Tiago Muck has submitted this change. (
https://gem5-review.googlesource.com/c/public/gem5/+/41862 )
Change subject: mem-ruby: dequeue rate limit for message buffers
......................................................................
mem-ruby: dequeue rate limit for message buffers
The 'max_dequeue_rate' parameter limits the rate at which messages can
be dequeued in a single cycle. When set, 'isReady' returns false if
after max_dequeue_rate is reached.
This can be used to fine tune the performance of cache controllers.
For the record, other ways of achieving a similar effect could be:
1) Modifying the SLICC compiler to limit message consumption in the
generated wakeup() function
2) Set the buffer size to max_dequeue_rate. This can potentially cut the
the expected throughput in half. For instance if a producer can
enqueue every cycle, and a consumer can dequeue every cycle, a
message can only be actually enqueued every two (assuming
buffer_size=1) since the buffer entries available after dequeue
are only visible in the next cycle (even if the consumer executes
before the producer).
JIRA: https://gem5.atlassian.net/browse/GEM5-920
Change-Id: I3a446c7276b80a0e3f409b4fbab0ab65ff5c1f81
Signed-off-by: Tiago Mück <[email protected]>
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/41862
Reviewed-by: Meatboy 106 <[email protected]>
Maintainer: Bobby Bruce <[email protected]>
Tested-by: kokoro <[email protected]>
---
M src/mem/ruby/network/MessageBuffer.cc
M src/mem/ruby/network/MessageBuffer.hh
M src/mem/ruby/network/MessageBuffer.py
3 files changed, 61 insertions(+), 5 deletions(-)
Approvals:
Meatboy 106: Looks good to me, approved
Bobby Bruce: Looks good to me, approved
kokoro: Regressions pass
diff --git a/src/mem/ruby/network/MessageBuffer.cc
b/src/mem/ruby/network/MessageBuffer.cc
index 5c0e116..85cc00c 100644
--- a/src/mem/ruby/network/MessageBuffer.cc
+++ b/src/mem/ruby/network/MessageBuffer.cc
@@ -58,8 +58,9 @@
using stl_helpers::operator<<;
MessageBuffer::MessageBuffer(const Params &p)
- : SimObject(p), m_stall_map_size(0),
- m_max_size(p.buffer_size), m_time_last_time_size_checked(0),
+ : SimObject(p), m_stall_map_size(0), m_max_size(p.buffer_size),
+ m_max_dequeue_rate(p.max_dequeue_rate), m_dequeues_this_cy(0),
+ m_time_last_time_size_checked(0),
m_time_last_time_enqueue(0), m_time_last_time_pop(0),
m_last_arrival_time(0), m_strict_fifo(p.ordered),
m_randomization(p.randomization),
@@ -312,7 +313,9 @@
m_size_at_cycle_start = m_prio_heap.size();
m_stalled_at_cycle_start = m_stall_map_size;
m_time_last_time_pop = current_time;
+ m_dequeues_this_cy = 0;
}
+ ++m_dequeues_this_cy;
pop_heap(m_prio_heap.begin(), m_prio_heap.end(),
std::greater<MsgPtr>());
m_prio_heap.pop_back();
@@ -507,8 +510,17 @@
bool
MessageBuffer::isReady(Tick current_time) const
{
- return ((m_prio_heap.size() > 0) &&
- (m_prio_heap.front()->getLastEnqueueTime() <= current_time));
+ assert(m_time_last_time_pop <= current_time);
+ bool can_dequeue = (m_max_dequeue_rate == 0) ||
+ (m_time_last_time_pop < current_time) ||
+ (m_dequeues_this_cy < m_max_dequeue_rate);
+ bool is_ready = (m_prio_heap.size() > 0) &&
+ (m_prio_heap.front()->getLastEnqueueTime() <=
current_time);
+ if (!can_dequeue && is_ready) {
+ // Make sure the Consumer executes next cycle to dequeue the ready
msg
+ m_consumer->scheduleEvent(Cycles(1));
+ }
+ return can_dequeue && is_ready;
}
uint32_t
diff --git a/src/mem/ruby/network/MessageBuffer.hh
b/src/mem/ruby/network/MessageBuffer.hh
index 7a48a1d..fb0ef43 100644
--- a/src/mem/ruby/network/MessageBuffer.hh
+++ b/src/mem/ruby/network/MessageBuffer.hh
@@ -240,6 +240,14 @@
*/
const unsigned int m_max_size;
+ /**
+ * When != 0, isReady returns false once m_max_dequeue_rate
+ * messages have been dequeued in the same cycle.
+ */
+ const unsigned int m_max_dequeue_rate;
+
+ unsigned int m_dequeues_this_cy;
+
Tick m_time_last_time_size_checked;
unsigned int m_size_last_time_size_checked;
diff --git a/src/mem/ruby/network/MessageBuffer.py
b/src/mem/ruby/network/MessageBuffer.py
index b30726e..80dc872 100644
--- a/src/mem/ruby/network/MessageBuffer.py
+++ b/src/mem/ruby/network/MessageBuffer.py
@@ -1,4 +1,4 @@
-# Copyright (c) 2020 ARM Limited
+# Copyright (c) 2020-2021 ARM Limited
# All rights reserved.
#
# The license below extends only to copyright in the software and shall
@@ -67,3 +67,6 @@
master = DeprecatedParam(out_port, '`master` is now called `out_port`')
in_port = ResponsePort("Response port from MessageBuffer sender")
slave = DeprecatedParam(in_port, '`slave` is now called `in_port`')
+ max_dequeue_rate = Param.Unsigned(0, "Maximum number of messages that
can \
+ be dequeued per cycle \
+ (0 allows dequeueing all ready
messages)")
--
To view, visit https://gem5-review.googlesource.com/c/public/gem5/+/41862
To unsubscribe, or for help writing mail filters, visit
https://gem5-review.googlesource.com/settings
Gerrit-Project: public/gem5
Gerrit-Branch: develop
Gerrit-Change-Id: I3a446c7276b80a0e3f409b4fbab0ab65ff5c1f81
Gerrit-Change-Number: 41862
Gerrit-PatchSet: 4
Gerrit-Owner: Tiago Muck <[email protected]>
Gerrit-Reviewer: Bobby Bruce <[email protected]>
Gerrit-Reviewer: Jason Lowe-Power <[email protected]>
Gerrit-Reviewer: Jason Lowe-Power <[email protected]>
Gerrit-Reviewer: John Alsop <[email protected]>
Gerrit-Reviewer: Lilia <[email protected]>
Gerrit-Reviewer: Meatboy 106 <[email protected]>
Gerrit-Reviewer: Tiago Muck <[email protected]>
Gerrit-Reviewer: kokoro <[email protected]>
Gerrit-CC: Giacomo Travaglini <[email protected]>
Gerrit-MessageType: merged
_______________________________________________
gem5-dev mailing list -- [email protected]
To unsubscribe send an email to [email protected]
%(web_page_url)slistinfo%(cgiext)s/%(_internal_name)s
