Aurora LTS 2523.40 - #574

Open

michel2323 wants to merge 29 commits into

mainfrom

aurora-lts-2523.40

Commits on Jul 27, 2026

Commits on Jul 30, 2026

Test wtih JuliaGPU/OpenCL.jl#main
michel2323
committed
Build against and run on the Intel GPU LTS stack

michel2323
and
claude
committed
Work around strided mapreducedim! miscompiles on the LTS IGC

michel2323
and
claude
committed
Harden command-queue lifetime to prevent banned-context faults on free

michel2323
and
claude
committed
Root oneMKL FFT descriptor arrays across the C call

michel2323
and
claude
committed
Add an opt-in per-submission synchronize for the LTS dropped-tail bug

michel2323
and
claude
committed
Configure the Aurora LTS self-hosted CI runner

michel2323
and
claude
committed
Gate the LTS workarounds behind a runtime ONEAPI_LTS flag (layers 1-2)

michel2323
and
claude
committed
Compat: SPIRVIntrsinics v1
michel2323
committed
test: fix BFloat16 math test on the Aurora LTS stack
michel2323
committed
examples: skip bfloat16.jl on the Aurora LTS stack
michel2323
committed
level-zero: fix sync-each-submission deadlock in kernel-execution test
michel2323
committed
level-zero: share one env-bool parser across ONEAPI_* flags
michel2323
committed
Register KA.priority! replacement queue in the queue registry
michel2323
committed
mapreduce: guard LTS strided-materialization fallback against host inputs
michel2323
committed
level-zero: bound the LTS queue-finalizer drain instead of waiting forever
michel2323
committed
mapreduce: document that mixed dims=(1,3) reductions need no workaround
michel2323
committed
Restore SPIRV_LLVM_Backend_jll so the rolling stack can compile kernels
michel2323
committed
Restore BFloat16 kernel support on the rolling stack
michel2323
committed
mkl: guard sparse setters against a stale oneAPI_Support_jll ABI
michel2323
committed
oneAPI: make in-process NEO discovery real, fix LD_LIBRARY_PATH comment
michel2323
committed
mapreduce: size the coalesced LTS kernel via launch_configuration
michel2323
committed
mapreduce: add exact-arithmetic regression test for strided mixed reductions
michel2323
committed
Set default ONEAPI_LTS=0
michel2323
committed
Declare the atomic-float extensions on the rolling-stack path too
michel2323
committed
Default to the rolling driver stack; LTS deployments override externally
michel2323
committed
CI: build the support library locally before running tests
michel2323
committed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

Aurora LTS 2523.40 - #574

Aurora LTS 2523.40#574
michel2323 wants to merge 29 commits into
mainfrom
aurora-lts-2523.40

Commits on Jul 27, 2026

Make floating-point atomics work on the LLVM back-end path

onemkl tests: accept oneMKL 2025.2+

Commits on Jul 30, 2026

Test wtih JuliaGPU/OpenCL.jl#main

Build against and run on the Intel GPU LTS stack

Work around strided mapreducedim! miscompiles on the LTS IGC

Harden command-queue lifetime to prevent banned-context faults on free

Root oneMKL FFT descriptor arrays across the C call

Add an opt-in per-submission synchronize for the LTS dropped-tail bug

Configure the Aurora LTS self-hosted CI runner

Gate the LTS workarounds behind a runtime ONEAPI_LTS flag (layers 1-2)

Compat: SPIRVIntrsinics v1

test: fix BFloat16 math test on the Aurora LTS stack

examples: skip bfloat16.jl on the Aurora LTS stack

level-zero: fix sync-each-submission deadlock in kernel-execution test

level-zero: share one env-bool parser across ONEAPI_* flags

Register KA.priority! replacement queue in the queue registry

mapreduce: guard LTS strided-materialization fallback against host inputs

level-zero: bound the LTS queue-finalizer drain instead of waiting forever

mapreduce: document that mixed dims=(1,3) reductions need no workaround

Restore SPIRV_LLVM_Backend_jll so the rolling stack can compile kernels

Restore BFloat16 kernel support on the rolling stack

mkl: guard sparse setters against a stale oneAPI_Support_jll ABI

oneAPI: make in-process NEO discovery real, fix LD_LIBRARY_PATH comment

mapreduce: size the coalesced LTS kernel via launch_configuration

mapreduce: add exact-arithmetic regression test for strided mixed reductions

Set default ONEAPI_LTS=0

Declare the atomic-float extensions on the rolling-stack path too

Default to the rolling driver stack; LTS deployments override externally

CI: build the support library locally before running tests