Add template-based operation dispatch - #998
Draft
davidbeckingsale wants to merge 35 commits into
Draft
Conversation
…r safety - Update reallocate dispatch to avoid platform-specific std::realloc - Fix memory safety issues in ResourceManager's reallocate implementations - Clean up asynchronous operation handling - Add safety checks to prevent reallocation of offset pointers 🤖 Generated with [Claude Code](https://claude.ai/code) Co-Authored-By: Claude <noreply@anthropic.com>
- Rename generic_reallocate to reallocate and use this implementation for all platforms - Fix missing return statements and pointer update issue in reallocate implementation - Remove platform-specific reallocate implementations to prevent unsafe std::realloc usage - Add safety mechanism to update pointers in the caller context - Fix argument consistency in async and typed versions - Improve documentation and error handling This ensures that all memory pools and allocators can safely reallocate memory using the allocate-copy-free pattern rather than platform-specific implementations which could fail with pooled memory. 🤖 Generated with [Claude Code](https://claude.ai/code) Co-Authored-By: Claude <noreply@anthropic.com>
…MP Target - Replace SYCL and OpenMP Target platform-specific reallocate implementations with comments - Ensure consistent use of the generic allocate-copy-free pattern for all platforms - Fix argument signature to match the updated reallocate implementation 🤖 Generated with [Claude Code](https://claude.ai/code) Co-Authored-By: Claude <noreply@anthropic.com>
davidbeckingsale
force-pushed
the
feature/operation-v2
branch
from
July 29, 2026 21:30
b6e3e4f to
2b3db65
Compare
The UMPIRE_ENABLE_DEVELOPER_BENCHMARKS path builds benchmarks/ without BLT's gbenchmark support enabled, so targets that include benchmark/benchmark.h fail to compile (seen on mac and Windows CI). Wrap those targets in an UMPIRE_ENABLE_BENCHMARKS block, matching develop's structure; stress tests remain unconditional. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
- cuda.hpp: the forward declaration of umpire_device_memset_kernel had its 'template <typename T>' outside the __CUDACC__ guard, so non-CUDA compilation of the header produced two consecutive template headers (shadowed parameter / too many template-parameter-lists) and left detail::device_memset undeclared. - cuda.hpp/hip.hpp/sycl.hpp: wrap typed-pointer arguments to fmt::format in fmt::ptr(); fmt disallows formatting non-void pointers, which broke the HIP build (amdclang, fmt static_assert). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
fmt v12 refuses to format unscoped enums without a formatter specialization; the raw get_dev_err argument in the prefetch error messages broke the CUDA CI build when instantiated from ResourceManager.cpp. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Declaring `using sycl = resource::sycl_platform` inside namespace umpire made every unqualified `sycl::queue`/`sycl::event`/`sycl::usm` lookup in umpire's SYCL code (op/sycl.hpp, ResourceManager.cpp, traits) resolve to the platform tag struct instead of the global ::sycl namespace, breaking the SYCL CI build. The alias was unused; the cuda/hip/omp_target aliases have no such collision and are kept. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…r shadow - src/tpl/CMakeLists.txt: define CAMP_ENABLE_SYCL on the camp target when UMPIRE_ENABLE_SYCL is on (mirrors the existing CAMP_HAVE_CUDA/HIP wiring); without it camp::resources::Sycl does not exist and op/sycl.hpp's get_queue() cannot compile. - sycl.hpp: camp's EventProxy has no (Resource, sycl::event) constructor -- construct it from the resource alone like the CUDA/HIP async impls; the resource's in-order queue provides ordering. - sycl.hpp: rename sycl_error_check's catch variable from `e` to `ex`; UMPIRE_ERROR declares a local `e`, so the message expression self-referenced the uninitialized local instead of the caught exception. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
The SYCL and OpenMP-target backends specialize the primary op templates (copy, memset, prefetch, device_memset) but never included the header that declares them, so compiling with those backends enabled failed with "explicit specialization of undeclared template struct". cuda.hpp, hip.hpp, and host.hpp already include operations.hpp; the gap was only masked because op.hpp happens to include host.hpp (which pulls in operations.hpp) before the platform headers. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
These tests call the public umpire::copy/memset/prefetch free functions but only included op/operations.hpp (the op tag declarations), not the umbrella umpire/op.hpp that brings in the dispatch layer defining those functions. The other op tests already include op.hpp. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
The SYCL sync device_memset and prefetch intentionally throw and direct callers to resource-taking overloads, but no public async umpire::device_memset(ptr, val, len, ctx) free function existed, so the SYCL tests (the documented usage) could not compile. Add the async overload to the dispatch layer plus the corresponding platform exec overloads (CUDA/HIP: wrap the synchronizing kernel and return a completed-event proxy; host: throw, mirroring the sync overload; SYCL and OpenMP-target already had async exec). Also fix the SYCL PrefetchSync test, which called prefetch with a missing device argument; it now verifies the documented throw and uses the async overload. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
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
2 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.