[dayatsin-amd]

Motivation

Add support for gfx1250 GPU

Test Plan

Tested that these patches pass ROCRtst and HIP catch tests on existing GPUs.

Test Result

Pass

Submission Checklist

• Look over the contributing guidelines at https://github.com/ROCm/ROCm/blob/develop/CONTRIBUTING.md#pull-requests.

[kentrussell]

Nothing worth blocking for. Just 2 kfdtest changes and an unneeded capitalization on a hex value. Ignore the hex change unless you have to do another upload anyways. kfdtest changes can be made after these land

[randomizedcoder]

@dayatsin-amd

As you may have seen from some of my recent PRs, I've been running static analysis tools across the codebase.

One minor thing flawfinder flagged (pre-existing, not introduced by your PR): PrefetchRange in runtime.h L732-739 — the 2-arg constructor doesn't initialize prev and next. The prefetch_lock_ mutex likely protects against any read-before-write in practice, but initializing them in the constructor would make it safe by construction:

PrefetchRange(size_t Bytes, PrefetchOp* Op) : bytes(Bytes), op(Op), prev{}, next{} {}

Not a blocker at all — just a small defensive improvement. Feel free to ignore if it's out of scope for this PR.

[randomizedcoder]

You are also touching fmm.c which has some tiny printf cppcheck issues
#5049

[dayatsin-amd]

@dayatsin-amd

As you may have seen from some of my recent PRs, I've been running static analysis tools across the codebase.

One minor thing flawfinder flagged (pre-existing, not introduced by your PR): PrefetchRange in runtime.h L732-739 — the 2-arg constructor doesn't initialize prev and next. The prefetch_lock_ mutex likely protects against any read-before-write in practice, but initializing them in the constructor would make it safe by construction:

PrefetchRange(size_t Bytes, PrefetchOp* Op) : bytes(Bytes), op(Op), prev{}, next{} {}

Not a blocker at all — just a small defensive improvement. Feel free to ignore if it's out of scope for this PR.

Please keep PR's focused on a single topic. I will address that change in another PR.

[randomizedcoder]

Filed #5051 as many of the files in this PR use the obsolete 'alloca'

[randomizedcoder]

Another PR on signal.cpp: #5052

[randomizedcoder]

This PR touches amd_hsa_code.cpp which also has memory corruption bug, although it's possibly unused code
#5059

[lumachad]

@dayatsin-amd FYI, this broke gfx1201 debugging on rocgdb.

Edit: More specifically this commit: 0853edd

[lancesix]

@dayatsin-amd FYI, this broke gfx1201 debugging on rocgdb.

Hi @dayatsin-amd, since this got pinned down to trap handler, I'll try to have a look shortly.