summaryrefslogtreecommitdiff
diff options
context:
space:
mode:
Diffstat
-rw-r--r--configs/example/ruby_random_test.py13
-rw-r--r--configs/ruby/MESI_Three_Level.py82
-rw-r--r--configs/ruby/MESI_Two_Level.py48
-rw-r--r--configs/ruby/MI_example.py37
-rw-r--r--configs/ruby/MOESI_CMP_directory.py37
-rw-r--r--configs/ruby/MOESI_CMP_token.py54
-rw-r--r--configs/ruby/MOESI_hammer.py40
-rw-r--r--src/cpu/testers/rubytest/Check.cc23
-rw-r--r--src/cpu/testers/rubytest/CheckTable.cc7
-rw-r--r--src/cpu/testers/rubytest/RubyTester.cc70
-rw-r--r--src/cpu/testers/rubytest/RubyTester.hh14
-rw-r--r--src/cpu/testers/rubytest/RubyTester.py5
-rw-r--r--src/mem/ruby/system/RubyPort.cc60
-rw-r--r--src/mem/ruby/system/RubyPort.hh17
-rw-r--r--src/mem/ruby/system/Sequencer.cc1
-rw-r--r--src/mem/ruby/system/Sequencer.py1
16 files changed, 326 insertions, 183 deletions
diff --git a/configs/example/ruby_random_test.py b/configs/example/ruby_random_test.py
index 225b3d23b..10d4318c7 100644
--- a/configs/example/ruby_random_test.py
+++ b/configs/example/ruby_random_test.py
@@ -125,10 +125,15 @@ for ruby_port in system.ruby._cpu_ports:
#
# Tie the ruby tester ports to the ruby cpu read and write ports
#
- if ruby_port.support_data_reqs:
- tester.cpuDataPort = ruby_port.slave
- if ruby_port.support_inst_reqs:
- tester.cpuInstPort = ruby_port.slave
+ if ruby_port.support_data_reqs and ruby_port.support_inst_reqs:
+ tester.cpuInstDataPort = ruby_port.slave
+ elif ruby_port.support_data_reqs:
+ tester.cpuDataPort = ruby_port.slave
+ elif ruby_port.support_inst_reqs:
+ tester.cpuInstPort = ruby_port.slave
+
+ # Do not automatically retry stalled Ruby requests
+ ruby_port.no_retry_on_stall = True
#
# Tell each sequencer this is the ruby tester so that it
diff --git a/configs/ruby/MESI_Three_Level.py b/configs/ruby/MESI_Three_Level.py
index 9db9d27cd..1d4b6ebf6 100644
--- a/configs/ruby/MESI_Three_Level.py
+++ b/configs/ruby/MESI_Three_Level.py
@@ -1,5 +1,5 @@
# Copyright (c) 2006-2007 The Regents of The University of Michigan
-# Copyright (c) 2009 Advanced Micro Devices, Inc.
+# Copyright (c) 2009,2015 Advanced Micro Devices, Inc.
# Copyright (c) 2013 Mark D. Hill and David A. Wood
# All rights reserved.
#
@@ -44,22 +44,24 @@ class L1Cache(RubyCache): pass
class L2Cache(RubyCache): pass
def define_options(parser):
- parser.add_option("--num-clusters", type="int", default=1,
- help="number of clusters in a design in which there are shared\
+ parser.add_option("--num-clusters", type = "int", default = 1,
+ help = "number of clusters in a design in which there are shared\
caches private to clusters")
return
def create_system(options, full_system, system, dma_ports, ruby_system):
if buildEnv['PROTOCOL'] != 'MESI_Three_Level':
- fatal("This script requires the MESI_Three_Level protocol to be built.")
+ fatal("This script requires the MESI_Three_Level protocol to be\
+ built.")
cpu_sequencers = []
#
# The ruby network creation expects the list of nodes in the system to be
- # consistent with the NetDest list. Therefore the l1 controller nodes must be
- # listed before the directory nodes and directory nodes before dma nodes, etc.
+ # consistent with the NetDest list. Therefore the l1 controller nodes
+ # must be listed before the directory nodes and directory nodes before
+ # dma nodes, etc.
#
l0_cntrl_nodes = []
l1_cntrl_nodes = []
@@ -94,30 +96,45 @@ def create_system(options, full_system, system, dma_ports, ruby_system):
start_index_bit = block_size_bits,
replacement_policy = LRUReplacementPolicy())
- l0_cntrl = L0Cache_Controller(version = i*num_cpus_per_cluster + j,
- Icache = l0i_cache, Dcache = l0d_cache,
- send_evictions = send_evicts(options),
- clk_domain=system.cpu[i].clk_domain,
- ruby_system = ruby_system)
+ # the ruby random tester reuses num_cpus to specify the
+ # number of cpu ports connected to the tester object, which
+ # is stored in system.cpu. because there is only ever one
+ # tester object, num_cpus is not necessarily equal to the
+ # size of system.cpu; therefore if len(system.cpu) == 1
+ # we use system.cpu[0] to set the clk_domain, thereby ensuring
+ # we don't index off the end of the cpu list.
+ if len(system.cpu) == 1:
+ clk_domain = system.cpu[0].clk_domain
+ else:
+ clk_domain = system.cpu[i].clk_domain
+
+ l0_cntrl = L0Cache_Controller(
+ version = i * num_cpus_per_cluster + j, Icache = l0i_cache,
+ Dcache = l0d_cache, send_evictions = send_evicts(options),
+ clk_domain = clk_domain, ruby_system = ruby_system)
cpu_seq = RubySequencer(version = i * num_cpus_per_cluster + j,
- icache = l0i_cache,
- clk_domain=system.cpu[i].clk_domain,
- dcache = l0d_cache, ruby_system = ruby_system)
+ icache = l0i_cache,
+ clk_domain = clk_domain,
+ dcache = l0d_cache,
+ ruby_system = ruby_system)
l0_cntrl.sequencer = cpu_seq
- l1_cache = L1Cache(size = options.l1d_size, assoc = options.l1d_assoc,
- start_index_bit = block_size_bits, is_icache = False)
+ l1_cache = L1Cache(size = options.l1d_size,
+ assoc = options.l1d_assoc,
+ start_index_bit = block_size_bits,
+ is_icache = False)
- l1_cntrl = L1Cache_Controller(version = i*num_cpus_per_cluster+j,
- cache = l1_cache, l2_select_num_bits = l2_bits,
- cluster_id = i, ruby_system = ruby_system)
+ l1_cntrl = L1Cache_Controller(
+ version = i * num_cpus_per_cluster + j,
+ cache = l1_cache, l2_select_num_bits = l2_bits,
+ cluster_id = i, ruby_system = ruby_system)
- exec("ruby_system.l0_cntrl%d = l0_cntrl" % (
- i*num_cpus_per_cluster+j))
- exec("ruby_system.l1_cntrl%d = l1_cntrl" % (
- i*num_cpus_per_cluster+j))
+ exec("ruby_system.l0_cntrl%d = l0_cntrl"
+ % ( i * num_cpus_per_cluster + j))
+ exec("ruby_system.l1_cntrl%d = l1_cntrl"
+ % ( i * num_cpus_per_cluster + j))
#
# Add controllers and sequencers to the appropriate lists
@@ -155,11 +172,11 @@ def create_system(options, full_system, system, dma_ports, ruby_system):
l2_cntrl = L2Cache_Controller(
version = i * num_l2caches_per_cluster + j,
L2cache = l2_cache, cluster_id = i,
- transitions_per_cycle=options.ports,
+ transitions_per_cycle = options.ports,
ruby_system = ruby_system)
- exec("ruby_system.l2_cntrl%d = l2_cntrl" % (
- i * num_l2caches_per_cluster + j))
+ exec("ruby_system.l2_cntrl%d = l2_cntrl"
+ % (i * num_l2caches_per_cluster + j))
l2_cntrl_nodes.append(l2_cntrl)
# Connect the L2 controllers and the network
@@ -185,8 +202,7 @@ def create_system(options, full_system, system, dma_ports, ruby_system):
# the ruby system
# clk_divider value is a fix to pass regression.
ruby_system.memctrl_clk_domain = DerivedClockDomain(
- clk_domain=ruby_system.clk_domain,
- clk_divider=3)
+ clk_domain = ruby_system.clk_domain, clk_divider = 3)
for i in xrange(options.num_dirs):
#
@@ -196,10 +212,9 @@ def create_system(options, full_system, system, dma_ports, ruby_system):
dir_size.value = mem_module_size
dir_cntrl = Directory_Controller(version = i,
- directory = RubyDirectoryMemory(
- version = i, size = dir_size),
- transitions_per_cycle = options.ports,
- ruby_system = ruby_system)
+ directory = RubyDirectoryMemory(version = i, size = dir_size),
+ transitions_per_cycle = options.ports,
+ ruby_system = ruby_system)
exec("ruby_system.dir_cntrl%d = dir_cntrl" % i)
dir_cntrl_nodes.append(dir_cntrl)
@@ -217,8 +232,7 @@ def create_system(options, full_system, system, dma_ports, ruby_system):
#
# Create the Ruby objects associated with the dma controller
#
- dma_seq = DMASequencer(version = i,
- ruby_system = ruby_system)
+ dma_seq = DMASequencer(version = i, ruby_system = ruby_system)
dma_cntrl = DMA_Controller(version = i,
dma_sequencer = dma_seq,
diff --git a/configs/ruby/MESI_Two_Level.py b/configs/ruby/MESI_Two_Level.py
index 9997a117b..4cfa54bd8 100644
--- a/configs/ruby/MESI_Two_Level.py
+++ b/configs/ruby/MESI_Two_Level.py
@@ -82,23 +82,33 @@ def create_system(options, full_system, system, dma_ports, ruby_system):
prefetcher = RubyPrefetcher.Prefetcher()
- l1_cntrl = L1Cache_Controller(version = i,
- L1Icache = l1i_cache,
+ # the ruby random tester reuses num_cpus to specify the
+ # number of cpu ports connected to the tester object, which
+ # is stored in system.cpu. because there is only ever one
+ # tester object, num_cpus is not necessarily equal to the
+ # size of system.cpu; therefore if len(system.cpu) == 1
+ # we use system.cpu[0] to set the clk_domain, thereby ensuring
+ # we don't index off the end of the cpu list.
+ if len(system.cpu) == 1:
+ clk_domain = system.cpu[0].clk_domain
+ else:
+ clk_domain = system.cpu[i].clk_domain
+
+ l1_cntrl = L1Cache_Controller(version = i, L1Icache = l1i_cache,
L1Dcache = l1d_cache,
l2_select_num_bits = l2_bits,
send_evictions = send_evicts(options),
prefetcher = prefetcher,
ruby_system = ruby_system,
- clk_domain=system.cpu[i].clk_domain,
- transitions_per_cycle=options.ports,
+ clk_domain = clk_domain,
+ transitions_per_cycle = options.ports,
enable_prefetch = False)
- cpu_seq = RubySequencer(version = i,
- icache = l1i_cache,
- dcache = l1d_cache,
- clk_domain=system.cpu[i].clk_domain,
+ cpu_seq = RubySequencer(version = i, icache = l1i_cache,
+ dcache = l1d_cache, clk_domain = clk_domain,
ruby_system = ruby_system)
+
l1_cntrl.sequencer = cpu_seq
exec("ruby_system.l1_cntrl%d = l1_cntrl" % i)
@@ -135,7 +145,7 @@ def create_system(options, full_system, system, dma_ports, ruby_system):
l2_cntrl = L2Cache_Controller(version = i,
L2cache = l2_cache,
- transitions_per_cycle=options.ports,
+ transitions_per_cycle = options.ports,
ruby_system = ruby_system)
exec("ruby_system.l2_cntrl%d = l2_cntrl" % i)
@@ -166,18 +176,17 @@ def create_system(options, full_system, system, dma_ports, ruby_system):
# the ruby system
# clk_divider value is a fix to pass regression.
ruby_system.memctrl_clk_domain = DerivedClockDomain(
- clk_domain=ruby_system.clk_domain,
- clk_divider=3)
+ clk_domain = ruby_system.clk_domain,
+ clk_divider = 3)
for i in xrange(options.num_dirs):
dir_size = MemorySize('0B')
dir_size.value = mem_module_size
dir_cntrl = Directory_Controller(version = i,
- directory = RubyDirectoryMemory(
- version = i, size = dir_size),
- transitions_per_cycle = options.ports,
- ruby_system = ruby_system)
+ directory = RubyDirectoryMemory(version = i, size = dir_size),
+ transitions_per_cycle = options.ports,
+ ruby_system = ruby_system)
exec("ruby_system.dir_cntrl%d = dir_cntrl" % i)
dir_cntrl_nodes.append(dir_cntrl)
@@ -194,12 +203,10 @@ def create_system(options, full_system, system, dma_ports, ruby_system):
for i, dma_port in enumerate(dma_ports):
# Create the Ruby objects associated with the dma controller
- dma_seq = DMASequencer(version = i,
- ruby_system = ruby_system,
+ dma_seq = DMASequencer(version = i, ruby_system = ruby_system,
slave = dma_port)
- dma_cntrl = DMA_Controller(version = i,
- dma_sequencer = dma_seq,
+ dma_cntrl = DMA_Controller(version = i, dma_sequencer = dma_seq,
transitions_per_cycle = options.ports,
ruby_system = ruby_system)
@@ -220,7 +227,8 @@ def create_system(options, full_system, system, dma_ports, ruby_system):
# Create the io controller and the sequencer
if full_system:
- io_seq = DMASequencer(version=len(dma_ports), ruby_system=ruby_system)
+ io_seq = DMASequencer(version = len(dma_ports),
+ ruby_system = ruby_system)
ruby_system._io_port = io_seq
io_controller = DMA_Controller(version = len(dma_ports),
dma_sequencer = io_seq,
diff --git a/configs/ruby/MI_example.py b/configs/ruby/MI_example.py
index 6f28c6ade..24b0f9716 100644
--- a/configs/ruby/MI_example.py
+++ b/configs/ruby/MI_example.py
@@ -74,21 +74,28 @@ def create_system(options, full_system, system, dma_ports, ruby_system):
assoc = options.l1d_assoc,
start_index_bit = block_size_bits)
- #
- # Only one unified L1 cache exists. Can cache instructions and data.
- #
- l1_cntrl = L1Cache_Controller(version = i,
- cacheMemory = cache,
- send_evictions = send_evicts(options),
- transitions_per_cycle = options.ports,
- clk_domain=system.cpu[i].clk_domain,
- ruby_system = ruby_system)
-
- cpu_seq = RubySequencer(version = i,
- icache = cache,
- dcache = cache,
- clk_domain=system.cpu[i].clk_domain,
- ruby_system = ruby_system)
+
+ # the ruby random tester reuses num_cpus to specify the
+ # number of cpu ports connected to the tester object, which
+ # is stored in system.cpu. because there is only ever one
+ # tester object, num_cpus is not necessarily equal to the
+ # size of system.cpu; therefore if len(system.cpu) == 1
+ # we use system.cpu[0] to set the clk_domain, thereby ensuring
+ # we don't index off the end of the cpu list.
+ if len(system.cpu) == 1:
+ clk_domain = system.cpu[0].clk_domain
+ else:
+ clk_domain = system.cpu[i].clk_domain
+
+ # Only one unified L1 cache exists. Can cache instructions and data.
+ l1_cntrl = L1Cache_Controller(version=i, cacheMemory=cache,
+ send_evictions=send_evicts(options),
+ transitions_per_cycle=options.ports,
+ clk_domain=clk_domain,
+ ruby_system=ruby_system)
+
+ cpu_seq = RubySequencer(version=i, icache=cache, dcache=cache,
+ clk_domain=clk_domain, ruby_system=ruby_system)
l1_cntrl.sequencer = cpu_seq
exec("ruby_system.l1_cntrl%d = l1_cntrl" % i)
diff --git a/configs/ruby/MOESI_CMP_directory.py b/configs/ruby/MOESI_CMP_directory.py
index cdf8688f0..a72b5b20e 100644
--- a/configs/ruby/MOESI_CMP_directory.py
+++ b/configs/ruby/MOESI_CMP_directory.py
@@ -80,20 +80,29 @@ def create_system(options, full_system, system, dma_ports, ruby_system):
start_index_bit = block_size_bits,
is_icache = False)
- l1_cntrl = L1Cache_Controller(version = i,
- L1Icache = l1i_cache,
- L1Dcache = l1d_cache,
- l2_select_num_bits = l2_bits,
- send_evictions = send_evicts(options),
- transitions_per_cycle = options.ports,
- clk_domain=system.cpu[i].clk_domain,
- ruby_system = ruby_system)
-
- cpu_seq = RubySequencer(version = i,
- icache = l1i_cache,
- dcache = l1d_cache,
- clk_domain=system.cpu[i].clk_domain,
- ruby_system = ruby_system)
+ # the ruby random tester reuses num_cpus to specify the
+ # number of cpu ports connected to the tester object, which
+ # is stored in system.cpu. because there is only ever one
+ # tester object, num_cpus is not necessarily equal to the
+ # size of system.cpu; therefore if len(system.cpu) == 1
+ # we use system.cpu[0] to set the clk_domain, thereby ensuring
+ # we don't index off the end of the cpu list.
+ if len(system.cpu) == 1:
+ clk_domain = system.cpu[0].clk_domain
+ else:
+ clk_domain = system.cpu[i].clk_domain
+
+ l1_cntrl = L1Cache_Controller(version=i, L1Icache=l1i_cache,
+ L1Dcache=l1d_cache,
+ l2_select_num_bits=l2_bits,
+ send_evictions=send_evicts(options),
+ transitions_per_cycle=options.ports,
+ clk_domain=clk_domain,
+ ruby_system=ruby_system)
+
+ cpu_seq = RubySequencer(version=i, icache=l1i_cache,
+ dcache=l1d_cache, clk_domain=clk_domain,
+ ruby_system=ruby_system)
l1_cntrl.sequencer = cpu_seq
exec("ruby_system.l1_cntrl%d = l1_cntrl" % i)
diff --git a/configs/ruby/MOESI_CMP_token.py b/configs/ruby/MOESI_CMP_token.py
index 23c6d9fef..7161544b7 100644
--- a/configs/ruby/MOESI_CMP_token.py
+++ b/configs/ruby/MOESI_CMP_token.py
@@ -91,29 +91,37 @@ def create_system(options, full_system, system, dma_ports, ruby_system):
assoc = options.l1d_assoc,
start_index_bit = block_size_bits)
- l1_cntrl = L1Cache_Controller(version = i,
- L1Icache = l1i_cache,
- L1Dcache = l1d_cache,
- l2_select_num_bits = l2_bits,
- N_tokens = n_tokens,
- retry_threshold = \
- options.l1_retries,
- fixed_timeout_latency = \
- options.timeout_latency,
- dynamic_timeout_enabled = \
- not options.disable_dyn_timeouts,
- no_mig_atomic = not \
- options.allow_atomic_migration,
- send_evictions = send_evicts(options),
- transitions_per_cycle = options.ports,
- clk_domain=system.cpu[i].clk_domain,
- ruby_system = ruby_system)
-
- cpu_seq = RubySequencer(version = i,
- icache = l1i_cache,
- dcache = l1d_cache,
- clk_domain=system.cpu[i].clk_domain,
- ruby_system = ruby_system)
+ # the ruby random tester reuses num_cpus to specify the
+ # number of cpu ports connected to the tester object, which
+ # is stored in system.cpu. because there is only ever one
+ # tester object, num_cpus is not necessarily equal to the
+ # size of system.cpu; therefore if len(system.cpu) == 1
+ # we use system.cpu[0] to set the clk_domain, thereby ensuring
+ # we don't index off the end of the cpu list.
+ if len(system.cpu) == 1:
+ clk_domain = system.cpu[0].clk_domain
+ else:
+ clk_domain = system.cpu[i].clk_domain
+
+ l1_cntrl = L1Cache_Controller(version=i, L1Icache=l1i_cache,
+ L1Dcache=l1d_cache,
+ l2_select_num_bits=l2_bits,
+ N_tokens=n_tokens,
+ retry_threshold=options.l1_retries,
+ fixed_timeout_latency=\
+ options.timeout_latency,
+ dynamic_timeout_enabled=\
+ not options.disable_dyn_timeouts,
+ no_mig_atomic=not \
+ options.allow_atomic_migration,
+ send_evictions=send_evicts(options),
+ transitions_per_cycle=options.ports,
+ clk_domain=clk_domain,
+ ruby_system=ruby_system)
+
+ cpu_seq = RubySequencer(version=i, icache=l1i_cache,
+ dcache=l1d_cache, clk_domain=clk_domain,
+ ruby_system=ruby_system)
l1_cntrl.sequencer = cpu_seq
exec("ruby_system.l1_cntrl%d = l1_cntrl" % i)
diff --git a/configs/ruby/MOESI_hammer.py b/configs/ruby/MOESI_hammer.py
index 0860fb682..6a1cfd70b 100644
--- a/configs/ruby/MOESI_hammer.py
+++ b/configs/ruby/MOESI_hammer.py
@@ -89,22 +89,30 @@ def create_system(options, full_system, system, dma_ports, ruby_system):
assoc = options.l2_assoc,
start_index_bit = block_size_bits)
- l1_cntrl = L1Cache_Controller(version = i,
- L1Icache = l1i_cache,
- L1Dcache = l1d_cache,
- L2cache = l2_cache,
- no_mig_atomic = not \
- options.allow_atomic_migration,
- send_evictions = send_evicts(options),
- transitions_per_cycle = options.ports,
- clk_domain=system.cpu[i].clk_domain,
- ruby_system = ruby_system)
-
- cpu_seq = RubySequencer(version = i,
- icache = l1i_cache,
- dcache = l1d_cache,
- clk_domain=system.cpu[i].clk_domain,
- ruby_system = ruby_system)
+ # the ruby random tester reuses num_cpus to specify the
+ # number of cpu ports connected to the tester object, which
+ # is stored in system.cpu. because there is only ever one
+ # tester object, num_cpus is not necessarily equal to the
+ # size of system.cpu; therefore if len(system.cpu) == 1
+ # we use system.cpu[0] to set the clk_domain, thereby ensuring
+ # we don't index off the end of the cpu list.
+ if len(system.cpu) == 1:
+ clk_domain = system.cpu[0].clk_domain
+ else:
+ clk_domain = system.cpu[i].clk_domain
+
+ l1_cntrl = L1Cache_Controller(version=i, L1Icache=l1i_cache,
+ L1Dcache=l1d_cache, L2cache=l2_cache,
+ no_mig_atomic=not \
+ options.allow_atomic_migration,
+ send_evictions=send_evicts(options),
+ transitions_per_cycle=options.ports,
+ clk_domain=clk_domain,
+ ruby_system=ruby_system)
+
+ cpu_seq = RubySequencer(version=i, icache=l1i_cache,
+ dcache=l1d_cache,clk_domain=clk_domain,
+ ruby_system=ruby_system)
l1_cntrl.sequencer = cpu_seq
if options.recycle_latency:
diff --git a/src/cpu/testers/rubytest/Check.cc b/src/cpu/testers/rubytest/Check.cc
index 4cdaf9b2f..c8e7816c3 100644
--- a/src/cpu/testers/rubytest/Check.cc
+++ b/src/cpu/testers/rubytest/Check.cc
@@ -94,7 +94,9 @@ Check::initiatePrefetch()
cmd = MemCmd::ReadReq;
// if necessary, make the request an instruction fetch
- if (m_tester_ptr->isInstReadableCpuPort(index)) {
+ if (m_tester_ptr->isInstOnlyCpuPort(index) ||
+ (m_tester_ptr->isInstDataCpuPort(index) &&
+ (random_mt.random(0, 0x1)))) {
flags.set(Request::INST_FETCH);
}
} else {
@@ -193,7 +195,7 @@ Check::initiateAction()
*writeData = m_value + m_store_count;
pkt->dataDynamic(writeData);
- DPRINTF(RubyTest, "data 0x%x check 0x%x\n",
+ DPRINTF(RubyTest, "Seq write: index %d data 0x%x check 0x%x\n", index,
*(pkt->getConstPtr<uint8_t>()), *writeData);
// push the subblock onto the sender state. The sequencer will
@@ -205,6 +207,7 @@ Check::initiateAction()
DPRINTF(RubyTest, "status before action update: %s\n",
(TesterStatus_to_string(m_status)).c_str());
m_status = TesterStatus_Action_Pending;
+ DPRINTF(RubyTest, "Check %s, State=Action_Pending\n", m_address);
} else {
// If the packet did not issue, must delete
// Note: No need to delete the data, the packet destructor
@@ -232,7 +235,9 @@ Check::initiateCheck()
Request::Flags flags;
// If necessary, make the request an instruction fetch
- if (m_tester_ptr->isInstReadableCpuPort(index)) {
+ if (m_tester_ptr->isInstOnlyCpuPort(index) ||
+ (m_tester_ptr->isInstDataCpuPort(index) &&
+ (random_mt.random(0, 0x1)))) {
flags.set(Request::INST_FETCH);
}
@@ -245,6 +250,8 @@ Check::initiateCheck()
uint8_t *dataArray = new uint8_t[CHECK_SIZE];
pkt->dataDynamic(dataArray);
+ DPRINTF(RubyTest, "Seq read: index %d\n", index);
+
// push the subblock onto the sender state. The sequencer will
// update the subblock on the return
pkt->senderState = new SenderState(m_address, req->getSize());
@@ -254,6 +261,7 @@ Check::initiateCheck()
DPRINTF(RubyTest, "status before check update: %s\n",
TesterStatus_to_string(m_status).c_str());
m_status = TesterStatus_Check_Pending;
+ DPRINTF(RubyTest, "Check %s, State=Check_Pending\n", m_address);
} else {
// If the packet did not issue, must delete
// Note: No need to delete the data, the packet destructor
@@ -291,8 +299,11 @@ Check::performCallback(NodeID proc, SubBlock* data, Cycles curTime)
m_store_count++;
if (m_store_count == CHECK_SIZE) {
m_status = TesterStatus_Ready;
+ DPRINTF(RubyTest, "Check %s, State=Ready\n", m_address);
} else {
m_status = TesterStatus_Idle;
+ DPRINTF(RubyTest, "Check %s, State=Idle store_count: %d\n",
+ m_address, m_store_count);
}
DPRINTF(RubyTest, "Action callback return data now %d\n",
data->getByte(0));
@@ -316,6 +327,7 @@ Check::performCallback(NodeID proc, SubBlock* data, Cycles curTime)
m_tester_ptr->incrementCheckCompletions();
m_status = TesterStatus_Idle;
+ DPRINTF(RubyTest, "Check %s, State=Idle\n", m_address);
pickValue();
} else {
@@ -335,6 +347,7 @@ Check::changeAddress(Addr address)
assert(m_status == TesterStatus_Idle || m_status == TesterStatus_Ready);
m_status = TesterStatus_Idle;
m_address = address;
+ DPRINTF(RubyTest, "Check %s, State=Idle\n", m_address);
m_store_count = 0;
}
@@ -342,7 +355,6 @@ void
Check::pickValue()
{
assert(m_status == TesterStatus_Idle);
- m_status = TesterStatus_Idle;
m_value = random_mt.random(0, 0xff); // One byte
m_store_count = 0;
}
@@ -353,7 +365,8 @@ Check::pickInitiatingNode()
assert(m_status == TesterStatus_Idle || m_status == TesterStatus_Ready);
m_status = TesterStatus_Idle;
m_initiatingNode = (random_mt.random(0, m_num_writers - 1));
- DPRINTF(RubyTest, "picked initiating node %d\n", m_initiatingNode);
+ DPRINTF(RubyTest, "Check %s, State=Idle, picked initiating node %d\n",
+ m_address, m_initiatingNode);
m_store_count = 0;
}
diff --git a/src/cpu/testers/rubytest/CheckTable.cc b/src/cpu/testers/rubytest/CheckTable.cc
index b75fd0a52..3bdd73f27 100644
--- a/src/cpu/testers/rubytest/CheckTable.cc
+++ b/src/cpu/testers/rubytest/CheckTable.cc
@@ -42,6 +42,7 @@ CheckTable::CheckTable(int _num_writers, int _num_readers, RubyTester* _tester)
const int size1 = 32;
const int size2 = 100;
+ DPRINTF(RubyTest, "Adding false sharing checks\n");
// The first set is to get some false sharing
physical = 1000;
for (int i = 0; i < size1; i++) {
@@ -50,6 +51,7 @@ CheckTable::CheckTable(int _num_writers, int _num_readers, RubyTester* _tester)
physical += CHECK_SIZE;
}
+ DPRINTF(RubyTest, "Adding cache conflict checks\n");
// The next two sets are to get some limited false sharing and
// cache conflicts
physical = 1000;
@@ -59,6 +61,7 @@ CheckTable::CheckTable(int _num_writers, int _num_readers, RubyTester* _tester)
physical += 256;
}
+ DPRINTF(RubyTest, "Adding cache conflict checks2\n");
physical = 1000 + CHECK_SIZE;
for (int i = 0; i < size2; i++) {
// Setup linear addresses
@@ -91,6 +94,8 @@ CheckTable::addCheck(Addr address)
}
}
+ DPRINTF(RubyTest, "Adding check for address: %s\n", address);
+
Check* check_ptr = new Check(address, 100 + m_check_vector.size(),
m_num_writers, m_num_readers, m_tester_ptr);
for (int i = 0; i < CHECK_SIZE; i++) {
@@ -110,7 +115,7 @@ CheckTable::getRandomCheck()
Check*
CheckTable::getCheck(const Addr address)
{
- DPRINTF(RubyTest, "Looking for check by address: %s", address);
+ DPRINTF(RubyTest, "Looking for check by address: %s\n", address);
auto i = m_lookup_map.find(address);
diff --git a/src/cpu/testers/rubytest/RubyTester.cc b/src/cpu/testers/rubytest/RubyTester.cc
index e0f30f552..5ed6d7f66 100644
--- a/src/cpu/testers/rubytest/RubyTester.cc
+++ b/src/cpu/testers/rubytest/RubyTester.cc
@@ -58,7 +58,8 @@ RubyTester::RubyTester(const Params *p)
m_num_readers(0),
m_wakeup_frequency(p->wakeup_frequency),
m_check_flush(p->check_flush),
- m_num_inst_ports(p->port_cpuInstPort_connection_count)
+ m_num_inst_only_ports(p->port_cpuInstPort_connection_count),
+ m_num_inst_data_ports(p->port_cpuInstDataPort_connection_count)
{
m_checks_completed = 0;
@@ -73,15 +74,25 @@ RubyTester::RubyTester(const Params *p)
// Note: the inst ports are the lowest elements of the readPort vector,
// then the data ports are added to the readPort vector
//
+ int idx = 0;
for (int i = 0; i < p->port_cpuInstPort_connection_count; ++i) {
readPorts.push_back(new CpuPort(csprintf("%s-instPort%d", name(), i),
- this, i));
+ this, i, idx));
+ idx++;
+ }
+ for (int i = 0; i < p->port_cpuInstDataPort_connection_count; ++i) {
+ CpuPort *port = new CpuPort(csprintf("%s-instDataPort%d", name(), i),
+ this, i, idx);
+ readPorts.push_back(port);
+ writePorts.push_back(port);
+ idx++;
}
for (int i = 0; i < p->port_cpuDataPort_connection_count; ++i) {
CpuPort *port = new CpuPort(csprintf("%s-dataPort%d", name(), i),
- this, i);
+ this, i, idx);
readPorts.push_back(port);
writePorts.push_back(port);
+ idx++;
}
// add the check start event to the event queue
@@ -108,6 +119,7 @@ RubyTester::init()
m_num_writers = writePorts.size();
m_num_readers = readPorts.size();
+ assert(m_num_readers == m_num_cpus);
m_checkTable_ptr = new CheckTable(m_num_writers, m_num_readers, this);
}
@@ -115,32 +127,45 @@ RubyTester::init()
BaseMasterPort &
RubyTester::getMasterPort(const std::string &if_name, PortID idx)
{
- if (if_name != "cpuInstPort" && if_name != "cpuDataPort") {
+ if (if_name != "cpuInstPort" && if_name != "cpuInstDataPort" &&
+ if_name != "cpuDataPort") {
// pass it along to our super class
return MemObject::getMasterPort(if_name, idx);
} else {
if (if_name == "cpuInstPort") {
- if (idx > m_num_inst_ports) {
- panic("RubyTester::getMasterPort: unknown inst port idx %d\n",
+ if (idx > m_num_inst_only_ports) {
+ panic("RubyTester::getMasterPort: unknown inst port %d\n",
idx);
}
//
- // inst ports directly map to the lowest readPort elements
+ // inst ports map to the lowest readPort elements
//
return *readPorts[idx];
+ } else if (if_name == "cpuInstDataPort") {
+ if (idx > m_num_inst_data_ports) {
+ panic("RubyTester::getMasterPort: unknown inst+data port %d\n",
+ idx);
+ }
+ int read_idx = idx + m_num_inst_only_ports;
+ //
+ // inst+data ports map to the next readPort elements
+ //
+ return *readPorts[read_idx];
} else {
assert(if_name == "cpuDataPort");
//
- // add the inst port offset to translate to the correct read port
- // index
+ // data only ports map to the final readPort elements
//
- int read_idx = idx + m_num_inst_ports;
- if (read_idx >= static_cast<PortID>(readPorts.size())) {
- panic("RubyTester::getMasterPort: unknown data port idx %d\n",
+ if (idx > (static_cast<int>(readPorts.size()) -
+ (m_num_inst_only_ports + m_num_inst_data_ports))) {
+ panic("RubyTester::getMasterPort: unknown data port %d\n",
idx);
}
+ int read_idx = idx + m_num_inst_only_ports + m_num_inst_data_ports;
return *readPorts[read_idx];
}
+ // Note: currently the Ruby Tester does not support write only ports
+ // but that could easily be added here
}
}
@@ -152,7 +177,7 @@ RubyTester::CpuPort::recvTimingResp(PacketPtr pkt)
safe_cast<RubyTester::SenderState*>(pkt->senderState);
SubBlock& subblock = senderState->subBlock;
- tester->hitCallback(id, &subblock);
+ tester->hitCallback(globalIdx, &subblock);
// Now that the tester has completed, delete the senderState
// (includes sublock) and the packet, then return
@@ -163,9 +188,16 @@ RubyTester::CpuPort::recvTimingResp(PacketPtr pkt)
}
bool
-RubyTester::isInstReadableCpuPort(int idx)
+RubyTester::isInstOnlyCpuPort(int idx)
+{
+ return idx < m_num_inst_only_ports;
+}
+
+bool
+RubyTester::isInstDataCpuPort(int idx)
{
- return idx < m_num_inst_ports;
+ return ((idx >= m_num_inst_only_ports) &&
+ (idx < (m_num_inst_only_ports + m_num_inst_data_ports)));
}
MasterPort*
@@ -190,13 +222,13 @@ RubyTester::hitCallback(NodeID proc, SubBlock* data)
// Mark that we made progress
m_last_progress_vector[proc] = curCycle();
- DPRINTF(RubyTest, "completed request for proc: %d\n", proc);
- DPRINTF(RubyTest, "addr: 0x%x, size: %d, data: ",
+ DPRINTF(RubyTest, "completed request for proc: %d", proc);
+ DPRINTFR(RubyTest, " addr: 0x%x, size: %d, data: ",
data->getAddress(), data->getSize());
for (int byte = 0; byte < data->getSize(); byte++) {
- DPRINTF(RubyTest, "%d", data->getByte(byte));
+ DPRINTFR(RubyTest, "%d ", data->getByte(byte));
}
- DPRINTF(RubyTest, "\n");
+ DPRINTFR(RubyTest, "\n");
// This tells us our store has 'completed' or for a load gives us
// back the data to make the check
diff --git a/src/cpu/testers/rubytest/RubyTester.hh b/src/cpu/testers/rubytest/RubyTester.hh
index 94a982e32..39e6d78a3 100644
--- a/src/cpu/testers/rubytest/RubyTester.hh
+++ b/src/cpu/testers/rubytest/RubyTester.hh
@@ -60,6 +60,8 @@ class RubyTester : public MemObject
{
private:
RubyTester *tester;
+ // index for m_last_progress_vector and hitCallback
+ PortID globalIdx;
public:
//
@@ -68,8 +70,10 @@ class RubyTester : public MemObject
// RubyPorts that support both types of requests, separate InstOnly
// and DataOnly CpuPorts will map to that RubyPort
- CpuPort(const std::string &_name, RubyTester *_tester, PortID _id)
- : MasterPort(_name, _tester, _id), tester(_tester)
+ CpuPort(const std::string &_name, RubyTester *_tester, PortID _id,
+ PortID _index)
+ : MasterPort(_name, _tester, _id), tester(_tester),
+ globalIdx(_index)
{}
protected:
@@ -93,7 +97,8 @@ class RubyTester : public MemObject
virtual BaseMasterPort &getMasterPort(const std::string &if_name,
PortID idx = InvalidPortID);
- bool isInstReadableCpuPort(int idx);
+ bool isInstOnlyCpuPort(int idx);
+ bool isInstDataCpuPort(int idx);
MasterPort* getReadableCpuPort(int idx);
MasterPort* getWritableCpuPort(int idx);
@@ -152,7 +157,8 @@ class RubyTester : public MemObject
int m_num_readers;
int m_wakeup_frequency;
bool m_check_flush;
- int m_num_inst_ports;
+ int m_num_inst_only_ports;
+ int m_num_inst_data_ports;
};
inline std::ostream&
diff --git a/src/cpu/testers/rubytest/RubyTester.py b/src/cpu/testers/rubytest/RubyTester.py
index 7af70cae0..f12485566 100644
--- a/src/cpu/testers/rubytest/RubyTester.py
+++ b/src/cpu/testers/rubytest/RubyTester.py
@@ -34,8 +34,9 @@ class RubyTester(MemObject):
type = 'RubyTester'
cxx_header = "cpu/testers/rubytest/RubyTester.hh"
num_cpus = Param.Int("number of cpus / RubyPorts")
- cpuDataPort = VectorMasterPort("the cpu data cache ports")
- cpuInstPort = VectorMasterPort("the cpu inst cache ports")
+ cpuInstDataPort = VectorMasterPort("cpu combo ports to inst & data caches")
+ cpuInstPort = VectorMasterPort("cpu ports to only inst caches")
+ cpuDataPort = VectorMasterPort("cpu ports to only data caches")
checks_to_complete = Param.Int(100, "checks to complete")
deadlock_threshold = Param.Int(50000, "how often to check for deadlock")
wakeup_frequency = Param.Int(10, "number of cycles between wakeups")
diff --git a/src/mem/ruby/system/RubyPort.cc b/src/mem/ruby/system/RubyPort.cc
index b2fb8d72d..95a83873c 100644
--- a/src/mem/ruby/system/RubyPort.cc
+++ b/src/mem/ruby/system/RubyPort.cc
@@ -11,7 +11,7 @@
* unmodified and in its entirety in all distributions of the software,
* modified or unmodified, in source code or in binary form.
*
- * Copyright (c) 2009 Advanced Micro Devices, Inc.
+ * Copyright (c) 2009-2013 Advanced Micro Devices, Inc.
* Copyright (c) 2011 Mark D. Hill and David A. Wood
* All rights reserved.
*
@@ -58,7 +58,8 @@ RubyPort::RubyPort(const Params *p)
pioSlavePort(csprintf("%s.pio-slave-port", name()), this),
memMasterPort(csprintf("%s.mem-master-port", name()), this),
memSlavePort(csprintf("%s-mem-slave-port", name()), this,
- p->ruby_system->getAccessBackingStore(), -1),
+ p->ruby_system->getAccessBackingStore(), -1,
+ p->no_retry_on_stall),
gotAddrRanges(p->port_master_connection_count)
{
assert(m_version != -1);
@@ -66,7 +67,8 @@ RubyPort::RubyPort(const Params *p)
// create the slave ports based on the number of connected ports
for (size_t i = 0; i < p->port_slave_connection_count; ++i) {
slave_ports.push_back(new MemSlavePort(csprintf("%s.slave%d", name(),
- i), this, p->ruby_system->getAccessBackingStore(), i));
+ i), this, p->ruby_system->getAccessBackingStore(),
+ i, p->no_retry_on_stall));
}
// create the master ports based on the number of connected ports
@@ -156,9 +158,11 @@ RubyPort::MemMasterPort::MemMasterPort(const std::string &_name,
}
RubyPort::MemSlavePort::MemSlavePort(const std::string &_name, RubyPort *_port,
- bool _access_backing_store, PortID id)
+ bool _access_backing_store, PortID id,
+ bool _no_retry_on_stall)
: QueuedSlavePort(_name, _port, queue, id), queue(*_port, *this),
- access_backing_store(_access_backing_store)
+ access_backing_store(_access_backing_store),
+ no_retry_on_stall(_no_retry_on_stall)
{
DPRINTF(RubyPort, "Created slave memport on ruby sequencer %s\n", _name);
}
@@ -267,21 +271,31 @@ RubyPort::MemSlavePort::recvTimingReq(PacketPtr pkt)
return true;
}
- //
- // Unless one is using the ruby tester, record the stalled M5 port for
- // later retry when the sequencer becomes free.
- //
- if (!ruby_port->m_usingRubyTester) {
- ruby_port->addToRetryList(this);
- }
DPRINTF(RubyPort, "Request for address %#x did not issued because %s\n",
pkt->getAddr(), RequestStatus_to_string(requestStatus));
+ addToRetryList();
+
return false;
}
void
+RubyPort::MemSlavePort::addToRetryList()
+{
+ RubyPort *ruby_port = static_cast<RubyPort *>(&owner);
+
+ //
+ // Unless the requestor do not want retries (e.g., the Ruby tester),
+ // record the stalled M5 port for later retry when the sequencer
+ // becomes free.
+ //
+ if (!no_retry_on_stall && !ruby_port->onRetryList(this)) {
+ ruby_port->addToRetryList(this);
+ }
+}
+
+void
RubyPort::MemSlavePort::recvFunctional(PacketPtr pkt)
{
DPRINTF(RubyPort, "Functional access for address: %#x\n", pkt->getAddr());
@@ -356,31 +370,33 @@ RubyPort::ruby_hit_callback(PacketPtr pkt)
port->hitCallback(pkt);
+ trySendRetries();
+}
+
+void
+RubyPort::trySendRetries()
+{
//
// If we had to stall the MemSlavePorts, wake them up because the sequencer
// likely has free resources now.
//
if (!retryList.empty()) {
- //
- // Record the current list of ports to retry on a temporary list before
- // calling sendRetry on those ports. sendRetry will cause an
- // immediate retry, which may result in the ports being put back on the
- // list. Therefore we want to clear the retryList before calling
- // sendRetry.
- //
+ // Record the current list of ports to retry on a temporary list
+ // before calling sendRetryReq on those ports. sendRetryReq will cause
+ // an immediate retry, which may result in the ports being put back on
+ // the list. Therefore we want to clear the retryList before calling
+ // sendRetryReq.
std::vector<MemSlavePort *> curRetryList(retryList);
retryList.clear();
for (auto i = curRetryList.begin(); i != curRetryList.end(); ++i) {
DPRINTF(RubyPort,
- "Sequencer may now be free. SendRetry to port %s\n",
+ "Sequencer may now be free. SendRetry to port %s\n",
(*i)->name());
(*i)->sendRetryReq();
}
}
-
- testDrainComplete();
}
void
diff --git a/src/mem/ruby/system/RubyPort.hh b/src/mem/ruby/system/RubyPort.hh
index 58d2558dd..07e0fde5a 100644
--- a/src/mem/ruby/system/RubyPort.hh
+++ b/src/mem/ruby/system/RubyPort.hh
@@ -11,7 +11,7 @@
* unmodified and in its entirety in all distributions of the software,
* modified or unmodified, in source code or in binary form.
*
- * Copyright (c) 2009 Advanced Micro Devices, Inc.
+ * Copyright (c) 2009-2013 Advanced Micro Devices, Inc.
* Copyright (c) 2011 Mark D. Hill and David A. Wood
* All rights reserved.
*
@@ -76,10 +76,12 @@ class RubyPort : public MemObject
private:
RespPacketQueue queue;
bool access_backing_store;
+ bool no_retry_on_stall;
public:
MemSlavePort(const std::string &_name, RubyPort *_port,
- bool _access_backing_store, PortID id);
+ bool _access_backing_store,
+ PortID id, bool _no_retry_on_stall);
void hitCallback(PacketPtr pkt);
void evictionCallback(Addr address);
@@ -94,6 +96,8 @@ class RubyPort : public MemObject
AddrRangeList getAddrRanges() const
{ AddrRangeList ranges; return ranges; }
+ void addToRetryList();
+
private:
bool isPhysMemAddress(Addr addr) const;
};
@@ -164,6 +168,7 @@ class RubyPort : public MemObject
DrainState drain() override;
protected:
+ void trySendRetries();
void ruby_hit_callback(PacketPtr pkt);
void testDrainComplete();
void ruby_eviction_callback(Addr address);
@@ -186,10 +191,14 @@ class RubyPort : public MemObject
System* system;
private:
+ bool onRetryList(MemSlavePort * port)
+ {
+ return (std::find(retryList.begin(), retryList.end(), port) !=
+ retryList.end());
+ }
void addToRetryList(MemSlavePort * port)
{
- if (std::find(retryList.begin(), retryList.end(), port) !=
- retryList.end()) return;
+ if (onRetryList(port)) return;
retryList.push_back(port);
}
diff --git a/src/mem/ruby/system/Sequencer.cc b/src/mem/ruby/system/Sequencer.cc
index 26db6b6f8..50418c700 100644
--- a/src/mem/ruby/system/Sequencer.cc
+++ b/src/mem/ruby/system/Sequencer.cc
@@ -491,6 +491,7 @@ Sequencer::hitCallback(SequencerRequest* srequest, DataBlock& data,
rs->m_cache_recorder->enqueueNextFlushRequest();
} else {
ruby_hit_callback(pkt);
+ testDrainComplete();
}
}
diff --git a/src/mem/ruby/system/Sequencer.py b/src/mem/ruby/system/Sequencer.py
index 7494986e9..7c90eb29c 100644
--- a/src/mem/ruby/system/Sequencer.py
+++ b/src/mem/ruby/system/Sequencer.py
@@ -45,6 +45,7 @@ class RubyPort(MemObject):
mem_slave_port = SlavePort("Ruby memory port")
using_ruby_tester = Param.Bool(False, "")
+ no_retry_on_stall = Param.Bool(False, "")
ruby_system = Param.RubySystem(Parent.any, "")
system = Param.System(Parent.any, "system object")
support_data_reqs = Param.Bool(True, "data cache requests supported")
generated by cgit v1.2.3 (git 2.25.1) at 2025-01-26 21:42:55 +0000