[PATCH] D156989: FloatingPointMode: Use -1 for "Dynamic"

foad added inline comments. Comment at: llvm/include/llvm/ADT/FloatingPointMode.h:39 + // Special values. + Invalid = -2, + Lost the `///<` comment here. Comment at: llvm/include/llvm/ADT/FloatingPointMode.h:41 + + ///< Denotes mode unknown

[PATCH] D74094: [IRGen] Emit lifetime intrinsics around temporary aggregate argument allocas

foad added a comment. Herald added a project: All. Hi @erik.pilkington, I see this got reverted: commit e26c24b849211f35a988d001753e0cd15e4a9d7b Author: Erik Pilkington Date: Wed Feb 12 12:02:58 2020 -0800 Revert "[IRGen] Emit lifetime intrinsics around temporary aggregate

[PATCH] D155429: [AMDGPU] Add targets gfx1150 and gfx1151

This revision was landed with ongoing or failed builds. This revision was automatically updated to reflect the committed changes. Closed by commit rG92542f2a4000: [AMDGPU] Add targets gfx1150 and gfx1151 (authored by foad). Repository: rG LLVM Github Monorepo CHANGES SINCE LAST ACTION

[PATCH] D155429: [AMDGPU] Add targets gfx1150 and gfx1151

foad created this revision. foad added a reviewer: AMDGPU. Herald added subscribers: libc-commits, mattd, gchakrabarti, asavonic, StephenFan, kerbowa, hiraditya, tpr, dstuttard, yaxunl, jvesely, kzhuravl, emaste, arsenm. Herald added a reviewer: jhenderson. Herald added a reviewer: MaskRay.

[PATCH] D153953: Revert "[AMDGPU] Mark mbcnt as convergent"

foad added a comment. In D153953#4458134 , @sameerds wrote: > @pravinjagtap @arsenm ... reverting the mbcnt intrinsic affects tests that > were added for atomic optimizations. In particular, the mbcnt is now being > moved across/into/out of control

[PATCH] D152901: AMDGPU: Add llvm.amdgcn.exp2 intrinsic

foad added inline comments. Comment at: llvm/lib/Target/AMDGPU/AMDGPUISelLowering.h:463 + // exp2, no denormal handling for f32. + EXP, + Is this used anywhere? CHANGES SINCE LAST ACTION https://reviews.llvm.org/D152901/new/

[PATCH] D152697: AMDGPU: Add llvm.amdgcn.log intrinsic

foad accepted this revision. foad added a comment. This revision is now accepted and ready to land. Seems fine. CHANGES SINCE LAST ACTION https://reviews.llvm.org/D152697/new/ https://reviews.llvm.org/D152697 ___ cfe-commits mailing list

[PATCH] D149776: Re-land "[AMDGPU] Define data layout entries for buffers""

foad added a comment. Hi, with the new datalayout we're hitting this crash: ; RUN: opt -passes=indvars -S < %s target datalayout =

[PATCH] D147732: [AMDGPU] Add f32 permlane{16, x16} builtin variants

foad added a comment. In D147732#4268661 , @rampitec wrote: > In D147732#4267553 , @foad wrote: > >> Changing the existing intrinsics to use type mangling could break clients >> like LLPC and Mesa. I've put up a

[PATCH] D147732: [AMDGPU] Add f32 permlane{16, x16} builtin variants

foad added a comment. Changing the existing intrinsics to use type mangling could break clients like LLPC and Mesa. I've put up a patch for LLPC to protect it against this change: https://github.com/GPUOpen-Drivers/llpc/pull/2404 Repository: rG LLVM Github Monorepo CHANGES SINCE LAST

[PATCH] D146808: [AMDGPU] Add clang builtin for __builtin_amdgcn_ds_atomic_fadd_v2f16

foad accepted this revision. foad added a comment. This revision is now accepted and ready to land. LGTM. Comment at: clang/include/clang/Basic/BuiltinsAMDGPU.def:234 TARGET_BUILTIN(__builtin_amdgcn_ds_atomic_fadd_v2bf16, "V2sV2s*3V2s", "t", "atomic-ds-pk-add-16-insts")

[PATCH] D146701: [AMDGPU] Create Subtarget Features for some of 16 bits atomic fadd instructions

foad accepted this revision. foad added a comment. This revision is now accepted and ready to land. LGTM, thanks! If you want to remove some of the other unnecessary predicates from Real instructions you could do that in a separate NFC patch. Repository: rG LLVM Github Monorepo CHANGES

[PATCH] D146701: [AMDGPU] Create Subtarget Features for some of 16 bits atomic fadd instructions

foad added inline comments. Comment at: llvm/lib/Target/AMDGPU/BUFInstructions.td:2891 +let SubtargetPredicate = HasAtomicFaddNoRtnInsts in { +defm BUFFER_ATOMIC_ADD_F32: MUBUF_Real_Atomic_vi <0x4d>; mariusz-sikora-at-amd wrote: > foad wrote: > > Is this

[PATCH] D146701: [AMDGPU] Create Subtarget Features for some of 16 bits atomic fadd instructions

foad added inline comments. Comment at: llvm/lib/Target/AMDGPU/BUFInstructions.td:2891 +let SubtargetPredicate = HasAtomicFaddNoRtnInsts in { +defm BUFFER_ATOMIC_ADD_F32: MUBUF_Real_Atomic_vi <0x4d>; Is this still required? Comment at:

[PATCH] D146701: [AMDGPU] Create Subtarget Features for some of 16 bits atomic fadd instructions

foad added inline comments. Comment at: llvm/lib/Target/AMDGPU/BUFInstructions.td:2889 -defm BUFFER_ATOMIC_ADD_F32: MUBUF_Real_Atomic_vi <0x4d>; +let SubtargetPredicate = HasAtomicBufferGlobalPkAddF16NoRtnInsts in { defm BUFFER_ATOMIC_PK_ADD_F16 : MUBUF_Real_Atomic_vi

[PATCH] D146701: [AMDGPU] Create Subtarget Features for some of 16 bits atomic fadd instructions

foad added inline comments. Comment at: clang/include/clang/Basic/BuiltinsAMDGPU.def:233 +TARGET_BUILTIN(__builtin_amdgcn_global_atomic_fadd_v2bf16, "V2sV2s*1V2s", "t", "atomic-global-pk-add-bf16-inst") +TARGET_BUILTIN(__builtin_amdgcn_ds_atomic_fadd_v2bf16, "V2sV2s*3V2s", "t",

[PATCH] D145441: [AMDGPU] Define data layout entries for buffers

foad added a comment. Just my 2p: it feels a bit premature to commit patches for this. It feels more like something you could prototype on a branch somewhere and come back when you have more experience with how it all works out in practice. But I don't actually object to the patch, if the

[PATCH] D142823: Intrinsics: Allow tablegen to mark parameters with dereferenceable

foad added a comment. In D142823#4093363 , @arsenm wrote: > In D142823#4093357 , @foad wrote: > >> I think the tablegen functionality should be a separate patch from the >> amdgpu changes. > > Maybe, but then

[PATCH] D142968: [NFC] Extract `CodeGenInstAlias` into its own *.h/*.cpp

foad accepted this revision. foad added a comment. This revision is now accepted and ready to land. Seems reasonable. Repository: rG LLVM Github Monorepo CHANGES SINCE LAST ACTION https://reviews.llvm.org/D142968/new/ https://reviews.llvm.org/D142968

[PATCH] D142823: Intrinsics: Allow tablegen to mark parameters with dereferenceable

foad added a comment. I think the tablegen functionality should be a separate patch from the amdgpu changes. CHANGES SINCE LAST ACTION https://reviews.llvm.org/D142823/new/ https://reviews.llvm.org/D142823 ___ cfe-commits mailing list

[PATCH] D142968: [NFC] Extract `CodeGenInstAlias` into its own *.h/*.cpp

foad added inline comments. Comment at: clang/docs/tools/clang-formatted-files.txt:7421 llvm/utils/not/not.cpp +llvm/Utils/TableGen/CodeGenInstAlias.cpp +llvm/Utils/TableGen/CodeGenInstAlias.h Should come after CodeBeads in alphabetical order? Repository:

[PATCH] D142968: [NFC] Extract `CodeGenInstAlias` into its own *.h/*.cpp

foad added a comment. Looks OK but what's the motivation for it? Repository: rG LLVM Github Monorepo CHANGES SINCE LAST ACTION https://reviews.llvm.org/D142968/new/ https://reviews.llvm.org/D142968 ___ cfe-commits mailing list

[PATCH] D141798: Remove ZeroBehavior of countLeadingZeros and the like (NFC)

foad added a comment. Herald added a subscriber: luke. In D141798#4055050 , @barannikov88 wrote: > It would be nice to have comments reflecting the new behavior in the case of > 0 / max value. +1 Repository: rG LLVM Github Monorepo CHANGES SINCE

[PATCH] D140467: [X86][Reduce] Preserve fast math flags when change it. NFCI

foad added inline comments. Comment at: clang/lib/CodeGen/CGBuiltin.cpp:14740 CGM.getIntrinsic(Intrinsic::vector_reduce_fadd, Ops[1]->getType()); +FastMathFlags FMF = Builder.getFastMathFlags(); Builder.getFastMathFlags().setAllowReassoc(); We

[PATCH] D140294: clang: Replace implementation of __builtin_isnormal

foad added inline comments. Comment at: clang/lib/CodeGen/CGBuiltin.cpp:3308 case Builtin::BI__builtin_isnormal: { -// isnormal(x) --> x == x && fabsf(x) < infinity && fabsf(x) >= float_min +// isnormal(x) --> fabs(x) < infinity && !(fabs(x) < float_min)

[PATCH] D137960: [Lexer] Speedup LexTokenInternal

foad added inline comments. Comment at: clang/lib/Lex/Lexer.cpp:3520-3521 +LexStart: + assert(!Result.needsCleaning() && "Result doesn't need cleaning"); + assert(!Result.hasPtrData() && "Result has been reset"); Messages are backwards. They should be like

[PATCH] D137524: clang/AMDGPU: Emit atomicrmw for atomic_inc/dec builtins

foad added a comment. In D137524#3911439 , @JonChesterfield wrote: > Do you know where the uinc_wrap etc were introduced? D137361 in the stack for this patch. CHANGES SINCE LAST ACTION

[PATCH] D137338: Fix dupe word typos

foad added a comment. I committed the lib/Target/AMDGPU parts as 5073ae2a883f . Repository: rG LLVM Github Monorepo CHANGES SINCE LAST ACTION https://reviews.llvm.org/D137338/new/ https://reviews.llvm.org/D137338

[PATCH] D125557: [APInt] Remove all uses of zextOrSelf, sextOrSelf and truncOrSelf

This revision was automatically updated to reflect the committed changes. Closed by commit rG6bec3e9303d6: [APInt] Remove all uses of zextOrSelf, sextOrSelf and truncOrSelf (authored by foad). Repository: rG LLVM Github Monorepo CHANGES SINCE LAST ACTION

[PATCH] D125557: [APInt] Remove all uses of zextOrSelf, sextOrSelf and truncOrSelf

foad added inline comments. Comment at: llvm/lib/IR/ConstantRange.cpp:724 auto BW = getBitWidth(); -APInt Min = APInt::getMinValue(BW).zextOrSelf(ResultBitWidth); -APInt Max = APInt::getMaxValue(BW).zextOrSelf(ResultBitWidth); +APInt Min =

[PATCH] D125557: [APInt] Remove all uses of zextOrSelf, sextOrSelf and truncOrSelf

foad marked 2 inline comments as done. foad added inline comments. Comment at: llvm/lib/Analysis/ConstantFolding.cpp:2884 if (IntrinsicID == Intrinsic::smul_fix_sat) { - APInt Max = APInt::getSignedMaxValue(Width).sextOrSelf(ExtendedWidth); - APInt Min =

[PATCH] D125557: [APInt] Remove all uses of zextOrSelf, sextOrSelf and truncOrSelf

foad updated this revision to Diff 429466. foad added a comment. Address some review comments. Repository: rG LLVM Github Monorepo CHANGES SINCE LAST ACTION https://reviews.llvm.org/D125557/new/ https://reviews.llvm.org/D125557 Files: clang/lib/AST/ExprConstant.cpp

[PATCH] D125557: [APInt] Remove all uses of zextOrSelf, sextOrSelf and truncOrSelf

foad created this revision. foad added reviewers: lattner, RKSimon, lebedev.ri, spatel. Herald added subscribers: kosarev, jsilvanus, hsmhsm, jeroen.dobbelaere, frasercrmck, ecnelises, martong, kerbowa, luismarques, apazos, sameer.abuasal, pengfei, s.egerton, Jim, jocewei, PkmX, the_o,

[PATCH] D124536: [AMDGPU] Add gfx11 subtarget ELF definition

foad accepted this revision. foad added a reviewer: t-tye. foad added a comment. This revision is now accepted and ready to land. LGTM, thanks! Repository: rG LLVM Github Monorepo CHANGES SINCE LAST ACTION https://reviews.llvm.org/D124536/new/ https://reviews.llvm.org/D124536

[PATCH] D114957: [AMDGPU] Change llvm.amdgcn.image.bvh.intersect.ray to take vec3 args

foad abandoned this revision. foad added a comment. Abandoned in favour of D115032 . Repository: rG LLVM Github Monorepo CHANGES SINCE LAST ACTION https://reviews.llvm.org/D114957/new/ https://reviews.llvm.org/D114957

[PATCH] D115032: [AMDGPU] Change llvm.amdgcn.image.bvh.intersect.ray to take vec3 args

This revision was landed with ongoing or failed builds. This revision was automatically updated to reflect the committed changes. Closed by commit rG2774bad11242: [AMDGPU] Change llvm.amdgcn.image.bvh.intersect.ray to take vec3 args (authored by foad). Repository: rG LLVM Github Monorepo

[PATCH] D114957: [AMDGPU] Change llvm.amdgcn.image.bvh.intersect.ray to take vec3 args

foad added a comment. In D114957#3167700 , @arsenm wrote: > I think this macro is purely terrible and should not be added (and at least > should be all caps?). If we can't just hard break users, I would rather just > leave the builtin signatures broken

[PATCH] D115032: [AMDGPU] Change llvm.amdgcn.image.bvh.intersect.ray to take vec3 args

foad added a comment. This is an alternative to D114957 that does not update the API of the OpenCL builtins. Repository: rG LLVM Github Monorepo CHANGES SINCE LAST ACTION https://reviews.llvm.org/D115032/new/ https://reviews.llvm.org/D115032

[PATCH] D115032: [AMDGPU] Change llvm.amdgcn.image.bvh.intersect.ray to take vec3 args

foad created this revision. foad added reviewers: arsenm, rampitec, yaxunl, critson, b-sumner. Herald added subscribers: kerbowa, hiraditya, t-tye, Anastasia, tpr, dstuttard, nhaehnle, jvesely, kzhuravl. foad requested review of this revision. Herald added subscribers: llvm-commits, cfe-commits,

[PATCH] D114957: [AMDGPU] Change llvm.amdgcn.image.bvh.intersect.ray to take vec3 args

foad updated this revision to Diff 391403. foad added a comment. Define __amdgcn_bvh_use_vec3__. Repository: rG LLVM Github Monorepo CHANGES SINCE LAST ACTION https://reviews.llvm.org/D114957/new/ https://reviews.llvm.org/D114957 Files: clang/include/clang/Basic/BuiltinsAMDGPU.def

[PATCH] D114957: [AMDGPU] Change llvm.amdgcn.image.bvh.intersect.ray to take vec3 args

foad added a comment. In D114957#3166948 , @b-sumner wrote: > In D114957#3166936 , @foad wrote: > >> In D114957#3166858 , @yaxunl wrote: >> >>> In D114957#3166817

[PATCH] D114957: [AMDGPU] Change llvm.amdgcn.image.bvh.intersect.ray to take vec3 args

foad added a comment. In D114957#3166858 , @yaxunl wrote: > In D114957#3166817 , @foad wrote: > >> This is a flag-day change to the signatures of the LLVM intrinsics and the >> OpenCL builtins. Is that OK? > >

[PATCH] D114957: [AMDGPU] Change llvm.amdgcn.image.bvh.intersect.ray to take vec3 args

foad added a comment. This is a flag-day change to the signatures of the LLVM intrinsics and the OpenCL builtins. Is that OK? Repository: rG LLVM Github Monorepo CHANGES SINCE LAST ACTION https://reviews.llvm.org/D114957/new/ https://reviews.llvm.org/D114957

[PATCH] D114957: [AMDGPU] Change llvm.amdgcn.image.bvh.intersect.ray to take vec3 args

foad created this revision. foad added reviewers: arsenm, rampitec, critson, yaxunl, b-sumner. Herald added subscribers: kerbowa, hiraditya, t-tye, tpr, dstuttard, nhaehnle, jvesely, kzhuravl. foad requested review of this revision. Herald added subscribers: llvm-commits, cfe-commits, wdng.

[PATCH] D112791: [IR] Merge createReplacementInstr into ConstantExpr::getAsInstruction

This revision was landed with ongoing or failed builds. This revision was automatically updated to reflect the committed changes. Closed by commit rG1b758925adf6: [IR] Merge createReplacementInstr into ConstantExpr::getAsInstruction (authored by foad). Repository: rG LLVM Github Monorepo

[PATCH] D112791: [IR] Merge createReplacementInstr into ConstantExpr::getAsInstruction

foad added inline comments. Comment at: llvm/include/llvm/IR/Constants.h:1317 /// would make it harder to remove ConstantExprs altogether. - Instruction *getAsInstruction() const; + Instruction *getAsInstruction(Instruction *InsertBefore = nullptr) const;

[PATCH] D112791: [IR] Merge createReplacementInstr into ConstantExpr::getAsInstruction

foad updated this revision to Diff 383341. foad added a comment. Add comment about InsertBefore. Repository: rG LLVM Github Monorepo CHANGES SINCE LAST ACTION https://reviews.llvm.org/D112791/new/ https://reviews.llvm.org/D112791 Files: clang/lib/CodeGen/CGCUDANV.cpp

[PATCH] D112791: [IR] Merge createReplacementInstr into ConstantExpr::getAsInstruction

foad created this revision. Herald added subscribers: ormris, dexonsmith, hiraditya. foad requested review of this revision. Herald added projects: clang, LLVM. Herald added subscribers: llvm-commits, cfe-commits. createReplacementInstr was a trivial wrapper around ConstantExpr::getAsInstruction,

[PATCH] D110808: [APInt] Stop using soft-deprecated constructors and methods in clang. NFC.

This revision was landed with ongoing or failed builds. This revision was automatically updated to reflect the committed changes. Closed by commit rGd933adeaca7b: [APInt] Stop using soft-deprecated constructors and methods in clang. NFC. (authored by foad). Repository: rG LLVM Github Monorepo

[PATCH] D110808: [APInt] Stop using soft-deprecated constructors and methods in clang. NFC.

foad created this revision. Herald added a subscriber: martong. foad requested review of this revision. Herald added projects: clang, clang-tools-extra. Herald added a subscriber: cfe-commits. Stop using APInt constructors and methods that were soft-deprecated in D109483

[PATCH] D109483: [APInt] Normalize naming on keep constructors / predicate methods.

foad added a comment. What is a "keep constructor"? Repository: rG LLVM Github Monorepo CHANGES SINCE LAST ACTION https://reviews.llvm.org/D109483/new/ https://reviews.llvm.org/D109483 ___ cfe-commits mailing list cfe-commits@lists.llvm.org

[PATCH] D81886: [AMDGPU] Add gfx1030 target

foad added inline comments. Herald added a subscriber: dexonsmith. Comment at: llvm/lib/Target/AMDGPU/AMDGPU.td:1245 + +def HasDsSrc2Insts : Predicate<"!Subtarget->hasDsSrc2Insts()">, + AssemblerPredicate<(all_of FeatureDsSrc2Insts)>; The `!` is obviously wrong

[PATCH] D104804: [AMDGPU] Add gfx1035 target

foad added a comment. Looks OK. Have you run check-llvm and check-clang? Repository: rG LLVM Github Monorepo CHANGES SINCE LAST ACTION https://reviews.llvm.org/D104804/new/ https://reviews.llvm.org/D104804 ___ cfe-commits mailing list

[PATCH] D104804: [AMDGPU] Add gfx1035 target

foad added inline comments. Comment at: clang/test/Driver/amdgpu-mcpu.cl:138 // GFX1034: "-target-cpu" "gfx1034" +// GFX1034: "-target-cpu" "gfx1035" Typo in check prefix. Why didn't this make the test fail? Comment at:

[PATCH] D104124: [IR] Simplify createReplacementInstr

This revision was automatically updated to reflect the committed changes. Closed by commit rG157473a58f02: [IR] Simplify createReplacementInstr (authored by foad). Repository: rG LLVM Github Monorepo CHANGES SINCE LAST ACTION https://reviews.llvm.org/D104124/new/

[PATCH] D104124: [IR] Simplify createReplacementInstr

foad added reviewers: yaxunl, robertlytton. foad added a comment. Given how simple createReplacementInstr is now, this does make me wonder if it was really worth creating ReplaceConstant.{cpp,h} in the first place. Repository: rG LLVM Github Monorepo CHANGES SINCE LAST ACTION

[PATCH] D104124: [IR] Simplify createReplacementInstr

foad created this revision. Herald added subscribers: dexonsmith, hiraditya. foad requested review of this revision. Herald added projects: clang, LLVM. Herald added subscribers: llvm-commits, cfe-commits. NFCI, although the test change shows that ConstantExpr::getAsInstruction is better than the

[PATCH] D103663: [AMDGPU] Add gfx1013 target

foad added inline comments. Comment at: llvm/lib/Target/AMDGPU/AMDGPU.td:471 + "true", + "Encoding format GFX10_A" +>; bcahoon wrote: > foad wrote: > > I realise you're just following the precedent set by GFX10_B, but is this > > terminology actually used in

[PATCH] D103663: [AMDGPU] Add gfx1013 target

foad accepted this revision. foad added a comment. LGTM anyway, with or without any action on my last couple of comments. CHANGES SINCE LAST ACTION https://reviews.llvm.org/D103663/new/ https://reviews.llvm.org/D103663 ___ cfe-commits mailing list

[PATCH] D103663: [AMDGPU] Add gfx1013 target

foad added inline comments. Comment at: llvm/lib/Target/AMDGPU/AMDGPU.td:471 + "true", + "Encoding format GFX10_A" +>; I realise you're just following the precedent set by GFX10_B, but is this terminology actually used in any documentation anywhere? And if

[PATCH] D103663: [AMDGPU] Add gfx1013 target

foad added inline comments. Comment at: llvm/docs/AMDGPUUsage.rst:389 - xnack scratch - *pal-amdpal* + ``gfx1013`` ``amdgcn`` dGPU - cumode - Absolute - *rocm-amdhsa*

[PATCH] D103663: [AMDGPU] Add gfx1013 target

foad added a comment. Please also update `llvm/test/Object/AMDGPU/elf-header-flags-mach.yaml`. Repository: rG LLVM Github Monorepo CHANGES SINCE LAST ACTION https://reviews.llvm.org/D103663/new/ https://reviews.llvm.org/D103663 ___ cfe-commits

[PATCH] D102306: Add gfx1034

foad added a comment. Can you also check for updates in: clang/lib/Basic/Cuda.cpp openmp/libomptarget/plugins/amdgpu/impl/get_elf_mach_gfx_name.cpp llvm/test/CodeGen/AMDGPU/directive-amdgcn-target.ll llvm/test/tools/llvm-objdump/ELF/AMDGPU/subtarget.ll

[PATCH] D69498: IR: Invert convergent attribute handling

foad added a comment. In D69498#2705441 , @sameerds wrote: > I would propose refining the definition of the `noconvergent` attribute as > follows: > >> noconvergent: >> >> Some targets with a parallel execution model provide cross-thread operations >>

[PATCH] D69498: IR: Invert convergent attribute handling

foad added a comment. > But in practice, the main issue for everyone is the effect on compile time > for targets that don't care about convergence/divergence. For such targets, > running even the divergence analysis is an unnecessary cost. LegacyDivergenceAnalysis::runOnFunction bails out

[PATCH] D69498: IR: Invert convergent attribute handling

foad added a comment. In D69498#2703317 , @sameerds wrote: > The way I see it, the notion of convergence is relevant only to a certain > class of targets (usually represented by GPUs) and it only affects certain > optimizations. Then why not have only

[PATCH] D96906: [AMDGPU] gfx90a support

foad added inline comments. Herald added a subscriber: mstorsjo. Comment at: llvm/lib/Target/AMDGPU/SIFoldOperands.cpp:100 bool tryFoldOMod(MachineInstr ); + bool tryFoldRegSeqence(MachineInstr ); + bool tryFoldLCSSAPhi(MachineInstr ); Spelling "sequence".

[PATCH] D98717: [AMDGPU] Split dot2-insts feature

This revision was landed with ongoing or failed builds. This revision was automatically updated to reflect the committed changes. Closed by commit rG967b64beb4bf: [AMDGPU] Split dot2-insts feature (authored by foad). Repository: rG LLVM Github Monorepo CHANGES SINCE LAST ACTION

[PATCH] D98717: [AMDGPU] Split dot2-insts feature

foad added inline comments. Comment at: llvm/lib/Target/AMDGPU/AMDGPU.td:511 +def FeatureDot7Insts : SubtargetFeature<"dot7-insts", + "HasDot7Insts", + "true", arsenm wrote: > I'm not sure where the "7" is coming from It's the next number after 6. I don't

[PATCH] D98717: [AMDGPU] Split dot2-insts feature

foad created this revision. foad added reviewers: rampitec, kzhuravl, b-sumner. Herald added subscribers: kerbowa, hiraditya, t-tye, tpr, dstuttard, yaxunl, nhaehnle, jvesely, arsenm. foad requested review of this revision. Herald added subscribers: llvm-commits, cfe-commits, wdng. Herald added

[PATCH] D97928: [AMDGPU] Restore the s_memtime instruction in gfx1030

This revision was landed with ongoing or failed builds. This revision was automatically updated to reflect the committed changes. Closed by commit rGfc28f600e558: [AMDGPU] Restore the s_memtime instruction in gfx1030 (authored by foad). Repository: rG LLVM Github Monorepo CHANGES SINCE LAST

[PATCH] D97928: [AMDGPU] Restore the s_memtime instruction in gfx1030

foad updated this revision to Diff 328429. foad added a comment. Herald added a project: clang. Herald added a subscriber: cfe-commits. Add clang changes. Repository: rG LLVM Github Monorepo CHANGES SINCE LAST ACTION https://reviews.llvm.org/D97928/new/ https://reviews.llvm.org/D97928

[PATCH] D61112: AMDGPU: Enable _Float16

foad added a comment. Herald added a subscriber: kerbowa. Should have updated https://clang.llvm.org/docs/LanguageExtensions.html#half-precision-floating-point "_Float16 is currently only supported on the following targets, with further targets pending ABI standardization: ..." CHANGES SINCE

[PATCH] D96906: [AMDGPU] gfx90a support

foad added a comment. In D96906#2573265 , @echristo wrote: > In D96906#2572842 , @msearles wrote: > >> In D96906#2572749 , @kzhuravl wrote: >> The point is that nobody

[PATCH] D96906: [AMDGPU] gfx90a support

foad added inline comments. Comment at: llvm/test/CodeGen/AMDGPU/llvm.amdgcn.workitem.id.ll:23 + +; CO-V3: .amdhsa_system_vgpr_workitem_id 0 +; PACKED-TID: .amdhsa_system_vgpr_workitem_id 0 CO-V3 isn't tested by any RUN line. I think FileCheck might complain

[PATCH] D90809: [amdgpu] Add `llvm.amdgcn.endpgm` support.

foad added inline comments. Comment at: llvm/include/llvm/IR/IntrinsicsAMDGPU.td:1580 +def int_amdgcn_endpgm : GCCBuiltin<"__builtin_amdgcn_endpgm">, + Intrinsic<[], [], [IntrNoReturn, IntrCold, IntrNoMem, IntrHasSideEffects] The intrinsic def needs a

[PATCH] D90419: [AMDGPU] Add gfx90c target

foad added inline comments. Comment at: llvm/docs/AMDGPUUsage.rst:261 + ``gfx90c`` ``amdgcn`` APU - xnack *TBA* + [off] +

[PATCH] D90447: [AMDGPU] Add gfx1033 target

foad added a comment. Don't you need to update lib/Object/ELFObjectFile.cpp and test/Object/AMDGPU/elf-header-flags-mach.yaml? Repository: rG LLVM Github Monorepo CHANGES SINCE LAST ACTION https://reviews.llvm.org/D90447/new/ https://reviews.llvm.org/D90447

[PATCH] D89487: [AMDGPU] gfx1032 target

foad added inline comments. Comment at: llvm/docs/AMDGPUUsage.rst:280 names. + ``gfx1032`` ``amdgcn`` dGPU - xnack *TBA* +

[PATCH] D81728: [InstCombine] Add target-specific inst combining

foad added inline comments. Comment at: llvm/include/llvm/Analysis/TargetTransformInfo.h:552-555 + /// \returns false to not do anything target specific or true to return the + /// value in \p ResultI from the InstCombiner. It is possible to return null + /// and stop further

[PATCH] D81728: [InstCombine] Add target-specific inst combining

foad added a subscriber: bogner. foad added inline comments. Comment at: llvm/lib/Transforms/InstCombine/InstCombineSimplifyDemanded.cpp:1444 + *this, *II, DemandedElts, UndefElts, UndefElts2, UndefElts3, + simplifyAndSetOp, )) +return V;

[PATCH] D82085: [TRE] markTails marks call sites as tailcalls though some of them are not.

foad added a comment. > markTails function set IsTailcall bit for functions which are not > last calls: It's OK to set "tail" on any call that satisfies these requirements (from https://llvm.org/docs/LangRef.html#call-instruction): "Both markers [tail and musttail] imply that the callee does

[PATCH] D81886: [AMDGPU] Add gfx1030 target

foad added inline comments. Comment at: llvm/docs/AMDGPUUsage.rst:266-267 names. + ``gfx1030`` ``amdgcn`` dGPU - xnack *TBA* +

[PATCH] D71213: [Alignment][NFC] CreateMemSet use MaybeAlign

foad added a comment. In D71213#1780088 , @gchatelet wrote: > In D71213#1779841 , @foad wrote: > > > @gchatelet in general would it be possible to make changes like this in a > > backwards-compatible way, or in

[PATCH] D71213: [Alignment][NFC] CreateMemSet use MaybeAlign

foad added a comment. @gchatelet in general would it be possible to make changes like this in a backwards-compatible way, or in two stages without a "flag day" change? We have out-of-tree users of CreateMemSet and it's awkward to change them all at exactly the same time as we merge in this