Thread local hack to make transRead faster...
authorbdemsky <bdemsky>
Fri, 27 Feb 2009 04:34:51 +0000 (04:34 +0000)
committerbdemsky <bdemsky>
Fri, 27 Feb 2009 04:34:51 +0000 (04:34 +0000)
commit97ffd10a70e49bd94cdf8fe41b07ba3d2778e00f
tree4b02f6716d32abe85511390ac8c64d415f3885d1
parent82b39745b4c18fe630dab8ca2d885d4b827771f1
Thread local hack to make transRead faster...
1) make transrecords threadlocal so we don't pass them around anymore
2) inline the clookup into the code...call transread only if we miss on the transaction
cache

This doesn't help much for the current benchmark set, but should make a significant
difference if benchmarks read fields a lot.
14 files changed:
Robust/src/IR/Flat/BuildCode.java
Robust/src/Runtime/DSTM/interface/abortreaders.c
Robust/src/Runtime/DSTM/interface/abortreaders.h
Robust/src/Runtime/DSTM/interface/addPrefetchEnhance.c
Robust/src/Runtime/DSTM/interface/addPrefetchEnhance.h
Robust/src/Runtime/DSTM/interface/clookup.c
Robust/src/Runtime/DSTM/interface/clookup.h
Robust/src/Runtime/DSTM/interface/dstm.h
Robust/src/Runtime/DSTM/interface/trans.c
Robust/src/Runtime/DSTM/interface/trans.h [new file with mode: 0644]
Robust/src/Runtime/garbage.c
Robust/src/Runtime/runtime.c
Robust/src/Runtime/runtime.h
Robust/src/Runtime/thread.c