impala-reviews mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Tim Armstrong (Code Review)" <ger...@cloudera.org>
Subject [Impala-ASF-CR] IMPALA-3884: Support TYPE TIMESTAMP for HashTableCtx::CodegenAssignNullValue()
Date Sat, 22 Oct 2016 03:19:45 GMT
Tim Armstrong has posted comments on this change.

Change subject: IMPALA-3884: Support TYPE_TIMESTAMP for HashTableCtx::CodegenAssignNullValue()
......................................................................


Patch Set 1:

(6 comments)

http://gerrit.cloudera.org:8080/#/c/4794/1/be/src/codegen/llvm-codegen.cc
File be/src/codegen/llvm-codegen.cc:

Line 586:   DCHECK_EQ(num_bytes & (num_bytes - 1), 0);
Could you factor this into BitUtil::IsPowerOfTwo() or something along those lines. That would
be generally useful - e.g. I know some cases in the buffer pool code where I'd use it.


Line 591:     return ConstantInt::get(context(), APInt(128, vals));
CodegenAnyVal::SetVal(int128_t) should use this approach. There is a TODO in there.


http://gerrit.cloudera.org:8080/#/c/4794/1/be/src/codegen/llvm-codegen.h
File be/src/codegen/llvm-codegen.h:

Line 387:   /// than 64-bit), an 128-bit value is formed by setting its upper and lower 64-bit
set
This behaviour of copying the value into the upper and lower bits is pretty confusing - I
would expect it to return an int128_t with all the upper bits 0. E.g. GetIntConstant(byte_size,
1) should always return 1 regardless of the byte_size. Maybe we should have a separate GetInt128Constant(int
bytes_size, int64_t upper, int64_t lower) method that does this so the behaviour is more "obvious".


http://gerrit.cloudera.org:8080/#/c/4794/1/be/src/exec/hash-table.cc
File be/src/exec/hash-table.cc:

Line 597:         // fall through to generate 128-bit 'fnv_seed'
May as well just call codegen->GetIntConstant() here, or codegen->GetInt128Constant()
here. It only saves a line of code.


PS1, Line 608: fnv_seed_float
Thank you! This bugged me when reading the code before. Optional, but you could do the search-and-replace
too in old-hash-table.cc


http://gerrit.cloudera.org:8080/#/c/4794/1/testdata/workloads/functional-query/queries/QueryTest/joins.test
File testdata/workloads/functional-query/queries/QueryTest/joins.test:

Line 723: ---- QUERY
I have a test in https://gerrit.cloudera.org/#/c/4655/6/tests/query_test/test_aggregation.py
that disables the check if codegen is enabled for timestamps. Depending on which patch lands
first we should make sure that's reenabled.


-- 
To view, visit http://gerrit.cloudera.org:8080/4794
To unsubscribe, visit http://gerrit.cloudera.org:8080/settings

Gerrit-MessageType: comment
Gerrit-Change-Id: I0211d38cbef46331e0006fa5ed0680e6e0867bc8
Gerrit-PatchSet: 1
Gerrit-Project: Impala-ASF
Gerrit-Branch: master
Gerrit-Owner: Michael Ho <kwho@cloudera.com>
Gerrit-Reviewer: Tim Armstrong <tarmstrong@cloudera.com>
Gerrit-HasComments: Yes

Mime
View raw message