i applied the patch to a gfx908 and a gfx906 system.
tests all passed , and the gfx906 should be the same as the amdgpu buildbot.

@doru1004 is interested in these tests being enabled.
thx for adding them.

In case folks are worried about this debug content:
this patch will have a short life span, once we see the dump from the BOT, we can revert it.

when the test detects the error, can we dump out all the arrays data1, data2, data3, and sum . and then break ...

LGTM

also seeing similar error in downstream windows build of llvm.

had not heard back , so reverted it for now
commit 8cf85a0cadb033fed3d96aa5283deb4bfbbaf2c8 (HEAD -> main, origin/main, origin/HEAD)
Author: Ron Lieberman <ron.lieberman@amd.com>
Date: Wed Mar 8 22:01:22 2023 -0600

seems to fail amdgpu buildbot
https://lab.llvm.org/buildbot/#/builders/193/builds/27692

LGTM, as this resolves our downstream assertions seen in our CI

LIT Test ?

i reverted to unblock our bot

Hi Jonathon,
our amdgpu buildbot seems to have been broken by this patch
https://lab.llvm.org/buildbot/#/builders/193/builds/26173
could you fix quickly? or revert and fix ?

Hi Janek, seeing buildbot failures: https://lab.llvm.org/buildbot/#/builders/193/builds/26047

hi, breaks our amdgpu buildbot, please fix ASAP , or revert and fix. thx
3 patches: https://lab.llvm.org/buildbot/#/builders/193/builds/25299

our nekbone app is failing from this patch.

Thanks DOru, looks way better than my botched commit

sorry about that, reverted.
will push a different fix now.

it does look simple :-) i pushed in a fix a while ago, (hour or so)

hi . bbot failure related to this patch https://lab.llvm.org/buildbot/#/builders/193/builds/24271

lgtm, passed the 5 internal tests that failed.

@JonChesterfield i am seeing the testcase bug50022.cpp occasionally randomly failing in amdgpu bbot
most recently: https://lab.llvm.org/buildbot/#/builders/193/builds/24230

seems like this broke the amdgpu buildbot , plz resolve
https://lab.llvm.org/buildbot/#/builders/193/builds/24122

Hi, amdgpu buildbot failing this test after this patch landed

libomptarget :: amdgcn-amd-amdhsa :: api/omp_dynamic_shared_memory_amdgpu.c

@jhuber6 please help with this one ? thanks

hello , this commit is breaking our buildbot : could you address please ? https://lab.llvm.org/buildbot/#/builders/193

In D135402#3981172, @ronlieb wrote:

this patch when applied on top of a previous successful build, passes.
however, if i start with a fresh build directory and cmake/build it fails.
normally our amdgpu buildbot would not notice.

buildbot https://lab.llvm.org/staging/#/builders/183/builds/2704

unfortunately our buildbot died 3-4 days ago, and had to be reimaged yesterday and is still in staging.
when i restarted the buildbot yesterday it picked up this change and started failing.
I am not sure why patch from JPeyton did not resovle this. looking into... @jhuber6 ?

Joseph Huber pushed in a fix that gets our bot to go green. thx
[OpenMP][libomp] Fix version scripts after undefined version script changes

this patch when applied on top of a previous successful build, passes.
however, if i start with a fresh build directory and cmake/build it fails.
normally our amdgpu buildbot would not notice.

buildbot https://lab.llvm.org/staging/#/builders/183/builds/2704

also breaks amdgpu hip and openmp buildbots

lets hold off landing for a few more days?
checking some internal related testing results

i applied the lld patch https://reviews.llvm.org/D135402 and this one, then ran the equivalent of our buildbot, all good, builds and passes.

thx for quick fix, much appreciated. love the color green!

hello, couple of our bots broke:
amdgpu hip bot:
https://lab.llvm.org/buildbot/#/builders/165/builds/25535

I am seeing intermittent failures with the new test case on our amdgpu buildbot

FAIL: ompd-test::ompd_parallel.c

Please revert this change, so our bots can go green.
we are struggling to figure out how to fix openmp/runtime

my build systems have ROCm installed, which might make a difference.
since i can reproduce the issue locally

ld.lld: error: openmp/runtime/src/CMakeFiles/omp.dir/kmp_csupport.cpp.o: symbol ompc_set_affinity_format@@VERSION has undefined version VERSION

This cmake recipe reproduces issue for me:

https://lab.llvm.org/buildbot/#/builders/193/builds/21412

amdgpu buildbot sees issues.

[AMD Official Use Only - General]

we are seeing the same issue compiling compiler-rt with these patches in trunk, are the various buildbots catching this also?
for awareness: @saiislam

hi, our amdgpu buildbot which uses lld has been failing since this commit landed
https://lab.llvm.org/buildbot/#/builders/193/builds/20055

hi, seeing our amdgpu buildbot broken with this patch. please look into ?
let me know if you any help on our end.
https://lab.llvm.org/buildbot/#/builders/193/builds/19744

LGTM

seeing a buildbot failure on amdgpu bot
https://lab.llvm.org/buildbot/#/builders/193/builds/18788

emailed a hopeful reproducer to you

hopefully will fix this ?
invalid cast kind for complex value
UNREACHABLE executed at /clang/lib/AST/ExprConstant.cpp:14263!
PLEASE submit a bug report to https://github.com/llvm/llvm-project/issues/ and include the crash backtrace, preprocessed source, and as
sociated run script.

i am seeing similar issues in openmp buildbot failing with this patch

In D132408#3773836, @fhahn wrote:

In D132408#3773174, @yaxunl wrote:

In D132408#3772225, @yaxunl wrote:

In D132408#3772184, @fhahn wrote:

In D132408#3757270, @yaxunl wrote:

In D132408#3756337, @nikic wrote:

Do you have any numbers for the impact of this change? I suspect (but haven't checked) that it may be quite significant, because the current tiny bonus inst threshold (1) is tuned for the current implementation, and this patch will result in much less common dest folding than would be desirable.

For a synthetic benchmark geared towards loops with small workloads, we saw around 16% performance gain due to avoiding always executing 31 extra load instructions by not folding 31 basic blocks coming from an unrolled loop.

For typical HIP programs, I did not see significant performance differences before and after this change. This is because the original bonus instruction threshold 2 is very small. It only allows folding small basic blocks which only contain one instruction other than the cmp/branch instructions. In practical applications, such basic blocks are rare and usually are not performance bottlenecks. Only in special situations (e.g. a fully unrolled loop of large loop counts ends up with a large number of foldable basic blocks) the performance penalty becomes significant.

I am not sure what typical HIP programs mean here. Did you measure the impact on general CPU benchmarks?

No. Can you recommend a free CPU benchmark that is suitable for this purpose? Thanks.

I will try running the SPEC CPU benchmark. There are many sub-tests. Do you recommend any sub-tests that are most relevant to this patch? Thanks.

For SPEC, I usually run SPEC2017rate (both integer and fp subsets).

seeing buildbot failures on hip, not sure if this patch caused it

thanks for the analysis , i agree its a coding error in our library.

hi, latest commit seems to have broken buildbot for amdgpu
https://lab.llvm.org/buildbot/#/builders/193/builds/16651

@FabioLuporini might find this interesting, memory leak.

In D131182#3700502, @tianshilei1992 wrote:

Will push a change to fix the printf issue on AMDGPU.

thx for fix.

i am seeing the same assert on the amdgpu target

Hi Peter

@saiislam for awareness

i forwarded a couple buildbot failures to Amit to look into

i pushed it in. will watch for build bot issues