inorder: redefine DynInst FP result type

Sharing the FP value w/the integer values was giving inconsistent results esp. when
their is a 32-bit integer register matched w/a 64-bit float value

7 files changed, 116 insertions, 67 deletions

diff --git a/src/cpu/inorder/cpu.cc b/src/cpu/inorder/cpu.cc
index a634535bc..104bb6ff7 100644
--- a/src/cpu/inorder/cpu.cc
+++ b/src/cpu/inorder/cpu.cc
@@ -195,6 +195,7 @@ InOrderCPU::InOrderCPU(Params *params)
       timeBuffer(2 , 2),
       removeInstsThisCycle(false),
       activityRec(params->name, NumStages, 10, params->activity),
+      stCondFails(0),
 #if FULL_SYSTEM
       system(params->system),
       physmem(system->physmem),
@@ -206,6 +207,7 @@ InOrderCPU::InOrderCPU(Params *params)
       switchCount(0),
       deferRegistration(false/*params->deferRegistration*/),
       stageTracing(params->stageTracing),
+      lastRunningCycle(0),
       instsPerSwitch(0)
 {    
     ThreadID active_threads;
@@ -258,6 +260,9 @@ InOrderCPU::InOrderCPU(Params *params)
     }
 
     for (ThreadID tid = 0; tid < numThreads; ++tid) {
+        pc[tid].set(0);
+        lastCommittedPC[tid].set(0);
+
 #if FULL_SYSTEM
         // SMT is not supported in FS mode yet.
         assert(numThreads == 1);
@@ -1170,12 +1175,18 @@ InOrderCPU::readIntReg(RegIndex reg_idx, ThreadID tid)
 FloatReg
 InOrderCPU::readFloatReg(RegIndex reg_idx, ThreadID tid)
 {
+    DPRINTF(FloatRegs, "[tid:%i]: Reading Float Reg %i as %x, %08f\n",
+            tid, reg_idx, floatRegs.i[tid][reg_idx], floatRegs.f[tid][reg_idx]);
+
     return floatRegs.f[tid][reg_idx];
 }
 
 FloatRegBits
 InOrderCPU::readFloatRegBits(RegIndex reg_idx, ThreadID tid)
-{;
+{
+    DPRINTF(FloatRegs, "[tid:%i]: Reading Float Reg %i as %x, %08f\n",
+            tid, reg_idx, floatRegs.i[tid][reg_idx], floatRegs.f[tid][reg_idx]);
+
     return floatRegs.i[tid][reg_idx];
 }
 
@@ -1199,6 +1210,11 @@ void
 InOrderCPU::setFloatReg(RegIndex reg_idx, FloatReg val, ThreadID tid)
 {
     floatRegs.f[tid][reg_idx] = val;
+    DPRINTF(FloatRegs, "[tid:%i]: Setting Float. Reg %i bits to "
+            "%x, %08f\n",
+            tid, reg_idx,
+            floatRegs.i[tid][reg_idx],
+            floatRegs.f[tid][reg_idx]);
 }
 
 
@@ -1206,6 +1222,11 @@ void
 InOrderCPU::setFloatRegBits(RegIndex reg_idx, FloatRegBits val, ThreadID tid)
 {
     floatRegs.i[tid][reg_idx] = val;
+    DPRINTF(FloatRegs, "[tid:%i]: Setting Float. Reg %i bits to "
+            "%x, %08f\n",
+            tid, reg_idx,
+            floatRegs.i[tid][reg_idx],
+            floatRegs.f[tid][reg_idx]);
 }
 
 uint64_t
@@ -1257,6 +1278,11 @@ InOrderCPU::readMiscRegNoEffect(int misc_reg, ThreadID tid)
 MiscReg
 InOrderCPU::readMiscReg(int misc_reg, ThreadID tid)
 {
+    DPRINTF(InOrderCPU, "MiscReg: %i\n", misc_reg);
+    DPRINTF(InOrderCPU, "tid: %i\n", tid);
+    DPRINTF(InOrderCPU, "tcBase: %x\n", tcBase(tid));
+    DPRINTF(InOrderCPU, "isa-tid: %x\n", &isa[tid]);
+
     return isa[tid].readMiscReg(misc_reg, tcBase(tid));
 }
 
diff --git a/src/cpu/inorder/cpu.hh b/src/cpu/inorder/cpu.hh
index a87030828..f0ac28adb 100644
--- a/src/cpu/inorder/cpu.hh
+++ b/src/cpu/inorder/cpu.hh
@@ -262,17 +262,11 @@ class InOrderCPU : public BaseCPU
      */
     unsigned fetchPortIdx;
 
-    /** Identifies the resource id that identifies a ITB       */
-    unsigned itbIdx;
-
     /** Identifies the resource id that identifies a data
      * access unit.
      */
     unsigned dataPortIdx;
 
-    /** Identifies the resource id that identifies a DTB       */
-    unsigned dtbIdx;
-
     /** The Pipeline Stages for the CPU */
     PipelineStage *pipelineStage[ThePipeline::NumStages];
 
diff --git a/src/cpu/inorder/inorder_dyn_inst.cc b/src/cpu/inorder/inorder_dyn_inst.cc
index ccde20da4..2926a9434 100644
--- a/src/cpu/inorder/inorder_dyn_inst.cc
+++ b/src/cpu/inorder/inorder_dyn_inst.cc
@@ -69,8 +69,6 @@ InOrderDynInst::InOrderDynInst(InOrderCPU *cpu,
     lqIdx(0), sqIdx(0), instListIt(NULL), onInstList(false)
 {
     for(int i = 0; i < MaxInstSrcRegs; i++) {
-        instSrc[i].integer = 0;
-        instSrc[i].dbl = 0;
         _readySrcRegIdx[i] = false;
         _srcRegIdx[i] = 0;
     }
@@ -124,9 +122,6 @@ InOrderDynInst::initVars()
 
     nextStage = 0;
 
-    for(int i = 0; i < MaxInstDestRegs; i++)
-        instResult[i].val.integer = 0;
-
     status.reset();
 
     memAddrReady = false;
@@ -355,39 +350,48 @@ InOrderDynInst::releaseReq(ResourceRequest* req)
 void
 InOrderDynInst::setIntSrc(int idx, uint64_t val)
 {
-    DPRINTF(InOrderDynInst, "[tid:%i]: [sn:%i] Source Value %i being set "
+    DPRINTF(InOrderDynInst, "[tid:%i]: [sn:%i] [src:%i] Int being set "
             "to %#x.\n", threadNumber, seqNum, idx, val);
-    instSrc[idx].integer = val;
+    instSrc[idx].intVal = val;
 }
 
 /** Records an fp register being set to a value. */
 void
 InOrderDynInst::setFloatSrc(int idx, FloatReg val)
 {
-    instSrc[idx].dbl = val;
+    instSrc[idx].fpVal.f = val;
+    DPRINTF(InOrderDynInst, "[tid:%i]: [sn:%i] [src:%i] FP being set "
+            "to %x, %08f...%08f\n", threadNumber, seqNum, idx,
+            instSrc[idx].fpVal.i, instSrc[idx].fpVal.f, val);
 }
 
 /** Records an fp register being set to an integer value. */
 void
-InOrderDynInst::setFloatRegBitsSrc(int idx, uint64_t val)
+InOrderDynInst::setFloatRegBitsSrc(int idx, FloatRegBits val)
 {
-    instSrc[idx].integer = val;
+    instSrc[idx].fpVal.i = val;
+    DPRINTF(InOrderDynInst, "[tid:%i]: [sn:%i] [src:%i] FPBits being set "
+            "to %x, %08f...%x\n", threadNumber, seqNum, idx,
+            instSrc[idx].fpVal.i, instSrc[idx].fpVal.f, val);
 }
 
 /** Reads a integer register. */
 IntReg
 InOrderDynInst::readIntRegOperand(const StaticInst *si, int idx, ThreadID tid)
 {
-    DPRINTF(InOrderDynInst, "[tid:%i]: [sn:%i] Source Value %i read as %#x.\n",
-            threadNumber, seqNum, idx, instSrc[idx].integer);
-    return instSrc[idx].integer;
+    DPRINTF(InOrderDynInst, "[tid:%i]: [sn:%i] [src:%i] IntVal read as %#x.\n",
+            threadNumber, seqNum, idx, instSrc[idx].intVal);
+    return instSrc[idx].intVal;
 }
 
 /** Reads a FP register. */
 FloatReg
 InOrderDynInst::readFloatRegOperand(const StaticInst *si, int idx)
 {
-    return instSrc[idx].dbl;
+    DPRINTF(InOrderDynInst, "[tid:%i]: [sn:%i] [src:%i] FPVal being read "
+            "as %x, %08f.\n", threadNumber, seqNum, idx,
+            instSrc[idx].fpVal.i, instSrc[idx].fpVal.f);
+    return instSrc[idx].fpVal.f;
 }
 
 
@@ -395,7 +399,10 @@ InOrderDynInst::readFloatRegOperand(const StaticInst *si, int idx)
 FloatRegBits
 InOrderDynInst::readFloatRegOperandBits(const StaticInst *si, int idx)
 {
-    return instSrc[idx].integer;
+    DPRINTF(InOrderDynInst, "[tid:%i]: [sn:%i] [src:%i] FPBits being read "
+            "as %x, %08f.\n", threadNumber, seqNum, idx,
+            instSrc[idx].fpVal.i, instSrc[idx].fpVal.f);
+    return instSrc[idx].fpVal.i;
 }
 
 /** Reads a miscellaneous register. */
@@ -414,8 +421,8 @@ InOrderDynInst::readMiscRegOperand(const StaticInst *si, int idx)
 {
     DPRINTF(InOrderDynInst, "[tid:%i]: [sn:%i] Misc. Reg Source Value %i"
             " read as %#x.\n", threadNumber, seqNum, idx,
-            instSrc[idx].integer);
-    return instSrc[idx].integer;
+            instSrc[idx].intVal);
+    return instSrc[idx].intVal;
 }
 
 
@@ -427,7 +434,7 @@ InOrderDynInst::setMiscRegOperand(const StaticInst *si, int idx,
                        const MiscReg &val)
 {
     instResult[idx].type = Integer;
-    instResult[idx].val.integer = val;
+    instResult[idx].res.intVal = val;
     instResult[idx].tick = curTick();
 
     DPRINTF(InOrderDynInst, "[tid:%i]: [sn:%i] Setting Misc Reg. Operand %i "
@@ -457,7 +464,7 @@ void
 InOrderDynInst::setIntRegOperand(const StaticInst *si, int idx, IntReg val)
 {
     instResult[idx].type = Integer;
-    instResult[idx].val.integer = val;
+    instResult[idx].res.intVal = val;
     instResult[idx].tick = curTick();
 
     DPRINTF(InOrderDynInst, "[tid:%i]: [sn:%i] Setting Result Int Reg. %i "
@@ -469,13 +476,13 @@ InOrderDynInst::setIntRegOperand(const StaticInst *si, int idx, IntReg val)
 void
 InOrderDynInst::setFloatRegOperand(const StaticInst *si, int idx, FloatReg val)
 {
-    instResult[idx].val.dbl = val;
+    instResult[idx].res.fpVal.f = val;
     instResult[idx].type = Float;
     instResult[idx].tick = curTick();
 
-    DPRINTF(InOrderDynInst, "[tid:%i]: [sn:%i] Setting Result Float Reg. %i "
-            "being set to %#x (result-tick:%i).\n",
-            threadNumber, seqNum, idx, val, instResult[idx].tick);
+    DPRINTF(InOrderDynInst, "[tid:%i]: [sn:%i] Result Float Reg. %i "
+            "being set to %#x, %08f (result-tick:%i).\n",
+            threadNumber, seqNum, idx, val, val, instResult[idx].tick);
 }
 
 /** Sets a FP register as a integer. */
@@ -484,10 +491,10 @@ InOrderDynInst::setFloatRegOperandBits(const StaticInst *si, int idx,
                               FloatRegBits val)
 {
     instResult[idx].type = Integer;
-    instResult[idx].val.integer = val;
+    instResult[idx].res.fpVal.i = val;
     instResult[idx].tick = curTick();
 
-    DPRINTF(InOrderDynInst, "[tid:%i]: [sn:%i] Setting Result Float Reg. %i "
+    DPRINTF(InOrderDynInst, "[tid:%i]: [sn:%i] Result Float Reg. Bits %i "
             "being set to %#x (result-tick:%i).\n",
             threadNumber, seqNum, idx, val, instResult[idx].tick);
 }
diff --git a/src/cpu/inorder/inorder_dyn_inst.hh b/src/cpu/inorder/inorder_dyn_inst.hh
index d14a7edfd..c896812d7 100644
--- a/src/cpu/inorder/inorder_dyn_inst.hh
+++ b/src/cpu/inorder/inorder_dyn_inst.hh
@@ -89,6 +89,8 @@ class InOrderDynInst : public FastAlloc, public RefCounted
     // Floating point register type.
     typedef TheISA::FloatReg FloatReg;
     // Floating point register type.
+    typedef TheISA::FloatRegBits FloatRegBits;
+    // Floating point register type.
     typedef TheISA::MiscReg MiscReg;
 
     typedef short int PhysRegIndex;
@@ -207,13 +209,6 @@ class InOrderDynInst : public FastAlloc, public RefCounted
     /** How many source registers are ready. */
     unsigned readyRegs;
 
-    /** An instruction src/dest has to be one of these types */
-    union InstValue {
-        uint64_t integer;
-        double dbl;
-    };
-
-    //@TODO: Naming Convention for Enums?
     enum ResultType {
         None,
         Integer,
@@ -221,19 +216,30 @@ class InOrderDynInst : public FastAlloc, public RefCounted
         Double
     };
 
+    /** An instruction src/dest has to be one of these types */
+    struct InstValue {
+        IntReg intVal;
+        union {
+            FloatReg f;
+            FloatRegBits i;
+        } fpVal;
+
+        InstValue()
+        {
+            intVal = 0;
+            fpVal.i = 0;
+        }
+    };
 
     /** Result of an instruction execution */
     struct InstResult {
         ResultType type;
-        InstValue val;
+        InstValue res;
         Tick tick;
 
         InstResult()
           : type(None), tick(0)
-        {
-          val.integer = 0;
-          val.dbl = 0;
-        }
+        { }
     };
 
     /** The source of the instruction; assumes for now that there's only one
@@ -852,10 +858,10 @@ class InOrderDynInst : public FastAlloc, public RefCounted
      *  source register to a value. */
     void setIntSrc(int idx, uint64_t val);
     void setFloatSrc(int idx, FloatReg val);
-    void setFloatRegBitsSrc(int idx, uint64_t val);
+    void setFloatRegBitsSrc(int idx, TheISA::FloatRegBits val);
 
-    uint64_t* getIntSrcPtr(int idx) { return &instSrc[idx].integer; }
-    uint64_t readIntSrc(int idx) { return instSrc[idx].integer; }
+    uint64_t* getIntSrcPtr(int idx) { return &instSrc[idx].intVal; }
+    uint64_t readIntSrc(int idx) { return instSrc[idx].intVal; }
 
     /** These Instructions read a integer/float/misc. source register
      *  value in the instruction. The instruction's execute function will
@@ -879,18 +885,17 @@ class InOrderDynInst : public FastAlloc, public RefCounted
 
     uint64_t readIntResult(int idx)
     {
-        return instResult[idx].val.integer;
+        return instResult[idx].res.intVal;
     }
 
-    /** Depending on type, return Float or Double */
-    double readFloatResult(int idx)
+    FloatReg readFloatResult(int idx)
     {
-       return instResult[idx].val.dbl;
+       return instResult[idx].res.fpVal.f;
     }
 
     Tick readResultTime(int idx) { return instResult[idx].tick; }
 
-    uint64_t* getIntResultPtr(int idx) { return &instResult[idx].val.integer; }
+    uint64_t* getIntResultPtr(int idx) { return &instResult[idx].res.intVal; }
 
     /** This is the interface that an instruction will use to write
      *  it's destination register.
diff --git a/src/cpu/inorder/reg_dep_map.cc b/src/cpu/inorder/reg_dep_map.cc
index 2cb6b7adb..2f8a2924c 100644
--- a/src/cpu/inorder/reg_dep_map.cc
+++ b/src/cpu/inorder/reg_dep_map.cc
@@ -104,7 +104,7 @@ RegDepMap::insert(DynInstPtr inst)
 
         inst->flattenDestReg(i, flat_idx);
 
-        if (flat_idx == TheISA::ZeroReg) {
+        if (flat_idx == TheISA::ZeroReg && reg_type == InOrderCPU::IntType) {
             DPRINTF(RegDepMap, "[sn:%i]: Ignoring Insert-Dependency tracking for "
                     "ISA-ZeroReg (Int. Reg %i).\n", inst->seqNum,
                     flat_idx);
@@ -143,15 +143,18 @@ RegDepMap::remove(DynInstPtr inst)
 
         for (int i = 0; i < dest_regs; i++) {
             RegIndex flat_idx = inst->flattenedDestRegIdx(i);
+            InOrderCPU::RegType reg_type = cpu->getRegType(inst->destRegIdx(i));
 
-            if (flat_idx == TheISA::ZeroReg) {
+            // Merge Dyn Inst & CPU Result Types
+            if (flat_idx == TheISA::ZeroReg &&
+                reg_type == InOrderCPU::IntType) {
                 DPRINTF(RegDepMap, "[sn:%i]: Ignoring Remove-Dependency tracking for "
                         "ISA-ZeroReg (Int. Reg %i).\n", inst->seqNum,
                         flat_idx);
                 continue;
             }
 
-            InOrderCPU::RegType reg_type = cpu->getRegType(inst->destRegIdx(i));
+
             remove(reg_type, flat_idx, inst);
         }
     }
@@ -174,7 +177,7 @@ RegDepMap::remove(uint8_t reg_type, RegIndex idx, DynInstPtr inst)
         }
         list_it++;
     }
-    panic("[sn:%i] Did not find entry for %i ", inst->seqNum, idx);
+    panic("[sn:%i] Did not find entry for %i, type:%i\n", inst->seqNum, idx, reg_type);
 }
 
 void
diff --git a/src/cpu/inorder/resources/execution_unit.cc b/src/cpu/inorder/resources/execution_unit.cc
index 77ab21b21..460ec32fe 100644
--- a/src/cpu/inorder/resources/execution_unit.cc
+++ b/src/cpu/inorder/resources/execution_unit.cc
@@ -142,6 +142,9 @@ ExecutionUnit::execute(int slot_num)
 
                 // Evaluate Branch
                 fault = inst->execute();
+
+                // Should unconditional control , pc relative count as an
+                // execution??? Probably not.
                 executions++;
 
                 if (fault == NoFault) {
@@ -204,11 +207,19 @@ ExecutionUnit::execute(int slot_num)
                 if (fault == NoFault) {
                     inst->setExecuted();
 
-                    DPRINTF(InOrderExecute, "[tid:%i]: [sn:%i]: The result "
-                            "of execution is 0x%x.\n", inst->readTid(),
-                            seq_num,
-                            (inst->resultType(0) == InOrderDynInst::Float) ?
-                            inst->readFloatResult(0) : inst->readIntResult(0));
+#if TRACING_ON
+                    for (int didx = 0; didx < inst->numDestRegs(); didx++)
+                        if (inst->resultType(didx) == InOrderDynInst::Float ||
+                            inst->resultType(didx) == InOrderDynInst::Double)
+                            DPRINTF(InOrderExecute, "[tid:%i]: [sn:%i]: Dest result %i "
+                                    "of FP execution is %08f (%x).\n", inst->readTid(),
+                                    seq_num, didx, inst->readFloatResult(didx),
+                                    inst->readIntResult(didx));
+                        else
+                            DPRINTF(InOrderExecute, "[tid:%i]: [sn:%i]: Dest result %i "
+                                    "of Int execution is 0x%x.\n", inst->readTid(),
+                                    seq_num, didx, inst->readIntResult(didx));
+#endif
 
 #if !FULL_SYSTEM
                     // The Syscall might change the PC, so conservatively
diff --git a/src/cpu/inorder/resources/use_def.cc b/src/cpu/inorder/resources/use_def.cc
index 1adee09db..78497f6e5 100644
--- a/src/cpu/inorder/resources/use_def.cc
+++ b/src/cpu/inorder/resources/use_def.cc
@@ -181,7 +181,7 @@ UseDefUnit::execute(int slot_idx)
             RegIndex flat_idx = cpu->flattenRegIdx(reg_idx, reg_type, tid);
             inst->flattenSrcReg(ud_idx, flat_idx);
             
-            if (flat_idx == TheISA::ZeroReg) {
+            if (flat_idx == TheISA::ZeroReg && reg_type == InOrderCPU::IntType) {
                 DPRINTF(InOrderUseDef, "[tid:%i]: [sn:%i]: Ignoring Reading of ISA-ZeroReg "
                         "(Int. Reg %i).\n", tid, inst->seqNum, flat_idx);
                 ud_req->done();
@@ -226,6 +226,9 @@ UseDefUnit::execute(int slot_idx)
                         inst->setFloatSrc(ud_idx,
                                           cpu->readFloatReg(flat_idx,
                                                             inst->readTid()));
+                        inst->setFloatRegBitsSrc(ud_idx,
+                                                 cpu->readFloatRegBits(flat_idx,
+                                                                       inst->readTid()));
                         floatRegFileReads++;
                     }
                     break;
@@ -238,7 +241,7 @@ UseDefUnit::execute(int slot_idx)
                                 tid, seq_num,
                                 reg_idx - Ctrl_Base_DepTag, flat_idx,
                                 cpu->readMiscReg(flat_idx,
-                                                 inst->readTid()));
+                                inst->readTid()));
                         inst->setIntSrc(ud_idx,
                                         cpu->readMiscReg(flat_idx,
                                                          inst->readTid()));
@@ -331,7 +334,7 @@ UseDefUnit::execute(int slot_idx)
             RegIndex reg_idx = inst->_destRegIdx[ud_idx];
             RegIndex flat_idx = cpu->flattenRegIdx(reg_idx, reg_type, tid);
 
-            if (flat_idx == TheISA::ZeroReg) {
+            if (flat_idx == TheISA::ZeroReg && reg_type == InOrderCPU::IntType) {
                 DPRINTF(IntRegs, "[tid:%i]: Ignoring Writing of ISA-ZeroReg "
                         "(Int. Reg %i)\n", tid, flat_idx);
                 ud_req->done();
@@ -374,7 +377,7 @@ UseDefUnit::execute(int slot_idx)
                         if (inst->resultType(ud_idx) == 
                             InOrderDynInst::Integer) {
                             DPRINTF(InOrderUseDef, "[tid:%i]: [sn:%i]: Writing FP-Bits "
-                                    "Result 0x%x (bits:0x%x) to register "
+                                    "Result %08f (bits:0x%x) to register "
                                     "idx %i (%i).\n",
                                     tid, seq_num,
                                     inst->readFloatResult(ud_idx), 
@@ -388,7 +391,7 @@ UseDefUnit::execute(int slot_idx)
                         } else if (inst->resultType(ud_idx) == 
                                    InOrderDynInst::Float) {
                             DPRINTF(InOrderUseDef, "[tid:%i]: [sn:%i]: Writing Float "
-                                    "Result 0x%x (bits:0x%x) to register "
+                                    "Result %08f (bits:0x%x) to register "
                                     "idx %i (%i).\n",
                                     tid, seq_num, inst->readFloatResult(ud_idx),
                                     inst->readIntResult(ud_idx), 
@@ -400,7 +403,7 @@ UseDefUnit::execute(int slot_idx)
                         } else if (inst->resultType(ud_idx) == 
                                    InOrderDynInst::Double) {
                             DPRINTF(InOrderUseDef, "[tid:%i]: [sn:%i]: Writing Double "
-                                    "Result 0x%x (bits:0x%x) to register "
+                                    "Result %08f (bits:0x%x) to register "
                                     "idx %i (%i).\n",
                                     tid, seq_num,
                                     inst->readFloatResult(ud_idx),