Use the cached task-local SYCL queue for oneMKL FFT plans

[michel2323]

Summary

_create_descriptor built fresh syclDevice/syclContext/syclQueue objects for every FFT plan. Once those wrappers become garbage, their finalizers (syclQueueDestroy etc.) tear down SYCL runtime state for the still-in-use underlying Level Zero queue, corrupting later DFT commits and crashing at process exit.

This switches _create_descriptor to the cached task-local sycl_queue(global_queue(...)) accessor that every other oneMKL wrapper (BLAS/LAPACK/sparse) already uses, so the plan shares the managed queue lifetime instead of owning a throwaway one.

Notes

Touches only the queue construction in _create_descriptor. The GC.@preserve rooting of lengths/strides already present on main (#582) is left untouched.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 80.90%. Comparing base (e995a63) to head (a1c8ade).
⚠️ Report is 5 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #586      +/-   ##
==========================================
- Coverage   80.92%   80.90%   -0.02%     
==========================================
  Files          48       48              
  Lines        3234     3232       -2     
==========================================
- Hits         2617     2615       -2     
  Misses        617      617              

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[github-actions]

Your PR requires formatting changes to meet the project's style guidelines.
Please consider running Runic (git runic main) to apply these changes.

Click here to view the suggested changes.

diff --git a/test/fft.jl b/test/fft.jl
index d441946..21cbc5b 100644
--- a/test/fft.jl
+++ b/test/fft.jl
@@ -80,26 +80,28 @@ end
     end
 end
 
-@testset "shared queue lifetime across plans" begin
-    # Plans must share the single cached task-local SYCL queue rather than each owning a
-    # throwaway one (whose finalizer would tear down shared SYCL/oneMKL state). Assert the
-    # shared handle deterministically, independent of whether a stale queue would crash.
-    cached_handle = Base.unsafe_convert(oneAPI.oneMKL.syclQueue_t,
-        oneAPI.sycl_queue(oneAPI.global_queue(oneAPI.context(), oneAPI.device())))
+    @testset "shared queue lifetime across plans" begin
+        # Plans must share the single cached task-local SYCL queue rather than each owning a
+        # throwaway one (whose finalizer would tear down shared SYCL/oneMKL state). Assert the
+        # shared handle deterministically, independent of whether a stale queue would crash.
+        cached_handle = Base.unsafe_convert(
+            oneAPI.oneMKL.syclQueue_t,
+            oneAPI.sycl_queue(oneAPI.global_queue(oneAPI.context(), oneAPI.device()))
+        )
 
-    dX1 = gpu(rand(ComplexF32, 8))
-    p1 = AbstractFFTs.plan_fft(dX1)
-    @test p1.queue == cached_handle
-    dY1 = p1 * dX1
-    p1i = AbstractFFTs.plan_ifft(dX1)
-    p1i * dY1
+        dX1 = gpu(rand(ComplexF32, 8))
+        p1 = AbstractFFTs.plan_fft(dX1)
+        @test p1.queue == cached_handle
+        dY1 = p1 * dX1
+        p1i = AbstractFFTs.plan_ifft(dX1)
+        p1i * dY1
 
-    GC.gc(true)  # run finalizers of any throwaway per-plan SYCL wrappers
+        GC.gc(true)  # run finalizers of any throwaway per-plan SYCL wrappers
 
-    X2 = rand(ComplexF32, 8, 32)
-    dX2 = gpu(X2)
-    p2 = AbstractFFTs.plan_fft(dX2)
-    @test p2.queue == cached_handle
-    cmp(p2 * dX2, fft(X2))
-end
+        X2 = rand(ComplexF32, 8, 32)
+        dX2 = gpu(X2)
+        p2 = AbstractFFTs.plan_fft(dX2)
+        @test p2.queue == cached_handle
+        cmp(p2 * dX2, fft(X2))
+    end
 end