Running Tests#

GLASS ships a pytest suite that compiles small CUDA binaries and checks their output against NumPy/SciPy references.

Running the suite#

cd test
pip install -r requirements.txt
pytest -v

The first run compiles the CUDA test binaries with nvcc. Subsequent runs skip recompilation unless the source changed (see the compile cache below). Compiled binaries land in test/build/.

Selecting tests#

pytest test_l1.py -v
pytest test_l1.py -k "simple"      # glass:: threadIdx variants only
pytest test_l1.py -k "cg"          # glass::cgrps:: cooperative-groups variants
pytest test_l1.py -k "simple_hs"   # high_speed warp-shuffle variants

The compile cache#

Compilation is driven by test/conftest.py. Binaries are compiled once per pytest session (a session-scoped bins fixture) and cached by a SHA-256 hash of the source set.

For each test binary, _hash_sources() hashes the .cu file together with every GLASS header it can pull in — glass.cuh, glass-cgrps.cuh, glass-nvidia.cuh, the src/base/** and src/nvidia/** headers (including tuning_table.cuh and query_simt.cuh), and the test helpers. The hash is written to test/build/<name>.hash. On the next run, if the hash file and binary both exist and the hash matches, compilation is skipped; otherwise the binary is rebuilt. Editing any hashed header therefore invalidates the cache and forces a recompile.

GPU-arch detection#

detect_arch() queries nvidia-smi --query-gpu=compute_cap and turns the reported compute capability into an nvcc arch flag (e.g. 8.6 → sm_86). If nvidia-smi is unavailable, it falls back to sm_75. Every binary is compiled with -std=c++17 -arch=<detected>.

Optional binaries (graceful skips)#

Some test binaries need extra dependencies, and the fixtures skip rather than fail when those aren’t present:

Binary	Requirement / behavior
`test_l3_nvidia`	Exercises the SIMT-only batched APIs (`gemm_batched_1d`, `gemm_strided_batched_1d`). Does not need cuBLASDx; tests skip only if compilation fails for some toolchain reason.
`test_nvidia_dispatch`	Round-2 auto-dispatch features. Needs `MATHDX_ROOT` (cuBLASDx) to compile; skipped otherwise.
`test_trailing_sync`	The `TRAILING_SYNC` surface. Compiles with or without cuBLASDx (it internally skips the cuBLASDx op when `GLASS_BENCH_CUBLASDX` isn’t defined).

When MATHDX_ROOT is set and cublasdx.hpp is found, the cuBLASDx-gated binaries are compiled with --expt-relaxed-constexpr -DGLASS_BENCH_CUBLASDX plus the MathDx include paths automatically.