arch: ISA parser additions of vector registers

Reiley's update :) of the isa parser definitions. My addition of the
vector element operand concept for the ISA parser. Nathanael's modification
creating a hierarchy between vector registers and its constituencies to the
isa parser.

Some fixes/updates on top to consider instructions as vectors instead of
floating when they use the VectorRF. Some counters added to all the
models to keep faithful counts.

Change-Id: Id8f162a525240dfd7ba884c5a4d9fa69f4050101
Reviewed-by: Andreas Sandberg <andreas.sandberg@arm.com>
Reviewed-on: https://gem5-review.googlesource.com/2706
Reviewed-by: Anthony Gutierrez <anthony.gutierrez@amd.com>
Maintainer: Andreas Sandberg <andreas.sandberg@arm.com>

4 files changed, 59 insertions, 9 deletions

diff --git a/src/cpu/o3/commit.hh b/src/cpu/o3/commit.hh
index 3cce7f69c..5977f94f3 100644
--- a/src/cpu/o3/commit.hh
+++ b/src/cpu/o3/commit.hh
@@ -517,6 +517,8 @@ class DefaultCommit
     Stats::Vector statComMembars;
     /** Total number of committed branches. */
     Stats::Vector statComBranches;
+    /** Total number of vector instructions */
+    Stats::Vector statComVector;
     /** Total number of floating point instructions */
     Stats::Vector statComFloating;
     /** Total number of integer instructions */
diff --git a/src/cpu/o3/commit_impl.hh b/src/cpu/o3/commit_impl.hh
index ea77f18fb..aba2696c2 100644
--- a/src/cpu/o3/commit_impl.hh
+++ b/src/cpu/o3/commit_impl.hh
@@ -260,6 +260,13 @@ DefaultCommit<Impl>::regStats()
         .flags(total)
         ;
 
+    statComVector
+        .init(cpu->numThreads)
+        .name(name() + ".vec_insts")
+        .desc("Number of committed Vector instructions.")
+        .flags(total)
+        ;
+
     statComInteger
         .init(cpu->numThreads)
         .name(name()+".int_insts")
@@ -1404,6 +1411,9 @@ DefaultCommit<Impl>::updateComInstStats(DynInstPtr &inst)
     // Floating Point Instruction
     if (inst->isFloating())
         statComFloating[tid]++;
+    // Vector Instruction
+    if (inst->isVector())
+        statComVector[tid]++;
 
     // Function Calls
     if (inst->isCall())
diff --git a/src/cpu/o3/inst_queue.hh b/src/cpu/o3/inst_queue.hh
index 23d8d416c..64f8aa1be 100644
--- a/src/cpu/o3/inst_queue.hh
+++ b/src/cpu/o3/inst_queue.hh
@@ -543,10 +543,14 @@ class InstructionQueue
     Stats::Scalar intInstQueueWakeupAccesses;
     Stats::Scalar fpInstQueueReads;
     Stats::Scalar fpInstQueueWrites;
-    Stats::Scalar fpInstQueueWakeupQccesses;
+    Stats::Scalar fpInstQueueWakeupAccesses;
+    Stats::Scalar vecInstQueueReads;
+    Stats::Scalar vecInstQueueWrites;
+    Stats::Scalar vecInstQueueWakeupAccesses;
 
     Stats::Scalar intAluAccesses;
     Stats::Scalar fpAluAccesses;
+    Stats::Scalar vecAluAccesses;
 };
 
 #endif //__CPU_O3_INST_QUEUE_HH__
diff --git a/src/cpu/o3/inst_queue_impl.hh b/src/cpu/o3/inst_queue_impl.hh
index 2b113ae04..3da72fd86 100644
--- a/src/cpu/o3/inst_queue_impl.hh
+++ b/src/cpu/o3/inst_queue_impl.hh
@@ -364,7 +364,7 @@ InstructionQueue<Impl>::regStats()
         .desc("Number of floating instruction queue writes")
         .flags(total);
 
-    fpInstQueueWakeupQccesses
+    fpInstQueueWakeupAccesses
         .name(name() + ".fp_inst_queue_wakeup_accesses")
         .desc("Number of floating instruction queue wakeup accesses")
         .flags(total);
@@ -567,7 +567,13 @@ template <class Impl>
 void
 InstructionQueue<Impl>::insert(DynInstPtr &new_inst)
 {
-    new_inst->isFloating() ? fpInstQueueWrites++ : intInstQueueWrites++;
+    if (new_inst->isFloating()) {
+        fpInstQueueWrites++;
+    } else if (new_inst->isVector()) {
+        vecInstQueueWrites++;
+    } else {
+        intInstQueueWrites++;
+    }
     // Make sure the instruction is valid
     assert(new_inst);
 
@@ -609,7 +615,13 @@ InstructionQueue<Impl>::insertNonSpec(DynInstPtr &new_inst)
 {
     // @todo: Clean up this code; can do it by setting inst as unable
     // to issue, then calling normal insert on the inst.
-    new_inst->isFloating() ? fpInstQueueWrites++ : intInstQueueWrites++;
+    if (new_inst->isFloating()) {
+        fpInstQueueWrites++;
+    } else if (new_inst->isVector()) {
+        vecInstQueueWrites++;
+    } else {
+        intInstQueueWrites++;
+    }
 
     assert(new_inst);
 
@@ -660,8 +672,10 @@ InstructionQueue<Impl>::getInstToExecute()
     assert(!instsToExecute.empty());
     DynInstPtr inst = instsToExecute.front();
     instsToExecute.pop_front();
-    if (inst->isFloating()){
+    if (inst->isFloating()) {
         fpInstQueueReads++;
+    } else if (inst->isVector()) {
+        vecInstQueueReads++;
     } else {
         intInstQueueReads++;
     }
@@ -783,7 +797,13 @@ InstructionQueue<Impl>::scheduleReadyInsts()
 
         DynInstPtr issuing_inst = readyInsts[op_class].top();
 
-        issuing_inst->isFloating() ? fpInstQueueReads++ : intInstQueueReads++;
+        if (issuing_inst->isFloating()) {
+            fpInstQueueReads++;
+        } else if (issuing_inst->isVector()) {
+            vecInstQueueReads++;
+        } else {
+            intInstQueueReads++;
+        }
 
         assert(issuing_inst->seqNum == (*order_it).oldestInst);
 
@@ -810,7 +830,13 @@ InstructionQueue<Impl>::scheduleReadyInsts()
 
         if (op_class != No_OpClass) {
             idx = fuPool->getUnit(op_class);
-            issuing_inst->isFloating() ? fpAluAccesses++ : intAluAccesses++;
+            if (issuing_inst->isFloating()) {
+                fpAluAccesses++;
+            } else if (issuing_inst->isVector()) {
+                vecAluAccesses++;
+            } else {
+                intAluAccesses++;
+            }
             if (idx > FUPool::NoFreeFU) {
                 op_latency = fuPool->getOpLatency(op_class);
             }
@@ -955,7 +981,9 @@ InstructionQueue<Impl>::wakeDependents(DynInstPtr &completed_inst)
 
     // The instruction queue here takes care of both floating and int ops
     if (completed_inst->isFloating()) {
-        fpInstQueueWakeupQccesses++;
+        fpInstQueueWakeupAccesses++;
+    } else if (completed_inst->isVector()) {
+        vecInstQueueWakeupAccesses++;
     } else {
         intInstQueueWakeupAccesses++;
     }
@@ -1189,7 +1217,13 @@ InstructionQueue<Impl>::doSquash(ThreadID tid)
            (*squash_it)->seqNum > squashedSeqNum[tid]) {
 
         DynInstPtr squashed_inst = (*squash_it);
-        squashed_inst->isFloating() ? fpInstQueueWrites++ : intInstQueueWrites++;
+        if (squashed_inst->isFloating()) {
+            fpInstQueueWrites++;
+        } else if (squashed_inst->isVector()) {
+            vecInstQueueWrites++;
+        } else {
+            intInstQueueWrites++;
+        }
 
         // Only handle the instruction if it actually is in the IQ and
         // hasn't already been squashed in the IQ.