[ci] rockrel testing gaps comparative to TheRock

[geomin12]

rockrel uses thin wrapper workflows that dispatch to TheRock's reusable workflows. However, several test workflows in TheRock don't have corresponding wrappers in rockrel, creating gaps in test coverage for releases (especially ASAN builds).

Background

rockrel's current test wrappers:

• test_artifacts.yml - Component tests per GPU family
• test_pytorch_wheels_full.yml - Full PyTorch test suite
• test_native_linux_packages_install.yml - DEB/RPM package install tests

These are called by multi_arch_release.yml and multi_arch_release_asan.yml which dispatch to TheRock.

Missing Test Wrappers

Workflow	Purpose	Priority
test_artifacts_structure.yml	Validates artifact archive integrity before publishing to release buckets	High
test_rocm_wheels.yml	Tests rocm[libraries], rocm[devel], rocm[profiler] wheel installs + rocm-sdk test	High
test_linux_jax_wheels.yml	Full JAX wheel testing with GPU validation (multi_device, core, scipy tests)	High
test_pytorch_wheels.yml	Quick PyTorch smoke tests (6 test suites vs full suite)	Medium
test_linux_jax_wheels_partial.yml	Partial JAX tests for multi-arch releases	Medium

ASAN-Specific Concerns

TheRock's test_component.yml sets ASAN-specific environment variables when build_variant: asan:

ASAN_OPTIONS=detect_odr_violation=0:quarantine_size_mb=600
HSA_XNACK=1
ASAN_SYMBOLIZER_PATH=<artifacts>/llvm/bin/llvm-symbolizer

• Verify build_variant: asan propagates correctly through rockrel -> TheRock dispatch chain
• Confirm ASAN env vars are applied during test runs
• Validate native package install tests work with ASAN-built packages
• Note: JAX wheels are intentionally skipped for ASAN builds in TheRock

Tasks

High Priority

• Create test_artifacts_structure.yml wrapper
• Create test_rocm_wheels.yml wrapper
• Create test_linux_jax_wheels.yml wrapper
• Verify ASAN test configuration end-to-end

Medium Priority

• Create test_pytorch_wheels.yml wrapper (for quick validation)
• Create test_linux_jax_wheels_partial.yml wrapper

Current Coverage

Test Type	Regular Release	ASAN Release	rockrel Wrapper?
Artifact Structure Validation	Yes	?	No
Component Tests (per GPU family)	Yes	Yes	Yes
Native DEB/RPM Install	Yes	Yes	Yes
ROCm SDK Wheels	Yes	?	No
PyTorch Wheels (smoke)	Yes	Yes	No
PyTorch Wheels (full)	Yes	Yes	Yes
JAX Wheels	Yes	N/A	No

? = needs verification, N/A = intentionally skipped

[ScottTodd]

See also this (pinned) issue:

• [CI] Expand CI workflows to build and test packages (ROCm Python, PyTorch, JAX, Native Linux, etc.) TheRock#3177

[jonatluu]

test_linux_jax_wheels.yml and .github/workflows/test_pytorch_wheels.yml will not need a wrapper as these are already covered. A single test rocm workflow will be introduced to limit workflow files and to cover the follow tests:

• artifact structure tests (CPU)
• rocm wheels (python packages, GPU)
• rocm native linux packages (CPU/GPU?)
• rocm native windows packages (when available, CPU/GPU?)

jax wheel already covered:
https://github.com/ROCm/TheRock/blob/a3e3f450db81b55f53a05581ebf52a21583c1126/.github/workflows/multi_arch_build_linux_jax_wheels.yml#L299-L306

pytorch wheel already covered:
https://github.com/ROCm/TheRock/blob/a3e3f450db81b55f53a05581ebf52a21583c1126/.github/workflows/multi_arch_build_portable_linux_pytorch_wheels.yml#L497-L519