LLVM

[wangli1426]

Low Level Virtual Machine (LLVM) is widely used in the efficient implementation of database system, e.g., impala, hyper, etc.

Given the inefficient implementation of expression calculation in Claims, it is worthy to consider LLVM for performance improvement.

[wangli1426]

I find an interesting page that introduces code generation through LLVM.
http://llvm-tutorial-cn.readthedocs.org/en/latest/chapter-3.html

[wangli1426]

I have already known the basic steps to generate code by LLVM. @scdong
The next job is to make sure which codes are needed to generate and how to generate.

In the following, I will list the critical section requiring code generation.

block->insert(tuple_in_block,tuple_from_child,bytes);
const unsigned bytes=state_.schema_->getTupleActualSize(tuple_from_child);
const void* key_addr=input_schema->getColumnAddess(state_.joinIndex_left[0],cur);
bn=op->getPartitionValue(key_addr,buckets);
input_schema->copyTuple(cur,tuple_in_hashtable);
operate->equal(key_in_input,key_in_hashtable);
cur=bsti->currentTuple();
bsti->increase_cur_();
ExpressionCalculator::calcuate(toCalc,result);

[wangli1426]

It is better to know better through performance test.

1. memcpy Generated code vs non-generated code: memory copy #39
2. filtering Generated code vs non-generated code: a case study on filtering #45