mslk-cuda-nightly 2026.1.19__cp310-cp310-manylinux_2_28_x86_64.whl

mslk/attention/flash_attn/flash_fwd_combine.py

@@ -0,0 +1,705 @@
+# @nolint # fbcode
+# Copyright (c) 2025, Jay Shah, Ganesh Bikshandi, Ying Zhang, Vijay Thakkar, Pradeep Ramani, Tri Dao.
+# A reimplementation of https://github.com/Dao-AILab/flash-attention/blob/main/hopper/flash_fwd_combine_kernel.h
+# from Cutlass C++ to Cute-DSL.
+import math
+import operator
+from typing import Type, Optional
+from functools import partial
+
+import cuda.bindings.driver as cuda
+
+import cutlass
+import cutlass.cute as cute
+from cutlass.cute.nvgpu import cpasync
+from cutlass import Float32, Int32, const_expr
+
+from mslk.attention.flash_attn import utils
+from mslk.attention.flash_attn.seqlen_info import SeqlenInfo
+from cutlass.cute import FastDivmodDivisor
+
+
+class FlashAttentionForwardCombine:
+    def __init__(
+        self,
+        dtype: Type[cutlass.Numeric],
+        dtype_partial: Type[cutlass.Numeric],
+        head_dim: int,
+        m_block_size: int = 8,
+        k_block_size: int = 64,
+        log_max_splits: int = 4,
+        num_threads: int = 256,
+        stages: int = 4,
+    ):
+        """
+        Forward combine kernel for split attention computation.
+
+        :param dtype: output data type
+        :param dtype_partial: partial accumulation data type
+        :param head_dim: head dimension
+        :param m_block_size: m block size
+        :param k_block_size: k block size
+        :param log_max_splits: log2 of maximum splits
+        :param num_threads: number of threads
+        :param varlen: whether using variable length sequences
+        :param stages: number of pipeline stages
+        """
+        self.dtype = dtype
+        self.dtype_partial = dtype_partial
+        self.head_dim = head_dim
+        self.m_block_size = m_block_size
+        self.k_block_size = k_block_size
+        self.max_splits = 1 << log_max_splits
+        self.num_threads = num_threads
+        self.is_even_k = head_dim % k_block_size == 0
+        self.stages = stages
+
+    @staticmethod
+    def can_implement(
+        dtype,
+        dtype_partial,
+        head_dim,
+        m_block_size,
+        k_block_size,
+        log_max_splits,
+        num_threads,
+    ) -> bool:
+        """Check if the kernel can be implemented with the given parameters."""
+        if dtype not in [cutlass.Float16, cutlass.BFloat16, cutlass.Float32]:
+            return False
+        if dtype_partial not in [cutlass.Float16, cutlass.BFloat16, Float32]:
+            return False
+        if head_dim % 8 != 0:
+            return False
+        if num_threads % 32 != 0:
+            return False
+        if m_block_size % 8 != 0:
+            return False
+        max_splits = 1 << log_max_splits
+        if max_splits > 256:
+            return False
+        if (m_block_size * max_splits) % num_threads != 0:
+            return False
+        return True
+
+    def _setup_attributes(self):
+        # GMEM copy setup for O partial
+        universal_copy_bits = 128
+        async_copy_elems = universal_copy_bits // self.dtype_partial.width
+        assert self.k_block_size % async_copy_elems == 0
+
+        k_block_gmem = (
+            128 if self.k_block_size % 128 == 0 else (64 if self.k_block_size % 64 == 0 else 32)
+        )
+        gmem_threads_per_row = k_block_gmem // async_copy_elems
+        assert self.num_threads % gmem_threads_per_row == 0
+
+        # Async copy atom for O partial load
+        atom_async_copy_partial = cute.make_copy_atom(
+            cpasync.CopyG2SOp(cache_mode=cpasync.LoadCacheMode.GLOBAL),
+            self.dtype_partial,
+            num_bits_per_copy=universal_copy_bits,
+        )
+        tOpartial_layout = cute.make_ordered_layout(
+            (self.num_threads // gmem_threads_per_row, gmem_threads_per_row),
+            order=(1, 0),
+        )
+        vOpartial_layout = cute.make_layout((1, async_copy_elems))  # 4 vals per load
+        self.gmem_tiled_copy_O_partial = cute.make_tiled_copy_tv(
+            atom_async_copy_partial, tOpartial_layout, vOpartial_layout
+        )
+
+        # GMEM copy setup for final O (use universal copy for store)
+        atom_universal_copy = cute.make_copy_atom(
+            cute.nvgpu.CopyUniversalOp(),
+            self.dtype,
+            num_bits_per_copy=async_copy_elems * self.dtype.width,
+        )
+        self.gmem_tiled_copy_O = cute.make_tiled_copy_tv(
+            atom_universal_copy,
+            tOpartial_layout,
+            vOpartial_layout,  # 4 vals per store
+        )
+
+        # LSE copy setup with async copy (alignment = 1)
+        lse_copy_bits = Float32.width  # 1 element per copy, width is in bits
+        m_block_smem = (
+            128
+            if self.m_block_size % 128 == 0
+            else (
+                64
+                if self.m_block_size % 64 == 0
+                else (
+                    32
+                    if self.m_block_size % 32 == 0
+                    else (16 if self.m_block_size % 16 == 0 else 8)
+                )
+            )
+        )
+        gmem_threads_per_row_lse = m_block_smem
+        assert self.num_threads % gmem_threads_per_row_lse == 0
+
+        # Async copy atom for LSE load
+        atom_async_copy_lse = cute.make_copy_atom(
+            cpasync.CopyG2SOp(cache_mode=cpasync.LoadCacheMode.ALWAYS),
+            Float32,
+            num_bits_per_copy=lse_copy_bits,
+        )
+        tLSE_layout = cute.make_ordered_layout(
+            (self.num_threads // gmem_threads_per_row_lse, gmem_threads_per_row_lse),
+            order=(1, 0),
+        )
+        vLSE_layout = cute.make_layout(1)
+        self.gmem_tiled_copy_LSE = cute.make_tiled_copy_tv(
+            atom_async_copy_lse, tLSE_layout, vLSE_layout
+        )
+
+        # ///////////////////////////////////////////////////////////////////////////////
+        # Shared memory
+        # ///////////////////////////////////////////////////////////////////////////////
+
+        # Shared memory to register copy for LSE
+        self.smem_threads_per_col_lse = self.num_threads // m_block_smem
+        assert 32 % self.smem_threads_per_col_lse == 0  # Must divide warp size
+
+        s2r_layout_atom_lse = cute.make_ordered_layout(
+            (self.smem_threads_per_col_lse, self.num_threads // self.smem_threads_per_col_lse),
+            order=(0, 1),
+        )
+        self.s2r_tiled_copy_LSE = cute.make_tiled_copy_tv(
+            cute.make_copy_atom(cute.nvgpu.CopyUniversalOp(), Float32),
+            s2r_layout_atom_lse,
+            cute.make_layout(1),
+        )
+
+        # LSE shared memory layout with swizzling to avoid bank conflicts
+        # This works for kBlockMSmem = 8, 16, 32, 64, 128, no bank conflicts
+        if const_expr(m_block_smem == 8):
+            smem_lse_swizzle = cute.make_swizzle(5, 0, 5)
+        elif const_expr(m_block_smem == 16):
+            smem_lse_swizzle = cute.make_swizzle(4, 0, 4)
+        else:
+            smem_lse_swizzle = cute.make_swizzle(3, 2, 3)
+        smem_layout_atom_lse = cute.make_composed_layout(
+            smem_lse_swizzle, 0, cute.make_ordered_layout((8, m_block_smem), order=(1, 0))
+        )
+        self.smem_layout_lse = cute.tile_to_shape(
+            smem_layout_atom_lse, (self.max_splits, self.m_block_size), (0, 1)
+        )
+
+        # O partial shared memory layout (simple layout for pipeline stages)
+        self.smem_layout_o = cute.make_ordered_layout(
+            (self.m_block_size, self.k_block_size, self.stages), order=(1, 0, 2)
+        )
+
+    @cute.jit
+    def __call__(
+        self,
+        mO_partial: cute.Tensor,
+        mLSE_partial: cute.Tensor,
+        mO: cute.Tensor,
+        mLSE: Optional[cute.Tensor] = None,
+        cu_seqlens: Optional[cute.Tensor] = None,
+        seqused: Optional[cute.Tensor] = None,
+        num_splits_dynamic_ptr: Optional[cute.Tensor] = None,
+        semaphore_to_reset: Optional[cute.Tensor] = None,
+        stream: cuda.CUstream = None,
+    ):
+        # Type checking
+        if const_expr(not (mO_partial.element_type == self.dtype_partial)):
+            raise TypeError("O partial tensor must match dtype_partial")
+        if const_expr(not (mO.element_type == self.dtype)):
+            raise TypeError("O tensor must match dtype")
+        if const_expr(mLSE_partial.element_type not in [Float32]):
+            raise TypeError("LSE partial tensor must be Float32")
+        if const_expr(mLSE is not None and mLSE.element_type not in [Float32]):
+            raise TypeError("LSE tensor must be Float32")
+
+        # Shape validation - input tensors are in user format, need to be converted to kernel format
+        if const_expr(len(mO_partial.shape) not in [4, 5]):
+            raise ValueError(
+                "O partial tensor must have 4 or 5 dimensions: (num_splits, batch, seqlen, nheads, headdim) or (num_splits, total_q, nheads, headdim)"
+            )
+        if const_expr(len(mLSE_partial.shape) not in [3, 4]):
+            raise ValueError(
+                "LSE partial tensor must have 3 or 4 dimensions: (num_splits, batch, seqlen, nheads) or (num_splits, total_q, nheads)"
+            )
+        if const_expr(len(mO.shape) not in [3, 4]):
+            raise ValueError(
+                "O tensor must have 3 or 4 dimensions: (batch, seqlen, nheads, headdim) or (total_q, nheads, headdim)"
+            )
+        if const_expr(mLSE is not None and len(mLSE.shape) not in [2, 3]):
+            raise ValueError(
+                "LSE tensor must have 2 or 3 dimensions: (batch, seqlen, nheads) or (total_q, nheads)"
+            )
+
+        # Assume all strides are divisible by 128 bits except the last stride
+        new_stride = lambda t: (
+            *(cute.assume(s, divby=128 // t.element_type.width) for s in t.stride[:-1]),
+            t.stride[-1],
+        )
+        mO_partial, mO = [
+            cute.make_tensor(t.iterator, cute.make_layout(t.shape, stride=new_stride(t)))
+            for t in (mO_partial, mO)
+        ]
+        # (num_splits, b, seqlen, h, d) -> (seqlen, d, num_splits, h, b)
+        # or (num_splits, total_q, h, d) -> (total_q, d, num_splits, h)
+        O_partial_layout_transpose = (
+            [2, 4, 0, 3, 1] if const_expr(cu_seqlens is None) else [1, 3, 0, 2]
+        )
+        # (b, seqlen, h, d) -> (seqlen, d, h, b) or (total_q, h, d) -> (total_q, d, h)
+        mO_partial = cute.make_tensor(
+            mO_partial.iterator, cute.select(mO_partial.layout, mode=O_partial_layout_transpose)
+        )
+        O_layout_transpose = [1, 3, 2, 0] if const_expr(cu_seqlens is None) else [0, 2, 1]
+        mO = cute.make_tensor(mO.iterator, cute.select(mO.layout, mode=O_layout_transpose))
+        # (num_splits, b, seqlen, h) -> (seqlen, num_splits, h, b)
+        # or (num_splits, total_q, h) -> (total_q, num_splits, h)
+        LSE_partial_layout_transpose = [2, 0, 3, 1] if const_expr(cu_seqlens is None) else [1, 0, 2]
+        mLSE_partial = cute.make_tensor(
+            mLSE_partial.iterator,
+            cute.select(mLSE_partial.layout, mode=LSE_partial_layout_transpose),
+        )
+        # (b, seqlen, h) -> (seqlen, h, b) or (total_q, h) -> (total_q, h)
+        LSE_layout_transpose = [1, 2, 0] if const_expr(cu_seqlens is None) else [0, 1]
+        mLSE = (
+            cute.make_tensor(mLSE.iterator, cute.select(mLSE.layout, mode=LSE_layout_transpose))
+            if mLSE is not None
+            else None
+        )
+
+        # Determine if we have variable length sequences
+        varlen = const_expr(cu_seqlens is not None or seqused is not None)
+
+        self._setup_attributes()
+
+        @cute.struct
+        class SharedStorage:
+            sLSE: cute.struct.Align[
+                cute.struct.MemRange[Float32, cute.cosize(self.smem_layout_lse)], 128
+            ]
+            sMaxValidSplit: cute.struct.Align[cute.struct.MemRange[Int32, self.m_block_size], 128]
+            sO: cute.struct.Align[
+                cute.struct.MemRange[self.dtype_partial, cute.cosize(self.smem_layout_o)], 128
+            ]
+
+        smem_size = SharedStorage.size_in_bytes()
+
+        # Grid dimensions: (ceil_div(seqlen, m_block), ceil_div(head_dim, k_block), num_head * batch)
+        seqlen = mO_partial.shape[0]
+        num_head = mO_partial.shape[3]
+        batch_size = (
+            mO_partial.shape[4]
+            if const_expr(cu_seqlens is None)
+            else Int32(cu_seqlens.shape[0] - 1)
+        )
+
+        # Create FastDivmodDivisor objects for efficient division
+        seqlen_divmod = FastDivmodDivisor(seqlen)
+        head_divmod = FastDivmodDivisor(num_head)
+
+        grid_dim = (
+            cute.ceil_div(seqlen * num_head, self.m_block_size),
+            cute.ceil_div(self.head_dim, self.k_block_size),
+            batch_size,
+        )
+
+        self.kernel(
+            mO_partial,
+            mLSE_partial,
+            mO,
+            mLSE,
+            cu_seqlens,
+            seqused,
+            num_splits_dynamic_ptr,
+            semaphore_to_reset,
+            SharedStorage,
+            self.smem_layout_lse,
+            self.smem_layout_o,
+            self.gmem_tiled_copy_O_partial,
+            self.gmem_tiled_copy_O,
+            self.gmem_tiled_copy_LSE,
+            self.s2r_tiled_copy_LSE,
+            seqlen_divmod,
+            head_divmod,
+            varlen,
+        ).launch(
+            grid=grid_dim,
+            block=[self.num_threads, 1, 1],
+            smem=smem_size,
+            stream=stream,
+        )
+
+    @cute.kernel
+    def kernel(
+        self,
+        mO_partial: cute.Tensor,
+        mLSE_partial: cute.Tensor,
+        mO: cute.Tensor,
+        mLSE: Optional[cute.Tensor],
+        cu_seqlens: Optional[cute.Tensor],
+        seqused: Optional[cute.Tensor],
+        num_splits_dynamic_ptr: Optional[cute.Tensor],
+        semaphore_to_reset: Optional[cute.Tensor],
+        SharedStorage: cutlass.Constexpr,
+        smem_layout_lse: cute.Layout | cute.ComposedLayout,
+        smem_layout_o: cute.Layout,
+        gmem_tiled_copy_O_partial: cute.TiledCopy,
+        gmem_tiled_copy_O: cute.TiledCopy,
+        gmem_tiled_copy_LSE: cute.TiledCopy,
+        s2r_tiled_copy_LSE: cute.TiledCopy,
+        seqlen_divmod: FastDivmodDivisor,
+        head_divmod: FastDivmodDivisor,
+        varlen: cutlass.Constexpr[bool],
+    ):
+        # Thread and block indices
+        tidx, _, _ = cute.arch.thread_idx()
+        m_block, k_block, batch_idx = cute.arch.block_idx()
+
+        # ///////////////////////////////////////////////////////////////////////////////
+        # Get shared memory buffer
+        # ///////////////////////////////////////////////////////////////////////////////
+        smem = cutlass.utils.SmemAllocator()
+        storage = smem.allocate(SharedStorage)
+        sLSE = storage.sLSE.get_tensor(smem_layout_lse)
+        sMaxValidSplit = storage.sMaxValidSplit.get_tensor((self.m_block_size,))
+        sO = storage.sO.get_tensor(smem_layout_o)
+
+        # Handle semaphore reset
+        if const_expr(semaphore_to_reset is not None):
+            if (
+                tidx == 0
+                and m_block == cute.arch.grid_dim()[0] - 1
+                and k_block == cute.arch.grid_dim()[1] - 1
+                and batch_idx == cute.arch.grid_dim()[2] - 1
+            ):
+                semaphore_to_reset[0] = 0
+
+        # Get number of splits
+        num_splits = (
+            num_splits_dynamic_ptr[batch_idx]
+            if const_expr(num_splits_dynamic_ptr is not None)
+            else mLSE_partial.shape[1]
+        )
+        # Handle variable length sequences using SeqlenInfo
+        seqlen_info = SeqlenInfo.create(
+            batch_idx=batch_idx,
+            seqlen_static=mO_partial.shape[0],
+            cu_seqlens=cu_seqlens,
+            seqused=seqused,
+        )
+        seqlen, offset = seqlen_info.seqlen, seqlen_info.offset
+
+        # Extract number of heads (head index will be determined dynamically)
+        num_head = mO_partial.shape[3]
+        max_idx = seqlen * num_head
+
+        # Early exit for single split if dynamic
+        if (const_expr(num_splits_dynamic_ptr is None) or num_splits > 1) and (
+            const_expr(not varlen) or m_block * self.m_block_size < max_idx
+        ):
+            # ===============================
+            # Step 1: Load LSE_partial from gmem to shared memory
+            # ===============================
+
+            if const_expr(cu_seqlens is None):
+                # mLSE_partial_cur = mLSE_partial[None, None, None, batch_idx]
+                mLSE_partial_cur = utils.coord_offset_i64(mLSE_partial, batch_idx, dim=3)
+            else:
+                # mLSE_partial_cur = cute.domain_offset((offset, 0, 0), mLSE_partial)
+                mLSE_partial_cur = utils.domain_offset_i64((offset, 0, 0), mLSE_partial)
+            mLSE_partial_copy = cute.tiled_divide(mLSE_partial_cur, (1,))
+
+            gmem_thr_copy_LSE = gmem_tiled_copy_LSE.get_slice(tidx)
+            tLSEsLSE = gmem_thr_copy_LSE.partition_D(sLSE)
+
+            # Create identity tensor for coordinate tracking
+            cLSE = cute.make_identity_tensor((self.max_splits, self.m_block_size))
+            tLSEcLSE = gmem_thr_copy_LSE.partition_S(cLSE)
+
+            # Load LSE partial values
+            for m in cutlass.range(cute.size(tLSEcLSE, mode=[2]), unroll_full=True):
+                mi = tLSEcLSE[0, 0, m][1]  # Get m coordinate
+                idx = m_block * self.m_block_size + mi
+                if idx < max_idx:
+                    # Calculate actual sequence position and head using FastDivmodDivisor
+                    if const_expr(not varlen):
+                        head_idx, m_idx = divmod(idx, seqlen_divmod)
+                    else:
+                        head_idx = idx // seqlen
+                        m_idx = idx - head_idx * seqlen
+                    mLSE_partial_cur_copy = mLSE_partial_copy[None, m_idx, None, head_idx]
+                    for s in cutlass.range(cute.size(tLSEcLSE, mode=[1]), unroll_full=True):
+                        si = tLSEcLSE[0, s, 0][0]  # Get split coordinate
+                        if si < num_splits:
+                            cute.copy(
+                                gmem_thr_copy_LSE,
+                                mLSE_partial_cur_copy[None, si],
+                                tLSEsLSE[None, s, m],
+                            )
+                        else:
+                            tLSEsLSE[None, s, m].fill(-Float32.inf)
+                # Don't need to zero out the rest of the LSEs, as we will not write the output to gmem
+            cute.arch.cp_async_commit_group()
+
+            # ===============================
+            # Step 2: Load O_partial for pipeline stages
+            # ===============================
+
+            gmem_thr_copy_O_partial = gmem_tiled_copy_O_partial.get_slice(tidx)
+            cO = cute.make_identity_tensor((self.m_block_size, self.k_block_size))
+            tOcO = gmem_thr_copy_O_partial.partition_D(cO)
+            tOsO_partial = gmem_thr_copy_O_partial.partition_D(sO)
+            if const_expr(cu_seqlens is None):
+                # mO_partial_cur = mO_partial[None, None, None, None, batch_idx]
+                mO_partial_cur = utils.coord_offset_i64(mO_partial, batch_idx, dim=4)
+            else:
+                # mO_partial_cur = cute.domain_offset((offset, 0, 0, 0), mO_partial)
+                mO_partial_cur = utils.domain_offset_i64((offset, 0, 0, 0), mO_partial)
+
+            # Precompute these values to avoid recomputing them in the loop
+            num_rows = const_expr(cute.size(tOcO, mode=[1]))
+            tOmidx = cute.make_fragment(num_rows, cutlass.Int32)
+            tOhidx = cute.make_fragment(num_rows, cutlass.Int32)
+            tOrOptr = cute.make_fragment(num_rows, cutlass.Int64)
+            for m in cutlass.range(num_rows, unroll_full=True):
+                mi = tOcO[0, m, 0][0]  # m coordinate
+                idx = m_block * self.m_block_size + mi
+                if const_expr(not varlen):
+                    tOhidx[m], tOmidx[m] = divmod(idx, seqlen_divmod)
+                else:
+                    tOhidx[m] = idx // seqlen
+                    tOmidx[m] = idx - tOhidx[m] * seqlen
+                tOrOptr[m] = utils.elem_pointer_i64(
+                    mO_partial_cur, (tOmidx[m], k_block * self.k_block_size, 0, tOhidx[m])
+                ).toint()
+                if idx >= max_idx:
+                    tOhidx[m] = -1
+
+            tOpO = cute.make_fragment(cute.size(tOcO, [2]), cutlass.Boolean)
+            if const_expr(not self.is_even_k):
+                for k in cutlass.range(cute.size(tOpO), unroll_full=True):
+                    tOpO[k] = tOcO[0, 0, k][1] < mO_partial.shape[1] - k_block * self.k_block_size
+            # if cute.arch.thread_idx()[0] == 0 and k_block == 1: cute.print_tensor(tOpO)
+
+            load_O_partial = partial(
+                self.load_O_partial,
+                gmem_tiled_copy_O_partial,
+                tOrOptr,
+                tOsO_partial,
+                tOhidx,
+                tOpO,
+                tOcO,
+                mO_partial_cur.layout,
+            )
+
+            # Load first few stages of O_partial
+            for stage in cutlass.range(self.stages - 1, unroll_full=True):
+                if stage < num_splits:
+                    load_O_partial(stage, stage)
+                cute.arch.cp_async_commit_group()
+
+            # ===============================
+            # Step 3: Load and transpose LSE from smem to registers
+            # ===============================
+
+            # Wait for LSE and initial O partial stages to complete
+            cute.arch.cp_async_wait_group(self.stages - 1)
+            cute.arch.sync_threads()
+            # if cute.arch.thread_idx()[0] == 0:
+            #     # cute.print_tensor(sLSE)
+            #     for i in range(64):
+            #         cute.printf("sLSE[%d, 0] = %f", i, sLSE[i, 0])
+            # cute.arch.sync_threads()
+
+            s2r_thr_copy_LSE = s2r_tiled_copy_LSE.get_slice(tidx)
+            ts2rsLSE = s2r_thr_copy_LSE.partition_S(sLSE)
+            ts2rrLSE = cute.make_fragment_like(ts2rsLSE)
+            cute.copy(s2r_tiled_copy_LSE, ts2rsLSE, ts2rrLSE)
+
+            # ===============================
+            # Step 4: Compute final LSE along split dimension
+            # ===============================
+
+            lse_sum = cute.make_fragment(cute.size(ts2rrLSE, mode=[2]), Float32)
+            ts2rcLSE = s2r_thr_copy_LSE.partition_D(cLSE)
+            # We compute the max valid split for each row to short-circuit the computation later
+            max_valid_split = cute.make_fragment(cute.size(ts2rrLSE, mode=[2]), Int32)
+            assert cute.size(ts2rrLSE, mode=[0]) == 1
+            # Compute max, scales, and final LSE for each row
+            for m in cutlass.range(cute.size(ts2rrLSE, mode=[2]), unroll_full=True):
+                # Find max LSE value across splits
+                threads_per_col = const_expr(self.smem_threads_per_col_lse)
+                lse_max = utils.warp_reduce(
+                    ts2rrLSE[None, None, m]
+                    .load()
+                    .reduce(cute.ReductionOp.MAX, init_val=-Float32.inf, reduction_profile=0),
+                    op=cute.arch.fmax,
+                    width=threads_per_col,
+                )
+                # if cute.arch.thread_idx()[0] == 0: cute.printf(lse_max)
+                # Find max valid split index
+                max_valid_idx = -1
+                for s in cutlass.range(cute.size(ts2rrLSE, mode=[1]), unroll_full=True):
+                    if ts2rrLSE[0, s, m] != -Float32.inf:
+                        max_valid_idx = ts2rcLSE[0, s, 0][0]  # Get split coordinate
+                # if cute.arch.thread_idx()[0] < 32: cute.printf(max_valid_idx)
+                max_valid_split[m] = utils.warp_reduce(max_valid_idx, max, width=threads_per_col)
+                # Compute exp scales and sum
+                lse_max_cur = (
+                    0.0 if lse_max == -Float32.inf else lse_max
+                )  # In case all local LSEs are -inf
+                LOG2_E = math.log2(math.e)
+                lse_sum_cur = 0.0
+                for s in cutlass.range(cute.size(ts2rrLSE, mode=[1]), unroll_full=True):
+                    scale = utils.exp2f(ts2rrLSE[0, s, m] * LOG2_E - (lse_max_cur * LOG2_E))
+                    lse_sum_cur += scale
+                    ts2rrLSE[0, s, m] = scale  # Store scale for later use
+                lse_sum_cur = utils.warp_reduce(lse_sum_cur, operator.add, width=threads_per_col)
+                lse_sum[m] = utils.logf(lse_sum_cur) + lse_max
+                # Normalize scales
+                inv_sum = (
+                    0.0 if (lse_sum_cur == 0.0 or lse_sum_cur != lse_sum_cur) else 1.0 / lse_sum_cur
+                )
+                ts2rrLSE[None, None, m].store(ts2rrLSE[None, None, m].load() * inv_sum)
+            # Store the scales exp(lse - lse_logsum) back to smem
+            cute.copy(s2r_tiled_copy_LSE, ts2rrLSE, ts2rsLSE)
+
+            # Store max valid split to smem
+            for m in cutlass.range(cute.size(ts2rrLSE, mode=[2]), unroll_full=True):
+                if ts2rcLSE[0, 0, m][0] == 0:  # Only thread responsible for s=0 writes
+                    mi = ts2rcLSE[0, 0, m][1]
+                    if mi < self.m_block_size:
+                        sMaxValidSplit[mi] = max_valid_split[m]
+
+            # ===============================
+            # Step 5: Store final LSE to gmem
+            # ===============================
+
+            if const_expr(mLSE is not None):
+                if const_expr(cu_seqlens is None):
+                    # mLSE_cur = mLSE[None, None, batch_idx]
+                    mLSE_cur = utils.coord_offset_i64(mLSE, batch_idx, dim=2)
+                else:
+                    # mLSE_cur = cute.domain_offset((offset, 0), mLSE)
+                    mLSE_cur = utils.domain_offset_i64((offset, 0), mLSE)
+                if k_block == 0:  # Only first k_block writes LSE when mLSE is provided
+                    for m in cutlass.range(cute.size(ts2rrLSE, mode=[2]), unroll_full=True):
+                        if ts2rcLSE[0, 0, m][0] == 0:  # Only thread responsible for s=0 writes
+                            mi = ts2rcLSE[0, 0, m][1]
+                            idx = m_block * self.m_block_size + mi
+                            if idx < max_idx:
+                                if const_expr(not varlen):
+                                    head_idx, m_idx = divmod(idx, seqlen_divmod)
+                                else:
+                                    head_idx = idx // seqlen
+                                    m_idx = idx - head_idx * seqlen
+                                mLSE_cur[m_idx, head_idx] = lse_sum[m]
+
+            # ===============================
+            # Step 6: Read O_partial and accumulate final O
+            # ===============================
+
+            cute.arch.sync_threads()
+
+            # Get max valid split for this thread
+            thr_max_valid_split = sMaxValidSplit[tOcO[0, 0, 0][0]]
+            for m in cutlass.range(1, cute.size(tOcO, mode=[1])):
+                thr_max_valid_split = max(thr_max_valid_split, sMaxValidSplit[tOcO[0, m, 0][0]])
+
+            tOrO_partial = cute.make_fragment_like(tOsO_partial[None, None, None, 0])
+            tOrO = cute.make_fragment_like(tOrO_partial, Float32)
+            tOrO.fill(0.0)
+
+            stage_load = self.stages - 1
+            stage_compute = 0
+
+            # Main accumulation loop
+            for s in cutlass.range(thr_max_valid_split + 1, unroll=4):
+                # Get scales for this split
+                scale = cute.make_fragment(num_rows, Float32)
+                for m in cutlass.range(num_rows, unroll_full=True):
+                    scale[m] = sLSE[s, tOcO[0, m, 0][0]]  # Get scale from smem
+
+                # Load next stage if needed
+                split_to_load = s + self.stages - 1
+                if split_to_load <= thr_max_valid_split:
+                    load_O_partial(split_to_load, stage_load)
+                cute.arch.cp_async_commit_group()
+                stage_load = 0 if stage_load == self.stages - 1 else stage_load + 1
+
+                # Wait for the current stage to be ready
+                cute.arch.cp_async_wait_group(self.stages - 1)
+                # We don't need __syncthreads() because each thread is just reading its own data from smem
+                # Copy from smem to registers
+                cute.autovec_copy(tOsO_partial[None, None, None, stage_compute], tOrO_partial)
+                stage_compute = 0 if stage_compute == self.stages - 1 else stage_compute + 1
+
+                # Accumulate scaled partial results
+                for m in cutlass.range(num_rows, unroll_full=True):
+                    if tOhidx[m] >= 0 and scale[m] > 0.0:
+                        tOrO[None, m, None].store(
+                            tOrO[None, m, None].load()
+                            + scale[m] * tOrO_partial[None, m, None].load().to(Float32)
+                        )
+
+            # ===============================
+            # Step 7: Write final O to gmem
+            # ===============================
+
+            rO = cute.make_fragment_like(tOrO, self.dtype)
+            rO.store(tOrO.load().to(self.dtype))
+            if const_expr(cu_seqlens is None):
+                # mO_cur = mO[None, None, None, batch_idx]
+                mO_cur = utils.coord_offset_i64(mO, batch_idx, dim=3)
+            else:
+                # mO_cur = cute.domain_offset((offset, 0, 0), mO)
+                mO_cur = utils.domain_offset_i64((offset, 0, 0), mO)
+            mO_cur = utils.domain_offset_aligned((0, k_block * self.k_block_size, 0), mO_cur)
+            elems_per_store = const_expr(cute.size(gmem_tiled_copy_O.layout_tv_tiled[1]))
+            # mO_cur_copy = cute.tiled_divide(mO_cur, (1, elems_per_store,))
+            gmem_thr_copy_O = gmem_tiled_copy_O.get_slice(tidx)
+            # Write final results
+            for m in cutlass.range(num_rows, unroll_full=True):
+                if tOhidx[m] >= 0:
+                    mO_cur_copy = cute.tiled_divide(
+                        mO_cur[tOmidx[m], None, tOhidx[m]], (elems_per_store,)
+                    )
+                    for k in cutlass.range(cute.size(tOcO, mode=[2]), unroll_full=True):
+                        k_idx = tOcO[0, 0, k][1] // elems_per_store
+                        if const_expr(self.is_even_k) or tOpO[k]:
+                            cute.copy(gmem_thr_copy_O, rO[None, m, k], mO_cur_copy[None, k_idx])
+
+    @cute.jit
+    def load_O_partial(
+        self,
+        gmem_tiled_copy_O_partial: cute.TiledCopy,
+        tOrOptr: cute.Tensor,
+        tOsO_partial: cute.Tensor,
+        tOhidx: cute.Tensor,
+        tOpO: cute.Tensor,
+        tOcO: cute.Tensor,
+        mO_cur_partial_layout: cute.Layout,
+        split: Int32,
+        stage: Int32,
+    ) -> None:
+        elems_per_load = const_expr(cute.size(gmem_tiled_copy_O_partial.layout_tv_tiled[1]))
+        tOsO_partial_cur = tOsO_partial[None, None, None, stage]
+        for m in cutlass.range(cute.size(tOcO, [1]), unroll_full=True):
+            if tOhidx[m] >= 0:
+                o_gmem_ptr = cute.make_ptr(
+                    tOsO_partial.element_type, tOrOptr[m], cute.AddressSpace.gmem, assumed_align=16
+                )
+                mO_partial_cur = cute.make_tensor(
+                    o_gmem_ptr, cute.slice_(mO_cur_partial_layout, (0, None, None, 0))
+                )
+                mO_partial_cur_copy = cute.tiled_divide(mO_partial_cur, (elems_per_load,))
+                for k in cutlass.range(cute.size(tOcO, mode=[2]), unroll_full=True):
+                    k_idx = tOcO[0, 0, k][1] // elems_per_load
+                    if const_expr(self.is_even_k) or tOpO[k]:
+                        cute.copy(
+                            gmem_tiled_copy_O_partial,
+                            # mO_partial_cur_copy[None, k_idx, split],
+                            utils.coord_offset_i64(mO_partial_cur_copy, split, dim=2)[None, k_idx],
+                            tOsO_partial_cur[None, m, k],
+                        )