import torch
import numpy as np
import math
import triton
import triton.language as tl
import pycuda.autoprimaryctx
from pycuda.compiler import SourceModule
from flash_attn import flash_attn_varlen_func
# @triton.autotune(
# configs=[
# triton.Config({}, num_stages=1, num_warps=4),
# triton.Config({}, num_stages=1, num_warps=8),
# triton.Config({}, num_stages=2, num_warps=4),
# triton.Config({}, num_stages=2, num_warps=8),
# triton.Config({}, num_stages=3, num_warps=4),
# triton.Config({}, num_stages=3, num_warps=8),
# triton.Config({}, num_stages=4, num_warps=4),
# triton.Config({}, num_stages=4, num_warps=8),
# triton.Config({}, num_stages=5, num_warps=4),
# triton.Config({}, num_stages=5, num_warps=8),
# ],
# key=['N_CTX'],
# )
@triton.jit
def triton_sparse_fwd_kernel(
Q, K, V, seqlens, sm_scale,
block_count, block_offset, column_count, column_index,
Out,
stride_qz, stride_qh, stride_qm, stride_qk,
stride_kz, stride_kh, stride_kn, stride_kk,
stride_vz, stride_vh, stride_vn, stride_vk,
stride_oz, stride_oh, stride_om, stride_ok,
Z, H, N_CTX,
NUM_ROWS, NNZ_S, NNZ_V,
BLOCK_M: tl.constexpr,
BLOCK_N: tl.constexpr,
BLOCK_DMODEL: tl.constexpr,
dtype: tl.constexpr,
):
start_m = tl.program_id(0)
off_hz = tl.program_id(1)
seqlen = tl.load(seqlens + off_hz // H)
if start_m * BLOCK_M >= seqlen:
return
# initialize offsets
offs_m = start_m * BLOCK_M + tl.arange(0, BLOCK_M)
offs_n = tl.arange(0, BLOCK_N)
offs_d = tl.arange(0, BLOCK_DMODEL)
qo_offset = (off_hz // H) * stride_qz + (off_hz % H) * stride_qh
kv_offset = (off_hz // H) * stride_kz + (off_hz % H) * stride_kh
q_ptrs = Q + qo_offset + offs_m[:, None] * stride_qm + offs_d[None, :] * stride_qk
k_ptrs = K + kv_offset + offs_d[:, None] * stride_kk
v_ptrs = V + kv_offset + offs_d[None, :] * stride_vk
o_ptrs = Out + qo_offset + offs_m[:, None] * stride_om + offs_d[None, :] * stride_ok
num_blks = tl.load(block_count + off_hz * NUM_ROWS + start_m)
blks_ptr = block_offset + (off_hz * NUM_ROWS + start_m) * NNZ_S
num_cols = tl.load(column_count + off_hz * NUM_ROWS + start_m)
cols_ptr = column_index + (off_hz * NUM_ROWS + start_m) * NNZ_V
# initialize pointer to m and l
m_i = tl.zeros([BLOCK_M], dtype=tl.float32) - float("inf")
l_i = tl.zeros([BLOCK_M], dtype=tl.float32)
acc = tl.zeros([BLOCK_M, BLOCK_DMODEL], dtype=tl.float32)
# scale sm_scale by log_2(e) and use
# 2^x instead of exp in the loop because CSE and LICM
# don't work as expected with `exp` in the loop
qk_scale = sm_scale * 1.44269504
# load q: it will stay in SRAM throughout
q = tl.load(q_ptrs)
q = (q * qk_scale).to(dtype)
# loop over k, v and update accumulator
m_mask = offs_m[:, None] < seqlen
for block_index in range(num_blks):
start_n = tl.load(blks_ptr + block_index)
cols = start_n + offs_n
n_mask = cols < seqlen
# -- load k, v --
k = tl.load(k_ptrs + cols[None, :] * stride_kn, mask=n_mask[None, :], other=0.0)
v = tl.load(v_ptrs + cols[:, None] * stride_vn, mask=n_mask[:, None], other=0.0)
# -- compute qk --
qk = tl.zeros([BLOCK_M, BLOCK_N], dtype=tl.float32)
causal_mask = cols[None, :] <= offs_m[:, None]
qk = tl.where(m_mask & causal_mask, qk, float("-inf"))
qk += tl.dot(q, k)
# -- compute scaling constant --
m_i_new = tl.maximum(m_i, tl.max(qk, 1))
alpha = tl.math.exp2(m_i - m_i_new)
p = tl.math.exp2(qk - m_i_new[:, None])
# -- scale and update acc --
acc_scale = l_i * 0 + alpha # workaround some compiler bug
acc *= acc_scale[:, None]
acc += tl.dot(p.to(dtype), v)
# -- update m_i and l_i --
l_i = l_i * alpha + tl.sum(p, 1)
m_i = m_i_new
for start_n in range(0, num_cols, BLOCK_N):
n_mask = start_n + offs_n < num_cols
cols = tl.load(cols_ptr + start_n + offs_n, mask=n_mask, other=0)
# -- load k, v --
k = tl.load(k_ptrs + cols[None, :] * stride_kn, mask=n_mask[None, :], other=0.0)
v = tl.load(v_ptrs + cols[:, None] * stride_vn, mask=n_mask[:, None], other=0.0)
# -- compute qk --
qk = tl.zeros([BLOCK_M, BLOCK_N], dtype=tl.float32)
qk = tl.where(m_mask & n_mask, qk, float("-inf"))
qk += tl.dot(q, k)
# -- compute scaling constant --
m_i_new = tl.maximum(m_i, tl.max(qk, 1))
alpha = tl.math.exp2(m_i - m_i_new)
p = tl.math.exp2(qk - m_i_new[:, None])
# -- scale and update acc --
acc_scale = l_i * 0 + alpha # workaround some compiler bug
acc *= acc_scale[:, None]
acc += tl.dot(p.to(dtype), v)
# -- update m_i and l_i --
l_i = l_i * alpha + tl.sum(p, 1)
m_i = m_i_new
# write back O
acc /= l_i[:, None]
# acc = tl.where(m_mask, acc / l_i[:, None], 0.0)
tl.store(o_ptrs, acc.to(dtype), mask=m_mask)
def triton_sparse_forward(
q: torch.Tensor, # [BATCH, N_HEADS, N_CTX, D_HEAD]
k: torch.Tensor, # [BATCH, N_HEADS, N_CTX, D_HEAD]
v: torch.Tensor, # [BATCH, N_HEADS, N_CTX, D_HEAD]
seqlens: torch.Tensor, # [BATCH, ]
block_count: torch.Tensor, # [BATCH, N_HEADS, cdiv(N_CTX, BLOCK_SIZE_M)]
block_offset: torch.Tensor, # [BATCH, N_HEADS, cdiv(N_CTX, BLOCK_SIZE_M), NNZ_S]
column_count: torch.Tensor, # [BATCH, N_HEADS, cdiv(N_CTX, BLOCK_SIZE_M)]
column_index: torch.Tensor, # [BATCH, N_HEADS, cdiv(N_CTX, BLOCK_SIZE_M), NNZ_V]
sm_scale: float,
block_size_M: int = 64,
block_size_N: int = 64,
) -> torch.Tensor:
# shape constraints
Lq, Lk, Lv = q.shape[-1], k.shape[-1], v.shape[-1]
assert Lq == Lk and Lk == Lv
assert Lk in {16, 32, 64, 128}
o = torch.zeros_like(q)
grid = (triton.cdiv(q.shape[2], block_size_M), q.shape[0] * q.shape[1], 1)
dtype = tl.bfloat16 if q.dtype == torch.bfloat16 else tl.float16
triton_sparse_fwd_kernel[grid](
q, k, v, seqlens, sm_scale,
block_count, block_offset, column_count, column_index,
o,
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),
q.shape[0], q.shape[1], q.shape[2],
block_count.shape[-1], block_offset.shape[-1], column_index.shape[-1],
BLOCK_M=block_size_M, BLOCK_N=block_size_N,
BLOCK_DMODEL=Lk,
dtype=dtype,
num_warps=4, num_stages=2,
)
return o
def torch_build_index(seqlens, vertical_indexes, slash_indexes, context_size, block_size_M=64, block_size_N=64):
device = seqlens.device
batch_size, num_heads, NNZ_S = slash_indexes.shape
NNZ_V = vertical_indexes.shape[-1]
num_rows = triton.cdiv(context_size, block_size_M)
block_count = torch.zeros((batch_size, num_heads, num_rows), dtype=torch.int32)
block_offset = torch.zeros((batch_size, num_heads, num_rows, NNZ_S), dtype=torch.int32)
column_count = torch.zeros((batch_size, num_heads, num_rows), dtype=torch.int32)
column_index = torch.zeros((batch_size, num_heads, num_rows, NNZ_V), dtype=torch.int32)
for b in range(batch_size):
seqlen = seqlens[b]
for h in range(num_heads):
for m, start_m in enumerate(range(0, seqlen, block_size_M)):
end_m = start_m + block_size_M
s = 0
while slash_indexes[b, h, s] >= end_m:
s += 1
s_idx = max(end_m - slash_indexes[b, h, s], block_size_M)
s += 1
range_start = s_idx - block_size_M
range_end = s_idx
tmp_blocks = []
while s < NNZ_S:
s_idx = max(end_m - slash_indexes[b, h, s], block_size_M)
if s_idx > range_end + block_size_M:
tmp_blocks += list(range(range_start, range_end, block_size_N))
range_start = s_idx - block_size_M
range_end = s_idx
elif s_idx > range_end:
range_end += block_size_M
s += 1
tmp_blocks += list(range(range_start, range_end, block_size_N))
block_count[b, h, m] = len(tmp_blocks)
block_offset[b, h, m, :len(tmp_blocks)] = torch.tensor(tmp_blocks, dtype=block_offset.dtype)
tmp_columns = vertical_indexes[b, h].cpu().numpy().tolist()
tmp_columns = [col for col in tmp_columns if col < range_end]
for range_start in tmp_blocks:
range_end = range_start + block_size_N
tmp_columns = [col for col in tmp_columns if col < range_start or col >= range_end]
column_count[b, h, m] = len(tmp_columns)
column_index[b, h, m, :len(tmp_columns)] = torch.tensor(tmp_columns, dtype=block_offset.dtype)
return block_count.to(device), block_offset.to(device), column_count.to(device), column_index.to(device)
PYCUDA_BUILD_INDEX_KERNEL_CODE = '''\
__device__ int min(int x, int y) {
return x < y ? x : y;
}
__device__ int max(int x, int y) {
return x > y ? x : y;
}
__device__ void save_blocks(int* block_offset, int range_start, int range_end, int block_size, int& block_count) {
for (int idx = range_start; idx < range_end; idx += block_size) {
block_offset[block_count++] = idx;
}
}
__global__ void PYCUDA_BUILD_INDEX_KERNEL(
const int* seqlens, // [BATCH, ]
const int* vertical_indexes, // [BATCH, N_HEADS, NNZ_V]
const int* slash_indexes, // [BATCH, N_HEADS, NNZ_S]
int* block_count, // [BATCH, N_HEADS, cdiv(N_CTX, BLOCK_SIZE_M)]
int* block_offset, // [BATCH, N_HEADS, cdiv(N_CTX, BLOCK_SIZE_M), NNZ_S]
int* column_count, // [BATCH, N_HEADS, cdiv(N_CTX, BLOCK_SIZE_M)]
int* column_index, // [BATCH, N_HEADS, cdiv(N_CTX, BLOCK_SIZE_M), NNZ_V]
int N_HEADS,
int N_ROWS,
int BLOCK_SIZE_M,
int BLOCK_SIZE_N,
int NNZ_V,
int NNZ_S
) {
const int batch_idx = blockIdx.y;
const int head_idx = blockIdx.x;
const int group_idx = blockIdx.z;
int seqlen = seqlens[batch_idx];
int block_idx_m = group_idx * blockDim.x + threadIdx.x;
int start_m = block_idx_m * BLOCK_SIZE_M;
if (start_m >= seqlen) {
return;
}
int end_m = start_m + BLOCK_SIZE_M;
vertical_indexes += (batch_idx * N_HEADS + head_idx) * NNZ_V;
slash_indexes += (batch_idx * N_HEADS + head_idx) * NNZ_S;
int row_offset = (batch_idx * N_HEADS + head_idx) * N_ROWS + block_idx_m;
block_count += row_offset;
block_offset += row_offset * NNZ_S;
column_count += row_offset;
column_index += row_offset * NNZ_V;
int tmp_col_cnt = 0, tmp_blk_cnt = 0;
int s = 0, v = 0;
int v_idx = vertical_indexes[v++];
int s_idx = slash_indexes[s++];
while (s_idx >= end_m) {
s_idx = slash_indexes[s++];
}
s_idx = max(end_m - s_idx, BLOCK_SIZE_M);
int range_start = s_idx - BLOCK_SIZE_M, range_end = s_idx;
while (1) {
if (v_idx < range_end) {
if (v_idx < range_start) {
column_index[tmp_col_cnt++] = v_idx;
}
if (v < NNZ_V) {
v_idx = vertical_indexes[v++];
} else {
v_idx = end_m + BLOCK_SIZE_M;
}
} else {
if (s < NNZ_S) {
s_idx = max(end_m - slash_indexes[s++], BLOCK_SIZE_M);
} else {
save_blocks(block_offset, range_start, range_end, BLOCK_SIZE_N, tmp_blk_cnt);
break;
}
if (s_idx > range_end + BLOCK_SIZE_M) {
save_blocks(block_offset, range_start, range_end, BLOCK_SIZE_N, tmp_blk_cnt);
range_start = s_idx - BLOCK_SIZE_M;
range_end = s_idx;
} else if (s_idx > range_end) {
range_end += BLOCK_SIZE_M;
}
}
}
block_count[0] = tmp_blk_cnt;
column_count[0] = tmp_col_cnt;
}
'''
PYCUDA_BUILD_INDEX_KERNEL = SourceModule(
PYCUDA_BUILD_INDEX_KERNEL_CODE,
options=['-std=c++14', '-O3'],
).get_function(f'PYCUDA_BUILD_INDEX_KERNEL')
def pycuda_build_index(seqlens, vertical_indexes, slash_indexes, context_size, block_size_M=64, block_size_N=64):
batch_size, num_heads, NNZ_S = slash_indexes.shape
NNZ_V = vertical_indexes.shape[-1]
num_rows = triton.cdiv(context_size, block_size_M)
block_count = torch.zeros((batch_size, num_heads, num_rows), dtype=torch.int32, device=seqlens.device)
block_offset = torch.zeros((batch_size, num_heads, num_rows, NNZ_S), dtype=torch.int32, device=seqlens.device)
column_count = torch.zeros((batch_size, num_heads, num_rows), dtype=torch.int32, device=seqlens.device)
column_index = torch.zeros((batch_size, num_heads, num_rows, NNZ_V), dtype=torch.int32, device=seqlens.device)
num_threads = 64
# import ipdb; ipdb.set_trace()
PYCUDA_BUILD_INDEX_KERNEL(
seqlens, vertical_indexes, slash_indexes,
block_count, block_offset, column_count, column_index,
np.int32(num_heads), np.int32(num_rows),
np.int32(block_size_M), np.int32(block_size_N),
np.int32(NNZ_V), np.int32(NNZ_S),
# grid=(triton.cdiv(num_rows, num_threads), N_HEADS, BATCH),
grid=(num_heads, batch_size, triton.cdiv(num_rows, num_threads)),
block=(num_threads, 1, 1),
)
return block_count, block_offset, column_count, column_index
def make_causal_mask(seqlens, device, context_size):
batch_size = seqlens.shape[0]
arange = torch.arange(context_size, dtype=torch.int32, device=device)
causal_mask = arange[None, None, :, None] >= arange[None, None, None, :]
causal_mask = causal_mask.repeat((batch_size, 1, 1, 1))
for b, seqlen in enumerate(seqlens):
causal_mask[b, :, seqlen:, :] = False
causal_mask[b, :, :, seqlen:] = False
return causal_mask
def make_finegrained_mask(vertical_indexes, slash_indexes, causal_mask, device):
batch_size, num_heads, _ = vertical_indexes.shape
context_size = causal_mask.shape[-1]
arange = torch.arange(context_size, dtype=torch.int32, device=device)
sparse_mask = torch.zeros((batch_size, num_heads, context_size, context_size), dtype=torch.bool, device=device)
for b in range(batch_size):
for h in range(num_heads):
for vertical_index in vertical_indexes[b, h]:
sparse_mask[b, h, :, vertical_index] = True
for slash_index in slash_indexes[b, h]:
sparse_mask[b, h].logical_or_(arange[:, None] - arange[None, :] == slash_index)
sparse_mask.logical_and_(causal_mask)
return sparse_mask
def make_block_mask(
block_count: torch.Tensor,
block_offset: torch.Tensor,
column_count: torch.Tensor,
column_index: torch.Tensor,
seqlens: torch.Tensor,
causal_mask: torch.Tensor,
device: torch.device,
block_size_M: int = 64,
block_size_N: int = 64.
):
batch_size, num_heads, _ = block_count.shape
context_size = causal_mask.shape[-1]
block_mask = torch.zeros((batch_size, num_heads, context_size, context_size), dtype=torch.bool, device=device)
for b in range(batch_size):
for h in range(num_heads):
for m, start_m in enumerate(range(0, seqlens[b], block_size_M)):
end_m = start_m + block_size_M
for col_idx in range(column_count[b, h, m]):
block_mask[b, h, start_m:end_m, column_index[b, h, m, col_idx]] = True
for blk_idx in range(block_count[b, h, m]):
blk_start = block_offset[b, h, m, blk_idx].item()
blk_end = blk_start + block_size_N
block_mask[b, h, start_m:end_m, blk_start:blk_end] = True
block_mask.logical_and_(causal_mask)
return block_mask
def plot_mask(mask, name, batch=0, head=0):
import matplotlib.pyplot as plt
import seaborn as sns
plt.figure(figsize=(16, 12))
plt.clf()
mask = mask[batch, head].cpu().numpy()
sns.heatmap(mask)
plt.savefig(name)
@triton.jit
def triton_dense_fwd_kernel(
Q, K, V, seqlens, sm_scale,
Out,
stride_qz, stride_qh, stride_qm, stride_qk,
stride_kz, stride_kh, stride_kn, stride_kk,
stride_vz, stride_vh, stride_vn, stride_vk,
stride_oz, stride_oh, stride_om, stride_ok,
Z, H, N_CTX,
BLOCK_M: tl.constexpr, BLOCK_DMODEL: tl.constexpr,
BLOCK_N: tl.constexpr,
dtype: tl.constexpr,
):
start_m = tl.program_id(0)
off_hz = tl.program_id(1)
seqlen = tl.load(seqlens + off_hz // H)
if start_m * BLOCK_M >= seqlen:
return
qo_offset = (off_hz // H) * stride_qz + (off_hz % H) * stride_qh
kv_offset = (off_hz // H) * stride_kz + (off_hz % H) * stride_kh
Q_block_ptr = tl.make_block_ptr(
base=Q + qo_offset,
shape=(N_CTX, BLOCK_DMODEL),
strides=(stride_qm, stride_qk),
offsets=(start_m * BLOCK_M, 0),
block_shape=(BLOCK_M, BLOCK_DMODEL),
order=(1, 0)
)
K_block_ptr = tl.make_block_ptr(
base=K + kv_offset,
shape=(BLOCK_DMODEL, N_CTX),
strides=(stride_kk, stride_kn),
offsets=(0, 0),
block_shape=(BLOCK_DMODEL, BLOCK_N),
order=(0, 1)
)
V_block_ptr = tl.make_block_ptr(
base=V + kv_offset,
shape=(N_CTX, BLOCK_DMODEL),
strides=(stride_vn, stride_vk),
offsets=(0, 0),
block_shape=(BLOCK_N, BLOCK_DMODEL),
order=(1, 0)
)
# initialize offsets
offs_m = start_m * BLOCK_M + tl.arange(0, BLOCK_M)
offs_n = tl.arange(0, BLOCK_N)
# initialize pointer to m and l
m_i = tl.zeros([BLOCK_M], dtype=tl.float32) - float("inf")
l_i = tl.zeros([BLOCK_M], dtype=tl.float32)
acc = tl.zeros([BLOCK_M, BLOCK_DMODEL], dtype=tl.float32)
# scale sm_scale by log_2(e) and use
# 2^x instead of exp in the loop because CSE and LICM
# don't work as expected with `exp` in the loop
qk_scale = sm_scale * 1.44269504
# load q: it will stay in SRAM throughout
q = tl.load(Q_block_ptr)
q = (q * qk_scale).to(dtype)
# loop over k, v and update accumulator
lo = 0
hi = (start_m + 1) * BLOCK_M
m_mask = offs_m[:, None] < seqlen
for start_n in range(lo, hi, BLOCK_N):
n_mask = (start_n + offs_n[None, :]) <= offs_m[:, None]
# -- load k, v --
k = tl.load(K_block_ptr)
v = tl.load(V_block_ptr)
# -- compute qk --
qk = tl.zeros([BLOCK_M, BLOCK_N], dtype=tl.float32)
qk = tl.where(m_mask & n_mask, qk, float("-inf"))
qk += tl.dot(q, k)
# -- compute scaling constant --
m_i_new = tl.maximum(m_i, tl.max(qk, 1))
alpha = tl.math.exp2(m_i - m_i_new)
p = tl.math.exp2(qk - m_i_new[:, None])
# -- scale and update acc --
acc_scale = l_i * 0 + alpha # workaround some compiler bug
acc *= acc_scale[:, None]
acc += tl.dot(p.to(dtype), v)
# -- update m_i and l_i --
l_i = l_i * alpha + tl.sum(p, 1)
m_i = m_i_new
# update pointers
K_block_ptr = tl.advance(K_block_ptr, (0, BLOCK_N))
V_block_ptr = tl.advance(V_block_ptr, (BLOCK_N, 0))
# write back O
acc = tl.where(m_mask, acc / l_i[:, None], 0.0)
O_block_ptr = tl.make_block_ptr(
base=Out + qo_offset,
shape=(N_CTX, BLOCK_DMODEL),
strides=(stride_om, stride_ok),
offsets=(start_m * BLOCK_M, 0),
block_shape=(BLOCK_M, BLOCK_DMODEL),
order=(1, 0)
)
tl.store(O_block_ptr, acc.to(dtype))
def triton_dense_forward(q, k, v, seqlens, sm_scale, block_size_M=128, block_size_N=64) -> torch.Tensor:
# shape constraints
Lq, Lk, Lv = q.shape[-1], k.shape[-1], v.shape[-1]
assert Lq == Lk and Lk == Lv
assert Lk in {16, 32, 64, 128}
o = torch.zeros_like(q)
grid = (triton.cdiv(q.shape[2], block_size_M), q.shape[0] * q.shape[1], 1)
num_warps = 4 if Lk <= 64 else 8 # 4
dtype = tl.bfloat16 if q.dtype == torch.bfloat16 else tl.float16
triton_dense_fwd_kernel[grid](
q, k, v, seqlens, sm_scale,
o,
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),
q.shape[0], q.shape[1], q.shape[2],
BLOCK_M=block_size_M, BLOCK_N=block_size_N,
BLOCK_DMODEL=Lk,
dtype=dtype,
num_warps=num_warps, num_stages=4,
)
return o
def flash_attn_forward(q, k, v, seqlens, sm_scale, context_size) -> torch.Tensor:
return flash_attn_varlen_func(
q,
k,
v,
cu_seqlens_q=seqlens,
cu_seqlens_k=seqlens,
max_seqlen_q=context_size,
max_seqlen_k=context_size,
dropout_p=0.0,
softmax_scale=sm_scale,
causal=True,
)
def torch_forward(
query: torch.Tensor,
key: torch.Tensor,
value: torch.Tensor,
mask: torch.Tensor,
sm_scale: float,
) -> torch.Tensor:
p = torch.einsum(f'bhmk, bhnk -> bhmn', query, key) * sm_scale
p = p.where(mask, -torch.inf)
p_max = p.max(-1, keepdim=True).values
p_max = torch.where(p_max < 0, 0.0, p_max)
p_exp = torch.exp(p - p_max)
s = p_exp / (p_exp.sum(-1, keepdim=True) + 1e-6)
out = torch.einsum(f'bhmn, bhnk -> bhmk', s, value)
return out
def profile(fn, total_flops, tag, warmup=25, rep=100):
ms = triton.testing.do_bench(fn, warmup=warmup, rep=rep)
gflops = total_flops / ms * 1e-9
print(f'{tag}: {ms:.3f} ms | {gflops:.3f} GFLOP/s')
def test_flash_attention(
query=None,
key=None,
value=None,
seqlens=None,
vertical_indexes=None,
slash_indexes=None,
dtype=torch.float16,
device="cuda",
torch_test=True,
batch_size=4,
num_heads=32,
context_size=2048,
head_dim=128,
nnz_v=100,
nnz_s=10,
block_size_M=64,
block_size_N=64,
):
print('========================================')
if query is None and key is None and value is None:
q = torch.randn((batch_size, num_heads, context_size, head_dim), dtype=dtype, device=device)
k = torch.randn((batch_size, num_heads, context_size, head_dim), dtype=dtype, device=device)
v = torch.randn((batch_size, num_heads, context_size, head_dim), dtype=dtype, device=device)
else:
q = torch.tensor(query, dtype=dtype, device=device)
k = torch.tensor(key, dtype=dtype, device=device)
v = torch.tensor(value, dtype=dtype, device=device)
batch_size, num_heads, context_size, head_dim = q.shape
print(f'BATCH={batch_size}, N_CTX={context_size}, N_HEADS={num_heads}, D_HEAD={head_dim}')
if seqlens is None:
seqlens = torch.randint(context_size // 2, context_size, (batch_size, ), dtype=torch.int32, device=device)
else:
seqlens = torch.tensor(seqlens, dtype=torch.int32, device=device)
print(seqlens)
dense_mask_nnz = seqlens.to(torch.float32).square().sum().item() * num_heads / 2
sm_scale = head_dim ** -0.5
if torch_test:
causal_mask = make_causal_mask(seqlens, device, context_size)
ref_o_dense = torch_forward(q, k, v, causal_mask, sm_scale)
if vertical_indexes is None or slash_indexes is None:
vertical_indexes = torch.stack([
torch.stack([
torch.randperm(seqlen, dtype=torch.int32, device=device)[:nnz_v].sort(descending=False)[0]
for _ in range(num_heads)
])
for seqlen in seqlens
])
slash_indexes = torch.concatenate([
torch.stack([
torch.stack([
torch.randperm(seqlen - 1, dtype=torch.int32, device=device)[:nnz_s - 1].sort(descending=True)[0] + 1
for _ in range(num_heads)
])
for seqlen in seqlens
]),
torch.zeros((batch_size, num_heads, 1), dtype=torch.int32, device=device)
], dim=-1)
pycuda_build_index_fn = lambda: pycuda_build_index(
seqlens, vertical_indexes, slash_indexes, context_size, block_size_M, block_size_N
)
indexes = pycuda_build_index_fn()
block_count, block_offset, column_count, column_index = indexes
if torch_test:
block_count_ref, block_offset_ref, column_count_ref, column_index_ref = torch_build_index(
seqlens, vertical_indexes, slash_indexes, context_size, block_size_M, block_size_N
)
torch.testing.assert_close(block_count_ref, block_count)
torch.testing.assert_close(block_offset_ref, block_offset)
torch.testing.assert_close(column_count_ref, column_count)
torch.testing.assert_close(column_index_ref, column_index)
sparse_mask_nnz = column_count.to(torch.float64).sum().item() * block_size_M + \
block_count.to(torch.float64).sum().item() * block_size_M * block_size_N
print(f'block mask sparsity: {1 - sparse_mask_nnz / dense_mask_nnz}')
pycuda_build_index_fn = lambda: pycuda_build_index(
seqlens, vertical_indexes, slash_indexes, context_size, block_size_M, block_size_N
)
profile(pycuda_build_index_fn, 0., 'pycuda-index')
if torch_test:
finegrained_mask = make_finegrained_mask(vertical_indexes, slash_indexes, causal_mask, device)
block_mask = make_block_mask(*indexes, seqlens, causal_mask, device, block_size_M, block_size_N)
plot_mask(finegrained_mask, 'mask.png', 0, 0)
plot_mask(block_mask, 'mask-1.png', 0, 0)
ref_o_sparse = torch_forward(q, k, v, block_mask, sm_scale)
triton_dense_fn = lambda: triton_dense_forward(q, k, v, seqlens, sm_scale)
output_triton_dense = triton_dense_fn()
if torch_test:
# Note: not correct for context_size % block_size_M != 0
torch.testing.assert_close(output_triton_dense, ref_o_dense, atol=1e-2, rtol=0)
profile(triton_dense_fn, 2. * head_dim * dense_mask_nnz, 'triton-dense')
triton_sparse_fn = lambda: triton_sparse_forward(q, k, v, seqlens, *indexes, sm_scale, block_size_M, block_size_N)
output_triton_sparse = triton_sparse_fn()
if torch_test:
torch.testing.assert_close(output_triton_sparse, ref_o_sparse, atol=1e-2, rtol=0)
profile(triton_sparse_fn, 2. * head_dim * sparse_mask_nnz, 'triton-sparse')
q = q.swapaxes(1, 2).contiguous()
k = k.swapaxes(1, 2).contiguous()
v = v.swapaxes(1, 2).contiguous()
q = torch.concatenate([q[i, :seqlen, :, :] for i, seqlen in enumerate(seqlens)])
k = torch.concatenate([k[i, :seqlen, :, :] for i, seqlen in enumerate(seqlens)])
v = torch.concatenate([v[i, :seqlen, :, :] for i, seqlen in enumerate(seqlens)])
seqlens = torch.nn.functional.pad(torch.cumsum(seqlens, dim=0, dtype=torch.int32), (1, 0))
flash_fn = lambda: flash_attn_forward(q, k, v, seqlens, sm_scale, context_size)
output_flash = flash_fn()
output_flash = torch.stack([
torch.nn.functional.pad(
output_flash[seqlens[i]:seqlens[i + 1], :, :],
(0, 0, 0, 0, 0, context_size + seqlens[i] - seqlens[i + 1])
)
for i in range(batch_size)
]).swapaxes(1, 2).contiguous()
if torch_test:
torch.testing.assert_close(output_flash, ref_o_dense, atol=1e-2, rtol=0)
profile(flash_fn, 2. * head_dim * dense_mask_nnz, 'flash-dense')
print('========================================\n')
if torch_test and sparse_mask_nnz >= dense_mask_nnz:
torch.testing.assert_close(output_flash, output_triton_sparse, atol=1e-2, rtol=0)
def vertical_slash_sparse_attention(
query: torch.Tensor, # [BATCH, N_HEADS, N_CTX, D_HEAD]
key: torch.Tensor, # [BATCH, N_HEADS, N_CTX, D_HEAD]
value: torch.Tensor, # [BATCH, N_HEADS, N_CTX, D_HEAD]
v_idx: torch.Tensor, # [BATCH, N_HEADS, NNZ_V]
s_idx: torch.Tensor, # [BATCH, N_HEADS, NNZ_S]
block_size_M: int = 64,
block_size_N: int = 64,
):
batch_size, num_heads, context_size, head_dim = query.shape
pad = block_size_M - (context_size & (block_size_M - 1))
query = torch.nn.functional.pad(query, [0, 0, 0, pad, 0, 0, 0, 0])
key = torch.nn.functional.pad(key, [0, 0, 0, pad, 0, 0, 0, 0])
value = torch.nn.functional.pad(value, [0, 0, 0, pad, 0, 0, 0, 0])
if head_dim not in [16, 32, 64, 128, 256, 512]:
target_dim = 2 ** math.ceil(math.log2(head_dim)) - head_dim
query = torch.nn.functional.pad(query, [0, target_dim, 0, 0, 0, 0, 0, 0])
key = torch.nn.functional.pad(key, [0, target_dim, 0, 0, 0, 0, 0, 0])
value = torch.nn.functional.pad(value, [0, target_dim, 0, 0, 0, 0, 0, 0])
v_idx = v_idx.to(torch.int32).reshape((batch_size, num_heads, -1)).sort(dim=-1, descending=False)[0]
s_idx = s_idx.to(torch.int32).reshape((batch_size, num_heads, -1)).sort(dim=-1, descending=True)[0]
seqlens = torch.tensor([context_size], dtype=torch.int32, device=query.device)
sm_scale = head_dim ** -0.5
block_count, block_offset, column_count, column_index = pycuda_build_index(
seqlens, v_idx, s_idx, context_size, block_size_M, block_size_N,
)
# if context_size > 700000:
# import ipdb; ipdb.set_trace()
# dense_mask_nnz = seqlens.to(torch.float32).square().sum().item() * num_heads / 2
# sparse_mask_nnz = column_count.to(torch.float64).sum().item() * block_size_M + \
# block_count.to(torch.float64).sum().item() * block_size_M * block_size_N
# print(f'block mask sparsity: {1 - sparse_mask_nnz / dense_mask_nnz}')
out = triton_sparse_forward(
query, key, value, seqlens,
block_count, block_offset, column_count, column_index,
sm_scale, block_size_M, block_size_N,
)
return out[..., :context_size, :head_dim]