wangbo/ComfyUI
1
0
Fork 0
mirror of https://github.com/comfyanonymous/ComfyUI.git synced 2026-05-24 16:07:30 +08:00
Code Issues Actions 19 Packages Projects Releases Wiki Activity

Note _calc_cond_batch and _calc_cond_batch_multigpu must stay in sync

Browse Source 
Per review feedback on #7063. The two functions share the conds-by-hooks
accumulation, memory-fit batching, and per-chunk output aggregation; the
multigpu variant adds per-device scheduling, .to(device) placement,
per-device patcher/control lookup, and thread-pool dispatch around the
inner loop. Documenting the relationship without extracting helpers --
extraction can land after the initial worksplit-multigpu release once
both paths have settled.

Amp-Thread-ID: https://ampcode.com/threads/T-019e4a00-fe3d-76bd-a2f2-a8c8c4040082
Co-authored-by: Amp <amp@ampcode.com>
This commit is contained in:
Kosinkadink 2026-05-21 11:47:53 -07:00
parent 1417b711ce
commit 822a3ecf73
1 changed files with 7 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

7
comfy/samplers.py
View File

@ -218,6 +218,9 @@ def _calc_cond_batch_outer(model: BaseModel, conds: list[list[dict]], x_in: torc
return executor.execute(model, conds, x_in, timestep, model_options)
def _calc_cond_batch(model: BaseModel, conds: list[list[dict]], x_in: torch.Tensor, timestep: torch.Tensor, model_options: dict[str]):
# NOTE: keep in sync with _calc_cond_batch_multigpu below. Shared logic
# (hooked_to_run accumulation, memory-fit batching, per-chunk output
# aggregation) is duplicated there with per-device scheduling layered on top.
if 'multigpu_clones' in model_options:
return _calc_cond_batch_multigpu(model, conds, x_in, timestep, model_options)
out_conds = []
@ -353,6 +356,10 @@ def _calc_cond_batch(model: BaseModel, conds: list[list[dict]], x_in: torch.Tens
return out_conds
def _calc_cond_batch_multigpu(model: BaseModel, conds: list[list[dict]], x_in: torch.Tensor, timestep: torch.Tensor, model_options: dict[str]):
# NOTE: keep in sync with _calc_cond_batch above. Same conds-by-hooks
# accumulation, memory-fit batching, and output aggregation, but adds a
# per-device scheduler, per-device patcher/control lookup, tensor .to(device)
# placement, and MultiGPUThreadPool dispatch around the inner loop.
out_conds = []
out_counts = []
# separate conds by matching hooks