diff --git a/lib/Target/X86/X86InstrSSE.td b/lib/Target/X86/X86InstrSSE.td
index 32358a3..fe66d8b 100644
--- a/lib/Target/X86/X86InstrSSE.td
+++ b/lib/Target/X86/X86InstrSSE.td
@@ -704,6 +704,24 @@
                  !strconcat(OpcodeStr, "sd\t{$src2, $dst|$dst, $src2}"),
                  [(set FR64:$dst, (OpNode FR64:$src1, (load addr:$src2)))]>;
 
+  def V#NAME#SSrm : VSSI<opc, MRMSrcMem, (outs FR32:$dst),
+                 (ins FR32:$src1, f32mem:$src2),
+                 !strconcat(OpcodeStr,
+                            "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
+                 []> {
+    let Constraints = "";
+    let isAsmParserOnly = 1;
+  }
+
+  def V#NAME#SDrm : VSDI<opc, MRMSrcMem, (outs FR64:$dst),
+                 (ins FR64:$src1, f64mem:$src2),
+                 !strconcat(OpcodeStr,
+                            "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
+                 []> {
+    let Constraints = "";
+    let isAsmParserOnly = 1;
+  }
+
   // Vector operation, reg+reg.
   def PSrr : PSI<opc, MRMSrcReg, (outs VR128:$dst),
                                  (ins VR128:$src1, VR128:$src2),
diff --git a/lib/Target/X86/X86MCCodeEmitter.cpp b/lib/Target/X86/X86MCCodeEmitter.cpp
index d105b5d..9a32b6ea0 100644
--- a/lib/Target/X86/X86MCCodeEmitter.cpp
+++ b/lib/Target/X86/X86MCCodeEmitter.cpp
@@ -349,6 +349,13 @@
   //
   unsigned char VEX_R = 0x1;
 
+  // VEX_X: equivalent to REX.X, only used when a
+  // register is used for index in SIB Byte.
+  //
+  //  1: Same as REX.X=0 (must be 1 in 32-bit mode)
+  //  0: Same as REX.X=1 (64-bit mode only)
+  unsigned char VEX_X = 0x1;
+
   // VEX_B:
   //
   //  1: Same as REX_B=0 (ignored in 32-bit mode)
@@ -415,9 +422,12 @@
   unsigned NumOps = MI.getNumOperands();
   unsigned i = 0;
   unsigned SrcReg = 0, SrcRegNum = 0;
+  bool IsSrcMem = false;
 
   switch (TSFlags & X86II::FormMask) {
   case X86II::MRMInitReg: assert(0 && "FIXME: Remove this!");
+  case X86II::MRMSrcMem:
+    IsSrcMem = true;
   case X86II::MRMSrcReg:
     if (MI.getOperand(0).isReg() &&
         X86InstrInfo::isX86_64ExtendedReg(MI.getOperand(0).getReg()))
@@ -447,6 +457,9 @@
       const MCOperand &MO = MI.getOperand(i);
       if (MO.isReg() && X86InstrInfo::isX86_64ExtendedReg(MO.getReg()))
         VEX_B = 0x0;
+      if (!VEX_B && MO.isReg() && IsSrcMem &&
+          X86InstrInfo::isX86_64ExtendedReg(MO.getReg()))
+        VEX_X = 0x0;
     }
     break;
   default:
@@ -464,11 +477,9 @@
   //    | C5h | | R | vvvv | L | pp |
   //    +-----+ +-------------------+
   //
-  // Note: VEX.X isn't used so far
-  //
   unsigned char LastByte = VEX_PP | (VEX_L << 2) | (VEX_4V << 3);
 
-  if (VEX_B /* & VEX_X */) { // 2 byte VEX prefix
+  if (VEX_B && VEX_X) { // 2 byte VEX prefix
     EmitByte(0xC5, CurByte, OS);
     EmitByte(LastByte | (VEX_R << 7), CurByte, OS);
     return;
@@ -476,7 +487,7 @@
 
   // 3 byte VEX prefix
   EmitByte(0xC4, CurByte, OS);
-  EmitByte(VEX_R << 7 | 1 << 6 /* VEX_X = 1 */ | VEX_5M, CurByte, OS);
+  EmitByte(VEX_R << 7 | VEX_X << 6 | VEX_5M, CurByte, OS);
   EmitByte(LastByte | (VEX_W << 7), CurByte, OS);
 }
 
@@ -751,8 +762,13 @@
       AddrOperands = X86AddrNumOperands - 1; // No segment register
     else
       AddrOperands = X86AddrNumOperands;
-    
-    EmitMemModRMByte(MI, CurOp+1, GetX86RegNum(MI.getOperand(CurOp)),
+
+    if (IsAVXForm)
+      AddrOperands++;
+
+    // Skip the register source (which is encoded in VEX_VVVV)
+    EmitMemModRMByte(MI, IsAVXForm ? CurOp+2 : CurOp+1,
+                     GetX86RegNum(MI.getOperand(CurOp)),
                      TSFlags, CurByte, OS, Fixups);
     CurOp += AddrOperands + 1;
     break;
diff --git a/test/MC/AsmParser/X86/x86_32-encoding.s b/test/MC/AsmParser/X86/x86_32-encoding.s
index 7f79b9d..237f67b 100644
--- a/test/MC/AsmParser/X86/x86_32-encoding.s
+++ b/test/MC/AsmParser/X86/x86_32-encoding.s
@@ -10084,3 +10084,36 @@
 // CHECK: vdivsd  %xmm4, %xmm6, %xmm2
 // CHECK:  encoding: [0xc5,0xcb,0x5e,0xd4]
           vdivsd  %xmm4, %xmm6, %xmm2
+
+// CHECK: vaddss  3735928559(%ebx,%ecx,8), %xmm2, %xmm5
+// CHECK:  encoding: [0xc5,0xea,0x58,0xac,0xcb,0xef,0xbe,0xad,0xde]
+          vaddss  3735928559(%ebx,%ecx,8), %xmm2, %xmm5
+
+// CHECK: vsubss  3735928559(%ebx,%ecx,8), %xmm2, %xmm5
+// CHECK:  encoding: [0xc5,0xea,0x5c,0xac,0xcb,0xef,0xbe,0xad,0xde]
+          vsubss  3735928559(%ebx,%ecx,8), %xmm2, %xmm5
+
+// CHECK: vmulss  3735928559(%ebx,%ecx,8), %xmm2, %xmm5
+// CHECK:  encoding: [0xc5,0xea,0x59,0xac,0xcb,0xef,0xbe,0xad,0xde]
+          vmulss  3735928559(%ebx,%ecx,8), %xmm2, %xmm5
+
+// CHECK: vdivss  3735928559(%ebx,%ecx,8), %xmm2, %xmm5
+// CHECK:  encoding: [0xc5,0xea,0x5e,0xac,0xcb,0xef,0xbe,0xad,0xde]
+          vdivss  3735928559(%ebx,%ecx,8), %xmm2, %xmm5
+
+// CHECK: vaddsd  3735928559(%ebx,%ecx,8), %xmm2, %xmm5
+// CHECK:  encoding: [0xc5,0xeb,0x58,0xac,0xcb,0xef,0xbe,0xad,0xde]
+          vaddsd  3735928559(%ebx,%ecx,8), %xmm2, %xmm5
+
+// CHECK: vsubsd  3735928559(%ebx,%ecx,8), %xmm2, %xmm5
+// CHECK:  encoding: [0xc5,0xeb,0x5c,0xac,0xcb,0xef,0xbe,0xad,0xde]
+          vsubsd  3735928559(%ebx,%ecx,8), %xmm2, %xmm5
+
+// CHECK: vmulsd  3735928559(%ebx,%ecx,8), %xmm2, %xmm5
+// CHECK:  encoding: [0xc5,0xeb,0x59,0xac,0xcb,0xef,0xbe,0xad,0xde]
+          vmulsd  3735928559(%ebx,%ecx,8), %xmm2, %xmm5
+
+// CHECK: vdivsd  3735928559(%ebx,%ecx,8), %xmm2, %xmm5
+// CHECK:  encoding: [0xc5,0xeb,0x5e,0xac,0xcb,0xef,0xbe,0xad,0xde]
+          vdivsd  3735928559(%ebx,%ecx,8), %xmm2, %xmm5
+
diff --git a/test/MC/AsmParser/X86/x86_64-encoding.s b/test/MC/AsmParser/X86/x86_64-encoding.s
index b259608..592db35 100644
--- a/test/MC/AsmParser/X86/x86_64-encoding.s
+++ b/test/MC/AsmParser/X86/x86_64-encoding.s
@@ -135,3 +135,35 @@
 // CHECK: vdivsd  %xmm8, %xmm9, %xmm10
 // CHECK:  encoding: [0xc4,0x41,0x33,0x5e,0xd0]
 vdivsd  %xmm8, %xmm9, %xmm10
+
+// CHECK:   vaddss  -4(%rcx,%rbx,8), %xmm10, %xmm11
+// CHECK:   encoding: [0xc5,0x2a,0x58,0x5c,0xd9,0xfc]
+vaddss  -4(%rcx,%rbx,8), %xmm10, %xmm11
+
+// CHECK:   vsubss  -4(%rcx,%rbx,8), %xmm10, %xmm11
+// CHECK:   encoding: [0xc5,0x2a,0x5c,0x5c,0xd9,0xfc]
+vsubss  -4(%rcx,%rbx,8), %xmm10, %xmm11
+
+// CHECK:   vmulss  -4(%rcx,%rbx,8), %xmm10, %xmm11
+// CHECK:   encoding: [0xc5,0x2a,0x59,0x5c,0xd9,0xfc]
+vmulss  -4(%rcx,%rbx,8), %xmm10, %xmm11
+
+// CHECK:   vdivss  -4(%rcx,%rbx,8), %xmm10, %xmm11
+// CHECK:   encoding: [0xc5,0x2a,0x5e,0x5c,0xd9,0xfc]
+vdivss  -4(%rcx,%rbx,8), %xmm10, %xmm11
+
+// CHECK:   vaddsd  -4(%rcx,%rbx,8), %xmm10, %xmm11
+// CHECK:   encoding: [0xc5,0x2b,0x58,0x5c,0xd9,0xfc]
+vaddsd  -4(%rcx,%rbx,8), %xmm10, %xmm11
+
+// CHECK:   vsubsd  -4(%rcx,%rbx,8), %xmm10, %xmm11
+// CHECK:   encoding: [0xc5,0x2b,0x5c,0x5c,0xd9,0xfc]
+vsubsd  -4(%rcx,%rbx,8), %xmm10, %xmm11
+
+// CHECK:   vmulsd  -4(%rcx,%rbx,8), %xmm10, %xmm11
+// CHECK:   encoding: [0xc5,0x2b,0x59,0x5c,0xd9,0xfc]
+vmulsd  -4(%rcx,%rbx,8), %xmm10, %xmm11
+
+// CHECK:   vdivsd  -4(%rcx,%rbx,8), %xmm10, %xmm11
+// CHECK:   encoding: [0xc5,0x2b,0x5e,0x5c,0xd9,0xfc]
+vdivsd  -4(%rcx,%rbx,8), %xmm10, %xmm11
diff --git a/utils/TableGen/X86RecognizableInstr.cpp b/utils/TableGen/X86RecognizableInstr.cpp
index 5fe2bfe..19b51cb 100644
--- a/utils/TableGen/X86RecognizableInstr.cpp
+++ b/utils/TableGen/X86RecognizableInstr.cpp
@@ -535,7 +535,8 @@
     HANDLE_OPERAND(rmRegister)
 
     if (HasVEX_4VPrefix)
-      // FIXME: encoding of registers in AVX is in 1's complement form.
+      // FIXME: In AVX, the register below becomes the one encoded
+      // in ModRMVEX and the one above the one in the VEX.VVVV field
       HANDLE_OPTIONAL(rmRegister)
     else
       HANDLE_OPTIONAL(immediate)
@@ -547,6 +548,12 @@
     assert(numPhysicalOperands >= 2 && numPhysicalOperands <= 3 &&
            "Unexpected number of operands for MRMSrcMemFrm");
     HANDLE_OPERAND(roRegister)
+
+    if (HasVEX_4VPrefix)
+      // FIXME: In AVX, the register below becomes the one encoded
+      // in ModRMVEX and the one above the one in the VEX.VVVV field
+      HANDLE_OPTIONAL(rmRegister)
+
     HANDLE_OPERAND(memory)
     HANDLE_OPTIONAL(immediate)
     break;