Skip to content

[ROCm][Perf] Replace cat to bmm's inplace write when aiter enabled #30611

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
ganyi1996ppo wants to merge 2 commits into
base: main
Choose a base branch
from
+25 −23
Open

[ROCm][Perf] Replace cat to bmm's inplace write when aiter enabled #30611

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
a6da514
replace cat in decode to bmm's inplace write
ganyi1996ppo Dec 13, 2025
ee1bf38
change mla back for safety
ganyi1996ppo Dec 13, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
48 changes: 25 additions & 23 deletions vllm/v1/attention/backends/mla/common.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -2007,14 +2007,29 @@ def forward(
decode_pe_padded.copy_(decode_q_pe)
decode_q_pe = decode_pe_padded

decode_q_shape = (
decode_q.shape[0],
self.num_heads,
self.kv_lora_rank + self.qk_rope_head_dim,
)
decode_q = torch.empty(
decode_q_shape,
device=decode_q_nope.device,
dtype=decode_q_nope.dtype,
)

if self.is_aiter_triton_fp8_bmm_enabled:
# Aiter path adopt strided write in bmm to prevent one
# additional index_copy or cat
decode_q[..., self.kv_lora_rank :].copy_(decode_q_pe)
# Multiply+Transpose (N, B, P)x(N, P, L)->(N, B, L)->(B, N, L)
decode_ql_nope = rocm_aiter_ops.triton_fp8_bmm(
decode_q_nope,
self.W_K,
self.W_K_scale,
group_size=128,
transpose_bm=True,
YQ=decode_q[..., : self.kv_lora_rank],
)
else:
# Pads the head_dim if necessary (for the underlying kernel)
Expand All @@ -2035,36 +2050,23 @@ def forward(
# Convert from (N, B, L) to (B, N, L)
decode_ql_nope = decode_ql_nope.transpose(0, 1)

if fp8_attention:
ql_nope_shape = decode_ql_nope.shape
q_pe_shape = decode_q_pe.shape
assert decode_ql_nope.shape[0] == decode_q_pe.shape[0]
assert decode_ql_nope.shape[1] == decode_q_pe.shape[1]
decode_q_shape = (
ql_nope_shape[0],
ql_nope_shape[1],
ql_nope_shape[2] + q_pe_shape[2],
)
# Using empty and copy since torch.cat introduces significant overhead.
decode_q0 = torch.empty(
decode_q_shape,
device=decode_ql_nope.device,
dtype=decode_ql_nope.dtype,
)
decode_q0[..., : ql_nope_shape[2]].copy_(decode_ql_nope)
decode_q0[..., ql_nope_shape[2] :].copy_(decode_q_pe)
if fp8_attention or self.dcp_world_size > 1:
# Make sure continuous tensor for fp8 or dcp case
decode_q[..., : decode_ql_nope.shape[2]].copy_(decode_ql_nope)
decode_q[..., decode_ql_nope.shape[2] :].copy_(decode_q_pe)
else:
# Otherwise, use tuple to avoid extra copy
decode_q = (decode_ql_nope, decode_q_pe)

if fp8_attention:
decode_q, _ = ops.scaled_fp8_quant(
decode_q0.view(decode_q_shape[0], -1),
decode_q.view(decode_q_shape[0], -1),
layer._q_scale,
)
decode_q = decode_q.view(decode_q_shape)
else:
decode_q = (decode_ql_nope, decode_q_pe)

if self.dcp_world_size > 1:
assert not fp8_attention, "DCP not support fp8 kvcache now."
# concatenate decode_ql_nope and decode_q_pe -> (B, N, L + P)
decode_q = torch.cat(decode_q, dim=-1)
# decode_q do allgather in head dim.
decode_q = get_dcp_group().all_gather(decode_q, dim=1)

Expand Down