Page MenuHomePhabricator

JonChesterfield (Jon Chesterfield)
User

Projects

User does not belong to any projects.

User Details

User Since
Aug 11 2015, 1:08 PM (296 w, 1 d)

Recent Activity

Today

JonChesterfield added inline comments to D100481: [AMDGPU] Disable forceful inline of non-kernel functions which use LDS..
Wed, Apr 14, 2:24 PM · Restricted Project
JonChesterfield added a comment to D99949: [AMDGPU][OpenMP] Add amdgpu-arch tool to list AMD GPUs installed.

Dependence on hsa is not necessary. The amdgpu and nvidia drivers both use PCI codes available in /sys . We should use architecture independent methods as much as possible.

Wed, Apr 14, 2:07 PM · Restricted Project
JonChesterfield accepted D100481: [AMDGPU] Disable forceful inline of non-kernel functions which use LDS..

LG, thanks

Wed, Apr 14, 11:48 AM · Restricted Project
JonChesterfield requested changes to D100481: [AMDGPU] Disable forceful inline of non-kernel functions which use LDS..

Change requested is to not change the handling of region_address. I haven't looked up what that is but I am sure the module lds pass doesn't do anything with it.

Wed, Apr 14, 9:51 AM · Restricted Project
JonChesterfield added a comment to D100481: [AMDGPU] Disable forceful inline of non-kernel functions which use LDS..

I agree with the intent. This would be a good change to run through the internal CI before landing.

Wed, Apr 14, 8:11 AM · Restricted Project
JonChesterfield added a comment to D99949: [AMDGPU][OpenMP] Add amdgpu-arch tool to list AMD GPUs installed.
  1. It does not provide the infrastructure to identify runtime capabilities to satisfy requirements of a compiled image.

I believe we only require a value for '-march=' to unblock running tests on CI machines. I'd guess you're referring to target id stuff where clang fills in reasonable defaults already.

Wed, Apr 14, 7:14 AM · Restricted Project
JonChesterfield accepted D100441: [AMDGPU] Rename "LDS lowering" pass name..

Yep, thanks for factoring this out

Wed, Apr 14, 5:27 AM · Restricted Project

Mon, Apr 12

JonChesterfield added a comment to D99949: [AMDGPU][OpenMP] Add amdgpu-arch tool to list AMD GPUs installed.

New tests look good to me, thanks.

Mon, Apr 12, 4:09 PM · Restricted Project
JonChesterfield added a comment to D100124: [Clang][NVPTX] Add NVPTX intrinsics and builtins for CUDA PTX redux.sync instructions.

Interesting. Reduction across lanes in warp? If so, this is probably a way to handle the last step reduction for openmp reductions

Mon, Apr 12, 3:03 PM · Restricted Project, Restricted Project
JonChesterfield added a comment to D91516: [AMDGPU] Replace uses of LDS globals within non-kernel functions by pointers..

The algorithm I had in mind was along the lines of:

for each LDS variable:
  if should-transform
    create 16 bit integer in LDS
    initialize that global with (constexpr) address of variable
    replace all uses of variable with a (constexpr) access through new pointer
Mon, Apr 12, 7:12 AM · Restricted Project
JonChesterfield added a comment to D91516: [AMDGPU] Replace uses of LDS globals within non-kernel functions by pointers..

This is much more complicated than I expected.
Is the large amount of comments largely from a previous patch doing different things that has been hammered into this one?

No, nothing is hammered from the previous patch. The current patch is implementing what is planned via internal emial discussion.

Mon, Apr 12, 5:27 AM · Restricted Project

Thu, Apr 8

JonChesterfield accepted D100144: [AMDGPU] Allow relaxed/consume memory order for atomic inc/dec.

Nice. Thank you for fixing the oversight.

Thu, Apr 8, 3:41 PM · Restricted Project
JonChesterfield added a comment to D99553: [libomptarget] Read standard notes for ELF offload images.

Note that in D93055 there is still a path for OPENMP_STANDALONE_BUILD. I believe we have not yet abandoned the out-of-tree builds (https://openmp.llvm.org/README.txt). And it seems like a big change to me.

Thu, Apr 8, 9:52 AM · Restricted Project
JonChesterfield added a comment to D99553: [libomptarget] Read standard notes for ELF offload images.

Hi Jon,
Regarding the LLVM component libraries linking into the libomptarget/plugins, I guess the bigger change is that OpenMP offload runtime will not build in out-of-tree mode. Should we actually start with disabling the out-of-tree builds, and see how people react to this?

Thu, Apr 8, 9:33 AM · Restricted Project
JonChesterfield added inline comments to D99949: [AMDGPU][OpenMP] Add amdgpu-arch tool to list AMD GPUs installed.
Thu, Apr 8, 6:28 AM · Restricted Project

Wed, Apr 7

JonChesterfield added a comment to D99553: [libomptarget] Read standard notes for ELF offload images.

Right, we don't really have minutes for the weekly call.

Wed, Apr 7, 12:14 PM · Restricted Project
JonChesterfield added a comment to D99949: [AMDGPU][OpenMP] Add amdgpu-arch tool to list AMD GPUs installed.

We'll have slightly indirect testing once this is used to enable D99656. There are two pieces that can be tested:

Wed, Apr 7, 7:43 AM · Restricted Project
JonChesterfield added a comment to D99553: [libomptarget] Read standard notes for ELF offload images.

Implementing GNU libelf interface as-is via LLVM ELF is problematic because this will require const casting of constant references/pointers exposed by LLVM ELF.

Wed, Apr 7, 7:30 AM · Restricted Project
JonChesterfield added a comment to D99949: [AMDGPU][OpenMP] Add amdgpu-arch tool to list AMD GPUs installed.

I'm happy with this as-is. @jdoerfert is this close enough to what you expected when we discussed this offline?

Wed, Apr 7, 6:59 AM · Restricted Project

Tue, Apr 6

JonChesterfield accepted D99447: [OpenMP] Define omp_is_initial_device() variants in omp.h.

LGTM, thanks!

Tue, Apr 6, 12:31 PM · Restricted Project, Restricted Project
JonChesterfield added a comment to D99949: [AMDGPU][OpenMP] Add amdgpu-arch tool to list AMD GPUs installed.

This change is partly motivated by wanting to check in runtime tests for openmp that execute on whatever hardware is available locally. It is functionally similar to an out of tree bash script called mygpu that contains manually curated tables of pci.ids and to a python script called rocm_agent_enumerator that calls a c++ tool called rocminfo and tries to parse the output, with a different table of pci.ids for when that fails.

Tue, Apr 6, 6:32 AM · Restricted Project

Sat, Apr 3

JonChesterfield added a comment to D99447: [OpenMP] Define omp_is_initial_device() variants in omp.h.

Nice, thanks! I think the function in the devicertl is dead with this change, maybe remove that too?

Sat, Apr 3, 1:09 AM · Restricted Project, Restricted Project

Wed, Mar 31

JonChesterfield added a comment to D99656: [AMDGPU][OpenMP] Enable Libomptarget runtime tests.

I guess there's no XFAIL equivalent here? In that case we should probably leave off the RUN line for the tests that can't work, as they'll otherwise break the build once an amdgpu CI machine goes live

Wed, Mar 31, 7:08 AM · Restricted Project

Tue, Mar 30

JonChesterfield added a reviewer for D99553: [libomptarget] Read standard notes for ELF offload images: JonChesterfield.
Tue, Mar 30, 3:13 AM · Restricted Project
JonChesterfield added a comment to D99553: [libomptarget] Read standard notes for ELF offload images.

That's a lot of code. Mostly straightforward, though the raw new/delete makes me slightly nervous. I think it's almost all here to abstract over libelf or llvm's elf library with a higher level interface. I haven't read through it carefully yet.

Tue, Mar 30, 3:13 AM · Restricted Project

Mon, Mar 29

JonChesterfield added a comment to D94565: [libomptarget][nvptx] Include omp_data.cu in bitcode deviceRTL.

Bug https://bugs.llvm.org/show_bug.cgi?id=49764 points at this change as increasing device memory use. Previously the bitcode and the archive were both linked, where this file was always in the archive. After this commit, no files will have been extracted from the archive (and shortly afterwards the archive was dropped entirely).

Mon, Mar 29, 4:14 PM · Restricted Project
JonChesterfield added a comment to D99297: [OPENMP]Fix PR49636: Assertion `(!Entry.getAddress() || Entry.getAddress() == Addr) && "Resetting with the new address."' failed..

Nice, thanks!

Mon, Mar 29, 8:55 AM · Restricted Project
JonChesterfield resigned from D99347: [AMDGPU] Set implicit arg attributes for indirect calls.

I don't understand why fine grained attributes would be introduced and then all the fine grained attributes set on the function. I also don't understand why setting the attributes on a function based on which they call is considered difficult. We don't seem to be reaching consensus. I'll leave the assessment of this patch to others.

Mon, Mar 29, 8:32 AM · Restricted Project
JonChesterfield accepted D99506: [OpenMP][NFC] Move the `noinline` to the parallel entry point.

Context in https://bugs.llvm.org/show_bug.cgi?id=49752 is that this resolves a regression in stack usage from D94315. This change looks good. I'm not totally sold on using a function call boundary to convey invariants on ICV, but that's an existing property.

Mon, Mar 29, 8:09 AM · Restricted Project

Fri, Mar 26

JonChesterfield added a comment to D99402: [AMDGPU][OpenMP] Add <install-prefix>/include to the search path.

The general problem looks harder but important to fix. Finding the right headers but the wrong shared library is bad, and iirc we currently have to use LD_LIBRARY_PATH to bodge the latter which is not a good UX.

Fri, Mar 26, 4:38 AM · Restricted Project
JonChesterfield added a comment to D55725: [OpenMP] Add libs to clang-dedicated directories.

Arrived here from D99402. Complicated! Perhaps simpler today than in 2019 due to the monorepo, and if I understand the above, somewhat separable.

Fri, Mar 26, 4:37 AM · Restricted Project

Thu, Mar 25

JonChesterfield added inline comments to D99347: [AMDGPU] Set implicit arg attributes for indirect calls.
Thu, Mar 25, 10:48 AM · Restricted Project
JonChesterfield added inline comments to D99347: [AMDGPU] Set implicit arg attributes for indirect calls.
Thu, Mar 25, 9:07 AM · Restricted Project

Tue, Mar 23

JonChesterfield added a reviewer for D91516: [AMDGPU] Replace uses of LDS globals within non-kernel functions by pointers.: JonChesterfield.
Tue, Mar 23, 3:43 AM · Restricted Project
JonChesterfield added a comment to D91516: [AMDGPU] Replace uses of LDS globals within non-kernel functions by pointers..

I can't work out which LDS variables you intend to replace with pointers from the code. Could you spell out what the condition under which you intend to replace one is?

Tue, Mar 23, 3:42 AM · Restricted Project
JonChesterfield added a comment to D91516: [AMDGPU] Replace uses of LDS globals within non-kernel functions by pointers..

This is much more complicated than I expected. Perhaps because it's been written intertwined with the lowering pass, instead of as an optimisation that executes beforehand?

Tue, Mar 23, 12:46 AM · Restricted Project

Mon, Mar 22

JonChesterfield added a reviewer for D98678: [OpenMP][DeviceRT] Remove eager allocation for dynamic schedule handling: ronlieb.

Adding Ron in case he can run this against the amd-stg-open branch, and as fore warning of merge conflicts if not.

Mon, Mar 22, 11:40 AM · Restricted Project
JonChesterfield added a comment to D98678: [OpenMP][DeviceRT] Remove eager allocation for dynamic schedule handling.

I'm not blocking. It doesn't merge cleanly into rocm so it's hard to run against aomp's testing.

Mon, Mar 22, 11:38 AM · Restricted Project
JonChesterfield added a comment to D98678: [OpenMP][DeviceRT] Remove eager allocation for dynamic schedule handling.

I failed to apply this to amd-stg-open to test, but may be able to run qmcpack against llvm main now. Will try that.

What turns out in your try?

Mon, Mar 22, 10:52 AM · Restricted Project

Fri, Mar 19

JonChesterfield abandoned D93135: [libomptarget][devicertl] Port amdgcn devicertl to openmp.

No longer useful as a reference, the good parts of this are in main now.

Fri, Mar 19, 7:27 AM · Restricted Project
JonChesterfield added a comment to D98746: [clang][amdgpu] Use implicit code object default.

Great context, thanks guys. I had missed that part of the compiler.

Fri, Mar 19, 7:19 AM · Restricted Project
JonChesterfield added a comment to D98832: [libomptarget] Tune the number of teams and threads for kernel launch..

...
Agreed. However, I don't see LDS usage in the metadata table in the image. Is it present there?

Fri, Mar 19, 6:31 AM · Restricted Project, Restricted Project

Thu, Mar 18

JonChesterfield added a comment to D98678: [OpenMP][DeviceRT] Remove eager allocation for dynamic schedule handling.

I failed to apply this to amd-stg-open to test, but may be able to run qmcpack against llvm main now. Will try that.

Thu, Mar 18, 5:25 PM · Restricted Project
JonChesterfield added a comment to D98838: [OpenMP] Fixed a crash in hidden helper thread.

Let me be direct for a second so we don't end up here again in a few months:
The patch was on phab for ~1 year, nobody cared, this is a very common phenomena.
It also has been merged for weeks. I get the fact that we want a stable release
but showing up last minute just saying we need to pull stuff is *not* helpful
from an overall perspective. I say this especially because the number of people/
organizations that develop and upstream complex features is very limited. If you
want to benefit from such efforts you should be prepared to help, IMHO. That does
mean to do some testing and reviewing *before* the last release candidate is due.
Not to say this was not tested, but the capabilities are arguably different here.

Thu, Mar 18, 1:46 PM · Restricted Project
JonChesterfield added a comment to D98829: [libomptarget] Add register usage info to kernel metadata.

Ah, apologies. My 'land it' script has clobbered the author field. Will need to look up how to not do that in future

Thu, Mar 18, 10:02 AM · Restricted Project
JonChesterfield committed rG626a31de1521: [libomptarget] Add register usage info to kernel metadata (authored by JonChesterfield).
[libomptarget] Add register usage info to kernel metadata
Thu, Mar 18, 10:01 AM
JonChesterfield closed D98829: [libomptarget] Add register usage info to kernel metadata.
Thu, Mar 18, 10:01 AM · Restricted Project
JonChesterfield added a comment to D98829: [libomptarget] Add register usage info to kernel metadata.

Sure. The process for requesting access has changed since I did it, can't remember what it is now.

Thu, Mar 18, 10:00 AM · Restricted Project
JonChesterfield added a comment to D98838: [OpenMP] Fixed a crash in hidden helper thread.

We should pull this from the 12 release. Lots of effort at the last minute to stop a complicated patch asserting, after it has been patched several times already, is unlikely to yield a stable release.

Thu, Mar 18, 9:57 AM · Restricted Project
JonChesterfield added a comment to D98865: [AMDGPU] Disable forceful inline of non-kernel functions which use LDS..

Looks good to me. This will presumably increase the uses of the lower module lds pass. Maybe prudent to run it through the gerrit CI infra before landing?

Thu, Mar 18, 9:01 AM · Restricted Project
JonChesterfield committed rG253f804debb3: [amdgpu] Update med3 combine to skip i64 (authored by JonChesterfield).
[amdgpu] Update med3 combine to skip i64
Thu, Mar 18, 8:57 AM
JonChesterfield closed D98872: [amdgpu] Update med3 combine to skip i64.
Thu, Mar 18, 8:57 AM · Restricted Project
JonChesterfield added a comment to D98872: [amdgpu] Update med3 combine to skip i64.

Reviewer list derived from git blame + frequency.

Thu, Mar 18, 8:27 AM · Restricted Project
JonChesterfield requested review of D98872: [amdgpu] Update med3 combine to skip i64.
Thu, Mar 18, 8:26 AM · Restricted Project
JonChesterfield added inline comments to D98832: [libomptarget] Tune the number of teams and threads for kernel launch..
Thu, Mar 18, 4:06 AM · Restricted Project, Restricted Project
JonChesterfield added a comment to D98832: [libomptarget] Tune the number of teams and threads for kernel launch..

This is really interesting. The idea seems to be to choose the dispatch parameters based on the kernel metadata and the limits of the machine.

Thu, Mar 18, 4:04 AM · Restricted Project, Restricted Project
JonChesterfield accepted D98829: [libomptarget] Add register usage info to kernel metadata.

Looks good to me, thanks.

Thu, Mar 18, 3:45 AM · Restricted Project

Wed, Mar 17

JonChesterfield added a comment to D74361: [Clang] Undef attribute for global variables.

This appears to have missed a case for openmp. Credit to @jdoerfert for the repro: https://godbolt.org/z/xWTYbv

Wed, Mar 17, 3:13 PM · Restricted Project
JonChesterfield added a reviewer for D75581: [libomptarget][amdgcn] Implement get_wtime: t-tye.
Wed, Mar 17, 9:42 AM · Restricted Project
JonChesterfield added a comment to D98658: [libomptarget] Build amdgcn devicertl by default.

Reverted, D98746 attracted requests for improvement that I don't have a timeline for. This will unblock people building openmp with a clang that doesn't have amdgpu enabled.

Wed, Mar 17, 4:36 AM · Restricted Project
JonChesterfield added a reverting change for rGe23f3502d9a2: [libomptarget] Build amdgcn devicertl by default: rGdbf8f2b089e8: Revert "[libomptarget] Build amdgcn devicertl by default".
Wed, Mar 17, 4:36 AM
JonChesterfield committed rGdbf8f2b089e8: Revert "[libomptarget] Build amdgcn devicertl by default" (authored by JonChesterfield).
Revert "[libomptarget] Build amdgcn devicertl by default"
Wed, Mar 17, 4:36 AM
JonChesterfield added a reverting change for D98658: [libomptarget] Build amdgcn devicertl by default: rGdbf8f2b089e8: Revert "[libomptarget] Build amdgcn devicertl by default".
Wed, Mar 17, 4:35 AM · Restricted Project

Tue, Mar 16

JonChesterfield added inline comments to D98746: [clang][amdgpu] Use implicit code object default.
Tue, Mar 16, 7:21 PM · Restricted Project
JonChesterfield added inline comments to D98746: [clang][amdgpu] Use implicit code object default.
Tue, Mar 16, 5:09 PM · Restricted Project
JonChesterfield added a comment to D77609: [OpenMP] Added the support for hidden helper task in RTL.

One of the drawbacks of limited trunk testing of openmp is that we're reliant on out of trunk people noticing something looks odd. I don't want to set a precedent of downstream forks reverting patches that fail local testing, as that'll remove a bunch of the ad hoc testing we do have.

Tue, Mar 16, 4:49 PM · Restricted Project
JonChesterfield added a comment to D77609: [OpenMP] Added the support for hidden helper task in RTL.

I'm starting to have doubts about the thread safety of this library in general so would lean towards removing the commit entirely such that the remainder is easier to reason about. That way we can be fairly sure we've removed whatever bug this introduced so have ~ one fewer race to try to pin down.

Tue, Mar 16, 4:16 PM · Restricted Project
JonChesterfield added a comment to D98746: [clang][amdgpu] Use implicit code object default.

Thanks. I'm going to wait for some of the rocm people to pass judgement too as this code path is shared with hip / opencl etc.

Tue, Mar 16, 4:05 PM · Restricted Project
JonChesterfield added a comment to D77609: [OpenMP] Added the support for hidden helper task in RTL.

That's three independent reports of stuff breaking after this patch. There are a bunch of locks and condition variables involved, and it looks suspicious to me that the introduced variables are volatile but not atomic.

Tue, Mar 16, 3:58 PM · Restricted Project
JonChesterfield requested review of D98746: [clang][amdgpu] Use implicit code object default.
Tue, Mar 16, 3:45 PM · Restricted Project
JonChesterfield added a comment to D98658: [libomptarget] Build amdgcn devicertl by default.

Do you know how to do so? I'm an hour into cmake documentation, stack overflow and trial and error with zero progress.

Tue, Mar 16, 11:45 AM · Restricted Project
JonChesterfield added a comment to D98658: [libomptarget] Build amdgcn devicertl by default.

There is a failure mode here. If clang is built with an llvm that has the amdgpu target disabled, it will fail to compile this library. This is because clang sets the flag amdhsa-code-object-version for amdgpu, but that flag is defined in llvm, and is excluded from the build if the amdgpu target is disabled. That can be fixed by either not building openmp, or by enabling the target, but the error message the user gets is poor:

Tue, Mar 16, 11:41 AM · Restricted Project
JonChesterfield added a comment to D94648: [amdgpu] Implement lower function LDS pass.

Note to self - there is ongoing interest in minimising the LDS usage of applications. This patch allocates the struct in every kernel (see the call to markUsedByKernel, it is applied exactly once to each kernel), in order to support calls to functions that make use of that struct.

Tue, Mar 16, 4:23 AM · Restricted Project

Mar 15 2021

JonChesterfield committed rGe23f3502d9a2: [libomptarget] Build amdgcn devicertl by default (authored by JonChesterfield).
[libomptarget] Build amdgcn devicertl by default
Mar 15 2021, 4:27 PM
JonChesterfield closed D98658: [libomptarget] Build amdgcn devicertl by default.
Mar 15 2021, 4:27 PM · Restricted Project
JonChesterfield committed rGbb38d7ff05c2: [libomptarget][nfc][amdgcn] Use precise triple for devicertl build (authored by JonChesterfield).
[libomptarget][nfc][amdgcn] Use precise triple for devicertl build
Mar 15 2021, 1:24 PM
JonChesterfield added inline comments to D98658: [libomptarget] Build amdgcn devicertl by default.
Mar 15 2021, 1:19 PM · Restricted Project
JonChesterfield requested review of D98658: [libomptarget] Build amdgcn devicertl by default.
Mar 15 2021, 1:18 PM · Restricted Project
JonChesterfield committed rGd0bc85f04a04: [libomptarget][nfc] Drop unused DEVICE macro (authored by JonChesterfield).
[libomptarget][nfc] Drop unused DEVICE macro
Mar 15 2021, 1:13 PM
JonChesterfield closed D98655: [libomptarget][nfc] Drop unused DEVICE macro.
Mar 15 2021, 1:13 PM · Restricted Project
JonChesterfield committed rG7da76aaaf41e: [libomptarget] Build amdgpu plugin by default (authored by JonChesterfield).
[libomptarget] Build amdgpu plugin by default
Mar 15 2021, 1:12 PM
JonChesterfield closed D98654: [libomptarget] Build amdgpu plugin by default.
Mar 15 2021, 1:12 PM · Restricted Project
JonChesterfield requested review of D98655: [libomptarget][nfc] Drop unused DEVICE macro.
Mar 15 2021, 1:06 PM · Restricted Project
JonChesterfield added a comment to D98651: [libomptarget] Fix devicertl build.

Yep. Also, DEVICE now always expands to whitespace, so that can go.

Code for AMD platform is the only reason we still have DEVICE in common parts. ;-)

Mar 15 2021, 12:58 PM · Restricted Project
JonChesterfield requested review of D98654: [libomptarget] Build amdgpu plugin by default.
Mar 15 2021, 12:57 PM · Restricted Project
JonChesterfield committed rGbcb3f0f867b2: [libomptarget] Fix devicertl build (authored by JonChesterfield).
[libomptarget] Fix devicertl build
Mar 15 2021, 12:53 PM
JonChesterfield closed D98651: [libomptarget] Fix devicertl build.
Mar 15 2021, 12:53 PM · Restricted Project
JonChesterfield added a comment to D98651: [libomptarget] Fix devicertl build.

Yep. Also, DEVICE now always expands to whitespace, so that can go.

Mar 15 2021, 12:49 PM · Restricted Project
JonChesterfield added inline comments to D98651: [libomptarget] Fix devicertl build.
Mar 15 2021, 12:41 PM · Restricted Project
JonChesterfield updated the diff for D98651: [libomptarget] Fix devicertl build.
  • drop accidental amdgcn part
Mar 15 2021, 12:40 PM · Restricted Project
JonChesterfield requested review of D98651: [libomptarget] Fix devicertl build.
Mar 15 2021, 12:39 PM · Restricted Project
JonChesterfield abandoned D96877: [libomptarget] Try a fallback devicertl if the preferred one is missing.

I'm going to abandon this. I'm not confident that a cuda toolkit that is newer than the compiler will work with it correctly and would prefer it take some jury rigging on the end users part to put the two together.

Mar 15 2021, 12:21 PM · Restricted Project
JonChesterfield committed rGf675b3df4848: [libomptarget] Drop assert.h, use freestanding for amdgcn devicertl (authored by JonChesterfield).
[libomptarget] Drop assert.h, use freestanding for amdgcn devicertl
Mar 15 2021, 11:50 AM
JonChesterfield closed D98649: [libomptarget] Drop assert.h, use freestanding for amdgcn devicertl.
Mar 15 2021, 11:50 AM · Restricted Project
JonChesterfield added inline comments to D95752: [OpenMP][DeviceRTL] Extract shuffle idiom and port it to declare variant.
Mar 15 2021, 11:48 AM · Restricted Project
JonChesterfield requested review of D98649: [libomptarget] Drop assert.h, use freestanding for amdgcn devicertl.
Mar 15 2021, 11:21 AM · Restricted Project
JonChesterfield added a comment to D97883: [libomptarget] Add allocator support for target memory.

Apologies for missing this. Looks reasonable to me. I'll wire up the amdgcn plugin when I find some time.

Mar 15 2021, 10:22 AM · Restricted Project
JonChesterfield accepted D96769: [OpenMP][AMDGPU] Skip backend and assemble phases for amdgcn.

Agreed. Lack of save temps is causing grief when debugging, so I keep on applying this patch locally. Let's go with this for now. and change to something better when we think of it.

Mar 15 2021, 10:18 AM · Restricted Project
JonChesterfield committed rG156842937f51: [libomptarget][amdgcn] Drop use of inttypes.h, moving closer to freestanding (authored by JonChesterfield).
[libomptarget][amdgcn] Drop use of inttypes.h, moving closer to freestanding
Mar 15 2021, 9:55 AM
JonChesterfield closed D98565: [libomptarget][amdgcn] Drop use of inttypes.h, moving closer to freestanding.
Mar 15 2021, 9:55 AM · Restricted Project