impala-reviews mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Tim Armstrong (Code Review)" <>
Subject [Impala-ASF-CR] IMPALA-6069: Fix CodegenAnyVal's handling of 'nan'
Date Thu, 07 Dec 2017 20:19:36 GMT
Tim Armstrong has posted comments on this change. ( )

Change subject: IMPALA-6069: Fix CodegenAnyVal's handling of 'nan'

Patch Set 1:


Seems like the right fix.
File be/src/codegen/
PS1, Line 679:       return builder_->CreateFCmpOEQ(GetVal(), other->GetVal(), "eq");
Can you add a comment explaining why this variant is used? It's subtle. I looked at the LLVM
language reference manual to make sure the semantics were correct:
File testdata/workloads/functional-query/queries/QueryTest/joins.test:
PS1, Line 775: # Test that 'nan' != 'nan' when joining.
Can we test a few more NaN variants just to get additional coverage? I think in practice the
code path is the same now, but if our constant replacement got more sophisticated that might

I'm able to get NaNs from expressions on alltypestiny in a few ways. I used this for ideas

  select sqrt(-int_col), float_col / double_col, 0 * 1/double_col, 1/double_col + -1/double_col,
log10(-double_col) from functional.alltypestiny

To view, visit
To unsubscribe, visit

Gerrit-Project: Impala-ASF
Gerrit-Branch: master
Gerrit-MessageType: comment
Gerrit-Change-Id: I1bb8e5074b3c939927dedc46bc9db63ca24486a1
Gerrit-Change-Number: 8790
Gerrit-PatchSet: 1
Gerrit-Owner: Thomas Tauber-Marshall <>
Gerrit-Reviewer: Tim Armstrong <>
Gerrit-Comment-Date: Thu, 07 Dec 2017 20:19:36 +0000
Gerrit-HasComments: Yes

  • Unnamed multipart/alternative (inline, 8-Bit, 0 bytes)
View raw message