The GitHub Actions job "tvm-bot" on tvm.git/main has succeeded.
Run started by GitHub user spectrometerHBH (triggered by spectrometerHBH).

Head commit for run:
abe2526b3e54bd47eaa880c0e832ae7edf75a918 / Akaash Parthasarathy 
<[email protected]>
[Web] Batch GPU-to-GPU copies, fix WebGPU synchronization, and add tests for 
command batching (#20059)

Batch GPU-to-GPU buffer copies with pending compute dispatches in the
same `GPUCommandEncoder` instead of finishing and submitting after every
copy. Since `copyBufferToBuffer` records an encoder command,
dependencies remain ordered while copies can be submitted at the next
required synchronization boundary (see
https://www.w3.org/TR/webgpu/#buffer-copies).

Preserve pending GPU-to-CPU readbacks separately from their queue-tail
status so `sync()` waits for both host readback completion and any later
queue work when necessary. Also flush pending commands before canvas
drawing to prevent rendering stale buffer contents.

Add WebGPU device mocks and tests covering:
  - Mixed compute and GPU-copy batching
  - Host-write, readback, deallocation, and canvas-draw flush boundaries
  - `deviceCopyFromGPU` followed by a batched GPU copy and `sync()`
  - Readback error propagation

Regarding importing from `../../src/webgpu` instead of `../../dist`:
`WebGPUContext` is not re-exported from `index.ts`, so it is unavailable
through the built package like for the other tests. Exporting it solely
for this test would unnecessarily expand the public API. Importing
`../../src/webgpu` is a relatively simple alternative.

Report URL: https://github.com/apache/tvm/actions/runs/30676675429

With regards,
GitHub Actions via GitBox


---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]

Reply via email to