For the case in this issue, Start and L are not enough, we also need Stride, to calculate correct MinStart, requiring getUnsignedRangeAtScope() take to SCEV* parameters, I think it's a little confusing.

Jul 23 2019, 11:08 AM · Restricted Project

zzheng updated the diff for D62990: [SCEV]When safe, compute MinStart as unsigned_min(Start - Stride) + Stride in computeMaxBECountForLT.

Added new test

Jul 23 2019, 11:00 AM · Restricted Project

Jun 20 2019

zzheng added a comment to D62989: [Unroll] Do NOT unroll a loop with small runtime upperbound.

ping...

Jun 20 2019, 1:07 PM · Restricted Project

zzheng retitled D62990: [SCEV]When safe, compute MinStart as unsigned_min(Start - Stride) + Stride in computeMaxBECountForLT from [SCEV]When safe, use Stride as MinStart in computeMaxBECountForLT to [SCEV]When safe, compute MinStart as unsigned_min(Start - Stride) + Stride in computeMaxBECountForLT.

Jun 20 2019, 12:16 PM · Restricted Project

zzheng updated the diff for D62990: [SCEV]When safe, compute MinStart as unsigned_min(Start - Stride) + Stride in computeMaxBECountForLT.

Jun 20 2019, 12:12 PM · Restricted Project

Jun 13 2019

zzheng added a comment to D62989: [Unroll] Do NOT unroll a loop with small runtime upperbound.

ping?

Jun 13 2019, 11:17 AM · Restricted Project

zzheng added a comment to D62990: [SCEV]When safe, compute MinStart as unsigned_min(Start - Stride) + Stride in computeMaxBECountForLT.

ping?

Jun 13 2019, 11:17 AM · Restricted Project

Jun 11 2019

zzheng updated the diff for D62990: [SCEV]When safe, compute MinStart as unsigned_min(Start - Stride) + Stride in computeMaxBECountForLT.

Jun 11 2019, 3:59 PM · Restricted Project

zzheng updated the summary of D62989: [Unroll] Do NOT unroll a loop with small runtime upperbound.

Jun 11 2019, 1:28 PM · Restricted Project

zzheng updated the diff for D62989: [Unroll] Do NOT unroll a loop with small runtime upperbound.

Renamed UnrollByMaxCount to FullUnrollMaxTripCount, note this is not to be confused with member variable UP.FullUnrollMaxCount.
Changed debug output to accurately reflect partial or runtime unrolling.

Jun 11 2019, 1:28 PM · Restricted Project

Jun 10 2019

zzheng added a reviewer for D62989: [Unroll] Do NOT unroll a loop with small runtime upperbound: dmgreen.

Jun 10 2019, 5:29 PM · Restricted Project

zzheng added inline comments to D62989: [Unroll] Do NOT unroll a loop with small runtime upperbound.

Jun 10 2019, 5:27 PM · Restricted Project

Jun 7 2019

zzheng added a comment to D62989: [Unroll] Do NOT unroll a loop with small runtime upperbound.

Wouldn't I want to (fully) unroll loops especially those with low trip count to reduce overhead/enable further optimizations?

Do you have performance/code size numbers?

Jun 7 2019, 4:59 PM · Restricted Project

zzheng added a comment to D62989: [Unroll] Do NOT unroll a loop with small runtime upperbound.

In D62989#1534692, @Meinersbur wrote:

What problem is this solving?

Jun 7 2019, 4:01 PM · Restricted Project

zzheng updated the diff for D62989: [Unroll] Do NOT unroll a loop with small runtime upperbound.

define dso_local void @hoge_3(i8 %arg) {
entry:
  %x = load i32, i32* @global, align 4
  %y = load i8*, i8** @global.1, align 4
  %0 = icmp ult i32 %x, 17
  br i1 %0, label %loop, label %exit

Jun 7 2019, 4:01 PM · Restricted Project

Jun 6 2019

zzheng updated the diff for D62990: [SCEV]When safe, compute MinStart as unsigned_min(Start - Stride) + Stride in computeMaxBECountForLT.

re-upload with full context

Jun 6 2019, 9:20 PM · Restricted Project

zzheng updated the summary of D62989: [Unroll] Do NOT unroll a loop with small runtime upperbound.

Jun 6 2019, 9:14 PM · Restricted Project

zzheng updated the diff for D62989: [Unroll] Do NOT unroll a loop with small runtime upperbound.

Re-upload patch with full context.

Jun 6 2019, 9:14 PM · Restricted Project

zzheng abandoned D27911: [LoopUnroll] Keep UP.Count when considering complete unroll.

Jun 6 2019, 6:24 PM · Restricted Project

zzheng updated the summary of D62990: [SCEV]When safe, compute MinStart as unsigned_min(Start - Stride) + Stride in computeMaxBECountForLT.

Jun 6 2019, 6:19 PM · Restricted Project

zzheng created D62990: [SCEV]When safe, compute MinStart as unsigned_min(Start - Stride) + Stride in computeMaxBECountForLT.

Jun 6 2019, 6:16 PM · Restricted Project

zzheng created D62989: [Unroll] Do NOT unroll a loop with small runtime upperbound.

Jun 6 2019, 5:52 PM · Restricted Project

Jun 3 2019

zzheng added a parent revision for D62833: [DAGCombine][Thumb] Use single CP entry for addressing GV with large offset: D60294: [DAGCombiner] [CodeGenPrepare] Split large offsets from base addresses.

Jun 3 2019, 5:21 PM · Restricted Project

zzheng added a child revision for D60294: [DAGCombiner] [CodeGenPrepare] Split large offsets from base addresses: D62833: [DAGCombine][Thumb] Use single CP entry for addressing GV with large offset.

Jun 3 2019, 5:21 PM · Restricted Project

zzheng created D62833: [DAGCombine][Thumb] Use single CP entry for addressing GV with large offset.

Jun 3 2019, 5:08 PM · Restricted Project

Oct 11 2018

zzheng added inline comments to D53005: Implement machine unroller utility class.

Oct 11 2018, 1:54 PM

Oct 10 2018

zzheng added inline comments to D33935: Allow rematerialization of ARM Thumb MOVi8 instruction in some contexts.

Oct 10 2018, 4:09 PM

Sep 25 2018

zzheng committed rL343053: Revert "Revert "[ConstHoist] Do not rebase single (or few) dependent constant"".

Revert "Revert "[ConstHoist] Do not rebase single (or few) dependent constant""

Sep 25 2018, 6:01 PM

zzheng committed rL342994: [ConstHoist] Do not rebase single (or few) dependent constant.

[ConstHoist] Do not rebase single (or few) dependent constant

Sep 25 2018, 10:49 AM

zzheng closed D52243: [ConstHoist] Do not rebase single (or few) dependent constant.

Sep 25 2018, 10:49 AM

Sep 24 2018

zzheng added inline comments to D52243: [ConstHoist] Do not rebase single (or few) dependent constant.

Sep 24 2018, 3:28 PM

zzheng updated the diff for D52243: [ConstHoist] Do not rebase single (or few) dependent constant.

Moved collecting dependent rebased and checking size of ToBeRebased prior to emitting bitcast of base constants
Added bb labels to check in the test case.

Sep 24 2018, 3:27 PM

zzheng committed rL342898: [Thumb1] Any imm8 should have cost of 1.

[Thumb1] Any imm8 should have cost of 1

Sep 24 2018, 9:17 AM

zzheng closed D52257: [Thumb1] Any imm of i8 type on Thumb1 should have cost of 1.

Sep 24 2018, 9:17 AM

Sep 21 2018

zzheng added a comment to D52243: [ConstHoist] Do not rebase single (or few) dependent constant.

In D52243#1241540, @dmgreen wrote:

This does seem, from the tests I ran, to reduce codesize on average. Especially on Thumb1.

Sep 21 2018, 5:26 PM

zzheng added inline comments to D52243: [ConstHoist] Do not rebase single (or few) dependent constant.

Sep 21 2018, 5:25 PM

zzheng updated the diff for D52243: [ConstHoist] Do not rebase single (or few) dependent constant.

Cleaned up test case.
Added opt tests.
Opt tests was put in the same test file to avoid duplication and maintenance burden.

Sep 21 2018, 5:17 PM

zzheng added a comment to D52257: [Thumb1] Any imm of i8 type on Thumb1 should have cost of 1.

In D52257#1241502, @SjoerdMeijer wrote:

Ah, but looking a bit closer now, I am not sure this is the right thing to do. This changes makes any 8 bit value cheap, including negative numbers. And I am not sure if this is the right thing to do, since the Thumb1 immediates are positive numbers. It looks this a workaround for store-merging interacting badly with constant hoisting.

Sep 21 2018, 3:58 PM

zzheng updated the diff for D52257: [Thumb1] Any imm of i8 type on Thumb1 should have cost of 1.

Cleaned up test case.

Sep 21 2018, 3:54 PM

Sep 19 2018

zzheng added a comment to D52243: [ConstHoist] Do not rebase single (or few) dependent constant.

ping...

Sep 19 2018, 4:33 PM

zzheng added a comment to D52257: [Thumb1] Any imm of i8 type on Thumb1 should have cost of 1.

ping...

Sep 19 2018, 4:32 PM

Sep 18 2018

zzheng created D52257: [Thumb1] Any imm of i8 type on Thumb1 should have cost of 1.

Sep 18 2018, 7:45 PM

zzheng added reviewers for D52257: [Thumb1] Any imm of i8 type on Thumb1 should have cost of 1: • zinob, apazos.

Sep 18 2018, 7:45 PM

zzheng added reviewers for D52243: [ConstHoist] Do not rebase single (or few) dependent constant: javed.absar, wmi, SjoerdMeijer, mehdi_amini.

Sep 18 2018, 7:32 PM

zzheng created D52243: [ConstHoist] Do not rebase single (or few) dependent constant.

Sep 18 2018, 11:51 AM

Sep 4 2018

zzheng committed rL341417: Revert "Revert r341269: [Constant Hoisting] Hoisting Constant GEP Expressions".

Revert "Revert r341269: [Constant Hoisting] Hoisting Constant GEP Expressions"

Sep 4 2018, 3:18 PM

zzheng closed D51654: Reland r341269: [Constant Hoisting] Hoisting Constant GEP Expressions.

Sep 4 2018, 3:18 PM

zzheng updated the summary of D51654: Reland r341269: [Constant Hoisting] Hoisting Constant GEP Expressions.

Sep 4 2018, 2:48 PM

zzheng updated the diff for D51654: Reland r341269: [Constant Hoisting] Hoisting Constant GEP Expressions.

Use std::stable_sort() as Eli suggested.

Sep 4 2018, 2:48 PM

zzheng created D51654: Reland r341269: [Constant Hoisting] Hoisting Constant GEP Expressions.

Sep 4 2018, 1:20 PM

Aug 31 2018

zzheng committed rL341269: [Constant Hoisting] Hoisting Constant GEP Expressions.

[Constant Hoisting] Hoisting Constant GEP Expressions

Aug 31 2018, 5:07 PM

zzheng closed D51396: [Constant Hoisting] Hoisting Constant GEP Expressions.

Aug 31 2018, 5:07 PM

zzheng updated the diff for D51396: [Constant Hoisting] Hoisting Constant GEP Expressions.

Addressed Eli's comments.

Aug 31 2018, 4:52 PM

zzheng added inline comments to D51396: [Constant Hoisting] Hoisting Constant GEP Expressions.

Aug 31 2018, 4:08 PM

zzheng updated the diff for D51396: [Constant Hoisting] Hoisting Constant GEP Expressions.

Updated version reflects discussion with Eli.

Aug 31 2018, 3:59 PM

Aug 29 2018

zzheng added inline comments to D51396: [Constant Hoisting] Hoisting Constant GEP Expressions.

Aug 29 2018, 4:13 PM

zzheng updated the diff for D51396: [Constant Hoisting] Hoisting Constant GEP Expressions.

Unfortunately bitcast + gep + bitcast will only change the constant GEP expr from multi-dimension to single dimension, and it's still a constant pool entry after lowering.

Aug 29 2018, 1:01 PM

Aug 28 2018

zzheng created D51396: [Constant Hoisting] Hoisting Constant GEP Expressions.

Aug 28 2018, 5:44 PM

zzheng committed rL340882: [QTOOL-37352] Consider isLegalAddressingImm in Constant Hoisting.

[QTOOL-37352] Consider isLegalAddressingImm in Constant Hoisting

Aug 28 2018, 4:03 PM

zzheng closed D50931: Consider isLegalAddressingImm in Constant Hoisting.

Aug 28 2018, 4:03 PM

zzheng updated the diff for D50931: Consider isLegalAddressingImm in Constant Hoisting.

Aug 28 2018, 3:46 PM

Aug 27 2018

zzheng updated the diff for D50931: Consider isLegalAddressingImm in Constant Hoisting.

Moved 'break;' under proper if statement.

Aug 27 2018, 4:57 PM

zzheng updated the diff for D50931: Consider isLegalAddressingImm in Constant Hoisting.

Previous version mistakenly checks MinValItr->ConstInst rather than CC->ConstInst against StoreInst User's pointer operand.

Aug 27 2018, 4:08 PM

Aug 24 2018

zzheng added inline comments to D50931: Consider isLegalAddressingImm in Constant Hoisting.

Aug 24 2018, 5:16 PM

zzheng updated the diff for D50931: Consider isLegalAddressingImm in Constant Hoisting.

Updated as Eli suggested.

Aug 24 2018, 5:14 PM

Aug 23 2018

zzheng added a comment to D50931: Consider isLegalAddressingImm in Constant Hoisting.

ping...

Aug 23 2018, 10:24 AM

Aug 21 2018

zzheng updated the diff for D50931: Consider isLegalAddressingImm in Constant Hoisting.

Updated test case as Eli suggested.

Aug 21 2018, 3:56 PM

Aug 20 2018

zzheng added a comment to D50931: Consider isLegalAddressingImm in Constant Hoisting.

ping?

Aug 20 2018, 11:24 AM

Aug 17 2018

zzheng edited reviewers for D50931: Consider isLegalAddressingImm in Constant Hoisting, added: srhines, ributzka; removed: • zinob, yinma.

Aug 17 2018, 4:11 PM

zzheng created D50931: Consider isLegalAddressingImm in Constant Hoisting.

Aug 17 2018, 4:05 PM

Apr 9 2018

zzheng closed D44257: [MemorySSAUpdater] Mark non-trivial Phi for not to optimize.

committed as r329621, spelling errors fixed in r329230

Apr 9 2018, 2:03 PM

zzheng committed rL329621: [MemorySSAUpdater] Mark Phi users of a node being moved as non-optimize.

[MemorySSAUpdater] Mark Phi users of a node being moved as non-optimize

Apr 9 2018, 1:58 PM

Apr 4 2018

zzheng added a comment to D44257: [MemorySSAUpdater] Mark non-trivial Phi for not to optimize.

I'll run a bunch of internal test to validate the patch. I'll commit it after testing if no one objects.

Apr 4 2018, 7:29 PM

zzheng added a comment to D44257: [MemorySSAUpdater] Mark non-trivial Phi for not to optimize.

Just one nit (inline) and a general question: is this new set meant to persist across multiple updates, should the erase in fixupDefs be taking care of everything added in a given query, or should we be clearing it at some point?

(If the answer is "the middle one," would it be easy to assert that the def is empty somewhere?)

Apr 4 2018, 7:28 PM

zzheng updated the diff for D44257: [MemorySSAUpdater] Mark non-trivial Phi for not to optimize.

Addressed comments by Eli and George.

Apr 4 2018, 7:25 PM

zzheng committed rL329230: [MemorySSA] Fix spelling errors in MemorySSA.cpp. NFC.

[MemorySSA] Fix spelling errors in MemorySSA.cpp. NFC

Apr 4 2018, 2:11 PM