Cached run outputs¶
Sanitized stdout/stderr from running curriculum labs on AMD Instinct MI350 series (gfx950)
with ROCm 7. These files back the ```text sample-output blocks in module READMEs and
exercise guides.
What is (and is not) in these files¶
| Included | Stripped |
|---|---|
gfx950 architecture label |
Hostnames, IPs, usernames |
ROCm 7 major version |
Detailed ROCm/HIP/HSA patch versions |
| PASS/FAIL states, bandwidth shapes, profiler table structure | Internal paths, PIDs, timestamps |
| Reproducible command transcripts | Conductor / SSH login banners |
Reproduce locally¶
git clone https://github.com/gahan9/ParallelProgramming
cd ParallelProgramming
export GPU_ARCH=gfx950 # MI350 series
export HIP_PATH=/opt/rocm # ROCm 7 install
bash scripts/collect_outputs.sh
Sanitized artifacts land in outputs/; raw captures (pre-strip) in outputs/_raw/ (gitignored).
Coverage notes¶
| Category | Cached on gfx950 / ROCm 7 |
Gap |
|---|---|---|
| Track A HIP labs + drills | A01–A07 | A08/A09 Triton need a torch wheel (make venv prints setup) |
| Track B HIP labs + drills | B01–B04 | — |
| Track B Python tools | B05–B07 | Host-only stdlib tools |
| Track C Python tools | C01–C06 | Host-only stdlib tools |
| NVIDIA CUDA / CuTe (A10) | — | Requires nvcc (not on AMD hosts) |
Re-run bash scripts/collect_outputs.sh after installing missing wheels or on an NVIDIA box for CUDA targets.
Refresh a single module¶
cd tracks/A-gpu-programming/A01.foundations-and-programming-model
make run-hip GPU_ARCH=gfx950
# pipe through the sanitizer before publishing:
make run-hip GPU_ARCH=gfx950 2>&1 | python3 ../../../scripts/sanitize_output.py
See manifest.json for the artifact index and capture date.