Page MenuHomePhabricator
Feed Advanced Search

Mon, Nov 29

igor.kirillov added inline comments to D110235: [LoopVectorize] Support reductions that store intermediary result.
Mon, Nov 29, 11:09 AM · Restricted Project
igor.kirillov updated the diff for D110235: [LoopVectorize] Support reductions that store intermediary result.

Lots of updates related to the recent comments

Mon, Nov 29, 11:03 AM · Restricted Project

Fri, Nov 26

igor.kirillov committed rG08d45e6f4da0: [AArch64][SVEIntrinsicOpts] Fix: predicated SVE mul/fmul are not commutative (authored by igor.kirillov).
[AArch64][SVEIntrinsicOpts] Fix: predicated SVE mul/fmul are not commutative
Fri, Nov 26, 4:42 AM
igor.kirillov closed D114577: [AArch64][SVEIntrinsicOpts] Fix: predicated SVE mul/fmul are not commutative.
Fri, Nov 26, 4:41 AM · Restricted Project

Thu, Nov 25

igor.kirillov updated the diff for D114577: [AArch64][SVEIntrinsicOpts] Fix: predicated SVE mul/fmul are not commutative.

Update comment

Thu, Nov 25, 11:42 AM · Restricted Project
igor.kirillov added reviewers for D114577: [AArch64][SVEIntrinsicOpts] Fix: predicated SVE mul/fmul are not commutative: bsmith, peterwaller-arm, paulwalker-arm.
Thu, Nov 25, 8:34 AM · Restricted Project
igor.kirillov requested review of D114577: [AArch64][SVEIntrinsicOpts] Fix: predicated SVE mul/fmul are not commutative.
Thu, Nov 25, 3:42 AM · Restricted Project

Tue, Nov 23

igor.kirillov updated the diff for D110235: [LoopVectorize] Support reductions that store intermediary result.

Fix typo

Tue, Nov 23, 2:48 AM · Restricted Project

Mon, Nov 22

igor.kirillov updated the diff for D110235: [LoopVectorize] Support reductions that store intermediary result.

Little extension in a new test

Mon, Nov 22, 6:49 AM · Restricted Project
igor.kirillov updated the diff for D110235: [LoopVectorize] Support reductions that store intermediary result.

Add scalable reduction test

Mon, Nov 22, 6:03 AM · Restricted Project

Thu, Nov 18

igor.kirillov added inline comments to D110235: [LoopVectorize] Support reductions that store intermediary result.
Thu, Nov 18, 7:22 AM · Restricted Project
igor.kirillov updated the diff for D110235: [LoopVectorize] Support reductions that store intermediary result.

Refactoring a bit

Thu, Nov 18, 6:58 AM · Restricted Project
igor.kirillov added inline comments to D110235: [LoopVectorize] Support reductions that store intermediary result.
Thu, Nov 18, 3:30 AM · Restricted Project
igor.kirillov updated the diff for D110235: [LoopVectorize] Support reductions that store intermediary result.

Add test

Thu, Nov 18, 3:30 AM · Restricted Project
igor.kirillov added inline comments to D110235: [LoopVectorize] Support reductions that store intermediary result.
Thu, Nov 18, 3:06 AM · Restricted Project
igor.kirillov updated the diff for D110235: [LoopVectorize] Support reductions that store intermediary result.

Update tests

Thu, Nov 18, 1:21 AM · Restricted Project

Wed, Nov 10

igor.kirillov committed rG4860f6cb25bd: [OpenMP] Fix: opposite attributes could be set by -fno-inline (authored by igor.kirillov).
[OpenMP] Fix: opposite attributes could be set by -fno-inline
Wed, Nov 10, 8:56 AM
igor.kirillov closed D112645: [OpenMP] Fix: opposite attributes could be set by -fno-inline.
Wed, Nov 10, 8:56 AM · Restricted Project

Tue, Nov 9

igor.kirillov added inline comments to D110235: [LoopVectorize] Support reductions that store intermediary result.
Tue, Nov 9, 7:24 AM · Restricted Project
igor.kirillov updated the diff for D110235: [LoopVectorize] Support reductions that store intermediary result.

Update intermediate store check for ordered fadd vectorization

Tue, Nov 9, 7:22 AM · Restricted Project

Sun, Nov 7

igor.kirillov updated the summary of D110235: [LoopVectorize] Support reductions that store intermediary result.
Sun, Nov 7, 5:02 AM · Restricted Project
igor.kirillov added a comment to D110235: [LoopVectorize] Support reductions that store intermediary result.

Hi @igor.kirillov, is it also possible to get this working for ordered reductions, i.e.

float sum = 0;
for(i=0..N) {
  sum += src[i];
  dst[42] = sum;
}

when building with -O3? I think it might mean updating checkOrderedReductions to look through the store. If it looks too difficult to do as part of this patch we can always follow-up with a patch later.

Sun, Nov 7, 5:01 AM · Restricted Project
igor.kirillov updated the diff for D110235: [LoopVectorize] Support reductions that store intermediary result.

Update commit message and comments
Move storeToSameAddress to ScalarEvolution
Add fadd fast test
Do not apply patch for ordered fadd reductions

Sun, Nov 7, 4:53 AM · Restricted Project

Nov 1 2021

igor.kirillov updated the diff for D112645: [OpenMP] Fix: opposite attributes could be set by -fno-inline.

Remove my old test for the change

Nov 1 2021, 2:42 PM · Restricted Project
igor.kirillov updated the diff for D112645: [OpenMP] Fix: opposite attributes could be set by -fno-inline.

Update test, remove redundant code

Nov 1 2021, 2:37 PM · Restricted Project
igor.kirillov added inline comments to D112645: [OpenMP] Fix: opposite attributes could be set by -fno-inline.
Nov 1 2021, 11:54 AM · Restricted Project
igor.kirillov updated subscribers of D112645: [OpenMP] Fix: opposite attributes could be set by -fno-inline.
Nov 1 2021, 3:50 AM · Restricted Project

Oct 27 2021

igor.kirillov added inline comments to D112645: [OpenMP] Fix: opposite attributes could be set by -fno-inline.
Oct 27 2021, 11:55 AM · Restricted Project
igor.kirillov requested review of D112645: [OpenMP] Fix: opposite attributes could be set by -fno-inline.
Oct 27 2021, 11:46 AM · Restricted Project

Oct 13 2021

igor.kirillov added a comment to D110235: [LoopVectorize] Support reductions that store intermediary result.

It seems like the main problem is that we potentially bail out too early at the moment when checking for reductions due to the store, but once we generate runtime checks, sinking the store may become legal (see inline comment about loads to the same address)? If that's the case, ideally we'd just sink any such loads/stores before detecting reductions once we know they can be sunk due to runtime checks, but unfortunately I do not think that's possible with the current structure/ordering.

Oct 13 2021, 2:15 PM · Restricted Project

Oct 10 2021

igor.kirillov added inline comments to D110235: [LoopVectorize] Support reductions that store intermediary result.
Oct 10 2021, 6:42 AM · Restricted Project
igor.kirillov updated the diff for D110235: [LoopVectorize] Support reductions that store intermediary result.

Update comment

Oct 10 2021, 6:25 AM · Restricted Project
igor.kirillov updated the diff for D110235: [LoopVectorize] Support reductions that store intermediary result.

Updating D110235

Oct 10 2021, 6:19 AM · Restricted Project

Sep 27 2021

igor.kirillov added a reviewer for D110235: [LoopVectorize] Support reductions that store intermediary result: fhahn.
Sep 27 2021, 9:13 AM · Restricted Project

Sep 22 2021

igor.kirillov added inline comments to D110235: [LoopVectorize] Support reductions that store intermediary result.
Sep 22 2021, 6:30 AM · Restricted Project
igor.kirillov requested review of D110235: [LoopVectorize] Support reductions that store intermediary result.
Sep 22 2021, 5:51 AM · Restricted Project