Skip to content

Enable VECT_MUL for blackwell attention #536

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
wants to merge 1 commit into
base: main
Choose a base branch
from
Open

Enable VECT_MUL for blackwell attention #536

Changes from all commits
Commits
Show all changes
1 commit
Select commit
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
5 changes: 3 additions & 2 deletions tritonbench/kernels/blackwell_triton_fused_attention_dp.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -72,7 +72,8 @@ def _attn_fwd_subtile(
qk -= m_ij[:, None]
else:
m_ij = tl.maximum(m_i, tl.max(qk, 1) * qk_scale)
if VECT_MUL:
# TODO: Figure out why vector FMA slows things down.
if VECT_MUL and False:
qk = _fma_f32x2(qk, qk_scale, -m_ij[:, None])
Copy link
Contributor

Choose a reason for hiding this comment

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

Wondering if we should make VECT_MUL an integer with each bit representing one vectorization so we can autotune.

Copy link
Contributor

Choose a reason for hiding this comment

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

I like that idea.

else:
qk = qk * qk_scale - m_ij[:, None]
Expand Down Expand Up @@ -262,7 +263,7 @@ def make_tile_config(BM, BN, occ, subtile, vectmul, add2reduce):
for BN in [64, 128]
for occ in [1, 2]
for subtile in [True]
for vectmul in [False]
for vectmul in [True]
for add2reduce in [False]
]
else:
Expand Down