♟ mkazantsev

Addressed comments, added a unit test to make sure that the computation works as intended.

Re-enable terminator folding in LoopSimplifyCFG: underlying bugs fixed

[LoopSimplifyCFG] Form LCSSA when a parent loop becomes a sibling

[LoopSimplifyCFG] Fix order of deletion of complex dead subloops

[NFC] Factor out some local vars

Re-uploaded with context.

[NFC] Remove some code duplication

[NFC] Remove obsolete enum RangeCheckKind

[NFC] Decrease if nest

[NFC] Move some functions to LoopUtils

Updated test to preserve current behavior (it creates *really* big SCEVs in test_05).

Rebase & reclaim to fix PR40292.

We need it to fix PR40292. Reclaiming.

[BasicBlockUtils] Generalize DeleteDeadBlock to deal with multiple dead blocks

[IPT] Drop cache less eagerly in GVN and LoopSafetyInfo

Drop SE cache early because loop parent can change in LoopSimplifyCFG

[NFC] Add failing test on LCSSA form preservation of LoopSimplifyCFG

Temporarily disable term folding in LoopSimplifyCFG, add tests

[LoopSimplifyCFG] Delete dead blocks in RPO

[NFC] Use utility function for guards detection

[NFC] Reuse variables instead of re-calling getParent

Revert rL350048 and rL350050

Rebased, ready for review since underlying patches are now merged.

[LoopSimplifyCFG] Delete dead exiting edges

Return "[LoopSimplifyCFG] Delete dead in-loop blocks"

Introduce llvm.experimental.widenable_condition intrinsic

[LoopSimplifyCFG] Do not deal with loops with irreducible CFG inside

Patch has been reverted, likely because it triggered a bug fixed by D55357 (underlying bug, not directly connected to this patch). Waiting for confirmation, if so, I will recommit it after the fix is merged.

Underlying patch merged, opening for review.

[LoopSimplifyCFG] Delete dead in-loop blocks

Added asserts that we do not delete current loop and its header while dealing with dead blocks.

[NFC] Verify memoryssa in test for PR39783

There is already a mode with memorySSA being constructed and verified in constant-fold-branch.ll:

; RUN: opt -S -enable-loop-simplifycfg-term-folding=true -loop-simplifycfg -enable-mssa-loop-dependency=true -verify-memoryssa -debug-only=loop-simplifycfg -verify-loop-info -verify-dom-info -verify-loop-lcssa 2>&1 < %s | FileCheck %s

The entire algorithm looks non-deterministic and depends on traversal order of uses. I suggest making it deterministic (I think my comment inline should help).

Rebased, added removal of blocks from MSSA.

[LoopSimplifyCFG] Update MemorySSA in terminator folding. PR39783

Rebased on top of simplified tests.

[NFC] Simplify and reduce tests for PR39783

Uh, actually I can. I will update the tests and rebase on top.

In D51207#1312345, @apilipenko wrote:

One of the alternatives naming schemes which was discussed is to call this intrinsic should_deoptimize. In this case the meaning of the returned value is inverted, so we need to or it with the condition which we want to widen.

Current definition in LangRef gives it much wider use cases than this. The point is, we have no obligation to *only* use this intrinsic in and condition with deopt in one of branches. For example, this usage is legit:

It will need a MemorySSA update when we delete blocks.

[NFC] Add two XFAIL tests from PR39783

mkazantsev (Max Kazantsev)
User

Projects

User Details

Recent Activity
View All

Fri, Jan 18

Thu, Jan 17

Wed, Jan 16

Tue, Jan 15

Mon, Jan 14

Sun, Jan 13

Thu, Jan 10

Tue, Jan 8

Fri, Dec 28

Thu, Dec 27

Wed, Dec 26

Tue, Dec 25

Mon, Dec 24

Sun, Dec 23

Dec 7 2018

Dec 6 2018

Dec 5 2018

Dec 4 2018

Nov 30 2018

Nov 29 2018

mkazantsev (Max Kazantsev)User