This is an archive of the discontinued LLVM Phabricator instance.

[CodeGen] Fix invalid shift in mul expansion
ClosedPublic

Authored by chfast on Sep 12 2016, 3:36 PM.

Diff Detail

Repository
rL LLVM

Event Timeline

chfast updated this revision to Diff 71064.Sep 12 2016, 3:36 PM
chfast retitled this revision from to [CodeGen] Fix invalid shift in mul expansion.
chfast updated this object.
chfast added reviewers: hfinkel, majnemer.
chfast set the repository for this revision to rL LLVM.
chfast added a subscriber: llvm-commits.
RKSimon added inline comments.
test/CodeGen/X86/mul-bigint.ll
1 ↗(On Diff #71064)

Call this file mul-i512.ll to match the existing filenames?

Test on 32-bit targets as well if you can and use utils/update_llc_test_checks.py to generate the output

chfast added inline comments.Sep 13 2016, 4:26 AM
test/CodeGen/X86/mul-bigint.ll
1 ↗(On Diff #71064)

To test on both 32-bit and 64-bit targets should I do something like:

; RUN llc < %s -mtriple=x86_64-unknown-linux-gnu
; RUN llc < %s -mtriple=x86-unknown-linux-gnu

?

chfast updated this revision to Diff 71147.Sep 13 2016, 5:26 AM

Update tests

RKSimon added inline comments.Sep 13 2016, 6:08 AM
test/CodeGen/X86/mul-i512.ll
5 ↗(On Diff #71147)

Adding nounwind should cleanup the codegen a little, same for mul-i1024.ll

chfast updated this revision to Diff 71224.Sep 13 2016, 12:54 PM

Add nounwind attribute to make tests smaller

RKSimon accepted this revision.Sep 13 2016, 1:54 PM
RKSimon added a reviewer: RKSimon.

LGTM

This revision is now accepted and ready to land.Sep 13 2016, 1:54 PM
This revision was automatically updated to reflect the committed changes.
chfast marked an inline comment as done.
llvm/trunk/lib/CodeGen/SelectionDAG/LegalizeIntegerTypes.cpp