isa: Add parameter to pick different decoder inside ISA

The decoder is responsible for splitting instructions in micro
operations (uops). Given that different micro architectures may split
operations differently, this patch allows to specify which micro
architecture each isa implements, so different cores in the system can
split instructions differently, also decoupling uop splitting
(microArch) from ISA (Arch). This is done making the decodification
calls templates that receive a type 'DecoderFlavour' that maps the
name of the operation to the class that implements it. This way there
is only one selection point (converting the command line enum to the
appropriate DecodeFeatures object). In addition, there is no explicit
code replication: template instantiation hides that, and the compiler
should be able to resolve a number of things at compile-time.

src/arch/alpha/decoder.hh

@@ -39,6 +39,7 @@
 namespace AlphaISA
 {
 
+class ISA;
 class Decoder
 {
   protected:
@@ -47,7 +48,7 @@ class Decoder
     bool instDone;
 
   public:
-    Decoder() : instDone(false)
+    Decoder(ISA* isa = nullptr) : instDone(false)
     {}
 
     void

src/arch/arm/ArmISA.py

@@ -1,4 +1,4 @@
-# Copyright (c) 2012-2013 ARM Limited
+# Copyright (c) 2012-2013, 2015 ARM Limited
 # All rights reserved.
 #
 # The license below extends only to copyright in the software and shall
@@ -42,6 +42,9 @@ from m5.SimObject import SimObject
 
 from ArmPMU import ArmPMU
 
+# Enum for DecoderFlavour
+class DecoderFlavour(Enum): vals = ['Generic']
+
 class ArmISA(SimObject):
     type = 'ArmISA'
     cxx_class = 'ArmISA::ISA'
@@ -50,6 +53,7 @@ class ArmISA(SimObject):
     system = Param.System(Parent.any, "System this ISA object belongs to")
 
     pmu = Param.ArmPMU(NULL, "Performance Monitoring Unit")
+    decoderFlavour = Param.DecoderFlavour('Generic', "Decoder flavour specification")
 
     midr = Param.UInt32(0x410fc0f0, "MIDR value")
 

src/arch/arm/decoder.cc

@@ -41,6 +41,8 @@
  */
 
 #include "arch/arm/decoder.hh"
+
+#include "arch/arm/isa.hh"
 #include "arch/arm/isa_traits.hh"
 #include "arch/arm/utility.hh"
 #include "base/trace.hh"
@@ -51,8 +53,10 @@ namespace ArmISA
 
 GenericISA::BasicDecodeCache Decoder::defaultCache;
 
-Decoder::Decoder()
+Decoder::Decoder(ISA* isa)
-    : data(0), fpscrLen(0), fpscrStride(0)
+    : data(0), fpscrLen(0), fpscrStride(0), decoderFlavour(isa
+            ? isa->decoderFlavour()
+            : Enums::Generic)
 {
     reset();
 }

src/arch/arm/decoder.hh

@@ -50,10 +50,12 @@
 #include "arch/generic/decode_cache.hh"
 #include "base/types.hh"
 #include "cpu/static_inst.hh"
+#include "enums/DecoderFlavour.hh"
 
 namespace ArmISA
 {
 
+class ISA;
 class Decoder
 {
   protected:
@@ -70,6 +72,8 @@ class Decoder
     int fpscrLen;
     int fpscrStride;
 
+    Enums::DecoderFlavour decoderFlavour;
+
     /// A cache of decoded instruction objects.
     static GenericISA::BasicDecodeCache defaultCache;
 
@@ -86,7 +90,7 @@ class Decoder
     void consumeBytes(int numBytes);
 
   public: // Decoder API
-    Decoder();
+    Decoder(ISA* isa = nullptr);
 
     /** Reset the decoders internal state. */
     void reset();

src/arch/arm/isa.cc

@@ -127,6 +127,7 @@ const struct ISA::MiscRegInitializerEntry
 ISA::ISA(Params *p)
     : SimObject(p),
       system(NULL),
+      _decoderFlavour(p->decoderFlavour),
       pmu(p->pmu),
       lookUpMiscReg(NUM_MISCREGS, {0,0})
 {

src/arch/arm/isa.hh

@@ -50,6 +50,7 @@
 #include "arch/arm/types.hh"
 #include "debug/Checkpoint.hh"
 #include "sim/sim_object.hh"
+#include "enums/DecoderFlavour.hh"
 
 struct ArmISAParams;
 struct DummyArmISADeviceParams;
@@ -132,6 +133,9 @@ namespace ArmISA
         // Parent system
         ArmSystem *system;
 
+        // Micro Architecture
+        const Enums::DecoderFlavour _decoderFlavour;
+
         /** Dummy device for to handle non-existing ISA devices */
         DummyISADevice dummyDevice;
 
@@ -429,6 +433,8 @@ namespace ArmISA
 
         void startup(ThreadContext *tc) {}
 
+        Enums::DecoderFlavour decoderFlavour() const { return _decoderFlavour; }
+
         /// Explicitly import the otherwise hidden startup
         using SimObject::startup;
 

src/arch/arm/isa/formats/aarch64.isa

@@ -1,4 +1,4 @@
-// Copyright (c) 2011-2014 ARM Limited
+// Copyright (c) 2011-2015 ARM Limited
 // All rights reserved
 //
 // The license below extends only to copyright in the software and shall
@@ -46,8 +46,10 @@ namespace Aarch64
     StaticInstPtr decodeLoadsStores(ExtMachInst machInst);
     StaticInstPtr decodeDataProcReg(ExtMachInst machInst);
 
+    template <typename DecoderFeatures>
     StaticInstPtr decodeFpAdvSIMD(ExtMachInst machInst);
     StaticInstPtr decodeFp(ExtMachInst machInst);
+    template <typename DecoderFeatures>
     StaticInstPtr decodeAdvSIMD(ExtMachInst machInst);
     StaticInstPtr decodeAdvSIMDScalar(ExtMachInst machInst);
 
@@ -1278,12 +1280,13 @@ namespace Aarch64
 output decoder {{
 namespace Aarch64
 {
+    template <typename DecoderFeatures>
     StaticInstPtr
     decodeAdvSIMD(ExtMachInst machInst)
     {
         if (bits(machInst, 24) == 1) {
             if (bits(machInst, 10) == 0) {
-                return decodeNeonIndexedElem(machInst);
+                return decodeNeonIndexedElem<DecoderFeatures>(machInst);
             } else if (bits(machInst, 23) == 1) {
                 return new Unknown64(machInst);
             } else {
@@ -1295,7 +1298,7 @@ namespace Aarch64
             }
         } else if (bits(machInst, 21) == 1) {
             if (bits(machInst, 10) == 1) {
-                return decodeNeon3Same(machInst);
+                return decodeNeon3Same<DecoderFeatures>(machInst);
             } else if (bits(machInst, 11) == 0) {
                 return decodeNeon3Diff(machInst);
             } else if (bits(machInst, 20, 17) == 0x0) {
@@ -1957,13 +1960,14 @@ namespace Aarch64
 output decoder {{
 namespace Aarch64
 {
+    template <typename DecoderFeatures>
     StaticInstPtr
     decodeFpAdvSIMD(ExtMachInst machInst)
     {
 
         if (bits(machInst, 28) == 0) {
             if (bits(machInst, 31) == 0) {
-                return decodeAdvSIMD(machInst);
+                return decodeAdvSIMD<DecoderFeatures>(machInst);
             } else {
                 return new Unknown64(machInst);
             }
@@ -1978,6 +1982,18 @@ namespace Aarch64
 }
 }};
 
+let {{
+    decoder_output ='''
+namespace Aarch64
+{'''
+    for decoderFlavour, type_dict in decoders.iteritems():
+        decoder_output +='''
+template StaticInstPtr decodeFpAdvSIMD<%(df)sDecoder>(ExtMachInst machInst);
+''' % { "df" : decoderFlavour }
+    decoder_output +='''
+}'''
+}};
+
 output decoder {{
 namespace Aarch64
 {
@@ -2041,7 +2057,10 @@ def format Aarch64() {{
             return decodeGem5Ops(machInst);
         } else {
             // bit 27:25=111
-            return decodeFpAdvSIMD(machInst);
+            switch(decoderFlavour){
+            default:
+                return decodeFpAdvSIMD<GenericDecoder>(machInst);
+            }
         }
     }
     '''

src/arch/arm/isa/formats/neon64.isa

@@ -40,51 +40,54 @@ output header {{
 namespace Aarch64
 {
     // AdvSIMD three same
+    template <typename DecoderFeatures>
     StaticInstPtr decodeNeon3Same(ExtMachInst machInst);
     // AdvSIMD three different
-    StaticInstPtr decodeNeon3Diff(ExtMachInst machInst);
+    inline StaticInstPtr decodeNeon3Diff(ExtMachInst machInst);
     // AdvSIMD two-reg misc
-    StaticInstPtr decodeNeon2RegMisc(ExtMachInst machInst);
+    inline StaticInstPtr decodeNeon2RegMisc(ExtMachInst machInst);
     // AdvSIMD across lanes
-    StaticInstPtr decodeNeonAcrossLanes(ExtMachInst machInst);
+    inline StaticInstPtr decodeNeonAcrossLanes(ExtMachInst machInst);
     // AdvSIMD copy
-    StaticInstPtr decodeNeonCopy(ExtMachInst machInst);
+    inline StaticInstPtr decodeNeonCopy(ExtMachInst machInst);
     // AdvSIMD vector x indexed element
+    template <typename DecoderFeatures>
     StaticInstPtr decodeNeonIndexedElem(ExtMachInst machInst);
     // AdvSIMD modified immediate
-    StaticInstPtr decodeNeonModImm(ExtMachInst machInst);
+    inline StaticInstPtr decodeNeonModImm(ExtMachInst machInst);
     // AdvSIMD shift by immediate
-    StaticInstPtr decodeNeonShiftByImm(ExtMachInst machInst);
+    inline StaticInstPtr decodeNeonShiftByImm(ExtMachInst machInst);
     // AdvSIMD TBL/TBX
-    StaticInstPtr decodeNeonTblTbx(ExtMachInst machInst);
+    inline StaticInstPtr decodeNeonTblTbx(ExtMachInst machInst);
     // AdvSIMD ZIP/UZP/TRN
-    StaticInstPtr decodeNeonZipUzpTrn(ExtMachInst machInst);
+    inline StaticInstPtr decodeNeonZipUzpTrn(ExtMachInst machInst);
     // AdvSIMD EXT
-    StaticInstPtr decodeNeonExt(ExtMachInst machInst);
+    inline StaticInstPtr decodeNeonExt(ExtMachInst machInst);
 
     // AdvSIMD scalar three same
-    StaticInstPtr decodeNeonSc3Same(ExtMachInst machInst);
+    inline StaticInstPtr decodeNeonSc3Same(ExtMachInst machInst);
     // AdvSIMD scalar three different
-    StaticInstPtr decodeNeonSc3Diff(ExtMachInst machInst);
+    inline StaticInstPtr decodeNeonSc3Diff(ExtMachInst machInst);
     // AdvSIMD scalar two-reg misc
-    StaticInstPtr decodeNeonSc2RegMisc(ExtMachInst machInst);
+    inline StaticInstPtr decodeNeonSc2RegMisc(ExtMachInst machInst);
     // AdvSIMD scalar pairwise
-    StaticInstPtr decodeNeonScPwise(ExtMachInst machInst);
+    inline StaticInstPtr decodeNeonScPwise(ExtMachInst machInst);
     // AdvSIMD scalar copy
-    StaticInstPtr decodeNeonScCopy(ExtMachInst machInst);
+    inline StaticInstPtr decodeNeonScCopy(ExtMachInst machInst);
     // AdvSIMD scalar x indexed element
-    StaticInstPtr decodeNeonScIndexedElem(ExtMachInst machInst);
+    inline StaticInstPtr decodeNeonScIndexedElem(ExtMachInst machInst);
     // AdvSIMD scalar shift by immediate
-    StaticInstPtr decodeNeonScShiftByImm(ExtMachInst machInst);
+    inline StaticInstPtr decodeNeonScShiftByImm(ExtMachInst machInst);
 
     // AdvSIMD load/store
-    StaticInstPtr decodeNeonMem(ExtMachInst machInst);
+    inline StaticInstPtr decodeNeonMem(ExtMachInst machInst);
 }
 }};
 
 output decoder {{
 namespace Aarch64
 {
+    template <typename DecoderFeatures>
     StaticInstPtr
     decodeNeon3Same(ExtMachInst machInst)
     {
@@ -1267,6 +1270,7 @@ namespace Aarch64
         }
     }
 
+    template <typename DecoderFeatures>
     StaticInstPtr
     decodeNeonIndexedElem(ExtMachInst machInst)
     {

src/arch/arm/isa/includes.isa

@@ -67,6 +67,7 @@ output header {{
 #include "arch/arm/isa_traits.hh"
 #include "mem/packet.hh"
 #include "sim/faults.hh"
+#include "enums/DecoderFlavour.hh"
 }};
 
 output decoder {{

src/arch/arm/isa/insts/neon64.isa

@@ -1,6 +1,6 @@
 // -*- mode: c++ -*-
 
-// Copyright (c) 2012-2013 ARM Limited
+// Copyright (c) 2012-2013, 2015 ARM Limited
 // All rights reserved
 //
 // The license below extends only to copyright in the software and shall
@@ -42,6 +42,7 @@ let {{
 
     header_output = ""
     exec_output = ""
+    decoders = { 'Generic' : {} }
 
     # FP types (FP operations always work with unsigned representations)
     floatTypes = ("uint32_t", "uint64_t")
@@ -49,9 +50,9 @@ let {{
 
     def threeEqualRegInstX(name, Name, opClass, types, rCount, op,
                            readDest=False, pairwise=False, scalar=False,
-                           byElem=False):
+                           byElem=False, decoder='Generic'):
         assert (not pairwise) or ((not byElem) and (not scalar))
-        global header_output, exec_output
+        global header_output, exec_output, decoders
         eWalkCode = simd64EnabledCheckCode + '''
         RegVect srcReg1, destReg;
         '''
@@ -3356,4 +3357,16 @@ let {{
     threeRegScrambleInstX("zip2", "Zip2QX", "SimdAluOp", unsignedTypes, 4,
                           zipCode % "eCount / 2")
 
+    for decoderFlavour, type_dict in decoders.iteritems():
+        header_output += '''
+        class %(decoder_flavour)sDecoder {
+        public:
+        ''' % { "decoder_flavour" : decoderFlavour }
+        for type,name in type_dict.iteritems():
+            header_output += '''
+            template<typename Elem> using %(type)s = %(new_name)s<Elem>;''' % {
+               "type" : type, "new_name" : name
+            }
+        header_output += '''
+        };'''
 }};

src/arch/mips/decoder.hh

@@ -40,6 +40,7 @@
 namespace MipsISA
 {
 
+class ISA;
 class Decoder
 {
   protected:
@@ -48,7 +49,7 @@ class Decoder
     bool instDone;
 
   public:
-    Decoder() : instDone(false)
+    Decoder(ISA* isa = nullptr) : instDone(false)
     {}
 
     void

src/arch/power/decoder.hh

@@ -38,6 +38,7 @@
 namespace PowerISA
 {
 
+class ISA;
 class Decoder
 {
   protected:
@@ -46,7 +47,7 @@ class Decoder
     bool instDone;
 
   public:
-    Decoder() : instDone(false)
+    Decoder(ISA* isa = nullptr) : instDone(false)
     {
     }
 

src/arch/sparc/decoder.hh

@@ -39,6 +39,7 @@
 namespace SparcISA
 {
 
+class ISA;
 class Decoder
 {
   protected:
@@ -48,7 +49,7 @@ class Decoder
     MiscReg asi;
 
   public:
-    Decoder() : instDone(false), asi(0)
+    Decoder(ISA* isa = nullptr) : instDone(false), asi(0)
     {}
 
     void process() {}

src/arch/x86/decoder.hh

@@ -47,6 +47,7 @@
 namespace X86ISA
 {
 
+class ISA;
 class Decoder
 {
   private:
@@ -230,7 +231,7 @@ class Decoder
     static InstCacheMap instCacheMap;
 
   public:
-    Decoder() : basePC(0), origPC(0), offset(0),
+    Decoder(ISA* isa = nullptr) : basePC(0), origPC(0), offset(0),
         outOfBytes(true), instDone(false),
         state(ResetState)
     {

src/cpu/o3/fetch_impl.hh

@@ -73,6 +73,7 @@
 #include "sim/eventq.hh"
 #include "sim/full_system.hh"
 #include "sim/system.hh"
+#include "cpu/o3/isa_specific.hh"
 
 using namespace std;
 
@@ -151,7 +152,7 @@ DefaultFetch<Impl>::DefaultFetch(O3CPU *_cpu, DerivO3CPUParams *params)
     branchPred = params->branchPred;
 
     for (ThreadID tid = 0; tid < numThreads; tid++) {
-        decoder[tid] = new TheISA::Decoder;
+        decoder[tid] = new TheISA::Decoder(params->isa[tid]);
         // Create space to buffer the cache line data,
         // which may not hold the entire cache line.
         fetchBuffer[tid] = new uint8_t[fetchBufferSize];