derek/gem5
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

mem: Adjust cache queue reserve to more conservative values

Browse Source 
The cache queue reserve is there as an overflow to give us enough
headroom based on when we block the cache, and how many transactions
we may already have accepted before actually blocking. The previous
values were probably chosen to be "big enough", when we actually know
that we check the MSHRs after every single allocation, and for the
write buffers we know that we implicitly may need one entry for every
outstanding MSHR.
* * *
mem: Adjust cache queue reserve to more conservative values

The cache queue reserve is there as an overflow to give us enough
headroom based on when we block the cache, and how many transactions
we may already have accepted before actually blocking. The previous
values were probably chosen to be "big enough", when we actually know
that we check the MSHRs after every single allocation, and for the
write buffers we know that we implicitly may need one entry for every
outstanding MSHR.
This commit is contained in:
Andreas Hansson
2016-03-17 09:51:22 -04:00
parent 7a40e7864a
commit abcbc4e51e
3 changed files with 15 additions and 7 deletions
Download Patch File Download Diff File Expand all files Collapse all files

10
src/mem/cache/base.cc vendored
View File

@@ -68,8 +68,8 @@ BaseCache::CacheSlavePort::CacheSlavePort(const std::string &_name,
BaseCache::BaseCache(const BaseCacheParams *p, unsigned blk_size)
: MemObject(p),
cpuSidePort(nullptr), memSidePort(nullptr),
mshrQueue("MSHRs", p->mshrs, 4, p->demand_mshr_reserve),
writeBuffer("write buffer", p->write_buffers, p->mshrs+1000),
mshrQueue("MSHRs", p->mshrs, 0, p->demand_mshr_reserve), // see below
writeBuffer("write buffer", p->write_buffers, p->mshrs), // see below
blkSize(blk_size),
lookupLatency(p->hit_latency),
forwardLatency(p->hit_latency),
@@ -85,6 +85,12 @@ BaseCache::BaseCache(const BaseCacheParams *p, unsigned blk_size)
addrRanges(p->addr_ranges.begin(), p->addr_ranges.end()),
system(p->system)
{
// the MSHR queue has no reserve entries as we check the MSHR
// queue on every single allocation, whereas the write queue has
// as many reserve entries as we have MSHRs, since every MSHR may
// eventually require a writeback, and we do not check the write
// buffer before committing to an MSHR
// forward snoops is overridden in init() once we can query
// whether the connected master is actually snooping or not
}