Exclude concurrent inames from get_iname_duplication_options because they are never explicitly 'entered'

Reduce logger chattiness during kernel creation

See merge request !244

Reduce cache-to-execution latency by using lazy unpickling of instructions.

Closes #67

See merge request inducer/loopy!242

of instructions.

This requires caching the generation of invokers (which itself should
also give a speed boost), since invoker generation needed to analyze
the instructions in the kernel.

allowing stride=0 in make_args and make_ref_args by calculating alloc_size as if stride=1

See merge request !241

Try hard to find a git revision to use as cache key

See merge request inducer/loopy!239

Stop using apparently-deprecated Set.add in save/load transform

See merge request !238

Speed up overlap checking (closes #131)

Closes #131

See merge request !237

computations when possible.

Adding all local strides to mem access

See merge request inducer/loopy!236

switched lid_strides from list to dict in order to distinguish between two cases: (1) where an lid was found with stride 0 and (2) where no lid was found

remember instructions sorting between recursions

See merge request inducer/loopy!229