Daniel Carvalho has submitted this change and it was merged. (
https://gem5-review.googlesource.com/c/public/gem5/+/14876 )
Change subject: mem-cache: Use header delay on latency calculation
......................................................................
mem-cache: Use header delay on latency calculation
Previously the bus delay was being ignored for the access latency
calculation, and then applied on top of the access latency. This
patch fixes the order, as first the packet must arrive before the
access starts.
Change-Id: I6d55299a911d54625c147814dd423bfc63ef1b65
Signed-off-by: Daniel R. Carvalho <[email protected]>
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/14876
Reviewed-by: Nikos Nikoleris <[email protected]>
Maintainer: Nikos Nikoleris <[email protected]>
---
M src/mem/cache/base.cc
M src/mem/cache/base.hh
2 files changed, 22 insertions(+), 13 deletions(-)
Approvals:
Nikos Nikoleris: Looks good to me, approved; Looks good to me, approved
diff --git a/src/mem/cache/base.cc b/src/mem/cache/base.cc
index 0949e49..d4e93c0 100644
--- a/src/mem/cache/base.cc
+++ b/src/mem/cache/base.cc
@@ -365,7 +365,7 @@
// The latency charged is just the value set by the access() function.
// In case of a hit we are neglecting response latency.
// In case of a miss we are neglecting forward latency.
- Tick request_time = clockEdge(lat) + pkt->headerDelay;
+ Tick request_time = clockEdge(lat);
// Here we reset the timing of the packet.
pkt->headerDelay = pkt->payloadDelay = 0;
@@ -889,27 +889,33 @@
//
/////////////////////////////////////////////////////
Cycles
-BaseCache::calculateAccessLatency(const CacheBlk* blk,
+BaseCache::calculateAccessLatency(const CacheBlk* blk, const uint32_t
delay,
const Cycles lookup_lat) const
{
- Cycles lat(lookup_lat);
+ Cycles lat(0);
if (blk != nullptr) {
- // First access tags, then data
+ // As soon as the access arrives, for sequential accesses first
access
+ // tags, then the data entry. In the case of parallel accesses the
+ // latency is dictated by the slowest of tag and data latencies.
if (sequentialAccess) {
- lat += dataLatency;
- // Latency is dictated by the slowest of tag and data latencies
+ lat = ticksToCycles(delay) + lookup_lat + dataLatency;
} else {
- lat = std::max(lookup_lat, dataLatency);
+ lat = ticksToCycles(delay) + std::max(lookup_lat, dataLatency);
}
// Check if the block to be accessed is available. If not, apply
the
// access latency on top of when the block is ready to be accessed.
+ const Tick tick = curTick() + delay;
const Tick when_ready = blk->getWhenReady();
- if (when_ready > curTick() &&
- ticksToCycles(when_ready - curTick()) > lat) {
- lat += ticksToCycles(when_ready - curTick());
+ if (when_ready > tick &&
+ ticksToCycles(when_ready - tick) > lat) {
+ lat += ticksToCycles(when_ready - tick);
}
+ } else {
+ // In case of a miss, apply lookup latency on top of the metadata
+ // delay, as the access can only start when it arrives.
+ lat = ticksToCycles(delay) + lookup_lat;
}
return lat;
@@ -930,8 +936,10 @@
Cycles tag_latency(0);
blk = tags->accessBlock(pkt->getAddr(), pkt->isSecure(), tag_latency);
- // Calculate access latency
- lat = calculateAccessLatency(blk, tag_latency);
+ // Calculate access latency on top of when the packet arrives. This
+ // takes into account the bus delay.
+ lat = calculateAccessLatency(blk, pkt->headerDelay,
+ tag_latency);
DPRINTF(Cache, "%s for %s %s\n", __func__, pkt->print(),
blk ? "hit " + blk->print() : "miss");
diff --git a/src/mem/cache/base.hh b/src/mem/cache/base.hh
index 1137241..7bf7411 100644
--- a/src/mem/cache/base.hh
+++ b/src/mem/cache/base.hh
@@ -426,10 +426,11 @@
* whether access was a hit or miss.
*
* @param blk The cache block that was accessed.
+ * @param delay The delay until the packet's metadata is present.
* @param lookup_lat Latency of the respective tag lookup.
* @return The number of ticks that pass due to a block access.
*/
- Cycles calculateAccessLatency(const CacheBlk* blk,
+ Cycles calculateAccessLatency(const CacheBlk* blk, const uint32_t
delay,
const Cycles lookup_lat) const;
/**
--
To view, visit https://gem5-review.googlesource.com/c/public/gem5/+/14876
To unsubscribe, or for help writing mail filters, visit
https://gem5-review.googlesource.com/settings
Gerrit-Project: public/gem5
Gerrit-Branch: master
Gerrit-Change-Id: I6d55299a911d54625c147814dd423bfc63ef1b65
Gerrit-Change-Number: 14876
Gerrit-PatchSet: 7
Gerrit-Owner: Daniel Carvalho <[email protected]>
Gerrit-Reviewer: Daniel Carvalho <[email protected]>
Gerrit-Reviewer: Jason Lowe-Power <[email protected]>
Gerrit-Reviewer: Nikos Nikoleris <[email protected]>
Gerrit-MessageType: merged
_______________________________________________
gem5-dev mailing list
[email protected]
http://m5sim.org/mailman/listinfo/gem5-dev
