Fix sparse CSC matrix multiplication - #602
Merged
Merged
Conversation
`dA * x` and `dA * B` returned Aᵀx and AᵀB for a oneSparseMatrixCSC. A CSC handle stores CSR(Aᵀ), so sparse_gemv! and sparse_gemm! apply flip_trans when calling into oneMKL and already present op(A) semantics to their callers. generic_matvecmul! and generic_matmatmul! applied flip_trans a second time, so 'N' reached oneMKL unchanged and it computed S*x = Aᵀx. Pass the transpose flag through instead. The existing sparse tests call sparse_gemv!/sparse_gemm! directly and never exercised `*` or `mul!` on a sparse matrix, which is why this went unnoticed; add a testset covering that path for both CSR and CSC.
michel2323
force-pushed
the
fix-sparse-csc-mul
branch
from
August 2, 2026 02:30
9ea1275 to
442b087
Compare
The CSC generic_matvecmul!/generic_matmatmul! methods were constrained to
BlasReal, so `dA * x` on a complex oneSparseMatrixCSC never reached them
and fell through to the AbstractArray fallback, failing with
CanonicalIndexError: getindex not defined for oneSparseMatrixCSC{ComplexF32, Int64}
The underlying sparse_gemv!/sparse_gemm! wrappers already carry complex
CSC specializations that handle the conjugation identities for the 'C'
case, so widening the constraint to BlasFloat, matching the CSR methods,
is all that is needed.
Extend the multiplication testset over identity/transpose/adjoint for
every supported element type, which covers the complex-only distinction
between transpose and adjoint.
michel2323
changed the title
michel2323
enabled auto-merge (squash)
August 2, 2026 02:43
Codecov Report✅ All modified and coverable lines are covered by tests. Additional details and impacted files@@ Coverage Diff @@
## main #602 +/- ##
==========================================
+ Coverage 79.15% 79.76% +0.61%
==========================================
Files 50 50
Lines 3391 3391
==========================================
+ Hits 2684 2705 +21
+ Misses 707 686 -21 ☔ View full report in Codecov by Harness. 🚀 New features to boost your workflow:
|
michel2323
referenced
this pull request
Aug 3, 2026
Aurora and other large deployments run Intel's long-term-servicing branch of the Compute Runtime rather than the rolling release oneAPI.jl targets. That branch predates a number of driver and IGC fixes, several of which corrupt results silently rather than raise an error. Add the workarounds behind a single opt-in switch, oneL0.LTS[], resolved from ONEAPI_LTS at load time and defaulting to off, so the rolling stack behaves exactly as before. Gated on that switch: * SPIR-V codegen goes through the Khronos translator instead of LLVM's SPIR-V back-end, whose output the LTS NEO/IGC runtime does not accept (#491). Both JLLs stay dependencies; GPUCompiler resolves the tool from the target's `backend` field, so the choice is made at compile time. * BFloat16 is forced off. The LTS SPIR-V stack cannot codegen native bfloat in generic kernels -- a kernel that merely keeps a bfloat value fails with InvalidIRError, and declaring SPV_KHR_bfloat16 crashes the runtime. _device_supports_bfloat16() reports hardware capability and does not capture this, so the test suite and examples/bfloat16.jl gate on the switch instead. * Reductions avoid the strided-read miscompile that silently breaks `sum(transpose(x))`, `a == transpose(b)` and `ishermitian`. Non-dense inputs are materialized before reducing, and reductions that keep the contiguous leading dimension use a coalesced one-work-item-per-slice kernel. Reductions that also reduce dim 1 keep a contiguous innermost axis and are left alone; an Int32 regression test covers that boundary. * Buffers are freed only after draining the queues that may still reference them. LTS NEO advertises ZE_extension_memory_free_policies but ignores BLOCKING_FREE, so a GC-driven free of in-flight work faults and bans the context, surfacing later as a ZE_RESULT_ERROR_UNKNOWN at an unrelated call. A per-(context, device) registry tracks every queue -- including the replacement KA.priority! installs -- and the queue finalizer drains before destroying, with a bounded wait so a task that dies mid-submission cannot hang finalization. Independently of that switch, ONEAPI_SYNC_EACH_SUBMISSION=1 synchronizes after every command-list submission, working around a dropped-tail corruption seen when several processes oversubscribe a single tile. It costs roughly 3x throughput and is off by default, with getter, setter and scoped forms available at runtime. Also, not specific to the LTS stack: dlopen the NEO driver by full path during __init__ so libsycl's bundled Level Zero loader finds it in-process. Setting LD_LIBRARY_PATH there never could -- glibc captures it once at process startup -- so it only ever served spawned worker processes. The self-hosted Aurora runner is configured to exercise the LTS path, with buildkite continuing to cover the rolling stack, and docs/src/lts.md documents both switches, what they change, and what they cost.
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
1 participant
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.
Two fixes to sparse CSC multiplication through the
LinearAlgebrainterface.Transposed result.
dA * xreturnedAᵀxfor aoneSparseMatrixCSC(same fordA * Bandmul!). A CSC handle storesCSR(Aᵀ), sosparse_gemv!/sparse_gemm!already applyflip_transinternally.generic_matvecmul!/generic_matmatmul!applied it a second time, so the two cancelled and oneMKL computedS*x = Aᵀx. CSR was unaffected.Complex element types. The CSC methods were constrained to
BlasReal, so complex CSC never reached them and failed withCanonicalIndexError: getindex not defined. The underlying wrappers already have complex CSC specializations handling the'C'conjugation identities, so widening toBlasFloatto match CSR is sufficient.