3 files changed, 66 insertions, 15 deletions

diff --git a/lib/Target/X86/AsmParser/X86AsmParser.cpp b/lib/Target/X86/AsmParser/X86AsmParser.cpp
index 417842b467..8905509143 100644
--- a/lib/Target/X86/AsmParser/X86AsmParser.cpp
+++ b/lib/Target/X86/AsmParser/X86AsmParser.cpp
@@ -331,6 +331,23 @@ struct X86Operand : public MCParsedAsmOperand {
     return Kind == Memory && (!Mem.Size || Mem.Size == 256);
   }
 
+  bool isMemVX32() const {
+    return Kind == Memory && (!Mem.Size || Mem.Size == 32) &&
+      getMemIndexReg() >= X86::XMM0 && getMemIndexReg() <= X86::XMM15;
+  }
+  bool isMemVY32() const {
+    return Kind == Memory && (!Mem.Size || Mem.Size == 32) &&
+      getMemIndexReg() >= X86::YMM0 && getMemIndexReg() <= X86::YMM15;
+  }
+  bool isMemVX64() const {
+    return Kind == Memory && (!Mem.Size || Mem.Size == 64) &&
+      getMemIndexReg() >= X86::XMM0 && getMemIndexReg() <= X86::XMM15;
+  }
+  bool isMemVY64() const {
+    return Kind == Memory && (!Mem.Size || Mem.Size == 64) &&
+      getMemIndexReg() >= X86::YMM0 && getMemIndexReg() <= X86::YMM15;
+  }
+
   bool isAbsMem() const {
     return Kind == Memory && !getMemSegReg() && !getMemBaseReg() &&
       !getMemIndexReg() && getMemScale() == 1;
@@ -377,6 +394,18 @@ struct X86Operand : public MCParsedAsmOperand {
   void addMem256Operands(MCInst &Inst, unsigned N) const {
     addMemOperands(Inst, N);
   }
+  void addMemVX32Operands(MCInst &Inst, unsigned N) const {
+    addMemOperands(Inst, N);
+  }
+  void addMemVY32Operands(MCInst &Inst, unsigned N) const {
+    addMemOperands(Inst, N);
+  }
+  void addMemVX64Operands(MCInst &Inst, unsigned N) const {
+    addMemOperands(Inst, N);
+  }
+  void addMemVY64Operands(MCInst &Inst, unsigned N) const {
+    addMemOperands(Inst, N);
+  }
 
   void addMemOperands(MCInst &Inst, unsigned N) const {
     assert((N == 5) && "Invalid number of operands!");
diff --git a/lib/Target/X86/X86InstrInfo.td b/lib/Target/X86/X86InstrInfo.td
index 2c4e393556..d293156c1f 100644
--- a/lib/Target/X86/X86InstrInfo.td
+++ b/lib/Target/X86/X86InstrInfo.td
@@ -292,6 +292,20 @@ def X86Mem256AsmOperand : AsmOperandClass {
   let Name = "Mem256"; let PredicateMethod = "isMem256";
 }
 
+// Gather mem operands
+def X86MemVX32Operand : AsmOperandClass {
+  let Name = "MemVX32"; let PredicateMethod = "isMemVX32";
+}
+def X86MemVY32Operand : AsmOperandClass {
+  let Name = "MemVY32"; let PredicateMethod = "isMemVY32";
+}
+def X86MemVX64Operand : AsmOperandClass {
+  let Name = "MemVX64"; let PredicateMethod = "isMemVX64";
+}
+def X86MemVY64Operand : AsmOperandClass {
+  let Name = "MemVY64"; let PredicateMethod = "isMemVY64";
+}
+
 def X86AbsMemAsmOperand : AsmOperandClass {
   let Name = "AbsMem";
   let SuperClasses = [X86MemAsmOperand];
@@ -330,12 +344,20 @@ def f128mem : X86MemOperand<"printf128mem"> {
   let ParserMatchClass = X86Mem128AsmOperand; }
 def f256mem : X86MemOperand<"printf256mem">{ 
   let ParserMatchClass = X86Mem256AsmOperand; }
-def v128mem : X86MemOperand<"printf128mem"> {
+
+// Gather mem operands
+def vx32mem : X86MemOperand<"printi32mem">{
   let MIOperandInfo = (ops ptr_rc, i8imm, VR128, i32imm, i8imm);
-  let ParserMatchClass = X86Mem128AsmOperand; }
-def v256mem : X86MemOperand<"printf256mem"> {
+  let ParserMatchClass = X86MemVX32Operand; }
+def vy32mem : X86MemOperand<"printi32mem">{
   let MIOperandInfo = (ops ptr_rc, i8imm, VR256, i32imm, i8imm);
-  let ParserMatchClass = X86Mem256AsmOperand; }
+  let ParserMatchClass = X86MemVY32Operand; }
+def vx64mem : X86MemOperand<"printi64mem">{
+  let MIOperandInfo = (ops ptr_rc, i8imm, VR128, i32imm, i8imm);
+  let ParserMatchClass = X86MemVX64Operand; }
+def vy64mem : X86MemOperand<"printi64mem">{
+  let MIOperandInfo = (ops ptr_rc, i8imm, VR256, i32imm, i8imm);
+  let ParserMatchClass = X86MemVY64Operand; }
 }
 
 // A version of i8mem for use on x86-64 that uses GR64_NOREX instead of
diff --git a/lib/Target/X86/X86InstrSSE.td b/lib/Target/X86/X86InstrSSE.td
index e4caace00c..10cc48377c 100644
--- a/lib/Target/X86/X86InstrSSE.td
+++ b/lib/Target/X86/X86InstrSSE.td
@@ -8036,10 +8036,10 @@ defm VPSRAVD : avx2_var_shift<0x46, "vpsravd", sra, v4i32, v8i32>;
 
 //===----------------------------------------------------------------------===//
 // VGATHER - GATHER Operations
-multiclass avx2_gather<bits<8> opc, string OpcodeStr,
-                       RegisterClass RC256, X86MemOperand memop256> {
+multiclass avx2_gather<bits<8> opc, string OpcodeStr, RegisterClass RC256,
+                       X86MemOperand memop128, X86MemOperand memop256> {
   def rm  : AVX28I<opc, MRMSrcMem, (outs VR128:$dst, VR128:$mask_wb),
-            (ins VR128:$src1, v128mem:$src2, VR128:$mask),
+            (ins VR128:$src1, memop128:$src2, VR128:$mask),
             !strconcat(OpcodeStr,
               "\t{$mask, $src2, $dst|$dst, $src2, $mask}"),
             []>, VEX_4VOp3;
@@ -8051,12 +8051,12 @@ multiclass avx2_gather<bits<8> opc, string OpcodeStr,
 }
 
 let Constraints = "$src1 = $dst, $mask = $mask_wb" in {
-  defm VGATHERDPD : avx2_gather<0x92, "vgatherdpd", VR256, v128mem>, VEX_W;
-  defm VGATHERQPD : avx2_gather<0x93, "vgatherqpd", VR256, v256mem>, VEX_W;
-  defm VGATHERDPS : avx2_gather<0x92, "vgatherdps", VR256, v256mem>;
-  defm VGATHERQPS : avx2_gather<0x93, "vgatherqps", VR128, v256mem>;
-  defm VPGATHERDQ : avx2_gather<0x90, "vpgatherdq", VR256, v128mem>, VEX_W;
-  defm VPGATHERQQ : avx2_gather<0x91, "vpgatherqq", VR256, v256mem>, VEX_W;
-  defm VPGATHERDD : avx2_gather<0x90, "vpgatherdd", VR256, v256mem>;
-  defm VPGATHERQD : avx2_gather<0x91, "vpgatherqd", VR128, v256mem>;
+  defm VGATHERDPD : avx2_gather<0x92, "vgatherdpd", VR256, vx64mem, vx64mem>, VEX_W;
+  defm VGATHERQPD : avx2_gather<0x93, "vgatherqpd", VR256, vx64mem, vy64mem>, VEX_W;
+  defm VGATHERDPS : avx2_gather<0x92, "vgatherdps", VR256, vx32mem, vy32mem>;
+  defm VGATHERQPS : avx2_gather<0x93, "vgatherqps", VR128, vx32mem, vy32mem>;
+  defm VPGATHERDQ : avx2_gather<0x90, "vpgatherdq", VR256, vx64mem, vx64mem>, VEX_W;
+  defm VPGATHERQQ : avx2_gather<0x91, "vpgatherqq", VR256, vx64mem, vy64mem>, VEX_W;
+  defm VPGATHERDD : avx2_gather<0x90, "vpgatherdd", VR256, vx32mem, vy32mem>;
+  defm VPGATHERQD : avx2_gather<0x91, "vpgatherqd", VR128, vx32mem, vy32mem>;
 }