[SYCL][E2E][Matrix] Restrict XFAIL in some tests to DG2 #17639
Conversation
Signed-off-by: Sarnie, Nick <[email protected]>
There was a problem hiding this comment.
@sarnex , are they indeed failing in pre/post-commit? Or are they actually skipped because there is an XFAIL and multiple backends?
What if you just remove XFAIL. Probably, it will just pass in pre/post-commit.
Could you please try that?
Also, are we using different gpu driver versions in nightly and pre/post-commit?
Signed-off-by: Sarnie, Nick <[email protected]>
|
Driver should be the same, testing that now |
ok, based on testing results - passed on PVC, failed on DG2. Hence, xfail needs to be there for 2 tests, but only for DG2 architecture, which makes sense to me (feature is implemented only for PVC/BMG for now). |
Signed-off-by: Sarnie, Nick <[email protected]>
|
done |
|
@sarnex , you may want to update title/description, before merging. Thanks for taking care of it! |
sarnex
changed the title
|
Done, thx |
|
@YuriPlyakhin They're failing on BMG |
Then, pls, add |
Signed-off-by: Sarnie, Nick <[email protected]>
| @@ -18,7 +18,7 @@ | |||
| // This tests support of col major layout for matrix B which does transpose and | |||
| // then VNNI transform. This is currently only available on AMX | |||
|
|
|||
| // XFAIL: gpu | |||
| // XFAIL: gpu-intel-dg2 || arch-intel_gpu_bmg_g21 | |||
| // XFAIL-TRACKER: GSD-5768 | |||
There was a problem hiding this comment.
The BMG part is fine.
But for dg2, we need add:
"
// SG size = 32 is not currently supported for SYCL Joint Matrix by IGC on DG2
// UNSUPPORTED: gpu-intel-dg2
"
There was a problem hiding this comment.
just pushed a new commit that should address this, thanks!
Signed-off-by: Sarnie, Nick <[email protected]>
|
Merging since nightly is broken, happy to address anything post-commit. |
XPASSing in [nightly](https://github.com/intel/llvm/actions/runs/14050841978) but failing in pre/postcommit, make them XFAIL only where they actually fail. --------- Signed-off-by: Sarnie, Nick <[email protected]>
XPASSing in [nightly](https://github.com/intel/llvm/actions/runs/14050841978) but failing in pre/postcommit, make them XFAIL only where they actually fail. --------- Signed-off-by: Sarnie, Nick <[email protected]>
Cherry-pick commits that reached the internal branch between intel/llvm cutoff and release branch pulldown. Patches included: [SYCL][E2E] Fix DeviceLib/assert-windows.cpp run-time errors (#17493) Patch-by: David Garcia Orozco <[email protected]> --- [SYCL] Fix memory leak. (#17632) According to https://github.com/llvm/llvm-project/blob/main/llvm/unittests/Demangle/OutputBufferTest.cpp, `OutputBuffer` has to be manually freed. Patch-by: Marcos Maronas <[email protected]> --- [SYCL][E2E][Matrix] Cleanup of xfail and unsupported (#17688) Patch-by: Dounia Khaldi <[email protected]> --- [SYCL][E2E][Matrix] Restrict XFAIL in some tests to DG2 (#17639) XPASSing in [nightly](https://github.com/intel/llvm/actions/runs/14050841978) but failing in pre/postcommit, make them XFAIL only where they actually fail. Patch-by: Nick Sarnie <[email protected]> --- [SYCL][ESIMD] Move RAII deleter init after buffer allocation. (#17706) The buffer seems to be allocated in the call to `NameNode->print(NameBuf)`, so we need to move the RAII deleter after that call, otherwise the buffer keeps leaking because when the RAII deleter is created, there's no buffer yet. Patch-by: Marcos Maronas <[email protected]> --- [SYCL] Add spill_memory_size unittest (#17657) This adds a unittest for the kernel queries extension according to #17593. Patch-by: Petr Kurapov <[email protected]> --- [SYCL][Bindless][E2E] Add 1D host USM tests (#17374) Adds 1D Host USM backed image tests. This patch also fixes device aspect queries for 1D & 2D USM backed image sampling support. Patch-by: przemektmalon <[email protected]>
XPASSing in nightly but failing in pre/postcommit, make them XFAIL only where they actually fail.