example_linear_attn_bwd.py 9.18 KB
Newer Older
1
import torch
2
import tilelang
3
4
5
6
import tilelang.language as T
from tilelang.profiler import do_bench
import argparse
from fla.ops.linear_attn import fused_chunk_linear_attn  # We compare with FLA
7
8
9
from fla.modules.l2norm import l2norm_fwd
from einops import rearrange
from typing import Optional, Tuple
10
11


12
@tilelang.jit(
13
    pass_configs={
14
15
        tilelang.PassConfigKey.TL_DISABLE_TMA_LOWER: True,
        tilelang.PassConfigKey.TL_DISABLE_WARP_SPECIALIZED: True,
16
    })
17
def tl_fused_chunk_bwd_kernel(
18
19
20
21
22
23
24
25
26
27
28
29
30
31
    B,
    S,
    H,
    DK,
    DV,
    dtype: str = 'float16',
    scale: float = None,
) -> torch.Tensor:

    if scale is None:
        scale = DK**-0.5
    accum_dtype = 'float'

    chunk_size = 64
32
    BK = BV = 64  # Set to 128 can be faster, but has some numerical differences with FLA
33
    assert S % chunk_size == 0 and DK % BK == 0 and DV % BV == 0
34
35
36
    NK = tilelang.cdiv(DK, BK)
    NV = tilelang.cdiv(DV, BV)
    NT = tilelang.cdiv(S, chunk_size)
37
38

    @T.prim_func
39
    def fused_chunk_linear_attn_bwd(
40
41
42
43
            Q: T.Tensor([B, S, H, DK], dtype),  # type: ignore
            K: T.Tensor([B, S, H, DK], dtype),  # type: ignore
            V: T.Tensor([B, S, H, DV], dtype),  # type: ignore
            dO: T.Tensor([B, S, H, DV], dtype),  # type: ignore
44
45
46
            dQ: T.Tensor([B, S, H, DK], accum_dtype),  # type: ignore
            dK: T.Tensor([B, S, H, DK], accum_dtype),  # type: ignore
            dV: T.Tensor([B, S, H, DV], accum_dtype),  # type: ignore
47
48
49
50
51
52
53
54
    ):
        with T.Kernel(NV, NK, B * H) as (i_v, i_k, i_bh):
            i_b = i_bh // H
            i_h = i_bh % H

            ds = T.alloc_fragment([chunk_size, chunk_size], accum_dtype)
            ds_shared = T.alloc_shared([chunk_size, chunk_size], dtype)
            dq = T.alloc_fragment([chunk_size, BK], accum_dtype)
55
            dq_shared = T.alloc_shared([chunk_size, BK], accum_dtype)
56
            dk = T.alloc_fragment([chunk_size, BK], accum_dtype)
57
            dk_shared = T.alloc_shared([chunk_size, BK], accum_dtype)
58
            dv = T.alloc_fragment([chunk_size, BV], accum_dtype)
59
            dv_shared = T.alloc_shared([chunk_size, BV], accum_dtype)
60
61
62
63
64
65
66
67
68
69
            q = T.alloc_shared([chunk_size, BK], dtype)
            k = T.alloc_shared([chunk_size, BK], dtype)
            v = T.alloc_shared([chunk_size, BV], dtype)
            do = T.alloc_shared([chunk_size, BV], dtype)
            h = T.alloc_fragment([BV, BK], accum_dtype)
            h_shared = T.alloc_shared([BV, BK], dtype)
            dh = T.alloc_fragment([BK, BV], accum_dtype)
            dh_shared = T.alloc_shared([BK, BV], dtype)

            T.annotate_layout({
70
71
72
                dq_shared: tilelang.layout.make_swizzled_layout(dq_shared),
                dk_shared: tilelang.layout.make_swizzled_layout(dk_shared),
                dv_shared: tilelang.layout.make_swizzled_layout(dv_shared)
73
            })
74
            T.use_swizzle(10)
75

76
77
78
            T.clear(h)
            T.clear(dh)

79
            # Calculate dQ
80
            for i in T.Pipelined(0, NT):
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
                T.copy(K[i_b, i * chunk_size:(i + 1) * chunk_size, i_h, i_k * BK:(i_k + 1) * BK], k)
                T.copy(V[i_b, i * chunk_size:(i + 1) * chunk_size, i_h, i_v * BV:(i_v + 1) * BV], v)
                T.copy(dO[i_b, i * chunk_size:(i + 1) * chunk_size, i_h, i_v * BV:(i_v + 1) * BV],
                       do)

                T.gemm(do, v, ds, transpose_B=True, clear_accum=True)
                for row, col in T.Parallel(chunk_size, chunk_size):
                    ds_shared[row, col] = T.if_then_else(row >= col, ds[row, col], 0)

                T.gemm(ds_shared, k, dq, clear_accum=True)
                T.copy(h, h_shared)
                T.gemm(do, h_shared, dq)
                T.gemm(v, k, h, transpose_A=True)
                for row, col in T.Parallel(chunk_size, BK):
                    dq[row, col] *= scale
96
97
98
99
                T.copy(dq, dq_shared)
                T.atomic_add(
                    dQ[i_b, i * chunk_size:(i + 1) * chunk_size, i_h, i_k * BK:(i_k + 1) * BK],
                    dq_shared)
100
101

            # Calculate dK, dV (reversely)
102
            for i in T.Pipelined(1, NT + 1):
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
                start = NT - i
                for row, col in T.Parallel(chunk_size, BK):
                    q[row, col] = Q[i_b, start * chunk_size + row, i_h, i_k * BK + col] * scale
                T.copy(
                    K[i_b, start * chunk_size:(start + 1) * chunk_size, i_h,
                      i_k * BK:(i_k + 1) * BK], k)
                T.copy(
                    V[i_b, start * chunk_size:(start + 1) * chunk_size, i_h,
                      i_v * BV:(i_v + 1) * BV], v)
                T.copy(
                    dO[i_b, start * chunk_size:(start + 1) * chunk_size, i_h,
                       i_v * BV:(i_v + 1) * BV], do)

                # Calculate dk
                T.gemm(
                    v, do, ds, transpose_B=True, clear_accum=True
                )  # ds here actually means `s`, but we simply reuse the buffer `ds`
                for row, col in T.Parallel(chunk_size, chunk_size):
                    ds_shared[row, col] = T.if_then_else(row <= col, ds[row, col], 0)
                T.gemm(ds_shared, q, dk, clear_accum=True)
123
                T.copy(dh, dh_shared)
124
125
126
127
128
129
130
131
132
133
134
135
                T.gemm(v, dh_shared, dk, transpose_B=True)

                # Calculate dv
                T.gemm(k, q, ds, transpose_B=True, clear_accum=True)
                for row, col in T.Parallel(chunk_size, chunk_size):
                    ds_shared[row, col] = T.if_then_else(row <= col, ds[row, col], 0)
                T.gemm(ds_shared, do, dv, clear_accum=True)
                T.gemm(k, dh_shared, dv)

                # Update dh
                T.gemm(q, do, dh, transpose_A=True)

136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
                T.copy(dk, dk_shared)
                T.atomic_add(
                    dK[i_b, start * chunk_size:(start + 1) * chunk_size, i_h,
                       i_k * BK:(i_k + 1) * BK], dk_shared)
                T.copy(dv, dv_shared)
                T.atomic_add(
                    dV[i_b, start * chunk_size:(start + 1) * chunk_size, i_h,
                       i_v * BV:(i_v + 1) * BV], dv_shared)

    return fused_chunk_linear_attn_bwd


def tl_fused_chunk_bwd(Q, K, V, dO):
    B, S, H, D = Q.shape
    kernel = tl_fused_chunk_bwd_kernel(B, S, H, D, D)
    dQ = torch.zeros_like(Q, dtype=torch.float32)
    dK = torch.zeros_like(K, dtype=torch.float32)
    dV = torch.zeros_like(V, dtype=torch.float32)
    kernel(Q, K, V, dO, dQ, dK, dV)
    return dQ.to(torch.float16), dK.to(torch.float16), dV.to(torch.float16)


def ref_program(q: torch.Tensor,
                k: torch.Tensor,
                v: torch.Tensor,
                scale: Optional[float] = None) -> Tuple[torch.Tensor, torch.Tensor]:
    q, k, v = q.float(), k.float(), v.float()
    if scale is None:
        scale = q.shape[-1]**-0.5
    chunk_size = 64
    q = rearrange(q, 'b (n c) h d -> b h n c d', c=chunk_size) * scale
    k = rearrange(k, 'b (n c) h d -> b h n c d', c=chunk_size)
    v = rearrange(v, 'b (n c) h d -> b h n c d', c=chunk_size)
    kv = k.transpose(-1, -2) @ v
    kv = kv.cumsum(2)
    h = kv[:, :, -1, :, :]
    kv = torch.cat([torch.zeros_like(kv[:, :, :1]), kv[:, :, :-1]], dim=2)
    inter = q @ kv
    intra = ((q @ k.transpose(-1, -2)).masked_fill_(
        torch.triu(torch.ones(chunk_size, chunk_size, dtype=bool, device=q.device), diagonal=1),
        0)) @ v
    o = inter + intra
    return rearrange(o, 'b h n c d -> b (n c) h d'), h


def main(B=1, S=1024, H=16, D=128):
182
183
184
185
186
    q = torch.randn((B, S, H, D), device='cuda', dtype=torch.float16, requires_grad=True)
    k = torch.randn((B, S, H, D), device='cuda', dtype=torch.float16, requires_grad=True)
    v = torch.randn((B, S, H, D), device='cuda', dtype=torch.float16, requires_grad=True)
    do = torch.randn((B, S, H, D), device='cuda', dtype=torch.float16)

187
188
189
190
191
192
193
    # qk norm is necessary for linear attn
    q = l2norm_fwd(q)[0].requires_grad_(True)
    k = l2norm_fwd(k)[0].requires_grad_(True)

    dq, dk, dv = tl_fused_chunk_bwd(q, k, v, do)
    q.grad = k.grad = v.grad = None
    o_ref, _ = ref_program(q, k, v)
194
    o_ref.backward(do, retain_graph=True)
195
196
197
198
199
200
201
202
203
204

    assert torch.allclose(
        dq, q.grad, atol=1e-2, rtol=1e-2), f'dq max err: {(dq - q.grad).abs().max()}'
    assert torch.allclose(
        dk, k.grad, atol=1e-2, rtol=1e-2), f'dk max err: {(dk - k.grad).abs().max()}'
    assert torch.allclose(
        dv, v.grad, atol=1e-2, rtol=1e-2), f'dv max err: {(dv - v.grad).abs().max()}'
    print('Passed all tests!✅')

    # Benchmark
205
    q.grad = k.grad = v.grad = None
206
    o_ref, _ = fused_chunk_linear_attn(q, k, v, output_final_state=True, normalize=False)
207
208
    t1 = do_bench(lambda: o_ref.backward(do, retain_graph=True), backend='cupti')
    t2 = do_bench(lambda: tl_fused_chunk_bwd(q, k, v, do), backend='cupti')
209
210
211
212
213
214
    print(f'Triton latency: {t1:.3f} ms')
    print(f'TileLang latency: {t2:.3f} ms')
    print(f'Speedup: {t1/t2:.3f}x')


if __name__ == '__main__':
215
216
217
218
219
220
221
222
    parser = argparse.ArgumentParser()
    parser.add_argument('--B', type=int, default=8, help='Batch size')
    parser.add_argument('--S', type=int, default=1024, help='Seq len')
    parser.add_argument('--H', type=int, default=32, help='Num heads')
    parser.add_argument('--D', type=int, default=128, help='Head dim')
    args = parser.parse_args()

    main(args.B, args.S, args.H, args.D)