intel crash reproducer

Author: majosm

.github/workflows/ci.yml

@@ -90,6 +90,34 @@ jobs:
                 export NO_COLOR=1
                 test_py_project
 
+    crash_reproducer_intel_cl:
+        name: Crash Reproducer Conda Py3 Intel
+        runs-on: ubuntu-latest
+        steps:
+        -   uses: actions/checkout@v6
+        -   name: "Main Script"
+            run: |
+                curl -L -O https://raw.githubusercontent.com/illinois-scicomp/machine-shop-maintenance/main/install-intel-icd.sh
+                sed -i \
+                    -e '0,/^VERSION=/s|^VERSION=.*|VERSION=oclcpuexp-2025.21.10.0.10_160000_rel|' \
+                    -e '0,/^RELEASE=/s|^RELEASE=.*|RELEASE="2025-WW45"|' \
+                    -e '0,/^TBB_VERSION=/s|^TBB_VERSION=.*|TBB_VERSION=2022.2.0|' \
+                    install-intel-icd.sh
+                sudo bash ./install-intel-icd.sh
+
+                CONDA_ENVIRONMENT=.test-conda-env-py3.yml
+                echo "- ocl-icd-system" >> "$CONDA_ENVIRONMENT"
+                sed -i "/pocl/ d" "$CONDA_ENVIRONMENT"
+                export PYOPENCL_TEST=intel
+                source /opt/enable-intel-cl.sh
+
+                EXTRA_INSTALL="scipy"
+                curl -L -O https://tiker.net/ci-support-v0
+                . ./ci-support-v0
+                build_py_project_in_conda_env
+                export NO_COLOR=1
+                python intel_crash_reproducer.py
+
     examples3:
         name: Examples Conda Py3
         runs-on: ubuntu-latest

intel_crash_reproducer.py

@@ -0,0 +1,49 @@
+#!/usr/bin/env python
+"""Generate device code for the ``frozen_result`` kernel.
+
+The kernel below is reconstructed inline from its loopy listing: a (10, 4)
+float64 array fill (``_ary = 0.0 + 1.0``) whose two axes have been split into
+group/local/chunk inames and tagged for an OpenCL-style execution.
+"""
+
+import numpy as np
+
+import loopy as lp
+import pyopencl as cl
+
+
+knl = lp.make_kernel(
+    "{ [_ary_dim0_chunk, _ary_dim0_group, _ary_dim0_local_one,"
+    "   _ary_dim1_chunk, _ary_dim1_local_zero] :"
+    " _ary_dim0_chunk = 0 and _ary_dim1_chunk = 0 and _ary_dim0_group >= 0"
+    " and 0 <= _ary_dim0_local_one <= 9 - 4*_ary_dim0_group"
+    " and _ary_dim0_local_one <= 3"
+    " and 0 <= _ary_dim1_local_zero <= 3 }",
+    """
+    _ary[_ary_dim0_local_one + _ary_dim0_group*4 + _ary_dim0_chunk*64,
+         _ary_dim1_local_zero + _ary_dim1_chunk*16] = 0.0 + 1.0  {id=_ary_store}
+    """,
+    [
+        lp.GlobalArg("_ary", dtype=np.float64, shape=(10, 4), dim_tags="N1,N0"),
+    ],
+    name="frozen_result",
+    target=lp.PyOpenCLTarget(),
+    lang_version=(2018, 2),
+)
+
+knl = lp.tag_inames(knl, {
+    "_ary_dim0_group": "g.0",
+    "_ary_dim0_local_one": "l.1",
+    "_ary_dim1_local_zero": "l.0",
+})
+
+print(knl)
+print()
+print(lp.generate_code_v2(knl).device_code())
+
+# Execute the kernel.
+ctx = cl.create_some_context(interactive=False)
+queue = cl.CommandQueue(ctx)
+
+_evt, (out,) = knl(queue)
+print(out.get())