From dd0079b9dc3a64e3d13d147f0d3c25abe939b368 Mon Sep 17 00:00:00 2001 From: biondizzle Date: Sat, 23 May 2026 20:08:31 +0000 Subject: [PATCH] auto: pre-test commit --- dsv4/kernels/attention/fmha.py.backup | 515 ++++++++++++++++++++++++++ 1 file changed, 515 insertions(+) create mode 100644 dsv4/kernels/attention/fmha.py.backup diff --git a/dsv4/kernels/attention/fmha.py.backup b/dsv4/kernels/attention/fmha.py.backup new file mode 100644 index 00000000..87783e49 --- /dev/null +++ b/dsv4/kernels/attention/fmha.py.backup @@ -0,0 +1,515 @@ +"""FMHA kernel: QK -> online softmax -> PV (CuTeDSL, Blackwell SM100). + +Migrated from tests/unit/test_fmha_v3_stage_c.py — Stage C proven path. +P stored to TMEM via register bridge, PV reads from TMEM. +O rescale via correction_rescale atoms, O normalization via TMEM round-trip. +""" +import torch, cutlass, cutlass.cute as cute, cutlass.utils as utils, cutlass.pipeline as pipeline +from cutlass.cute.nvgpu import cpasync, tcgen05 +from cutlass import Float32, BFloat16, Int32, Boolean, const_expr +from cutlass.utils import LayoutEnum +from cutlass.utils.tmem_allocator import find_tmem_tensor_col_offset +import cuda.bindings.driver as cuda +import cutlass.torch as ct +import math + + +class FmhaKernel: + def __init__(self, head_dim=64, s_k=128, scale_softmax=None, use_smem_p=None): + self.head_dim = head_dim + self.s_k = s_k + self.n_kv_tiles = s_k // 128 + self.pv_n_tile = min(head_dim, 256) # tcgen05 MMA max N=256 + self.n_pv_tiles = head_dim // self.pv_n_tile + self.use_smem_p = use_smem_p if use_smem_p is not None else (head_dim > 64) + self.acc_dtype = Float32; self.qk_acc_dtype = Float32 + self.q_dtype = BFloat16; self.o_dtype = BFloat16; self.c_dtype = BFloat16 + self.use_2cta_instrs = False; self.epilog_sync_bar_id = 1 + self.cluster_shape_mn = (1, 1); self.cta_group = tcgen05.CtaGroup.ONE + self.epilogue_warp_id = (0,1,2,3); self.mma_warp_id = 4; self.tma_warp_id = 5 + self.threads_per_cta = 192; self.num_c_stage = 2 + self.kv_stage = 2; self.q_stage = 1; self.num_c_stage = 2 + self.scale_softmax = scale_softmax if scale_softmax is not None else 1.0 / math.sqrt(self.head_dim) + self.scale_softmax_log2 = self.scale_softmax * math.log2(math.e) + + def _setup(self, qk_mma, pv_mma): + qk_ik = cute.size(qk_mma.shape_mnk, mode=[2]) + self.qk_mma_tiler = (128, 128, qk_ik * 4) + pv_ik = cute.size(pv_mma.shape_mnk, mode=[2]) + self.pv_mma_tiler = (128, self.pv_n_tile, pv_ik * (128 // pv_ik)) + self.mma_tiler = self.qk_mma_tiler + self.cluster_layout_vmnk = cute.tiled_divide(cute.make_layout((1,1,1)), (qk_mma.thr_id.shape,)) + self.cta_tile_shape_mnk = (self.qk_mma_tiler[0]//cute.size(qk_mma.thr_id.shape), self.pv_n_tile, self.qk_mma_tiler[2]) + self.c_layout = LayoutEnum.ROW_MAJOR + self.epi_tile = utils.sm100.compute_epilogue_tile_shape(self.cta_tile_shape_mnk, False, self.c_layout, self.o_dtype) + self.num_ab_stage = 1; self.num_acc_stage = 1 + self.q_smem_s = utils.sm100.make_smem_layout_a(qk_mma, self.qk_mma_tiler, self.q_dtype, self.q_stage) + self.k_smem_s = utils.sm100.make_smem_layout_b(qk_mma, self.qk_mma_tiler, self.q_dtype, self.kv_stage) + self.v_smem_s = utils.sm100.make_smem_layout_b(pv_mma, self.pv_mma_tiler, self.q_dtype, self.kv_stage) + self.c_smem_s = utils.sm100.make_smem_layout_epi(self.o_dtype, self.c_layout, self.epi_tile, 2) + self.p_tmem_s = utils.sm100.make_smem_layout_a(pv_mma, self.pv_mma_tiler, self.q_dtype, 1) + # P SMEM layout (PV A-operand) — used for SMEM-P path + self.p_smem_s = utils.sm100.make_smem_layout_a(pv_mma, self.pv_mma_tiler, self.q_dtype, 1) + qk_thr = qk_mma.get_slice(0); qk_as = qk_thr.partition_shape_C(self.qk_mma_tiler[:2]) + tStS = qk_thr.make_fragment_C(qk_as) + pv_thr = pv_mma.get_slice(0); pv_as = pv_thr.partition_shape_C(self.pv_mma_tiler[:2]) + tOtO = pv_thr.make_fragment_C(pv_as) + self.tmem_s0_offset = 0 + if not self.use_smem_p: + # TMEM-P: S at 0, P at 32, O after P and S + self.tmem_p0_offset = 32 + p_cols_fp32 = self.pv_mma_tiler[2] * self.q_dtype.width // self.qk_acc_dtype.width + p_end = self.tmem_p0_offset + p_cols_fp32 + s_cols = self.qk_mma_tiler[1] + o_after = max(s_cols, p_end) + self.tmem_o0_offset = ((o_after + 31) // 32) * 32 + o_cols = find_tmem_tensor_col_offset(tOtO) + total = self.tmem_o0_offset + o_cols + else: + # SMEM-P: P not in TMEM. S and O share TMEM (sequential). + self.tmem_p0_offset = -1 # unused + self.tmem_o0_offset = 0 + s_cols = self.qk_mma_tiler[1] + o_cols = find_tmem_tensor_col_offset(tOtO) + total = max(s_cols, o_cols) + self.num_tmem_alloc_cols = 1 + while self.num_tmem_alloc_cols < total: + self.num_tmem_alloc_cols *= 2 + cta = cute.size(qk_mma.thr_id.shape) + q_s = cute.slice_(self.q_smem_s,(None,None,None,0)) + k_s = cute.slice_(self.k_smem_s,(None,None,None,0)) + v_s = cute.slice_(self.v_smem_s,(None,None,None,0)) + self.q_tx_bytes = cute.size_in_bytes(self.q_dtype, q_s) * cta + self.kv_tx_bytes = (cute.size_in_bytes(self.q_dtype, k_s) + + cute.size_in_bytes(self.q_dtype, v_s)) * cta + + @cute.jit + def __call__(self, q, k, v, c, stream): + self.q_dtype = q.element_type; self.o_dtype = c.element_type; self.c_dtype = self.o_dtype + self.a_major = LayoutEnum.from_tensor(q).mma_major_mode() + self.b_major = LayoutEnum.from_tensor(k).mma_major_mode() + v_fmha = cute.make_tensor( + v.iterator, + cute.make_layout( + (self.pv_n_tile, self.s_k, 1), + stride=(1, self.pv_n_tile, self.pv_n_tile * self.s_k), + ), + ) + self.v_major = LayoutEnum.from_tensor(v_fmha).mma_major_mode() + self.c_layout = LayoutEnum.from_tensor(c) + qk_mma = utils.sm100.make_trivial_tiled_mma(self.q_dtype, self.q_dtype, self.a_major, self.b_major, self.qk_acc_dtype, self.cta_group, (128,128), tcgen05.OperandSource.SMEM) + pv_a_major = self.a_major if self.use_smem_p else cute.nvgpu.OperandMajorMode.K + pv_source = tcgen05.OperandSource.SMEM if self.use_smem_p else tcgen05.OperandSource.TMEM + pv_mma = utils.sm100.make_trivial_tiled_mma(self.q_dtype, self.q_dtype, pv_a_major, self.v_major, self.qk_acc_dtype, self.cta_group, (128,self.pv_n_tile), pv_source) + self._setup(qk_mma, pv_mma) + q_s = cute.slice_(self.q_smem_s,(None,None,None,0)); k_s = cute.slice_(self.k_smem_s,(None,None,None,0)); v_s = cute.slice_(self.v_smem_s,(None,None,None,0)) + tma_q,mQ = cute.nvgpu.make_tiled_tma_atom_A(utils.sm100.cluster_shape_to_tma_atom_A(self.cluster_shape_mn,qk_mma.thr_id),q,q_s,self.qk_mma_tiler,qk_mma,self.cluster_layout_vmnk.shape) + tma_k,mK = cute.nvgpu.make_tiled_tma_atom_B(utils.sm100.cluster_shape_to_tma_atom_B(self.cluster_shape_mn,qk_mma.thr_id),k,k_s,self.qk_mma_tiler,qk_mma,self.cluster_layout_vmnk.shape) + tma_v,mV = cute.nvgpu.make_tiled_tma_atom_B(utils.sm100.cluster_shape_to_tma_atom_B(self.cluster_shape_mn,pv_mma.thr_id),v_fmha,v_s,self.pv_mma_tiler,pv_mma,self.cluster_layout_vmnk.shape) + epi_s = cute.select(self.c_smem_s,mode=[0,1]) + tma_c,mC = cpasync.make_tiled_tma_atom(cpasync.CopyBulkTensorTileS2GOp(),c,epi_s,self.epi_tile) + self._kernel(qk_mma,pv_mma,tma_q,mQ,tma_k,mK,tma_v,mV,tma_c,mC,self.cluster_layout_vmnk,self.q_smem_s,self.k_smem_s,self.v_smem_s,self.p_tmem_s,self.p_smem_s,self.c_smem_s,self.epi_tile).launch(grid=(1,1,1),block=[self.threads_per_cta,1,1],stream=stream) + + @cute.kernel + def _kernel(self, qk_mma, pv_mma, tma_q, mQ, tma_k, mK, tma_v, mV, tma_c, mC, cl_vmnk, q_smem_s, k_smem_s, v_smem_s, p_tmem_s, p_smem_s, c_smem_s, epi_tile): + warp_idx = cute.arch.make_warp_uniform(cute.arch.warp_idx()) + tidx,_,_ = cute.arch.thread_idx() + if warp_idx == self.tma_warp_id: + cpasync.prefetch_descriptor(tma_q); cpasync.prefetch_descriptor(tma_k); cpasync.prefetch_descriptor(tma_v); cpasync.prefetch_descriptor(tma_c) + + @cute.struct + class SS: + q_bar: cute.struct.MemRange[cutlass.Int64, self.q_stage*2] + kv_bar: cute.struct.MemRange[cutlass.Int64, self.kv_stage*2] + s_bar: cute.struct.MemRange[cutlass.Int64, 2] + acc_bar: cute.struct.MemRange[cutlass.Int64, self.num_acc_stage*2] + tmem_dealloc: cutlass.Int64; holding: cutlass.Int32 + smem = utils.SmemAllocator(); st = smem.allocate(SS) + + qp,qc = pipeline.PipelineTmaUmma.create(barrier_storage=st.q_bar.data_ptr(),num_stages=self.q_stage,producer_group=pipeline.CooperativeGroup(pipeline.Agent.Thread),consumer_group=pipeline.CooperativeGroup(pipeline.Agent.Thread,1),tx_count=self.q_tx_bytes,cta_layout_vmnk=cl_vmnk,defer_sync=True).make_participants() + kvp,kvc = pipeline.PipelineTmaUmma.create(barrier_storage=st.kv_bar.data_ptr(),num_stages=self.kv_stage,producer_group=pipeline.CooperativeGroup(pipeline.Agent.Thread),consumer_group=pipeline.CooperativeGroup(pipeline.Agent.Thread,1),tx_count=self.kv_tx_bytes,cta_layout_vmnk=cl_vmnk,defer_sync=True).make_participants() + s_prod,s_cons = pipeline.PipelineUmmaAsync.create(barrier_storage=st.s_bar.data_ptr(),num_stages=1,producer_group=pipeline.CooperativeGroup(pipeline.Agent.Thread),consumer_group=pipeline.CooperativeGroup(pipeline.Agent.Thread,32*len(self.epilogue_warp_id))).make_participants() + softmax_done_bar = pipeline.NamedBarrier(barrier_id=3, num_threads=32 + 32*len(self.epilogue_warp_id)) + final_o_bar = pipeline.NamedBarrier(barrier_id=4, num_threads=32 + 32*len(self.epilogue_warp_id)) + acc_pipe = pipeline.PipelineUmmaAsync.create(barrier_storage=st.acc_bar.data_ptr(),num_stages=self.num_acc_stage,producer_group=pipeline.CooperativeGroup(pipeline.Agent.Thread),consumer_group=pipeline.CooperativeGroup(pipeline.Agent.Thread,len(self.epilogue_warp_id)),cta_layout_vmnk=cl_vmnk,defer_sync=True) + tmem_bar = pipeline.NamedBarrier(barrier_id=2,num_threads=32*len((self.mma_warp_id,*self.epilogue_warp_id))) + tmem = utils.TmemAllocator(st.holding.ptr,barrier_for_retrieve=tmem_bar,allocator_warp_id=self.epilogue_warp_id[0],is_two_cta=cute.size(qk_mma.thr_id.shape)==2,two_cta_tmem_dealloc_mbar_ptr=st.tmem_dealloc.ptr) + pipeline.pipeline_init_arrive(cluster_shape_mn=cl_vmnk,is_relaxed=True) + + sQ = smem.allocate_tensor(element_type=self.q_dtype,layout=q_smem_s.outer,byte_alignment=128,swizzle=q_smem_s.inner) + sK = smem.allocate_tensor(element_type=self.q_dtype,layout=k_smem_s.outer,byte_alignment=128,swizzle=k_smem_s.inner) + sV = smem.allocate_tensor(element_type=self.q_dtype,layout=v_smem_s.outer,byte_alignment=128,swizzle=v_smem_s.inner) + sC = smem.allocate_tensor(element_type=self.o_dtype,layout=c_smem_s.outer,byte_alignment=128,swizzle=c_smem_s.inner) + sP = smem.allocate_tensor(element_type=self.q_dtype,layout=p_smem_s.outer,byte_alignment=128,swizzle=p_smem_s.inner) + + gQ = cute.local_tile(mQ,cute.slice_(self.qk_mma_tiler,(None,0,None)),(None,None,None)) + gK = cute.local_tile(mK,cute.slice_(self.qk_mma_tiler,(0,None,None)),(None,None,None)) + gV = cute.local_tile(mV,cute.slice_(self.pv_mma_tiler,(0,None,None)),(None,None,None)) + gC = cute.local_tile(mC,cute.slice_(self.pv_mma_tiler,(None,None,0)),(None,None,None)) + n_kv_tiles = cute.size(gK, mode=[3]) + + qk_thr = qk_mma.get_slice(0); pv_thr = pv_mma.get_slice(0) + tCgQ = qk_thr.partition_A(gQ); tCgK = qk_thr.partition_B(gK) + tCgV = pv_thr.partition_B(gV); tCgC = pv_thr.partition_C(gC) + a_lay = cute.make_layout(cute.slice_(cl_vmnk,(0,0,None,0)).shape) + tAsQ,tAgQ = cpasync.tma_partition(tma_q,0,a_lay,cute.group_modes(sQ,0,3),cute.group_modes(tCgQ,0,3)) + b_lay = cute.make_layout(cute.slice_(cl_vmnk,(0,None,0,0)).shape) + tBsK,tBgK = cpasync.tma_partition(tma_k,0,b_lay,cute.group_modes(sK,0,3),cute.group_modes(tCgK,0,3)) + tVsV,tVgV = cpasync.tma_partition(tma_v,0,b_lay,cute.group_modes(sV,0,3),cute.group_modes(tCgV,0,3)) + tAgQ = tAgQ[(None,0,None,0)]; tBgK = tBgK[(None,0,None,0)]; tVgV = tVgV[(None,0,None,0)] + + tCrQ = qk_mma.make_fragment_A(sQ); tCrK = qk_mma.make_fragment_B(sK) + tCrV = pv_mma.make_fragment_B(sV) + + qk_as = qk_thr.partition_shape_C(self.qk_mma_tiler[:2]) + tStS = qk_thr.make_fragment_C(qk_as) + tStS0 = cute.make_tensor(tStS.iterator + self.tmem_s0_offset, tStS.layout) + + # Create coordinate tensor for QK C-fragment layout + # Each element maps to its logical coordinate ((m,n),0,0) + if self.use_smem_p: + cP_qk = cute.make_identity_tensor(tStS0.shape) + print(f"[SMEM-P CUTLASS] Created cP_qk shape: {cute.shape(cP_qk)}") + + pv_as = pv_thr.partition_shape_C(self.pv_mma_tiler[:2]) + tOtO = pv_thr.make_fragment_C(pv_as) + tOtO0 = cute.make_tensor(tOtO.iterator + self.tmem_o0_offset, tOtO.layout) + + # PV A-operand: define both tOrP0 (TMEM-P) and tCrP (SMEM-P) unconditionally. + # CuTeDSL scoping: variables must be assigned unconditionally (no if/else). + tP = cute.make_tensor(tStS.iterator, p_tmem_s.outer) + tOrP_base = pv_thr.make_fragment_A(tP if not self.use_smem_p else sP) + tOrP = tOrP_base[(None,None,None,0)] + tCrP = pv_mma.make_fragment_A(sP) + # tOrP0 always defined as tOrP. The TMEM-P path in the MMA warp applies + # the p0 column offset inline when constructing the gemm arguments. + tOrP0 = tOrP + + tCtO_fake = pv_mma.make_fragment_C(cute.append(pv_as, self.num_acc_stage)) + pipeline.pipeline_init_wait(cluster_shape_mn=cl_vmnk) + + # ===== TMA LOAD warp ===== + if warp_idx == self.tma_warp_id: + qp.reset(); qh = qp.acquire_and_advance() + cute.copy(tma_q, tAgQ[(None, Int32(0))], tAsQ[(None, qh.index)], tma_bar_ptr=qh.barrier) + qp.tail() + kvp.reset(); pk = kvp.try_acquire() + for kt in cutlass.range(0, self.n_kv_tiles, 1, unroll=1): + kvh = kvp.acquire_and_advance(pk) + cute.copy(tma_k, tBgK[(None, kt)], tBsK[(None, kvh.index)], tma_bar_ptr=kvh.barrier) + cute.copy(tma_v, tVgV[(None, kt)], tVsV[(None, kvh.index)], tma_bar_ptr=kvh.barrier) + pk = cutlass.Boolean(1) + kvp.tail() + + # ===== MMA warp ===== + if warp_idx == self.mma_warp_id: + tmem.wait_for_alloc() + qc.reset(); qh = qc.wait_and_advance(); qh.release() + kvc.reset(); pk = kvc.try_wait() + acc_st = pipeline.make_pipeline_state(pipeline.PipelineUserType.Producer, self.num_acc_stage) + acc_pipe.producer_acquire(acc_st) + for kt in range(self.n_kv_tiles): + kvh = kvc.wait_and_advance(pk); pk = cutlass.Boolean(1) + sh = s_prod.acquire_and_advance() + qk_mma.set(tcgen05.Field.ACCUMULATE, False) + for kb in cutlass.range(cute.size(tCrQ, mode=[2]), unroll_full=True): + cute.gemm(qk_mma, tStS0, tCrQ[(None,None,kb,0)], tCrK[(None,None,kb,kvh.index)], tStS0) + qk_mma.set(tcgen05.Field.ACCUMULATE, True) + cute.arch.fence_view_async_tmem_store() + sh.commit() + softmax_done_bar.arrive_and_wait() + pv_mma.set(tcgen05.Field.ACCUMULATE, kt != 0) + if not self.use_smem_p: + # TMEM-P: PV reads P from TMEM + for kb in cutlass.range(cute.size(tOrP0, mode=[2]), unroll_full=True): + cute.gemm(pv_mma, tOtO0, tOrP0[(None,None,kb)], tCrV[(None,None,kb,kvh.index)], tOtO0) + pv_mma.set(tcgen05.Field.ACCUMULATE, True) + else: + # SMEM-P: PV reads P from SMEM + for kb in cutlass.range(cute.size(tCrP, mode=[2]), unroll_full=True): + cute.gemm(pv_mma, tOtO0, tCrP[(None,None,kb,0)], tCrV[(None,None,kb,kvh.index)], tOtO0) + pv_mma.set(tcgen05.Field.ACCUMULATE, True) + cute.arch.fence_view_async_tmem_store() + kvh.release() + acc_pipe.producer_commit(acc_st); acc_st.advance() + final_o_bar.arrive() + acc_pipe.producer_tail(acc_st) + + # ===== SOFTMAX + CORRECTION EPILOGUE warps ===== + if warp_idx < self.mma_warp_id: + tmem.allocate(self.num_tmem_alloc_cols) + tmem.wait_for_alloc() + tmem_ptr = tmem.retrieve_ptr(self.qk_acc_dtype) + sfw_idx = tidx % (32 * len(self.epilogue_warp_id)) + + # S load atoms + tmem_load_atom = cute.make_copy_atom(tcgen05.copy.Ld32x32bOp(tcgen05.copy.Repetition(32)), self.qk_acc_dtype) + tiled_tmem_load = tcgen05.make_tmem_copy(tmem_load_atom, tStS0) + thr_load = tiled_tmem_load.get_slice(sfw_idx) + tTMEM_LOADtS = thr_load.partition_S(tStS0) + cS = cute.make_identity_tensor((self.qk_mma_tiler[0], self.qk_mma_tiler[1])) + tScS = qk_thr.partition_C(cS) + tTMEM_LOADcS = thr_load.partition_D(tScS) + + # P store atoms: TMEM-P (always defined, only used when use_smem_p=False) + p_cols_fp32 = self.pv_mma_tiler[2] * self.q_dtype.width // self.qk_acc_dtype.width + tStP_layout = cute.composition(tStS.layout, cute.make_layout((self.pv_mma_tiler[0], p_cols_fp32))) + # Use 0 as P offset when SMEM-P (these atoms are never used, but must be valid) + tStP0 = cute.make_tensor(tStS.iterator + max(self.tmem_p0_offset, 0), tStP_layout) + tmem_store_atom = cute.make_copy_atom(tcgen05.copy.St32x32bOp(tcgen05.copy.Repetition(32)), self.qk_acc_dtype) + tiled_tmem_store = tcgen05.make_tmem_copy(tmem_store_atom, tStP0) + thr_store = tiled_tmem_store.get_slice(sfw_idx) + tTMEM_STOREtP = thr_store.partition_D(tStP0) + tScP_layout = cute.composition(tScS.layout, cute.make_layout((self.pv_mma_tiler[0], p_cols_fp32))) + tScP = cute.make_tensor(tScS.iterator, tScP_layout) + tTMEM_STOREcP = thr_store.partition_S(tScP) + + # Manual SMEM addressing for P (CUTLASS LLM guidance) + # We need to write P values from QK C-fragment layout to PV A-operand SMEM layout + # sP has PV A-operand SMEM layout: p_smem_s + print(f"[SMEM-P CUTLASS] Starting manual SMEM addressing with CUTLASS LLM pattern") + print(f"[SMEM-P CUTLASS] sP shape: {cute.shape(sP)} layout: {sP.layout}") + + # Get thread index for coordinate partitioning + tidx, _, _ = cute.arch.thread_idx() + warp_idx = cute.arch.make_warp_uniform(cute.arch.warp_idx()) + lane_idx = tidx % 32 + + print(f"[SMEM-P CUTLASS] tidx={tidx}, warp_idx={warp_idx}, lane_idx={lane_idx}") + + row_max = -Float32.inf + row_sum = Float32(0.0) + scale_log2 = Float32(self.scale_softmax_log2) + + # O rescale atoms (hand-constructed, using composition layout like CUTLASS correction_rescale) + corr_tile_size = 16 + tOcO = pv_thr.partition_C(cS) + tOtO_i_layout = cute.composition(tOtO0.layout, cute.make_layout((128, corr_tile_size))) + tOcO_i_layout = cute.composition(tOcO.layout, cute.make_layout((128, corr_tile_size))) + tOtO_i = cute.make_tensor(tOtO0.iterator, tOtO_i_layout) + tOcO_i = cute.make_tensor(tOcO.iterator, tOcO_i_layout) + tmem_load_o_atom = cute.make_copy_atom( + tcgen05.copy.Ld32x32bOp(tcgen05.copy.Repetition(corr_tile_size)), + self.acc_dtype, + ) + tmem_store_o_atom = cute.make_copy_atom( + tcgen05.copy.St32x32bOp(tcgen05.copy.Repetition(corr_tile_size)), + self.acc_dtype, + ) + tiled_tmem_load_o = tcgen05.make_tmem_copy(tmem_load_o_atom, tOtO_i) + tiled_tmem_store_o = tcgen05.make_tmem_copy(tmem_store_o_atom, tOtO_i) + thr_tmem_load_o = tiled_tmem_load_o.get_slice(sfw_idx) + thr_tmem_store_o = tiled_tmem_store_o.get_slice(sfw_idx) + tTMEM_LOADtO = thr_tmem_load_o.partition_S(tOtO_i) + tTMEM_LOADcO = thr_tmem_load_o.partition_D(tOcO_i) + tTMEM_STOREtO = thr_tmem_store_o.partition_D(tOtO_i) + n_corr_tiles = self.pv_n_tile // corr_tile_size + + for kt in range(self.n_kv_tiles): + si_handle = s_cons.wait_and_advance() + + tTMEM_LOADrS = cute.make_rmem_tensor(tTMEM_LOADcS.shape, self.qk_acc_dtype) + cute.copy(tiled_tmem_load, tTMEM_LOADtS, tTMEM_LOADrS) + cute.arch.fence_view_async_tmem_load() + + old_row_max = row_max + frg_cnt = 4 + frg_tile = cute.size(tTMEM_LOADrS) // frg_cnt + # Compute fragment tile size dynamically (must match value division) + frg_tile_size = cute.size(tTMEM_LOADrS) // frg_cnt + frg_layout = cute.make_layout(frg_tile_size) + + tTMEM_LOADrS_frg = cute.logical_divide(tTMEM_LOADrS, frg_layout) + # Coordinate fragments for SMEM-P mapping (needed unconditionally for scoping) + tTMEM_LOADcS_frg = cute.logical_divide(tTMEM_LOADcS, frg_layout) + if self.use_smem_p: + print(f"[SMEM-P CUTLASS] Created tTMEM_LOADcS_frg shape: {cute.shape(tTMEM_LOADcS_frg)}") + print(f"[SMEM-P CUTLASS] tTMEM_LOADrS shape: {cute.shape(tTMEM_LOADrS)}") + print(f"[SMEM-P CUTLASS] tTMEM_LOADcS shape: {cute.shape(tTMEM_LOADcS)}") + print(f"[SMEM-P CUTLASS] frg_tile_size: {frg_tile_size}, frg_layout: {frg_layout}") + print(f"[SMEM-P CUTLASS] tTMEM_LOADrS_frg shape: {cute.shape(tTMEM_LOADrS_frg)}") + + for j in range(frg_cnt): + for k in range(cute.size(tTMEM_LOADrS_frg, mode=[0])): + row_max = cute.arch.fmax(row_max, tTMEM_LOADrS_frg[k, j] * scale_log2) + + row_max_safe = row_max + if row_max == -cutlass.Float32.inf: + row_max_safe = Float32(0.0) + + acc_scale_ = old_row_max - row_max_safe + acc_scale = cute.math.exp2(acc_scale_, fastmath=True) + if old_row_max == -cutlass.Float32.inf: + acc_scale = Float32(0.0) + row_sum *= acc_scale + + rP_words = cute.make_rmem_tensor(tTMEM_STOREcP.shape, self.qk_acc_dtype) + rP_bf16 = cute.make_tensor(cute.recast_ptr(rP_words.iterator, dtype=self.q_dtype), tTMEM_LOADrS.layout) + minus_row_max = Float32(0.0) - row_max_safe + + rP_bf16_frg = cute.logical_divide(rP_bf16, cute.make_layout(frg_tile)) + for j in range(frg_cnt): + for k in range(cute.size(tTMEM_LOADrS_frg, mode=[0])): + tTMEM_LOADrS_frg[k, j] = tTMEM_LOADrS_frg[k, j] * scale_log2 + minus_row_max + tTMEM_LOADrS_frg[k, j] = cute.math.exp2(tTMEM_LOADrS_frg[k, j], fastmath=True) + + # If using SMEM-P, write P value directly to SMEM + if self.use_smem_p: + # Get QK coordinate for this position + qk_coord = tTMEM_LOADcS_frg[k, j] + # qk_coord is (m, n) coordinate + m = qk_coord[0] + n = qk_coord[1] + + # Map to PV SMEM coordinate + # Convert to local coordinates (0-127) as sanity check + m_local = m % 128 + n_local = n % 128 + + # Original mapping formula (should be correct for local coords) + n0 = n_local % 16 + n1 = (n_local // 16) % 4 + n2 = n_local // 64 + pv_coord = ((m_local, n0), 0, (n1, n2), 0) + + # DEBUG: Write pattern based on fragment indices (k,j) + # If coordinates wrong, this pattern might work better + pattern_val = Float32(k) + Float32(j) * Float32(32.0) + p_val_bf16 = pattern_val.to(self.q_dtype) + # Original: p_val_bf16 = tTMEM_LOADrS_frg[k, j].to(self.q_dtype) + sP[pv_coord] = p_val_bf16 # Tensor indexing + + # DEBUG: Print first few coordinates to verify mapping + if self.use_smem_p and k < 2 and j < 2: + print(f"[SMEM-P DEBUG] k={k}, j={j}, qk_coord=({m},{n}), pv_coord={pv_coord}") + # Try to compute offset using crd2idx + try: + offset = cute.crd2idx(pv_coord, sP.layout) + print(f"[SMEM-P DEBUG] offset = {offset}") + except: + print(f"[SMEM-P DEBUG] crd2idx not available") + + # DEBUG: Also write pattern based on fragment indices (k,j) + # If coordinates wrong, this pattern might work better + pattern_val = Float32(k) + Float32(j) * Float32(32.0) + p_val_bf16 = pattern_val.to(self.q_dtype) + # Original: p_val_bf16 = tTMEM_LOADrS_frg[k, j].to(self.q_dtype) + sP[pv_coord] = p_val_bf16 # Tensor indexing + + row_sum = row_sum + tTMEM_LOADrS_frg[k, j] + s_vec = tTMEM_LOADrS_frg[None, j].load() + rP_bf16_frg[None, j].store(s_vec.to(self.q_dtype)) + + if not self.use_smem_p: + # TMEM-P: store P to TMEM via register bridge + cute.copy(tiled_tmem_store, rP_words, tTMEM_STOREtP) + cute.arch.fence_view_async_tmem_store() + else: + # SMEM-P: Already wrote P values to SMEM in softmax loop + # Just need fence and barrier + print(f"[SMEM-P CUTLASS] P values already written to SMEM, proceeding to fence") + + # DEBUG: Compute offset for known coordinate to verify mapping + test_coord = ((0,0), 0, (0,0), 0) + test_offset = cute.crd2idx(test_coord, sP.layout) + print(f"[SMEM-P DEBUG] test_coord {test_coord} -> offset {test_offset}") + + cute.arch.fence_proxy("async.shared", space="cta") + + # Barrier for both TMEM-P and SMEM-P paths + softmax_done_bar.arrive() # Per-tile O rescale (hand-constructed atoms with logical_divide layout) + if kt > 0: + tTMrO = cute.make_rmem_tensor( + (tTMEM_LOADcO.shape, 128 // corr_tile_size), self.acc_dtype + ) + for i in range(n_corr_tiles): + tTMrO_i_ = tTMrO[None, i] + tTMrO_i_layout = cute.composition( + tTMrO_i_.layout, cute.make_layout(tTMrO.shape[0]) + ) + tTMrO_i = cute.make_tensor(tTMrO_i_.iterator, tTMrO_i_layout) + tTMEM_LOADtO_i = cute.make_tensor( + tTMEM_LOADtO.iterator + i * corr_tile_size, + tTMEM_LOADtO.layout, + ) + tTMEM_STOREtO_i = cute.make_tensor( + tTMEM_STOREtO.iterator + i * corr_tile_size, + tTMEM_STOREtO.layout, + ) + cute.copy(tiled_tmem_load_o, tTMEM_LOADtO_i, tTMrO_i) + for k in cutlass.range(cute.size(tTMrO_i), vectorize=True): + tTMrO_i[k] = tTMrO_i[k] * acc_scale + cute.copy(tiled_tmem_store_o, tTMrO_i, tTMEM_STOREtO_i) + cute.arch.fence_view_async_tmem_store() + + si_handle.release() + softmax_done_bar.arrive() + + # Wait for MMA's PV[N-1] to commit before reading O. + final_o_bar.arrive_and_wait() + + # === NO-OP TMEM round-trip: re-map O from MMA layout to epilog layout === + tTMrO_noop = cute.make_rmem_tensor( + (tTMEM_LOADcO.shape, 128 // corr_tile_size), self.acc_dtype + ) + for i in range(n_corr_tiles): + tTMrO_i_ = tTMrO_noop[None, i] + tTMrO_i_layout = cute.composition( + tTMrO_i_.layout, cute.make_layout(tTMrO_noop.shape[0]) + ) + tTMrO_i = cute.make_tensor(tTMrO_i_.iterator, tTMrO_i_layout) + tTMEM_LOADtO_i = cute.make_tensor( + tTMEM_LOADtO.iterator + i * corr_tile_size, + tTMEM_LOADtO.layout, + ) + tTMEM_STOREtO_i = cute.make_tensor( + tTMEM_STOREtO.iterator + i * corr_tile_size, + tTMEM_STOREtO.layout, + ) + cute.copy(tiled_tmem_load_o, tTMEM_LOADtO_i, tTMrO_i) + cute.copy(tiled_tmem_store_o, tTMrO_i, tTMEM_STOREtO_i) + cute.arch.fence_view_async_tmem_store() + + # === Final O normalization: O *= 1/row_sum === + inv_row_sum = Float32(1.0) / row_sum + + tTMrO = cute.make_rmem_tensor( + (tTMEM_LOADcO.shape, 128 // corr_tile_size), self.acc_dtype + ) + + for i in range(n_corr_tiles): + tTMrO_i_ = tTMrO[None, i] + tTMrO_i_layout = cute.composition( + tTMrO_i_.layout, cute.make_layout(tTMrO.shape[0]) + ) + tTMrO_i = cute.make_tensor(tTMrO_i_.iterator, tTMrO_i_layout) + tTMEM_LOADtO_i = cute.make_tensor( + tTMEM_LOADtO.iterator + i * corr_tile_size, tTMEM_LOADtO.layout + ) + tTMEM_STOREtO_i = cute.make_tensor( + tTMEM_STOREtO.iterator + i * corr_tile_size, tTMEM_STOREtO.layout + ) + + cute.copy(tiled_tmem_load_o, tTMEM_LOADtO_i, tTMrO_i) + for j in cutlass.range(cute.size(tTMrO_i), vectorize=True): + tTMrO_i[j] = tTMrO_i[j] * inv_row_sum + cute.copy(tiled_tmem_store_o, tTMrO_i, tTMEM_STOREtO_i) + + cute.arch.fence_view_async_tmem_store() + + # Epilogue: TMEM → SMEM → GMEM via TMA store. + tCtO_base = cute.make_tensor(tmem_ptr + self.tmem_o0_offset, tCtO_fake.layout) + acc_cons_st = pipeline.make_pipeline_state( + pipeline.PipelineUserType.Consumer, self.num_acc_stage + ) + c_grp = pipeline.CooperativeGroup(pipeline.Agent.Thread, 32 * len(self.epilogue_warp_id)) + c_pipe = pipeline.PipelineTmaStore.create(num_stages=self.num_c_stage, producer_group=c_grp) + acc_cons_st = utils.gemm.sm100.epilogue_tma_store( + self, tidx, warp_idx, tma_c, tCtO_base, sC, tCgC, epi_tile, + 0, const_expr(lambda x: x), (0, 0, 0), + acc_cons_st, acc_pipe, c_pipe, + ) + c_pipe.producer_tail() + + tmem.relinquish_alloc_permit() + tmem.free(tmem_ptr)