Skip to content

amd warp - #38684

Open
YassineYousfi wants to merge 5 commits into
masterfrom
amd-warp
Open

amd warp#38684
YassineYousfi wants to merge 5 commits into
masterfrom
amd-warp

Conversation

@YassineYousfi

@YassineYousfi YassineYousfi commented Aug 21, 2026

Copy link
Copy Markdown
Contributor
  • skip the trailing kernel/guard allocation from VisionIpc (806,912 bytes per mici frame)
  • pack all non frame inputs inputs to avoid tiny copies
  • merge warp and net into 1 graph
id4-model-replay-packed-timings

@github-actions

Copy link
Copy Markdown
Contributor

Process replay diff report

Replays driving segments through this PR and compares the behavior to master.
Please review any changes carefully to ensure they are expected.

✅ 0 changed, 66 passed, 0 errors

Comment thread openpilot/selfdrive/modeld/SConscript Outdated
USBGPU = usbgpu_present()
if USBGPU:
usbgpu_tg_flags = f'DEBUG=2 DEV=USB+AMD:LLVM WARP_DEV={tg_backend} FLOAT16=1 JIT_BATCH_SIZE=0 GMMU=0 TC_OPT=2'
usbgpu_tg_flags = 'DEBUG=2 DEV=USB+AMD:LLVM FRAME_DEV=CPU FLOAT16=1 JIT_BATCH_SIZE=0 GMMU=0 TC_OPT=2'

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

DEBUG=1 should give a nice speedup

@YassineYousfi

Copy link
Copy Markdown
Contributor Author

needs retry on LIBUSB_ERROR_INTERRUPTED and thread local SQLite cons for the model fallback
fixing here
https://github.com/YassineYousfi/tinygrad/tree/sq-usb-fixes

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants