Repository navigation
Refactor Binary Identity Simplification Macro - #7195
Conversation
2c1fda2 to
f4e05e8
Compare
|
It looks like there are some build errors on Windows and macOS. |
c8a8d5d to
092686b
Compare
|
@hzongaro ready for another look! |
092686b to
b901254
Compare
|
@a7ehuo, I haven't managed to spend the time needed to review this pull request properly. May I ask you to review? |
|
@jmesyou Does the second commit almost rewrite the core change made in the first commit? It might be hard to review since any comment added to the first commit might not be applicable any more after the second commit. And some of the comments might need to refer to the code in the first commit but the suggestion would apply to the second commit since the structure of the design has been changed in the second commit. Could the two commits be squashed into one to make it easier for review? Thank you! Could the commit comment and the PR description be updated to reflect more accurately on the content since this PR doesn't solve the whole issue but it addresses 2 of the 3 macros mentioned in the issue #7166 |
0664dc2 to
6795c73
Compare
|
@a7ehuo I squashed the commits together. Hopefully that will make the PR and changes easier to read. I've also updated the |
There was a problem hiding this comment.
Thank you very much for making this change to clean up the code and squashing the two commits! I just have some minor comments.
Meanwhile could the description of this PR be updated with the latest commit message? I'd also recommend running some regression tests such as sanity.functional, extended.functional, or sanity.system, sanity.openjdk etc. on different platforms with this change to make sure there is no regression.
d1827d5 to
cbfa9ff
Compare
e9d2b08 to
0152754
Compare
There was a problem hiding this comment.
Thank you for the update!
LGTM. I assume you're running some regression tests (sanity.functional, extended.functional, sanity.openjdk etc). Once you confirm the regression tests complete successfully and all the commits are squashed into one. I'll approve the PR. Thanks again!
0152754 to
060d4ca
Compare
060d4ca to
45ea2ab
Compare
|
Internal build 21052 is passing and commits squashed |
45ea2ab to
217a709
Compare
hzongaro
left a comment
There was a problem hiding this comment.
I think the changes look good. I have a number of cosmetic comments, mostly on the existing code.
hzongaro
left a comment
There was a problem hiding this comment.
Thanks for the updates. All your changes look good - just one further comment. Once addressed, you can go ahead and squash the commits. Thanks!
11dac46 to
0eb4905
Compare
|
Jenkins build all |
|
It looks like the z/OS and AIX compilers are having trouble with part of these changes. |
This commit cleans up the BINARY_IDENTITY_OP and BIANRY_IDENTITY_OR_ZERO macros and replaces them with inline equivalents by implementing a struct to replace template instantions of the functions tryAndSimplifyBinaryOp*. This change limits the number of template instantiations potential call sites have to create as well as abstracts some captured variables inside the struct to minimize the number of arguments passed to actual calls. A closure is emulated by the struct. Through escape analysis, scalar replacement and inlining, the overhead of creating this struct should be minimal on -O3 Related: eclipse-omr#7166 Signed-off-by: James You <james.you@protonmail.com>
0eb4905 to
5030f48
Compare
|
@hzongaro had to rebase and force push to run the Jenkins pipeline, sorry about that 😅 |
|
Jenkins build all |
This commit cleans up the BINARY_IDENTITY_OP and
BIANRY_IDENTITY_OR_ZERO macros and replaces them
with inline equivalents by implementing a struct to replace template
instantions of the functions tryAndSimplifyBinaryOp*.
This change limits the number of template instantiations
potential call sites have to create as well as abstracts
some captured variables inside the struct to minimize
the number of arguments passed to actual calls.
A closure is emulated by the struct.
Through escape analysis, scalar replacement and inlining,
the overhead of creating this struct should be minimal
on -O3
Related: #7166
Signed-off-by: James You james.you@protonmail.com