RomanNum3ral
/
ComfyUI
mirror of https://github.com/Comfy-Org/ComfyUI.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

attention_pytorch: skip the chunk buffer/copy when steps==1 

Addresses CodeRabbit feedback on #14838: the chunking loop always
allocated a temporary output buffer and copied SDPA's result into it,
even in the common case (any non-MPS device, or MPS below the size
threshold) where steps==1 and the loop only ever runs once. Adds a
fast path that calls scaled_dot_product_attention directly and
reshapes its result, matching the original pre-fix code for that case
-- no extra allocation or copy. The chunked path (steps>1) is
unchanged.

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01HAMqHgFBD6e9wjU8k8U6uW
This commit is contained in:
Scott Yewell 2026-07-08 16:31:12 -04:00
parent 7b7d48faa2
commit 55760cdeb5
1 changed files with 19 additions and 16 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

35
comfy/ldm/modules/attention.py
View File

@ -618,28 +618,31 @@ def attention_pytorch(q, k, v, heads, mask=None, attn_precision=None, skip_resha
if SDP_BATCH_LIMIT >= b: if SDP_BATCH_LIMIT >= b:
# See _mps_forced_attn_chunk_steps: on MPS a single attention matrix over # See _mps_forced_attn_chunk_steps: on MPS a single attention matrix over
# ~2^31 elements silently corrupts, so chunk over seq_q to stay under that. # ~2^31 elements silently corrupts, so chunk over seq_q to stay under that.
# steps stays 1 (one iteration over the full sequence, identical to the
# previous unchunked call) everywhere this doesn't apply.
steps = 1 steps = 1
if q.device.type == "mps": if q.device.type == "mps":
elements_full = q.shape[0] * q.shape[1] * q.shape[2] * k.shape[2] elements_full = q.shape[0] * q.shape[1] * q.shape[2] * k.shape[2]
steps = _mps_forced_attn_chunk_steps(elements_full, steps) steps = _mps_forced_attn_chunk_steps(elements_full, steps)
slice_size = math.ceil(q.shape[2] / steps)
raw_out = torch.empty((b, q.shape[1], q.shape[2], dim_head), dtype=q.dtype, layout=q.layout, device=q.device) if steps == 1:
for i in range(0, q.shape[2], slice_size): out = comfy.ops.scaled_dot_product_attention(q, k, v, attn_mask=mask, dropout_p=0.0, is_causal=False, **sdpa_extra)
end = i + slice_size if not skip_output_reshape:
mask_chunk = mask out = out.transpose(1, 2).reshape(b, -1, heads * dim_head)
if mask is not None and mask.shape[-2] != 1:
mask_chunk = mask[..., i:end, :]
raw_out[:, :, i:end] = comfy.ops.scaled_dot_product_attention(
q[:, :, i:end], k, v, attn_mask=mask_chunk, dropout_p=0.0, is_causal=False, **sdpa_extra
)
if skip_output_reshape:
out = raw_out
else: else:
out = raw_out.transpose(1, 2).reshape(b, -1, heads * dim_head) slice_size = math.ceil(q.shape[2] / steps)
raw_out = torch.empty((b, q.shape[1], q.shape[2], dim_head), dtype=q.dtype, layout=q.layout, device=q.device)
for i in range(0, q.shape[2], slice_size):
end = i + slice_size
mask_chunk = mask
if mask is not None and mask.shape[-2] != 1:
mask_chunk = mask[..., i:end, :]
raw_out[:, :, i:end] = comfy.ops.scaled_dot_product_attention(
q[:, :, i:end], k, v, attn_mask=mask_chunk, dropout_p=0.0, is_causal=False, **sdpa_extra
)
if skip_output_reshape:
out = raw_out
else:
out = raw_out.transpose(1, 2).reshape(b, -1, heads * dim_head)
else: else:
out = torch.empty((b, q.shape[2], heads * dim_head), dtype=q.dtype, layout=q.layout, device=q.device) out = torch.empty((b, q.shape[2], heads * dim_head), dtype=q.dtype, layout=q.layout, device=q.device)
for i in range(0, b, SDP_BATCH_LIMIT): for i in range(0, b, SDP_BATCH_LIMIT):