| .github | |
| benchmarks | |
| benchviz | |
| build | |
| docs | |
| generated | |
| docs | |
| algorithms | |
| design | |
| developer | |
| guide | |
| pages | |
| perf | |
| evaluation | |
| tuning | |
| examples | |
| consumer | |
| include | |
| batchlas | |
| blas | |
| device | |
| detail | |
| group_blas_common.hh | |
| group_blas_fill.hh | |
| group_blas_gemm.hh | |
| group_blas_gemv.hh | |
| group_blas_ger.hh | |
| group_blas_rankk.hh | |
| group_blas_subgroup_common.hh | |
| group_blas_symm.hh | |
| group_blas_symv.hh | |
| group_blas_trmm.hh | |
| group_blas_trmv.hh | |
| group_blas_vector.hh | |
| group_blas.hh | All device group BLAS operations; include <batchlas/blas/device.hh> instead |
| functions | |
| gemm.hh | |
| gemv.hh | |
| geqrf.hh | Batched Householder QR factorization (geqrf) and its workspace query |
| gesv.hh | Batched general linear solve (gesv) by LU with partial pivoting |
| gesvd.hh | |
| getrf.hh | Batched LU factorization with partial pivoting (getrf) and its workspace query |
| getri.hh | Batched matrix inverse from LU factors (getri) and its workspace query |
| getrs.hh | Batched solve with LU factors from getrf (getrs) and its workspace query |
| hemm.hh | |
| her2k.hh | |
| herk.hh | |
| iluk.hh | |
| orgqr.hh | Batched formation of Q from geqrf reflectors (orgqr/ungqr) and its workspace query |
| ormqr.hh | Batched application of Q from geqrf reflectors (ormqr/unmqr) and its workspace query |
| posv.hh | Batched Hermitian positive-definite solve (posv) by Cholesky |
| potrf.hh | Batched Cholesky factorization (potrf) and its workspace query |
| spmm.hh | |
| syev.hh | |
| symm.hh | |
| syr2k.hh | |
| syrk.hh | |
| trmm.hh | |
| trsm.hh | |
| csr_generators.hh | Random CSR test-matrix generators (batchlas::csr_generators) |
| device.hh | Umbrella header for the in-kernel group BLAS (namespace batchlas::device) |
| enums.hh | Enumerations, scalar traits and scalar concepts shared by every public header |
| extensions.hh | |
| extra.hh | Matrix norms, condition numbers, conditioned test-matrix generators and transpose |
| functions.hh | All positional dense entry points (blas/functions/*.hh) plus their option-struct spellings |
| linalg-ops.hh | The batchlas::linalg convenience layer: elementwise operations and value-returning wrappers |
| linalg.hh | Everything <batchlas.hh> provides: containers, entry points, extensions and the linalg layer |
| matrix.hh | The batched containers: Matrix (owning), MatrixView (non-owning), KernelMatrixView (device-side), Vector and VectorView |
| options.hh | Option-struct spellings of the dense BLAS and LAPACK-style entry points |
| queue-dispatch.hh | Backend deduction from the Queue and the USM pointer check |
| internal | |
| ormqr_blocked.hh | Blocked (WY) application of Q from geqrf reflectors |
| util | |
| env.hh | The shared environment-variable parsers and the ScopedEnvVar RAII setter |
| group-collectives.hh | batchlas::portable:: group broadcast, shuffles and sums for any trivially copyable or complex T |
| group-invoke.hh | Leader-only invocation and leader broadcast for SYCL groups and sub-group partitions |
| kernel-heuristics.hh | Launch-size heuristics used by the library's own kernels |
| mempool.hh | BumpAllocator, the linear sub-allocator over a workspace span, and workspace_bytes() |
| miniacc.hh | Miniacc: a header-only accuracy-benchmark harness (register cases, sample, summarise, CSV) |
| sycl-device-queue.hh | Batchlas::Device, batchlas::Event and batchlas::Queue: where work runs and how it is ordered |
| sycl-local-accessor-helpers.hh | Util::get_raw_ptr() for sycl::local_accessor, used to pass device-BLAS workspaces |
| sycl-span.hh | Batchlas::Span, the non-owning pointer + length every entry point takes for 1-D data |
| sycl-vector.hh | Batchlas::UnifiedVector, an owning growable array in USM shared memory |
| workspace.hh | WorkspaceLease: a scoped borrow of scratch memory from a Queue's workspace arena |
| verify | |
| reference.hh | |
| tolerance.hh | |
| error.hh | The BatchLAS exception hierarchy |
| no_route.hh | |
| settings.hh | Every BATCHLAS_* environment variable the library reads, captured once into batchlas::Settings, plus the programmatic override configure() |
| sycl_interop.hh | SYCL interop: batchlas::Queue and batchlas::Event to and from sycl::queue and sycl::event |
| tuning_params.hh | Tuning constants, bucketed by problem order n, with runtime BATCHLAS_TUNE_* overrides |
| batchlas.hh | The umbrella header: #include <batchlas.hh> gives the whole public API |
| python | |
| examples | |
| src | |
| ops | |
| gemm | |
| kernels | |
| accessors.hh | |
| epilogue_linear.hh | |
| load_policies.hh | |
| register_128x128.hh | |
| register_64x64_k16_wide.hh | |
| register_launchers.hh | |
| register_tiled_common.hh | |
| register_wide_transposed.hh | |
| small_batched.hh | |
| tiled_general.hh | |
| vec4.hh | |
| choice.hh | Gemm: direct, tiled, small, reg, wide, vendor |
| gemm_kernels.hh | |
| gemv | |
| choice.hh | Gemv: cta, direct, vendor |
| gemv_native.hh | |
| geqrf | |
| kernels | |
| geqrf_tiny_device.hh | |
| can_run.hh | Geqrf's can_run, a header so host tests can pass a synthetic select::Device |
| choice.hh | Geqrf: tiny, cta, blocked, vendor |
| geqrf.hh | |
| geqrf_cta_device.hh | |
| geqrf_native.hh | |
| geqrf_panel_reg_device.hh | |
| gesv | |
| choice.hh | Gesv: tiny, blocked |
| gesvd | |
| choice.hh | Gesvd: jacobi, cta, blocked, vendor |
| gesvd_native.hh | |
| getrf | |
| kernels | |
| getrf_panel_reg_device.hh | |
| choice.hh | Getrf: tiny, cta, blocked, vendor |
| getrf_cta_device.hh | |
| getrf_native.hh | |
| getri | |
| choice.hh | Getri: blocked, vendor |
| getri_native.hh | |
| getrs | |
| choice.hh | Getrs: cta, blocked, vendor |
| getrs_native.hh | |
| orgqr | |
| choice.hh | Orgqr: blocked, vendor |
| orgqr_native.hh | |
| ormqr | |
| choice.hh | Ormqr: blocked, vendor |
| vendor.hh | The ormqr vendor call for callers that bypass selection, gated on the library |
| posv | |
| choice.hh | Posv: tiny, cta, blocked |
| potrf | |
| kernels | |
| potrf_cta_device.hh | |
| potrf_lpanel_device.hh | |
| symmetric_product_fold.hh | |
| choice.hh | Potrf: tiny, cta, lpanel:panel, blocked, vendor |
| potrf_native.hh | |
| potrf_slm_hole.hh | |
| spmm | |
| choice.hh | Spmm: direct, vendor |
| spmm_native.hh | |
| syev | |
| stedc | |
| stedc_internal.hh | |
| stedc_levels_plan.hh | |
| stedc_merge_kernels.hh | |
| stedc_secular.hh | |
| steqr | |
| steqr_cta_device.hh | |
| steqr_internal.hh | |
| sytrd | |
| sytrd_cta_device.hh | |
| sytrd_sb2st_cta.hh | |
| sytrd_sb2st_cta_instantiations.hh | |
| sytrd_sb2st_hh.hh | |
| choice.hh | Syev: cta, cta_fused, jacobi, blocked, two_stage, vendor |
| uplo_mirror.hh | |
| vendor.hh | The syev vendor call for callers that bypass selection, gated on the library |
| syevx | |
| syevx.hh | |
| symm | |
| choice.hh | Symm: expand, vendor |
| syr2k | |
| kernels | |
| syr2k_triangular_tiles.hh | |
| choice.hh | Syr2k: triangular, vendor |
| syrk | |
| kernels | |
| syrk_gram_tiles.hh | |
| syrk_triangular_tiles.hh | |
| choice.hh | Syrk: gram, triangular, vendor |
| trmm | |
| kernels | |
| trmm_triangular_tiles.hh | |
| choice.hh | Trmm: triangular, expand, vendor |
| trsm | |
| choice.hh | Trsm: cta, sg_left, blocked, vendor |
| trsm_canonical.hh | |
| trsm_native.hh | |
| select | |
| coverage.hh | What this build links and what a run reached, as CSV in BATCHLAS_COVERAGE_OUT.pid; the columns are a contract with scripts/, tools/tune and benchviz |
| select.hh | Flat kernel selection: one choose() per op over measured per-device tables |
| vendor.hh | Link facts for if constexpr gates: is an op's vendor library compiled in for a backend, and is the native level-3 tile kernel linked |
| tests | |
| tools | |
| tune | |
| tuned | |
1.18.0
