# Copyright (c) Microsoft Corporation. # SPDX-License-Identifier: Apache-2.0 # DeepSpeed Team """ Inspired by original Triton implementation: https://github.com/openai/triton/blob/b244db06da24a87453a40ad35b085ee37dac3705/python/tutorials/06-fused-attention.py """ import torch import triton import triton.language as tl @triton.jit def _fwd_kernel( Q, K, V, sm_scale, TMP, Out, stride_qz, stride_qh, stride_qm, stride_qk, stride_kz, stride_kh, stride_kn, stride_kk, stride_vz, stride_vh, stride_vk, stride_vn, stride_oz, stride_oh, stride_om, stride_on, Z, H, N_CTX, BLOCK_M: tl.constexpr, BLOCK_DMODEL: tl.constexpr, BLOCK_N: tl.constexpr, ): start_m = tl.program_id(0) off_hz = tl.program_id(1) # 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) off_q = off_hz * stride_qh + offs_m[:, None] * stride_qm + offs_d[None, :] * stride_qk off_k = off_hz * stride_kh + offs_n[:, None] * stride_kn + offs_d[None, :] * stride_kk off_v = off_hz * stride_vh + offs_n[:, None] * stride_qm + offs_d[None, :] * stride_qk # Initialize pointers to Q, K, V q_ptrs = Q + off_q k_ptrs = K + off_k v_ptrs = V + off_v # initialize pointer to m and l t_ptrs = TMP + off_hz * N_CTX + offs_m 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) # load q: it will stay in SRAM throughout q = tl.load(q_ptrs) # loop over k, v and update accumulator for start_n in range(0, N_CTX, BLOCK_N): start_n = tl.multiple_of(start_n, BLOCK_N) # -- compute qk ---- k = tl.load(k_ptrs + start_n * stride_kn) qk = tl.zeros([BLOCK_M, BLOCK_N], dtype=tl.float32) qk += tl.dot(q, k, trans_b=True) qk *= sm_scale # -- compute m_ij, p, l_ij m_ij = tl.max(qk, 1) p = tl.exp(qk - m_ij[:, None]) l_ij = tl.sum(p, 1) # -- update m_i and l_i m_i_new = tl.maximum(m_i, m_ij) alpha = tl.exp(m_i - m_i_new) beta = tl.exp(m_ij - m_i_new) l_i_new = alpha * l_i + beta * l_ij # -- update output accumulator -- # scale p p_scale = beta / l_i_new p = p * p_scale[:, None] # scale acc acc_scale = l_i / l_i_new * alpha tl.store(t_ptrs, acc_scale) acc_scale = tl.load(t_ptrs) # BUG: have to store and immediately load acc = acc * acc_scale[:, None] # update acc v = tl.load(v_ptrs + start_n * stride_vk) p = p.to(tl.float16) acc += tl.dot(p, v) # update m_i and l_i l_i = l_i_new m_i = m_i_new # initialize pointers to output offs_n = tl.arange(0, BLOCK_DMODEL) off_o = off_hz * stride_oh + offs_m[:, None] * stride_om + offs_n[None, :] * stride_on out_ptrs = Out + off_o tl.store(out_ptrs, acc) class triton_flash_attn(torch.nn.Module): def __init__(self, ): super(triton_flash_attn, self).__init__() def forward(self, q, k, v, sm_scale, block_128=True): BLOCK = 128 if block_128 else 64 # shape constraints Lq, Lk, Lv = q.shape[-1], k.shape[-1], v.shape[-1] o = torch.empty_like(q) grid = (triton.cdiv(q.shape[2], BLOCK), q.shape[0] * q.shape[1]) tmp = torch.empty((q.shape[0] * q.shape[1], q.shape[2]), device=q.device, dtype=torch.float32) num_warps = 4 if Lk <= 64 else 8 _fwd_kernel[grid]( q, k, v, sm_scale, tmp, 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), k.shape[0], k.shape[1], k.shape[2], BLOCK_M=BLOCK, BLOCK_N=BLOCK, BLOCK_DMODEL=Lk, num_warps=num_warps, num_stages=1, ) return o