Page MenuHomePhabricator

csigg (Christian Sigg)
User

Projects

User does not belong to any projects.

User Details

User Since
Nov 25 2019, 10:10 AM (130 w, 4 d)

Recent Activity

Wed, May 25

csigg added inline comments to D126199: [MLIR][GPU] Expose GpuParallelLoopMapping as non-test pass..
Wed, May 25, 12:28 PM · Restricted Project, Restricted Project
csigg updated the diff for D126199: [MLIR][GPU] Expose GpuParallelLoopMapping as non-test pass..

Clean up namespacing.

Wed, May 25, 12:27 PM · Restricted Project, Restricted Project
csigg updated the diff for D126158: [MLIR][GPU] Replace fdiv on fp16 with promoted (fp32) multiplication with reciprocal plus one (conditional) Newton iteration..

Rebase.

Wed, May 25, 12:08 PM · Restricted Project, Restricted Project, Restricted Project, Restricted Project
csigg committed rGc4bc416418a2: [LLVM] Add rcp.approx.ftz.f32 intrinsic (authored by csigg).
[LLVM] Add rcp.approx.ftz.f32 intrinsic
Wed, May 25, 12:05 PM · Restricted Project, Restricted Project, Restricted Project
csigg closed D126369: [LLVM] Add rcp.approx.ftz.f32 intrinsic.
Wed, May 25, 12:05 PM · Restricted Project, Restricted Project, Restricted Project
csigg added a comment to D126369: [LLVM] Add rcp.approx.ftz.f32 intrinsic.

Thanks Artem. I think I should be able to land it myself.

Wed, May 25, 12:03 PM · Restricted Project, Restricted Project, Restricted Project
csigg added a comment to D126158: [MLIR][GPU] Replace fdiv on fp16 with promoted (fp32) multiplication with reciprocal plus one (conditional) Newton iteration..

I would suggest separating it into separate LLVM and MLIR patches.

Wed, May 25, 4:42 AM · Restricted Project, Restricted Project, Restricted Project, Restricted Project
csigg requested review of D126369: [LLVM] Add rcp.approx.ftz.f32 intrinsic.
Wed, May 25, 4:42 AM · Restricted Project, Restricted Project, Restricted Project

Mon, May 23

csigg updated the diff for D126199: [MLIR][GPU] Expose GpuParallelLoopMapping as non-test pass..

Comment fix.

Mon, May 23, 6:10 AM · Restricted Project, Restricted Project
csigg requested review of D126199: [MLIR][GPU] Expose GpuParallelLoopMapping as non-test pass..
Mon, May 23, 5:55 AM · Restricted Project, Restricted Project

Sun, May 22

csigg requested review of D126158: [MLIR][GPU] Replace fdiv on fp16 with promoted (fp32) multiplication with reciprocal plus one (conditional) Newton iteration..
Sun, May 22, 1:00 AM · Restricted Project, Restricted Project, Restricted Project, Restricted Project

Sat, May 14

csigg committed rG0e3d1ca54a61: [MLIR][GPU] NFC: simplify kernel operand accessor implementations. (authored by csigg).
[MLIR][GPU] NFC: simplify kernel operand accessor implementations.
Sat, May 14, 5:15 AM · Restricted Project, Restricted Project
csigg closed D125112: [MLIR][GPU] NFC: simplify kernel operand accessor implementations..
Sat, May 14, 5:15 AM · Restricted Project, Restricted Project

Fri, May 6

csigg requested review of D125112: [MLIR][GPU] NFC: simplify kernel operand accessor implementations..
Fri, May 6, 11:33 AM · Restricted Project, Restricted Project

Mon, May 2

csigg accepted D124257: [MLIR][GPU] Add canonicalizer for gpu.memcpy.

Looks good from my side but please address Uday's comments.

Mon, May 2, 10:43 PM · Restricted Project, Restricted Project

Apr 26 2022

csigg added inline comments to D124188: [AsmParser/Printer] Rework sourceloc support for function arguments..
Apr 26 2022, 7:32 AM · Restricted Project, Restricted Project

Apr 22 2022

csigg added a comment to D121279: [MLIR][GPU] Add canonicalizer for gpu.memcpy.

Here is a repro:

Apr 22 2022, 1:03 AM · Restricted Project, Restricted Project

Apr 21 2022

csigg accepted D123499: Add async dependencies support for gpu.launch op.
Apr 21 2022, 1:02 AM · Restricted Project, Restricted Project

Apr 12 2022

csigg added a comment to D123499: Add async dependencies support for gpu.launch op.

async with 0 async dep tokens wouldn't appear to be a meaningful configuration for the op. The lowering does check for it and fails, but should it just be disallowed?

I'm not sure. I would say the semantics are clear if an op uses async (the host does not wait for the op to complete) but no dependencies (it can run immediately without waiting for anything else), and it's OK for the current lowering to be limited in what it can handle and rely on gpu-async-region to bring it into lowering-compatible form. I kind of like the symmetry of these ops (including gpu.wait, where gpu.wait async [] needs to be valid).

Apr 12 2022, 8:01 AM · Restricted Project, Restricted Project
csigg updated subscribers of D123482: Fix CUDA runtime wrapper for GPU mem alloc/free to async.

The CUDA_VERSION comes from the cuda.h header file.

Apr 12 2022, 7:28 AM · Restricted Project, Restricted Project
csigg added inline comments to D121878: [MLIR][GPU] Add canonicalization patterns for folding simple gpu.wait ops..
Apr 12 2022, 6:41 AM · Restricted Project, Restricted Project
csigg abandoned D123603: Fix BUILD.bazel after 83bd4fe2e83c66796afaeb18386de249133c6732..
Apr 12 2022, 6:25 AM · Restricted Project, Restricted Project
csigg requested review of D123603: Fix BUILD.bazel after 83bd4fe2e83c66796afaeb18386de249133c6732..
Apr 12 2022, 6:21 AM · Restricted Project, Restricted Project
csigg accepted D123576: Support constructing empty function_ref from other callables that can be "empty".
Apr 12 2022, 2:08 AM · Restricted Project, Restricted Project
csigg added inline comments to D123576: Support constructing empty function_ref from other callables that can be "empty".
Apr 12 2022, 1:33 AM · Restricted Project, Restricted Project

Apr 11 2022

csigg added a comment to D123499: Add async dependencies support for gpu.launch op.

The gpu-async-region pass simply chains together sequences of gpu ops, with the intention of using async.execute to separate independent work that runs on separate streams. For that case, gpu ops can be synchronous during lowering from higher dialects because the async.execute regions specify which gpu ops should run in sequence and which ones can run in parallel.

Apr 11 2022, 8:11 AM · Restricted Project, Restricted Project
csigg added a comment to D123499: Add async dependencies support for gpu.launch op.

Not opposed to this change at all, but what's the motivation for allowing gpu-async-region to run before gpu-kernel-outlining?

Apr 11 2022, 6:19 AM · Restricted Project, Restricted Project
csigg added a comment to D123482: Fix CUDA runtime wrapper for GPU mem alloc/free to async.

This should be sufficient:

Apr 11 2022, 5:16 AM · Restricted Project, Restricted Project
csigg committed rGa0e1c13caef2: Remove deprecated `parseSourceFile/String()` overloads. (authored by csigg).
Remove deprecated `parseSourceFile/String()` overloads.
Apr 11 2022, 5:11 AM · Restricted Project, Restricted Project
csigg closed D123490: Remove deprecated `parseSourceFile/String()` overloads..
Apr 11 2022, 5:11 AM · Restricted Project, Restricted Project
csigg requested review of D123490: Remove deprecated `parseSourceFile/String()` overloads..
Apr 11 2022, 2:30 AM · Restricted Project, Restricted Project
csigg added inline comments to D121878: [MLIR][GPU] Add canonicalization patterns for folding simple gpu.wait ops..
Apr 11 2022, 12:14 AM · Restricted Project, Restricted Project

Apr 10 2022

csigg accepted D123482: Fix CUDA runtime wrapper for GPU mem alloc/free to async.

This API was introduced in CUDA 11.2 (December 2020).
Should we #if CUDA_VERSION >= 11020 around this?

Apr 10 2022, 11:48 PM · Restricted Project, Restricted Project

Mar 16 2022

csigg added inline comments to D121279: [MLIR][GPU] Add canonicalizer for gpu.memcpy.
Mar 16 2022, 11:49 PM · Restricted Project, Restricted Project
csigg added a comment to D121878: [MLIR][GPU] Add canonicalization patterns for folding simple gpu.wait ops..

Thanks for taking care of this!

Mar 16 2022, 11:40 PM · Restricted Project, Restricted Project

Mar 15 2022

csigg added a comment to D121279: [MLIR][GPU] Add canonicalizer for gpu.memcpy.

Would this fit logically in this revision or a separate revision for gpu.wait canonicalizer? This revision is meant to erase away trivial gpu.memcpy and ancillary stuff. A full-fledged gpu.wait folder/canonicalizer should ideally go into a separate commit.

Mar 15 2022, 2:50 AM · Restricted Project, Restricted Project

Mar 14 2022

csigg added a comment to D121279: [MLIR][GPU] Add canonicalizer for gpu.memcpy.

I think the gpu.wait canonicalizer could be cleaning up more cases:

Mar 14 2022, 12:59 AM · Restricted Project, Restricted Project

Mar 13 2022

csigg committed rG732c39fa3b05: [MLIR][NFC] Fix deprecation message. (authored by csigg).
[MLIR][NFC] Fix deprecation message.
Mar 13 2022, 6:19 AM · Restricted Project

Mar 9 2022

csigg added inline comments to D121279: [MLIR][GPU] Add canonicalizer for gpu.memcpy.
Mar 9 2022, 4:12 AM · Restricted Project, Restricted Project

Mar 8 2022

csigg committed rGdfaadf6b12b8: Update more `parseSourceString()` call sites. (authored by csigg).
Update more `parseSourceString()` call sites.
Mar 8 2022, 4:26 AM · Restricted Project
csigg closed D121102: Update more `parseSourceString()` call sites..
Mar 8 2022, 4:26 AM · Restricted Project, Restricted Project
csigg added a comment to D121102: Update more `parseSourceString()` call sites..

Hello @csigg, there seems to be another use of the deprecated parseSourceString function in mlir/lib/CAPI/IR/IR.cpp that is causing a deprecated error in our builds. Would you mind please updating that use as well? Thanks!

Mar 8 2022, 4:02 AM · Restricted Project, Restricted Project
csigg updated the diff for D121102: Update more `parseSourceString()` call sites..

Fix mlir/lib/CAPI/IR/IR.cpp as well.

Mar 8 2022, 3:59 AM · Restricted Project, Restricted Project

Mar 7 2022

csigg updated the diff for D121102: Update more `parseSourceString()` call sites..

Rebase.

Mar 7 2022, 6:27 AM · Restricted Project, Restricted Project
csigg committed rG4c3c107b54ea: [MLIR] Fix bazel BUILD file after Parser.h moved. (authored by csigg).
[MLIR] Fix bazel BUILD file after Parser.h moved.
Mar 7 2022, 4:20 AM · Restricted Project
csigg closed D121094: [MLIR] Fix bazel BUILD file after Parser.h moved..
Mar 7 2022, 4:20 AM · Restricted Project, Restricted Project
csigg updated the diff for D121094: [MLIR] Fix bazel BUILD file after Parser.h moved..

Rebase.

Mar 7 2022, 4:11 AM · Restricted Project, Restricted Project
csigg updated the diff for D121094: [MLIR] Fix bazel BUILD file after Parser.h moved..

Fix.

Mar 7 2022, 4:04 AM · Restricted Project, Restricted Project
csigg updated the diff for D121094: [MLIR] Fix bazel BUILD file after Parser.h moved..

Fix.

Mar 7 2022, 3:49 AM · Restricted Project, Restricted Project
csigg requested review of D121102: Update more `parseSourceString()` call sites..
Mar 7 2022, 3:43 AM · Restricted Project, Restricted Project
csigg updated the diff for D121094: [MLIR] Fix bazel BUILD file after Parser.h moved..

Fix.

Mar 7 2022, 3:40 AM · Restricted Project, Restricted Project
csigg updated the diff for D121094: [MLIR] Fix bazel BUILD file after Parser.h moved..

And again, more fixes.

Mar 7 2022, 3:38 AM · Restricted Project, Restricted Project
csigg updated the diff for D121094: [MLIR] Fix bazel BUILD file after Parser.h moved..

More fixes.

Mar 7 2022, 3:21 AM · Restricted Project, Restricted Project
csigg updated the summary of D121094: [MLIR] Fix bazel BUILD file after Parser.h moved..
Mar 7 2022, 3:07 AM · Restricted Project, Restricted Project
csigg updated the diff for D121094: [MLIR] Fix bazel BUILD file after Parser.h moved..

Rebase.

Mar 7 2022, 3:03 AM · Restricted Project, Restricted Project
csigg updated the diff for D121094: [MLIR] Fix bazel BUILD file after Parser.h moved..

Add fixes for other file moves.

Mar 7 2022, 2:54 AM · Restricted Project, Restricted Project
csigg updated the diff for D121094: [MLIR] Fix bazel BUILD file after Parser.h moved..

Rebase.

Mar 7 2022, 2:33 AM · Restricted Project, Restricted Project
csigg requested review of D121094: [MLIR] Fix bazel BUILD file after Parser.h moved..
Mar 7 2022, 2:21 AM · Restricted Project, Restricted Project

Mar 6 2022

csigg committed rG0dc66b76fe4c: [MLIR] Change call sites from deprecated `parseSourceFile()` to… (authored by csigg).
[MLIR] Change call sites from deprecated `parseSourceFile()` to…
Mar 6 2022, 9:50 PM · Restricted Project
csigg closed D121075: [MLIR] Change call sites from deprecated `parseSourceFile()` to `parseSourceFile<ModuleOp>()`..
Mar 6 2022, 9:50 PM · Restricted Project, Restricted Project
csigg requested review of D121075: [MLIR] Change call sites from deprecated `parseSourceFile()` to `parseSourceFile<ModuleOp>()`..
Mar 6 2022, 11:56 AM · Restricted Project, Restricted Project

Feb 2 2022

csigg abandoned D118318: Remove `Commutative` interface from `fmin/fmax`.

We have settled on marking addf/mulf commutative as well (D118600) instead of removing it from minf/maxf. Abandoning this revision.

Feb 2 2022, 9:03 AM · Restricted Project

Jan 31 2022

csigg committed rG9b078f8fd26a: [MLIR][arith] Mark addf/mulf as commutative (authored by csigg).
[MLIR][arith] Mark addf/mulf as commutative
Jan 31 2022, 11:34 PM
csigg closed D118600: [MLIR][arith] Mark addf/mulf as commutative.
Jan 31 2022, 11:34 PM · Restricted Project
csigg updated the diff for D118600: [MLIR][arith] Mark addf/mulf as commutative.

Rebase.

Jan 31 2022, 11:33 PM · Restricted Project
csigg updated the diff for D118600: [MLIR][arith] Mark addf/mulf as commutative.

Update tests.

Jan 31 2022, 10:17 PM · Restricted Project
csigg updated the diff for D118600: [MLIR][arith] Mark addf/mulf as commutative.

Rebase.

Jan 31 2022, 10:50 AM · Restricted Project
csigg committed rGf278cf9cbc3e: [MLIR][arith] More float op folders (authored by csigg).
[MLIR][arith] More float op folders
Jan 31 2022, 10:32 AM
csigg closed D118244: [MLIR][arith] More float op folders.
Jan 31 2022, 10:31 AM · Restricted Project
csigg requested review of D118600: [MLIR][arith] Mark addf/mulf as commutative.
Jan 31 2022, 5:17 AM · Restricted Project
csigg retitled D118244: [MLIR][arith] More float op folders from More arith float op folders to [MLIR][arith] More float op folders.
Jan 31 2022, 5:10 AM · Restricted Project
csigg updated the diff for D118244: [MLIR][arith] More float op folders.

Simplify 1.0 matcher, add test for minf(inf, %x) -> %x.

Jan 31 2022, 5:08 AM · Restricted Project
csigg updated the summary of D118244: [MLIR][arith] More float op folders.
Jan 31 2022, 4:49 AM · Restricted Project
csigg updated the diff for D118244: [MLIR][arith] More float op folders.

Also fold minf(%x, +inf) -> %x, fix test.

Jan 31 2022, 4:34 AM · Restricted Project
csigg updated the summary of D118244: [MLIR][arith] More float op folders.
Jan 31 2022, 3:39 AM · Restricted Project
csigg updated the summary of D118244: [MLIR][arith] More float op folders.
Jan 31 2022, 3:39 AM · Restricted Project
csigg updated the diff for D118244: [MLIR][arith] More float op folders.

Rebase.

Jan 31 2022, 3:19 AM · Restricted Project

Jan 28 2022

csigg updated the diff for D118244: [MLIR][arith] More float op folders.

Rebase.

Jan 28 2022, 4:43 AM · Restricted Project

Jan 27 2022

csigg added a comment to D118318: Remove `Commutative` interface from `fmin/fmax`.

I understand all this as "keeping commutative is fine" isn't it?

Jan 27 2022, 11:28 PM · Restricted Project
csigg committed rGf7da4a5d4dee: [NVPTX] Remove fmin/fmax.NaN.f64 again (authored by csigg).
[NVPTX] Remove fmin/fmax.NaN.f64 again
Jan 27 2022, 10:46 PM
csigg closed D118398: [NVPTX] Remove fmin/fmax.NaN.f64 again.
Jan 27 2022, 10:46 PM · Restricted Project
csigg added a comment to D118398: [NVPTX] Remove fmin/fmax.NaN.f64 again.

Is it fine to leave the change in llvm/lib/Target/NVPTX/NVPTXInstrInfo.td as is?

I'd add a comment around FMINNAN that f64 variants do not actually exist. As long as we never lower to it it's not worth the trouble of refactoring tablegen to deal with a minor exception.

Jan 27 2022, 1:36 PM · Restricted Project
csigg updated the summary of D118398: [NVPTX] Remove fmin/fmax.NaN.f64 again.
Jan 27 2022, 1:36 PM · Restricted Project
csigg updated the diff for D118398: [NVPTX] Remove fmin/fmax.NaN.f64 again.

Add comment.

Jan 27 2022, 1:33 PM · Restricted Project
csigg requested review of D118398: [NVPTX] Remove fmin/fmax.NaN.f64 again.
Jan 27 2022, 11:49 AM · Restricted Project
csigg updated the diff for D118126: Correct instcombine of fcmp+select..

Rebase.

Jan 27 2022, 3:14 AM · Restricted Project
csigg added a comment to D118318: Remove `Commutative` interface from `fmin/fmax`.

So it seems that you're assuming that one of the input value is returned when there is a NaN, and it would consistently be based on its position in the argument list. Is this how we define it?
Another definition could be that payload isn't guaranteed to be carried over and that we return a non-specific NaN that can be different from either of the argument.

Jan 27 2022, 2:36 AM · Restricted Project
csigg added a comment to D118318: Remove `Commutative` interface from `fmin/fmax`.

Can you elaborate a bit with an example of why the commutativity isn't there with NaN?

Jan 27 2022, 12:02 AM · Restricted Project

Jan 26 2022

csigg added a comment to D118318: Remove `Commutative` interface from `fmin/fmax`.

Something I'm missing here is why it would be legal to swap operands if the operation isn't commutative.

Jan 26 2022, 11:28 PM · Restricted Project
csigg updated the diff for D118244: [MLIR][arith] More float op folders.

Rebase

Jan 26 2022, 11:06 PM · Restricted Project
csigg added a reviewer for D118318: Remove `Commutative` interface from `fmin/fmax`: mehdi_amini.
Jan 26 2022, 11:05 PM · Restricted Project
csigg added a comment to D118244: [MLIR][arith] More float op folders.

I would even say that it deserve a separate revision from adding folders on simple arithmetic.

Jan 26 2022, 11:04 PM · Restricted Project
csigg requested review of D118318: Remove `Commutative` interface from `fmin/fmax`.
Jan 26 2022, 11:00 PM · Restricted Project
csigg updated the diff for D118244: [MLIR][arith] More float op folders.

Addressing reviewer comments:

Jan 26 2022, 10:58 PM · Restricted Project
csigg added inline comments to D118244: [MLIR][arith] More float op folders.
Jan 26 2022, 8:57 AM · Restricted Project
csigg updated the diff for D118244: [MLIR][arith] More float op folders.

Adding float value matchers and using the same pattern for int value matchers for consistency.

Jan 26 2022, 8:54 AM · Restricted Project
csigg committed rGdc441d776f9f: [NVPTX] NFC: Remove unused arguments and attribute from test (authored by csigg).
[NVPTX] NFC: Remove unused arguments and attribute from test
Jan 26 2022, 6:58 AM
csigg requested review of D118244: [MLIR][arith] More float op folders.
Jan 26 2022, 6:27 AM · Restricted Project

Jan 25 2022

csigg added a reviewer for D118126: Correct instcombine of fcmp+select.: spatel.
Jan 25 2022, 6:53 AM · Restricted Project
csigg requested review of D118126: Correct instcombine of fcmp+select..
Jan 25 2022, 4:46 AM · Restricted Project