Hahnfeld (Jonas Hahnfeld)
User

Projects

User does not belong to any projects.

User Details

User Since: Apr 2 2015, 4:52 AM (116 w, 3 d)

IT Center
Group: High Performance Computing
Division: Computational Science and Engineering
RWTH Aachen University

Recent Activity

Mon, Jun 5

Hahnfeld updated subscribers of D33851: [OpenMP] libomptarget: eliminate compiler warnings at build.

Because I've seen this on other revisions as well: Please always subscribe the relevant mailing list!

Mon, Jun 5, 10:56 PM

Wed, May 31

Hahnfeld added a comment to D33227: [OpenMP] libomptarget: move debugging dumps under control of environment variable OMP_TARGET_DEBUG.

In D33227#769276, @sdmitriev wrote:

I can change the name for this environment variable it as you suggested, but it will make it consistent with other environment variable's name that is already used by libomptarget - OMP_TARGET_OFFLOAD (look at openmp/libomptarget/src/omptarget.cpp, line 285).

Wed, May 31, 11:04 PM · Restricted Project

May 10 2017

Hahnfeld updated subscribers of D33028: [OpenMP] Changes in the plugin interface.

Please always add the relevant -commits mailing list

May 10 2017, 5:43 AM · Restricted Project

Hahnfeld accepted D33028: [OpenMP] Changes in the plugin interface.

LGTM. Have you run clang-format on omptargetplugin.h? Some lines look weirdly wrapped...

May 10 2017, 5:42 AM · Restricted Project

May 8 2017

Hahnfeld added a comment to D32734: [CMake][runtimes] Set default directory for runtime libraries.

In D32734#748977, @beanz wrote:

What is the problem you're trying to solve here? This seems odd to me.

May 8 2017, 11:12 PM

May 4 2017

Hahnfeld added a comment to D29647: [OpenMP] Extend CLANG target options with device offloading kind..

One minor drive-by comment. I think there is still one outstanding from Justin...

May 4 2017, 10:39 AM

May 3 2017

Hahnfeld committed rL302118: [libcxxabi][CMake] Set library dir when built under runtimes/.

[libcxxabi][CMake] Set library dir when built under runtimes/

May 3 2017, 11:17 PM

Hahnfeld closed D32735: [libcxxabi][CMake] Set library dir when built under runtimes/ by committing rL302118: [libcxxabi][CMake] Set library dir when built under runtimes/.

May 3 2017, 11:17 PM

Hahnfeld committed rL302117: [libcxx][CMake] Set library dir when built under runtimes/.

[libcxx][CMake] Set library dir when built under runtimes/

May 3 2017, 11:16 PM

Hahnfeld closed D32736: [libcxx][CMake] Set library dir when built under runtimes/ by committing rL302117: [libcxx][CMake] Set library dir when built under runtimes/.

May 3 2017, 11:16 PM

May 2 2017

Hahnfeld created D32736: [libcxx][CMake] Set library dir when built under runtimes/.

May 2 2017, 12:29 AM

Hahnfeld created D32735: [libcxxabi][CMake] Set library dir when built under runtimes/.

May 2 2017, 12:28 AM

Hahnfeld created D32734: [CMake][runtimes] Set default directory for runtime libraries.

May 2 2017, 12:28 AM

Apr 25 2017

Hahnfeld accepted D32321: [OpenMP] Optimized default kernel launch parameters in CUDA plugin.

LGTM

Apr 25 2017, 9:41 AM · Restricted Project

Apr 22 2017

Hahnfeld accepted D32326: [OpenMP] libomptarget: Set ref count for global objects to positive infinity.

LGTM

Apr 22 2017, 4:48 AM · Restricted Project

Apr 20 2017

Hahnfeld added a comment to D32321: [OpenMP] Optimized default kernel launch parameters in CUDA plugin.

Does this change result in a lower runtime? Last time I tested clang-ykt on Pascal GPUs, 1024 threads were really the best thing to do...

Apr 20 2017, 11:15 PM · Restricted Project

Hahnfeld added inline comments to D32326: [OpenMP] libomptarget: Set ref count for global objects to positive infinity.

Apr 20 2017, 11:08 PM · Restricted Project

Hahnfeld accepted D32325: [OpenMP] libomptarget: Remove obsolete negative device IDs -2/-3.

LGTM, not currently used in Clang.

Apr 20 2017, 10:55 PM · Restricted Project

Hahnfeld accepted D29658: [OpenMP] Customize CUDA-based tool chain selection.

LGTM with one small note

Apr 20 2017, 2:13 AM

Hahnfeld accepted D29904: [OpenMP] Prevent emission of exception handling code when using OpenMP to offload to NVIDIA devices..

LGTM unless Alexey still has objections

Apr 20 2017, 1:57 AM

Apr 19 2017

Hahnfeld committed rL300689: [Driver] Unify linking of OpenMP runtime. NFCI..

[Driver] Unify linking of OpenMP runtime. NFCI.

Apr 19 2017, 7:08 AM

Hahnfeld closed D30087: [Driver] Unify linking of OpenMP runtime. NFCI. by committing rL300689: [Driver] Unify linking of OpenMP runtime. NFCI..

Apr 19 2017, 7:08 AM

Apr 18 2017

Hahnfeld added a comment to D30087: [Driver] Unify linking of OpenMP runtime. NFCI..

Ping

Apr 18 2017, 11:40 PM

Apr 13 2017

Hahnfeld added a comment to D29658: [OpenMP] Customize CUDA-based tool chain selection.

Is there a way to actually test the changed code? The current test changes seem unrelated...

Apr 13 2017, 8:09 AM

Apr 6 2017

Hahnfeld added inline comments to D30087: [Driver] Unify linking of OpenMP runtime. NFCI..

Apr 6 2017, 7:50 AM

Hahnfeld updated the diff for D30087: [Driver] Unify linking of OpenMP runtime. NFCI..

Apr 6 2017, 7:49 AM

Apr 3 2017

Hahnfeld added a comment to D30087: [Driver] Unify linking of OpenMP runtime. NFCI..

Ping!

Apr 3 2017, 11:17 PM

Hahnfeld committed rL299415: Align all scalar numbers to LLVM_YAML_IS_FLOW_SEQUENCE_VECTOR.

Align all scalar numbers to LLVM_YAML_IS_FLOW_SEQUENCE_VECTOR

Apr 3 2017, 11:15 PM

Hahnfeld closed D30508: Align all scalar numbers to LLVM_YAML_IS_FLOW_SEQUENCE_VECTOR by committing rL299415: Align all scalar numbers to LLVM_YAML_IS_FLOW_SEQUENCE_VECTOR.

Apr 3 2017, 11:15 PM

Hahnfeld added reviewers for D30508: Align all scalar numbers to LLVM_YAML_IS_FLOW_SEQUENCE_VECTOR: sbc100, davide.

Ping! This gets pretty annoying...

Apr 3 2017, 12:23 AM

Hahnfeld updated the diff for D30508: Align all scalar numbers to LLVM_YAML_IS_FLOW_SEQUENCE_VECTOR.

Catch one more occurence.

Apr 3 2017, 12:22 AM

Mar 31 2017

Hahnfeld requested changes to D29658: [OpenMP] Customize CUDA-based tool chain selection.

Mar 31 2017, 2:48 AM

Hahnfeld added a reviewer for D29654: [OpenMP] Integrate OpenMP target region cubin into host binary: rnk.

Please format all comments as full sentences.

Mar 31 2017, 2:33 AM

Hahnfeld accepted D29339: [OpenMP] Add support for auxiliary triple specification.

LGTM

Mar 31 2017, 1:37 AM

Hahnfeld requested changes to D29904: [OpenMP] Prevent emission of exception handling code when using OpenMP to offload to NVIDIA devices..

Out of interest: This is a limition of the implementation, right? Because the standard only says: A throw executed inside a target region must cause execution to resume within the same target region, and the same thread that threw the exception must catch it.

Mar 31 2017, 12:07 AM

Mar 30 2017

Hahnfeld added a comment to D29644: [OpenMP] Pass -v to PTXAS if it was passed to the driver..

Just noticed that this isn't specific to OpenMP. Could you move the test to Driver/cuda-external-tools.cu where the rest of ptxas is tested?

Mar 30 2017, 11:34 PM

Hahnfeld accepted D29642: [OpenMP] Make OpenMP generated code for the NVIDIA device relocatable by default.

LGTM after fixing the check-prefix

Mar 30 2017, 11:25 PM

Mar 29 2017

Hahnfeld added a comment to D29904: [OpenMP] Prevent emission of exception handling code when using OpenMP to offload to NVIDIA devices..

LGTM. Please run clang-format before committing!

Mar 29 2017, 11:08 AM

Hahnfeld added a comment to D29659: [OpenMP] Add flag for disabling the default generation of relocatable OpenMP target code for NVIDIA GPUs..

Having something with dashes behind other used prefixes (CHK-PTXAS, CHK-PTXAS-C, CHK-PTXAS-C-RELO) might not be optimal and break when other suffixes like -NOT or -SAME are added to lit. Please see inline about my suggestions but feel free to use others...

Mar 29 2017, 11:03 AM

Hahnfeld added inline comments to D29644: [OpenMP] Pass -v to PTXAS if it was passed to the driver..

Mar 29 2017, 11:02 AM

Hahnfeld accepted D29645: [OpenMP] Pass -fopenmp-is-device to preprocessing and machine specific code generation stages.

LGTM

Mar 29 2017, 10:45 AM

Hahnfeld added inline comments to D29642: [OpenMP] Make OpenMP generated code for the NVIDIA device relocatable by default.

Mar 29 2017, 10:42 AM

Mar 27 2017

Hahnfeld updated the diff for D30508: Align all scalar numbers to LLVM_YAML_IS_FLOW_SEQUENCE_VECTOR.

Rebase and ping!

Mar 27 2017, 11:39 PM

Hahnfeld updated the diff for D30087: [Driver] Unify linking of OpenMP runtime. NFCI..

Rebase and ping.

Mar 27 2017, 11:38 PM

Mar 23 2017

Hahnfeld added inline comments to D31071: GOMP compatibility: add missing OMP4.0 taskdeps handling code.

Mar 23 2017, 4:08 AM

Mar 22 2017

Hahnfeld accepted D31206: [OpenMP] CUDA plugin: More descriptive error messages.

LGTM

Mar 22 2017, 10:41 AM · Restricted Project

Hahnfeld accepted D31205: [OpenMP] Allow multiple weak symbols to be loaded from the fat binary.

LGTM

Mar 22 2017, 12:53 AM · Restricted Project

Hahnfeld added inline comments to D31206: [OpenMP] CUDA plugin: More descriptive error messages.

Mar 22 2017, 12:52 AM · Restricted Project

Hahnfeld accepted D31207: [OpenMP] CUDA plugin: add include directory for libelf.

LGTM

Mar 22 2017, 12:49 AM · Restricted Project

Mar 21 2017

Hahnfeld accepted D30739: [OpenMP] "declare simd" for AArch64 Advanced SIMD..

In D30739#706304, @fpetrogalli wrote:

AFAIK, none of the machinery required in LLVM to expose the mangled names in the vectorizer is present in trunk. There is a patch for x86 under review that does that: https://reviews.llvm.org/D22792

Mar 21 2017, 6:18 AM

Hahnfeld added a comment to D30739: [OpenMP] "declare simd" for AArch64 Advanced SIMD..

In principal looks good to me although I'm not really familiar with this part. Does that work for you if you have the declare simd in a header file and the implementation in another file? On x86_64 I currently get:

remark: loop not vectorized: call instruction cannot be vectorized

Mar 21 2017, 5:48 AM

Hahnfeld accepted D31055: [OpenMP] libomptarget: Disable on MacOS X.

I thought the library would just compile fine and that the CUDA plugin should work without ELF support - turns out both don't because the CUDA plugin uses ELF to verify the binary header. This could certainly be solved differently but that's future work if someone is interested.

Mar 21 2017, 12:18 AM

Mar 20 2017

Hahnfeld added a comment to D30508: Align all scalar numbers to LLVM_YAML_IS_FLOW_SEQUENCE_VECTOR.

Ping

Mar 20 2017, 2:13 AM

Hahnfeld added a comment to D30087: [Driver] Unify linking of OpenMP runtime. NFCI..

Another ping

Mar 20 2017, 2:13 AM

Hahnfeld abandoned D30733: [Driver] Add arch-specific rpath for libc++.

Mar 20 2017, 12:10 AM

Hahnfeld requested changes to D31055: [OpenMP] libomptarget: Disable on MacOS X.

Mar 20 2017, 12:09 AM

Mar 17 2017

Hahnfeld added a comment to D31055: [OpenMP] libomptarget: Disable on MacOS X.

Ideally, we should check if the linker supports --version-script, but currently this yields no benefit.

Mar 17 2017, 12:09 AM

Mar 9 2017

Hahnfeld added a comment to D30733: [Driver] Add arch-specific rpath for libc++.

In D30733#697108, @jroelofs wrote:

As I said on D30214, it is inappropriate to be installing libc++ in the resource directory... please do not do that.

Mar 9 2017, 10:55 PM

Mar 8 2017

Hahnfeld created D30733: [Driver] Add arch-specific rpath for libc++.

Mar 8 2017, 4:32 AM

Hahnfeld added a comment to D30700: [Driver] Add flag to request arch-specific-subdir in -rpath.

No build system will ever set -frtlib-add-rpath to enable this "feature". I'm for keeping this opt-out until we have configuration files to set this by default. Making it opt-in would weaken its main reason of existence: Not to break simple binaries for the user, and we can just drop it.

Mar 8 2017, 4:31 AM

Hahnfeld committed rL297271: [XRay] Use AddCXXStdlibLibArgs. NFCI..

[XRay] Use AddCXXStdlibLibArgs. NFCI.

Mar 8 2017, 4:18 AM

Hahnfeld updated the diff for D30087: [Driver] Unify linking of OpenMP runtime. NFCI..

Rebase for recent refactoring and ping.

Mar 8 2017, 1:50 AM

Hahnfeld committed rL297264: [test] Unbreak OpenMP/linking.c with arch-specific libdir.

[test] Unbreak OpenMP/linking.c with arch-specific libdir

Mar 8 2017, 1:19 AM

Hahnfeld updated the diff for D30508: Align all scalar numbers to LLVM_YAML_IS_FLOW_SEQUENCE_VECTOR.

Rebase and adapt test added meanwhile.

Mar 8 2017, 12:55 AM

Hahnfeld committed rL297260: [Support] Remove unit test for fs::is_local.

[Support] Remove unit test for fs::is_local

Mar 8 2017, 12:48 AM

Mar 7 2017

Hahnfeld added a comment to D30470: [cmake] Include openmp with add_llvm_external_project.

In D30470#694814, @jlpeyton wrote:

Yes, OpenMP would be under the runtimes/ directory.

Mar 7 2017, 11:13 PM

Mar 2 2017

Hahnfeld abandoned D30214: [Driver] Search for libc++ headers in ResourceDir.

http://lists.llvm.org/pipermail/cfe-dev/2017-January/052512.html suggested to install runtime libraries (like libomp, but I think libc++ in certain cases may fall under the same category) to ResourceDir and I wanted the headers to be co-located. However, I can live with installing them to normal include

Mar 2 2017, 11:21 PM

Hahnfeld updated the diff for D30214: [Driver] Search for libc++ headers in ResourceDir.

Adapt and add tests.

Mar 2 2017, 1:03 AM

Hahnfeld updated the diff for D30087: [Driver] Unify linking of OpenMP runtime. NFCI..

Reword comment

Mar 2 2017, 1:01 AM

Mar 1 2017

Hahnfeld added a comment to D30450: OpenMP version 5.0 added.

In D30450#689560, @AndreyChurbanov wrote:

(1) We had examples of changed semantics of omp_in_parallel() API from OpenMP 2.0 --> 2.5. Recent incompatibility of OpenMP 4.0 and 4.5 is probably pure compiler related, but anyway, codes written for OpenMP 4.0 don't work for OpenMP 4.5 (because of changed implicit data sharing attribute of scalars in the target construct).
(2) It can be worthwhile to be able to build library for various OpenMP versions support, e.g. for matching the version supported by compiler the library is built for. It looks inelegant and confusing for customers when compiler and library declare different supported versions.

Just my 2 cents.

Mar 1 2017, 11:51 PM

Hahnfeld added a comment to D30214: [Driver] Search for libc++ headers in ResourceDir.

In D30214#690010, @jroelofs wrote:

libc++ headers should not be installed in the resource dir.

Mar 1 2017, 11:48 PM

Hahnfeld added a comment to D30214: [Driver] Search for libc++ headers in ResourceDir.

Ping

Mar 1 2017, 1:58 PM

Hahnfeld added a comment to D30087: [Driver] Unify linking of OpenMP runtime. NFCI..

Ping

Mar 1 2017, 1:57 PM

Hahnfeld created D30508: Align all scalar numbers to LLVM_YAML_IS_FLOW_SEQUENCE_VECTOR.

Mar 1 2017, 1:57 PM

Hahnfeld added a comment to D30450: OpenMP version 5.0 added.

In D30450#689455, @omalyshe wrote:

In D30450#688282, @Hahnfeld wrote:

Do we really need this? In other words: Is there a use case for not building the OpenMP runtime for the latest standard?

For example, if it is supposed to use the runtime library with the compiler which does not support the latest standard.

Mar 1 2017, 3:21 AM

Feb 28 2017

Hahnfeld added a comment to D30450: OpenMP version 5.0 added.

Do we really need this? In other words: Is there a use case for not building the OpenMP runtime for the latest standard?

Feb 28 2017, 2:10 AM

Feb 21 2017

Hahnfeld committed rL295808: [OpenMP] Generate better diagnostics for cancel and cancellation point.

[OpenMP] Generate better diagnostics for cancel and cancellation point

Feb 21 2017, 11:01 PM

Hahnfeld closed D30135: [OpenMP] Generate better diagnostics for cancel and cancellation point by committing rL295808: [OpenMP] Generate better diagnostics for cancel and cancellation point.

Feb 21 2017, 11:01 PM

Hahnfeld created D30214: [Driver] Search for libc++ headers in ResourceDir.

Feb 21 2017, 10:16 AM

Feb 20 2017

Hahnfeld updated the diff for D30087: [Driver] Unify linking of OpenMP runtime. NFCI..

rebase

Feb 20 2017, 6:10 AM

Hahnfeld updated the diff for D30135: [OpenMP] Generate better diagnostics for cancel and cancellation point.

Address review comment's and apply new naming style to checkNestingOfRegions

Feb 20 2017, 4:48 AM

Hahnfeld added inline comments to D30135: [OpenMP] Generate better diagnostics for cancel and cancellation point.

Feb 20 2017, 4:02 AM

Hahnfeld added inline comments to D30135: [OpenMP] Generate better diagnostics for cancel and cancellation point.

Feb 20 2017, 3:48 AM

Hahnfeld updated the diff for D30135: [OpenMP] Generate better diagnostics for cancel and cancellation point.

new static function CheckCancelRegion

Feb 20 2017, 3:12 AM

Hahnfeld added a comment to D30135: [OpenMP] Generate better diagnostics for cancel and cancellation point.

In D30135#681354, @ABataev wrote:

Not sure that this is better because at first, we need to be sure that this nesting is allowed. Why do we need to perform some additional analysis if nesting is not allowed at all?

Feb 20 2017, 1:32 AM