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

Revert "[KVCACHE] Improved schedule for prefill attention" #17466

Merged
merged 1 commit into from
Oct 15, 2024
Merged

Revert "[KVCACHE] Improved schedule for prefill attention" #17466

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
60 changes: 11 additions & 49 deletions python/tvm/relax/frontend/nn/llm/kv_cache.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -925,12 +925,8 @@ def _attention_decode(

THREAD_LIMIT = 512
TILE_SIZE_PER_BDX = 2
if target.kind.name == "opencl" and (
("android" in str(target.host)) or ("adreno" in str(target.attrs))
):
# Keeping lower thread limit for this kernel on adreno target
# to avoid register spill
THREAD_LIMIT = 256
if target.kind.name == "opencl" and "android" in str(target.host):
THREAD_LIMIT = 256 if H_kv < 8 else 512
TILE_SIZE_PER_BDX = 1
max_num_threads_per_block = get_max_num_threads_per_block(target)
thread_limit = min(max_num_threads_per_block, THREAD_LIMIT)
Expand Down Expand Up @@ -1574,11 +1570,7 @@ def _attention_prefill_ragged(h_kv, h_q, d, dtype, rope_scaling: Dict[str, Any],

bdx = 32
num_warps = 4
tile_x, tile_y, tile_z = (
64 // ((DataType(dtype).bits + 7) // 8) // max(d // 128, 1),
d,
64 // ((DataType(dtype).bits + 7) // 8) // max(d // 128, 1),
)
tile_x, tile_y, tile_z = 64 // ((DataType(dtype).bits + 7) // 8) // max(d // 128, 1), d, 16

# Otherwise we would exceed maxComputeWorkgroupStorageSize
if (
Expand All @@ -1588,12 +1580,6 @@ def _attention_prefill_ragged(h_kv, h_q, d, dtype, rope_scaling: Dict[str, Any],
tile_z = 8
num_warps = 2

if target.kind.name == "opencl" and (
("android" in str(target.host)) or ("adreno" in str(target.attrs))
):
LOAD_VEC = 16 // ((DataType(dtype).bits + 7) // 8) # 16 bytes
NUM_BLKS = group_size * 8

# fmt: off
@T.prim_func
def batch_prefill_ragged_kv( # pylint: disable=too-many-branches
Expand Down Expand Up @@ -1722,6 +1708,8 @@ def batch_prefill_ragged_kv( # pylint: disable=too-many-branches
for lz, ly in T.grid(tile_z, tile_y):
with T.block("K_load"):
i, j = T.axis.remap("SS", [lz, ly])
T.reads()
T.writes()
cur_L = L_kv_start + i
if cur_L < kv_chunk_len[0]:
K_smem[i, j] = T.if_then_else(
Expand Down Expand Up @@ -1836,14 +1824,6 @@ def batch_prefill_ragged_kv( # pylint: disable=too-many-branches
# fmt: on
# pylint: enable=line-too-long,too-many-branches
sch = tir.Schedule(batch_prefill_ragged_kv)
get_extent = lambda *lps: [int(sch.get(lp).extent) for lp in lps]

def get_vecsize(extent):
return min(LOAD_VEC, (extent & ~(extent - 1)))

def getxy_vecsize(x, y, t):
assert (x * y) % t == 0
return min(get_vecsize(y), get_vecsize(x * y // t))

def get_tile_size(x, y, t):
cnt = (x * y) // t
Expand All @@ -1857,37 +1837,26 @@ def get_tile_size(x, y, t):

def apply_to_qkv_load(sch: tir.Schedule, block):
loop_x, loop_y = sch.get_loops(block)[-2:]
x_extent, y_extent = get_extent(loop_x, loop_y)
vec_size = getxy_vecsize(x_extent, y_extent, bdx * num_warps)
yo, yv = sch.split(loop_y, [None, vec_size])
yo_extent = y_extent // vec_size
tile_x, tile_y = get_tile_size(x_extent, yo_extent, (bdx * num_warps))
xo, xi = sch.split(loop_x, [tile_x, None])
yo, yi = sch.split(yo, [tile_y, None])
sch.reorder(xi, yi, xo, yo)
t = sch.fuse(xi, yi)
ty, tx = sch.split(t, [num_warps, bdx])
loop = sch.fuse(loop_x, loop_y)
_, ty, tx, vec = sch.split(
loop, factors=[None, num_warps, bdx, LOAD_VEC], preserve_unit_iters=True
)
sch.bind(ty, "threadIdx.y")
sch.bind(tx, "threadIdx.x")
sch.vectorize(yv)
sch.vectorize(vec)

def apply_to_so_ewise(sch: tir.Schedule, block, tile):
loop_x, loop_y = sch.get_loops(block)[-2:]
xo, xi = sch.split(loop_x, factors=[None, tile[0]])
yo, yi = sch.split(loop_y, factors=[None, tile[1]])
sch.reorder(xo, yo, xi, yi)
sch.unroll(xi)
yiv_extent = get_vecsize(tile[1])
yio, yiv = sch.split(yi, [None, yiv_extent])
sch.unroll(yio)
sch.vectorize(yiv)
t = sch.fuse(xo, yo)
ty, tx = sch.split(t, factors=[None, bdx])
sch.bind(ty, "threadIdx.y")
sch.bind(tx, "threadIdx.x")

def apply_to_gemm( # pylint: disable=unused-argument
sch: tir.Schedule, block, tile, read_0, read_1, r_len=16, k_major=False
sch: tir.Schedule, block, tile, read_0, read_1, r_len=8, k_major=False
):
loop_x, loop_y, loop_z = sch.get_loops(block)[-3:]
xo, xi = sch.split(loop_x, factors=[None, tile[0]])
Expand All @@ -1903,12 +1872,6 @@ def apply_to_gemm( # pylint: disable=unused-argument
sch.reorder(ko, xi, yi, ki)
else:
sch.reorder(ko, ki, xi, yi)
yiv_extent = get_vecsize(tile[1])
yio, yiv = sch.split(yi, [None, yiv_extent])
sch.unroll(yio)
sch.vectorize(yiv)
sch.unroll(xi)
sch.unroll(ki)
sch.decompose_reduction(block, ty)

def apply_to_md(sch, block):
Expand All @@ -1917,7 +1880,6 @@ def apply_to_md(sch, block):
sch.bind(ty, "threadIdx.y")
sch.bind(tx, "threadIdx.x")

sch.transform_layout("K_load", ("write", 0), lambda i, j: (j, i))
tile_s = get_tile_size(tile_x, tile_z, bdx * num_warps)
tile_o = get_tile_size(tile_x, tile_y, bdx * num_warps)
apply_to_gemm(sch, sch.get_block("S_gemm"), tile_s, 0, 1, k_major=True)
Expand Down
Loading