kernel
File size: 13,627 Bytes
eb8ddce
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
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
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
/******************************************************************************
 * Copyright (c) 2024, Jay Shah, Ganesh Bikshandi, Ying Zhang, Vijay Thakkar, Pradeep Ramani, Tri Dao.
 ******************************************************************************/

#pragma once

#include "cute/tensor.hpp"

#include <cutlass/cutlass.h>
#include <cutlass/array.h>
#include <cutlass/numeric_types.h>
#include <cutlass/numeric_conversion.h>
#include "cutlass/arch/barrier.h"

#include "seqlen.h"
#include "utils.h"

namespace flash {

using namespace cute;

template <class TileShape_MK_, class Element, class ElementAccum, class ArchTag_, int kNThreads, class TiledMma, bool dQ_swapAB>
class FlashAttnBwdPostprocessConvertdQ {

public:

    // Type Aliases
    using TileShape_MK = TileShape_MK_;
    using ArchTag = ArchTag_;

    static_assert(ArchTag::kMinComputeCapability >= 75);
    static constexpr bool IsSm90 = ArchTag::kMinComputeCapability >= 90;

    static constexpr uint32_t MaxThreadsPerBlock = kNThreads;
    static constexpr uint32_t MinBlocksPerMultiprocessor = 2;

    static constexpr int kBlockM = get<0>(TileShape_MK{});
    static constexpr int kHeadDim = get<1>(TileShape_MK{});
    static_assert(!IsSm90 || kNThreads % cutlass::NumThreadsPerWarpGroup == 0, "kNThreads must be a multiple of NumThreadsPerWarpGroup");
    static constexpr int NumdQWarpGgroups = kNThreads / cutlass::NumThreadsPerWarpGroup;
    using R2SLayoutAtomdQaccum = std::conditional_t<
        IsSm90,
        Layout<Shape<Int<cutlass::NumThreadsPerWarpGroup>, Int<NumdQWarpGgroups>>>,
        Layout<Shape<Int<kNThreads>>>
    >;
    using R2STiledCopydQaccum = decltype(make_tiled_copy(Copy_Atom<AutoVectorizingCopyWithAssumedAlignment<128>, ElementAccum>{}, R2SLayoutAtomdQaccum{},
                                                         Layout<Shape<Int<IsSm90 ? 4 : 1>>>{}));  // Val layout, 1 or 4 vals per read
    using G2SLayoutAtomdQaccum = Layout<Shape<Int<kNThreads>>>;
    // UniversalCopy instead of AutoVectorizingCopyWithAssumedAlignment as the latter generates cp.async instructions
    using G2STiledCopydQaccum = decltype(make_tiled_copy(Copy_Atom<UniversalCopy<uint128_t>, ElementAccum>{}, G2SLayoutAtomdQaccum{},
                                                         Layout<Shape<_4>>{}));  // Val layout, 4 vals per read
    // We don't do bound checking for the gmem -> smem load so we just assert here.
    static_assert(IsSm90 || (kBlockM * kHeadDim) % (kNThreads * 4) == 0);
    static constexpr int SmemdQaccumSize = size(TileShape_MK{});
    using SmemLayoutdQaccumFlat = Layout<Shape<Int<SmemdQaccumSize>>>;
    using SmemLayoutdQaccum = std::conditional_t<
        IsSm90,
        Layout<Shape<Int<kBlockM * kHeadDim / NumdQWarpGgroups>, Int<NumdQWarpGgroups>>>,
        Layout<Shape<Int<kBlockM * kHeadDim>>>
    >;

    // We can't just use kHeadDim here. E.g. if MMA shape is 64 x 96 but split across 2 WGs,
    // then setting kBlockKSmem to 32 will cause "Static shape_div failure".
    // We want to treat it as 64 x 48, so kBlockKSmem should be 16.
    static constexpr int MmaShapeN = get<1>(typename TiledMma::AtomShape_MNK{});
    static constexpr int kBlockKSmem = MmaShapeN % 64 == 0 ? 64 : (MmaShapeN % 32 == 0 ? 32 : 16);
    static constexpr int kSwizzle = kBlockKSmem == 64 ? 3 : (kBlockKSmem == 32 ? 2 : 1);
    using SmemLayoutAtomdQ =
        decltype(composition(Swizzle<kSwizzle, 3, 3>{},
                 Layout<Shape<Int<8>, Int<kBlockKSmem>>,
                 Stride<Int<kBlockKSmem>, _1>>{}));
    using SmemLayoutdQ = decltype(tile_to_shape(SmemLayoutAtomdQ{}, TileShape_MK{}));
    using SmemLayoutdQt =
        decltype(cute::composition(SmemLayoutdQ{},
                                   make_layout(make_shape(get<1>(TileShape_MK{}), get<0>(TileShape_MK{})),
                                               make_stride(Int<get<0>(TileShape_MK{})>{}, _1{}))));

    using SmemCopyAtomdQ = Copy_Atom<
        std::conditional_t<
            IsSm90,
            std::conditional_t<!dQ_swapAB, cute::SM90_U32x4_STSM_N, cute::SM90_U16x8_STSM_T>,
            AutoVectorizingCopyWithAssumedAlignment<128>
        >,
        Element>;

    static constexpr int kGmemElemsPerLoad = sizeof(cute::uint128_t) / sizeof(Element);
    static_assert(kHeadDim % kGmemElemsPerLoad == 0, "Headdim must be a multiple of kGmemElemsPerLoad");
    static constexpr int kGmemThreadsPerRow = cutlass::gcd(kHeadDim / kGmemElemsPerLoad, int(MaxThreadsPerBlock));
    static_assert(MaxThreadsPerBlock % kGmemThreadsPerRow == 0, "MaxThreadsPerBlock must be a multiple of kGmemThreadsPerRow");
    using GmemLayoutAtom = Layout<Shape <Int<MaxThreadsPerBlock / kGmemThreadsPerRow>, Int<kGmemThreadsPerRow>>,
                                  Stride<Int<kGmemThreadsPerRow>, _1>>;
    using GmemTiledCopy = decltype(
        make_tiled_copy(Copy_Atom<AutoVectorizingCopyWithAssumedAlignment<128>, Element>{},
                        GmemLayoutAtom{},
                        Layout<Shape<_1, Int<kGmemElemsPerLoad>>>{}));  // Val layout, 8 or 16 vals per load

    struct SharedStorage : cute::aligned_struct<128> {
        cute::array_aligned<ElementAccum, cute::cosize_v<SmemLayoutdQaccum>> smem_dqacc;
        cute::array_aligned<Element, cute::cosize_v<SmemLayoutdQ>> smem_dq;
        alignas(16) cutlass::arch::ClusterTransactionBarrier barrier_dQaccum;
    };

    static constexpr int SharedStorageSize = sizeof(SharedStorage);

    using ShapedQ = cute::Shape<int32_t, int32_t, int32_t, int32_t>;   // (seqlen_q, d, head, batch)
    using StridedQ = cute::Stride<int64_t, _1, int64_t, int64_t>;
    using ShapedQaccum = cute::Shape<int32_t, int32_t, int32_t>;  // (seqlen_q * d, head, batch)
    using StridedQaccum = cute::Stride<_1, int64_t, int64_t>;

    // Device side arguments
    struct Arguments {
        ElementAccum const* ptr_dQaccum;
        ShapedQaccum const shape_dQaccum;
        StridedQaccum const stride_dQaccum;
        Element* ptr_dQ;
        ShapedQ const shape_dQ;
        StridedQ const stride_dQ;
        float const softmax_scale;
        int const* cu_seqlens = nullptr;
        int const* seqused = nullptr;
    };

    // Kernel entry point API
    struct Params {
        ElementAccum const* ptr_dQaccum;
        ShapedQaccum const shape_dQaccum;
        StridedQaccum const stride_dQaccum;
        Element* ptr_dQ;
        ShapedQ const shape_dQ;
        StridedQ const stride_dQ;
        float const softmax_scale;
        int const* cu_seqlens = nullptr;
        int const* seqused = nullptr;
    };

    // Convert to underlying arguments. In this case, a simple copy for the aliased type.
    static
    Params
    to_underlying_arguments(Arguments const& args) {
        return {
            args.ptr_dQaccum,
            args.shape_dQaccum,
            args.stride_dQaccum,
            args.ptr_dQ,
            args.shape_dQ,
            args.stride_dQ,
            args.softmax_scale,
            args.cu_seqlens,
            args.seqused
        };
    }

    CUTLASS_DEVICE
    void
    operator()(Params const& params, char* smem_buf) {

        static constexpr int kBlockM = get<0>(TileShape_MK{});
        SharedStorage& shared_storage = *reinterpret_cast<SharedStorage*>(smem_buf);

        Tensor sdQaccum = make_tensor(make_smem_ptr(shared_storage.smem_dqacc.data()), SmemLayoutdQaccum{});
        Tensor sdQaccum_flat = make_tensor(make_smem_ptr(shared_storage.smem_dqacc.data()), SmemLayoutdQaccumFlat{});
        Tensor sdQ = make_tensor(make_smem_ptr(shared_storage.smem_dq.data()), SmemLayoutdQ{});
        Tensor sdQt = make_tensor(make_smem_ptr(shared_storage.smem_dq.data()), SmemLayoutdQt{});

        int const thread_idx = threadIdx.x;
        int const m_block = blockIdx.x;
        int const bidh = blockIdx.y;
        int const bidb = blockIdx.z;

        flash::SeqlenInfo<true /*Varlen*/, kBlockM> seqlen_info(bidb, size<0>(params.shape_dQ), params.cu_seqlens, params.seqused);
        bool const is_varlen = params.cu_seqlens;
        if (is_varlen && m_block * kBlockM >= seqlen_info.seqlen) { return; }

        // Step 1: load dQaccum from gmem to smem
        Tensor mdQaccum = make_tensor(make_gmem_ptr(reinterpret_cast<ElementAccum const*>(params.ptr_dQaccum)),
                                      params.shape_dQaccum, params.stride_dQaccum)(_, bidh, !is_varlen ? bidb : 0);
        Tensor gdQaccum = local_tile(domain_offset(make_coord(seqlen_info.offset_padded * kHeadDim), mdQaccum), Shape<Int<kBlockM * kHeadDim>>{}, make_coord(m_block));  // (M * K)
        if constexpr (IsSm90) {  // Use BulkCopy
            static constexpr uint32_t TmaTransactionBytesdQaccum = static_cast<uint32_t>(size(SmemLayoutdQaccumFlat{}) * cute::sizeof_bits_v<ElementAccum> / 8);
            auto bulk_copy = Copy_Traits<SM90_BULK_COPY_AUTO>{};
            // if (thread0()) { print(gdQaccum); printf("\n"); print(sdQaccum_flat); printf("\n"); }
            if (thread_idx == 0) {
                shared_storage.barrier_dQaccum.init(1 /*numThreads*/);
                shared_storage.barrier_dQaccum.arrive_and_expect_tx(TmaTransactionBytesdQaccum);
                copy(bulk_copy.with(*reinterpret_cast<uint64_t*>(&shared_storage.barrier_dQaccum)), gdQaccum, sdQaccum_flat);
            }
            __syncthreads();
            shared_storage.barrier_dQaccum.wait(0);
        } else {
            G2STiledCopydQaccum g2s_tiled_copy_dQaccum;
            auto g2s_thr_copy_dQaccum = g2s_tiled_copy_dQaccum.get_thread_slice(thread_idx);
            Tensor tdQgdQaccumg2s = g2s_thr_copy_dQaccum.partition_S(gdQaccum);
            Tensor tdQsdQaccumg2s = g2s_thr_copy_dQaccum.partition_D(sdQaccum);
            cute::copy(g2s_tiled_copy_dQaccum, tdQgdQaccumg2s, tdQsdQaccumg2s);
            __syncthreads();
        }

        // __syncthreads(); if (cute::thread0()) { print_tensor(sdQaccum); }

        // Step 2: Load dQaccum from smem to register, then convert fp32 -> fp16/bf16
        R2STiledCopydQaccum s2r_tiled_copy_dQaccum;
        auto s2r_thr_copy_dQaccum = s2r_tiled_copy_dQaccum.get_thread_slice(thread_idx);
        Tensor tdQsdQaccum = s2r_thr_copy_dQaccum.partition_S(sdQaccum);
        TiledMma tiled_mma_dQ;
        Tensor taccdQrdQaccum = partition_fragment_C(tiled_mma_dQ, select<!dQ_swapAB ? 0 : 1, !dQ_swapAB ? 1 : 0>(TileShape_MK{}));
        // if (blockIdx.x == 0 && blockIdx.y == 0 && threadIdx.x == 1) { print(tiled_mma_dQ); printf("\n"); }
        // if (blockIdx.x == 0 && blockIdx.y == 0 && threadIdx.x == 1) { print(tdQsdQaccum); }
        // if (blockIdx.x == 0 && blockIdx.y == 0 && threadIdx.x == 1) { print(taccdQrdQaccum); }
        CUTE_STATIC_ASSERT_V(size(taccdQrdQaccum) == size(tdQsdQaccum));
        Tensor tdQrdQaccum = s2r_thr_copy_dQaccum.retile_D(taccdQrdQaccum);
        cute::copy(s2r_tiled_copy_dQaccum, tdQsdQaccum, tdQrdQaccum);
        #pragma unroll
        for (int i = 0; i < size(taccdQrdQaccum); ++i) { taccdQrdQaccum(i) *= params.softmax_scale; }
        // Convert tdQrdQ from fp32 to fp16
        Tensor rdQ = make_tensor_like<Element>(taccdQrdQaccum);
        flash::convert_type_out(taccdQrdQaccum, rdQ);

        // Step 3: Copy dQ from register to smem
        auto smem_tiled_copy_dQ = make_tiled_copy_C(SmemCopyAtomdQ{}, tiled_mma_dQ);
        auto smem_thr_copy_dQ = smem_tiled_copy_dQ.get_thread_slice(thread_idx);
        Tensor taccdQrdQ = smem_thr_copy_dQ.retile_S(rdQ);  // ((Atom,AtomNum), MMA_N, MMA_N)
        // if (cute::thread0()) { print(smem_tiled_copy_dQ); }
        // if (cute::thread0()) { print(smem_thr_copy_dQ); }
        // if (cute::thread0()) { print(sdQ); }
        Tensor taccdQsdQ = smem_thr_copy_dQ.partition_D(cute::conditional_return<!dQ_swapAB>(sdQ, sdQt));  // ((Atom,AtomNum),PIPE_M,PIPE_N)
        cute::copy(smem_tiled_copy_dQ, taccdQrdQ, taccdQsdQ);
        __syncthreads();

        // Step 4: Copy dQ from smem to register to prepare for coalesced write to gmem
        Tensor mdQ = make_tensor(make_gmem_ptr(params.ptr_dQ), params.shape_dQ, params.stride_dQ)(_, _, bidh, !is_varlen ? bidb : 0);
        Tensor gdQ = local_tile(domain_offset(make_coord(seqlen_info.offset, _0{}), mdQ), TileShape_MK{}, make_coord(m_block, _0{}));  // (M, K)
        GmemTiledCopy gmem_tiled_copy_dQ;
        auto gmem_thr_copy_dQ = gmem_tiled_copy_dQ.get_thread_slice(thread_idx);
        Tensor tdQsdQ = gmem_thr_copy_dQ.partition_S(sdQ);    // ((Atom,AtomNum),ATOM_M,ATOM_N)
        Tensor tdQgdQ = gmem_thr_copy_dQ.partition_D(gdQ);

        Tensor tdQrdQ = make_fragment_like(tdQsdQ);
        Tensor tdQcdQ = gmem_thr_copy_dQ.partition_D(cute::make_identity_tensor(TileShape_MK{}));
        Tensor tdQpdQ = make_tensor<bool>(make_shape(size<2>(tdQgdQ)));
        #pragma unroll
        for (int k = 0; k < size(tdQpdQ); ++k) { tdQpdQ(k) = get<1>(tdQcdQ(_0{}, _0{}, k)) < get<1>(params.shape_dQ); }
        // Need to check OOB when reading from smem if kBlockM isn't evenly tiled
        static constexpr bool EvenM = kBlockM % CUTE_STATIC_V(size<0>(GmemLayoutAtom{})) == 0;
        flash::copy</*Is_even_MN=*/EvenM, /*Is_even_K=*/true, /*Clear_OOB_MN=*/false>(
            gmem_tiled_copy_dQ, tdQsdQ, tdQrdQ, tdQcdQ, tdQpdQ, kBlockM);

        // Step 5: Copy dQ from register to gmem
        // Clear_OOB_K must be false since we don't want to write zeros to gmem
        flash::copy</*Is_even_MN=*/false, /*Is_even_K=*/false, /*Clear_OOB_MN=*/false, /*Clear_OOB_K=*/false>(
            gmem_tiled_copy_dQ, tdQrdQ, tdQgdQ, tdQcdQ, tdQpdQ, std::min(seqlen_info.seqlen - m_block * kBlockM, kBlockM)
        );
    }

};

} // namespace flash