Todd Lipcon has posted comments on this change.

Change subject: WIP: flatbuffers-based codegen
......................................................................


Patch Set 2:

(3 comments)

thanks for taking a look Vlad!

http://gerrit.cloudera.org:8080/#/c/2903/2/src/kudu/codegen/precompiled.cc
File src/kudu/codegen/precompiled.cc:

Line 111: 0
> #if 0 ?
this was is a hackathon-produced WIP so some messiness in here still needs a 
cleanup. You found one such messiness :) (this should probably just be an 
NDEBUG check)


Line 142:   for (auto dst_col_idx : *fb->default_cols()) {
> Why no #pragma unroll here? Why #pragma unroll at all? Shouldn't LLVM figur
ah, probably should be one.

The #pragma unroll is because clang's heuristic is to not unroll loops with 
known trip counts if the trip count is more than 8. So for schemas with <8 
columns, it'll unroll anyway, but with more, you have to tell it to do so. It's 
not always smart enough to see that by unrolling the loop it gets to eliminate 
a lot of constants.


http://gerrit.cloudera.org:8080/#/c/2903/2/src/kudu/codegen/row_projector.cc
File src/kudu/codegen/row_projector.cc:

Line 286:  // TODO: it seems like there's a bug here since we're using the 
pointer value
        :   // of the read_default in the case of STRINGs, which could get 
reused.
> Why would that be a bug? If it gets reused then the pointer is still valid 
I guess you're right - if it got reused for the same type in the same location 
then the code would still work. I was thinking somehow that we'd actually 
inlined the _value_ of the default, which isn't the case of course (and if we 
had, then we wouldnt' care if the pointer changed)


-- 
To view, visit http://gerrit.cloudera.org:8080/2903
To unsubscribe, visit http://gerrit.cloudera.org:8080/settings

Gerrit-MessageType: comment
Gerrit-Change-Id: I583ef1edad379bb13b34362f121263de28c8334c
Gerrit-PatchSet: 2
Gerrit-Project: kudu
Gerrit-Branch: master
Gerrit-Owner: Todd Lipcon <[email protected]>
Gerrit-Reviewer: Anonymous Coward #174
Gerrit-Reviewer: Kudu Jenkins
Gerrit-Reviewer: Todd Lipcon <[email protected]>
Gerrit-HasComments: Yes

Reply via email to