From: Colin LeMahieu <colinl@codeaurora.org>
Date: Tue, 3 Feb 2015 19:15:11 +0000 (+0000)
Subject: [Hexagon] Adding missing vector multiply instruction encodings.  Converting multiply... 
X-Git-Url: http://demsky.eecs.uci.edu/git/?a=commitdiff_plain;h=6217146dce813640056927011f97ad2590ba9a4a;p=oota-llvm.git

[Hexagon] Adding missing vector multiply instruction encodings.  Converting multiply intrinsics and updating tests.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@228010 91177308-0d34-0410-b5e6-96231b3b80d8
---

diff --git a/lib/Target/Hexagon/HexagonInstrInfo.td b/lib/Target/Hexagon/HexagonInstrInfo.td
index b415bf88627..6b8bc766983 100644
--- a/lib/Target/Hexagon/HexagonInstrInfo.td
+++ b/lib/Target/Hexagon/HexagonInstrInfo.td
@@ -2862,6 +2862,10 @@ def A2_vrsadub: T_XTYPE_Vect <"vrsadub", 0b010, 0b010, 0>;
 def A2_vrsadub_acc: T_XTYPE_Vect_acc <"vrsadub", 0b010, 0b010, 0>;
 }
 
+// Vector absolute difference words: Rdd=vabsdiffw(Rtt,Rss)
+let isCodeGenOnly = 0 in
+def M2_vabsdiffw: T_XTYPE_Vect_diff<0b001, "vabsdiffw">;
+
 // Vector absolute difference: Rdd=vabsdiffh(Rtt,Rss)
 let isCodeGenOnly = 0 in
 def M2_vabsdiffh: T_XTYPE_Vect_diff<0b011, "vabsdiffh">;
@@ -2951,6 +2955,22 @@ def M2_mmachs_rs1: T_M2_vmpy_acc_sat <"vmpywoh", 0b101, 0b111, 1, 1>;
 def M2_mmachs_rs0: T_M2_vmpy_acc_sat <"vmpywoh", 0b001, 0b111, 0, 1>;
 }
 
+// Vector multiply word by unsigned half with accumulation
+// Rxx+=vmpyw[eo]uh(Rss,Rtt)[:<<1][:rnd]:sat
+let isCodeGenOnly = 0 in {
+def M2_mmaculs_s1:  T_M2_vmpy_acc_sat <"vmpyweuh", 0b110, 0b101, 1, 0>;
+def M2_mmaculs_s0:  T_M2_vmpy_acc_sat <"vmpyweuh", 0b010, 0b101, 0, 0>;
+def M2_mmaculs_rs1: T_M2_vmpy_acc_sat <"vmpyweuh", 0b111, 0b101, 1, 1>;
+def M2_mmaculs_rs0: T_M2_vmpy_acc_sat <"vmpyweuh", 0b011, 0b101, 0, 1>;
+}
+
+let isCodeGenOnly = 0 in {
+def M2_mmacuhs_s1:  T_M2_vmpy_acc_sat <"vmpywouh", 0b110, 0b111, 1, 0>;
+def M2_mmacuhs_s0:  T_M2_vmpy_acc_sat <"vmpywouh", 0b010, 0b111, 0, 0>;
+def M2_mmacuhs_rs1: T_M2_vmpy_acc_sat <"vmpywouh", 0b111, 0b111, 1, 1>;
+def M2_mmacuhs_rs0: T_M2_vmpy_acc_sat <"vmpywouh", 0b011, 0b111, 0, 1>;
+}
+
 // Vector multiply even halfwords with accumulation
 // Rxx+=vmpyeh(Rss,Rtt)[:<<1][:sat]
 let isCodeGenOnly = 0 in {
diff --git a/lib/Target/Hexagon/HexagonIntrinsics.td b/lib/Target/Hexagon/HexagonIntrinsics.td
index 419c21e309f..8f3171719a6 100644
--- a/lib/Target/Hexagon/HexagonIntrinsics.td
+++ b/lib/Target/Hexagon/HexagonIntrinsics.td
@@ -345,6 +345,52 @@ def : T_PP_pat <M2_vcmpy_s0_sat_i, int_hexagon_M2_vcmpy_s0_sat_i>;
 def : T_PP_pat <M2_vcmpy_s1_sat_r, int_hexagon_M2_vcmpy_s1_sat_r>;
 def : T_PP_pat <M2_vcmpy_s0_sat_r, int_hexagon_M2_vcmpy_s0_sat_r>;
 
+// Vector dual multiply: Rdd=vdmpy(Rss,Rtt)[:<<1]:sat
+def : T_PP_pat <M2_vdmpys_s1, int_hexagon_M2_vdmpys_s1>;
+def : T_PP_pat <M2_vdmpys_s0, int_hexagon_M2_vdmpys_s0>;
+
+// Vector multiply even halfwords: Rdd=vmpyeh(Rss,Rtt)[:<<1]:sat
+def : T_PP_pat <M2_vmpy2es_s1, int_hexagon_M2_vmpy2es_s1>;
+def : T_PP_pat <M2_vmpy2es_s0, int_hexagon_M2_vmpy2es_s0>;
+
+//Rdd=vmpywoh(Rss,Rtt)[:<<1][:rnd]:sat
+def : T_PP_pat <M2_mmpyh_s0,  int_hexagon_M2_mmpyh_s0>;
+def : T_PP_pat <M2_mmpyh_s1,  int_hexagon_M2_mmpyh_s1>;
+def : T_PP_pat <M2_mmpyh_rs0, int_hexagon_M2_mmpyh_rs0>;
+def : T_PP_pat <M2_mmpyh_rs1, int_hexagon_M2_mmpyh_rs1>;
+
+//Rdd=vmpyweh(Rss,Rtt)[:<<1][:rnd]:sat
+def : T_PP_pat <M2_mmpyl_s0,  int_hexagon_M2_mmpyl_s0>;
+def : T_PP_pat <M2_mmpyl_s1,  int_hexagon_M2_mmpyl_s1>;
+def : T_PP_pat <M2_mmpyl_rs0, int_hexagon_M2_mmpyl_rs0>;
+def : T_PP_pat <M2_mmpyl_rs1, int_hexagon_M2_mmpyl_rs1>;
+
+//Rdd=vmpywouh(Rss,Rtt)[:<<1][:rnd]:sat
+def : T_PP_pat <M2_mmpyuh_s0,  int_hexagon_M2_mmpyuh_s0>;
+def : T_PP_pat <M2_mmpyuh_s1,  int_hexagon_M2_mmpyuh_s1>;
+def : T_PP_pat <M2_mmpyuh_rs0, int_hexagon_M2_mmpyuh_rs0>;
+def : T_PP_pat <M2_mmpyuh_rs1, int_hexagon_M2_mmpyuh_rs1>;
+
+//Rdd=vmpyweuh(Rss,Rtt)[:<<1][:rnd]:sat
+def : T_PP_pat <M2_mmpyul_s0,  int_hexagon_M2_mmpyul_s0>;
+def : T_PP_pat <M2_mmpyul_s1,  int_hexagon_M2_mmpyul_s1>;
+def : T_PP_pat <M2_mmpyul_rs0, int_hexagon_M2_mmpyul_rs0>;
+def : T_PP_pat <M2_mmpyul_rs1, int_hexagon_M2_mmpyul_rs1>;
+
+// Vector reduce add unsigned bytes: Rdd32[+]=vrmpybu(Rss32,Rtt32)
+def : T_PP_pat  <A2_vraddub,     int_hexagon_A2_vraddub>;
+def : T_PPP_pat <A2_vraddub_acc, int_hexagon_A2_vraddub_acc>;
+
+// Vector sum of absolute differences unsigned bytes: Rdd=vrsadub(Rss,Rtt)
+def : T_PP_pat  <A2_vrsadub,     int_hexagon_A2_vrsadub>;
+def : T_PPP_pat <A2_vrsadub_acc, int_hexagon_A2_vrsadub_acc>;
+
+// Vector absolute difference: Rdd=vabsdiffh(Rtt,Rss)
+def : T_PP_pat <M2_vabsdiffh, int_hexagon_M2_vabsdiffh>;
+
+// Vector absolute difference words: Rdd=vabsdiffw(Rtt,Rss)
+def : T_PP_pat <M2_vabsdiffw, int_hexagon_M2_vabsdiffw>;
+
 // Vector reduce complex multiply real or imaginary:
 // Rdd[+]=vrcmpy[ir](Rss,Rtt[*])
 def : T_PP_pat  <M2_vrcmpyi_s0,  int_hexagon_M2_vrcmpyi_s0>;
@@ -362,6 +408,43 @@ def : T_PPP_pat <M2_vrcmacr_s0c, int_hexagon_M2_vrcmacr_s0c>;
 def : T_PP_pat  <M2_vrmpy_s0, int_hexagon_M2_vrmpy_s0>;
 def : T_PPP_pat <M2_vrmac_s0, int_hexagon_M2_vrmac_s0>;
 
+//===----------------------------------------------------------------------===//
+// Vector Multipy with accumulation
+//===----------------------------------------------------------------------===//
+
+// Vector multiply word by signed half with accumulation
+// Rxx+=vmpyw[eo]h(Rss,Rtt)[:<<1][:rnd]:sat
+def : T_PPP_pat <M2_mmacls_s1, int_hexagon_M2_mmacls_s1>;
+def : T_PPP_pat <M2_mmacls_s0, int_hexagon_M2_mmacls_s0>;
+def : T_PPP_pat <M2_mmacls_rs1, int_hexagon_M2_mmacls_rs1>;
+def : T_PPP_pat <M2_mmacls_rs0, int_hexagon_M2_mmacls_rs0>;
+def : T_PPP_pat <M2_mmachs_s1, int_hexagon_M2_mmachs_s1>;
+def : T_PPP_pat <M2_mmachs_s0, int_hexagon_M2_mmachs_s0>;
+def : T_PPP_pat <M2_mmachs_rs1, int_hexagon_M2_mmachs_rs1>;
+def : T_PPP_pat <M2_mmachs_rs0, int_hexagon_M2_mmachs_rs0>;
+
+// Vector multiply word by unsigned half with accumulation
+// Rxx+=vmpyw[eo]uh(Rss,Rtt)[:<<1][:rnd]:sat
+def : T_PPP_pat <M2_mmaculs_s1, int_hexagon_M2_mmaculs_s1>;
+def : T_PPP_pat <M2_mmaculs_s0, int_hexagon_M2_mmaculs_s0>;
+def : T_PPP_pat <M2_mmaculs_rs1, int_hexagon_M2_mmaculs_rs1>;
+def : T_PPP_pat <M2_mmaculs_rs0, int_hexagon_M2_mmaculs_rs0>;
+def : T_PPP_pat <M2_mmacuhs_s1, int_hexagon_M2_mmacuhs_s1>;
+def : T_PPP_pat <M2_mmacuhs_s0, int_hexagon_M2_mmacuhs_s0>;
+def : T_PPP_pat <M2_mmacuhs_rs1, int_hexagon_M2_mmacuhs_rs1>;
+def : T_PPP_pat <M2_mmacuhs_rs0, int_hexagon_M2_mmacuhs_rs0>;
+
+// Vector multiply even halfwords with accumulation
+// Rxx+=vmpyeh(Rss,Rtt)[:<<1][:sat]
+def : T_PPP_pat <M2_vmac2es, int_hexagon_M2_vmac2es>;
+def : T_PPP_pat <M2_vmac2es_s1, int_hexagon_M2_vmac2es_s1>;
+def : T_PPP_pat <M2_vmac2es_s0, int_hexagon_M2_vmac2es_s0>;
+
+// Vector dual multiply with accumulation
+// Rxx+=vdmpy(Rss,Rtt)[:sat]
+def : T_PPP_pat <M2_vdmacs_s1, int_hexagon_M2_vdmacs_s1>;
+def : T_PPP_pat <M2_vdmacs_s0, int_hexagon_M2_vdmacs_s0>;
+
 // Vector complex multiply real or imaginary with accumulation
 // Rxx+=vcmpy[ir](Rss,Rtt):sat
 def : T_PPP_pat <M2_vcmac_s0_sat_r, int_hexagon_M2_vcmac_s0_sat_r>;
@@ -784,6 +867,11 @@ def : T_RR_pat <M2_cmpysc_s0,  int_hexagon_M2_cmpysc_s0>;
 def : T_RR_pat <M2_cmpys_s1,   int_hexagon_M2_cmpys_s1>;
 def : T_RR_pat <M2_cmpysc_s1,  int_hexagon_M2_cmpysc_s1>;
 
+// Vector multiply halfwords
+// Rdd=vmpyh(Rs,Rt)[:<<1]:sat
+def : T_RR_pat <M2_vmpy2s_s0,  int_hexagon_M2_vmpy2s_s0>;
+def : T_RR_pat <M2_vmpy2s_s1,  int_hexagon_M2_vmpy2s_s1>;
+
 // Rxx[+-]= mpy[u](Rs,Rt)
 def : T_PRR_pat <M2_dpmpyss_acc_s0, int_hexagon_M2_dpmpyss_acc_s0>;
 def : T_PRR_pat <M2_dpmpyss_nac_s0, int_hexagon_M2_dpmpyss_nac_s0>;
@@ -806,6 +894,11 @@ def : T_PRR_pat <M2_cnacsc_s1, int_hexagon_M2_cnacsc_s1>;
 def : T_PRR_pat <M2_cmaci_s0, int_hexagon_M2_cmaci_s0>;
 def : T_PRR_pat <M2_cmacr_s0, int_hexagon_M2_cmacr_s0>;
 
+// Rxx+=vmpyh(Rs,Rt)[:<<1][:sat]
+def : T_PRR_pat <M2_vmac2, int_hexagon_M2_vmac2>;
+def : T_PRR_pat <M2_vmac2s_s0, int_hexagon_M2_vmac2s_s0>;
+def : T_PRR_pat <M2_vmac2s_s1, int_hexagon_M2_vmac2s_s1>;
+
 /********************************************************************
 *            CR                                                     *
 *********************************************************************/
@@ -848,6 +941,19 @@ class MType_R32_pat <Intrinsic IntID, InstHexagon OutputInst> :
       Pat <(IntID IntRegs:$src1, IntRegs:$src2),
            (OutputInst IntRegs:$src1, IntRegs:$src2)>;
 
+// Vector dual multiply with round and pack
+
+def : Pat <(int_hexagon_M2_vdmpyrs_s0 DoubleRegs:$src1, DoubleRegs:$src2),
+           (M2_vdmpyrs_s0 DoubleRegs:$src1, DoubleRegs:$src2)>;
+
+def : Pat <(int_hexagon_M2_vdmpyrs_s1 DoubleRegs:$src1, DoubleRegs:$src2),
+           (M2_vdmpyrs_s1 DoubleRegs:$src1, DoubleRegs:$src2)>;
+
+// Vector multiply halfwords with round and pack
+
+def : MType_R32_pat <int_hexagon_M2_vmpy2s_s0pack, M2_vmpy2s_s0pack>;
+def : MType_R32_pat <int_hexagon_M2_vmpy2s_s1pack, M2_vmpy2s_s1pack>;
+
 // Multiply and use lower result
 def : MType_R32_pat <int_hexagon_M2_mpyi, M2_mpyi>;
 def : T_RI_pat<M2_mpysmi, int_hexagon_M2_mpysmi>;
@@ -946,6 +1052,15 @@ def: T_RR_pat<S2_togglebit_r, int_hexagon_S2_togglebit_r>;
 def: T_RI_pat<S2_tstbit_i,    int_hexagon_S2_tstbit_i>;
 def: T_RR_pat<S2_tstbit_r,    int_hexagon_S2_tstbit_r>;
 
+/********************************************************************
+*            STYPE/COMPLEX                                          *
+*********************************************************************/
+// Vector Complex conjugate
+def : T_P_pat <A2_vconj, int_hexagon_A2_vconj>;
+
+// Vector Complex rotate
+def : T_PR_pat <S2_vcrotate, int_hexagon_S2_vcrotate>;
+
 /********************************************************************
 *            STYPE/SHIFT                                            *
 *********************************************************************/
@@ -1542,191 +1657,6 @@ class di_LDInstPI_diu4<string opc, Intrinsic IntID>
            [],
            "$src1 = $dst">;
 
-
-// ALU64 / VB / Vector maximum/minimum unsigned bytes.
-def HEXAGON_A2_vmaxub:
-  di_ALU64_didi                   <"vmaxub",   int_hexagon_A2_vmaxub>;
-def HEXAGON_A2_vminub:
-  di_ALU64_didi                   <"vminub",   int_hexagon_A2_vminub>;
-
-// ALU64 / VB / Vector subtract unsigned bytes.
-def HEXAGON_A2_vsubub:
-  di_ALU64_didi                   <"vsubub",   int_hexagon_A2_vsubub>;
-def HEXAGON_A2_vsububs:
-  di_ALU64_didi_sat               <"vsubub",   int_hexagon_A2_vsububs>;
-
-/********************************************************************
-*            MTYPE/ALU                                              *
-*********************************************************************/
-
-// MTYPE / ALU / Vector absolute difference.
-def HEXAGON_M2_vabsdiffh:
-  di_MInst_didi                   <"vabsdiffh",int_hexagon_M2_vabsdiffh>;
-def HEXAGON_M2_vabsdiffw:
-  di_MInst_didi                   <"vabsdiffw",int_hexagon_M2_vabsdiffw>;
-
-/********************************************************************
-*            MTYPE/MPYH                                             *
-*********************************************************************/
-
-// MTYPE / MPYH / Multiply word by half (32x16).
-//Rdd[+]=vmpywoh(Rss,Rtt)[:<<1][:rnd][:sat]
-//Rdd[+]=vmpyweh(Rss,Rtt)[:<<1][:rnd][:sat]
-def HEXAGON_M2_mmpyl_rs1:
-  di_MInst_didi_s1_rnd_sat        <"vmpyweh",  int_hexagon_M2_mmpyl_rs1>;
-def HEXAGON_M2_mmpyl_s1:
-  di_MInst_didi_s1_sat            <"vmpyweh",  int_hexagon_M2_mmpyl_s1>;
-def HEXAGON_M2_mmpyl_rs0:
-  di_MInst_didi_rnd_sat           <"vmpyweh",  int_hexagon_M2_mmpyl_rs0>;
-def HEXAGON_M2_mmpyl_s0:
-  di_MInst_didi_sat               <"vmpyweh",  int_hexagon_M2_mmpyl_s0>;
-def HEXAGON_M2_mmpyh_rs1:
-  di_MInst_didi_s1_rnd_sat        <"vmpywoh",  int_hexagon_M2_mmpyh_rs1>;
-def HEXAGON_M2_mmpyh_s1:
-  di_MInst_didi_s1_sat            <"vmpywoh",  int_hexagon_M2_mmpyh_s1>;
-def HEXAGON_M2_mmpyh_rs0:
-  di_MInst_didi_rnd_sat           <"vmpywoh",  int_hexagon_M2_mmpyh_rs0>;
-def HEXAGON_M2_mmpyh_s0:
-  di_MInst_didi_sat               <"vmpywoh",  int_hexagon_M2_mmpyh_s0>;
-def HEXAGON_M2_mmacls_rs1:
-  di_MInst_dididi_acc_s1_rnd_sat  <"vmpyweh",  int_hexagon_M2_mmacls_rs1>;
-def HEXAGON_M2_mmacls_s1:
-  di_MInst_dididi_acc_s1_sat      <"vmpyweh",  int_hexagon_M2_mmacls_s1>;
-def HEXAGON_M2_mmacls_rs0:
-  di_MInst_dididi_acc_rnd_sat     <"vmpyweh",  int_hexagon_M2_mmacls_rs0>;
-def HEXAGON_M2_mmacls_s0:
-  di_MInst_dididi_acc_sat         <"vmpyweh",  int_hexagon_M2_mmacls_s0>;
-def HEXAGON_M2_mmachs_rs1:
-  di_MInst_dididi_acc_s1_rnd_sat  <"vmpywoh",  int_hexagon_M2_mmachs_rs1>;
-def HEXAGON_M2_mmachs_s1:
-  di_MInst_dididi_acc_s1_sat      <"vmpywoh",  int_hexagon_M2_mmachs_s1>;
-def HEXAGON_M2_mmachs_rs0:
-  di_MInst_dididi_acc_rnd_sat     <"vmpywoh",  int_hexagon_M2_mmachs_rs0>;
-def HEXAGON_M2_mmachs_s0:
-  di_MInst_dididi_acc_sat         <"vmpywoh",  int_hexagon_M2_mmachs_s0>;
-
-// MTYPE / MPYH / Multiply word by unsigned half (32x16).
-//Rdd[+]=vmpywouh(Rss,Rtt)[:<<1][:rnd][:sat]
-//Rdd[+]=vmpyweuh(Rss,Rtt)[:<<1][:rnd][:sat]
-def HEXAGON_M2_mmpyul_rs1:
-  di_MInst_didi_s1_rnd_sat        <"vmpyweuh", int_hexagon_M2_mmpyul_rs1>;
-def HEXAGON_M2_mmpyul_s1:
-  di_MInst_didi_s1_sat            <"vmpyweuh", int_hexagon_M2_mmpyul_s1>;
-def HEXAGON_M2_mmpyul_rs0:
-  di_MInst_didi_rnd_sat           <"vmpyweuh", int_hexagon_M2_mmpyul_rs0>;
-def HEXAGON_M2_mmpyul_s0:
-  di_MInst_didi_sat               <"vmpyweuh", int_hexagon_M2_mmpyul_s0>;
-def HEXAGON_M2_mmpyuh_rs1:
-  di_MInst_didi_s1_rnd_sat        <"vmpywouh", int_hexagon_M2_mmpyuh_rs1>;
-def HEXAGON_M2_mmpyuh_s1:
-  di_MInst_didi_s1_sat            <"vmpywouh", int_hexagon_M2_mmpyuh_s1>;
-def HEXAGON_M2_mmpyuh_rs0:
-  di_MInst_didi_rnd_sat           <"vmpywouh", int_hexagon_M2_mmpyuh_rs0>;
-def HEXAGON_M2_mmpyuh_s0:
-  di_MInst_didi_sat               <"vmpywouh", int_hexagon_M2_mmpyuh_s0>;
-def HEXAGON_M2_mmaculs_rs1:
-  di_MInst_dididi_acc_s1_rnd_sat  <"vmpyweuh", int_hexagon_M2_mmaculs_rs1>;
-def HEXAGON_M2_mmaculs_s1:
-  di_MInst_dididi_acc_s1_sat      <"vmpyweuh", int_hexagon_M2_mmaculs_s1>;
-def HEXAGON_M2_mmaculs_rs0:
-  di_MInst_dididi_acc_rnd_sat     <"vmpyweuh", int_hexagon_M2_mmaculs_rs0>;
-def HEXAGON_M2_mmaculs_s0:
-  di_MInst_dididi_acc_sat         <"vmpyweuh", int_hexagon_M2_mmaculs_s0>;
-def HEXAGON_M2_mmacuhs_rs1:
-  di_MInst_dididi_acc_s1_rnd_sat  <"vmpywouh", int_hexagon_M2_mmacuhs_rs1>;
-def HEXAGON_M2_mmacuhs_s1:
-  di_MInst_dididi_acc_s1_sat      <"vmpywouh", int_hexagon_M2_mmacuhs_s1>;
-def HEXAGON_M2_mmacuhs_rs0:
-  di_MInst_dididi_acc_rnd_sat     <"vmpywouh", int_hexagon_M2_mmacuhs_rs0>;
-def HEXAGON_M2_mmacuhs_s0:
-  di_MInst_dididi_acc_sat         <"vmpywouh", int_hexagon_M2_mmacuhs_s0>;
-
-/********************************************************************
-*            MTYPE/VB                                               *
-*********************************************************************/
-
-// MTYPE / VB / Vector reduce add unsigned bytes.
-def HEXAGON_A2_vraddub:
-  di_MInst_didi                   <"vraddub", int_hexagon_A2_vraddub>;
-def HEXAGON_A2_vraddub_acc:
-  di_MInst_dididi_acc             <"vraddub", int_hexagon_A2_vraddub_acc>;
-
-// MTYPE / VB / Vector sum of absolute differences unsigned bytes.
-def HEXAGON_A2_vrsadub:
-  di_MInst_didi                   <"vrsadub", int_hexagon_A2_vrsadub>;
-def HEXAGON_A2_vrsadub_acc:
-  di_MInst_dididi_acc             <"vrsadub", int_hexagon_A2_vrsadub_acc>;
-
-/********************************************************************
-*            MTYPE/VH                                               *
-*********************************************************************/
-
-// MTYPE / VH / Vector dual multiply.
-def HEXAGON_M2_vdmpys_s1:
-  di_MInst_didi_s1_sat            <"vdmpy",   int_hexagon_M2_vdmpys_s1>;
-def HEXAGON_M2_vdmpys_s0:
-  di_MInst_didi_sat               <"vdmpy",   int_hexagon_M2_vdmpys_s0>;
-def HEXAGON_M2_vdmacs_s1:
-  di_MInst_dididi_acc_s1_sat      <"vdmpy",   int_hexagon_M2_vdmacs_s1>;
-def HEXAGON_M2_vdmacs_s0:
-  di_MInst_dididi_acc_sat         <"vdmpy",   int_hexagon_M2_vdmacs_s0>;
-
-// MTYPE / VH / Vector dual multiply with round and pack.
-def HEXAGON_M2_vdmpyrs_s0:
-  si_MInst_didi_rnd_sat           <"vdmpy",   int_hexagon_M2_vdmpyrs_s0>;
-def HEXAGON_M2_vdmpyrs_s1:
-  si_MInst_didi_s1_rnd_sat        <"vdmpy",   int_hexagon_M2_vdmpyrs_s1>;
-
-// MTYPE / VH / Vector multiply even halfwords.
-def HEXAGON_M2_vmpy2es_s1:
-  di_MInst_didi_s1_sat            <"vmpyeh",  int_hexagon_M2_vmpy2es_s1>;
-def HEXAGON_M2_vmpy2es_s0:
-  di_MInst_didi_sat               <"vmpyeh",  int_hexagon_M2_vmpy2es_s0>;
-def HEXAGON_M2_vmac2es:
-  di_MInst_dididi_acc             <"vmpyeh",  int_hexagon_M2_vmac2es>;
-def HEXAGON_M2_vmac2es_s1:
-  di_MInst_dididi_acc_s1_sat      <"vmpyeh",  int_hexagon_M2_vmac2es_s1>;
-def HEXAGON_M2_vmac2es_s0:
-  di_MInst_dididi_acc_sat         <"vmpyeh",  int_hexagon_M2_vmac2es_s0>;
-
-// MTYPE / VH / Vector multiply halfwords.
-def HEXAGON_M2_vmpy2s_s0:
-  di_MInst_sisi_sat               <"vmpyh",   int_hexagon_M2_vmpy2s_s0>;
-def HEXAGON_M2_vmpy2s_s1:
-  di_MInst_sisi_s1_sat            <"vmpyh",   int_hexagon_M2_vmpy2s_s1>;
-def HEXAGON_M2_vmac2:
-  di_MInst_disisi_acc             <"vmpyh",   int_hexagon_M2_vmac2>;
-def HEXAGON_M2_vmac2s_s0:
-  di_MInst_disisi_acc_sat         <"vmpyh",   int_hexagon_M2_vmac2s_s0>;
-def HEXAGON_M2_vmac2s_s1:
-  di_MInst_disisi_acc_s1_sat      <"vmpyh",   int_hexagon_M2_vmac2s_s1>;
-
-// MTYPE / VH / Vector multiply halfwords with round and pack.
-def HEXAGON_M2_vmpy2s_s0pack:
-  si_MInst_sisi_rnd_sat           <"vmpyh",   int_hexagon_M2_vmpy2s_s0pack>;
-def HEXAGON_M2_vmpy2s_s1pack:
-  si_MInst_sisi_s1_rnd_sat        <"vmpyh",   int_hexagon_M2_vmpy2s_s1pack>;
-
-// MTYPE / VH / Vector reduce multiply halfwords.
-// Rxx32+=vrmpyh(Rss32,Rtt32)
-def HEXAGON_M2_vrmpy_s0:
-  di_MInst_didi                   <"vrmpyh",  int_hexagon_M2_vrmpy_s0>;
-def HEXAGON_M2_vrmac_s0:
-  di_MInst_dididi_acc             <"vrmpyh",  int_hexagon_M2_vrmac_s0>;
-
-/********************************************************************
-*            STYPE/COMPLEX                                          *
-*********************************************************************/
-
-// STYPE / COMPLEX / Vector Complex conjugate.
-def HEXAGON_A2_vconj:
-  di_SInst_di_sat                 <"vconj",   int_hexagon_A2_vconj>;
-
-// STYPE / COMPLEX / Vector Complex rotate.
-def HEXAGON_S2_vcrotate:
-  di_SInst_disi                   <"vcrotate",int_hexagon_S2_vcrotate>;
-
-
 /********************************************************************
 *            STYPE/PERM                                             *
 *********************************************************************/
diff --git a/lib/Target/Hexagon/HexagonIntrinsicsV4.td b/lib/Target/Hexagon/HexagonIntrinsicsV4.td
index f93d0e8c540..556e19d2cc0 100644
--- a/lib/Target/Hexagon/HexagonIntrinsicsV4.td
+++ b/lib/Target/Hexagon/HexagonIntrinsicsV4.td
@@ -12,6 +12,38 @@
 // 80-V9418-12 Rev. A
 // June 15, 2010
 
+// Vector reduce multiply word by signed half (32x16)
+//Rdd=vrmpyweh(Rss,Rtt)[:<<1]
+def : T_PP_pat <M4_vrmpyeh_s0, int_hexagon_M4_vrmpyeh_s0>;
+def : T_PP_pat <M4_vrmpyeh_s1, int_hexagon_M4_vrmpyeh_s1>;
+
+//Rdd=vrmpywoh(Rss,Rtt)[:<<1]
+def : T_PP_pat <M4_vrmpyoh_s0, int_hexagon_M4_vrmpyoh_s0>;
+def : T_PP_pat <M4_vrmpyoh_s1, int_hexagon_M4_vrmpyoh_s1>;
+
+//Rdd+=vrmpyweh(Rss,Rtt)[:<<1]
+def : T_PPP_pat <M4_vrmpyeh_acc_s0, int_hexagon_M4_vrmpyeh_acc_s0>;
+def : T_PPP_pat <M4_vrmpyeh_acc_s1, int_hexagon_M4_vrmpyeh_acc_s1>;
+
+//Rdd=vrmpywoh(Rss,Rtt)[:<<1]
+def : T_PPP_pat <M4_vrmpyoh_acc_s0, int_hexagon_M4_vrmpyoh_acc_s0>;
+def : T_PPP_pat <M4_vrmpyoh_acc_s1, int_hexagon_M4_vrmpyoh_acc_s1>;
+
+// Vector multiply halfwords, signed by unsigned
+// Rdd=vmpyhsu(Rs,Rt)[:<<1]:sat
+def : T_RR_pat <M2_vmpy2su_s0, int_hexagon_M2_vmpy2su_s0>;
+def : T_RR_pat <M2_vmpy2su_s1, int_hexagon_M2_vmpy2su_s1>;
+
+// Rxx+=vmpyhsu(Rs,Rt)[:<<1]:sat
+def : T_PRR_pat <M2_vmac2su_s0, int_hexagon_M2_vmac2su_s0>;
+def : T_PRR_pat <M2_vmac2su_s1, int_hexagon_M2_vmac2su_s1>;
+
+// Vector polynomial multiply halfwords
+// Rdd=vpmpyh(Rs,Rt)
+def : T_RR_pat <M4_vpmpyh, int_hexagon_M4_vpmpyh>;
+// Rxx[^]=vpmpyh(Rs,Rt)
+def : T_PRR_pat <M4_vpmpyh_acc, int_hexagon_M4_vpmpyh_acc>;
+
 // Polynomial multiply words
 // Rdd=pmpyw(Rs,Rt)
 def : T_RR_pat <M4_pmpyw, int_hexagon_M4_pmpyw>;
diff --git a/test/CodeGen/Hexagon/intrinsics/xtype_mpy.ll b/test/CodeGen/Hexagon/intrinsics/xtype_mpy.ll
index f9c93970894..5ed761a613f 100644
--- a/test/CodeGen/Hexagon/intrinsics/xtype_mpy.ll
+++ b/test/CodeGen/Hexagon/intrinsics/xtype_mpy.ll
@@ -37,6 +37,120 @@ define i32 @M4_mpyrr_addr(i32 %a, i32 %b, i32 %c) {
 }
 ; CHECK: r1 = add(r0, mpyi(r1, r2))
 
+; Vector multiply word by signed half (32x16)
+declare i64 @llvm.hexagon.M2.mmpyl.s0(i64, i64)
+define i64 @M2_mmpyl_s0(i64 %a, i64 %b) {
+  %z = call i64 @llvm.hexagon.M2.mmpyl.s0(i64 %a, i64 %b)
+  ret i64 %z
+}
+; CHECK: r1:0 = vmpyweh(r1:0, r3:2):sat
+
+declare i64 @llvm.hexagon.M2.mmpyl.s1(i64, i64)
+define i64 @M2_mmpyl_s1(i64 %a, i64 %b) {
+  %z = call i64 @llvm.hexagon.M2.mmpyl.s1(i64 %a, i64 %b)
+  ret i64 %z
+}
+; CHECK: r1:0 = vmpyweh(r1:0, r3:2):<<1:sat
+
+declare i64 @llvm.hexagon.M2.mmpyh.s0(i64, i64)
+define i64 @M2_mmpyh_s0(i64 %a, i64 %b) {
+  %z = call i64 @llvm.hexagon.M2.mmpyh.s0(i64 %a, i64 %b)
+  ret i64 %z
+}
+; CHECK: r1:0 = vmpywoh(r1:0, r3:2):sat
+
+declare i64 @llvm.hexagon.M2.mmpyh.s1(i64, i64)
+define i64 @M2_mmpyh_s1(i64 %a, i64 %b) {
+  %z = call i64 @llvm.hexagon.M2.mmpyh.s1(i64 %a, i64 %b)
+  ret i64 %z
+}
+; CHECK: r1:0 = vmpywoh(r1:0, r3:2):<<1:sat
+
+declare i64 @llvm.hexagon.M2.mmpyl.rs0(i64, i64)
+define i64 @M2_mmpyl_rs0(i64 %a, i64 %b) {
+  %z = call i64 @llvm.hexagon.M2.mmpyl.rs0(i64 %a, i64 %b)
+  ret i64 %z
+}
+; CHECK: r1:0 = vmpyweh(r1:0, r3:2):rnd:sat
+
+declare i64 @llvm.hexagon.M2.mmpyl.rs1(i64, i64)
+define i64 @M2_mmpyl_rs1(i64 %a, i64 %b) {
+  %z = call i64 @llvm.hexagon.M2.mmpyl.rs1(i64 %a, i64 %b)
+  ret i64 %z
+}
+; CHECK: r1:0 = vmpyweh(r1:0, r3:2):<<1:rnd:sat
+
+declare i64 @llvm.hexagon.M2.mmpyh.rs0(i64, i64)
+define i64 @M2_mmpyh_rs0(i64 %a, i64 %b) {
+  %z = call i64 @llvm.hexagon.M2.mmpyh.rs0(i64 %a, i64 %b)
+  ret i64 %z
+}
+; CHECK: r1:0 = vmpywoh(r1:0, r3:2):rnd:sat
+
+declare i64 @llvm.hexagon.M2.mmpyh.rs1(i64, i64)
+define i64 @M2_mmpyh_rs1(i64 %a, i64 %b) {
+  %z = call i64 @llvm.hexagon.M2.mmpyh.rs1(i64 %a, i64 %b)
+  ret i64 %z
+}
+; CHECK: r1:0 = vmpywoh(r1:0, r3:2):<<1:rnd:sat
+
+; Vector multiply word by unsigned half (32x16)
+declare i64 @llvm.hexagon.M2.mmpyul.s0(i64, i64)
+define i64 @M2_mmpyul_s0(i64 %a, i64 %b) {
+  %z = call i64 @llvm.hexagon.M2.mmpyul.s0(i64 %a, i64 %b)
+  ret i64 %z
+}
+; CHECK: r1:0 = vmpyweuh(r1:0, r3:2):sat
+
+declare i64 @llvm.hexagon.M2.mmpyul.s1(i64, i64)
+define i64 @M2_mmpyul_s1(i64 %a, i64 %b) {
+  %z = call i64 @llvm.hexagon.M2.mmpyul.s1(i64 %a, i64 %b)
+  ret i64 %z
+}
+; CHECK: r1:0 = vmpyweuh(r1:0, r3:2):<<1:sat
+
+declare i64 @llvm.hexagon.M2.mmpyuh.s0(i64, i64)
+define i64 @M2_mmpyuh_s0(i64 %a, i64 %b) {
+  %z = call i64 @llvm.hexagon.M2.mmpyuh.s0(i64 %a, i64 %b)
+  ret i64 %z
+}
+; CHECK: r1:0 = vmpywouh(r1:0, r3:2):sat
+
+declare i64 @llvm.hexagon.M2.mmpyuh.s1(i64, i64)
+define i64 @M2_mmpyuh_s1(i64 %a, i64 %b) {
+  %z = call i64 @llvm.hexagon.M2.mmpyuh.s1(i64 %a, i64 %b)
+  ret i64 %z
+}
+; CHECK: r1:0 = vmpywouh(r1:0, r3:2):<<1:sat
+
+declare i64 @llvm.hexagon.M2.mmpyul.rs0(i64, i64)
+define i64 @M2_mmpyul_rs0(i64 %a, i64 %b) {
+  %z = call i64 @llvm.hexagon.M2.mmpyul.rs0(i64 %a, i64 %b)
+  ret i64 %z
+}
+; CHECK: r1:0 = vmpyweuh(r1:0, r3:2):rnd:sat
+
+declare i64 @llvm.hexagon.M2.mmpyul.rs1(i64, i64)
+define i64 @M2_mmpyul_rs1(i64 %a, i64 %b) {
+  %z = call i64 @llvm.hexagon.M2.mmpyul.rs1(i64 %a, i64 %b)
+  ret i64 %z
+}
+; CHECK: r1:0 = vmpyweuh(r1:0, r3:2):<<1:rnd:sat
+
+declare i64 @llvm.hexagon.M2.mmpyuh.rs0(i64, i64)
+define i64 @M2_mmpyuh_rs0(i64 %a, i64 %b) {
+  %z = call i64 @llvm.hexagon.M2.mmpyuh.rs0(i64 %a, i64 %b)
+  ret i64 %z
+}
+; CHECK: r1:0 = vmpywouh(r1:0, r3:2):rnd:sat
+
+declare i64 @llvm.hexagon.M2.mmpyuh.rs1(i64, i64)
+define i64 @M2_mmpyuh_rs1(i64 %a, i64 %b) {
+  %z = call i64 @llvm.hexagon.M2.mmpyuh.rs1(i64 %a, i64 %b)
+  ret i64 %z
+}
+; CHECK: r1:0 = vmpywouh(r1:0, r3:2):<<1:rnd:sat
+
 ; Multiply signed halfwords
 declare i64 @llvm.hexagon.M2.mpyd.ll.s0(i32, i32)
 define i64 @M2_mpyd_ll_s0(i32 %a, i32 %b) {
@@ -1006,6 +1120,63 @@ define i64 @M4_pmpyw_acc(i64 %a, i32 %b, i32 %c) {
 }
 ; CHECK: r1:0 ^= pmpyw(r2, r3)
 
+; Vector reduce multiply word by signed half
+declare i64 @llvm.hexagon.M4.vrmpyoh.s0(i64, i64)
+define i64 @M4_vrmpyoh_s0(i64 %a, i64 %b) {
+  %z = call i64 @llvm.hexagon.M4.vrmpyoh.s0(i64 %a, i64 %b)
+  ret i64 %z
+}
+; CHECK: r1:0 = vrmpywoh(r1:0, r3:2)
+
+declare i64 @llvm.hexagon.M4.vrmpyoh.s1(i64, i64)
+define i64 @M4_vrmpyoh_s1(i64 %a, i64 %b) {
+  %z = call i64 @llvm.hexagon.M4.vrmpyoh.s1(i64 %a, i64 %b)
+  ret i64 %z
+}
+; CHECK: r1:0 = vrmpywoh(r1:0, r3:2):<<1
+
+declare i64 @llvm.hexagon.M4.vrmpyeh.s0(i64, i64)
+define i64 @M4_vrmpyeh_s0(i64 %a, i64 %b) {
+  %z = call i64 @llvm.hexagon.M4.vrmpyeh.s0(i64 %a, i64 %b)
+  ret i64 %z
+}
+; CHECK: r1:0 = vrmpyweh(r1:0, r3:2)
+
+declare i64 @llvm.hexagon.M4.vrmpyeh.s1(i64, i64)
+define i64 @M4_vrmpyeh_s1(i64 %a, i64 %b) {
+  %z = call i64 @llvm.hexagon.M4.vrmpyeh.s1(i64 %a, i64 %b)
+  ret i64 %z
+}
+; CHECK: r1:0 = vrmpyweh(r1:0, r3:2):<<1
+
+declare i64 @llvm.hexagon.M4.vrmpyoh.acc.s0(i64, i64, i64)
+define i64 @M4_vrmpyoh_acc_s0(i64 %a, i64 %b, i64 %c) {
+  %z = call i64 @llvm.hexagon.M4.vrmpyoh.acc.s0(i64 %a, i64 %b, i64 %c)
+  ret i64 %z
+}
+; CHECK: r1:0 += vrmpywoh(r3:2, r5:4)
+
+declare i64 @llvm.hexagon.M4.vrmpyoh.acc.s1(i64, i64, i64)
+define i64 @M4_vrmpyoh_acc_s1(i64 %a, i64 %b, i64 %c) {
+  %z = call i64 @llvm.hexagon.M4.vrmpyoh.acc.s1(i64 %a, i64 %b, i64 %c)
+  ret i64 %z
+}
+; CHECK: r1:0 += vrmpywoh(r3:2, r5:4):<<1
+
+declare i64 @llvm.hexagon.M4.vrmpyeh.acc.s0(i64, i64, i64)
+define i64 @M4_vrmpyeh_acc_s0(i64 %a, i64 %b, i64 %c) {
+  %z = call i64 @llvm.hexagon.M4.vrmpyeh.acc.s0(i64 %a, i64 %b, i64 %c)
+  ret i64 %z
+}
+; CHECK: r1:0 += vrmpyweh(r3:2, r5:4)
+
+declare i64 @llvm.hexagon.M4.vrmpyeh.acc.s1(i64, i64, i64)
+define i64 @M4_vrmpyeh_acc_s1(i64 %a, i64 %b, i64 %c) {
+  %z = call i64 @llvm.hexagon.M4.vrmpyeh.acc.s1(i64 %a, i64 %b, i64 %c)
+  ret i64 %z
+}
+; CHECK: r1:0 += vrmpyweh(r3:2, r5:4):<<1
+
 ; Multiply and use upper result
 declare i32 @llvm.hexagon.M2.dpmpyss.rnd.s0(i32, i32)
 define i32 @M2_dpmpyss_rnd_s0(i32 %a, i32 %b) {
@@ -1133,3 +1304,222 @@ define i64 @M2_dpmpyuu_nac_s0(i64 %a, i32 %b, i32 %c) {
   ret i64 %z
 }
 ; CHECK: r1:0 -= mpyu(r2, r3)
+
+; Vector dual multiply
+declare i64 @llvm.hexagon.M2.vdmpys.s0(i64, i64)
+define i64 @M2_vdmpys_s0(i64 %a, i64 %b) {
+  %z = call i64 @llvm.hexagon.M2.vdmpys.s0(i64 %a, i64 %b)
+  ret i64 %z
+}
+; CHECK: r1:0 = vdmpy(r1:0, r3:2):sat
+
+declare i64 @llvm.hexagon.M2.vdmpys.s1(i64, i64)
+define i64 @M2_vdmpys_s1(i64 %a, i64 %b) {
+  %z = call i64 @llvm.hexagon.M2.vdmpys.s1(i64 %a, i64 %b)
+  ret i64 %z
+}
+; CHECK: r1:0 = vdmpy(r1:0, r3:2):<<1:sat
+
+; Vector reduce multiply bytes
+declare i64 @llvm.hexagon.M5.vrmpybuu(i64, i64)
+define i64 @M5_vrmpybuu(i64 %a, i64 %b) {
+  %z = call i64 @llvm.hexagon.M5.vrmpybuu(i64 %a, i64 %b)
+  ret i64 %z
+}
+; CHECK: r1:0 = vrmpybu(r1:0, r3:2)
+
+declare i64 @llvm.hexagon.M5.vrmpybsu(i64, i64)
+define i64 @M5_vrmpybsu(i64 %a, i64 %b) {
+  %z = call i64 @llvm.hexagon.M5.vrmpybsu(i64 %a, i64 %b)
+  ret i64 %z
+}
+; CHECK: r1:0 = vrmpybsu(r1:0, r3:2)
+
+declare i64 @llvm.hexagon.M5.vrmacbuu(i64, i64, i64)
+define i64 @M5_vrmacbuu(i64 %a, i64 %b, i64 %c) {
+  %z = call i64 @llvm.hexagon.M5.vrmacbuu(i64 %a, i64 %b, i64 %c)
+  ret i64 %z
+}
+; CHECK: r1:0 += vrmpybu(r3:2, r5:4)
+
+declare i64 @llvm.hexagon.M5.vrmacbsu(i64, i64, i64)
+define i64 @M5_vrmacbsu(i64 %a, i64 %b, i64 %c) {
+  %z = call i64 @llvm.hexagon.M5.vrmacbsu(i64 %a, i64 %b, i64 %c)
+  ret i64 %z
+}
+; CHECK: r1:0 += vrmpybsu(r3:2, r5:4)
+
+; Vector dual multiply signed by unsigned bytes
+declare i64 @llvm.hexagon.M5.vdmpybsu(i64, i64)
+define i64 @M5_vdmpybsu(i64 %a, i64 %b) {
+  %z = call i64 @llvm.hexagon.M5.vdmpybsu(i64 %a, i64 %b)
+  ret i64 %z
+}
+; CHECK: r1:0 = vdmpybsu(r1:0, r3:2):sat
+
+declare i64 @llvm.hexagon.M5.vdmacbsu(i64, i64, i64)
+define i64 @M5_vdmacbsu(i64 %a, i64 %b, i64 %c) {
+  %z = call i64 @llvm.hexagon.M5.vdmacbsu(i64 %a, i64 %b, i64 %c)
+  ret i64 %z
+}
+; CHECK: r1:0 += vdmpybsu(r3:2, r5:4):sat
+
+; Vector multiply even halfwords
+declare i64 @llvm.hexagon.M2.vmpy2es.s0(i64, i64)
+define i64 @M2_vmpy2es_s0(i64 %a, i64 %b) {
+  %z = call i64 @llvm.hexagon.M2.vmpy2es.s0(i64 %a, i64 %b)
+  ret i64 %z
+}
+; CHECK: r1:0 = vmpyeh(r1:0, r3:2):sat
+
+declare i64 @llvm.hexagon.M2.vmpy2es.s1(i64, i64)
+define i64 @M2_vmpy2es_s1(i64 %a, i64 %b) {
+  %z = call i64 @llvm.hexagon.M2.vmpy2es.s1(i64 %a, i64 %b)
+  ret i64 %z
+}
+; CHECK: r1:0 = vmpyeh(r1:0, r3:2):<<1:sat
+
+declare i64 @llvm.hexagon.M2.vmac2es(i64, i64, i64)
+define i64 @M2_vmac2es(i64 %a, i64 %b, i64 %c) {
+  %z = call i64 @llvm.hexagon.M2.vmac2es(i64 %a, i64 %b, i64 %c)
+  ret i64 %z
+}
+; CHECK: r1:0 += vmpyeh(r3:2, r5:4)
+
+declare i64 @llvm.hexagon.M2.vmac2es.s0(i64, i64, i64)
+define i64 @M2_vmac2es_s0(i64 %a, i64 %b, i64 %c) {
+  %z = call i64 @llvm.hexagon.M2.vmac2es.s0(i64 %a, i64 %b, i64 %c)
+  ret i64 %z
+}
+; CHECK: r1:0 += vmpyeh(r3:2, r5:4):sat
+
+declare i64 @llvm.hexagon.M2.vmac2es.s1(i64, i64, i64)
+define i64 @M2_vmac2es_s1(i64 %a, i64 %b, i64 %c) {
+  %z = call i64 @llvm.hexagon.M2.vmac2es.s1(i64 %a, i64 %b, i64 %c)
+  ret i64 %z
+}
+; CHECK: r1:0 += vmpyeh(r3:2, r5:4):<<1:sat
+
+; Vector multiply halfwords
+declare i64 @llvm.hexagon.M2.vmpy2s.s0(i32, i32)
+define i64 @M2_vmpy2s_s0(i32 %a, i32 %b) {
+  %z = call i64 @llvm.hexagon.M2.vmpy2s.s0(i32 %a, i32 %b)
+  ret i64 %z
+}
+; CHECK: r1:0 = vmpyh(r0, r1):sat
+
+declare i64 @llvm.hexagon.M2.vmpy2s.s1(i32, i32)
+define i64 @M2_vmpy2s_s1(i32 %a, i32 %b) {
+  %z = call i64 @llvm.hexagon.M2.vmpy2s.s1(i32 %a, i32 %b)
+  ret i64 %z
+}
+; CHECK: r1:0 = vmpyh(r0, r1):<<1:sat
+
+declare i64 @llvm.hexagon.M2.vmac2(i64, i32, i32)
+define i64 @M2_vmac2(i64 %a, i32 %b, i32 %c) {
+  %z = call i64 @llvm.hexagon.M2.vmac2(i64 %a, i32 %b, i32 %c)
+  ret i64 %z
+}
+; CHECK: r1:0 += vmpyh(r2, r3)
+
+declare i64 @llvm.hexagon.M2.vmac2s.s0(i64, i32, i32)
+define i64 @M2_vmac2s_s0(i64 %a, i32 %b, i32 %c) {
+  %z = call i64 @llvm.hexagon.M2.vmac2s.s0(i64 %a, i32 %b, i32 %c)
+  ret i64 %z
+}
+; CHECK: r1:0 += vmpyh(r2, r3):sat
+
+declare i64 @llvm.hexagon.M2.vmac2s.s1(i64, i32, i32)
+define i64 @M2_vmac2s_s1(i64 %a, i32 %b, i32 %c) {
+  %z = call i64 @llvm.hexagon.M2.vmac2s.s1(i64 %a, i32 %b, i32 %c)
+  ret i64 %z
+}
+; CHECK: r1:0 += vmpyh(r2, r3):<<1:sat
+
+; Vector multiply halfwords signed by unsigned
+declare i64 @llvm.hexagon.M2.vmpy2su.s0(i32, i32)
+define i64 @M2_vmpy2su_s0(i32 %a, i32 %b) {
+  %z = call i64 @llvm.hexagon.M2.vmpy2su.s0(i32 %a, i32 %b)
+  ret i64 %z
+}
+; CHECK: r1:0 = vmpyhsu(r0, r1):sat
+
+declare i64 @llvm.hexagon.M2.vmpy2su.s1(i32, i32)
+define i64 @M2_vmpy2su_s1(i32 %a, i32 %b) {
+  %z = call i64 @llvm.hexagon.M2.vmpy2su.s1(i32 %a, i32 %b)
+  ret i64 %z
+}
+; CHECK: r1:0 = vmpyhsu(r0, r1):<<1:sat
+
+declare i64 @llvm.hexagon.M2.vmac2su.s0(i64, i32, i32)
+define i64 @M2_vmac2su_s0(i64 %a, i32 %b, i32 %c) {
+  %z = call i64 @llvm.hexagon.M2.vmac2su.s0(i64 %a, i32 %b, i32 %c)
+  ret i64 %z
+}
+; CHECK: r1:0 += vmpyhsu(r2, r3):sat
+
+declare i64 @llvm.hexagon.M2.vmac2su.s1(i64, i32, i32)
+define i64 @M2_vmac2su_s1(i64 %a, i32 %b, i32 %c) {
+  %z = call i64 @llvm.hexagon.M2.vmac2su.s1(i64 %a, i32 %b, i32 %c)
+  ret i64 %z
+}
+; CHECK: r1:0 += vmpyhsu(r2, r3):<<1:sat
+
+; Vector reduce multiply halfwords
+declare i64 @llvm.hexagon.M2.vrmpy.s0(i64, i64)
+define i64 @M2_vrmpy_s0(i64 %a, i64 %b) {
+  %z = call i64 @llvm.hexagon.M2.vrmpy.s0(i64 %a, i64 %b)
+  ret i64 %z
+}
+; CHECK: r1:0 = vrmpyh(r1:0, r3:2)
+
+declare i64 @llvm.hexagon.M2.vrmac.s0(i64, i64, i64)
+define i64 @M2_vrmac_s0(i64 %a, i64 %b, i64 %c) {
+  %z = call i64 @llvm.hexagon.M2.vrmac.s0(i64 %a, i64 %b, i64 %c)
+  ret i64 %z
+}
+; CHECK: r1:0 += vrmpyh(r3:2, r5:4)
+
+; Vector multiply bytes
+declare i64 @llvm.hexagon.M5.vmpybsu(i32, i32)
+define i64 @M2_vmpybsu(i32 %a, i32 %b) {
+  %z = call i64 @llvm.hexagon.M5.vmpybsu(i32 %a, i32 %b)
+  ret i64 %z
+}
+; CHECK: r1:0 = vmpybsu(r0, r1)
+
+declare i64 @llvm.hexagon.M5.vmpybuu(i32, i32)
+define i64 @M2_vmpybuu(i32 %a, i32 %b) {
+  %z = call i64 @llvm.hexagon.M5.vmpybuu(i32 %a, i32 %b)
+  ret i64 %z
+}
+; CHECK: r1:0 = vmpybu(r0, r1)
+
+declare i64 @llvm.hexagon.M5.vmacbuu(i64, i32, i32)
+define i64 @M2_vmacbuu(i64 %a, i32 %b, i32 %c) {
+  %z = call i64 @llvm.hexagon.M5.vmacbuu(i64 %a, i32 %b, i32 %c)
+  ret i64 %z
+}
+; CHECK: r1:0 += vmpybu(r2, r3)
+
+declare i64 @llvm.hexagon.M5.vmacbsu(i64, i32, i32)
+define i64 @M2_vmacbsu(i64 %a, i32 %b, i32 %c) {
+  %z = call i64 @llvm.hexagon.M5.vmacbsu(i64 %a, i32 %b, i32 %c)
+  ret i64 %z
+}
+; CHECK: r1:0 += vmpybsu(r2, r3)
+
+; Vector polynomial multiply halfwords
+declare i64 @llvm.hexagon.M4.vpmpyh(i32, i32)
+define i64 @M4_vpmpyh(i32 %a, i32 %b) {
+  %z = call i64 @llvm.hexagon.M4.vpmpyh(i32 %a, i32 %b)
+  ret i64 %z
+}
+; CHECK: r1:0 = vpmpyh(r0, r1)
+
+declare i64 @llvm.hexagon.M4.vpmpyh.acc(i64, i32, i32)
+define i64 @M4_vpmpyh_acc(i64 %a, i32 %b, i32 %c) {
+  %z = call i64 @llvm.hexagon.M4.vpmpyh.acc(i64 %a, i32 %b, i32 %c)
+  ret i64 %z
+}
+; CHECK: r1:0 ^= vpmpyh(r2, r3)
diff --git a/test/MC/Disassembler/Hexagon/xtype_alu.txt b/test/MC/Disassembler/Hexagon/xtype_alu.txt
index d44787d69f5..03d0f0518a3 100644
--- a/test/MC/Disassembler/Hexagon/xtype_alu.txt
+++ b/test/MC/Disassembler/Hexagon/xtype_alu.txt
@@ -219,6 +219,10 @@
 0x10 0xd4 0x7e 0xe8
 # CHECK: r17:16 = vabsdiffh(r21:20, r31:30)
 
+# Vector absolute difference words
+0x10 0xd4 0x3e 0xe8
+# CHECK: r17:16 = vabsdiffw(r21:20, r31:30)
+
 # Vector add halfwords
 0x50 0xde 0x14 0xd3
 # CHECK: r17:16 = vaddh(r21:20, r31:30)
diff --git a/test/MC/Disassembler/Hexagon/xtype_mpy.txt b/test/MC/Disassembler/Hexagon/xtype_mpy.txt
index 970c379f35f..ada32162a81 100644
--- a/test/MC/Disassembler/Hexagon/xtype_mpy.txt
+++ b/test/MC/Disassembler/Hexagon/xtype_mpy.txt
@@ -42,6 +42,22 @@
 # CHECK: r17:16 = vmpywoh(r21:20, r31:30):rnd:sat
 0xf0 0xde 0xb4 0xe8
 # CHECK: r17:16 = vmpywoh(r21:20, r31:30):<<1:rnd:sat
+0xb0 0xde 0x14 0xea
+# CHECK: r17:16 += vmpyweh(r21:20, r31:30):sat
+0xb0 0xde 0x94 0xea
+# CHECK: r17:16 += vmpyweh(r21:20, r31:30):<<1:sat
+0xf0 0xde 0x14 0xea
+# CHECK: r17:16 += vmpywoh(r21:20, r31:30):sat
+0xf0 0xde 0x94 0xea
+# CHECK: r17:16 += vmpywoh(r21:20, r31:30):<<1:sat
+0xb0 0xde 0x34 0xea
+# CHECK: r17:16 += vmpyweh(r21:20, r31:30):rnd:sat
+0xb0 0xde 0xb4 0xea
+# CHECK: r17:16 += vmpyweh(r21:20, r31:30):<<1:rnd:sat
+0xf0 0xde 0x34 0xea
+# CHECK: r17:16 += vmpywoh(r21:20, r31:30):rnd:sat
+0xf0 0xde 0xb4 0xea
+# CHECK: r17:16 += vmpywoh(r21:20, r31:30):<<1:rnd:sat
 
 # Vector multiply word by unsigned half (32x16)
 0xb0 0xde 0x54 0xe8
@@ -60,6 +76,22 @@
 # CHECK: r17:16 = vmpywouh(r21:20, r31:30):rnd:sat
 0xf0 0xde 0xf4 0xe8
 # CHECK: r17:16 = vmpywouh(r21:20, r31:30):<<1:rnd:sat
+0xb0 0xde 0x54 0xea
+# CHECK: r17:16 += vmpyweuh(r21:20, r31:30):sat
+0xb0 0xde 0xd4 0xea
+# CHECK: r17:16 += vmpyweuh(r21:20, r31:30):<<1:sat
+0xf0 0xde 0x54 0xea
+# CHECK: r17:16 += vmpywouh(r21:20, r31:30):sat
+0xf0 0xde 0xd4 0xea
+# CHECK: r17:16 += vmpywouh(r21:20, r31:30):<<1:sat
+0xb0 0xde 0x74 0xea
+# CHECK: r17:16 += vmpyweuh(r21:20, r31:30):rnd:sat
+0xb0 0xde 0xf4 0xea
+# CHECK: r17:16 += vmpyweuh(r21:20, r31:30):<<1:rnd:sat
+0xf0 0xde 0x74 0xea
+# CHECK: r17:16 += vmpywouh(r21:20, r31:30):rnd:sat
+0xf0 0xde 0xf4 0xea
+# CHECK: r17:16 += vmpywouh(r21:20, r31:30):<<1:rnd:sat
 
 # Multiply signed halfwords
 0x10 0xdf 0x95 0xe4