aheejin (Heejin Ahn)
User

Projects

User does not belong to any projects.

User Details

User Since
Jul 29 2016, 12:33 AM (112 w, 4 d)

Recent Activity

Today

aheejin added inline comments to D52325: [WebAssembly] SIMD min and max.
Tue, Sep 25, 3:48 PM
aheejin added a comment to D52324: [ValueTracking] Allow select patterns to work on vectors in more places.

Do we need to handle integer vectors too?

Tue, Sep 25, 3:31 PM
aheejin added inline comments to D52324: [ValueTracking] Allow select patterns to work on vectors in more places.
Tue, Sep 25, 1:48 PM
aheejin committed rL343018: Unify landing pad information adding routines (NFC).
Unify landing pad information adding routines (NFC)
Tue, Sep 25, 1:00 PM
aheejin closed D52428: Unify landing pad information adding routines (NFC).
Tue, Sep 25, 1:00 PM
aheejin added a comment to D52450: [Intrinsic] Add llvm.minnan and llvm.maxnan instrinsic functions.

Are these all places we need to handle these new intrinsics? Searching by minnum or maxnum in the llvm directory yields other occurrences in files like lib/Analysis/ValueTracking.cpp, lib/Analysis/VectorUtils.cpp, and lib/Analysis/InstructionSimplify.cpp. (There are more). For test cases as well.

Tue, Sep 25, 12:28 PM
aheejin accepted D52512: [WebAssembly] Restore proposed SIMD conversion names.
Tue, Sep 25, 12:03 PM
aheejin accepted D52372: [WebAssembly] SIMD conversions.
Tue, Sep 25, 12:02 PM
aheejin added a comment to D52428: Unify landing pad information adding routines (NFC).

Thank you!

Tue, Sep 25, 11:57 AM
aheejin committed rL342952: [llvm-exegesis] Add MCParser to LLVM_LINK_COMPONENTS.
[llvm-exegesis] Add MCParser to LLVM_LINK_COMPONENTS
Tue, Sep 25, 1:27 AM

Yesterday

aheejin accepted D52387: [WebAssembly] SIMD sqrt.
Mon, Sep 24, 6:16 PM
aheejin added a comment to D52325: [WebAssembly] SIMD min and max.

Then maybe it's better to add test for ordered ones too..?

Mon, Sep 24, 6:16 PM
aheejin added inline comments to D52324: [ValueTracking] Allow select patterns to work on vectors in more places.
Mon, Sep 24, 6:09 PM
aheejin updated the diff for D52428: Unify landing pad information adding routines (NFC).
  • Address comments
Mon, Sep 24, 5:44 PM
aheejin added a comment to D52372: [WebAssembly] SIMD conversions.

@sunfish Oh I see. Thank you!

Mon, Sep 24, 5:31 PM
aheejin added inline comments to D52372: [WebAssembly] SIMD conversions.
Mon, Sep 24, 5:12 PM
aheejin added a comment to D52372: [WebAssembly] SIMD conversions.
  1. Is it correct to translate those llvm instructions to saturating versions of wasm instructions? For example, for scalars, we had a patch that implemented the saturating behavior. Don't we need a similar thing for vectors too?
  2. Is this behavior also dependent on nontrapping-fptoint feature, as the scalar instructions?
Mon, Sep 24, 5:12 PM
aheejin added a comment to D52325: [WebAssembly] SIMD min and max.

What happens if we use ordered comparisons? They fail to be selected?

Mon, Sep 24, 3:17 PM
aheejin added a comment to D52324: [ValueTracking] Allow select patterns to work on vectors in more places.
  1. What happens when only some of the elements in a vector are NaN?
  2. What happens we have -0.0 in a vector?
Mon, Sep 24, 3:01 PM
aheejin added inline comments to D52428: Unify landing pad information adding routines (NFC).
Mon, Sep 24, 11:36 AM
aheejin created D52428: Unify landing pad information adding routines (NFC).
Mon, Sep 24, 10:56 AM
aheejin accepted D52388: [WebAssembly][NFC] Fix hardcoded stack indices in tests.
Mon, Sep 24, 10:43 AM

Thu, Sep 20

aheejin accepted D52333: [WebAssembly][NFC] Rename simd-conversions test to simd-bitcasts.
Thu, Sep 20, 7:14 PM
aheejin accepted D52105: [WebAssembly] Add V128 value type to binary format.

I was not sure if what @dschuff asked was done. If it was I’m ok with the change.

Thu, Sep 20, 2:57 PM

Wed, Sep 19

aheejin accepted D52282: [WebAssembly] Renumber SIMD ops.
Wed, Sep 19, 4:19 PM
aheejin added inline comments to D52105: [WebAssembly] Add V128 value type to binary format.
Wed, Sep 19, 11:47 AM
aheejin added inline comments to D52105: [WebAssembly] Add V128 value type to binary format.
Wed, Sep 19, 11:29 AM
aheejin added inline comments to D52108: [WebAssembly] Handle V128RegClass in more places.
Wed, Sep 19, 11:28 AM
aheejin added inline comments to D52105: [WebAssembly] Add V128 value type to binary format.
Wed, Sep 19, 11:21 AM
aheejin accepted D52254: [WebAssembly][NFC] Add missing tests for indirect calls.
Wed, Sep 19, 11:20 AM

Tue, Sep 18

aheejin added inline comments to D52105: [WebAssembly] Add V128 value type to binary format.
Tue, Sep 18, 5:31 PM
aheejin accepted D52246: [WebAssembly] v4f32.abs and v2f64.abs.
Tue, Sep 18, 2:06 PM

Fri, Sep 14

aheejin accepted D52130: [WebAssembly][NFC] Generalize operand numbers in SIMD tests.
Fri, Sep 14, 6:06 PM
aheejin accepted D51906: [WebAssembly] SIMD shifts.
Fri, Sep 14, 5:42 PM
aheejin added a comment to D51906: [WebAssembly] SIMD shifts.

Oh, what I meant is, the right operand is created not with insertelement + shufflevector but with just a constant vector whose all elements are the same integer. My comment was not very accurate because I think I remember you included that constant pattern as a splat too. So I guess it would work, but having a test case for that wouldn't hurt.

Fri, Sep 14, 4:13 PM
aheejin added a comment to D51906: [WebAssembly] SIMD shifts.

Does this work when the first argument is not in a splat pattern as well? If so, maybe add a test case for that?

Fri, Sep 14, 3:35 PM
aheejin accepted D52009: [WebAssembly] SIMD neg.
Fri, Sep 14, 3:11 PM
aheejin added inline comments to D52009: [WebAssembly] SIMD neg.
Fri, Sep 14, 2:25 PM
aheejin accepted D52007: [WebAssembly][NFC] Move SIMD encoding tests to dedicated file.
Fri, Sep 14, 2:07 PM
aheejin accepted D52106: [WebAssembly] Add v128 value type.
Fri, Sep 14, 1:54 PM
aheejin added a comment to D52007: [WebAssembly][NFC] Move SIMD encoding tests to dedicated file.

Thanks! Could you please update diff for this CL then so it does not contain D52110?

Fri, Sep 14, 1:41 PM
aheejin accepted D52110: [WebAssembly][NFC] Fix unconventional test names.
Fri, Sep 14, 12:03 PM
aheejin added inline comments to D52106: [WebAssembly] Add v128 value type.
Fri, Sep 14, 11:18 AM

Thu, Sep 13

aheejin added a comment to D52007: [WebAssembly][NFC] Move SIMD encoding tests to dedicated file.

Nit: Can we move the file name changing part to a separate CL?

Thu, Sep 13, 2:01 PM

Wed, Sep 12

aheejin added a comment to D52009: [WebAssembly] SIMD neg.

I think this is dependent on D52007. Fixed it. By the way, the diff looks like it contains D52007's diff too. Could you fix it? You can do relative diff by
arc diff --update D52009 DIVISION_TO_COMPARE

Wed, Sep 12, 7:00 PM
aheejin added a dependent revision for D52007: [WebAssembly][NFC] Move SIMD encoding tests to dedicated file: D52009: [WebAssembly] SIMD neg.
Wed, Sep 12, 6:58 PM
aheejin removed a dependent revision for D5007: [BUG] Fix regression in the instruction combiner after r214385: D52009: [WebAssembly] SIMD neg.
Wed, Sep 12, 6:58 PM
aheejin edited dependencies for D52009: [WebAssembly] SIMD neg, added: 1; removed: 1.
Wed, Sep 12, 6:58 PM
aheejin accepted D51518: [WebAssembly] Add missing SIMD instruction attributes.
Wed, Sep 12, 6:57 PM
aheejin committed rL342084: [WebAssembly] Make tied inline asm operands work again.
[WebAssembly] Make tied inline asm operands work again
Wed, Sep 12, 2:35 PM
aheejin closed D51991: [WebAssembly] Make tied inline asm operands work again.
Wed, Sep 12, 2:35 PM
aheejin added a comment to D51991: [WebAssembly] Make tied inline asm operands work again.

@dschuff Re: about your question asked in person:
If a reg operand gets tied to another operand in instruction selection, they become the same register in here in TwoAddressInstructionPass. So they are assigned to a same local in this ExplicitLocal pass, so we don't need to worry two tied operands ending up in different locals.

Wed, Sep 12, 2:28 PM
aheejin updated the summary of D51991: [WebAssembly] Make tied inline asm operands work again.
Wed, Sep 12, 11:04 AM
aheejin added a comment to D51991: [WebAssembly] Make tied inline asm operands work again.

Without this, this assertion fails.

Wed, Sep 12, 11:03 AM
aheejin added a reviewer for D51991: [WebAssembly] Make tied inline asm operands work again: tlively.
Wed, Sep 12, 11:00 AM
aheejin updated the diff for D51991: [WebAssembly] Make tied inline asm operands work again.
  • Add comment
Wed, Sep 12, 11:00 AM
aheejin created D51991: [WebAssembly] Make tied inline asm operands work again.
Wed, Sep 12, 9:01 AM
aheejin committed rL342037: [clangd] Add missing clangBasic target_link_libraries.
[clangd] Add missing clangBasic target_link_libraries
Wed, Sep 12, 2:43 AM
aheejin committed rCTE342037: [clangd] Add missing clangBasic target_link_libraries.
[clangd] Add missing clangBasic target_link_libraries
Wed, Sep 12, 2:43 AM

Tue, Sep 11

aheejin accepted D51765: [WebAssembly] SIMD comparisons.
Tue, Sep 11, 2:26 PM

Mon, Sep 10

aheejin committed rL341899: [XRay] Add missing LLVMTestingSupport target_link_libraries.
[XRay] Add missing LLVMTestingSupport target_link_libraries
Mon, Sep 10, 9:49 PM
aheejin updated subscribers of D51891: [WebAssembly][NFC] Enable explicit locals in simd-arith.ll.

I'm not actually sure if removing -wasm-disable-explicit-locals or -wasm-keep-registers from test cases is a direction we should head towards, because that generally makes test cases harder to read, as @aardappel D51459. Especially removing -wasm-keep-registers makes writing test cases very hard. Leaving these options as they are and assign 0 as a fake opcode to TEE_LOCAL might be an option too, as I suggested the other day. But that's just how I feel, and I'm not sure if we agreed on gradually removing those options from the test cases. What do you think, @dschuff and @aardappel ?

Mon, Sep 10, 6:37 PM
aheejin added inline comments to D51765: [WebAssembly] SIMD comparisons.
Mon, Sep 10, 6:32 PM
aheejin added a comment to D51891: [WebAssembly][NFC] Enable explicit locals in simd-arith.ll.

How did you resolve it in D51498 then?

Mon, Sep 10, 3:26 PM
aheejin added inline comments to D51765: [WebAssembly] SIMD comparisons.
Mon, Sep 10, 3:23 PM
aheejin added a comment to D51891: [WebAssembly][NFC] Enable explicit locals in simd-arith.ll.

Which aspect of the upcoming CL breaks it?

Mon, Sep 10, 3:18 PM

Fri, Sep 7

aheejin added inline comments to D51765: [WebAssembly] SIMD comparisons.
Fri, Sep 7, 4:20 PM
aheejin added inline comments to D51765: [WebAssembly] SIMD comparisons.
Fri, Sep 7, 3:24 PM
aheejin accepted D51659: [WebAssembly] v8x16.shuffle.
Fri, Sep 7, 2:48 PM
aheejin added inline comments to D51765: [WebAssembly] SIMD comparisons.
Fri, Sep 7, 9:53 AM
aheejin added a comment to D51659: [WebAssembly] v8x16.shuffle.

I'm checking. So according to the current shuffle spec, only half the number of elements from the two operands can be selected, right?

Fri, Sep 7, 9:53 AM
aheejin accepted D51766: [WebAssembly] Change SIMD lane indices to vec_i8imm_op.
Fri, Sep 7, 9:53 AM

Thu, Sep 6

aheejin added inline comments to D51659: [WebAssembly] v8x16.shuffle.
Thu, Sep 6, 3:49 PM

Tue, Sep 4

aheejin committed rL341439: [WebAssembly] clang-format (NFC).
[WebAssembly] clang-format (NFC)
Tue, Sep 4, 6:29 PM
aheejin closed D51447: [WebAssembly] clang-format (NFC).
Tue, Sep 4, 6:28 PM
aheejin updated the diff for D51447: [WebAssembly] clang-format (NFC).

Rebase

Tue, Sep 4, 6:27 PM
aheejin added a comment to D51447: [WebAssembly] clang-format (NFC).

Thank you everyone for comments! I'll land this then. :)

Tue, Sep 4, 6:04 PM
aheejin added a comment to D51447: [WebAssembly] clang-format (NFC).

@sunfish Thanks. I agree that we have to respect the original file's style when we make changes to existing files, especially those that are not exclusive to wasm. But in these wasm-only files we didn't particularly have another style that deviates from LLVM, other than a few places of switch-cases. So I thought it might be possible to ensure clang-formattedness for wasm-exclusive code for now and we can recommend that for future committers as well.

Tue, Sep 4, 6:03 PM
aheejin accepted D51656: [WebAssembly][NFC] Fix formatting and tests.
Tue, Sep 4, 1:58 PM
aheejin added a comment to D51447: [WebAssembly] clang-format (NFC).

Ping. Does anyone have more opinions on this? Do you think I can land this or this is not a good idea?

Tue, Sep 4, 10:30 AM

Fri, Aug 31

aheejin committed rC341254: [WebAssembly] clang-format (NFC).
[WebAssembly] clang-format (NFC)
Fri, Aug 31, 1:58 PM
aheejin committed rL341254: [WebAssembly] clang-format (NFC).
[WebAssembly] clang-format (NFC)
Fri, Aug 31, 1:58 PM
aheejin closed D51448: [WebAssembly] clang-format (NFC).
Fri, Aug 31, 1:58 PM
aheejin closed D51448: [WebAssembly] clang-format (NFC).
Fri, Aug 31, 1:57 PM

Thu, Aug 30

aheejin added a comment to D51448: [WebAssembly] clang-format (NFC).

Ping pong

Thu, Aug 30, 4:55 PM
aheejin added inline comments to D51518: [WebAssembly] Add missing SIMD instruction attributes.
Thu, Aug 30, 4:55 PM
aheejin accepted D51516: [WebAssembly] Update utility functions with SIMD types.
Thu, Aug 30, 3:02 PM
aheejin accepted D51383: [WebAssembly] SIMD loads and stores.
Thu, Aug 30, 2:26 PM
aheejin added inline comments to D51383: [WebAssembly] SIMD loads and stores.
Thu, Aug 30, 2:17 PM
aheejin added a comment to D51448: [WebAssembly] clang-format (NFC).

Ping! Can we land this because this is fairly small?

Thu, Aug 30, 11:33 AM
aheejin added a comment to D51447: [WebAssembly] clang-format (NFC).

I'm totally OK with not landing this if there are any objections. :) But if we are ever gonna pursue clang-formattedness on wasm backend, I think doing it in a single CL that everyone is aware of makes it actually easier than having dozens of reformatting CLs when you trace history back using git blame or something. We can of course not do this, which is totally fine.

Thu, Aug 30, 11:32 AM

Wed, Aug 29

aheejin added inline comments to D51383: [WebAssembly] SIMD loads and stores.
Wed, Aug 29, 6:18 PM
aheejin added inline comments to D51459: [WebAssembly][WIP] Vector conversions.
Wed, Aug 29, 3:09 PM
aheejin committed rL340970: [WebAssembly] clang-format (NFC).
[WebAssembly] clang-format (NFC)
Wed, Aug 29, 2:04 PM
aheejin committed rLLD340970: [WebAssembly] clang-format (NFC).
[WebAssembly] clang-format (NFC)
Wed, Aug 29, 2:04 PM
aheejin closed D51449: [WebAssembly] clang-format (NFC).
Wed, Aug 29, 2:04 PM
aheejin added reviewers for D51448: [WebAssembly] clang-format (NFC): sunfish, dschuff, jgravelle-google.
Wed, Aug 29, 12:48 PM
aheejin added a comment to D51447: [WebAssembly] clang-format (NFC).

Also, among the hundreds of lines in this patch, it looks 80% of them are from switch-cases, when people prefer

switch SOMETHING: break;

over

Wed, Aug 29, 12:47 PM
aheejin added reviewers for D51449: [WebAssembly] clang-format (NFC): sbc100, ruiu.
Wed, Aug 29, 12:44 PM
aheejin added reviewers for D51447: [WebAssembly] clang-format (NFC): sunfish, dschuff, sbc100, jgravelle-google, tlively, aardappel, yurydelendik.
Wed, Aug 29, 12:43 PM