Page MenuHomePhabricator

rampitec (Stanislav Mekhanoshin)
User

Projects

User does not belong to any projects.

User Details

User Since
Apr 4 2014, 4:14 AM (272 w, 3 d)

Recent Activity

Yesterday

rampitec accepted D63749: AMDGPU: Make amdgcn.s.get.waveid.in.workgroup inaccessiblememonly.

LGTM

Mon, Jun 24, 7:32 PM
rampitec added inline comments to D63209: [AMDGPU] gfx1010 wave32 clang support.
Mon, Jun 24, 4:10 PM · Restricted Project
rampitec added inline comments to D63209: [AMDGPU] gfx1010 wave32 clang support.
Mon, Jun 24, 3:21 PM · Restricted Project
rampitec updated subscribers of D63209: [AMDGPU] gfx1010 wave32 clang support.
Mon, Jun 24, 2:47 PM · Restricted Project
rampitec added inline comments to D63731: [AMDGPU] Prevent VGPR copies from moving across the EXEC mask definitions.
Mon, Jun 24, 1:18 PM
rampitec added a comment to D63731: [AMDGPU] Prevent VGPR copies from moving across the EXEC mask definitions.

Please do not forget to add llvm-commits.

Mon, Jun 24, 11:25 AM
rampitec added a reviewer for D63731: [AMDGPU] Prevent VGPR copies from moving across the EXEC mask definitions: arsenm.
Mon, Jun 24, 11:25 AM
rampitec added inline comments to D63709: [AMDGPU] Add peephole to optimize MOV.
Mon, Jun 24, 9:40 AM · Restricted Project
rampitec accepted D63724: [AMDGPU] Allow any value in unused src0 field in v_nop.

LGTM

Mon, Jun 24, 9:36 AM · Restricted Project
rampitec requested changes to D63709: [AMDGPU] Add peephole to optimize MOV.

Patch does not take control flow divergence into account.

Mon, Jun 24, 8:42 AM · Restricted Project
rampitec accepted D63716: AMDGPU/GFX10: implement ds_ordered_count changes.

LGTM

Mon, Jun 24, 8:33 AM · Restricted Project

Fri, Jun 21

rampitec accepted D63682: AMDGPU: Fold frame index into MUBUF.

LGTM

Fri, Jun 21, 7:14 PM
rampitec accepted D63673: AMDGPU: Cleanup checking when spills need emergency slots.

LGTM

Fri, Jun 21, 7:10 PM
rampitec committed rGbdf7f81b89d1: [AMDGPU] hazard recognizer for fp atomic to s_denorm_mode (authored by rampitec).
[AMDGPU] hazard recognizer for fp atomic to s_denorm_mode
Fri, Jun 21, 9:28 AM
rampitec committed rL364074: [AMDGPU] hazard recognizer for fp atomic to s_denorm_mode.
[AMDGPU] hazard recognizer for fp atomic to s_denorm_mode
Fri, Jun 21, 9:27 AM
rampitec closed D63619: [AMDGPU] hazard recognizer for fp atomic to s_denorm_mode.
Fri, Jun 21, 9:27 AM · Restricted Project
rampitec accepted D63649: AMDGPU: Fix target builtins for gfx10.

LGTM

Fri, Jun 21, 9:20 AM
rampitec added a comment to D63619: [AMDGPU] hazard recognizer for fp atomic to s_denorm_mode.

LGTM assuming the DS atomics really aren't impacted

Fri, Jun 21, 9:01 AM · Restricted Project

Thu, Jun 20

rampitec added inline comments to D63494: [AMDGPU] Fix for branch offset hardware workaround.
Thu, Jun 20, 3:22 PM · Restricted Project
rampitec added a comment to D63494: [AMDGPU] Fix for branch offset hardware workaround.

LGTM, except one indent.

Thu, Jun 20, 3:07 PM · Restricted Project
rampitec updated the diff for D63619: [AMDGPU] hazard recognizer for fp atomic to s_denorm_mode.
Thu, Jun 20, 3:03 PM · Restricted Project
rampitec added inline comments to D63494: [AMDGPU] Fix for branch offset hardware workaround.
Thu, Jun 20, 1:36 PM · Restricted Project
rampitec added inline comments to D63619: [AMDGPU] hazard recognizer for fp atomic to s_denorm_mode.
Thu, Jun 20, 12:54 PM · Restricted Project
rampitec updated the diff for D63619: [AMDGPU] hazard recognizer for fp atomic to s_denorm_mode.
Thu, Jun 20, 12:54 PM · Restricted Project
rampitec created D63619: [AMDGPU] hazard recognizer for fp atomic to s_denorm_mode.
Thu, Jun 20, 12:30 PM · Restricted Project
rampitec accepted D63611: AMDGPU: Fix ignoring DisableFramePointerElim in leaf calls.

LGTM

Thu, Jun 20, 9:54 AM
rampitec accepted D63605: AMDGPU: Always use s33 for global scratch wave offset.

LGTM

Thu, Jun 20, 9:27 AM
rampitec committed rGe917b3b4b813: [AMDGPU] gfx10 tests. NFC. (authored by rampitec).
[AMDGPU] gfx10 tests. NFC.
Thu, Jun 20, 9:27 AM
rampitec committed rL363946: [AMDGPU] gfx10 tests. NFC..
[AMDGPU] gfx10 tests. NFC.
Thu, Jun 20, 9:27 AM
rampitec committed rL363934: [AMDGPU] gfx1010 core wave32 changes.
[AMDGPU] gfx1010 core wave32 changes
Thu, Jun 20, 8:39 AM
rampitec committed rG0846c125f98b: [AMDGPU] gfx1010 core wave32 changes (authored by rampitec).
[AMDGPU] gfx1010 core wave32 changes
Thu, Jun 20, 8:09 AM
rampitec closed D63204: [AMDGPU] gfx1010 core wave32 changes.
Thu, Jun 20, 8:08 AM · Restricted Project
rampitec accepted D63560: AMDGPU: Treat undef as an inline immediate.

LGTM

Thu, Jun 20, 8:03 AM
rampitec accepted D63578: AMDGPU: Add DS GWS sema builtins.

LGTM

Thu, Jun 20, 8:03 AM
rampitec committed rL363931: Virtualize TargetInstrInfo::getRegClass().
Virtualize TargetInstrInfo::getRegClass()
Thu, Jun 20, 7:58 AM
rampitec committed rG082ffa9960cc: Virtualize TargetInstrInfo::getRegClass() (authored by rampitec).
Virtualize TargetInstrInfo::getRegClass()
Thu, Jun 20, 7:57 AM
rampitec closed D63205: Virtualize TargetInstrInfo::getRegClass().
Thu, Jun 20, 7:56 AM · Restricted Project
rampitec added inline comments to D63494: [AMDGPU] Fix for branch offset hardware workaround.
Thu, Jun 20, 7:29 AM · Restricted Project

Wed, Jun 19

rampitec accepted D63576: AMDGPU: Add intrinsics for DS GWS semaphore instructions.

LGTM

Wed, Jun 19, 3:35 PM
rampitec accepted D63558: AMDGPU: Undo sub x, c canonicalization for v2i16.

LGTM

Wed, Jun 19, 2:27 PM
rampitec added inline comments to D63560: AMDGPU: Treat undef as an inline immediate.
Wed, Jun 19, 2:24 PM
rampitec added inline comments to D63558: AMDGPU: Undo sub x, c canonicalization for v2i16.
Wed, Jun 19, 2:18 PM
rampitec accepted D63566: AMDGPU: Fix folding immediate into readfirstlane through reg_sequence .

LGTM

Wed, Jun 19, 1:36 PM
rampitec added inline comments to D63566: AMDGPU: Fix folding immediate into readfirstlane through reg_sequence .
Wed, Jun 19, 11:28 AM
rampitec added inline comments to D63494: [AMDGPU] Fix for branch offset hardware workaround.
Wed, Jun 19, 11:21 AM · Restricted Project

Tue, Jun 18

rampitec added inline comments to D63494: [AMDGPU] Fix for branch offset hardware workaround.
Tue, Jun 18, 4:40 PM · Restricted Project
rampitec added a comment to D63205: Virtualize TargetInstrInfo::getRegClass().

Maybe I'm missing something, but the override implemented in D63204 looks like a subset of the default TII::getRegClass that falls through to TRI::getRegClass, just like the default does. Is this change actually necessary?

It will fallback to TRI::getRegClass(), but to the base class of it through the pointer to TRI in the InstrEmitter, so call goes to the default implementation. Either TII::getRegClass() needs to be virtual or TRI::getRegClass().
I have preferred to override in TII because it is called less frequently than from TRI.

Tue, Jun 18, 1:06 PM · Restricted Project
rampitec committed rGbb1c8b6f5cae: [AMDGPU] gfx10 wave32 patterns (authored by rampitec).
[AMDGPU] gfx10 wave32 patterns
Tue, Jun 18, 12:59 PM
rampitec updated the diff for D63204: [AMDGPU] gfx1010 core wave32 changes.

Rebased.

Tue, Jun 18, 12:59 PM · Restricted Project
rampitec committed rL363729: [AMDGPU] gfx10 wave32 patterns.
[AMDGPU] gfx10 wave32 patterns
Tue, Jun 18, 12:57 PM
rampitec closed D63511: [AMDGPU] gfx10 wave32 patterns.
Tue, Jun 18, 12:57 PM · Restricted Project
rampitec created D63511: [AMDGPU] gfx10 wave32 patterns.
Tue, Jun 18, 12:41 PM · Restricted Project
rampitec added a comment to D63494: [AMDGPU] Fix for branch offset hardware workaround.

Can we have a disasm test? I want to see that we do not mess with decoding.

Something other than the -disassemble checks?

Tue, Jun 18, 12:14 PM · Restricted Project
rampitec updated the diff for D63204: [AMDGPU] gfx1010 core wave32 changes.

Rebased.

Tue, Jun 18, 12:10 PM · Restricted Project
rampitec committed rGab4f2ea79318: [AMDGPU] gfx1010 disassembler changes for wave32 (authored by rampitec).
[AMDGPU] gfx1010 disassembler changes for wave32
Tue, Jun 18, 12:09 PM
rampitec committed rL363721: [AMDGPU] gfx1010 disassembler changes for wave32.
[AMDGPU] gfx1010 disassembler changes for wave32
Tue, Jun 18, 12:09 PM
rampitec closed D63506: [AMDGPU] gfx1010 disassembler changes for wave32.
Tue, Jun 18, 12:09 PM · Restricted Project
rampitec created D63506: [AMDGPU] gfx1010 disassembler changes for wave32.
Tue, Jun 18, 11:29 AM · Restricted Project
rampitec abandoned D62937: [AMDGPU] Mark s_endpgm convergent.
Tue, Jun 18, 11:03 AM
rampitec added a comment to D63489: [InstSimplify] LCSSA PHIs should not be simplified away.

This causes generation of incorrect code in AMDGPU backend.

This sounds like some other check is missing elsewhere?
What happens if you feed it such an ir as-if after this transform, but manually written?
("that will result in broken asm/crashes" is hopefully not the answer)

That being said, why is LCSSAPass not sufficient?
It's already supposed to undo transforms like this.

It will result in syntactically correct asm and no crashes. In runtime we'll get incorrect result though :)
Adding LCSSA pass again later on is difficult in the sense of the pass dependencies.
So, it's better to fix the explicit bug in SimplifyPHI....

Aha, so it's not -instsimplify pass itself, but how it's used during transition into backend.

  1. You certainly don't want to make this blacklist unconditional, it should still run when the -instsimplify pass itself is run. (+instsimplify test)
  2. How does this affect other targets (backends)? Does this need some TLI hook?
Tue, Jun 18, 10:04 AM · Restricted Project
rampitec added a comment to D63494: [AMDGPU] Fix for branch offset hardware workaround.

Can we have a disasm test? I want to see that we do not mess with decoding.

Tue, Jun 18, 10:03 AM · Restricted Project

Mon, Jun 17

rampitec accepted D63465: AMDGPU: Fix iterator crash in AMDGPUPromoteAlloca.

LGTM

Mon, Jun 17, 6:58 PM
rampitec accepted D63464: AMDGPU: Insert mem_viol check loop around GWS pre-GFX9.

Tentative LGTM for the code itself. Please check with Brian for feasibility.

Mon, Jun 17, 5:10 PM
rampitec added a comment to D63464: AMDGPU: Insert mem_viol check loop around GWS pre-GFX9.

Two questions:

  1. Should that only be enabled if xnack enabled? Was it even ever on before gfx9, I think no?
  2. Should we support it on ASICs older than gfx9? This may be more a question to Brian.
Mon, Jun 17, 4:57 PM
rampitec updated the diff for D63204: [AMDGPU] gfx1010 core wave32 changes.

Rebased and removed unused S_ADDC_U64_PSEUDO/S_SUBC_U64_PSEUDO.

Mon, Jun 17, 3:45 PM · Restricted Project
rampitec committed rG121956108f27: [AMDGPU] Use custom inserter for gfx10 VOP2b (authored by rampitec).
[AMDGPU] Use custom inserter for gfx10 VOP2b
Mon, Jun 17, 3:36 PM
rampitec committed rL363625: [AMDGPU] Use custom inserter for gfx10 VOP2b.
[AMDGPU] Use custom inserter for gfx10 VOP2b
Mon, Jun 17, 3:35 PM
rampitec committed rGca42687d62a8: [AMDGPU] gfx1010 subvector test. NFC. (authored by rampitec).
[AMDGPU] gfx1010 subvector test. NFC.
Mon, Jun 17, 2:54 PM
rampitec committed rL363623: [AMDGPU] gfx1010 subvector test. NFC..
[AMDGPU] gfx1010 subvector test. NFC.
Mon, Jun 17, 2:52 PM
rampitec accepted D63225: AMDGPU: Fold readlane from copy of SGPR or imm.

LGTM if precheckin will succeed for this and for D63456.

Mon, Jun 17, 1:44 PM
rampitec committed rG313827828730: [AMDGPU] Propagate function attributes thru bitcasts (authored by rampitec).
[AMDGPU] Propagate function attributes thru bitcasts
Mon, Jun 17, 1:42 PM
rampitec accepted D63456: AMDGPU: Remove unnecessary check for virtual register.

LGTM, but please run precheckin before submit. Same for D63225 if you sure dest is always virtual.

Mon, Jun 17, 1:41 PM
rampitec committed rL363614: [AMDGPU] Propagate function attributes thru bitcasts.
[AMDGPU] Propagate function attributes thru bitcasts
Mon, Jun 17, 1:39 PM
rampitec closed D63455: [AMDGPU] Propagate function attributes thru bitcasts.
Mon, Jun 17, 1:39 PM · Restricted Project
rampitec created D63455: [AMDGPU] Propagate function attributes thru bitcasts.
Mon, Jun 17, 1:31 PM · Restricted Project
rampitec added inline comments to D63225: AMDGPU: Fold readlane from copy of SGPR or imm.
Mon, Jun 17, 1:00 PM
rampitec added a comment to D63205: Virtualize TargetInstrInfo::getRegClass().

Maybe I'm missing something, but the override implemented in D63204 looks like a subset of the default TII::getRegClass that falls through to TRI::getRegClass, just like the default does. Is this change actually necessary?

Mon, Jun 17, 12:31 PM · Restricted Project
rampitec added inline comments to D63225: AMDGPU: Fold readlane from copy of SGPR or imm.
Mon, Jun 17, 11:23 AM
rampitec committed rGa9191c8492ab: [AMDGPU] gfx1010 wavefrontsize intrinsic folding (authored by rampitec).
[AMDGPU] gfx1010 wavefrontsize intrinsic folding
Mon, Jun 17, 10:57 AM
rampitec committed rL363588: [AMDGPU] gfx1010 wavefrontsize intrinsic folding.
[AMDGPU] gfx1010 wavefrontsize intrinsic folding
Mon, Jun 17, 10:57 AM
rampitec closed D63206: [AMDGPU] gfx1010 wavefrontsize intrinsic folding.
Mon, Jun 17, 10:57 AM · Restricted Project
rampitec removed a child revision for D63204: [AMDGPU] gfx1010 core wave32 changes: D63206: [AMDGPU] gfx1010 wavefrontsize intrinsic folding.
Mon, Jun 17, 10:56 AM · Restricted Project
rampitec removed a parent revision for D63206: [AMDGPU] gfx1010 wavefrontsize intrinsic folding: D63204: [AMDGPU] gfx1010 core wave32 changes.
Mon, Jun 17, 10:56 AM · Restricted Project
rampitec committed rGad04e7ad4266: [AMDGPU] Pass to propagate ABI attributes from kernels to the functions (authored by rampitec).
[AMDGPU] Pass to propagate ABI attributes from kernels to the functions
Mon, Jun 17, 10:47 AM
rampitec committed rL363586: [AMDGPU] Pass to propagate ABI attributes from kernels to the functions.
[AMDGPU] Pass to propagate ABI attributes from kernels to the functions
Mon, Jun 17, 10:44 AM
rampitec closed D63208: [AMDGPU] Pass to propagate ABI attributes from kernels to the functions.
Mon, Jun 17, 10:44 AM · Restricted Project
rampitec removed a parent revision for D63208: [AMDGPU] Pass to propagate ABI attributes from kernels to the functions: D63204: [AMDGPU] gfx1010 core wave32 changes.
Mon, Jun 17, 10:44 AM · Restricted Project
rampitec removed a child revision for D63204: [AMDGPU] gfx1010 core wave32 changes: D63208: [AMDGPU] Pass to propagate ABI attributes from kernels to the functions.
Mon, Jun 17, 10:44 AM · Restricted Project
rampitec updated the diff for D63204: [AMDGPU] gfx1010 core wave32 changes.

Rebased. Reenabled HSA metadata test disabled with previous commit due to dependency.

Mon, Jun 17, 10:02 AM · Restricted Project
rampitec committed rG5d00c3060e11: [AMDGPU] gfx1010 wave32 metadata (authored by rampitec).
[AMDGPU] gfx1010 wave32 metadata
Mon, Jun 17, 9:46 AM
rampitec committed rL363577: [AMDGPU] gfx1010 wave32 metadata.
[AMDGPU] gfx1010 wave32 metadata
Mon, Jun 17, 9:45 AM
rampitec closed D63207: [AMDGPU] gfx1010 wave32 metadata.
Mon, Jun 17, 9:45 AM · Restricted Project
rampitec removed a parent revision for D63207: [AMDGPU] gfx1010 wave32 metadata: D63204: [AMDGPU] gfx1010 core wave32 changes.
Mon, Jun 17, 9:45 AM · Restricted Project
rampitec removed a child revision for D63204: [AMDGPU] gfx1010 core wave32 changes: D63207: [AMDGPU] gfx1010 wave32 metadata.
Mon, Jun 17, 9:45 AM · Restricted Project
rampitec updated the diff for D63207: [AMDGPU] gfx1010 wave32 metadata.

Rebased to master and disabled one run line in the hsa-metadata-kernel-code-props-v3.ll to decouple the change from parent chain. That should complete support for wave64 at least until D63205 is submitted.

Mon, Jun 17, 9:45 AM · Restricted Project
rampitec accepted D63421: AMDGPU: Make getreg intrinsic inaccessiblememonly.

LGTM

Mon, Jun 17, 9:39 AM
rampitec added inline comments to D63427: AMDGPU/GFX10: Don't generate s_code_end padding in the asm-printer.
Mon, Jun 17, 9:05 AM · Restricted Project
rampitec updated the diff for D63206: [AMDGPU] gfx1010 wavefrontsize intrinsic folding.

Rebased.

Mon, Jun 17, 9:04 AM · Restricted Project
rampitec accepted D63427: AMDGPU/GFX10: Don't generate s_code_end padding in the asm-printer.

LGTM

Mon, Jun 17, 8:55 AM · Restricted Project
rampitec updated the diff for D63204: [AMDGPU] gfx1010 core wave32 changes.

Rebased.

Mon, Jun 17, 8:49 AM · Restricted Project