Conversation
Reworks the stream implementation for `/gpu/hip/gen` to avoid creating and destroying streams on every operator apply. Updates `hipblas` calls to only sync stream, this matters on MI300A since `hipblas` seems to use an async stream. Avoids a full device sync. Also makes working vectors come from the `Vector` object delegate to avoid bad ref behavior.
* Fixes for oneAPI-2025 * Neat fix for oneAPI 2025 * style changes
ex - add PETSc BP1+3 and BP2+4 examples
test - fix path to stay in folder
rust - update rust examples to match c/python
Use HTML comments as doc inclusion markers
Add ncomp=4 as common CPU restriction variant
Co-authored-by: Peter Munch <peterrmuench@gmail.com>
ex - use deal.II fe to build libCEED Basis
cpu - remove vectorize warnings for Clang -O3
Layla name change
* doc - improve internal API documentation * doc - clarify delegation * doc - update wording for clarity Co-authored-by: James Wright <james@jameswright.xyz> * doc - clarify memcheck==valgrind * doc - clarify shared gpu backends * doc - clarify l/e/q-vecs * doc - clarify occa backend status * doc - more inheritance clarification * doc - minor claification to QF fields * minor - remove unneeded variables * doc - update inheritance language * doc - update shared description * doc - more gpu notes * doc - update fallback description * doc - clarify fallback is only for pc support * doc - minor updates to user facing GPU section * doc - minor reordering of major sections --------- Co-authored-by: James Wright <james@jameswright.xyz>
minor - typo
interface: fix uninitialized use in CeedOperatorMultigridLevelCreateSingle_Core
cov - add missing exclusion markers
Better flags for cov accuracy on Noether
ci - add missing exclusion markers
* contributing - draft LLM statement * contributing - deleniate permitted and disallowed LLM usage * contributing - grammar Co-authored-by: Valeria Barra <39932030+valeriabarra@users.noreply.github.com> * dev - add PR template * contributing - wording consistency between PR template and CONTRIBUTING.md * minor - typo * contributing - refrence sf dora * minor - fix spelling Co-authored-by: Yohann <dudouit1@llnl.gov> --------- Co-authored-by: Valeria Barra <39932030+valeriabarra@users.noreply.github.com> Co-authored-by: Yohann <dudouit1@llnl.gov>
dealii CI Error
rust - allow setting ceed OPT flags for rust, mostly for CI/cov
pvelesko
force-pushed
the
chipStar
branch
2 times, most recently
from
d2d8709 to
a65eb92
Compare
Detect HIP platform at build time via hipconfig output: - __HIP_PLATFORM_SPIRV__ → HIP_LIB_NAME=CHIP (chipStar) - __HIP_PLATFORM_HCC__/__HIP_PLATFORM_AMD__ → HIP_LIB_NAME=amdhip64 Move ROCM_DIR and HIP_ARCH defaults to the top of the file where other tool-path defaults live. Use HIP_LIB_NAME in the library detection glob and in PKG_LIBS. Remove the subst=,, stripping from HIPCONFIG_CPPFLAGS so flags are passed through unmodified.
When SYCL backends are built, libceed.so must be linked with icpx (SYCLCXX) rather than g++, and -fsycl must appear in CEED_LDFLAGS (before object files) so icpx can merge the SYCL fat binary device sections. Without this, libceed.so lacks NEEDED: libsycl.so.7 and SYCL kernels fail to load at runtime.
Replace the outer element for-loop with a single element assignment and guard all memory accesses with if (elem < num_elem). Shared memory operations (Interp, Grad, etc.) must execute unconditionally across all threads so __syncthreads() is reached uniformly; only the load/store steps are guarded. Also guard qfunction calls with thread-id bounds checks and comment out the pragma unroll that triggered miscompilation on chipStar's LLVM.
pvelesko
force-pushed
the
chipStar
branch
2 times, most recently
from
6a5b77b to
e8e7c92
Compare
…r gcc chipStar's hipconfig -C outputs --offload=spirv64, -nohipwrapperinc, --hip-path=, and --target= which are clang-only flags. When CC=gcc is used for .c files (or CXX != HIPCC for .cpp files), these flags cause build failures. Add HIPCONFIG_CPPFLAGS_C that filters the clang-only flags and adds an explicit -I$(ROCM_DIR)/include (since -nohipwrapperinc was suppressing the wrapper that would have pulled in hip_runtime.h).
Add a generic --env KEY=VAL CLI option (repeatable) to tests/junit.py
that injects environment variables into the test subprocess environment.
The flag mutates os.environ in the parent before mp.Pool is created, so
worker processes inherit the values via the existing init_process()
copy.
Useful for backend-specific runtime knobs that affect test behavior, for
example silencing chipStar runtime informational/warning lines that
JUnit otherwise classifies as test failures:
make junit BACKENDS='/gpu/hip/ref ...' \
JUNIT_ARGS='--env CHIP_LOGLEVEL=crit'
No changes to tests/junit_common.py.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
12 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
No description provided.