summaryrefslogtreecommitdiff
path: root/src/cpu/ozone/lw_lsq_impl.hh
diff options
context:
space:
mode:
authorKevin Lim <ktlim@umich.edu>2006-09-30 23:43:23 -0400
committerKevin Lim <ktlim@umich.edu>2006-09-30 23:43:23 -0400
commit4ed184eadefb16627f2807cb3dc7886bb1b920d1 (patch)
tree7704b61afbb10876c86329665fe7d35865ba611b /src/cpu/ozone/lw_lsq_impl.hh
parent51425382ca079290411e033acc8cf14dde36c82b (diff)
parent30b719fd768115bb26e848a02096350c11c1b0bd (diff)
downloadgem5-4ed184eadefb16627f2807cb3dc7886bb1b920d1.tar.xz
Merge ktlim@zamp:./local/clean/o3-merge/m5
into zamp.eecs.umich.edu:/z/ktlim2/clean/o3-merge/newmem configs/boot/micro_memlat.rcS: configs/boot/micro_tlblat.rcS: src/arch/alpha/ev5.cc: src/arch/alpha/isa/decoder.isa: src/arch/alpha/isa_traits.hh: src/cpu/base.cc: src/cpu/base.hh: src/cpu/base_dyn_inst.hh: src/cpu/checker/cpu.hh: src/cpu/checker/cpu_impl.hh: src/cpu/o3/alpha/cpu_impl.hh: src/cpu/o3/alpha/params.hh: src/cpu/o3/checker_builder.cc: src/cpu/o3/commit_impl.hh: src/cpu/o3/cpu.cc: src/cpu/o3/decode_impl.hh: src/cpu/o3/fetch_impl.hh: src/cpu/o3/iew.hh: src/cpu/o3/iew_impl.hh: src/cpu/o3/inst_queue.hh: src/cpu/o3/lsq.hh: src/cpu/o3/lsq_impl.hh: src/cpu/o3/lsq_unit.hh: src/cpu/o3/lsq_unit_impl.hh: src/cpu/o3/regfile.hh: src/cpu/o3/rename_impl.hh: src/cpu/o3/thread_state.hh: src/cpu/ozone/checker_builder.cc: src/cpu/ozone/cpu.hh: src/cpu/ozone/cpu_impl.hh: src/cpu/ozone/front_end.hh: src/cpu/ozone/front_end_impl.hh: src/cpu/ozone/lw_back_end.hh: src/cpu/ozone/lw_back_end_impl.hh: src/cpu/ozone/lw_lsq.hh: src/cpu/ozone/lw_lsq_impl.hh: src/cpu/ozone/thread_state.hh: src/cpu/simple/base.cc: src/cpu/simple_thread.cc: src/cpu/simple_thread.hh: src/cpu/thread_state.hh: src/dev/ide_disk.cc: src/python/m5/objects/O3CPU.py: src/python/m5/objects/Root.py: src/python/m5/objects/System.py: src/sim/pseudo_inst.cc: src/sim/pseudo_inst.hh: src/sim/system.hh: util/m5/m5.c: Hand merge. --HG-- rename : arch/alpha/ev5.cc => src/arch/alpha/ev5.cc rename : arch/alpha/freebsd/system.cc => src/arch/alpha/freebsd/system.cc rename : arch/alpha/isa/decoder.isa => src/arch/alpha/isa/decoder.isa rename : arch/alpha/isa/mem.isa => src/arch/alpha/isa/mem.isa rename : arch/alpha/isa_traits.hh => src/arch/alpha/isa_traits.hh rename : arch/alpha/linux/system.cc => src/arch/alpha/linux/system.cc rename : arch/alpha/system.cc => src/arch/alpha/system.cc rename : arch/alpha/tru64/system.cc => src/arch/alpha/tru64/system.cc rename : cpu/base.cc => src/cpu/base.cc rename : cpu/base.hh => src/cpu/base.hh rename : cpu/base_dyn_inst.hh => src/cpu/base_dyn_inst.hh rename : cpu/checker/cpu.hh => src/cpu/checker/cpu.hh rename : cpu/checker/cpu.cc => src/cpu/checker/cpu_impl.hh rename : cpu/o3/alpha_cpu_builder.cc => src/cpu/o3/alpha/cpu_builder.cc rename : cpu/checker/o3_cpu_builder.cc => src/cpu/o3/checker_builder.cc rename : cpu/o3/commit_impl.hh => src/cpu/o3/commit_impl.hh rename : cpu/o3/cpu.cc => src/cpu/o3/cpu.cc rename : cpu/o3/fetch_impl.hh => src/cpu/o3/fetch_impl.hh rename : cpu/o3/iew.hh => src/cpu/o3/iew.hh rename : cpu/o3/iew_impl.hh => src/cpu/o3/iew_impl.hh rename : cpu/o3/inst_queue.hh => src/cpu/o3/inst_queue.hh rename : cpu/o3/inst_queue_impl.hh => src/cpu/o3/inst_queue_impl.hh rename : cpu/o3/lsq_impl.hh => src/cpu/o3/lsq_impl.hh rename : cpu/o3/lsq_unit.hh => src/cpu/o3/lsq_unit.hh rename : cpu/o3/lsq_unit_impl.hh => src/cpu/o3/lsq_unit_impl.hh rename : cpu/o3/mem_dep_unit_impl.hh => src/cpu/o3/mem_dep_unit_impl.hh rename : cpu/o3/rename.hh => src/cpu/o3/rename.hh rename : cpu/o3/rename_impl.hh => src/cpu/o3/rename_impl.hh rename : cpu/o3/thread_state.hh => src/cpu/o3/thread_state.hh rename : cpu/o3/tournament_pred.cc => src/cpu/o3/tournament_pred.cc rename : cpu/o3/tournament_pred.hh => src/cpu/o3/tournament_pred.hh rename : cpu/checker/cpu_builder.cc => src/cpu/ozone/checker_builder.cc rename : cpu/ozone/cpu.hh => src/cpu/ozone/cpu.hh rename : cpu/ozone/cpu_builder.cc => src/cpu/ozone/cpu_builder.cc rename : cpu/ozone/cpu_impl.hh => src/cpu/ozone/cpu_impl.hh rename : cpu/ozone/front_end.hh => src/cpu/ozone/front_end.hh rename : cpu/ozone/front_end_impl.hh => src/cpu/ozone/front_end_impl.hh rename : cpu/ozone/inorder_back_end_impl.hh => src/cpu/ozone/inorder_back_end_impl.hh rename : cpu/ozone/inst_queue_impl.hh => src/cpu/ozone/inst_queue_impl.hh rename : cpu/ozone/lw_back_end.hh => src/cpu/ozone/lw_back_end.hh rename : cpu/ozone/lw_back_end_impl.hh => src/cpu/ozone/lw_back_end_impl.hh rename : cpu/ozone/lw_lsq.hh => src/cpu/ozone/lw_lsq.hh rename : cpu/ozone/lw_lsq_impl.hh => src/cpu/ozone/lw_lsq_impl.hh rename : cpu/ozone/simple_params.hh => src/cpu/ozone/simple_params.hh rename : cpu/ozone/thread_state.hh => src/cpu/ozone/thread_state.hh rename : cpu/simple/cpu.cc => src/cpu/simple/base.cc rename : cpu/cpu_exec_context.cc => src/cpu/simple_thread.cc rename : cpu/thread_state.hh => src/cpu/thread_state.hh rename : dev/ide_disk.hh => src/dev/ide_disk.hh rename : python/m5/objects/BaseCPU.py => src/python/m5/objects/BaseCPU.py rename : python/m5/objects/AlphaFullCPU.py => src/python/m5/objects/O3CPU.py rename : python/m5/objects/OzoneCPU.py => src/python/m5/objects/OzoneCPU.py rename : python/m5/objects/Root.py => src/python/m5/objects/Root.py rename : python/m5/objects/System.py => src/python/m5/objects/System.py rename : sim/eventq.hh => src/sim/eventq.hh rename : sim/pseudo_inst.cc => src/sim/pseudo_inst.cc rename : sim/pseudo_inst.hh => src/sim/pseudo_inst.hh rename : sim/serialize.cc => src/sim/serialize.cc rename : sim/stat_control.cc => src/sim/stat_control.cc rename : sim/stat_control.hh => src/sim/stat_control.hh rename : sim/system.hh => src/sim/system.hh extra : convert_revision : 135d90e43f6cea89f9460ba4e23f4b0b85886e7d
Diffstat (limited to 'src/cpu/ozone/lw_lsq_impl.hh')
-rw-r--r--src/cpu/ozone/lw_lsq_impl.hh41
1 files changed, 32 insertions, 9 deletions
diff --git a/src/cpu/ozone/lw_lsq_impl.hh b/src/cpu/ozone/lw_lsq_impl.hh
index 7eef4b11f..31ffa9d67 100644
--- a/src/cpu/ozone/lw_lsq_impl.hh
+++ b/src/cpu/ozone/lw_lsq_impl.hh
@@ -132,7 +132,7 @@ OzoneLWLSQ<Impl>::completeDataAccess(PacketPtr pkt)
template <class Impl>
OzoneLWLSQ<Impl>::OzoneLWLSQ()
: switchedOut(false), dcachePort(this), loads(0), stores(0),
- storesToWB(0), stalled(false), isStoreBlocked(false),
+ storesToWB(0), storesInFlight(0), stalled(false), isStoreBlocked(false),
isLoadBlocked(false), loadBlockedHandled(false)
{
}
@@ -173,6 +173,11 @@ OzoneLWLSQ<Impl>::name() const
template<class Impl>
void
+OzoneLWLSQ<Impl>::regStats()
+{
+ lsqMemOrderViolation
+ .name(name() + ".memOrderViolation")
+ .desc("Number of memory ordering violations");
OzoneLWLSQ<Impl>::setCPU(OzoneCPU *cpu_ptr)
{
cpu = cpu_ptr;
@@ -321,7 +326,7 @@ unsigned
OzoneLWLSQ<Impl>::numFreeEntries()
{
unsigned free_lq_entries = LQEntries - loads;
- unsigned free_sq_entries = SQEntries - stores;
+ unsigned free_sq_entries = SQEntries - (stores + storesInFlight);
// Both the LQ and SQ entries have an extra dummy entry to differentiate
// empty/full conditions. Subtract 1 from the free entries.
@@ -385,6 +390,9 @@ OzoneLWLSQ<Impl>::executeLoad(DynInstPtr &inst)
// Actually probably want the oldest faulting load
if (load_fault != NoFault) {
DPRINTF(OzoneLSQ, "Load [sn:%lli] has a fault\n", inst->seqNum);
+ if (!(inst->req->flags & UNCACHEABLE && !inst->isAtCommit())) {
+ inst->setExecuted();
+ }
// Maybe just set it as can commit here, although that might cause
// some other problems with sending traps to the ROB too quickly.
be->instToCommit(inst);
@@ -461,6 +469,7 @@ OzoneLWLSQ<Impl>::executeStore(DynInstPtr &store_inst)
// A load incorrectly passed this store. Squash and refetch.
// For now return a fault to show that it was unsuccessful.
memDepViolator = (*lq_it);
+ ++lsqMemOrderViolation;
return TheISA::genMachineCheckFault();
}
@@ -553,8 +562,8 @@ OzoneLWLSQ<Impl>::writebackStores()
if ((*sq_it).size == 0 && !(*sq_it).completed) {
sq_it--;
- completeStore(inst->sqIdx);
-
+ removeStore(inst->sqIdx);
+ completeStore(inst);
continue;
}
@@ -626,6 +635,8 @@ OzoneLWLSQ<Impl>::writebackStores()
inst->sqIdx,inst->readPC(),
req->paddr, *(req->data),
inst->seqNum);
+ DPRINTF(OzoneLSQ, "StoresInFlight: %i\n",
+ storesInFlight + 1);
if (dcacheInterface) {
assert(!req->completionEvent);
@@ -687,6 +698,8 @@ OzoneLWLSQ<Impl>::writebackStores()
}
sq_it--;
}
+ ++storesInFlight;
+// removeStore(inst->sqIdx);
} else {
panic("Must HAVE DCACHE!!!!!\n");
}
@@ -704,7 +717,7 @@ void
OzoneLWLSQ<Impl>::squash(const InstSeqNum &squashed_num)
{
DPRINTF(OzoneLSQ, "Squashing until [sn:%lli]!"
- "(Loads:%i Stores:%i)\n",squashed_num,loads,stores);
+ "(Loads:%i Stores:%i)\n",squashed_num,loads,stores+storesInFlight);
LQIt lq_it = loadQueue.begin();
@@ -881,7 +894,7 @@ OzoneLWLSQ<Impl>::writeback(DynInstPtr &inst, PacketPtr pkt)
template <class Impl>
void
-OzoneLWLSQ<Impl>::completeStore(int store_idx)
+OzoneLWLSQ<Impl>::removeStore(int store_idx)
{
SQHashIt sq_hash_it = SQItHash.find(store_idx);
assert(sq_hash_it != SQItHash.end());
@@ -891,8 +904,6 @@ OzoneLWLSQ<Impl>::completeStore(int store_idx)
(*sq_it).completed = true;
DynInstPtr inst = (*sq_it).inst;
- --storesToWB;
-
if (isStalled() &&
inst->seqNum == stallingStoreIsn) {
DPRINTF(OzoneLSQ, "Unstalling, stalling store [sn:%lli] "
@@ -910,6 +921,13 @@ OzoneLWLSQ<Impl>::completeStore(int store_idx)
SQItHash.erase(sq_hash_it);
SQIndices.push(inst->sqIdx);
storeQueue.erase(sq_it);
+}
+
+template <class Impl>
+void
+OzoneLWLSQ<Impl>::completeStore(DynInstPtr &inst)
+{
+ --storesToWB;
--stores;
inst->setCompleted();
@@ -935,9 +953,14 @@ OzoneLWLSQ<Impl>::switchOut()
switchedOut = true;
// Clear the queue to free up resources
+ assert(stores == 0);
+ assert(storeQueue.empty());
+ assert(loads == 0);
+ assert(loadQueue.empty());
+ assert(storesInFlight == 0);
storeQueue.clear();
loadQueue.clear();
- loads = stores = storesToWB = 0;
+ loads = stores = storesToWB = storesInFlight = 0;
}
template <class Impl>