Skip to content
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

Jump to bottom

Divide inputs using BlockSpecs when possible #23287

Merged
merged 1 commit into from
Aug 28, 2024
Merged

Divide inputs using BlockSpecs when possible #23287

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
58 changes: 26 additions & 32 deletions jax/experimental/pallas/ops/gpu/attention.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -41,7 +41,7 @@ def mha_forward_kernel(
block_d: int,
block_k: int,
):
seq_len = q_ref.shape[0]
seq_len = k_ref.shape[0]
start_q = pl.program_id(0)

# o is the buffer where we accumulate the output on sram.
Expand All @@ -55,7 +55,7 @@ def mha_forward_kernel(
# read, compute, and write all in 2d chunks. 1 element ~= 1 CUDA thread index.
# q tile has shape [block_q, block_d], block_d == head_dim.
curr_q_slice = pl.dslice(start_q * block_q, block_q)
q = pl.load(q_ref, (curr_q_slice, pl.dslice(None)))
q = q_ref[...]
q_segment_ids = (
None
if segment_ids_ref is None
Expand Down Expand Up @@ -123,12 +123,9 @@ def body(start_k, carry):

if residual_refs:
lse_ref = residual_refs[0]
lse_i = m_i + jnp.log(l_i)
pl.store(lse_ref, (curr_q_slice,), lse_i)
lse_ref[...] = m_i + jnp.log(l_i)
# Write output to dram.
o = o.astype(o_ref.dtype)
pl.store(o_ref, (curr_q_slice, pl.dslice(None)), o)

o_ref[...] = o.astype(o_ref.dtype)

def segment_mask(
q_segment_ids: jax.Array,
Expand Down Expand Up @@ -197,7 +194,7 @@ def mha(

in_specs = [
pl.BlockSpec(
(None, seq_len, None, head_dim), lambda _, j, k: (j, 0, k, 0)
(None, block_q, None, head_dim), lambda i, j, k: (j, i, k, 0)
),
pl.BlockSpec(
(None, seq_len, None, head_dim), lambda _, j, k: (j, 0, k, 0)
Expand All @@ -217,7 +214,7 @@ def mha(
grid=grid_,
in_specs=in_specs,
out_specs=pl.BlockSpec(
(None, seq_len, None, head_dim), lambda _, j, k: (j, 0, k, 0)
(None, block_q, None, head_dim), lambda i, j, k: (j, i, k, 0)
),
compiler_params=dict(
triton=dict(num_warps=num_warps_, num_stages=num_stages)
Expand Down Expand Up @@ -268,7 +265,7 @@ def _mha_forward(
]
in_specs = [
pl.BlockSpec(
(None, seq_len, None, head_dim), lambda _, j, k: (j, 0, k, 0)
(None, block_q, None, head_dim), lambda i, j, k: (j, i, k, 0)
),
pl.BlockSpec(
(None, seq_len, None, head_dim), lambda _, j, k: (j, 0, k, 0)
Expand All @@ -288,9 +285,9 @@ def _mha_forward(
in_specs=in_specs,
out_specs=[
pl.BlockSpec(
(None, seq_len, None, head_dim), lambda _, j, k: (j, 0, k, 0)
(None, block_q, None, head_dim), lambda i, j, k: (j, i, k, 0)
),
pl.BlockSpec((None, None, seq_len), lambda _, j, k: (j, k, 0)),
pl.BlockSpec((None, None, block_q), lambda i, j, k: (j, k, i)),
],
compiler_params=dict(
triton=dict(num_warps=num_warps_, num_stages=num_stages)
Expand All @@ -303,35 +300,32 @@ def _mha_forward(
return out, (q, k, v, segment_ids, out, lse)


def _preprocess_backward_kernel(out_ref, dout_ref, delta_ref, *, block_q: int):
pid_m = pl.program_id(0)

off_m = pl.ds(pid_m * block_q, block_q)
def _preprocess_backward_kernel(out_ref, dout_ref, delta_ref):
# load
o = pl.load(out_ref, (off_m, slice(None))).astype(jnp.float32)
do = pl.load(dout_ref, (off_m, slice(None))).astype(jnp.float32)
o = out_ref[...].astype(jnp.float32)
do = dout_ref[...].astype(jnp.float32)
# compute
delta = jnp.sum(o * do, axis=1)
# write-back
pl.store(delta_ref, (off_m,), delta.astype(delta_ref.dtype))
delta_ref[...] = delta.astype(delta_ref.dtype)

@jax.named_scope("preprocess_backward")
def _preprocess_backward(out, do, lse, block_q: int,
debug: bool, interpret: bool):
batch_size, seq_len, num_heads, head_dim = out.shape
out_shape = jax.ShapeDtypeStruct(lse.shape, lse.dtype)
delta = pl.pallas_call(
functools.partial(_preprocess_backward_kernel, block_q=block_q),
_preprocess_backward_kernel,
grid=(pl.cdiv(seq_len, block_q), batch_size, num_heads),
in_specs=[
pl.BlockSpec(
(None, seq_len, None, head_dim), lambda _, j, k: (j, 0, k, 0)
(None, block_q, None, head_dim), lambda i, j, k: (j, i, k, 0)
),
pl.BlockSpec(
(None, seq_len, None, head_dim), lambda _, j, k: (j, 0, k, 0)
(None, block_q, None, head_dim), lambda i, j, k: (j, i, k, 0)
),
],
out_specs=pl.BlockSpec((None, None, seq_len), lambda _, j, k: (j, k, 0)),
out_specs=pl.BlockSpec((None, None, block_q), lambda i, j, k: (j, k, i)),
compiler_params=dict(triton=dict(num_warps=4, num_stages=3)),
out_shape=out_shape,
debug=debug,
Expand Down Expand Up @@ -431,8 +425,8 @@ def inner_loop_dkdv(start_q, carry):
dv, dk = lax.fori_loop(
lower_bound, pl.cdiv(seq_len, block_q1), inner_loop_dkdv, (dv, dk)
)
pl.store(dv_ref, (curr_k_slice, slice(None)), dv.astype(dv_ref.dtype))
pl.store(dk_ref, (curr_k_slice, slice(None)), dk.astype(dk_ref.dtype))
dv_ref[...] = dv.astype(dv_ref.dtype)
dk_ref[...] = dk.astype(dk_ref.dtype)

del dv, dk

Expand Down Expand Up @@ -495,7 +489,7 @@ def inner_loop_dq(start_k, dq):
upper_bound = pl.cdiv(seq_len, block_k2)

dq = lax.fori_loop(0, upper_bound, inner_loop_dq, (dq))
pl.store(dq_ref, (curr_q_slice, slice(None)), dq.astype(dq_ref.dtype))
dq_ref[...] = dq.astype(dq_ref.dtype)


def _mha_backward(sm_scale: float, causal: bool, block_q: int, block_k: int,
Expand Down Expand Up @@ -566,16 +560,16 @@ def _mha_backward(sm_scale: float, causal: bool, block_q: int, block_k: int,
grid=grid,
out_specs=[
pl.BlockSpec(
(None, seq_len, None, head_dim),
lambda i, j, _: (i, 0, j, 0), # dq
(None, block_q, None, head_dim),
lambda i, j, k: (i, k, j, 0), # dq
),
pl.BlockSpec(
(None, seq_len, None, head_dim),
lambda i, j, _: (i, 0, j, 0), # dk
(None, block_k, None, head_dim),
lambda i, j, k: (i, k, j, 0), # dk
),
pl.BlockSpec(
(None, seq_len, None, head_dim),
lambda i, j, _: (i, 0, j, 0), # dv
(None, block_k, None, head_dim),
lambda i, j, k: (i, k, j, 0), # dv
),
],
name="mha_backward",
Expand Down