o3: issue excl. prefetch as soon as store address is available
Review Request #2277 - Created May 13, 2014 and updated
| Information | |
|---|---|
| Steve Reinhardt | |
| gem5 | |
| default | |
| Reviewers | |
| Default | |
Changeset 10229:4da865dce0da --------------------------- o3: issue excl. prefetch as soon as store address is available Issue StoreAccess to acquire permission for store after translation. In Classic: StoreAccess gets converted to HardPFExReq and shares the existing prefetch queue. Therefore, it only affects performance if the prefetcher is enabled in classic. In Ruby: StoreAccess gets converted to RubyStore with NULL data field. This work was done while Binh was an intern at AMD Research. This patch includes bug fixes by Marc Orr.
Issue Summary
9
5
4
0
| Description | From | Last Updated | Status |
|---|---|---|---|
| What happens to the packet? Memory leak, or is it: 1) kept somewhere, or 2) deallocated somewhere? | Andreas Hansson | May 15, 2014, 3:43 p.m. | Open |
| Should this not be part of the if-statement above if (prefetcher)...? | Andreas Hansson | May 15, 2014, 3:43 p.m. | Open |
| How could it every be empty? | Andreas Hansson | May 15, 2014, 3:43 p.m. | Open |
| I don't find the name overly descriptive. Is it just me? | Andreas Hansson | May 15, 2014, 3:43 p.m. | Open |
| It would be nice to not change this here, and rather deal with it at the call site somehow. | Andreas Hansson | May 15, 2014, 3:43 p.m. | Open |
Posted (May 15, 2014, 3:43 p.m.)
-
src/cpu/o3/lsq_unit_impl.hh (Diff revision 1) -
What happens to the packet? Memory leak, or is it: 1) kept somewhere, or 2) deallocated somewhere?
-
src/mem/cache/cache_impl.hh (Diff revision 1) -
Should this not be part of the if-statement above if (prefetcher)...?
-
src/mem/cache/prefetch/base.cc (Diff revision 1) -
Should be removed
-
src/mem/cache/prefetch/base.cc (Diff revision 1) -
How could it every be empty?
-
src/mem/coherent_bus.cc (Diff revision 1) -
Spurious whitespace
-
src/mem/packet.hh (Diff revision 1) -
I don't find the name overly descriptive. Is it just me?
-
src/mem/packet.hh (Diff revision 1) -
It would be nice to not change this here, and rather deal with it at the call site somehow.
-
src/mem/ruby/system/Sequencer.cc (Diff revision 1) -
80 char?
-
src/mem/ruby/system/Sequencer.cc (Diff revision 1) -
A few of these in this patch with else being on a new line.