-
Notifications
You must be signed in to change notification settings - Fork 29
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.
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
Aot change merge #549
base: triton-mlir
Are you sure you want to change the base?
Aot change merge #549
Conversation
@@ -544,132 +591,198 @@ def _bwd_kernel_dk_dv( | |||
stride_qz, stride_qh, stride_qm, stride_qk, | |||
stride_kz, stride_kh, stride_kn, stride_kk, | |||
stride_vz, stride_vh, stride_vk, stride_vn, | |||
stride_oz, stride_oh, stride_om, stride_ok, |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
dkdv kernel needs strides from do, dk and dv (o's not used, contradicting to its name)
q = tl.load(Q_block_ptr, boundary_check=(0,1), padding_option="zero") | ||
do = tl.load(DO_block_ptr, boundary_check=(0,1), padding_option="zero") | ||
else: | ||
q = tl.load(Q_block_ptr, boundary_check=(0,), padding_option="zero") |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Needs one more level of branching.
However if we are aiming for the performance we should consider commenting out the boundary_check
for now
@@ -680,82 +793,118 @@ def _bwd_kernel_dq( | |||
stride_qz, stride_qh, stride_qm, stride_qk, | |||
stride_kz, stride_kh, stride_kn, stride_kk, | |||
stride_vz, stride_vh, stride_vk, stride_vn, | |||
seqlen_q, seqlen_k, dropout_p, philox_seed, philox_offset_base, | |||
stride_oz, stride_oh, stride_om, stride_ok, |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Similar to dkdv, dq kernel needs strides from do and dq (again, not strides from o)
strides=(stride_qm, stride_qk), | ||
offsets=(start_m, 0), | ||
block_shape=(BLOCK_M, BLOCK_DMODEL), | ||
order=(1, 0) | ||
) | ||
tl.store(DQ_block_ptr, (dq * sm_scale).to(DQ_block_ptr.type.element_ty)) | ||
tl.store(DQ_block_ptr, (dq * sm_scale).to(DQ_block_ptr.type.element_ty), boundary_check=(0,1)) |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
tl.store also cause some performance penalties with boundary checks, although it shouldn't.
q, k, v, ctx.sm_scale, | ||
o, do_scaled, | ||
dk, dv, | ||
L, delta, | ||
q.stride(0), q.stride(1), q.stride(2), q.stride(3), | ||
k.stride(0), k.stride(1), k.stride(2), k.stride(3), | ||
v.stride(0), v.stride(1), v.stride(2), v.stride(3), | ||
o.stride(0), o.stride(1), o.stride(2), o.stride(3), |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
One more reminder, o.strides are not used by backward kernels.
@@ -893,28 +1046,41 @@ def backward(ctx, do, _): | |||
seqlen_q = q.shape[2] |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
remove these two assertions above, since they are not need anymore.
q = torch.empty((Z, H, N_CTX, D_HEAD), dtype=dtype, device="cuda").normal_(mean=0., std=0.5).requires_grad_() | ||
k = torch.empty((Z, H, N_CTX, D_HEAD), dtype=dtype, device="cuda").normal_(mean=0., std=0.5).requires_grad_() | ||
v = torch.empty((Z, H, N_CTX, D_HEAD), dtype=dtype, device="cuda").normal_(mean=0., std=0.5).requires_grad_() | ||
dropout_p = 0 |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
dropout_p = 0?
@@ -1186,45 +1361,88 @@ def test_op_varlen_mqa_fwd(Z, HQ, HK, N_CTX, D_HEAD, causal, dtype=torch.float16 | |||
(4, 48, 2048, 64), | |||
(4, 48, 4096, 64), | |||
(1, 16, 8192, 64), | |||
(1, 16, 128, 32), |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
One major concern is the UT coverage here.
Merge in AOTriton backwards kernel changes
Bring in changes from AOT, unedited.