Skip to content

[gdpa] Turn on on-host TMA for TLX fwd #365

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

[gdpa] Turn on on-host TMA for TLX fwd #365

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
8 changes: 4 additions & 4 deletions tritonbench/operators/gdpa/gdpa_blackwell_tlx.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1217,8 +1217,8 @@ def gdpa_forward_tlx(
NUM_SMS = (
get_num_sms() or 1000000
) * 8 # if num sms is None, use a large number so that it is a no-op
print("NUM_SMS", NUM_SMS)
print(triton.cdiv(max_seq_len_q, 256) * BATCH * nheads)
# print("NUM_SMS", NUM_SMS)
# print(triton.cdiv(max_seq_len_q, 256) * BATCH * nheads)

q = expect_contiguous(query)
k = expect_contiguous(key)
Expand All @@ -1233,7 +1233,7 @@ def gdpa_forward_tlx(
H = nheads
y_dim = N_CTX_KV * Z
x_dim = HEAD_DIM * H // G
USE_ON_DEVICE_TMA = True
USE_ON_DEVICE_TMA = False
if not USE_ON_DEVICE_TMA:
desc_q = TensorDescriptor(
q,
Expand Down Expand Up @@ -1268,7 +1268,7 @@ def grid_tma_persistent(META):
)

activation_enum_int = activation_string_to_int(activation)
print(q.shape, k.shape, v.shape)
# print(q.shape, k.shape, v.shape)
# print("activation_enum_int", activation, activation_enum_int)
# print(query_offset)
# print(key_offset)
Expand Down
Loading