Enhancement: Add rocprofv2 trace support for AMD GPUs - #817

Open

Shenghsun Cho (shcho) wants to merge 14 commits into

microsoft:mainfrom

shcho:dev/choshenghsun/reliaguard-tracegen-mi300x

Commits on May 19, 2026

Enhancement: Add rocprofv2 trace support for AMD GPUs
Shenghsun Cho
committed
20ebb38
Use shlex.quote() for trace output paths in runner.py
Shenghsun Cho
committed
9b1dd82
Rename nsys_prefix to trace_prefix for tool-agnostic naming
Shenghsun Cho
committed
a9485ef

Commits on Jul 7, 2026

Use getattr for safe torch.version.cuda/hip access in pytorch_base.py
Shenghsun Cho
committed
8a22f3e
Add -- separator after rocprofv2 options in runner.py
Shenghsun Cho
committed
a351951

Commits on Jul 22, 2026

Escape single quotes so trace paths survive bash -lc wrapper
Shenghsun Cho
committed
04c55b0
Warn when both SB_ENABLE_NSYS and SB_ENABLE_ROCPROF are set
Shenghsun Cho
committed
5cca6cc
Document that torch profiler CUDA APIs alias to HIP on ROCm builds
Shenghsun Cho
committed
497abd9
Factor duplicated trace-command construction into a shared helper
Shenghsun Cho
committed
da4635c
Fix yapf formatting and flake8 D301 in runner.py / pytorch_base.py
Shenghsun Cho
committed
b4b5e54

Commits on Jul 23, 2026

Remove -- end-of-options separator from rocprofv2 command
Shenghsun Cho
committed
eb02d3e

Commits on Sep 1, 2026

Add regression tests for nested-shell quoting and profiler command paths
Shenghsun Cho
committed
1bf03e8

Commits on Sep 2, 2026

Merge branch 'main' into dev/choshenghsun/reliaguard-tracegen-mi300x
Hongtao Zhang (polarG)
authored
fad5602

Commits on Sep 5, 2026

Merge branch 'main' into dev/choshenghsun/reliaguard-tracegen-mi300x
Hongtao Zhang (polarG)
authored
f249cf7