2026-10-07 — docs: Took the publish list’s summary from the README, passed secrets to the called CI and pinned docs bundle with tests (docs/publish.toml, release.yml, cli/tests/test_cli.py).
2026-10-07 — ci: Added the release job that builds the documentation bundle after CI and attaches it to the tag’s release (.github/workflows/release.yml, ci.yml).
2026-10-07 — docs: Added the frequently asked questions page (docs/guides/FAQ.md).
2026-10-07 — cli: Added sr docs bundle and made docs a command group registered from SushiCore (cli/sushiruntime/cli.py, register_docs_commands).
2026-10-07 — docs: Added the publish list and turned on Doxygen’s XML output for the documentation bundle (docs/publish.toml, Doxyfile).
2026-10-07 — graph: Added add_emitter, a recording form that hands the caller the backing graph, to Graph and Region (graph.hpp, region.hpp, test_region_emitter.cpp).
2026-10-07 — docs: Moved the Doxygen main page into the manual and the generated site into the build tree (docs/reference/API_MAINPAGE.md, Doxyfile, services/project.py, ci.yml).
2026-10-07 — licence: Named Mustafa Garip and Sushi Systems as the copyright holders in the licence and notice files (LICENSE, NOTICE.md).
2026-10-06 — licence: Restored the three-section license box and named both holders in the copyright line of every source file (tools/licensing/write_license_block.py, tools/documentation/check_source_comments.py).
2026-10-05 — execution: Added RuntimeContext::live_usm_count, the number of USM allocations recorded and not yet freed (runtime_context.hpp, runtime_context.cpp, test_topology.cpp).
2026-10-05 — docs: Moved design reasoning out of source comments into folder READMEs and gave every source file its file block (include/SushiRuntime/, src/, tests/, examples/, cli/).
2026-10-05 — docs: Moved the manual into the documentation tree, merged the two changelogs, added the root README and the required documents, and copied the checkers (README.md, docs/, tools/).
2026-10-05 — cli: Renamed sr doxygen to sr docs, keeping the old word as a hidden alias until 2027-03-25 (cli/sushiruntime/cli.py, cli/sushiruntime/aliases.py).
2026-10-05 — cli: Renamed sr docker to sr container, keeping the old word as a hidden alias until 2027-03-25 (cli/sushiruntime/cli.py, cli/sushiruntime/aliases.py).
2026-10-05 — cli: Made sr package --prefix refuse a protected or foreign non-empty directory without the new --force (services/packaging.py, services/stage_tree.py).
2026-10-05 — cli: Took the project root from sushi-module.toml, so a directory holding only CMakeLists.txt is no longer a checkout (cli/sushiruntime/config.py).
2026-10-05 — cli: Made sr run --distributed reject a role other than master or worker with exit code 2 (cli/sushiruntime/cli.py, DistributedRole).
2026-10-05 — cli: Added --version, --describe and a main() entry point (cli/sushiruntime/cli.py).
2026-10-05 — cli: Took sr config from sushicore (cli/sushiruntime/cli.py, services/diag.py).
2026-10-05 — cli: Named sr setup --toolchain in every missing-toolchain message (services/install_hint.py, services/project.py, services/select.py).
2026-10-05 — docs: Corrected the hub install-cli, --no-cuda and Windows ASan flag descriptions (docs/guides/CLI_GUIDE.md, README.md).
2026-10-05 — licence: Changed the licence from Apache-2.0 WITH LLVM-exception to PolyForm Noncommercial 1.0.0 and removed docs/LICENSE (LICENSE, COMMERCIAL.md, NOTICE.md, source headers).
2026-10-04 — cli: Named sr setup –toolchain as the way to install a missing toolchain (cli/sushiruntime/services/select.py).
2026-10-04 — cli: Added setup, doctor, link and unlink from sushicore and stopped reading a repo-local dependencies folder (cli/sushiruntime/cli.py, cli/sushiruntime/config.py).
2026-09-22 — cli: Described this module in its own sushi-module.toml (sushi-module.toml).
2026-09-22 — cli: Took sushicore from PyPI instead of a sibling checkout (cli/pyproject.toml).
2026-09-22 — cli: Drew sr --help as sushicore’s grouped help page, with examples on every command (cli/sushiruntime/cli.py, console.py).
2026-09-22 — cli: Took the theme’s style names for the CLI’s colours in place of raw Rich colour names (services/select.py, services/project.py).
2026-09-22 — cli: Raised the sushicore lower bound to 0.4.0 (cli/pyproject.toml).
2026-09-21 — graph: Fixed the late-bound add_reduce overload’s and add_prefix_scan’s zero-capacity emits reporting WIDTH_UNKNOWN (graph.hpp).
2026-09-21 — scheduler: Fixed the scheduler path snapshotting dispatch widths with profiling off (scheduler.cpp).
2026-09-20 — graph: Added API::Graph::rule_registration, which returns the registration index a recorded rule took, the identity RunReport::node_events numbers by (api/graph/graph.hpp, graph/task_graph.hpp).
2026-09-20 — execution: Submitted the chain layer’s unprofiled nodes through ext::oneapi::experimental::submit, which keeps no per-node event (ChainRun::submit_node_discarding, src/execution/chain_run.cpp).
2026-09-20 — tests: Fixed the prefix scan chain test folding its digest over the whole capacity, so the stale tail held it constant (tests/functional/integration/test_device_sized_primitives.cpp).
2026-09-19 — execution: Added the in-order chain that runs a chainable plan from the calling thread, with RuntimeSettings::force_scheduler and RunReport::layer (execution/chain_plan.cpp, execution/chain_run.cpp, graph.hpp).
2026-09-19 — scheduler: Kept a worker holding in-flight events on yield instead of the 100 us sleep (TaskScheduler::worker_loop, src/scheduler/scheduler.cpp).
2026-09-19 — graph: Changed the scan, counting sort, late-bound reductions and sized_from_device tasks to size themselves on the device, so their plans take the chain (graph.hpp, chain_plan.cpp, device_fault.hpp).
2026-09-19 — tests: Restated the steady-state allocation tests’ rule as a bare-SYCL baseline comparison, not an absolute per-tick count (tests/functional/integration/test_steady_state_allocation.cpp).
2026-09-19 — scheduler: Removed the per-node dependency allocation in TaskScheduler::execute_task by iterating SmallEventVector inline (scheduler/scheduler.cpp, graph/node.hpp).
2026-09-17 — memory: Changed Memory::UsmRegistry::record and RuntimeContext::adopt_usm to take the allocation’s length beside its device index (memory/usm_registry.hpp, execution/runtime_context.hpp).
2026-09-17 — memory: Added UsmRegistry::containing_allocation and RuntimeContext::containing_usm_base, which resolve an interior pointer to the allocation holding it (memory/usm_registry.hpp, runtime_context.cpp).
2026-09-17 — graph: Added add_prefix_scan and add_counting_sort to API::Region (api/graph/region.hpp).
2026-09-17 — graph: Added DynamicGraph::set_profiling_requested, stamped into its plan and mirrored by a node_index renumbered densely by region key then add() order at every recompose (api/graph/dynamic_graph.hpp).
2026-09-17 — graph: Changed Graph::add_counting_sort to run only the digits its live key range needs, skipping the rest (graph.hpp, api/detail/fixed_order_scan.hpp).
2026-09-17 — graph: Fixed Graph::add rejecting an interior pointer into a Residency::Host slot, and named the fault when a late-bound slot does not resolve (api/graph/graph.hpp).
2026-09-17 — graph: Fixed add_counting_sort passing sized_from_device an interior pointer into its radix plan, which threw on a CUDA device (api/graph/graph.hpp).
2026-09-16 — graph: Added Graph::add_counting_sort, a stable late-bound radix argsort writing indices, exclusive offsets and the total (graph.hpp).
2026-09-16 — graph: Added API::LateCount and late-bound overloads of add_reduce and add_segmented_reduce, planned from a capacity and resolved per run (graph.hpp, api/detail/fixed_order_scan.hpp).
2026-09-16 — graph: Added Graph::add_prefix_scan, a late-bound exclusive uint32 scan over add_reduce’s tile tree (graph.hpp).
2026-09-16 — core: Added per-dispatch node event records with a run clock pairing, and derived NodeTiming from them (core/profiling_ledger.hpp, event_polling.hpp, execution_engine.cpp).
2026-09-16 — graph: Added Graph::set_profiling_requested, which routes a graph’s device nodes to the profiling queue and gates its timing per graph instead of runtime-wide (graph.hpp, scheduler.cpp).
2026-09-15 — core: Added Core::NodeEventRecord, Core::ClockPairing and RunReport::node_events/clock for per-dispatch node timing (core/run_report.hpp).
2026-09-15 — graph: Added a registration index on every node, shared by the tile levels of one add_reduce, and TaskGraph::reserve_registration() (graph/node.hpp, graph/task_graph.hpp).
2026-09-15 — execution: Added a lazily built profiling queue per device on the plain queue’s context (RuntimeContext::profiling_queue, runtime_context.cpp).
v0.3.0 — 2026-09-15
2026-09-15 — cmake: Added the sushiruntime_gpu_offload interface target and its SushiRuntime::gpu_offload alias, publishing -fsycl-targets and backend options to consumers outside the runtime’s tree (cmake/gpu/GpuBackend.cmake).
2026-09-14 — graph: Made Graph::add reject a late-bound device counter that is not host USM, throwing Errc::invalid_access for Shared and Device counters that worked before (buffer.hpp, dynamic.hpp, runtime_context.cpp).
2026-09-14 — interop: Added API::Interop for host memory export and device memory import (interop/, Runtime::interop()).
2026-09-14 — memory: Added Memory::Allocation, IAllocationOwner and UsmAllocationOwner (memory/allocation.hpp, memory/usm_allocation_owner.hpp).
2026-09-14 — api: Made Buffer<T> own its memory through a Memory::Allocation (api/data/buffer.hpp).
2026-09-14 — cmake: Added a per-vendor GPU toolkit locator layer that DetectGpu and Cuda request vendors through (cmake/gpu/locate/CudaToolkit.cmake, IntelToolkit.cmake, RocmToolkit.cmake).
2026-09-14 — cmake: Changed the Windows CUDA locator to fall back to the machine registry when the shell’s CUDA_PATH predates the install (cmake/gpu/locate/CudaToolkit.cmake).
2026-09-14 — memory: Added Residency::Host over pinned host USM and RuntimeContext::malloc_host_on (buffer.hpp, dynamic.hpp, runtime_context.cpp).
2026-09-14 — cli: Fixed the TSan message of sr build to match the Windows exclusion in cmake/Sanitizers.cmake instead of naming icx-cl as the untested case (cli/sushiruntime/services/project.py, docs/architecture/ARCHITECTURE.md).
2026-09-07 — docs: Renamed the hub’s command from ss to hub in the manual (docs/).
2026-08-30 — graph: Added API::DynamicGraph::find_region(key), a const lookup returning the live region or nullptr without marking it for rebuild (api/graph/dynamic_graph.hpp).
2026-08-25 — cli: Renamed the shared CLI layer’s import from sushicli to sushicore across every import site (cli/config.toml, cli/sushiruntime/config.py, cli/sushiruntime/console.py, cli/sushiruntime/services/diag.py).
2026-08-18 — memory: Removed the pooled USM allocation path with BitmaskMemoryPool, ThreadLocalMagazine and USMAllocator (include/SushiRuntime/memory/).
2026-08-18 — memory: Removed IAllocatorStrategy, create_allocator and RuntimeContext::get_allocator() with the pool (execution/runtime_context.hpp).
2026-08-18 — memory: Moved USM pointer ownership out of RuntimeContext into Memory::UsmRegistry and made free_usm resolve and forget in one locked step (memory/usm_registry.hpp, Unit_UsmRegistry).
2026-08-18 — cli: Moved package verification into sr test --suite package, removed sr package --no-verify and changed its default prefix to build/dist (tests/package/).
2026-08-18 — cmake: Included every CMake module by absolute path and prefixed ProjectOptions.cmake and Vcpkg.cmake, whose names collided with SushiEngine’s (cmake/, docs/CONTRIBUTING.md).
2026-08-18 — cmake: Fixed the determinism target needing CMake 3.30 by detecting the frontend at configure time where $<CXX_COMPILER_FRONTEND_VARIANT:...> is missing (cmake/Determinism.cmake).
