- [Bugfix] add powi to the list of instructions that the NVPTX backend cannot lower.
- [NFC] remove debug code
Aug 23 2017
Aug 22 2017
- [Diff fix] Upload correct arc diff.
- Use LegacyInlinerBase as the starting point for our inliner pass.
- Cache inlining results because polyhedral analysis is expensive.
Aug 21 2017
Test now passes. Probably succeeds due to rL311042
Abandoned because rL311259 contains changes that this patch was supposed to perform.
- [NFC] fix whitespace and debug print (Fix screwed up diff due to rebase).
- [NFC] fix whitespace and debug print
@grosser - review please.
- Update code and add test case.
Aug 20 2017
Thanks a lot for the patch! I have some stylistic comments to make. However, I do not understand the related codebase well enough, so I'll leave that for the others :).
Aug 19 2017
Please add the comment I requested for cc->isLatestScalarKind() ? isl_bool_true : isl_bool_false; and the test cases. LGTM.
It's cool that invariant-load-hoisting-with-failing-scop.ll now succeeds, but that's not the purpose of the test case :)
- [NFC] reorder statements
- [NFC] remove empty line
- Don't call DeleteValue on the ConstantExpr because there can be multiple use sites
- Invalidate Cur by deleting it and setting it to nullptr
- [NFC] remove blank line
@PhilippSchaad - Can you confirm that this now allows you to link on a machine without CUDA? Sorry for letting this slip by, I mixed up the runtime/driver APIs.
Thanks! Other than nits, LGTM.
I would like to discuss the isl::space ParamSpace issue before LGTM'ing this patch.
Aug 18 2017
Does this patch depend on some other patch? If so, please record this information by creating a Parent Revision. (Edit Related Revisions -> Edit Parent Revision). If some other patch depends on this, then make this a parent of that patch. Helps when looking up revisions later on.
Other than comments, LGTM :)
Aug 17 2017
- [NFC] remove metadata from tests
- [Bugfix, Debug] Actually skip declarations and add better debug prints.
- [NFC] run through check-polly, slightly restructure code.
- [NFC] rename function as what Tobias suggested and add comment explaining why
@grosser Dont, added tests.
- [NFC] add testcase
- [NFC] fix style comments, make ScopInliner check that -polly-allow-full-func is enabled
- [NFC] remove 0 address space because that is the default value.
- [NFC] Discuss algorithm with phillip offline, he helped to simplify it further.
- [Linkage] Update linkage code to use the correct linker options as well as the ignore linkage flag. Update test case to match this change
- [ReplaceUsesOfWith] remove double-loop that was not required.
- [Merge] Merged with master, hoping that GPUJIT does not show up from arc diff this time.
- [Re-upload] diff against the newest HEAD.
@philip.pfaffe Another round of review, please?
- [Bugfix] Send bytes, not bits. Also, move the alloca function extraction to the correct place.
Aug 16 2017
Nit: Consider using IRBuilder? Not sure if this is needed, but maybe it avoids the extra parameter. YMMV. Other than that LGTM.
- [NFC] Rewrite based on Phillip & Tobias' comments
Aug 15 2017
- [NFC] remove TODO with actual code
@Meinersbur - Second round of review, please :)
- [NFC] undo all changes to GPUJIT.
@singam-sanjay Review please.
- [NFC] document why we care about size 0 allocs
- [WIP] Update according to Michael's comments.
Aug 13 2017
- [UNDEBUG] remove debug printing in GPUJIT
- [NFC] check polly and fix test case to reduce size by x100
- [Code dump] Dump of all changes that now allow global arrays to be rewritten
Aug 11 2017
- [WIP] version now works on test code, going to run it on COSMO to detect bugs in implementation.
Aug 10 2017
Example of changes induced by this patch on a .ll file
- [WIP] not sure why this doesn't work, code seems reasonable
@grosser: could you change your status to "Requesting changes"? That way, you'll see when I push