[Hexagon] Adding doubleworld accumulating multiplies of halfwords.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@224267 91177308-0d34-0410-b5e6-96231b3b80d8

lib/Target/Hexagon/HexagonInstrInfo.td

@@ -1934,6 +1934,80 @@ def M2_mpy_nac_sat_hh_s1: T_M2_mpy_acc <0b11, 1, 1, 1, 0>;
 def M2_mpy_nac_sat_hh_s0: T_M2_mpy_acc <0b11, 1, 1, 0, 0>;
 }
 
+//===----------------------------------------------------------------------===//
+// Template Class
+// MPYS / Multipy signed/unsigned halfwords and add/subtract the
+// result from the 64-bit destination register.
+//Rxx [-+]= mpy[u](Rs.[H|L],Rt.[H|L])[:<<1][:sat]
+//===----------------------------------------------------------------------===//
+
+class T_M2_mpyd_acc < bits<2> LHbits, bit isNac, bit hasShift, bit isUnsigned>
+  : MInst_acc<(outs DoubleRegs:$Rxx),
+              (ins DoubleRegs:$dst2, IntRegs:$Rs, IntRegs:$Rt),
+  "$Rxx "#!if(isNac,"-= ","+= ")#!if(isUnsigned,"mpyu","mpy")
+                                #"($Rs."#!if(LHbits{1},"h","l")
+                                #", $Rt."#!if(LHbits{0},"h)","l)")
+                                #!if(hasShift,":<<1",""),
+  [], "$dst2 = $Rxx", M_tc_3x_SLOT23 > {
+    bits<5> Rxx;
+    bits<5> Rs;
+    bits<5> Rt;
+
+    let IClass = 0b1110;
+
+    let Inst{27-24} = 0b0110;
+    let Inst{23} = hasShift;
+    let Inst{22} = isUnsigned;
+    let Inst{21} = isNac;
+    let Inst{7} = 0;
+    let Inst{6-5} = LHbits;
+    let Inst{4-0} = Rxx;
+    let Inst{20-16} = Rs;
+    let Inst{12-8} = Rt;
+  }
+
+let isCodeGenOnly = 0 in {
+def M2_mpyd_acc_hh_s0: T_M2_mpyd_acc <0b11, 0, 0, 0>;
+def M2_mpyd_acc_hl_s0: T_M2_mpyd_acc <0b10, 0, 0, 0>;
+def M2_mpyd_acc_lh_s0: T_M2_mpyd_acc <0b01, 0, 0, 0>;
+def M2_mpyd_acc_ll_s0: T_M2_mpyd_acc <0b00, 0, 0, 0>;
+
+def M2_mpyd_acc_hh_s1: T_M2_mpyd_acc <0b11, 0, 1, 0>;
+def M2_mpyd_acc_hl_s1: T_M2_mpyd_acc <0b10, 0, 1, 0>;
+def M2_mpyd_acc_lh_s1: T_M2_mpyd_acc <0b01, 0, 1, 0>;
+def M2_mpyd_acc_ll_s1: T_M2_mpyd_acc <0b00, 0, 1, 0>;
+
+def M2_mpyd_nac_hh_s0: T_M2_mpyd_acc <0b11, 1, 0, 0>;
+def M2_mpyd_nac_hl_s0: T_M2_mpyd_acc <0b10, 1, 0, 0>;
+def M2_mpyd_nac_lh_s0: T_M2_mpyd_acc <0b01, 1, 0, 0>;
+def M2_mpyd_nac_ll_s0: T_M2_mpyd_acc <0b00, 1, 0, 0>;
+
+def M2_mpyd_nac_hh_s1: T_M2_mpyd_acc <0b11, 1, 1, 0>;
+def M2_mpyd_nac_hl_s1: T_M2_mpyd_acc <0b10, 1, 1, 0>;
+def M2_mpyd_nac_lh_s1: T_M2_mpyd_acc <0b01, 1, 1, 0>;
+def M2_mpyd_nac_ll_s1: T_M2_mpyd_acc <0b00, 1, 1, 0>;
+
+def M2_mpyud_acc_hh_s0: T_M2_mpyd_acc <0b11, 0, 0, 1>;
+def M2_mpyud_acc_hl_s0: T_M2_mpyd_acc <0b10, 0, 0, 1>;
+def M2_mpyud_acc_lh_s0: T_M2_mpyd_acc <0b01, 0, 0, 1>;
+def M2_mpyud_acc_ll_s0: T_M2_mpyd_acc <0b00, 0, 0, 1>;
+
+def M2_mpyud_acc_hh_s1: T_M2_mpyd_acc <0b11, 0, 1, 1>;
+def M2_mpyud_acc_hl_s1: T_M2_mpyd_acc <0b10, 0, 1, 1>;
+def M2_mpyud_acc_lh_s1: T_M2_mpyd_acc <0b01, 0, 1, 1>;
+def M2_mpyud_acc_ll_s1: T_M2_mpyd_acc <0b00, 0, 1, 1>;
+
+def M2_mpyud_nac_hh_s0: T_M2_mpyd_acc <0b11, 1, 0, 1>;
+def M2_mpyud_nac_hl_s0: T_M2_mpyd_acc <0b10, 1, 0, 1>;
+def M2_mpyud_nac_lh_s0: T_M2_mpyd_acc <0b01, 1, 0, 1>;
+def M2_mpyud_nac_ll_s0: T_M2_mpyd_acc <0b00, 1, 0, 1>;
+
+def M2_mpyud_nac_hh_s1: T_M2_mpyd_acc <0b11, 1, 1, 1>;
+def M2_mpyud_nac_hl_s1: T_M2_mpyd_acc <0b10, 1, 1, 1>;
+def M2_mpyud_nac_lh_s1: T_M2_mpyd_acc <0b01, 1, 1, 1>;
+def M2_mpyud_nac_ll_s1: T_M2_mpyd_acc <0b00, 1, 1, 1>;
+}
+
 // Multiply and use lower result.
 // Rd=+mpyi(Rs,#u8)
 let isExtendable = 1, opExtendable = 2, isExtentSigned = 0, opExtentBits = 8 in

test/MC/Disassembler/Hexagon/xtype_mpy.txt

@@ -1,5 +1,21 @@
 # RUN: llvm-mc --triple hexagon -disassemble < %s | FileCheck %s
 
+0x10 0xdf 0x95 0xe6
+# CHECK: r17:16 += mpy(r21.l, r31.l):<<1
+0x30 0xdf 0x95 0xe6
+# CHECK: r17:16 += mpy(r21.l, r31.h):<<1
+0x50 0xdf 0x95 0xe6
+# CHECK: r17:16 += mpy(r21.h, r31.l):<<1
+0x70 0xdf 0x95 0xe6
+# CHECK: r17:16 += mpy(r21.h, r31.h):<<1
+0x10 0xdf 0xb5 0xe6
+# CHECK: r17:16 -= mpy(r21.l, r31.l):<<1
+0x30 0xdf 0xb5 0xe6
+# CHECK: r17:16 -= mpy(r21.l, r31.h):<<1
+0x50 0xdf 0xb5 0xe6
+# CHECK: r17:16 -= mpy(r21.h, r31.l):<<1
+0x70 0xdf 0xb5 0xe6
+# CHECK: r17:16 -= mpy(r21.h, r31.h):<<1
 0x11 0xdf 0x95 0xec
 # CHECK: r17 = mpy(r21.l, r31.l):<<1
 0x31 0xdf 0x95 0xec
@@ -64,6 +80,22 @@
 # CHECK: r17 -= mpy(r21.h, r31.l):<<1:sat
 0xf1 0xdf 0xb5 0xee
 # CHECK: r17 -= mpy(r21.h, r31.h):<<1:sat
+0x10 0xdf 0xd5 0xe6
+# CHECK: r17:16 += mpyu(r21.l, r31.l):<<1
+0x30 0xdf 0xd5 0xe6
+# CHECK: r17:16 += mpyu(r21.l, r31.h):<<1
+0x50 0xdf 0xd5 0xe6
+# CHECK: r17:16 += mpyu(r21.h, r31.l):<<1
+0x70 0xdf 0xd5 0xe6
+# CHECK: r17:16 += mpyu(r21.h, r31.h):<<1
+0x10 0xdf 0xf5 0xe6
+# CHECK: r17:16 -= mpyu(r21.l, r31.l):<<1
+0x30 0xdf 0xf5 0xe6
+# CHECK: r17:16 -= mpyu(r21.l, r31.h):<<1
+0x50 0xdf 0xf5 0xe6
+# CHECK: r17:16 -= mpyu(r21.h, r31.l):<<1
+0x70 0xdf 0xf5 0xe6
+# CHECK: r17:16 -= mpyu(r21.h, r31.h):<<1
 0x11 0xdf 0xd5 0xec
 # CHECK: r17 = mpyu(r21.l, r31.l):<<1
 0x31 0xdf 0xd5 0xec