Page MenuHomePhabricator

pengfei (Pengfei Wang)
User

Projects

User does not belong to any projects.

User Details

User Since
Dec 12 2018, 5:57 PM (132 w, 1 d)

Recent Activity

Today

pengfei added inline comments to D104766: [X86] Zero some outputs of Keylocker intrinsics in error case.
Thu, Jun 24, 1:32 AM

Yesterday

pengfei accepted D103943: [X86] Add -mgeneral-regs-only support..

LGTM.

Wed, Jun 23, 12:00 AM · Restricted Project

Tue, Jun 22

pengfei accepted D104692: [X86] Add a test to reveal a bug in CMOV conversion..

LGTM.

Tue, Jun 22, 1:53 AM · Restricted Project

Mon, Jun 21

pengfei added inline comments to D104623: [X86] Update MachineLoopInfo in CMOV conversion..
Mon, Jun 21, 5:50 AM · Restricted Project
pengfei added inline comments to D103943: [X86] Add -mgeneral-regs-only support..
Mon, Jun 21, 12:47 AM · Restricted Project

Sun, Jun 20

pengfei accepted D104604: [clang] NFC: add line break at the end of if expressions.

LGTM.

Sun, Jun 20, 7:19 PM · Restricted Project
pengfei committed rG735ad67a4ce8: [clang] NFC: adjust indentation of statements with more than one lines (authored by zhouyizhou).
[clang] NFC: adjust indentation of statements with more than one lines
Sun, Jun 20, 7:17 PM
pengfei closed D104145: [clang] NFC: adjust indentation of statements with more than one lines.
Sun, Jun 20, 7:17 PM · Restricted Project

Fri, Jun 18

pengfei added a comment to D104247: [DAGCombine] reassoc flag shouldn't enable contract.

In code that I've looked at (mostly C compiled with -ffast-math), we always have contract when we have reassoc, so I don't see much practical difference.
Can you explain more how we could benefit from this change - in the example in the description, we would have 4 instructions rather than 3 if we use FMA - is that better?

Yes, you are right, the performance of having FMA should be better. However, we have quite some scenarios that users care about precision more than performance, they want to precise control of when FMA can be generated. So The major motivation of this is to ensure that we respect the IR semantics. For users that care about performance, we still can get them through default global option or emitting respect flag in IR.

Fri, Jun 18, 7:13 PM · Restricted Project

Thu, Jun 17

pengfei added inline comments to D101074: [X86] Canonicalize SGT/UGT compares with constants to use SGE/UGE to reduce the number of EFLAGs reads. (PR48760).
Thu, Jun 17, 7:09 AM · Restricted Project

Tue, Jun 15

pengfei added inline comments to D87981: [X86] AMX programming model..
Tue, Jun 15, 8:33 PM · Restricted Project, Restricted Project
pengfei added inline comments to D101074: [X86] Canonicalize SGT/UGT compares with constants to use SGE/UGE to reduce the number of EFLAGs reads. (PR48760).
Tue, Jun 15, 8:13 PM · Restricted Project

Sat, Jun 12

pengfei added inline comments to D104178: [X86] Add ISD::FREEZE and ISD::AssertAlign to the list of opcodes that don't guarantee upper 32 bits are zero..
Sat, Jun 12, 1:47 AM · Restricted Project
pengfei accepted D104145: [clang] NFC: adjust indentation of statements with more than one lines.

LG

Sat, Jun 12, 1:28 AM · Restricted Project

Fri, Jun 11

pengfei added a comment to D104037: [X86] Check immediate before get it..

Maybe we can use an MIR test?

Fri, Jun 11, 1:59 AM · Restricted Project
pengfei accepted D104037: [X86] Check immediate before get it..

LGTM but let's wait one or two days to see opinions from others.

Fri, Jun 11, 1:53 AM · Restricted Project

Thu, Jun 10

pengfei added inline comments to D104037: [X86] Check immediate before get it..
Thu, Jun 10, 10:35 PM · Restricted Project

Wed, Jun 9

pengfei updated subscribers of D104001: [X86] avoid assert with varargs, soft float, and no-implicit-float.
Wed, Jun 9, 7:47 PM · Restricted Project
pengfei added inline comments to D99675: [llvm][clang] Create new intrinsic llvm.arithmetic.fence to control FP optimization at expression level.
Wed, Jun 9, 7:34 AM · Restricted Project

Tue, Jun 8

pengfei added a comment to D56387: [DAGCombiner] Enable SimplifyDemandedBits vector support for TRUNCATE.

Seems it still causes crash https://bugs.llvm.org/show_bug.cgi?id=50609

Tue, Jun 8, 3:50 AM · Restricted Project

Mon, Jun 7

pengfei added inline comments to D99675: [llvm][clang] Create new intrinsic llvm.arithmetic.fence to control FP optimization at expression level.
Mon, Jun 7, 10:28 PM · Restricted Project

Thu, Jun 3

pengfei added a comment to D80344: [Windows SEH]: HARDWARE EXCEPTION HANDLING (MSVC -EHa) - Part 1.

We found another crush: https://godbolt.org/z/vcbvddEKr

Thu, Jun 3, 5:31 AM · Restricted Project, Restricted Project
pengfei added a reviewer for D99675: [llvm][clang] Create new intrinsic llvm.arithmetic.fence to control FP optimization at expression level: craig.topper.
Thu, Jun 3, 12:18 AM · Restricted Project

Tue, Jun 1

pengfei added reviewers for D99675: [llvm][clang] Create new intrinsic llvm.arithmetic.fence to control FP optimization at expression level: LuoYuanke, LiuChen3.
Tue, Jun 1, 5:38 PM · Restricted Project

Mon, May 31

pengfei added a comment to D103274: [X86] Split FeatureFastVariableShuffle tuning into Lane-Crossing and Per-Lane variants.

@pengfei @craig.topper @RKSimon does anyone feel strongly regarding the avx512 question and the question about not having fast-variable-crosslane-shuffle?

Mon, May 31, 6:29 PM · Restricted Project

Sun, May 30

pengfei added inline comments to D103274: [X86] Split FeatureFastVariableShuffle tuning into Lane-Crossing and Per-Lane variants.
Sun, May 30, 5:11 PM · Restricted Project
pengfei added a comment to D103274: [X86] Split FeatureFastVariableShuffle tuning into Lane-Crossing and Per-Lane variants.

I'm happy with this but I do think we should always enable variable shuffles (both types) on all AVX512 targets as well (and maybe land that first to simplify this patch) - @pengfei @craig.topper what do you think?

Sun, May 30, 2:28 AM · Restricted Project

Thu, May 27

pengfei added inline comments to D103280: [SDAG] try harder to fold casts into vector compare.
Thu, May 27, 8:55 PM · Restricted Project
pengfei added a comment to D103128: [GlobalISel] Don't emit lost debug location remarks when legalizing tail calls.

Sorry for the noise. I reran the bisect. The test in question is still failing, but it turns out to be failing unreliably so git-blame is fairly useless without a reliable test.

Thu, May 27, 6:57 PM · Restricted Project
pengfei accepted D98895: [X86][Draft] Disable long double type for -mno-x87 option.

LGTM. But let's wait one or more days to see if others have more comments.

Thu, May 27, 4:57 AM · Restricted Project

Wed, May 26

pengfei added reviewers for D99675: [llvm][clang] Create new intrinsic llvm.arithmetic.fence to control FP optimization at expression level: kpn, cameron.mcinally, uweigand.
Wed, May 26, 6:36 PM · Restricted Project
pengfei accepted D103145: [X86][AMX] Fix a bug on tile config..

LGTM. Thanks!

Wed, May 26, 6:08 AM · Restricted Project
pengfei added inline comments to D103145: [X86][AMX] Fix a bug on tile config..
Wed, May 26, 4:45 AM · Restricted Project

May 25 2021

pengfei added inline comments to D98895: [X86][Draft] Disable long double type for -mno-x87 option.
May 25 2021, 7:07 AM · Restricted Project
pengfei added inline comments to D98895: [X86][Draft] Disable long double type for -mno-x87 option.
May 25 2021, 12:08 AM · Restricted Project

May 21 2021

pengfei accepted D102588: [UpdateTestChecks] Default --x86_scrub_rip to False.

Updated LCPI regex to include Darwin style LCPI_[0-9]+_[0-9]+ (no
leading dot).

Do we have a test for it?

llvm/test/CodeGen/X86/WidenArith.ll

May 21 2021, 7:18 PM · Restricted Project
pengfei committed rGe6ec7ab5820f: NFC: TreeTransform.h: make the function TransformNestedNameSpecifierLoc more… (authored by zhouyizhou).
NFC: TreeTransform.h: make the function TransformNestedNameSpecifierLoc more…
May 21 2021, 3:28 AM
pengfei closed D102577: NFC: TreeTransform.h: make the function TransformNestedNameSpecifierLoc more beautiful .
May 21 2021, 3:28 AM · Restricted Project
pengfei committed rGe2815398ce0c: Fix warning: comparison of integer expressions of different signedness. NFC (authored by lbenes).
Fix warning: comparison of integer expressions of different signedness. NFC
May 21 2021, 3:24 AM
pengfei closed D102792: Fix warning: comparison of integer expressions of different signedness.
May 21 2021, 3:24 AM · Restricted Project

May 20 2021

pengfei added a comment to D102588: [UpdateTestChecks] Default --x86_scrub_rip to False.

Updated LCPI regex to include Darwin style LCPI_[0-9]+_[0-9]+ (no
leading dot).

May 20 2021, 6:24 PM · Restricted Project

May 19 2021

pengfei added inline comments to D98895: [X86][Draft] Disable long double type for -mno-x87 option.
May 19 2021, 7:13 PM · Restricted Project
pengfei accepted D102792: Fix warning: comparison of integer expressions of different signedness.

LG. Thanks for fixing it.
(do you want me to land this?)

May 19 2021, 5:22 PM · Restricted Project
pengfei committed rG9d09d20448e4: Reapply "[X86] Limit X86InterleavedAccessGroup to handle the same type case… (authored by pengfei).
Reapply "[X86] Limit X86InterleavedAccessGroup to handle the same type case…
May 19 2021, 7:51 AM
pengfei closed D102751: [X86] Limit X86InterleavedAccessGroup to handle the same type case only.
May 19 2021, 7:50 AM · Restricted Project
pengfei reopened D102751: [X86] Limit X86InterleavedAccessGroup to handle the same type case only.

Thanks @RKSimon , I will investigate it.

May 19 2021, 5:37 AM · Restricted Project
pengfei added a reverting change for rGca23a38e3731: [X86] Limit X86InterleavedAccessGroup to handle the same type case only: rG66513e2f20d9: Revert "[X86] Limit X86InterleavedAccessGroup to handle the same type case only".
May 19 2021, 5:36 AM
pengfei added a reverting change for D102751: [X86] Limit X86InterleavedAccessGroup to handle the same type case only: rG66513e2f20d9: Revert "[X86] Limit X86InterleavedAccessGroup to handle the same type case only".
May 19 2021, 5:36 AM · Restricted Project
pengfei committed rG66513e2f20d9: Revert "[X86] Limit X86InterleavedAccessGroup to handle the same type case only" (authored by pengfei).
Revert "[X86] Limit X86InterleavedAccessGroup to handle the same type case only"
May 19 2021, 5:36 AM
pengfei committed rGca23a38e3731: [X86] Limit X86InterleavedAccessGroup to handle the same type case only (authored by pengfei).
[X86] Limit X86InterleavedAccessGroup to handle the same type case only
May 19 2021, 3:39 AM
pengfei closed D102751: [X86] Limit X86InterleavedAccessGroup to handle the same type case only.
May 19 2021, 3:39 AM · Restricted Project
pengfei requested review of D102751: [X86] Limit X86InterleavedAccessGroup to handle the same type case only.
May 19 2021, 1:47 AM · Restricted Project

May 18 2021

pengfei added inline comments to D98895: [X86][Draft] Disable long double type for -mno-x87 option.
May 18 2021, 10:43 PM · Restricted Project
pengfei added inline comments to D102702: Experimental: inline assembly operands .
May 18 2021, 7:53 PM · Restricted Project

May 17 2021

pengfei accepted D102577: NFC: TreeTransform.h: make the function TransformNestedNameSpecifierLoc more beautiful .

LGTM. Thanks for improving it.

May 17 2021, 2:58 AM · Restricted Project
pengfei added a comment to D102577: NFC: TreeTransform.h: make the function TransformNestedNameSpecifierLoc more beautiful .

I think you can add a NFC in the title.

May 17 2021, 1:40 AM · Restricted Project

May 16 2021

pengfei committed rG976a3e5f6141: [SelectionDAG] Make fast and linearize visible by clang -pre-RA-sched (authored by TaoPan).
[SelectionDAG] Make fast and linearize visible by clang -pre-RA-sched
May 16 2021, 8:58 PM
pengfei closed D101601: [SelectionDAG] Make fast and linearize visible by clang -pre-RA-sched.
May 16 2021, 8:58 PM · Restricted Project, Restricted Project

May 10 2021

pengfei accepted D74730: [FPEnv][X86] Implement lowering of llvm.set.rounding.
May 10 2021, 6:03 PM · Restricted Project

May 7 2021

pengfei accepted D101601: [SelectionDAG] Make fast and linearize visible by clang -pre-RA-sched.

I see. LGTM, but let's wait one or more days to see if others object it.

May 7 2021, 7:24 PM · Restricted Project, Restricted Project

May 6 2021

pengfei added a comment to D101718: [X86] Fold load with tail call more aggressively.

I don't find any problem in the tests. So I think it looks good. But I'm not familiar with call lowering, I'd like others to sign off.

May 6 2021, 9:33 PM · Restricted Project
pengfei added inline comments to D101601: [SelectionDAG] Make fast and linearize visible by clang -pre-RA-sched.
May 6 2021, 9:26 PM · Restricted Project, Restricted Project

Apr 29 2021

pengfei added a comment to D101338: [MS] Preserve base register %rbx around cpuid.

Seems reasonable, thanks.

We really ought to solve the general problem one day. It is kind of pig-headed that the backend just picks two arbitrary CSRs, the base and frame pointers, and doesn't bother checking if any instructions clobber those registers.

Apr 29 2021, 7:18 PM · Restricted Project
pengfei committed rGe0c7db7d8ce7: [MS] Preserve base register %rbx around cpuid (authored by pengfei).
[MS] Preserve base register %rbx around cpuid
Apr 29 2021, 7:17 PM
pengfei closed D101338: [MS] Preserve base register %rbx around cpuid.
Apr 29 2021, 7:16 PM · Restricted Project

Apr 28 2021

pengfei committed rGf69adfb87f99: [X86][AMX][NFC] Add more comments and remove unnecessary check found by Clocwork (authored by pengfei).
[X86][AMX][NFC] Add more comments and remove unnecessary check found by Clocwork
Apr 28 2021, 1:35 AM

Apr 27 2021

pengfei accepted D101059: [X86][AMX] Add description for AMX new interface..

LGTM.

Apr 27 2021, 12:38 AM · Restricted Project
pengfei updated subscribers of D101311: Basic block sections for functions with implicit-section-name attribute.
Apr 27 2021, 12:34 AM · Restricted Project

Apr 26 2021

pengfei updated subscribers of D98895: [X86][Draft] Disable long double type for -mno-x87 option.
Apr 26 2021, 11:42 PM · Restricted Project
pengfei updated subscribers of D101338: [MS] Preserve base register %rbx around cpuid.
Apr 26 2021, 11:04 PM · Restricted Project
pengfei retitled D101338: [MS] Preserve base register %rbx around cpuid from [MS] Fix crash when calling __cpuid with /EHsc and -mavx2 to [MS] Preserve base register %rbx around cpuid.
Apr 26 2021, 10:58 PM · Restricted Project
pengfei updated the diff for D101338: [MS] Preserve base register %rbx around cpuid.

Yes, it is a pretty implementation. Thanks Craig.

Apr 26 2021, 10:56 PM · Restricted Project
pengfei updated the diff for D101338: [MS] Preserve base register %rbx around cpuid.

Update for test.

Apr 26 2021, 8:32 PM · Restricted Project
pengfei requested review of D101338: [MS] Preserve base register %rbx around cpuid.
Apr 26 2021, 7:51 PM · Restricted Project
pengfei committed rG016092d786f2: Reapply "[X86][AMX] Try to hoist AMX shapes' def" (authored by pengfei).
Reapply "[X86][AMX] Try to hoist AMX shapes' def"
Apr 26 2021, 7:28 PM
pengfei closed D101067: [X86][AMX] Try to hoist AMX shapes' def.
Apr 26 2021, 7:28 PM · Restricted Project
pengfei added a comment to D101067: [X86][AMX] Try to hoist AMX shapes' def.

Ping~

Apr 26 2021, 6:45 PM · Restricted Project

Apr 24 2021

pengfei updated the diff for D101067: [X86][AMX] Try to hoist AMX shapes' def.

Reapply the minor refactor.
Fix MSan build failures.

Apr 24 2021, 8:55 PM · Restricted Project
pengfei reopened D101067: [X86][AMX] Try to hoist AMX shapes' def.

Looks like this patch broke the MSan buildbots: https://lab.llvm.org/buildbot/#/builders/5/builds/6967/steps/9/logs/stdio

Reproing MSan is a little tricky as it can require a multistage compilation, so you might find the instructions at https://github.com/google/sanitizers/wiki/SanitizerBotReproduceBuild helpful.

Apr 24 2021, 8:50 PM · Restricted Project

Apr 23 2021

pengfei committed rG151e244fe687: [X86][AMX][NFC] Make comparison operators to be complete (authored by pengfei).
[X86][AMX][NFC] Make comparison operators to be complete
Apr 23 2021, 2:39 AM

Apr 22 2021

pengfei added a comment to D101067: [X86][AMX] Try to hoist AMX shapes' def.

I happened to commit this patch with another one. Since Xiang had +1 for it, @LuoYuanke , do you think if I need to revert it or not?

Apr 22 2021, 9:20 PM · Restricted Project
pengfei committed rG90118563ad0f: [X86][AMX] Try to hoist AMX shapes' def (authored by pengfei).
[X86][AMX] Try to hoist AMX shapes' def
Apr 22 2021, 9:17 PM
pengfei committed rG53673fd1bf6f: [X86][AMX][NFC] Avoid assert for the same immidiate value (authored by pengfei).
[X86][AMX][NFC] Avoid assert for the same immidiate value
Apr 22 2021, 9:17 PM
pengfei closed D101124: [X86][AMX][NFC] Avoid assert for the same immidiate value.
Apr 22 2021, 9:17 PM · Restricted Project
pengfei closed D101067: [X86][AMX] Try to hoist AMX shapes' def.
Apr 22 2021, 9:17 PM · Restricted Project
pengfei committed rGe8bce8399631: [X86] Enable compilation of user interrupt handlers. (authored by pengfei).
[X86] Enable compilation of user interrupt handlers.
Apr 22 2021, 8:44 PM
pengfei closed D99708: [X86] Enable compilation of user interrupt handlers..
Apr 22 2021, 8:44 PM · Restricted Project, Restricted Project
pengfei updated the diff for D101067: [X86][AMX] Try to hoist AMX shapes' def.

Minor refactor.

Apr 22 2021, 8:28 PM · Restricted Project
pengfei added inline comments to D101067: [X86][AMX] Try to hoist AMX shapes' def.
Apr 22 2021, 8:00 PM · Restricted Project
pengfei requested review of D101124: [X86][AMX][NFC] Avoid assert for the same immidiate value.
Apr 22 2021, 7:09 PM · Restricted Project
pengfei updated the diff for D101067: [X86][AMX] Try to hoist AMX shapes' def.

Always insert before AMX instruction.

Apr 22 2021, 8:19 AM · Restricted Project
pengfei added a comment to D74730: [FPEnv][X86] Implement lowering of llvm.set.rounding.

This patch LGTM, but I'd like other sign off.

Apr 22 2021, 8:02 AM · Restricted Project
pengfei updated the diff for D101067: [X86][AMX] Try to hoist AMX shapes' def.

Reset ShapeBBs[MBB] once we hoist any shapes in MBB.

Apr 22 2021, 7:58 AM · Restricted Project
pengfei requested review of D101067: [X86][AMX] Try to hoist AMX shapes' def.
Apr 22 2021, 7:21 AM · Restricted Project
pengfei committed rGaafb6d81cf25: [X86][AMX][NFC] Remove assert for comparison between different BBs. (authored by pengfei).
[X86][AMX][NFC] Remove assert for comparison between different BBs.
Apr 22 2021, 5:42 AM
pengfei closed D101039: [X86][AMX][NFC] Remove assert for comparison between different BBs..
Apr 22 2021, 5:42 AM · Restricted Project
pengfei added a comment to D101039: [X86][AMX][NFC] Remove assert for comparison between different BBs..

Is there any test case for it?

Apr 22 2021, 4:29 AM · Restricted Project
pengfei requested review of D101039: [X86][AMX][NFC] Remove assert for comparison between different BBs..
Apr 22 2021, 2:37 AM · Restricted Project
pengfei added inline comments to D74730: [FPEnv][X86] Implement lowering of llvm.set.rounding.
Apr 22 2021, 12:58 AM · Restricted Project

Apr 20 2021

pengfei accepted D100026: [X86] Support AMX fast register allocation.

The implementation is good to me in general. Let's wait a few days to see opinions from community and other reviewers.

Apr 20 2021, 7:48 PM · Restricted Project