @@
                                              "VPUNPCKLWDZ128rm(b?)",
                                              "(V?)PUNPCKLWDrm")>;
 
+def SKXWriteResGroup92a : SchedWriteRes<[SKXPort5,SKXPort23]> {
+  let Latency = 6;
+  let NumMicroOps = 2;
+  let ResourceCycles = [1,1];
+}
+def: InstRW<[SKXWriteResGroup92a], (instregex "MMX_PSHUFBrm")>;
+
 def SKXWriteResGroup93 : SchedWriteRes<[SKXPort5,SKXPort015]> {
   let Latency = 7;
   let NumMicroOps = 2;
@@ -2629,22 +2638,6 @@
                                               "FCOMP64m",
                                               "MMX_PSADBWirm",
                                               "VFPCLASSSDrm(b?)",
-                                              "VPACKSSDWYrm",
-                                              "VPACKSSDWZ256rm(b?)",
-                                              "VPACKSSDWZrm(b?)",
-                                              "VPACKSSWBYrm",
-                                              "VPACKSSWBZ256rm(b?)",
-                                              "VPACKSSWBZrm(b?)",
-                                              "VPACKUSDWYrm",
-                                              "VPACKUSDWZ256rm(b?)",
-                                              "VPACKUSDWZrm(b?)",
-                                              "VPACKUSWBYrm",
-                                              "VPACKUSWBZ256rm(b?)",
-                                              "VPACKUSWBZrm(b?)",
-                                              "VPALIGNRYrmi",
-                                              "VPALIGNRZ256rmi(b?)",
-                                              "VPALIGNRZrmi(b?)",
-                                              "VPBLENDWYrmi",
                                               "VPBROADCASTBYrm",
                                               "VPBROADCASTBZ256m(b?)",
                                               "VPBROADCASTBZm(b?)",
@@ -2653,47 +2646,7 @@
                                               "VPBROADCASTWZm(b?)",
                                               "VPMOVSXBDYrm",
                                               "VPMOVSXBQYrm",
-                                              "VPMOVSXWQYrm",
-                                              "VPSHUFBYrm",
-                                              "VPSHUFBZ256rm(b?)",
-                                              "VPSHUFBZrm(b?)",
-                                              "VPSHUFDYmi",
-                                              "VPSHUFDZ256m(b?)i",
-                                              "VPSHUFDZm(b?)i",
-                                              "VPSHUFHWYmi",
-                                              "VPSHUFHWZ256mi(b?)",
-                                              "VPSHUFHWZmi(b?)",
-                                              "VPSHUFLWYmi",
-                                              "VPSHUFLWZ256mi(b?)",
-                                              "VPSHUFLWZmi(b?)",
-                                              "VPSLLDQZ256rm(b?)",
-                                              "VPSLLDQZrm(b?)",
-                                              "VPSRLDQZ256rm(b?)",
-                                              "VPSRLDQZrm(b?)",
-                                              "VPUNPCKHBWYrm",
-                                              "VPUNPCKHBWZ256rm(b?)",
-                                              "VPUNPCKHBWZrm(b?)",
-                                              "VPUNPCKHDQYrm",
-                                              "VPUNPCKHDQZ256rm(b?)",
-                                              "VPUNPCKHDQZrm(b?)",
-                                              "VPUNPCKHQDQYrm",
-                                              "VPUNPCKHQDQZ256rm(b?)",
-                                              "VPUNPCKHQDQZrm(b?)",
-                                              "VPUNPCKHWDYrm",
-                                              "VPUNPCKHWDZ256rm(b?)",
-                                              "VPUNPCKHWDZrm(b?)",
-                                              "VPUNPCKLBWYrm",
-                                              "VPUNPCKLBWZ256rm(b?)",
-                                              "VPUNPCKLBWZrm(b?)",
-                                              "VPUNPCKLDQYrm",
-                                              "VPUNPCKLDQZ256rm(b?)",
-                                              "VPUNPCKLDQZrm(b?)",
-                                              "VPUNPCKLQDQYrm",
-                                              "VPUNPCKLQDQZ256rm(b?)",
-                                              "VPUNPCKLQDQZrm(b?)",
-                                              "VPUNPCKLWDYrm",
-                                              "VPUNPCKLWDZ256rm(b?)",
-                                              "VPUNPCKLWDZrm(b?)")>;
+                                              "VPMOVSXWQYrm")>;
 
 def SKXWriteResGroup120 : SchedWriteRes<[SKXPort01,SKXPort23]> {
   let Latency = 8;
diff --git a/llvm/lib/Target/X86/X86Schedule.td b/llvm/lib/Target/X86/X86Schedule.td
index 433f5b1..a264086 100644
--- a/llvm/lib/Target/X86/X86Schedule.td
+++ b/llvm/lib/Target/X86/X86Schedule.td
@@ -139,9 +139,13 @@
 defm WriteVecIMul  : X86SchedWritePair; // Vector integer multiply.
 defm WritePMULLD : X86SchedWritePair; // PMULLD
 defm WriteShuffle  : X86SchedWritePair; // Vector shuffles.
+defm WriteShuffleY : X86SchedWritePair; // Vector shuffles (YMM/ZMM).
 defm WriteVarShuffle  : X86SchedWritePair; // Vector variable shuffles.
+defm WriteVarShuffleY : X86SchedWritePair; // Vector variable shuffles (YMM/ZMM).
 defm WriteBlend  : X86SchedWritePair; // Vector blends.
+defm WriteBlendY : X86SchedWritePair; // Vector blends (YMM/ZMM).
 defm WriteVarBlend  : X86SchedWritePair; // Vector variable blends.
+defm WriteVarBlendY : X86SchedWritePair; // Vector variable blends (YMM/ZMM).
 defm WritePSADBW : X86SchedWritePair; // Vector PSADBW.
 defm WriteMPSAD : X86SchedWritePair; // Vector MPSAD.
 defm WritePHMINPOS : X86SchedWritePair; // Vector PHMINPOS.
@@ -264,15 +268,15 @@
 
 def SchedWriteShuffle
  : X86SchedWriteWidths<WriteShuffle, WriteShuffle,
-                       WriteShuffle, WriteShuffle>;
+                       WriteShuffleY, WriteShuffleY>;
 def SchedWriteVarShuffle
  : X86SchedWriteWidths<WriteVarShuffle, WriteVarShuffle,
-                       WriteVarShuffle, WriteVarShuffle>;
+                       WriteVarShuffleY, WriteVarShuffleY>;
 def SchedWriteBlend
- : X86SchedWriteWidths<WriteBlend, WriteBlend, WriteBlend, WriteBlend>;
+ : X86SchedWriteWidths<WriteBlend, WriteBlend, WriteBlendY, WriteBlendY>;
 def SchedWriteVarBlend
  : X86SchedWriteWidths<WriteVarBlend, WriteVarBlend,
-                       WriteVarBlend, WriteVarBlend>;
+                       WriteVarBlendY, WriteVarBlendY>;
 
 //===----------------------------------------------------------------------===//
 // Generic Processor Scheduler Models.
diff --git a/llvm/lib/Target/X86/X86ScheduleAtom.td b/llvm/lib/Target/X86/X86ScheduleAtom.td
index d022266..3eeeec0 100644
--- a/llvm/lib/Target/X86/X86ScheduleAtom.td
+++ b/llvm/lib/Target/X86/X86ScheduleAtom.td
@@ -261,9 +261,13 @@
 defm : AtomWriteResPair<WriteMPSAD,        [AtomPort01],  [AtomPort0], 1, 1>;
 defm : AtomWriteResPair<WritePSADBW,        [AtomPort0],  [AtomPort0], 5, 5, [5], [5]>;
 defm : AtomWriteResPair<WriteShuffle,       [AtomPort0],  [AtomPort0], 1, 1>;
+defm : AtomWriteResPair<WriteShuffleY,      [AtomPort0],  [AtomPort0], 1, 1>;
 defm : AtomWriteResPair<WriteVarShuffle,   [AtomPort01], [AtomPort01], 4, 5, [4], [5]>;
+defm : AtomWriteResPair<WriteVarShuffleY,  [AtomPort01], [AtomPort01], 4, 5, [4], [5]>;
 defm : AtomWriteResPair<WriteBlend,         [AtomPort0],  [AtomPort0]>; // NOTE: Doesn't exist on Atom.
+defm : AtomWriteResPair<WriteBlendY,        [AtomPort0],  [AtomPort0]>; // NOTE: Doesn't exist on Atom.
 defm : AtomWriteResPair<WriteVarBlend,      [AtomPort0],  [AtomPort0]>; // NOTE: Doesn't exist on Atom.
+defm : AtomWriteResPair<WriteVarBlendY,     [AtomPort0],  [AtomPort0]>; // NOTE: Doesn't exist on Atom.
 defm : AtomWriteResPair<WriteShuffle256,    [AtomPort0],  [AtomPort0]>; // NOTE: Doesn't exist on Atom.
 defm : AtomWriteResPair<WriteVarShuffle256, [AtomPort0],  [AtomPort0]>; // NOTE: Doesn't exist on Atom.
 defm : AtomWriteResPair<WriteVarVecShift,   [AtomPort0],  [AtomPort0]>; // NOTE: Doesn't exist on Atom.
diff --git a/llvm/lib/Target/X86/X86ScheduleBtVer2.td b/llvm/lib/Target/X86/X86ScheduleBtVer2.td
index 7d32bc6..9740722 100644
--- a/llvm/lib/Target/X86/X86ScheduleBtVer2.td
+++ b/llvm/lib/Target/X86/X86ScheduleBtVer2.td
@@ -411,9 +411,13 @@
 defm : JWriteResFpuPair<WritePSADBW,      [JFPU01, JVALU], 2>;
 defm : JWriteResFpuPair<WritePHMINPOS,    [JFPU0,  JVALU], 2>;
 defm : JWriteResFpuPair<WriteShuffle,     [JFPU01, JVALU], 1>;
+defm : JWriteResFpuPair<WriteShuffleY,    [JFPU01, JVALU], 1>;
 defm : JWriteResFpuPair<WriteVarShuffle,  [JFPU01, JVALU], 2, [1, 4], 3>;
+defm : JWriteResFpuPair<WriteVarShuffleY, [JFPU01, JVALU], 2, [1, 4], 3>;
 defm : JWriteResFpuPair<WriteBlend,       [JFPU01, JVALU], 1>;
+defm : JWriteResFpuPair<WriteBlendY,      [JFPU01, JVALU], 1>;
 defm : JWriteResFpuPair<WriteVarBlend,    [JFPU01, JVALU], 2, [1, 4], 3>;
+defm : JWriteResFpuPair<WriteVarBlendY,   [JFPU01, JVALU], 2, [1, 4], 3>;
 defm : JWriteResFpuPair<WriteVecLogic,    [JFPU01, JVALU], 1>;
 defm : JWriteResFpuPair<WriteVecLogicY,   [JFPU01, JVALU], 1>; // NOTE: Doesn't exist on Jaguar.
 defm : JWriteResFpuPair<WriteShuffle256,  [JFPU01, JVALU], 1>;
diff --git a/llvm/lib/Target/X86/X86ScheduleSLM.td b/llvm/lib/Target/X86/X86ScheduleSLM.td
index c0df1ff..1bc522f 100644
--- a/llvm/lib/Target/X86/X86ScheduleSLM.td
+++ b/llvm/lib/Target/X86/X86ScheduleSLM.td
@@ -171,8 +171,11 @@
 //defm : SLMWriteResPair<WritePMULLD,  [SLM_FPC_RSV0],   11, [11], 7>;
 defm : SLMWriteResPair<WritePMULLD,  [SLM_FPC_RSV0],   4>;
 defm : SLMWriteResPair<WriteShuffle,  [SLM_FPC_RSV0],  1>;
+defm : SLMWriteResPair<WriteShuffleY, [SLM_FPC_RSV0],  1>;
 defm : SLMWriteResPair<WriteVarShuffle,  [SLM_FPC_RSV0],  1>;
+defm : SLMWriteResPair<WriteVarShuffleY, [SLM_FPC_RSV0],  1>;
 defm : SLMWriteResPair<WriteBlend,  [SLM_FPC_RSV0],  1>;
+defm : SLMWriteResPair<WriteBlendY, [SLM_FPC_RSV0],  1>;
 defm : SLMWriteResPair<WriteMPSAD,  [SLM_FPC_RSV0],  7>;
 defm : SLMWriteResPair<WritePSADBW, [SLM_FPC_RSV0],  4>;
 defm : SLMWriteResPair<WritePHMINPOS,  [SLM_FPC_RSV0],   4>;
@@ -290,6 +293,7 @@
 def  : WriteRes<WriteIMulH, [SLM_FPC_RSV0]>;
 defm : SLMWriteResPair<WriteFBlendY, [SLM_FPC_RSV0],  1>;
 defm : SLMWriteResPair<WriteVarBlend, [SLM_FPC_RSV0], 1>;
+defm : SLMWriteResPair<WriteVarBlendY,[SLM_FPC_RSV0], 1>;
 defm : SLMWriteResPair<WriteFVarBlend, [SLM_FPC_RSV0], 1>;
 defm : SLMWriteResPair<WriteFVarBlendY, [SLM_FPC_RSV0], 1>;
 defm : SLMWriteResPair<WriteFShuffle256, [SLM_FPC_RSV0],  1>;
diff --git a/llvm/lib/Target/X86/X86ScheduleZnver1.td b/llvm/lib/Target/X86/X86ScheduleZnver1.td
index e8230d8..7c70e34 100644
--- a/llvm/lib/Target/X86/X86ScheduleZnver1.td
+++ b/llvm/lib/Target/X86/X86ScheduleZnver1.td
@@ -203,6 +203,7 @@
 defm : ZnWriteResFpuPair<WriteFVarBlend, [ZnFPU01], 1>;
 defm : ZnWriteResFpuPair<WriteFVarBlendY,[ZnFPU01], 1>;
 defm : ZnWriteResFpuPair<WriteVarBlend,  [ZnFPU0],  1>;
+defm : ZnWriteResFpuPair<WriteVarBlendY, [ZnFPU0],  1>;
 defm : ZnWriteResFpuPair<WriteCvtI2F,    [ZnFPU3],  5>;
 defm : ZnWriteResFpuPair<WriteCvtF2F,    [ZnFPU3],  5>;
 defm : ZnWriteResFpuPair<WriteCvtF2I,    [ZnFPU3],  5>;
@@ -241,8 +242,11 @@
 defm : ZnWriteResFpuPair<WriteVecIMul,    [ZnFPU0],  4>;
 defm : ZnWriteResFpuPair<WritePMULLD,     [ZnFPU0],  4>; // FIXME
 defm : ZnWriteResFpuPair<WriteShuffle,    [ZnFPU],   1>;
+defm : ZnWriteResFpuPair<WriteShuffleY,   [ZnFPU],   1>;
 defm : ZnWriteResFpuPair<WriteVarShuffle, [ZnFPU],   1>;
+defm : ZnWriteResFpuPair<WriteVarShuffleY,[ZnFPU],   1>;
 defm : ZnWriteResFpuPair<WriteBlend,      [ZnFPU01], 1>;
+defm : ZnWriteResFpuPair<WriteBlendY,     [ZnFPU01], 1>;
 defm : ZnWriteResFpuPair<WriteShuffle256, [ZnFPU],   2>;
 defm : ZnWriteResFpuPair<WriteVarShuffle256, [ZnFPU],   2>;
 defm : ZnWriteResFpuPair<WritePSADBW,     [ZnFPU0],  3>;