Page MenuHomePhabricator
Feed Advanced Search

Thu, Jul 29

mbrkusanin committed rG971f4173f82d: [AMDGPU][GlobalISel] Insert an and with exec before s_cbranch_vccnz if necessary (authored by mbrkusanin).
[AMDGPU][GlobalISel] Insert an and with exec before s_cbranch_vccnz if necessary
Thu, Jul 29, 2:23 AM
mbrkusanin closed D105709: [AMDGPU][GlobalISel] Insert an and with exec before s_cbranch_vccnz if necessary.
Thu, Jul 29, 2:23 AM · Restricted Project

Wed, Jul 28

mbrkusanin updated the diff for D105709: [AMDGPU][GlobalISel] Insert an and with exec before s_cbranch_vccnz if necessary.
  • Remove amdgcn_ballot case
  • Use TRI.getBoolRC()
Wed, Jul 28, 9:13 AM · Restricted Project
mbrkusanin updated the diff for D105709: [AMDGPU][GlobalISel] Insert an and with exec before s_cbranch_vccnz if necessary.
  • Covered case with amdgcn_ballot.
Wed, Jul 28, 8:45 AM · Restricted Project
mbrkusanin updated the diff for D105709: [AMDGPU][GlobalISel] Insert an and with exec before s_cbranch_vccnz if necessary.
  • Removed amdgcn_icmp and amdgcn_fcmp
Wed, Jul 28, 8:33 AM · Restricted Project

Fri, Jul 23

mbrkusanin added inline comments to D105709: [AMDGPU][GlobalISel] Insert an and with exec before s_cbranch_vccnz if necessary.
Fri, Jul 23, 8:13 AM · Restricted Project
mbrkusanin updated the diff for D105709: [AMDGPU][GlobalISel] Insert an and with exec before s_cbranch_vccnz if necessary.

Alternatively we could always insert and with exec and try to remove it in SIOptimizeExecMaskingPreRA (something similar to optimizeVcndVcmpPair).

Fri, Jul 23, 8:13 AM · Restricted Project

Fri, Jul 9

mbrkusanin requested review of D105709: [AMDGPU][GlobalISel] Insert an and with exec before s_cbranch_vccnz if necessary.
Fri, Jul 9, 9:22 AM · Restricted Project

Jun 4 2021

mbrkusanin committed rG35ef4c940bea: [AMDGPU][GlobalISel] Legalize G_ABS (authored by mbrkusanin).
[AMDGPU][GlobalISel] Legalize G_ABS
Jun 4 2021, 5:50 AM
mbrkusanin closed D102391: [AMDGPU][GlobalISel] Legalize G_ABS.
Jun 4 2021, 5:50 AM · Restricted Project

Jun 3 2021

mbrkusanin updated the diff for D102391: [AMDGPU][GlobalISel] Legalize G_ABS.
  • Suggested renaming
  • Removed extra parentheses
Jun 3 2021, 9:06 AM · Restricted Project
mbrkusanin updated the diff for D102391: [AMDGPU][GlobalISel] Legalize G_ABS.
  • Add lowerAbsToMaxSub as a new LegalizerHelper function to be used by AMDGPURegisterBankInfo
  • Move default code for lowering G_ABS to lowerAbs to make it more clear there are two ways of legalizing.
Jun 3 2021, 7:57 AM · Restricted Project

May 28 2021

mbrkusanin requested review of D102391: [AMDGPU][GlobalISel] Legalize G_ABS.
May 28 2021, 8:52 AM · Restricted Project
mbrkusanin updated the diff for D102391: [AMDGPU][GlobalISel] Legalize G_ABS.
  • Added legalizeABS() as a custom lowering to be used by RegBankSelect.
May 28 2021, 8:52 AM · Restricted Project

May 26 2021

mbrkusanin added inline comments to D98515: [AMDGPU][GlobalISel] Stop foldInsertEltToCmpSelect from changing reg banks.
May 26 2021, 3:01 AM · Restricted Project
mbrkusanin committed rG9601849984a7: [AMDGPU][GlobalISel] Stop foldInsertEltToCmpSelect from changing reg banks (authored by mbrkusanin).
[AMDGPU][GlobalISel] Stop foldInsertEltToCmpSelect from changing reg banks
May 26 2021, 3:00 AM
mbrkusanin added a reverting change for rG18c544470289: [AMDGPU][GlobalISel] Stop foldInsertEltToCmpSelect from changing reg banks: rG7386ad4e9e26: Revert "[AMDGPU][GlobalISel] Stop foldInsertEltToCmpSelect from changing reg….
May 26 2021, 3:00 AM
mbrkusanin committed rG7386ad4e9e26: Revert "[AMDGPU][GlobalISel] Stop foldInsertEltToCmpSelect from changing reg… (authored by mbrkusanin).
Revert "[AMDGPU][GlobalISel] Stop foldInsertEltToCmpSelect from changing reg…
May 26 2021, 3:00 AM

May 25 2021

mbrkusanin closed D98515: [AMDGPU][GlobalISel] Stop foldInsertEltToCmpSelect from changing reg banks.

18c5444702893fd63b0a99ec7133dd714284f9d2

May 25 2021, 10:38 AM · Restricted Project
mbrkusanin committed rG18c544470289: [AMDGPU][GlobalISel] Stop foldInsertEltToCmpSelect from changing reg banks (authored by mbrkusanin).
[AMDGPU][GlobalISel] Stop foldInsertEltToCmpSelect from changing reg banks
May 25 2021, 10:36 AM

May 21 2021

mbrkusanin updated the diff for D102391: [AMDGPU][GlobalISel] Legalize G_ABS.
  • Merged G_ABS with G_SMIN, G_SMAX, G_UMIN, G_UMAX in LegalizerInfo
May 21 2021, 6:57 AM · Restricted Project

May 20 2021

mbrkusanin added inline comments to D102391: [AMDGPU][GlobalISel] Legalize G_ABS.
May 20 2021, 2:59 AM · Restricted Project
mbrkusanin updated the diff for D102391: [AMDGPU][GlobalISel] Legalize G_ABS.
  • Widen S16 to S32 for GFX6 + GFX7
May 20 2021, 2:36 AM · Restricted Project

May 18 2021

mbrkusanin added inline comments to D98515: [AMDGPU][GlobalISel] Stop foldInsertEltToCmpSelect from changing reg banks.
May 18 2021, 6:49 AM · Restricted Project
mbrkusanin updated the diff for D98515: [AMDGPU][GlobalISel] Stop foldInsertEltToCmpSelect from changing reg banks.
May 18 2021, 6:48 AM · Restricted Project
mbrkusanin updated the diff for D102391: [AMDGPU][GlobalISel] Legalize G_ABS.
  • Addressed comments
May 18 2021, 6:45 AM · Restricted Project

May 14 2021

mbrkusanin updated the diff for D98515: [AMDGPU][GlobalISel] Stop foldInsertEltToCmpSelect from changing reg banks.

The function would look something like this.

May 14 2021, 8:56 AM · Restricted Project

May 13 2021

mbrkusanin requested review of D102391: [AMDGPU][GlobalISel] Legalize G_ABS.
May 13 2021, 2:38 AM · Restricted Project

Apr 8 2021

mbrkusanin abandoned D98491: [AMDGPU] Split GCN subtarget features for unaligned access.
Apr 8 2021, 2:50 AM · Restricted Project

Apr 1 2021

mbrkusanin added a comment to D98515: [AMDGPU][GlobalISel] Stop foldInsertEltToCmpSelect from changing reg banks.

%1:vgpr(s32) = COPY $sgpr0

I wonder if we should ban this in the verifier. It's not wrong, but it sure feels like bad form to allow cross bank copies involving physical registers

Apr 1 2021, 5:30 AM · Restricted Project

Mar 31 2021

mbrkusanin added a comment to D98515: [AMDGPU][GlobalISel] Stop foldInsertEltToCmpSelect from changing reg banks.

Specifically in the new test:
%4:sgpr(s32) = G_AMDGPU_S_BUFFER_LOAD %0(<4 x s32>), %3(s32), 0 :: (dereferenceable invariant load 4)
was changed into
%4:vgpr(s32) = G_AMDGPU_S_BUFFER_LOAD %0(<4 x s32>), %3(s32), 0 :: (dereferenceable invariant load 4)
after foldInsertEltToCmpSelect

Mar 31 2021, 3:45 AM · Restricted Project
mbrkusanin updated the diff for D98515: [AMDGPU][GlobalISel] Stop foldInsertEltToCmpSelect from changing reg banks.

The issue is in G_AMDGPU_S_BUFFER_LOAD where first we select sgpr for %4. However when selecting a bank for G_INSERT_VECTOR_ELT then foldInsertEltToCmpSelect will change it to vgpr and instruction-select will fail later on.

Mar 31 2021, 3:42 AM · Restricted Project

Mar 12 2021

mbrkusanin requested review of D98515: [AMDGPU][GlobalISel] Stop foldInsertEltToCmpSelect from changing reg banks.
Mar 12 2021, 8:39 AM · Restricted Project
mbrkusanin requested review of D98491: [AMDGPU] Split GCN subtarget features for unaligned access.
Mar 12 2021, 4:08 AM · Restricted Project

Mar 5 2021

mbrkusanin requested review of D98040: [AMDGPU][GlobalISel] Improve constant offset lookup for llvm.amdgcn.s.buffer.
Mar 5 2021, 6:09 AM · Restricted Project

Jan 28 2021

mbrkusanin committed rG3c979ae9ecdd: [AMDGPU][GlobalISel] Remove redundant cmp when copying constant to vcc (authored by mbrkusanin).
[AMDGPU][GlobalISel] Remove redundant cmp when copying constant to vcc
Jan 28 2021, 2:21 AM
mbrkusanin committed rG4b422708bac7: [AMDGPU][GlobalISel] Handle G_PTR_ADD when looking for constant offset (authored by mbrkusanin).
[AMDGPU][GlobalISel] Handle G_PTR_ADD when looking for constant offset
Jan 28 2021, 2:21 AM
mbrkusanin closed D95540: [AMDGPU][GlobalISel] Remove redundant cmp when copying constant to vcc.
Jan 28 2021, 2:21 AM · Restricted Project
mbrkusanin closed D95242: [AMDGPU][GlobalISel] Handle G_PTR_ADD when looking for constant offset.
Jan 28 2021, 2:21 AM · Restricted Project

Jan 27 2021

mbrkusanin updated the diff for D95242: [AMDGPU][GlobalISel] Handle G_PTR_ADD when looking for constant offset.
  • Rebase and rename test
Jan 27 2021, 9:25 AM · Restricted Project
mbrkusanin requested review of D95540: [AMDGPU][GlobalISel] Remove redundant cmp when copying constant to vcc.
Jan 27 2021, 9:25 AM · Restricted Project

Jan 26 2021

mbrkusanin committed rG608ac62540ad: [AMDGPU] Fix use of HasModifiers in VopProfile (authored by mbrkusanin).
[AMDGPU] Fix use of HasModifiers in VopProfile
Jan 26 2021, 6:23 AM
mbrkusanin closed D94795: [AMDGPU] Fix use of HasModifiers in VopProfile.
Jan 26 2021, 6:23 AM · Restricted Project

Jan 22 2021

mbrkusanin updated the summary of D95242: [AMDGPU][GlobalISel] Handle G_PTR_ADD when looking for constant offset.
Jan 22 2021, 8:33 AM · Restricted Project
mbrkusanin requested review of D95242: [AMDGPU][GlobalISel] Handle G_PTR_ADD when looking for constant offset.
Jan 22 2021, 8:32 AM · Restricted Project

Jan 20 2021

mbrkusanin committed rGa6a72dfdf2e1: [AMDGPU][GlobalISel] Avoid selecting S_PACK with constants (authored by mbrkusanin).
[AMDGPU][GlobalISel] Avoid selecting S_PACK with constants
Jan 20 2021, 2:56 AM
mbrkusanin closed D92219: [AMDGPU][GlobalISel] Avoid selecting S_PACK with constants.
Jan 20 2021, 2:56 AM · Restricted Project
mbrkusanin added a comment to D94795: [AMDGPU] Fix use of HasModifiers in VopProfile.

Do we really need HasModifiers if breaking this down to src modifiers and omod?

Jan 20 2021, 2:28 AM · Restricted Project
mbrkusanin updated the diff for D94795: [AMDGPU] Fix use of HasModifiers in VopProfile.
  • Undid few changes to fix disassemble tests.
Jan 20 2021, 2:26 AM · Restricted Project

Jan 15 2021

mbrkusanin added a comment to D94795: [AMDGPU] Fix use of HasModifiers in VopProfile.

"HasModifiers should not be true if at least one modifier is used."

Do you mean "*should* be true"?

Jan 15 2021, 9:22 AM · Restricted Project
mbrkusanin updated the summary of D94795: [AMDGPU] Fix use of HasModifiers in VopProfile.
Jan 15 2021, 9:22 AM · Restricted Project
mbrkusanin updated the diff for D92219: [AMDGPU][GlobalISel] Avoid selecting S_PACK with constants.

Rename + G_ANYEXT will now be treated same as G_SEXT instead of G_ZEXT.

Jan 15 2021, 9:15 AM · Restricted Project
mbrkusanin added a comment to D94795: [AMDGPU] Fix use of HasModifiers in VopProfile.

I would prefer to include HasClamp inside "field bit HasModifiers =" but VOPProfile allows clamp to be explicitly enabled through "bit _EnableClamp = 0" which still messes up the logic a little bit (like in class getIns64). Not sure if there is some other reason to keeping clamp separate from other modifiers.

Jan 15 2021, 9:14 AM · Restricted Project
mbrkusanin requested review of D94795: [AMDGPU] Fix use of HasModifiers in VopProfile.
Jan 15 2021, 9:13 AM · Restricted Project

Dec 10 2020

mbrkusanin committed rG0c7cce54eba3: [AMDGPU] Resolve issues when picking between ds_read/write and ds_read2/write2 (authored by mbrkusanin).
[AMDGPU] Resolve issues when picking between ds_read/write and ds_read2/write2
Dec 10 2020, 3:50 AM
mbrkusanin closed D92767: [AMDGPU] Resolve issues when picking between ds_read/write and ds_read2/write2.
Dec 10 2020, 3:50 AM · Restricted Project

Dec 8 2020

mbrkusanin added inline comments to D92767: [AMDGPU] Resolve issues when picking between ds_read/write and ds_read2/write2.
Dec 8 2020, 1:55 AM · Restricted Project
mbrkusanin updated the diff for D92767: [AMDGPU] Resolve issues when picking between ds_read/write and ds_read2/write2.
Dec 8 2020, 1:55 AM · Restricted Project

Dec 7 2020

mbrkusanin requested review of D92767: [AMDGPU] Resolve issues when picking between ds_read/write and ds_read2/write2.
Dec 7 2020, 7:50 AM · Restricted Project

Nov 27 2020

mbrkusanin added inline comments to D92219: [AMDGPU][GlobalISel] Avoid selecting S_PACK with constants.
Nov 27 2020, 5:58 AM · Restricted Project
mbrkusanin updated the summary of D92219: [AMDGPU][GlobalISel] Avoid selecting S_PACK with constants.
Nov 27 2020, 5:51 AM · Restricted Project
mbrkusanin updated the diff for D92219: [AMDGPU][GlobalISel] Avoid selecting S_PACK with constants.

Specific case from one of the tests:

%5:sgpr(s16) = G_FCONSTANT half 0xH4000
%7:sgpr(s32) = G_ANYEXT %5:sgpr(s16)
%4:sreg_32(<2 x s16>) = G_BUILD_VECTOR_TRUNC %7:sgpr(s32), %7:sgpr(s32)
Nov 27 2020, 5:26 AM · Restricted Project
mbrkusanin requested review of D92219: [AMDGPU][GlobalISel] Avoid selecting S_PACK with constants.
Nov 27 2020, 3:09 AM · Restricted Project

Nov 16 2020

mbrkusanin committed rG4cf6dd518ece: [AMDGPU][GlobalISel] Fix lowerShlSat (authored by mbrkusanin).
[AMDGPU][GlobalISel] Fix lowerShlSat
Nov 16 2020, 8:46 AM
mbrkusanin closed D91437: [AMDGPU][GlobalISel] Fix lowerShlSat.
Nov 16 2020, 8:46 AM · Restricted Project
mbrkusanin added inline comments to D91437: [AMDGPU][GlobalISel] Fix lowerShlSat.
Nov 16 2020, 8:29 AM · Restricted Project
mbrkusanin updated the diff for D91437: [AMDGPU][GlobalISel] Fix lowerShlSat.
Nov 16 2020, 8:29 AM · Restricted Project

Nov 13 2020

mbrkusanin requested review of D91437: [AMDGPU][GlobalISel] Fix lowerShlSat.
Nov 13 2020, 9:04 AM · Restricted Project

Nov 10 2020

mbrkusanin committed rGa75d6178b809: [GlobalISel] Add combine for (x | mask) -> x when (x | mask) == x (authored by mbrkusanin).
[GlobalISel] Add combine for (x | mask) -> x when (x | mask) == x
Nov 10 2020, 2:38 AM
mbrkusanin closed D90952: [GlobalISel] Add combine for (x | mask) -> x when (x | mask) == x.
Nov 10 2020, 2:38 AM · Restricted Project
mbrkusanin committed rGfb36ab0a42f4: [GlobalISel] Expand combine for (x & mask) -> x when (x & mask) == x (authored by mbrkusanin).
[GlobalISel] Expand combine for (x & mask) -> x when (x & mask) == x
Nov 10 2020, 2:38 AM
mbrkusanin closed D90674: [GlobalISel] Expand combine for (x & mask) -> x when (x & mask) == x.
Nov 10 2020, 2:37 AM · Restricted Project
mbrkusanin committed rG53ae95c946c0: [AMDGPU][GlobalISel] Combine shift + logic + shift with constant operands (authored by mbrkusanin).
[AMDGPU][GlobalISel] Combine shift + logic + shift with constant operands
Nov 10 2020, 2:37 AM
mbrkusanin closed D90223: [AMDGPU][GlobalISel] Combine shift + logic + shift with constant operands.
Nov 10 2020, 2:37 AM · Restricted Project
mbrkusanin committed rGde719586a8fa: [AMDGPU][GlobalISel] Fold a chain of two shift instructions with constant… (authored by mbrkusanin).
[AMDGPU][GlobalISel] Fold a chain of two shift instructions with constant…
Nov 10 2020, 2:37 AM
mbrkusanin closed D90217: [AMDGPU][GlobalISel] Fold a chain of two shift instructions with constant operands.
Nov 10 2020, 2:37 AM · Restricted Project
mbrkusanin updated the diff for D90952: [GlobalISel] Add combine for (x | mask) -> x when (x | mask) == x.
  • Rename
  • Updated comments
Nov 10 2020, 2:36 AM · Restricted Project
mbrkusanin updated the diff for D90674: [GlobalISel] Expand combine for (x & mask) -> x when (x & mask) == x.
  • Renamed
  • Updated comments
Nov 10 2020, 2:36 AM · Restricted Project
mbrkusanin updated the diff for D90223: [AMDGPU][GlobalISel] Combine shift + logic + shift with constant operands.
  • Rebase
Nov 10 2020, 2:35 AM · Restricted Project
mbrkusanin updated the diff for D90217: [AMDGPU][GlobalISel] Fold a chain of two shift instructions with constant operands.
  • Rebase
  • Updated comments
Nov 10 2020, 2:35 AM · Restricted Project

Nov 6 2020

mbrkusanin added inline comments to D90674: [GlobalISel] Expand combine for (x & mask) -> x when (x & mask) == x.
Nov 6 2020, 8:54 AM · Restricted Project
mbrkusanin requested review of D90952: [GlobalISel] Add combine for (x | mask) -> x when (x | mask) == x.
Nov 6 2020, 8:50 AM · Restricted Project
mbrkusanin updated the diff for D90674: [GlobalISel] Expand combine for (x & mask) -> x when (x & mask) == x.
  • Refactored match function
  • Updated tests
Nov 6 2020, 8:50 AM · Restricted Project
mbrkusanin updated the diff for D90223: [AMDGPU][GlobalISel] Combine shift + logic + shift with constant operands.
  • Rebase
Nov 6 2020, 8:50 AM · Restricted Project
mbrkusanin updated the diff for D90217: [AMDGPU][GlobalISel] Fold a chain of two shift instructions with constant operands.
  • Avoid combine for G_USHLSAT when sum exceeds scalar size.
Nov 6 2020, 8:50 AM · Restricted Project

Nov 3 2020

mbrkusanin requested review of D90674: [GlobalISel] Expand combine for (x & mask) -> x when (x & mask) == x.
Nov 3 2020, 2:51 AM · Restricted Project
mbrkusanin updated the diff for D90223: [AMDGPU][GlobalISel] Combine shift + logic + shift with constant operands.
  • Updated comments.
Nov 3 2020, 2:51 AM · Restricted Project
mbrkusanin updated the diff for D90217: [AMDGPU][GlobalISel] Fold a chain of two shift instructions with constant operands.
  • Fix handling of G_SSHLSAT and G_USHLSAT.
  • Removed new MachineIRBuilder.
Nov 3 2020, 2:51 AM · Restricted Project

Oct 30 2020

mbrkusanin updated the diff for D90223: [AMDGPU][GlobalISel] Combine shift + logic + shift with constant operands.
  • Supported G_SSHLSAT and G_USHLSAT and addressed other comments.
  • Added .mir test for G_SSHLSAT and G_USHLSAT.
Oct 30 2020, 9:16 AM · Restricted Project
mbrkusanin updated the diff for D90217: [AMDGPU][GlobalISel] Fold a chain of two shift instructions with constant operands.
  • Using RegisterImmPair, removed ShiftChain struct.
  • Supported G_SSHLSAT and G_USHLSAT.
  • Added .mir tests
Oct 30 2020, 9:14 AM · Restricted Project

Oct 27 2020

mbrkusanin updated the diff for D90223: [AMDGPU][GlobalISel] Combine shift + logic + shift with constant operands.
  • clang-format
Oct 27 2020, 6:37 AM · Restricted Project
mbrkusanin added inline comments to D90217: [AMDGPU][GlobalISel] Fold a chain of two shift instructions with constant operands.
Oct 27 2020, 6:08 AM · Restricted Project
mbrkusanin updated the diff for D90217: [AMDGPU][GlobalISel] Fold a chain of two shift instructions with constant operands.
  • Addressed review comments.
  • Updated comment at the start of match function.
Oct 27 2020, 6:08 AM · Restricted Project
mbrkusanin added a comment to D90223: [AMDGPU][GlobalISel] Combine shift + logic + shift with constant operands.

This patch is basically a global-isel version of combineShiftOfShiftedLogic() from DAGCombiner.

Oct 27 2020, 4:46 AM · Restricted Project
mbrkusanin requested review of D90223: [AMDGPU][GlobalISel] Combine shift + logic + shift with constant operands.
Oct 27 2020, 4:41 AM · Restricted Project
mbrkusanin retitled D90217: [AMDGPU][GlobalISel] Fold a chain of two shift instructions with constant operands from [GlobalISel] Fold a chain of two shift instructions with constant operands to [AMDGPU][GlobalISel] Fold a chain of two shift instructions with constant operands.
Oct 27 2020, 4:14 AM · Restricted Project
mbrkusanin requested review of D90217: [AMDGPU][GlobalISel] Fold a chain of two shift instructions with constant operands.
Oct 27 2020, 3:59 AM · Restricted Project

Oct 13 2020

mbrkusanin committed rG52ba4fa6aa21: [GlobalISel] Avoid making G_PTR_ADD with nullptr (authored by mbrkusanin).
[GlobalISel] Avoid making G_PTR_ADD with nullptr
Oct 13 2020, 4:04 AM
mbrkusanin closed D87140: [GlobalISel] Avoid making G_PTR_ADD with nullptr.
Oct 13 2020, 4:04 AM · Restricted Project
mbrkusanin updated the diff for D87140: [GlobalISel] Avoid making G_PTR_ADD with nullptr.
  • Addressed comments.
Oct 13 2020, 2:55 AM · Restricted Project