diff --git a/lib/Target/ARM/ARMCodeEmitter.cpp b/lib/Target/ARM/ARMCodeEmitter.cpp
index 895ad7322d9..fc12ddc1d26 100644
--- a/lib/Target/ARM/ARMCodeEmitter.cpp
+++ b/lib/Target/ARM/ARMCodeEmitter.cpp
@@ -225,6 +225,8 @@ namespace {
       const { return 0; }
     unsigned getBitfieldInvertedMaskOpValue(const MachineInstr &MI,
                                             unsigned Op) const { return 0; }
+    unsigned getMsbOpValue(const MachineInstr &MI,
+                           unsigned Op) const { return 0; }
     uint32_t getLdStmModeOpValue(const MachineInstr &MI, unsigned OpIdx)
       const {return 0; }
     uint32_t getLdStSORegOpValue(const MachineInstr &MI, unsigned OpIdx)
diff --git a/lib/Target/ARM/ARMInstrInfo.td b/lib/Target/ARM/ARMInstrInfo.td
index 9909dd0d021..15949c0698f 100644
--- a/lib/Target/ARM/ARMInstrInfo.td
+++ b/lib/Target/ARM/ARMInstrInfo.td
@@ -443,6 +443,18 @@ def bf_inv_mask_imm : Operand<i32>,
   let PrintMethod = "printBitfieldInvMaskImmOperand";
 }
 
+/// lsb_pos_imm - position of the lsb bit, used by BFI4p and t2BFI4p
+def lsb_pos_imm : Operand<i32>, PatLeaf<(imm), [{
+  return isInt<5>(N->getSExtValue());
+}]>;
+
+/// width_imm - number of bits to be copied, used by BFI4p and t2BFI4p
+def width_imm : Operand<i32>, PatLeaf<(imm), [{
+  return N->getSExtValue() > 0 &&  N->getSExtValue() <= 32;
+}] > {
+  let EncoderMethod = "getMsbOpValue";
+}
+
 // Define ARM specific addressing modes.
 
 
@@ -2463,6 +2475,25 @@ def BFI    : I<(outs GPR:$Rd), (ins GPR:$src, GPR:$Rn, bf_inv_mask_imm:$imm),
   let Inst{3-0}   = Rn;
 }
 
+// GNU as only supports this form of bfi (w/ 4 arguments)
+let isAsmParserOnly = 1 in
+def BFI4p : I<(outs GPR:$Rd), (ins GPR:$src, GPR:$Rn,
+                                   lsb_pos_imm:$lsb, width_imm:$width),
+               AddrMode1, Size4Bytes, IndexModeNone, DPFrm, IIC_iUNAsi,
+               "bfi", "\t$Rd, $Rn, $lsb, $width", "$src = $Rd",
+               []>, Requires<[IsARM, HasV6T2]> {
+  bits<4> Rd;
+  bits<4> Rn;
+  bits<5> lsb;
+  bits<5> width;
+  let Inst{27-21} = 0b0111110;
+  let Inst{6-4}   = 0b001; // Rn: Inst{3-0} != 15
+  let Inst{15-12} = Rd;
+  let Inst{11-7}  = lsb;
+  let Inst{20-16} = width; // Custom encoder => lsb+width-1
+  let Inst{3-0}   = Rn;
+}
+
 def  MVNr  : AsI1<0b1111, (outs GPR:$Rd), (ins GPR:$Rm), DPFrm, IIC_iMVNr,
                   "mvn", "\t$Rd, $Rm",
                   [(set GPR:$Rd, (not GPR:$Rm))]>, UnaryDP {
diff --git a/lib/Target/ARM/ARMInstrThumb2.td b/lib/Target/ARM/ARMInstrThumb2.td
index fe143af934a..1bd853005e1 100644
--- a/lib/Target/ARM/ARMInstrThumb2.td
+++ b/lib/Target/ARM/ARMInstrThumb2.td
@@ -2152,20 +2152,39 @@ def t2UBFX: T2TwoRegBitFI<
 }
 
 // A8.6.18  BFI - Bitfield insert (Encoding T1)
-let Constraints = "$src = $Rd" in
-def t2BFI : T2TwoRegBitFI<(outs rGPR:$Rd),
-                (ins rGPR:$src, rGPR:$Rn, bf_inv_mask_imm:$imm),
-                IIC_iBITi, "bfi", "\t$Rd, $Rn, $imm",
-                [(set rGPR:$Rd, (ARMbfi rGPR:$src, rGPR:$Rn,
-                                 bf_inv_mask_imm:$imm))]> {
-  let Inst{31-27} = 0b11110;
-  let Inst{25} = 1;
-  let Inst{24-20} = 0b10110;
-  let Inst{15} = 0;
+let Constraints = "$src = $Rd" in {
+  def t2BFI : T2TwoRegBitFI<(outs rGPR:$Rd),
+                  (ins rGPR:$src, rGPR:$Rn, bf_inv_mask_imm:$imm),
+                  IIC_iBITi, "bfi", "\t$Rd, $Rn, $imm",
+                  [(set rGPR:$Rd, (ARMbfi rGPR:$src, rGPR:$Rn,
+                                   bf_inv_mask_imm:$imm))]> {
+    let Inst{31-27} = 0b11110;
+    let Inst{25} = 1;
+    let Inst{24-20} = 0b10110;
+    let Inst{15} = 0;
 
-  bits<10> imm;
-  let msb{4-0} = imm{9-5};
-  let lsb{4-0} = imm{4-0};
+    bits<10> imm;
+    let msb{4-0} = imm{9-5};
+    let lsb{4-0} = imm{4-0};
+  }
+
+  // GNU as only supports this form of bfi (w/ 4 arguments)
+  let isAsmParserOnly = 1 in
+  def t2BFI4p : T2TwoRegBitFI<(outs rGPR:$Rd),
+                  (ins rGPR:$src, rGPR:$Rn, lsb_pos_imm:$lsbit,
+                       width_imm:$width),
+                  IIC_iBITi, "bfi", "\t$Rd, $Rn, $lsbit, $width",
+                  []> {
+    let Inst{31-27} = 0b11110;
+    let Inst{25} = 1;
+    let Inst{24-20} = 0b10110;
+    let Inst{15} = 0;
+
+    bits<5> lsbit;
+    bits<5> width;
+    let msb{4-0} = width; // Custom encoder => lsb+width-1
+    let lsb{4-0} = lsbit;
+  }
 }
 
 defm t2ORN  : T2I_bin_irs<0b0011, "orn",
diff --git a/lib/Target/ARM/ARMMCCodeEmitter.cpp b/lib/Target/ARM/ARMMCCodeEmitter.cpp
index 3aa40dbae70..a309122160f 100644
--- a/lib/Target/ARM/ARMMCCodeEmitter.cpp
+++ b/lib/Target/ARM/ARMMCCodeEmitter.cpp
@@ -262,6 +262,9 @@ public:
   unsigned getBitfieldInvertedMaskOpValue(const MCInst &MI, unsigned Op,
                                       SmallVectorImpl<MCFixup> &Fixups) const;
 
+  unsigned getMsbOpValue(const MCInst &MI, unsigned Op,
+                         SmallVectorImpl<MCFixup> &Fixups) const;
+
   unsigned getRegisterListOpValue(const MCInst &MI, unsigned Op,
                                   SmallVectorImpl<MCFixup> &Fixups) const;
   unsigned getAddrMode6AddressOpValue(const MCInst &MI, unsigned Op,
@@ -1066,6 +1069,17 @@ getBitfieldInvertedMaskOpValue(const MCInst &MI, unsigned Op,
   return lsb | (msb << 5);
 }
 
+unsigned ARMMCCodeEmitter::
+getMsbOpValue(const MCInst &MI, unsigned Op,
+              SmallVectorImpl<MCFixup> &Fixups) const {
+  // MSB - 5 bits.
+  uint32_t lsb = MI.getOperand(Op-1).getImm();
+  uint32_t width = MI.getOperand(Op).getImm();
+  uint32_t msb = lsb+width-1;
+  assert (width != 0 && msb < 32 && "Illegal bit width!");
+  return msb;
+}
+
 unsigned ARMMCCodeEmitter::
 getRegisterListOpValue(const MCInst &MI, unsigned Op,
                        SmallVectorImpl<MCFixup> &Fixups) const {
diff --git a/test/MC/ARM/arm_instructions.s b/test/MC/ARM/arm_instructions.s
index 43cad2f6524..002e5eb2627 100644
--- a/test/MC/ARM/arm_instructions.s
+++ b/test/MC/ARM/arm_instructions.s
@@ -124,3 +124,6 @@
 @ may depend on flags.
 @ CHECK-FIXME:: mlas	r1, r2, r3, r4 @ encoding: [0x92,0x43,0x31,0xe0]
 @        mlas r1,r2,r3,r4
+
+@ CHECK: bfi  r0, r0, #5, #7 @ encoding: [0x90,0x02,0xcb,0xe7]
+        bfi  r0, r0, #5, #7
diff --git a/test/MC/ARM/thumb2.s b/test/MC/ARM/thumb2.s
index ae88d4806ec..ceaf0a424f9 100644
--- a/test/MC/ARM/thumb2.s
+++ b/test/MC/ARM/thumb2.s
@@ -162,3 +162,6 @@
   ldrsb.w	r0, [r0]
 @ CHECK: ldrsh.w	r0, [r0]                @ encoding: [0x00,0x00,0xb0,0xf9]
   ldrsh.w	r0, [r0]
+@ CHECK: bfi  r0, r0, #5, #7 @ encoding: [0x60,0xf3,0x4b,0x10]
+  bfi  r0, r0, #5, #7
+
diff --git a/utils/TableGen/ARMDecoderEmitter.cpp b/utils/TableGen/ARMDecoderEmitter.cpp
index 0c88a316580..21af62050d7 100644
--- a/utils/TableGen/ARMDecoderEmitter.cpp
+++ b/utils/TableGen/ARMDecoderEmitter.cpp
@@ -1560,6 +1560,10 @@ ARMDEBackend::populateInstruction(const CodeGenInstruction &CGI,
   // which is a better design and less fragile than the name matchings.
   if (Bits.allInComplete()) return false;
 
+  // Ignore "asm parser only" instructions.
+  if (Def.getValueAsBit("isAsmParserOnly"))
+    return false;
+
   if (TN == TARGET_ARM) {
     // FIXME: what about Int_MemBarrierV6 and Int_SyncBarrierV6?
     if ((Name != "Int_MemBarrierV7" && Name != "Int_SyncBarrierV7") &&
diff --git a/utils/TableGen/EDEmitter.cpp b/utils/TableGen/EDEmitter.cpp
index c4391437474..353dbd6340c 100644
--- a/utils/TableGen/EDEmitter.cpp
+++ b/utils/TableGen/EDEmitter.cpp
@@ -566,6 +566,8 @@ static int ARMFlagFromOpName(LiteralConstantEmitter *type,
   IMM("i32imm");
   IMM("i32imm_hilo16");
   IMM("bf_inv_mask_imm");
+  IMM("lsb_pos_imm");
+  IMM("width_imm");
   IMM("jtblock_operand");
   IMM("nohash_imm");
   IMM("p_imm");