diff --git a/tests/unit/test_fmha_v3_debug.py b/tests/unit/test_fmha_v3_debug.py new file mode 100644 index 00000000..91ae2a68 --- /dev/null +++ b/tests/unit/test_fmha_v3_debug.py @@ -0,0 +1,469 @@ +""" +FMHA v3 + Stage C: QK -> online softmax -> PV with KV-tile interleaving. +Stage C: row_max, exp2, O rescale, row_sum, final normalization. +FMHA pattern P store preserved from Stage B. +""" +import math +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 + +HEAD_DIM = 64 + +class FmhaV3Softmax: + def __init__(self): + 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 + + 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, HEAD_DIM, 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), HEAD_DIM, 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) + 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; self.tmem_p0_offset = 32 + # P occupies [tmem_p0_offset, tmem_p0_offset + p_cols_fp32) + # S occupies [0, qk_mma_tiler[1]) = [0, 128) + # O must NOT overlap P. Place O after max(S end, P end), aligned to 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 # 32 + 64 = 96 + s_cols = self.qk_mma_tiler[1] # 128 + o_after = max(s_cols, p_end) # 128 + self.tmem_o0_offset = ((o_after + 31) // 32) * 32 + self.tmem_vec_offset = 0 # Reuse S region for per-row inv_row_sum vector # align to 32 = 128 + self.tmem_vec_offset = 0 # Reuse S region (free after softmax loop) + o_cols = find_tmem_tensor_col_offset(tOtO) # footprint of O + total = self.tmem_o0_offset + o_cols + # Must be multiple of 32 AND power of 2 + 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)) + 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) * cta + self.scale_softmax_log2 = Float32(1.0 / math.sqrt(HEAD_DIM) * math.log2(math.e)) + + @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() + # # s_k hardcoded # BROKEN in @cute.jit + # FMHA-style V: reconstruct as (HEAD_DIM, s_k, 1) MN-major + v_fmha = cute.make_tensor( + v.iterator, + cute.make_layout( + (HEAD_DIM, 128, 1), + stride=(1, HEAD_DIM, HEAD_DIM * 128), + ), + ) + 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_mma = utils.sm100.make_trivial_tiled_mma(self.q_dtype, self.q_dtype, cute.nvgpu.OperandMajorMode.K, self.v_major, self.qk_acc_dtype, self.cta_group, (128,HEAD_DIM), tcgen05.OperandSource.TMEM) + 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.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, 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)) + pv_done_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) + + 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) + 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 read view (for MMA only, NOT for softmax store) --- + tP = cute.make_tensor(tStS.iterator, p_tmem_s.outer) + tOrP_base = pv_thr.make_fragment_A(tP) + tOrP = tOrP_base[(None,None,None,0)] + tOrP0 = cute.make_tensor( + tOrP.iterator + self.qk_acc_dtype.width // self.q_dtype.width * self.tmem_p0_offset, + tOrP.layout) + + tCtS_fake = qk_mma.make_fragment_C(cute.append(qk_as, self.num_acc_stage)) + 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 + if warp_idx == self.tma_warp_id: + qp.reset(); qh = qp.acquire_and_advance() + cute.copy(tma_q,tAgQ[(None,qh.count)],tAsQ[(None,qh.index)],tma_bar_ptr=qh.barrier) + qp.tail() + kvp.reset(); pk = kvp.try_acquire() + for kt in cutlass.range(n_kv_tiles,unroll=1): + kh = kvp.acquire_and_advance(pk) + cute.copy(tma_k,tBgK[(None,kh.count)],tBsK[(None,kh.index)],tma_bar_ptr=kh.barrier) + pk = cutlass.Boolean(1) + vh = kvp.acquire_and_advance(pk) + cute.copy(tma_v,tVgV[(None,vh.count)],tVsV[(None,vh.index)],tma_bar_ptr=vh.barrier) + pk = cutlass.Boolean(1) + kvp.tail() + + # MMA + 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(n_kv_tiles): + kh = 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,kh.index)], tStS0) + qk_mma.set(tcgen05.Field.ACCUMULATE, True) + cute.arch.fence_view_async_tmem_store() + sh.commit(); kh.release() + softmax_done_bar.arrive_and_wait() + vh = kvc.wait_and_advance(pk); pk = cutlass.Boolean(1) + pv_mma.set(tcgen05.Field.ACCUMULATE, kt != 0) + 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,vh.index)], tOtO0) + pv_mma.set(tcgen05.Field.ACCUMULATE, True) + cute.arch.fence_view_async_tmem_store() + vh.release() + pv_done_bar.arrive() + acc_pipe.producer_commit(acc_st); acc_st.advance() + acc_pipe.producer_tail(acc_st) + + # ===================== EPILOGUE WARPS (STAGE C: ONLINE SOFTMAX) ===================== + 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 (QK C-fragment) --- + 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 (QK C-fragment composition, FMHA pattern) --- + 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))) + tStP0 = cute.make_tensor(tStS.iterator + self.tmem_p0_offset, 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) + + # --- Vector TMEM (per-row row_sum storage, FMHA pattern) --- + # composition(tStS.layout, (128, 2)) = 2 FP32 columns per logical row + # vec[0] = row_sum (final, after loop), vec[1] = unused + # Reuses S TMEM region (offset 0), free after softmax loop writes + + tStS_vec_layout = cute.composition(tStS.layout, cute.make_layout((128, 2))) + tStS_vec = cute.make_tensor(tStS.iterator + self.tmem_vec_offset, tStS_vec_layout) + tScS_vec_layout = cute.composition(tScS.layout, cute.make_layout((128, 2))) + tScS_vec = cute.make_tensor(tScS.iterator, tScS_vec_layout) + tmem_store_vec_atom = cute.make_copy_atom(tcgen05.copy.St32x32bOp(tcgen05.copy.Repetition(2)), self.qk_acc_dtype) + tiled_tmem_store_vec = tcgen05.make_tmem_copy(tmem_store_vec_atom, tStS_vec) + thr_tmem_store_vec = tiled_tmem_store_vec.get_slice(sfw_idx) + tTMEM_STORE_VECtS = thr_tmem_store_vec.partition_D(tStS_vec) + tTMEM_STORE_VECcS = thr_tmem_store_vec.partition_S(tScS_vec) + tmem_load_vec_atom = cute.make_copy_atom(tcgen05.copy.Ld32x32bOp(tcgen05.copy.Repetition(2)), self.qk_acc_dtype) + tiled_tmem_load_vec = tcgen05.make_tmem_copy(tmem_load_vec_atom, tStS_vec) + thr_tmem_load_vec = tiled_tmem_load_vec.get_slice(sfw_idx) + tTMEM_LOAD_VECtS = thr_tmem_load_vec.partition_S(tStS_vec) + tTMEM_LOAD_VECcS = thr_tmem_load_vec.partition_D(tScS_vec) + + # --- C6: O TMEM load/store for rescale (correction_rescale pattern) --- + corr_tile_size = 16 + cO = cute.make_identity_tensor((self.pv_mma_tiler[0], self.pv_mma_tiler[1])) + tOcO = pv_thr.partition_C(cO) + o_tmem_load_atom = cute.make_copy_atom(tcgen05.copy.Ld32x32bOp(tcgen05.copy.Repetition(corr_tile_size)), self.qk_acc_dtype) + o_tmem_store_atom = cute.make_copy_atom(tcgen05.copy.St32x32bOp(tcgen05.copy.Repetition(corr_tile_size)), self.qk_acc_dtype) + 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) + o_tiled_tmem_load = tcgen05.make_tmem_copy(o_tmem_load_atom, tOtO_i) + o_tiled_tmem_store = tcgen05.make_tmem_copy(o_tmem_store_atom, tOtO_i) + o_thr_load = o_tiled_tmem_load.get_slice(sfw_idx) + o_thr_store = o_tiled_tmem_store.get_slice(sfw_idx) + tTMEM_LOADtO = o_thr_load.partition_S(tOtO_i) + tTMEM_LOADcO = o_thr_load.partition_D(tOcO_i) + tTMEM_STOREtO = o_thr_store.partition_D(tOtO_i) + o_col_tiles = self.pv_mma_tiler[1] // corr_tile_size + + # --- C2: Per-thread row state (persist across KV tiles) --- + row_max = -cutlass.Float32.inf + row_sum = cutlass.Float32(0.0) + + # --- C3: QK scale = 1/sqrt(HEAD_DIM) * log2(e) for exp2 --- + scale = self.scale_softmax_log2 + + # ============================================================= + # Per-KV-tile online softmax loop + # ============================================================= + for kt in range(n_kv_tiles): + si_handle = s_cons.wait_and_advance() + + # Load S from TMEM (FP32, QK C-fragment layout) + tTMEM_LOADrS = cute.make_rmem_tensor(tTMEM_LOADcS.shape, self.qk_acc_dtype) + cute.copy(tiled_tmem_load, tTMEM_LOADtS, tTMEM_LOADrS) + + # --- C4: Compute tile_max via .reduce(MAX) --- + old_row_max = row_max + row_max = tTMEM_LOADrS.load().reduce(cute.ReductionOp.MAX, row_max, 0) + row_max_safe = row_max + if row_max == -cutlass.Float32.inf: + row_max_safe = cutlass.Float32(0.0) + + # --- C5: Compute rescale factor --- + acc_scale = cute.math.exp2(scale * (old_row_max - row_max_safe), fastmath=True) + + # --- C6: Rescale O in TMEM (load O, multiply by acc_scale, store O) --- + # acc_scale belongs to QK row (N//4), but O rows are in PV partition (N). + # Store acc_scale to vector by QK row, read by PV row. + if kt > 0: + pv_done_bar.arrive_and_wait() + + # Store acc_scale to vector indexed by QK logical row + qk_row_c6 = tTMEM_LOADcS[0][0] + thr_vs_c6 = tiled_tmem_store_vec.get_slice(qk_row_c6) + tVStore_c6 = thr_vs_c6.partition_D(tStS_vec) + tVStoreSrc_c6 = thr_vs_c6.partition_S(tScS_vec) + tVStoreRmem_c6 = cute.make_rmem_tensor(tVStoreSrc_c6.shape, self.qk_acc_dtype) + tVStoreRmem_c6[0] = acc_scale + cute.copy(tiled_tmem_store_vec, tVStoreRmem_c6, tVStore_c6) + cute.arch.fence_view_async_tmem_store() + + # Read acc_scale from vector indexed by PV logical row + pv_row_c6 = tTMEM_LOADcO[0][0] + thr_vl_c6 = tiled_tmem_load_vec.get_slice(pv_row_c6) + tVLoad_c6 = thr_vl_c6.partition_S(tStS_vec) + tVLoadDst_c6 = thr_vl_c6.partition_D(tScS_vec) + tVLoadRmem_c6 = cute.make_rmem_tensor(tVLoadDst_c6.shape, self.qk_acc_dtype) + cute.copy(tiled_tmem_load_vec, tVLoad_c6, tVLoadRmem_c6) + cute.arch.fence_view_async_tmem_load() + acc_scale_pv = tVLoadRmem_c6[0] + + tTMrO = cute.make_rmem_tensor((tTMEM_LOADcO.shape, o_col_tiles), self.qk_acc_dtype) + for i in range(o_col_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(o_tiled_tmem_load, tTMEM_LOADtO_i, tTMrO_i) + for j in cutlass.range(cute.size(tTMrO_i), vectorize=True): + tTMrO_i[j] = tTMrO_i[j] * acc_scale_pv + cute.copy(o_tiled_tmem_store, tTMrO_i, tTMEM_STOREtO_i) + cute.arch.fence_view_async_tmem_store() + + # Rescale row_sum + row_sum = row_sum * acc_scale + + # --- C7: Compute P = exp2((S - row_max_safe) * scale) --- + minus_row_max_scale = (cutlass.Float32(0.0) - row_max_safe) * scale + + # Register bridge (FMHA pattern: FP32 backing + BF16 view) + 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) + + frg_cnt = 4 + frg_tile = cute.size(tTMEM_LOADrS) // frg_cnt + tTMEM_LOADrS_frg = cute.logical_divide(tTMEM_LOADrS, cute.make_layout(frg_tile)) + rP_bf16_frg = cute.logical_divide(rP_bf16, cute.make_layout(frg_tile)) + + # Scale S, compute exp2, store through register bridge + for j in range(frg_cnt): + for k in cutlass.range(cute.size(tTMEM_LOADrS_frg, mode=[0]), vectorize=True): + tTMEM_LOADrS_frg[k, j] = tTMEM_LOADrS_frg[k, j] * scale + minus_row_max_scale + tTMEM_LOADrS_frg[k, j] = cute.math.exp2(tTMEM_LOADrS_frg[k, j], fastmath=True) + s_vec = tTMEM_LOADrS_frg[None, j].load() + rP_bf16_frg[None, j].store(s_vec.to(self.q_dtype)) + + # Store P to TMEM + cute.copy(tiled_tmem_store, rP_words, tTMEM_STOREtP) + cute.arch.fence_view_async_tmem_store() + si_handle.release() + softmax_done_bar.arrive() + + # --- C8: Row sum accumulation (CUTLASS FMHA packed f32x2 pattern) --- + # P values still in tTMEM_LOADrS registers. + # 4 accumulators for 4 reduction_unroll columns. + local_row_sum_0 = (cutlass.Float32(0.0), cutlass.Float32(0.0)) + local_row_sum_1 = (cutlass.Float32(0.0), cutlass.Float32(0.0)) + local_row_sum_2 = (cutlass.Float32(0.0), cutlass.Float32(0.0)) + local_row_sum_3 = (cutlass.Float32(0.0), cutlass.Float32(0.0)) + + reduction_unroll = 4 + rfrg_tile = cute.size(tTMEM_LOADrS) // reduction_unroll + tTMEM_LOADrS_rfrg = cute.logical_divide(tTMEM_LOADrS, cute.make_layout(rfrg_tile)) + + for j in cutlass.range_constexpr(0, cute.size(tTMEM_LOADrS_rfrg, mode=[0]), 2): + local_row_sum_0 = cute.arch.add_packed_f32x2( + local_row_sum_0, (tTMEM_LOADrS_rfrg[j, 0], tTMEM_LOADrS_rfrg[j + 1, 0])) + local_row_sum_1 = cute.arch.add_packed_f32x2( + local_row_sum_1, (tTMEM_LOADrS_rfrg[j, 1], tTMEM_LOADrS_rfrg[j + 1, 1])) + local_row_sum_2 = cute.arch.add_packed_f32x2( + local_row_sum_2, (tTMEM_LOADrS_rfrg[j, 2], tTMEM_LOADrS_rfrg[j + 1, 2])) + local_row_sum_3 = cute.arch.add_packed_f32x2( + local_row_sum_3, (tTMEM_LOADrS_rfrg[j, 3], tTMEM_LOADrS_rfrg[j + 1, 3])) + + local_row_sum_0 = cute.arch.add_packed_f32x2(local_row_sum_0, local_row_sum_1) + local_row_sum_2 = cute.arch.add_packed_f32x2(local_row_sum_2, local_row_sum_3) + local_row_sum_0 = cute.arch.add_packed_f32x2(local_row_sum_0, local_row_sum_2) + tile_sum = local_row_sum_0[0] + local_row_sum_0[1] + + row_sum = row_sum + tile_sum + + # --- C9: Final normalization via O TMEM rescale --- + pv_done_bar.arrive_and_wait() + + # Use QK-accumulated row_sum directly (DEBUG: check if row mapping matches PV) + inv_row_sum = cutlass.Float32(1.0) / row_sum + + # Normalize O in TMEM using PV-correct inv_row_sum + tTMrO_final = cute.make_rmem_tensor((tTMEM_LOADcO.shape, o_col_tiles), self.qk_acc_dtype) + for i in range(o_col_tiles): + tTMrO_i_ = tTMrO_final[None, i] + tTMrO_i_layout = cute.composition(tTMrO_i_.layout, cute.make_layout(tTMrO_final.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(o_tiled_tmem_load, 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(o_tiled_tmem_store, tTMrO_i, tTMEM_STOREtO_i) + cute.arch.fence_view_async_tmem_store() + + # Now O in TMEM is normalized. Use standard epilogue_tma_store with identity. + 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) + + +def test(): + import math + torch.manual_seed(42) + for n in [128, 256, 384]: + m, hd = 128, HEAD_DIM + q = torch.randn(m, hd, 1, dtype=torch.bfloat16, device="cuda") + k = torch.randn(n, hd, 1, dtype=torch.bfloat16, device="cuda") + v = torch.randn(n, hd, dtype=torch.bfloat16, device="cuda") + v_kernel = v.unsqueeze(-1) + c = torch.zeros(m, hd, 1, dtype=torch.bfloat16, device="cuda") + qf = q[:,:,0].float(); kf = k[:,:,0].float() + attn = qf @ kf.T / math.sqrt(hd) + ref = torch.softmax(attn, dim=-1) @ v.float() + mQ = ct.from_dlpack(q).mark_layout_dynamic(leading_dim=ct.get_leading_dim(q)) + mK = ct.from_dlpack(k).mark_layout_dynamic(leading_dim=ct.get_leading_dim(k)) + mV = ct.from_dlpack(v_kernel).mark_layout_dynamic(leading_dim=ct.get_leading_dim(v_kernel)) + mC = ct.from_dlpack(c).mark_layout_dynamic(leading_dim=ct.get_leading_dim(c)) + stream = cuda.CUstream(torch.cuda.current_stream().cuda_stream) + kernel = FmhaV3Softmax() + print(f"n={n}: Compiling...", flush=True) + compiled = cute.compile(kernel, mQ, mK, mV, mC, stream) + print(f"n={n}: tmem: s0={kernel.tmem_s0_offset} p0={kernel.tmem_p0_offset} o0={kernel.tmem_o0_offset} vec={kernel.tmem_vec_offset} alloc={kernel.num_tmem_alloc_cols}", flush=True) + print(f"n={n}: Running...", flush=True) + compiled(mQ, mK, mV, mC, stream) + torch.cuda.synchronize() + out = c[:,:,0].float() + cos = torch.nn.functional.cosine_similarity(out.flatten().unsqueeze(0), ref.flatten().unsqueeze(0)).item() + max_err = (out - ref).abs().max().item() + print(f"FMHA softmax n={n}: cosine {cos:.6f} max_err {max_err:.6f} {'PASS' if cos >= 0.999 else 'FAIL'}", flush=True) + +if __name__ == "__main__": + test() + + diff --git a/tests/unit/test_fmha_v3_noop_c9.py b/tests/unit/test_fmha_v3_noop_c9.py new file mode 100644 index 00000000..b9a11c6c --- /dev/null +++ b/tests/unit/test_fmha_v3_noop_c9.py @@ -0,0 +1,469 @@ +""" +FMHA v3 + Stage C: QK -> online softmax -> PV with KV-tile interleaving. +Stage C: row_max, exp2, O rescale, row_sum, final normalization. +FMHA pattern P store preserved from Stage B. +""" +import math +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 + +HEAD_DIM = 64 + +class FmhaV3Softmax: + def __init__(self): + 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 + + 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, HEAD_DIM, 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), HEAD_DIM, 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) + 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; self.tmem_p0_offset = 32 + # P occupies [tmem_p0_offset, tmem_p0_offset + p_cols_fp32) + # S occupies [0, qk_mma_tiler[1]) = [0, 128) + # O must NOT overlap P. Place O after max(S end, P end), aligned to 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 # 32 + 64 = 96 + s_cols = self.qk_mma_tiler[1] # 128 + o_after = max(s_cols, p_end) # 128 + self.tmem_o0_offset = ((o_after + 31) // 32) * 32 + self.tmem_vec_offset = 0 # Reuse S region for per-row inv_row_sum vector # align to 32 = 128 + self.tmem_vec_offset = 0 # Reuse S region (free after softmax loop) + o_cols = find_tmem_tensor_col_offset(tOtO) # footprint of O + total = self.tmem_o0_offset + o_cols + # Must be multiple of 32 AND power of 2 + 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)) + 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) * cta + self.scale_softmax_log2 = Float32(1.0 / math.sqrt(HEAD_DIM) * math.log2(math.e)) + + @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() + # # s_k hardcoded # BROKEN in @cute.jit + # FMHA-style V: reconstruct as (HEAD_DIM, s_k, 1) MN-major + v_fmha = cute.make_tensor( + v.iterator, + cute.make_layout( + (HEAD_DIM, 128, 1), + stride=(1, HEAD_DIM, HEAD_DIM * 128), + ), + ) + 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_mma = utils.sm100.make_trivial_tiled_mma(self.q_dtype, self.q_dtype, cute.nvgpu.OperandMajorMode.K, self.v_major, self.qk_acc_dtype, self.cta_group, (128,HEAD_DIM), tcgen05.OperandSource.TMEM) + 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.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, 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)) + pv_done_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) + + 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) + 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 read view (for MMA only, NOT for softmax store) --- + tP = cute.make_tensor(tStS.iterator, p_tmem_s.outer) + tOrP_base = pv_thr.make_fragment_A(tP) + tOrP = tOrP_base[(None,None,None,0)] + tOrP0 = cute.make_tensor( + tOrP.iterator + self.qk_acc_dtype.width // self.q_dtype.width * self.tmem_p0_offset, + tOrP.layout) + + tCtS_fake = qk_mma.make_fragment_C(cute.append(qk_as, self.num_acc_stage)) + 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 + if warp_idx == self.tma_warp_id: + qp.reset(); qh = qp.acquire_and_advance() + cute.copy(tma_q,tAgQ[(None,qh.count)],tAsQ[(None,qh.index)],tma_bar_ptr=qh.barrier) + qp.tail() + kvp.reset(); pk = kvp.try_acquire() + for kt in cutlass.range(n_kv_tiles,unroll=1): + kh = kvp.acquire_and_advance(pk) + cute.copy(tma_k,tBgK[(None,kh.count)],tBsK[(None,kh.index)],tma_bar_ptr=kh.barrier) + pk = cutlass.Boolean(1) + vh = kvp.acquire_and_advance(pk) + cute.copy(tma_v,tVgV[(None,vh.count)],tVsV[(None,vh.index)],tma_bar_ptr=vh.barrier) + pk = cutlass.Boolean(1) + kvp.tail() + + # MMA + 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(n_kv_tiles): + kh = 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,kh.index)], tStS0) + qk_mma.set(tcgen05.Field.ACCUMULATE, True) + cute.arch.fence_view_async_tmem_store() + sh.commit(); kh.release() + softmax_done_bar.arrive_and_wait() + vh = kvc.wait_and_advance(pk); pk = cutlass.Boolean(1) + pv_mma.set(tcgen05.Field.ACCUMULATE, kt != 0) + 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,vh.index)], tOtO0) + pv_mma.set(tcgen05.Field.ACCUMULATE, True) + cute.arch.fence_view_async_tmem_store() + vh.release() + pv_done_bar.arrive() + acc_pipe.producer_commit(acc_st); acc_st.advance() + acc_pipe.producer_tail(acc_st) + + # ===================== EPILOGUE WARPS (STAGE C: ONLINE SOFTMAX) ===================== + 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 (QK C-fragment) --- + 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 (QK C-fragment composition, FMHA pattern) --- + 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))) + tStP0 = cute.make_tensor(tStS.iterator + self.tmem_p0_offset, 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) + + # --- Vector TMEM (per-row row_sum storage, FMHA pattern) --- + # composition(tStS.layout, (128, 2)) = 2 FP32 columns per logical row + # vec[0] = row_sum (final, after loop), vec[1] = unused + # Reuses S TMEM region (offset 0), free after softmax loop writes + + tStS_vec_layout = cute.composition(tStS.layout, cute.make_layout((128, 2))) + tStS_vec = cute.make_tensor(tStS.iterator + self.tmem_vec_offset, tStS_vec_layout) + tScS_vec_layout = cute.composition(tScS.layout, cute.make_layout((128, 2))) + tScS_vec = cute.make_tensor(tScS.iterator, tScS_vec_layout) + tmem_store_vec_atom = cute.make_copy_atom(tcgen05.copy.St32x32bOp(tcgen05.copy.Repetition(2)), self.qk_acc_dtype) + tiled_tmem_store_vec = tcgen05.make_tmem_copy(tmem_store_vec_atom, tStS_vec) + thr_tmem_store_vec = tiled_tmem_store_vec.get_slice(sfw_idx) + tTMEM_STORE_VECtS = thr_tmem_store_vec.partition_D(tStS_vec) + tTMEM_STORE_VECcS = thr_tmem_store_vec.partition_S(tScS_vec) + tmem_load_vec_atom = cute.make_copy_atom(tcgen05.copy.Ld32x32bOp(tcgen05.copy.Repetition(2)), self.qk_acc_dtype) + tiled_tmem_load_vec = tcgen05.make_tmem_copy(tmem_load_vec_atom, tStS_vec) + thr_tmem_load_vec = tiled_tmem_load_vec.get_slice(sfw_idx) + tTMEM_LOAD_VECtS = thr_tmem_load_vec.partition_S(tStS_vec) + tTMEM_LOAD_VECcS = thr_tmem_load_vec.partition_D(tScS_vec) + + # --- C6: O TMEM load/store for rescale (correction_rescale pattern) --- + corr_tile_size = 16 + cO = cute.make_identity_tensor((self.pv_mma_tiler[0], self.pv_mma_tiler[1])) + tOcO = pv_thr.partition_C(cO) + o_tmem_load_atom = cute.make_copy_atom(tcgen05.copy.Ld32x32bOp(tcgen05.copy.Repetition(corr_tile_size)), self.qk_acc_dtype) + o_tmem_store_atom = cute.make_copy_atom(tcgen05.copy.St32x32bOp(tcgen05.copy.Repetition(corr_tile_size)), self.qk_acc_dtype) + 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) + o_tiled_tmem_load = tcgen05.make_tmem_copy(o_tmem_load_atom, tOtO_i) + o_tiled_tmem_store = tcgen05.make_tmem_copy(o_tmem_store_atom, tOtO_i) + o_thr_load = o_tiled_tmem_load.get_slice(sfw_idx) + o_thr_store = o_tiled_tmem_store.get_slice(sfw_idx) + tTMEM_LOADtO = o_thr_load.partition_S(tOtO_i) + tTMEM_LOADcO = o_thr_load.partition_D(tOcO_i) + tTMEM_STOREtO = o_thr_store.partition_D(tOtO_i) + o_col_tiles = self.pv_mma_tiler[1] // corr_tile_size + + # --- C2: Per-thread row state (persist across KV tiles) --- + row_max = -cutlass.Float32.inf + row_sum = cutlass.Float32(0.0) + + # --- C3: QK scale = 1/sqrt(HEAD_DIM) * log2(e) for exp2 --- + scale = self.scale_softmax_log2 + + # ============================================================= + # Per-KV-tile online softmax loop + # ============================================================= + for kt in range(n_kv_tiles): + si_handle = s_cons.wait_and_advance() + + # Load S from TMEM (FP32, QK C-fragment layout) + tTMEM_LOADrS = cute.make_rmem_tensor(tTMEM_LOADcS.shape, self.qk_acc_dtype) + cute.copy(tiled_tmem_load, tTMEM_LOADtS, tTMEM_LOADrS) + + # --- C4: Compute tile_max via .reduce(MAX) --- + old_row_max = row_max + row_max = tTMEM_LOADrS.load().reduce(cute.ReductionOp.MAX, row_max, 0) + row_max_safe = row_max + if row_max == -cutlass.Float32.inf: + row_max_safe = cutlass.Float32(0.0) + + # --- C5: Compute rescale factor --- + acc_scale = cute.math.exp2(scale * (old_row_max - row_max_safe), fastmath=True) + + # --- C6: Rescale O in TMEM (load O, multiply by acc_scale, store O) --- + # acc_scale belongs to QK row (N//4), but O rows are in PV partition (N). + # Store acc_scale to vector by QK row, read by PV row. + if kt > 0: + pv_done_bar.arrive_and_wait() + + # Store acc_scale to vector indexed by QK logical row + qk_row_c6 = tTMEM_LOADcS[0][0] + thr_vs_c6 = tiled_tmem_store_vec.get_slice(qk_row_c6) + tVStore_c6 = thr_vs_c6.partition_D(tStS_vec) + tVStoreSrc_c6 = thr_vs_c6.partition_S(tScS_vec) + tVStoreRmem_c6 = cute.make_rmem_tensor(tVStoreSrc_c6.shape, self.qk_acc_dtype) + tVStoreRmem_c6[0] = acc_scale + cute.copy(tiled_tmem_store_vec, tVStoreRmem_c6, tVStore_c6) + cute.arch.fence_view_async_tmem_store() + + # Read acc_scale from vector indexed by PV logical row + pv_row_c6 = tTMEM_LOADcO[0][0] + thr_vl_c6 = tiled_tmem_load_vec.get_slice(pv_row_c6) + tVLoad_c6 = thr_vl_c6.partition_S(tStS_vec) + tVLoadDst_c6 = thr_vl_c6.partition_D(tScS_vec) + tVLoadRmem_c6 = cute.make_rmem_tensor(tVLoadDst_c6.shape, self.qk_acc_dtype) + cute.copy(tiled_tmem_load_vec, tVLoad_c6, tVLoadRmem_c6) + cute.arch.fence_view_async_tmem_load() + acc_scale_pv = tVLoadRmem_c6[0] + + tTMrO = cute.make_rmem_tensor((tTMEM_LOADcO.shape, o_col_tiles), self.qk_acc_dtype) + for i in range(o_col_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(o_tiled_tmem_load, tTMEM_LOADtO_i, tTMrO_i) + for j in cutlass.range(cute.size(tTMrO_i), vectorize=True): + tTMrO_i[j] = tTMrO_i[j] * acc_scale_pv + cute.copy(o_tiled_tmem_store, tTMrO_i, tTMEM_STOREtO_i) + cute.arch.fence_view_async_tmem_store() + + # Rescale row_sum + row_sum = row_sum * acc_scale + + # --- C7: Compute P = exp2((S - row_max_safe) * scale) --- + minus_row_max_scale = (cutlass.Float32(0.0) - row_max_safe) * scale + + # Register bridge (FMHA pattern: FP32 backing + BF16 view) + 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) + + frg_cnt = 4 + frg_tile = cute.size(tTMEM_LOADrS) // frg_cnt + tTMEM_LOADrS_frg = cute.logical_divide(tTMEM_LOADrS, cute.make_layout(frg_tile)) + rP_bf16_frg = cute.logical_divide(rP_bf16, cute.make_layout(frg_tile)) + + # Scale S, compute exp2, store through register bridge + for j in range(frg_cnt): + for k in cutlass.range(cute.size(tTMEM_LOADrS_frg, mode=[0]), vectorize=True): + tTMEM_LOADrS_frg[k, j] = tTMEM_LOADrS_frg[k, j] * scale + minus_row_max_scale + tTMEM_LOADrS_frg[k, j] = cute.math.exp2(tTMEM_LOADrS_frg[k, j], fastmath=True) + s_vec = tTMEM_LOADrS_frg[None, j].load() + rP_bf16_frg[None, j].store(s_vec.to(self.q_dtype)) + + # Store P to TMEM + cute.copy(tiled_tmem_store, rP_words, tTMEM_STOREtP) + cute.arch.fence_view_async_tmem_store() + si_handle.release() + softmax_done_bar.arrive() + + # --- C8: Row sum accumulation (CUTLASS FMHA packed f32x2 pattern) --- + # P values still in tTMEM_LOADrS registers. + # 4 accumulators for 4 reduction_unroll columns. + local_row_sum_0 = (cutlass.Float32(0.0), cutlass.Float32(0.0)) + local_row_sum_1 = (cutlass.Float32(0.0), cutlass.Float32(0.0)) + local_row_sum_2 = (cutlass.Float32(0.0), cutlass.Float32(0.0)) + local_row_sum_3 = (cutlass.Float32(0.0), cutlass.Float32(0.0)) + + reduction_unroll = 4 + rfrg_tile = cute.size(tTMEM_LOADrS) // reduction_unroll + tTMEM_LOADrS_rfrg = cute.logical_divide(tTMEM_LOADrS, cute.make_layout(rfrg_tile)) + + for j in cutlass.range_constexpr(0, cute.size(tTMEM_LOADrS_rfrg, mode=[0]), 2): + local_row_sum_0 = cute.arch.add_packed_f32x2( + local_row_sum_0, (tTMEM_LOADrS_rfrg[j, 0], tTMEM_LOADrS_rfrg[j + 1, 0])) + local_row_sum_1 = cute.arch.add_packed_f32x2( + local_row_sum_1, (tTMEM_LOADrS_rfrg[j, 1], tTMEM_LOADrS_rfrg[j + 1, 1])) + local_row_sum_2 = cute.arch.add_packed_f32x2( + local_row_sum_2, (tTMEM_LOADrS_rfrg[j, 2], tTMEM_LOADrS_rfrg[j + 1, 2])) + local_row_sum_3 = cute.arch.add_packed_f32x2( + local_row_sum_3, (tTMEM_LOADrS_rfrg[j, 3], tTMEM_LOADrS_rfrg[j + 1, 3])) + + local_row_sum_0 = cute.arch.add_packed_f32x2(local_row_sum_0, local_row_sum_1) + local_row_sum_2 = cute.arch.add_packed_f32x2(local_row_sum_2, local_row_sum_3) + local_row_sum_0 = cute.arch.add_packed_f32x2(local_row_sum_0, local_row_sum_2) + tile_sum = local_row_sum_0[0] + local_row_sum_0[1] + + row_sum = row_sum + tile_sum + + # --- C9: Final normalization via O TMEM rescale --- + pv_done_bar.arrive_and_wait() + + # DEBUG: hardcoded inv_row_sum = 1.0 (no normalization) + inv_row_sum = cutlass.Float32(1.0) + + # Normalize O in TMEM using PV-correct inv_row_sum + tTMrO_final = cute.make_rmem_tensor((tTMEM_LOADcO.shape, o_col_tiles), self.qk_acc_dtype) + for i in range(o_col_tiles): + tTMrO_i_ = tTMrO_final[None, i] + tTMrO_i_layout = cute.composition(tTMrO_i_.layout, cute.make_layout(tTMrO_final.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(o_tiled_tmem_load, 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(o_tiled_tmem_store, tTMrO_i, tTMEM_STOREtO_i) + cute.arch.fence_view_async_tmem_store() + + # Now O in TMEM is normalized. Use standard epilogue_tma_store with identity. + 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) + + +def test(): + import math + torch.manual_seed(42) + for n in [128, 256, 384]: + m, hd = 128, HEAD_DIM + q = torch.randn(m, hd, 1, dtype=torch.bfloat16, device="cuda") + k = torch.randn(n, hd, 1, dtype=torch.bfloat16, device="cuda") + v = torch.randn(n, hd, dtype=torch.bfloat16, device="cuda") + v_kernel = v.unsqueeze(-1) + c = torch.zeros(m, hd, 1, dtype=torch.bfloat16, device="cuda") + qf = q[:,:,0].float(); kf = k[:,:,0].float() + attn = qf @ kf.T / math.sqrt(hd) + ref = torch.softmax(attn, dim=-1) @ v.float() + mQ = ct.from_dlpack(q).mark_layout_dynamic(leading_dim=ct.get_leading_dim(q)) + mK = ct.from_dlpack(k).mark_layout_dynamic(leading_dim=ct.get_leading_dim(k)) + mV = ct.from_dlpack(v_kernel).mark_layout_dynamic(leading_dim=ct.get_leading_dim(v_kernel)) + mC = ct.from_dlpack(c).mark_layout_dynamic(leading_dim=ct.get_leading_dim(c)) + stream = cuda.CUstream(torch.cuda.current_stream().cuda_stream) + kernel = FmhaV3Softmax() + print(f"n={n}: Compiling...", flush=True) + compiled = cute.compile(kernel, mQ, mK, mV, mC, stream) + print(f"n={n}: tmem: s0={kernel.tmem_s0_offset} p0={kernel.tmem_p0_offset} o0={kernel.tmem_o0_offset} vec={kernel.tmem_vec_offset} alloc={kernel.num_tmem_alloc_cols}", flush=True) + print(f"n={n}: Running...", flush=True) + compiled(mQ, mK, mV, mC, stream) + torch.cuda.synchronize() + out = c[:,:,0].float() + cos = torch.nn.functional.cosine_similarity(out.flatten().unsqueeze(0), ref.flatten().unsqueeze(0)).item() + max_err = (out - ref).abs().max().item() + print(f"FMHA softmax n={n}: cosine {cos:.6f} max_err {max_err:.6f} {'PASS' if cos >= 0.999 else 'FAIL'}", flush=True) + +if __name__ == "__main__": + test() + + diff --git a/tests/unit/test_fmha_v3_proper.py b/tests/unit/test_fmha_v3_proper.py new file mode 100644 index 00000000..3db346e2 --- /dev/null +++ b/tests/unit/test_fmha_v3_proper.py @@ -0,0 +1,416 @@ +""" +FMHA v3 Proper: 11-warp with correction warp group + epilogue warp. +Warp layout: softmax(0-3), correction(4-7), MMA(8), TMA(9), epilogue(10) +""" +import math, 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, cutlass.torch as ct + +HEAD_DIM = 64 + +class FmhaV3Proper: + def __init__(self): + self.qk_acc_dtype = Float32; self.pv_acc_dtype = Float32 + self.q_dtype = BFloat16; self.o_dtype = BFloat16; self.c_dtype = BFloat16 + self.use_2cta_instrs = False; self.cluster_shape_mn = (1, 1) + self.cta_group = tcgen05.CtaGroup.ONE + self.softmax_warp_ids = (0,1,2,3) + self.correction_warp_ids = (4,5,6,7) + self.mma_warp_id = 8; self.tma_warp_id = 9; self.epilogue_warp_id = 10 + self.threads_per_cta = 352 + self.q_stage = 1; self.kv_stage = 2; self.num_acc_stage = 1 + self.mma_softmax_stage = 1; self.softmax_corr_stage = 1 + self.mma_corr_stage = 2; self.epi_stage = 2; self.num_c_stage = 2 + self.scale_softmax_log2 = Float32(1.0 / math.sqrt(HEAD_DIM) * 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, HEAD_DIM, 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), HEAD_DIM, 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.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) + 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; self.tmem_p0_offset = 32; self.tmem_vec0_offset = 0 + p_cols_fp32 = self.pv_mma_tiler[2] * self.q_dtype.width // self.qk_acc_dtype.width + o_after = max(self.qk_mma_tiler[1], self.tmem_p0_offset + p_cols_fp32) + self.tmem_o0_offset = ((o_after + 31) // 32) * 32 + o_cols = find_tmem_tensor_col_offset(tOtO) + total = self.tmem_o0_offset + 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)) + 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) * 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((HEAD_DIM, 128, 1), stride=(1, HEAD_DIM, HEAD_DIM * 128))) + 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_mma = utils.sm100.make_trivial_tiled_mma(self.q_dtype, self.q_dtype, cute.nvgpu.OperandMajorMode.K, self.v_major, self.pv_acc_dtype, self.cta_group, (128,HEAD_DIM), tcgen05.OperandSource.TMEM) + 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.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, 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] + mma_si_bar: cute.struct.MemRange[cutlass.Int64, self.mma_softmax_stage*2] + si_corr_bar: cute.struct.MemRange[cutlass.Int64, self.softmax_corr_stage*2] + mma_corr_bar: cute.struct.MemRange[cutlass.Int64, self.mma_corr_stage*2] + corr_epi_bar: cute.struct.MemRange[cutlass.Int64, self.epi_stage*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() + mma_si_prod,mma_si_cons = pipeline.PipelineUmmaAsync.create(barrier_storage=st.mma_si_bar.data_ptr(),num_stages=self.mma_softmax_stage,producer_group=pipeline.CooperativeGroup(pipeline.Agent.Thread),consumer_group=pipeline.CooperativeGroup(pipeline.Agent.Thread,32*len(self.softmax_warp_ids))).make_participants() + si_corr_prod,si_corr_cons = pipeline.PipelineAsync.create(barrier_storage=st.si_corr_bar.data_ptr(),num_stages=self.softmax_corr_stage,producer_group=pipeline.CooperativeGroup(pipeline.Agent.Thread,32*len(self.softmax_warp_ids)),consumer_group=pipeline.CooperativeGroup(pipeline.Agent.Thread,32*len(self.correction_warp_ids))).make_participants() + mma_corr_prod,mma_corr_cons = pipeline.PipelineUmmaAsync.create(barrier_storage=st.mma_corr_bar.data_ptr(),num_stages=self.mma_corr_stage,producer_group=pipeline.CooperativeGroup(pipeline.Agent.Thread),consumer_group=pipeline.CooperativeGroup(pipeline.Agent.Thread,32*len(self.correction_warp_ids))).make_participants() + corr_epi_prod,corr_epi_cons = pipeline.PipelineAsync.create(barrier_storage=st.corr_epi_bar.data_ptr(),num_stages=self.epi_stage,producer_group=pipeline.CooperativeGroup(pipeline.Agent.Thread,32*len(self.correction_warp_ids)),consumer_group=pipeline.CooperativeGroup(pipeline.Agent.Thread,32)).make_participants() + 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.correction_warp_ids)),cta_layout_vmnk=cl_vmnk,defer_sync=True) + tmem_bar = pipeline.NamedBarrier(barrier_id=2,num_threads=32*len((self.mma_warp_id,*self.softmax_warp_ids))) + tmem = utils.TmemAllocator(st.holding.ptr,barrier_for_retrieve=tmem_bar,allocator_warp_id=self.softmax_warp_ids[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) + 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) + 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) + tP = cute.make_tensor(tStS.iterator, p_tmem_s.outer) + tOrP_base = pv_thr.make_fragment_A(tP); tOrP = tOrP_base[(None,None,None,0)] + tOrP0 = cute.make_tensor(tOrP.iterator + self.qk_acc_dtype.width // self.q_dtype.width * self.tmem_p0_offset, tOrP.layout) + 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 + if warp_idx == self.tma_warp_id: + qp.reset(); qh = qp.acquire_and_advance() + cute.copy(tma_q,tAgQ[(None,qh.count)],tAsQ[(None,qh.index)],tma_bar_ptr=qh.barrier); qp.tail() + kvp.reset(); pk = kvp.try_acquire() + for kt in cutlass.range(n_kv_tiles,unroll=1): + kh = kvp.acquire_and_advance(pk); cute.copy(tma_k,tBgK[(None,kh.count)],tBsK[(None,kh.index)],tma_bar_ptr=kh.barrier); pk = cutlass.Boolean(1) + vh = kvp.acquire_and_advance(pk); cute.copy(tma_v,tVgV[(None,vh.count)],tVsV[(None,vh.index)],tma_bar_ptr=vh.barrier); pk = cutlass.Boolean(1) + kvp.tail() + + # MMA + 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(n_kv_tiles): + kh = kvc.wait_and_advance(pk); pk = cutlass.Boolean(1) + sh = mma_si_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,kh.index)], tStS0) + qk_mma.set(tcgen05.Field.ACCUMULATE, True) + cute.arch.fence_view_async_tmem_store(); sh.commit(); kh.release() + if kt > 0: + o_handle = mma_corr_cons.wait_and_advance(); o_handle.release() + sh2 = mma_si_prod.acquire_and_advance() + vh = kvc.wait_and_advance(pk); pk = cutlass.Boolean(1) + pv_mma.set(tcgen05.Field.ACCUMULATE, kt != 0) + 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,vh.index)], tOtO0) + pv_mma.set(tcgen05.Field.ACCUMULATE, True) + cute.arch.fence_view_async_tmem_store(); vh.release() + o_prod_h = mma_corr_prod.acquire_and_advance(); o_prod_h.commit() + o_handle = mma_corr_cons.wait_and_advance(); o_handle.release() + acc_pipe.producer_commit(acc_st); acc_st.advance(); acc_pipe.producer_tail(acc_st) + + # SOFTMAX (warps 0-3) + if warp_idx < len(self.softmax_warp_ids): + 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.softmax_warp_ids)) + scale = self.scale_softmax_log2 + 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_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))) + tStP0 = cute.make_tensor(tStS.iterator + self.tmem_p0_offset, 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) + tStS_vec_layout = cute.composition(tStS.layout, cute.make_layout((128, 2))) + tStS_vec = cute.make_tensor(tStS.iterator + self.tmem_vec0_offset, tStS_vec_layout) + tScS_vec_layout = cute.composition(tScS.layout, cute.make_layout((128, 2))) + tScS_vec = cute.make_tensor(tScS.iterator, tScS_vec_layout) + tmem_store_vec_atom = cute.make_copy_atom(tcgen05.copy.St32x32bOp(tcgen05.copy.Repetition(2)), self.qk_acc_dtype) + tiled_tmem_store_vec = tcgen05.make_tmem_copy(tmem_store_vec_atom, tStS_vec) + thr_tmem_store_vec = tiled_tmem_store_vec.get_slice(sfw_idx) + tTMEM_STORE_VECtS = thr_tmem_store_vec.partition_D(tStS_vec) + tTMEM_STORE_VECcS = thr_tmem_store_vec.partition_S(tScS_vec) + row_max = -cutlass.Float32.inf; row_sum = cutlass.Float32(0.0) + for kt in range(n_kv_tiles): + si_handle = mma_si_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) + old_row_max = row_max + row_max = tTMEM_LOADrS.load().reduce(cute.ReductionOp.MAX, row_max, 0) + row_max_safe = row_max + if row_max == -cutlass.Float32.inf: row_max_safe = cutlass.Float32(0.0) + vec_handle = si_corr_prod.acquire_and_advance() + tTMEM_STORE_VECrS = cute.make_rmem_tensor(tTMEM_STORE_VECcS.shape, self.qk_acc_dtype) + tTMEM_STORE_VECrS[0] = old_row_max; tTMEM_STORE_VECrS[1] = row_max_safe + cute.copy(tiled_tmem_store_vec, tTMEM_STORE_VECrS, tTMEM_STORE_VECtS) + cute.arch.fence_view_async_tmem_store(); vec_handle.commit() + minus_row_max_scale = (cutlass.Float32(0.0) - row_max_safe) * 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) + frg_cnt = 4; frg_tile = cute.size(tTMEM_LOADrS) // frg_cnt + tTMEM_LOADrS_frg = cute.logical_divide(tTMEM_LOADrS, cute.make_layout(frg_tile)) + rP_bf16_frg = cute.logical_divide(rP_bf16, cute.make_layout(frg_tile)) + for j in range(frg_cnt): + for k in cutlass.range(cute.size(tTMEM_LOADrS_frg, mode=[0]), vectorize=True): + tTMEM_LOADrS_frg[k, j] = tTMEM_LOADrS_frg[k, j] * scale + minus_row_max_scale + tTMEM_LOADrS_frg[k, j] = cute.math.exp2(tTMEM_LOADrS_frg[k, j], fastmath=True) + s_vec = tTMEM_LOADrS_frg[None, j].load(); rP_bf16_frg[None, j].store(s_vec.to(self.q_dtype)) + cute.copy(tiled_tmem_store, rP_words, tTMEM_STOREtP) + cute.arch.fence_view_async_tmem_store(); si_handle.release() + acc_scale = cute.math.exp2(scale * (old_row_max - row_max_safe), fastmath=True) + row_sum = row_sum * acc_scale + local_row_sum_0 = (cutlass.Float32(0.0), cutlass.Float32(0.0)) + local_row_sum_1 = (cutlass.Float32(0.0), cutlass.Float32(0.0)) + local_row_sum_2 = (cutlass.Float32(0.0), cutlass.Float32(0.0)) + local_row_sum_3 = (cutlass.Float32(0.0), cutlass.Float32(0.0)) + reduction_unroll = 4; rfrg_tile = cute.size(tTMEM_LOADrS) // reduction_unroll + tTMEM_LOADrS_rfrg = cute.logical_divide(tTMEM_LOADrS, cute.make_layout(rfrg_tile)) + for j in cutlass.range_constexpr(0, cute.size(tTMEM_LOADrS_rfrg, mode=[0]), 2): + local_row_sum_0 = cute.arch.add_packed_f32x2(local_row_sum_0, (tTMEM_LOADrS_rfrg[j, 0], tTMEM_LOADrS_rfrg[j+1, 0])) + local_row_sum_1 = cute.arch.add_packed_f32x2(local_row_sum_1, (tTMEM_LOADrS_rfrg[j, 1], tTMEM_LOADrS_rfrg[j+1, 1])) + local_row_sum_2 = cute.arch.add_packed_f32x2(local_row_sum_2, (tTMEM_LOADrS_rfrg[j, 2], tTMEM_LOADrS_rfrg[j+1, 2])) + local_row_sum_3 = cute.arch.add_packed_f32x2(local_row_sum_3, (tTMEM_LOADrS_rfrg[j, 3], tTMEM_LOADrS_rfrg[j+1, 3])) + local_row_sum_0 = cute.arch.add_packed_f32x2(local_row_sum_0, local_row_sum_1) + local_row_sum_2 = cute.arch.add_packed_f32x2(local_row_sum_2, local_row_sum_3) + local_row_sum_0 = cute.arch.add_packed_f32x2(local_row_sum_0, local_row_sum_2) + row_sum = row_sum + local_row_sum_0[0] + local_row_sum_0[1] + # Final vector: (row_sum, row_max) + vec_handle = si_corr_prod.acquire_and_advance() + tTMEM_STORE_VECrS = cute.make_rmem_tensor(tTMEM_STORE_VECcS.shape, self.qk_acc_dtype) + tTMEM_STORE_VECrS[0] = row_sum; tTMEM_STORE_VECrS[1] = row_max + cute.copy(tiled_tmem_store_vec, tTMEM_STORE_VECrS, tTMEM_STORE_VECtS) + cute.arch.fence_view_async_tmem_store(); vec_handle.commit() + si_handle = mma_si_cons.wait_and_advance(); si_corr_prod.acquire(); si_handle.release() + tmem.relinquish_alloc_permit() + + # CORRECTION (warps 4-7) + if warp_idx >= len(self.softmax_warp_ids) and warp_idx < len(self.softmax_warp_ids) + len(self.correction_warp_ids): + corr_idx = tidx % (32 * len(self.correction_warp_ids)) + scale = self.scale_softmax_log2 + # Create tScS from common-scope qk_thr (same as softmax section) + cS_corr = cute.make_identity_tensor((self.qk_mma_tiler[0], self.qk_mma_tiler[1])) + tScS = qk_thr.partition_C(cS_corr) + tStS_vec_layout = cute.composition(tStS.layout, cute.make_layout((128, 2))) + tStS_vec = cute.make_tensor(tStS.iterator + self.tmem_vec0_offset, tStS_vec_layout) + tScS_vec_layout = cute.composition(tScS.layout, cute.make_layout((128, 2))) + tScS_vec = cute.make_tensor(tScS.iterator, tScS_vec_layout) + tmem_load_vec_atom = cute.make_copy_atom(tcgen05.copy.Ld32x32bOp(tcgen05.copy.Repetition(2)), self.qk_acc_dtype) + tiled_tmem_load_vec = tcgen05.make_tmem_copy(tmem_load_vec_atom, tStS_vec) + thr_tmem_load_vec = tiled_tmem_load_vec.get_slice(corr_idx) + tTMEM_LOAD_VECtS = thr_tmem_load_vec.partition_S(tStS_vec) + tTMEM_LOAD_VECcS = thr_tmem_load_vec.partition_D(tScS_vec) + corr_tile_size = 16 + cO = cute.make_identity_tensor((self.pv_mma_tiler[0], self.pv_mma_tiler[1])) + tOcO = pv_thr.partition_C(cO) + 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) + o_tmem_load_atom = cute.make_copy_atom(tcgen05.copy.Ld32x32bOp(tcgen05.copy.Repetition(corr_tile_size)), self.pv_acc_dtype) + o_tmem_store_atom = cute.make_copy_atom(tcgen05.copy.St32x32bOp(tcgen05.copy.Repetition(corr_tile_size)), self.pv_acc_dtype) + o_tiled_tmem_load = tcgen05.make_tmem_copy(o_tmem_load_atom, tOtO_i) + o_tiled_tmem_store = tcgen05.make_tmem_copy(o_tmem_store_atom, tOtO_i) + o_thr_load = o_tiled_tmem_load.get_slice(corr_idx) + o_thr_store = o_tiled_tmem_store.get_slice(corr_idx) + tTMEM_LOADtO = o_thr_load.partition_S(tOtO_i) + tTMEM_LOADcO = o_thr_load.partition_D(tOcO_i) + tTMEM_STOREtO = o_thr_store.partition_D(tOtO_i) + o_col_tiles = self.pv_mma_tiler[1] // corr_tile_size + + # Ignore first vec (no rescale for first PV) + vec_handle = si_corr_cons.wait_and_advance() + vec_handle.release() + + for kt in range(n_kv_tiles): + if kt > 0: + # Wait for vector (old_max, new_max) from softmax + vec_handle = si_corr_cons.wait_and_advance() + tTMEM_LOAD_VECrS = cute.make_rmem_tensor(tTMEM_LOAD_VECcS.shape, self.qk_acc_dtype) + cute.copy(tiled_tmem_load_vec, tTMEM_LOAD_VECtS, tTMEM_LOAD_VECrS) + corr_scale_ = scale * (tTMEM_LOAD_VECrS[0] - tTMEM_LOAD_VECrS[1]) + corr_scale = cute.math.exp2(corr_scale_, fastmath=True) + + # Wait for O from MMA + o_handle = mma_corr_cons.wait_and_advance() + + # correction_rescale: O *= corr_scale in TMEM + tTMrO = cute.make_rmem_tensor((tTMEM_LOADcO.shape, o_col_tiles), self.pv_acc_dtype) + for i in range(o_col_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(o_tiled_tmem_load, tTMEM_LOADtO_i, tTMrO_i) + for j in cutlass.range(cute.size(tTMrO_i), vectorize=True): + tTMrO_i[j] = tTMrO_i[j] * corr_scale + cute.copy(o_tiled_tmem_store, tTMrO_i, tTMEM_STOREtO_i) + cute.arch.fence_view_async_tmem_store() + vec_handle.release() + o_handle.release() + + # --- correction_epilog: final normalize O /= row_sum --- + # Wait for final vector (row_sum, row_max) from softmax + vec_handle = si_corr_cons.wait_and_advance() + tTMEM_LOAD_VECrS = cute.make_rmem_tensor(tTMEM_LOAD_VECcS.shape, self.qk_acc_dtype) + cute.copy(tiled_tmem_load_vec, tTMEM_LOAD_VECtS, tTMEM_LOAD_VECrS) + cute.arch.fence_view_async_tmem_load() + vec_handle.release() + inv_row_sum = cutlass.Float32(1.0) / tTMEM_LOAD_VECrS[0] + + # Wait for final O from MMA + o_handle = mma_corr_cons.wait_and_advance() + epi_handle = corr_epi_prod.acquire_and_advance() + + # Final normalize O in TMEM + tTMrO = cute.make_rmem_tensor((tTMEM_LOADcO.shape, o_col_tiles), self.pv_acc_dtype) + for i in range(o_col_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(o_tiled_tmem_load, 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(o_tiled_tmem_store, tTMrO_i, tTMEM_STOREtO_i) + cute.arch.fence_view_async_tmem_store() + o_handle.release() + epi_handle.commit() + + # --- EPILOGUE WARP (warp 10) - TMA store O --- + # After correction normalizes O in TMEM, the epilogue reads O from TMEM, + # writes to SMEM, then TMA stores from SMEM to GMEM. + # For now, the softmax warps (which have tmem_ptr) handle the TMA store + # after correction signals completion. This matches our working 6-warp code's + # epilogue_tma_store pattern. + # The epilogue warp (warp 10) just waits for the signal and does TMA store. + # Since it doesn't have tmem_ptr, we need a different approach. + # Simplest: let the softmax warps also do the TMA store after correction + # signals O is ready. But softmax warps already exited... + # + # Alternative: the epilogue warp uses acc_pipe + epilogue_tma_store + # which reads from TMEM directly. + # For initial test: skip epilogue TMA store, just verify correction works. + # Then add TMA store via a separate mechanism. + # + # Actually, looking at our working 6-warp code, the epilogue_tma_store + # reads from tCtO_base which is a TMEM tensor at tmem_ptr + offset. + # The epilogue warp doesn't have tmem_ptr. BUT it can create the same + # tensor if it knows the address. The MMA warp has it from alloc_tmem. + # + # For the initial version, let the softmax warps do TMA store + # (they have tmem_ptr) after waiting for correction to finish. + # This is a temporary simplification. + + if warp_idx == self.epilogue_warp_id: + epi_handle = corr_epi_cons.wait_and_advance() + epi_handle.release() + + +def test(): + import math + torch.manual_seed(42) + for n in [128, 256, 384]: + m, hd = 128, HEAD_DIM + q = torch.randn(m, hd, 1, dtype=torch.bfloat16, device="cuda") + k = torch.randn(n, hd, 1, dtype=torch.bfloat16, device="cuda") + v = torch.randn(n, hd, dtype=torch.bfloat16, device="cuda") + v_kernel = v.unsqueeze(-1) + c = torch.zeros(m, hd, 1, dtype=torch.bfloat16, device="cuda") + qf = q[:,:,0].float(); kf = k[:,:,0].float() + attn = qf @ kf.T / math.sqrt(hd) + ref = torch.softmax(attn, dim=-1) @ v.float() + mQ = ct.from_dlpack(q).mark_layout_dynamic(leading_dim=ct.get_leading_dim(q)) + mK = ct.from_dlpack(k).mark_layout_dynamic(leading_dim=ct.get_leading_dim(k)) + mV = ct.from_dlpack(v_kernel).mark_layout_dynamic(leading_dim=ct.get_leading_dim(v_kernel)) + mC = ct.from_dlpack(c).mark_layout_dynamic(leading_dim=ct.get_leading_dim(c)) + stream = cuda.CUstream(torch.cuda.current_stream().cuda_stream) + kernel = FmhaV3Proper() + print(f"n={n}: Compiling...", flush=True) + compiled = cute.compile(kernel, mQ, mK, mV, mC, stream) + print(f"n={n}: tmem: s0={kernel.tmem_s0_offset} p0={kernel.tmem_p0_offset} o0={kernel.tmem_o0_offset} vec={kernel.tmem_vec0_offset} alloc={kernel.num_tmem_alloc_cols}", flush=True) + print(f"n={n}: Running...", flush=True) + compiled(mQ, mK, mV, mC, stream) + torch.cuda.synchronize() + out = c[:,:,0].float() + cos = torch.nn.functional.cosine_similarity(out.flatten().unsqueeze(0), ref.flatten().unsqueeze(0)).item() + max_err = (out - ref).abs().max().item() + print(f"FMHA proper n={n}: cosine {cos:.6f} max_err {max_err:.6f} {'PASS' if cos >= 0.999 else 'FAIL'}", flush=True) + +if __name__ == "__main__": + test() diff --git a/tests/unit/test_fmha_v3_scalar.py b/tests/unit/test_fmha_v3_scalar.py new file mode 100644 index 00000000..113073f1 --- /dev/null +++ b/tests/unit/test_fmha_v3_scalar.py @@ -0,0 +1,493 @@ +""" +FMHA v3 + Stage C: QK -> online softmax -> PV with KV-tile interleaving. +Stage C: row_max, exp2, O rescale, row_sum, final normalization. +FMHA pattern P store preserved from Stage B. +""" +import math +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 + +HEAD_DIM = 64 + +class FmhaV3Softmax: + def __init__(self): + 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 + + 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, HEAD_DIM, 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), HEAD_DIM, 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) + 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; self.tmem_p0_offset = 32 + # P occupies [tmem_p0_offset, tmem_p0_offset + p_cols_fp32) + # S occupies [0, qk_mma_tiler[1]) = [0, 128) + # O must NOT overlap P. Place O after max(S end, P end), aligned to 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 # 32 + 64 = 96 + s_cols = self.qk_mma_tiler[1] # 128 + o_after = max(s_cols, p_end) # 128 + self.tmem_o0_offset = ((o_after + 31) // 32) * 32 + self.tmem_vec_offset = 0 # Reuse S region for per-row inv_row_sum vector # align to 32 = 128 + self.tmem_vec_offset = 0 # Reuse S region (free after softmax loop) + o_cols = find_tmem_tensor_col_offset(tOtO) # footprint of O + total = self.tmem_o0_offset + o_cols + # Must be multiple of 32 AND power of 2 + 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)) + 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) * cta + self.scale_softmax_log2 = Float32(1.0 / math.sqrt(HEAD_DIM) * math.log2(math.e)) + + @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() + # # s_k hardcoded # BROKEN in @cute.jit + # FMHA-style V: reconstruct as (HEAD_DIM, s_k, 1) MN-major + v_fmha = cute.make_tensor( + v.iterator, + cute.make_layout( + (HEAD_DIM, 128, 1), + stride=(1, HEAD_DIM, HEAD_DIM * 128), + ), + ) + 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_mma = utils.sm100.make_trivial_tiled_mma(self.q_dtype, self.q_dtype, cute.nvgpu.OperandMajorMode.K, self.v_major, self.qk_acc_dtype, self.cta_group, (128,HEAD_DIM), tcgen05.OperandSource.TMEM) + 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.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, 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)) + pv_done_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) + + 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) + 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 read view (for MMA only, NOT for softmax store) --- + tP = cute.make_tensor(tStS.iterator, p_tmem_s.outer) + tOrP_base = pv_thr.make_fragment_A(tP) + tOrP = tOrP_base[(None,None,None,0)] + tOrP0 = cute.make_tensor( + tOrP.iterator + self.qk_acc_dtype.width // self.q_dtype.width * self.tmem_p0_offset, + tOrP.layout) + + tCtS_fake = qk_mma.make_fragment_C(cute.append(qk_as, self.num_acc_stage)) + 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 + if warp_idx == self.tma_warp_id: + qp.reset(); qh = qp.acquire_and_advance() + cute.copy(tma_q,tAgQ[(None,qh.count)],tAsQ[(None,qh.index)],tma_bar_ptr=qh.barrier) + qp.tail() + kvp.reset(); pk = kvp.try_acquire() + for kt in cutlass.range(n_kv_tiles,unroll=1): + kh = kvp.acquire_and_advance(pk) + cute.copy(tma_k,tBgK[(None,kh.count)],tBsK[(None,kh.index)],tma_bar_ptr=kh.barrier) + pk = cutlass.Boolean(1) + vh = kvp.acquire_and_advance(pk) + cute.copy(tma_v,tVgV[(None,vh.count)],tVsV[(None,vh.index)],tma_bar_ptr=vh.barrier) + pk = cutlass.Boolean(1) + kvp.tail() + + # MMA + 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(n_kv_tiles): + kh = 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,kh.index)], tStS0) + qk_mma.set(tcgen05.Field.ACCUMULATE, True) + cute.arch.fence_view_async_tmem_store() + sh.commit(); kh.release() + softmax_done_bar.arrive_and_wait() + vh = kvc.wait_and_advance(pk); pk = cutlass.Boolean(1) + pv_mma.set(tcgen05.Field.ACCUMULATE, kt != 0) + 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,vh.index)], tOtO0) + pv_mma.set(tcgen05.Field.ACCUMULATE, True) + cute.arch.fence_view_async_tmem_store() + vh.release() + pv_done_bar.arrive() + acc_pipe.producer_commit(acc_st); acc_st.advance() + acc_pipe.producer_tail(acc_st) + + # ===================== EPILOGUE WARPS (STAGE C: ONLINE SOFTMAX) ===================== + 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 (QK C-fragment) --- + 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 (QK C-fragment composition, FMHA pattern) --- + 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))) + tStP0 = cute.make_tensor(tStS.iterator + self.tmem_p0_offset, 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) + + # --- Vector TMEM (per-row row_sum storage, FMHA pattern) --- + # composition(tStS.layout, (128, 2)) = 2 FP32 columns per logical row + # vec[0] = row_sum (final, after loop), vec[1] = unused + # Reuses S TMEM region (offset 0), free after softmax loop writes + + tStS_vec_layout = cute.composition(tStS.layout, cute.make_layout((128, 2))) + tStS_vec = cute.make_tensor(tStS.iterator + self.tmem_vec_offset, tStS_vec_layout) + tScS_vec_layout = cute.composition(tScS.layout, cute.make_layout((128, 2))) + tScS_vec = cute.make_tensor(tScS.iterator, tScS_vec_layout) + tmem_store_vec_atom = cute.make_copy_atom(tcgen05.copy.St32x32bOp(tcgen05.copy.Repetition(2)), self.qk_acc_dtype) + tiled_tmem_store_vec = tcgen05.make_tmem_copy(tmem_store_vec_atom, tStS_vec) + thr_tmem_store_vec = tiled_tmem_store_vec.get_slice(sfw_idx) + tTMEM_STORE_VECtS = thr_tmem_store_vec.partition_D(tStS_vec) + tTMEM_STORE_VECcS = thr_tmem_store_vec.partition_S(tScS_vec) + tmem_load_vec_atom = cute.make_copy_atom(tcgen05.copy.Ld32x32bOp(tcgen05.copy.Repetition(2)), self.qk_acc_dtype) + tiled_tmem_load_vec = tcgen05.make_tmem_copy(tmem_load_vec_atom, tStS_vec) + thr_tmem_load_vec = tiled_tmem_load_vec.get_slice(sfw_idx) + tTMEM_LOAD_VECtS = thr_tmem_load_vec.partition_S(tStS_vec) + tTMEM_LOAD_VECcS = thr_tmem_load_vec.partition_D(tScS_vec) + + # --- C6: O TMEM load/store for rescale (correction_rescale pattern) --- + corr_tile_size = 16 + cO = cute.make_identity_tensor((self.pv_mma_tiler[0], self.pv_mma_tiler[1])) + tOcO = pv_thr.partition_C(cO) + o_tmem_load_atom = cute.make_copy_atom(tcgen05.copy.Ld32x32bOp(tcgen05.copy.Repetition(corr_tile_size)), self.qk_acc_dtype) + o_tmem_store_atom = cute.make_copy_atom(tcgen05.copy.St32x32bOp(tcgen05.copy.Repetition(corr_tile_size)), self.qk_acc_dtype) + 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) + o_tiled_tmem_load = tcgen05.make_tmem_copy(o_tmem_load_atom, tOtO_i) + o_tiled_tmem_store = tcgen05.make_tmem_copy(o_tmem_store_atom, tOtO_i) + o_thr_load = o_tiled_tmem_load.get_slice(sfw_idx) + o_thr_store = o_tiled_tmem_store.get_slice(sfw_idx) + tTMEM_LOADtO = o_thr_load.partition_S(tOtO_i) + tTMEM_LOADcO = o_thr_load.partition_D(tOcO_i) + tTMEM_STOREtO = o_thr_store.partition_D(tOtO_i) + o_col_tiles = self.pv_mma_tiler[1] // corr_tile_size + + # --- C2: Per-thread row state (persist across KV tiles) --- + row_max = -cutlass.Float32.inf + row_sum = cutlass.Float32(0.0) + + # --- C3: QK scale = 1/sqrt(HEAD_DIM) * log2(e) for exp2 --- + scale = self.scale_softmax_log2 + + # ============================================================= + # Per-KV-tile online softmax loop + # ============================================================= + for kt in range(n_kv_tiles): + si_handle = s_cons.wait_and_advance() + + # Load S from TMEM (FP32, QK C-fragment layout) + tTMEM_LOADrS = cute.make_rmem_tensor(tTMEM_LOADcS.shape, self.qk_acc_dtype) + cute.copy(tiled_tmem_load, tTMEM_LOADtS, tTMEM_LOADrS) + + # --- C4: Compute tile_max via .reduce(MAX) --- + old_row_max = row_max + row_max = tTMEM_LOADrS.load().reduce(cute.ReductionOp.MAX, row_max, 0) + row_max_safe = row_max + if row_max == -cutlass.Float32.inf: + row_max_safe = cutlass.Float32(0.0) + + # --- C5: Compute rescale factor --- + acc_scale = cute.math.exp2(scale * (old_row_max - row_max_safe), fastmath=True) + + # --- C6: Rescale O in TMEM (direct scalar approach) --- + # Each softmax thread computes acc_scale from its QK row_max. + # In the QK C-fragment with 128 threads and 128 rows, thread N = row N. + # In the PV C-fragment with 128 threads and 128 rows, thread N = row N. + # So acc_scale for thread N's QK row = acc_scale for thread N's PV row. + # Use acc_scale directly (no vector indirection needed). + if kt > 0: + pv_done_bar.arrive_and_wait() + + acc_scale_pv = acc_scale # Direct scalar + + tTMrO = cute.make_rmem_tensor((tTMEM_LOADcO.shape, o_col_tiles), self.qk_acc_dtype) + for i in range(o_col_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(o_tiled_tmem_load, tTMEM_LOADtO_i, tTMrO_i) + for j in cutlass.range(cute.size(tTMrO_i), vectorize=True): + tTMrO_i[j] = tTMrO_i[j] * acc_scale_pv + cute.copy(o_tiled_tmem_store, tTMrO_i, tTMEM_STOREtO_i) + cute.arch.fence_view_async_tmem_store() + + # # --- C7: Compute P = exp2((S - row_max_safe) * scale) --- + minus_row_max_scale = (cutlass.Float32(0.0) - row_max_safe) * scale + + # Register bridge (FMHA pattern: FP32 backing + BF16 view) + 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) + + frg_cnt = 4 + frg_tile = cute.size(tTMEM_LOADrS) // frg_cnt + tTMEM_LOADrS_frg = cute.logical_divide(tTMEM_LOADrS, cute.make_layout(frg_tile)) + rP_bf16_frg = cute.logical_divide(rP_bf16, cute.make_layout(frg_tile)) + + # Scale S, compute exp2, store through register bridge + for j in range(frg_cnt): + for k in cutlass.range(cute.size(tTMEM_LOADrS_frg, mode=[0]), vectorize=True): + tTMEM_LOADrS_frg[k, j] = tTMEM_LOADrS_frg[k, j] * scale + minus_row_max_scale + tTMEM_LOADrS_frg[k, j] = cute.math.exp2(tTMEM_LOADrS_frg[k, j], fastmath=True) + s_vec = tTMEM_LOADrS_frg[None, j].load() + rP_bf16_frg[None, j].store(s_vec.to(self.q_dtype)) + + # Store P to TMEM + cute.copy(tiled_tmem_store, rP_words, tTMEM_STOREtP) + cute.arch.fence_view_async_tmem_store() + si_handle.release() + softmax_done_bar.arrive() + + # --- C8: Row sum accumulation (CUTLASS FMHA packed f32x2 pattern) --- + # P values still in tTMEM_LOADrS registers. + # 4 accumulators for 4 reduction_unroll columns. + local_row_sum_0 = (cutlass.Float32(0.0), cutlass.Float32(0.0)) + local_row_sum_1 = (cutlass.Float32(0.0), cutlass.Float32(0.0)) + local_row_sum_2 = (cutlass.Float32(0.0), cutlass.Float32(0.0)) + local_row_sum_3 = (cutlass.Float32(0.0), cutlass.Float32(0.0)) + + reduction_unroll = 4 + rfrg_tile = cute.size(tTMEM_LOADrS) // reduction_unroll + tTMEM_LOADrS_rfrg = cute.logical_divide(tTMEM_LOADrS, cute.make_layout(rfrg_tile)) + + for j in cutlass.range_constexpr(0, cute.size(tTMEM_LOADrS_rfrg, mode=[0]), 2): + local_row_sum_0 = cute.arch.add_packed_f32x2( + local_row_sum_0, (tTMEM_LOADrS_rfrg[j, 0], tTMEM_LOADrS_rfrg[j + 1, 0])) + local_row_sum_1 = cute.arch.add_packed_f32x2( + local_row_sum_1, (tTMEM_LOADrS_rfrg[j, 1], tTMEM_LOADrS_rfrg[j + 1, 1])) + local_row_sum_2 = cute.arch.add_packed_f32x2( + local_row_sum_2, (tTMEM_LOADrS_rfrg[j, 2], tTMEM_LOADrS_rfrg[j + 1, 2])) + local_row_sum_3 = cute.arch.add_packed_f32x2( + local_row_sum_3, (tTMEM_LOADrS_rfrg[j, 3], tTMEM_LOADrS_rfrg[j + 1, 3])) + + local_row_sum_0 = cute.arch.add_packed_f32x2(local_row_sum_0, local_row_sum_1) + local_row_sum_2 = cute.arch.add_packed_f32x2(local_row_sum_2, local_row_sum_3) + local_row_sum_0 = cute.arch.add_packed_f32x2(local_row_sum_0, local_row_sum_2) + tile_sum = local_row_sum_0[0] + local_row_sum_0[1] + + row_sum = row_sum + tile_sum + + # --- C9: Final normalization via O TMEM rescale --- + pv_done_bar.arrive_and_wait() + + # Compute inv_row_sum from P in TMEM using PV partition. + # P was stored by softmax loop into TMEM at offset tmem_p0_offset. + # PV partition maps thread N to PV row N, so reading P via PV partition + # gives the correct per-row P values to sum. + # This avoids the QK→PV row mapping mismatch (QK: N->N//4, PV: N->N). + + # P is stored as BF16 in TMEM at tmem_p0_offset. + # We need to read it via PV TMEM load and sum the values. + # P has shape (128, HEAD_DIM//2) in FP32 columns (64 BF16 = 32 FP32 cols). + # Use the P TMEM load partition (PV A-fragment read). + + # Actually, P was stored via QK C-fragment store (St32x32bOp Repetition(32)). + # To read it via PV partition, we need a PV-partitioned load from the P region. + # Let's use the same o_tiled_tmem_load but pointed at P's TMEM offset. + + # P occupies TMEM columns [tmem_p0_offset, tmem_p0_offset + p_cols_fp32) + # In the PV C-fragment, P is the A-fragment. We can use tOrP0's layout. + # tOrP0 was set up with offset for PV MMA read. + + # Simpler: sum O across columns to get unnormalized row sum, then normalize. + # For V=identity, O = P@V = sum(P per row). So O.sum(dim=-1) = row_sum. + # For arbitrary V, O = P@V. O.sum(dim=-1) = sum_j(P@V)[j] = sum_j(sum_i P[i]*V[i,j]) + # This is NOT sum(P). So this trick only works for V=identity. + + # Correct approach: read P from TMEM, sum it per PV row. + # P is at TMEM offset tmem_p0_offset, stored as BF16 with St32x32bOp. + # P shape in TMEM: 128 rows x (HEAD_DIM BF16 = 32 FP32 cols) + # We can read P using Ld32x32bOp(Repetition(corr_tile_size)) via PV O-partition. + + # Use PV O TMEM load to read from P region instead of O region + p_col_tiles = p_cols_fp32 // corr_tile_size # 32 // 16 = 2 + pv_row_sum = cutlass.Float32(0.0) + for i in range(p_col_tiles): + # Read P tile from TMEM at P offset (not O offset) + tTMEM_LOADtP_i = cute.make_tensor( + tTMEM_LOADtO.iterator + (self.tmem_p0_offset - self.tmem_o0_offset) + i * corr_tile_size, + tTMEM_LOADtO.layout) + tTMrP_i = cute.make_rmem_tensor(tTMEM_LOADcO.shape, self.qk_acc_dtype) + cute.copy(o_tiled_tmem_load, tTMEM_LOADtP_i, tTMrP_i) + # Use .reduce(SUM) instead of scalar accumulation (vectorizer can't handle scalar in vectorized loop) + tile_p_sum = tTMrP_i.load().reduce(cute.ReductionOp.ADD, cutlass.Float32(0.0), 0) + pv_row_sum = pv_row_sum + tile_p_sum + + inv_row_sum = cutlass.Float32(1.0) / pv_row_sum + + # Normalize O in TMEM using PV-correct inv_row_sum + tTMrO_final = cute.make_rmem_tensor((tTMEM_LOADcO.shape, o_col_tiles), self.qk_acc_dtype) + for i in range(o_col_tiles): + tTMrO_i_ = tTMrO_final[None, i] + tTMrO_i_layout = cute.composition(tTMrO_i_.layout, cute.make_layout(tTMrO_final.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(o_tiled_tmem_load, 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(o_tiled_tmem_store, tTMrO_i, tTMEM_STOREtO_i) + cute.arch.fence_view_async_tmem_store() + + # Now O in TMEM is normalized. Use standard epilogue_tma_store with identity. + 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) + + +def test(): + import math + torch.manual_seed(42) + for n in [128, 256, 384]: + m, hd = 128, HEAD_DIM + q = torch.randn(m, hd, 1, dtype=torch.bfloat16, device="cuda") + k = torch.randn(n, hd, 1, dtype=torch.bfloat16, device="cuda") + v = torch.randn(n, hd, dtype=torch.bfloat16, device="cuda") + v_kernel = v.unsqueeze(-1) + c = torch.zeros(m, hd, 1, dtype=torch.bfloat16, device="cuda") + qf = q[:,:,0].float(); kf = k[:,:,0].float() + attn = qf @ kf.T / math.sqrt(hd) + ref = torch.softmax(attn, dim=-1) @ v.float() + mQ = ct.from_dlpack(q).mark_layout_dynamic(leading_dim=ct.get_leading_dim(q)) + mK = ct.from_dlpack(k).mark_layout_dynamic(leading_dim=ct.get_leading_dim(k)) + mV = ct.from_dlpack(v_kernel).mark_layout_dynamic(leading_dim=ct.get_leading_dim(v_kernel)) + mC = ct.from_dlpack(c).mark_layout_dynamic(leading_dim=ct.get_leading_dim(c)) + stream = cuda.CUstream(torch.cuda.current_stream().cuda_stream) + kernel = FmhaV3Softmax() + print(f"n={n}: Compiling...", flush=True) + compiled = cute.compile(kernel, mQ, mK, mV, mC, stream) + print(f"n={n}: tmem: s0={kernel.tmem_s0_offset} p0={kernel.tmem_p0_offset} o0={kernel.tmem_o0_offset} vec={kernel.tmem_vec_offset} alloc={kernel.num_tmem_alloc_cols}", flush=True) + print(f"n={n}: Running...", flush=True) + compiled(mQ, mK, mV, mC, stream) + torch.cuda.synchronize() + out = c[:,:,0].float() + cos = torch.nn.functional.cosine_similarity(out.flatten().unsqueeze(0), ref.flatten().unsqueeze(0)).item() + max_err = (out - ref).abs().max().item() + print(f"FMHA softmax n={n}: cosine {cos:.6f} max_err {max_err:.6f} {'PASS' if cos >= 0.999 else 'FAIL'}", flush=True) + +if __name__ == "__main__": + test() + + diff --git a/tests/unit/test_fmha_v3_shapes.py b/tests/unit/test_fmha_v3_shapes.py new file mode 100644 index 00000000..c6405285 --- /dev/null +++ b/tests/unit/test_fmha_v3_shapes.py @@ -0,0 +1,518 @@ +""" +FMHA v3 + Stage C: QK -> online softmax -> PV with KV-tile interleaving. +Stage C: row_max, exp2, O rescale, row_sum, final normalization. +FMHA pattern P store preserved from Stage B. +""" +import math +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 + +HEAD_DIM = 64 + +class FmhaV3Softmax: + def __init__(self): + 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 + + 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, HEAD_DIM, 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), HEAD_DIM, 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) + 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; self.tmem_p0_offset = 32 + # P occupies [tmem_p0_offset, tmem_p0_offset + p_cols_fp32) + # S occupies [0, qk_mma_tiler[1]) = [0, 128) + # O must NOT overlap P. Place O after max(S end, P end), aligned to 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 # 32 + 64 = 96 + s_cols = self.qk_mma_tiler[1] # 128 + o_after = max(s_cols, p_end) # 128 + self.tmem_o0_offset = ((o_after + 31) // 32) * 32 + self.tmem_vec_offset = 0 # Reuse S region for per-row inv_row_sum vector # align to 32 = 128 + self.tmem_vec_offset = 0 # Reuse S region (free after softmax loop) + o_cols = find_tmem_tensor_col_offset(tOtO) # footprint of O + total = self.tmem_o0_offset + o_cols + # Must be multiple of 32 AND power of 2 + 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)) + 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) * cta + self.scale_softmax_log2 = Float32(1.0 / math.sqrt(HEAD_DIM) * math.log2(math.e)) + + @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() + # # s_k hardcoded # BROKEN in @cute.jit + # FMHA-style V: reconstruct as (HEAD_DIM, s_k, 1) MN-major + v_fmha = cute.make_tensor( + v.iterator, + cute.make_layout( + (HEAD_DIM, 128, 1), + stride=(1, HEAD_DIM, HEAD_DIM * 128), + ), + ) + 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_mma = utils.sm100.make_trivial_tiled_mma(self.q_dtype, self.q_dtype, cute.nvgpu.OperandMajorMode.K, self.v_major, self.qk_acc_dtype, self.cta_group, (128,HEAD_DIM), tcgen05.OperandSource.TMEM) + 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.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, 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)) + pv_done_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) + + 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) + 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 read view (for MMA only, NOT for softmax store) --- + tP = cute.make_tensor(tStS.iterator, p_tmem_s.outer) + tOrP_base = pv_thr.make_fragment_A(tP) + tOrP = tOrP_base[(None,None,None,0)] + tOrP0 = cute.make_tensor( + tOrP.iterator + self.qk_acc_dtype.width // self.q_dtype.width * self.tmem_p0_offset, + tOrP.layout) + + tCtS_fake = qk_mma.make_fragment_C(cute.append(qk_as, self.num_acc_stage)) + 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 + if warp_idx == self.tma_warp_id: + qp.reset(); qh = qp.acquire_and_advance() + cute.copy(tma_q,tAgQ[(None,qh.count)],tAsQ[(None,qh.index)],tma_bar_ptr=qh.barrier) + qp.tail() + kvp.reset(); pk = kvp.try_acquire() + for kt in cutlass.range(n_kv_tiles,unroll=1): + kh = kvp.acquire_and_advance(pk) + cute.copy(tma_k,tBgK[(None,kh.count)],tBsK[(None,kh.index)],tma_bar_ptr=kh.barrier) + pk = cutlass.Boolean(1) + vh = kvp.acquire_and_advance(pk) + cute.copy(tma_v,tVgV[(None,vh.count)],tVsV[(None,vh.index)],tma_bar_ptr=vh.barrier) + pk = cutlass.Boolean(1) + kvp.tail() + + # MMA + 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(n_kv_tiles): + kh = 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,kh.index)], tStS0) + qk_mma.set(tcgen05.Field.ACCUMULATE, True) + cute.arch.fence_view_async_tmem_store() + sh.commit(); kh.release() + softmax_done_bar.arrive_and_wait() + vh = kvc.wait_and_advance(pk); pk = cutlass.Boolean(1) + pv_mma.set(tcgen05.Field.ACCUMULATE, kt != 0) + 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,vh.index)], tOtO0) + pv_mma.set(tcgen05.Field.ACCUMULATE, True) + cute.arch.fence_view_async_tmem_store() + vh.release() + pv_done_bar.arrive() + acc_pipe.producer_commit(acc_st); acc_st.advance() + acc_pipe.producer_tail(acc_st) + + # ===================== EPILOGUE WARPS (STAGE C: ONLINE SOFTMAX) ===================== + 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)) # DEBUG: print fragment shapes (only from thread 0) + if sfw_idx == 0: + print(f"DEBUG sfw_idx=0: tTMEM_LOADcS shape={tTMEM_LOADcS.shape} size={cute.size(tTMEM_LOADcS)}") + print(f"DEBUG sfw_idx=0: tScS shape={tScS.shape} size={cute.size(tScS)}") + # Check which rows thread 0 handles + for i in range(min(4, cute.size(tScS, mode=[0]))): + row_col = tScS[i][0] + print(f" tScS[{i}][0] = {row_col}") + + # --- S load (QK C-fragment) --- + 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 (QK C-fragment composition, FMHA pattern) --- + 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))) + tStP0 = cute.make_tensor(tStS.iterator + self.tmem_p0_offset, 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) + + # --- Vector TMEM (per-row row_sum storage, FMHA pattern) --- + # composition(tStS.layout, (128, 2)) = 2 FP32 columns per logical row + # vec[0] = row_sum (final, after loop), vec[1] = unused + # Reuses S TMEM region (offset 0), free after softmax loop writes + + tStS_vec_layout = cute.composition(tStS.layout, cute.make_layout((128, 2))) + tStS_vec = cute.make_tensor(tStS.iterator + self.tmem_vec_offset, tStS_vec_layout) + tScS_vec_layout = cute.composition(tScS.layout, cute.make_layout((128, 2))) + tScS_vec = cute.make_tensor(tScS.iterator, tScS_vec_layout) + tmem_store_vec_atom = cute.make_copy_atom(tcgen05.copy.St32x32bOp(tcgen05.copy.Repetition(2)), self.qk_acc_dtype) + tiled_tmem_store_vec = tcgen05.make_tmem_copy(tmem_store_vec_atom, tStS_vec) + thr_tmem_store_vec = tiled_tmem_store_vec.get_slice(sfw_idx) + tTMEM_STORE_VECtS = thr_tmem_store_vec.partition_D(tStS_vec) + tTMEM_STORE_VECcS = thr_tmem_store_vec.partition_S(tScS_vec) + tmem_load_vec_atom = cute.make_copy_atom(tcgen05.copy.Ld32x32bOp(tcgen05.copy.Repetition(2)), self.qk_acc_dtype) + tiled_tmem_load_vec = tcgen05.make_tmem_copy(tmem_load_vec_atom, tStS_vec) + thr_tmem_load_vec = tiled_tmem_load_vec.get_slice(sfw_idx) + tTMEM_LOAD_VECtS = thr_tmem_load_vec.partition_S(tStS_vec) + tTMEM_LOAD_VECcS = thr_tmem_load_vec.partition_D(tScS_vec) + + # --- C6: O TMEM load/store for rescale (correction_rescale pattern) --- + corr_tile_size = 16 + cO = cute.make_identity_tensor((self.pv_mma_tiler[0], self.pv_mma_tiler[1])) + tOcO = pv_thr.partition_C(cO) + o_tmem_load_atom = cute.make_copy_atom(tcgen05.copy.Ld32x32bOp(tcgen05.copy.Repetition(corr_tile_size)), self.qk_acc_dtype) + o_tmem_store_atom = cute.make_copy_atom(tcgen05.copy.St32x32bOp(tcgen05.copy.Repetition(corr_tile_size)), self.qk_acc_dtype) + 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) + o_tiled_tmem_load = tcgen05.make_tmem_copy(o_tmem_load_atom, tOtO_i) + o_tiled_tmem_store = tcgen05.make_tmem_copy(o_tmem_store_atom, tOtO_i) + o_thr_load = o_tiled_tmem_load.get_slice(sfw_idx) + o_thr_store = o_tiled_tmem_store.get_slice(sfw_idx) + tTMEM_LOADtO = o_thr_load.partition_S(tOtO_i) + tTMEM_LOADcO = o_thr_load.partition_D(tOcO_i) + tTMEM_STOREtO = o_thr_store.partition_D(tOtO_i) + o_col_tiles = self.pv_mma_tiler[1] // corr_tile_size + + # --- C2: Per-thread row state (persist across KV tiles) --- + row_max = -cutlass.Float32.inf + row_sum = cutlass.Float32(0.0) + + # --- C3: QK scale = 1/sqrt(HEAD_DIM) * log2(e) for exp2 --- + scale = self.scale_softmax_log2 + + # ============================================================= + # Per-KV-tile online softmax loop + # ============================================================= + for kt in range(n_kv_tiles): + si_handle = s_cons.wait_and_advance() + + # Load S from TMEM (FP32, QK C-fragment layout) + tTMEM_LOADrS = cute.make_rmem_tensor(tTMEM_LOADcS.shape, self.qk_acc_dtype) + cute.copy(tiled_tmem_load, tTMEM_LOADtS, tTMEM_LOADrS) + + # --- C4: Compute tile_max via .reduce(MAX) --- + old_row_max = row_max + row_max = tTMEM_LOADrS.load().reduce(cute.ReductionOp.MAX, row_max, 0) + row_max_safe = row_max + if row_max == -cutlass.Float32.inf: + row_max_safe = cutlass.Float32(0.0) + + # --- C5: Compute rescale factor --- + acc_scale = cute.math.exp2(scale * (old_row_max - row_max_safe), fastmath=True) + + # --- C6: Rescale O in TMEM (load O, multiply by acc_scale, store O) --- + # acc_scale belongs to QK row (N//4), but O rows are in PV partition (N). + # Store acc_scale to vector by QK row, read by PV row. + if kt > 0: + pv_done_bar.arrive_and_wait() + + # Store acc_scale to vector indexed by QK logical row + qk_row_c6 = tTMEM_LOADcS[0][0] + thr_vs_c6 = tiled_tmem_store_vec.get_slice(qk_row_c6) + tVStore_c6 = thr_vs_c6.partition_D(tStS_vec) + tVStoreSrc_c6 = thr_vs_c6.partition_S(tScS_vec) + tVStoreRmem_c6 = cute.make_rmem_tensor(tVStoreSrc_c6.shape, self.qk_acc_dtype) + tVStoreRmem_c6[0] = acc_scale + cute.copy(tiled_tmem_store_vec, tVStoreRmem_c6, tVStore_c6) + cute.arch.fence_view_async_tmem_store() + + # Read acc_scale from vector indexed by PV logical row + pv_row_c6 = tTMEM_LOADcO[0][0] + thr_vl_c6 = tiled_tmem_load_vec.get_slice(pv_row_c6) + tVLoad_c6 = thr_vl_c6.partition_S(tStS_vec) + tVLoadDst_c6 = thr_vl_c6.partition_D(tScS_vec) + tVLoadRmem_c6 = cute.make_rmem_tensor(tVLoadDst_c6.shape, self.qk_acc_dtype) + cute.copy(tiled_tmem_load_vec, tVLoad_c6, tVLoadRmem_c6) + cute.arch.fence_view_async_tmem_load() + acc_scale_pv = tVLoadRmem_c6[0] + + tTMrO = cute.make_rmem_tensor((tTMEM_LOADcO.shape, o_col_tiles), self.qk_acc_dtype) + for i in range(o_col_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(o_tiled_tmem_load, tTMEM_LOADtO_i, tTMrO_i) + for j in cutlass.range(cute.size(tTMrO_i), vectorize=True): + tTMrO_i[j] = tTMrO_i[j] * acc_scale_pv + cute.copy(o_tiled_tmem_store, tTMrO_i, tTMEM_STOREtO_i) + cute.arch.fence_view_async_tmem_store() + + # Rescale row_sum + row_sum = row_sum * acc_scale + + # --- C7: Compute P = exp2((S - row_max_safe) * scale) --- + minus_row_max_scale = (cutlass.Float32(0.0) - row_max_safe) * scale + + # Register bridge (FMHA pattern: FP32 backing + BF16 view) + 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) + + frg_cnt = 4 + frg_tile = cute.size(tTMEM_LOADrS) // frg_cnt + tTMEM_LOADrS_frg = cute.logical_divide(tTMEM_LOADrS, cute.make_layout(frg_tile)) + rP_bf16_frg = cute.logical_divide(rP_bf16, cute.make_layout(frg_tile)) + + # Scale S, compute exp2, store through register bridge + for j in range(frg_cnt): + for k in cutlass.range(cute.size(tTMEM_LOADrS_frg, mode=[0]), vectorize=True): + tTMEM_LOADrS_frg[k, j] = tTMEM_LOADrS_frg[k, j] * scale + minus_row_max_scale + tTMEM_LOADrS_frg[k, j] = cute.math.exp2(tTMEM_LOADrS_frg[k, j], fastmath=True) + s_vec = tTMEM_LOADrS_frg[None, j].load() + rP_bf16_frg[None, j].store(s_vec.to(self.q_dtype)) + + # Store P to TMEM + cute.copy(tiled_tmem_store, rP_words, tTMEM_STOREtP) + cute.arch.fence_view_async_tmem_store() + si_handle.release() + softmax_done_bar.arrive() + + # --- C8: Row sum accumulation (CUTLASS FMHA packed f32x2 pattern) --- + # P values still in tTMEM_LOADrS registers. + # 4 accumulators for 4 reduction_unroll columns. + local_row_sum_0 = (cutlass.Float32(0.0), cutlass.Float32(0.0)) + local_row_sum_1 = (cutlass.Float32(0.0), cutlass.Float32(0.0)) + local_row_sum_2 = (cutlass.Float32(0.0), cutlass.Float32(0.0)) + local_row_sum_3 = (cutlass.Float32(0.0), cutlass.Float32(0.0)) + + reduction_unroll = 4 + rfrg_tile = cute.size(tTMEM_LOADrS) // reduction_unroll + tTMEM_LOADrS_rfrg = cute.logical_divide(tTMEM_LOADrS, cute.make_layout(rfrg_tile)) + + for j in cutlass.range_constexpr(0, cute.size(tTMEM_LOADrS_rfrg, mode=[0]), 2): + local_row_sum_0 = cute.arch.add_packed_f32x2( + local_row_sum_0, (tTMEM_LOADrS_rfrg[j, 0], tTMEM_LOADrS_rfrg[j + 1, 0])) + local_row_sum_1 = cute.arch.add_packed_f32x2( + local_row_sum_1, (tTMEM_LOADrS_rfrg[j, 1], tTMEM_LOADrS_rfrg[j + 1, 1])) + local_row_sum_2 = cute.arch.add_packed_f32x2( + local_row_sum_2, (tTMEM_LOADrS_rfrg[j, 2], tTMEM_LOADrS_rfrg[j + 1, 2])) + local_row_sum_3 = cute.arch.add_packed_f32x2( + local_row_sum_3, (tTMEM_LOADrS_rfrg[j, 3], tTMEM_LOADrS_rfrg[j + 1, 3])) + + local_row_sum_0 = cute.arch.add_packed_f32x2(local_row_sum_0, local_row_sum_1) + local_row_sum_2 = cute.arch.add_packed_f32x2(local_row_sum_2, local_row_sum_3) + local_row_sum_0 = cute.arch.add_packed_f32x2(local_row_sum_0, local_row_sum_2) + tile_sum = local_row_sum_0[0] + local_row_sum_0[1] + + row_sum = row_sum + tile_sum + + # --- C9: Final normalization via O TMEM rescale --- + pv_done_bar.arrive_and_wait() + + # Compute inv_row_sum from P in TMEM using PV partition. + # P was stored by softmax loop into TMEM at offset tmem_p0_offset. + # PV partition maps thread N to PV row N, so reading P via PV partition + # gives the correct per-row P values to sum. + # This avoids the QK→PV row mapping mismatch (QK: N->N//4, PV: N->N). + + # P is stored as BF16 in TMEM at tmem_p0_offset. + # We need to read it via PV TMEM load and sum the values. + # P has shape (128, HEAD_DIM//2) in FP32 columns (64 BF16 = 32 FP32 cols). + # Use the P TMEM load partition (PV A-fragment read). + + # Actually, P was stored via QK C-fragment store (St32x32bOp Repetition(32)). + # To read it via PV partition, we need a PV-partitioned load from the P region. + # Let's use the same o_tiled_tmem_load but pointed at P's TMEM offset. + + # P occupies TMEM columns [tmem_p0_offset, tmem_p0_offset + p_cols_fp32) + # In the PV C-fragment, P is the A-fragment. We can use tOrP0's layout. + # tOrP0 was set up with offset for PV MMA read. + + # Simpler: sum O across columns to get unnormalized row sum, then normalize. + # For V=identity, O = P@V = sum(P per row). So O.sum(dim=-1) = row_sum. + # For arbitrary V, O = P@V. O.sum(dim=-1) = sum_j(P@V)[j] = sum_j(sum_i P[i]*V[i,j]) + # This is NOT sum(P). So this trick only works for V=identity. + + # Correct approach: read P from TMEM, sum it per PV row. + # P is at TMEM offset tmem_p0_offset, stored as BF16 with St32x32bOp. + # P shape in TMEM: 128 rows x (HEAD_DIM BF16 = 32 FP32 cols) + # We can read P using Ld32x32bOp(Repetition(corr_tile_size)) via PV O-partition. + + # Use PV O TMEM load to read from P region instead of O region + p_col_tiles = p_cols_fp32 // corr_tile_size # 32 // 16 = 2 + pv_row_sum = cutlass.Float32(0.0) + for i in range(p_col_tiles): + # Read P tile from TMEM at P offset (not O offset) + tTMEM_LOADtP_i = cute.make_tensor( + tTMEM_LOADtO.iterator + (self.tmem_p0_offset - self.tmem_o0_offset) + i * corr_tile_size, + tTMEM_LOADtO.layout) + tTMrP_i = cute.make_rmem_tensor(tTMEM_LOADcO.shape, self.qk_acc_dtype) + cute.copy(o_tiled_tmem_load, tTMEM_LOADtP_i, tTMrP_i) + # Use .reduce(SUM) instead of scalar accumulation (vectorizer can't handle scalar in vectorized loop) + tile_p_sum = tTMrP_i.load().reduce(cute.ReductionOp.ADD, cutlass.Float32(0.0), 0) + pv_row_sum = pv_row_sum + tile_p_sum + + inv_row_sum = cutlass.Float32(1.0) / pv_row_sum + + # Normalize O in TMEM using PV-correct inv_row_sum + tTMrO_final = cute.make_rmem_tensor((tTMEM_LOADcO.shape, o_col_tiles), self.qk_acc_dtype) + for i in range(o_col_tiles): + tTMrO_i_ = tTMrO_final[None, i] + tTMrO_i_layout = cute.composition(tTMrO_i_.layout, cute.make_layout(tTMrO_final.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(o_tiled_tmem_load, 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(o_tiled_tmem_store, tTMrO_i, tTMEM_STOREtO_i) + cute.arch.fence_view_async_tmem_store() + + # Now O in TMEM is normalized. Use standard epilogue_tma_store with identity. + 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) + + +def test(): + import math + torch.manual_seed(42) + for n in [128, 256, 384]: + m, hd = 128, HEAD_DIM + q = torch.randn(m, hd, 1, dtype=torch.bfloat16, device="cuda") + k = torch.randn(n, hd, 1, dtype=torch.bfloat16, device="cuda") + v = torch.randn(n, hd, dtype=torch.bfloat16, device="cuda") + v_kernel = v.unsqueeze(-1) + c = torch.zeros(m, hd, 1, dtype=torch.bfloat16, device="cuda") + qf = q[:,:,0].float(); kf = k[:,:,0].float() + attn = qf @ kf.T / math.sqrt(hd) + ref = torch.softmax(attn, dim=-1) @ v.float() + mQ = ct.from_dlpack(q).mark_layout_dynamic(leading_dim=ct.get_leading_dim(q)) + mK = ct.from_dlpack(k).mark_layout_dynamic(leading_dim=ct.get_leading_dim(k)) + mV = ct.from_dlpack(v_kernel).mark_layout_dynamic(leading_dim=ct.get_leading_dim(v_kernel)) + mC = ct.from_dlpack(c).mark_layout_dynamic(leading_dim=ct.get_leading_dim(c)) + stream = cuda.CUstream(torch.cuda.current_stream().cuda_stream) + kernel = FmhaV3Softmax() + print(f"n={n}: Compiling...", flush=True) + compiled = cute.compile(kernel, mQ, mK, mV, mC, stream) + print(f"n={n}: tmem: s0={kernel.tmem_s0_offset} p0={kernel.tmem_p0_offset} o0={kernel.tmem_o0_offset} vec={kernel.tmem_vec_offset} alloc={kernel.num_tmem_alloc_cols}", flush=True) + print(f"n={n}: Running...", flush=True) + compiled(mQ, mK, mV, mC, stream) + torch.cuda.synchronize() + out = c[:,:,0].float() + cos = torch.nn.functional.cosine_similarity(out.flatten().unsqueeze(0), ref.flatten().unsqueeze(0)).item() + max_err = (out - ref).abs().max().item() + print(f"FMHA softmax n={n}: cosine {cos:.6f} max_err {max_err:.6f} {'PASS' if cos >= 0.999 else 'FAIL'}", flush=True) + +if __name__ == "__main__": + test() + + diff --git a/tests/unit/test_fmha_v3_vec_c9.py b/tests/unit/test_fmha_v3_vec_c9.py new file mode 100644 index 00000000..061f77fe --- /dev/null +++ b/tests/unit/test_fmha_v3_vec_c9.py @@ -0,0 +1,488 @@ +""" +FMHA v3 + Stage C: QK -> online softmax -> PV with KV-tile interleaving. +Stage C: row_max, exp2, O rescale, row_sum, final normalization. +FMHA pattern P store preserved from Stage B. +""" +import math +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 + +HEAD_DIM = 64 + +class FmhaV3Softmax: + def __init__(self): + 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 + + 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, HEAD_DIM, 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), HEAD_DIM, 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) + 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; self.tmem_p0_offset = 32 + # P occupies [tmem_p0_offset, tmem_p0_offset + p_cols_fp32) + # S occupies [0, qk_mma_tiler[1]) = [0, 128) + # O must NOT overlap P. Place O after max(S end, P end), aligned to 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 # 32 + 64 = 96 + s_cols = self.qk_mma_tiler[1] # 128 + o_after = max(s_cols, p_end) # 128 + self.tmem_o0_offset = ((o_after + 31) // 32) * 32 + self.tmem_vec_offset = 0 # Reuse S region for per-row inv_row_sum vector # align to 32 = 128 + self.tmem_vec_offset = 0 # Reuse S region (free after softmax loop) + o_cols = find_tmem_tensor_col_offset(tOtO) # footprint of O + total = self.tmem_o0_offset + o_cols + # Must be multiple of 32 AND power of 2 + 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)) + 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) * cta + self.scale_softmax_log2 = Float32(1.0 / math.sqrt(HEAD_DIM) * math.log2(math.e)) + + @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() + # # s_k hardcoded # BROKEN in @cute.jit + # FMHA-style V: reconstruct as (HEAD_DIM, s_k, 1) MN-major + v_fmha = cute.make_tensor( + v.iterator, + cute.make_layout( + (HEAD_DIM, 128, 1), + stride=(1, HEAD_DIM, HEAD_DIM * 128), + ), + ) + 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_mma = utils.sm100.make_trivial_tiled_mma(self.q_dtype, self.q_dtype, cute.nvgpu.OperandMajorMode.K, self.v_major, self.qk_acc_dtype, self.cta_group, (128,HEAD_DIM), tcgen05.OperandSource.TMEM) + 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.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, 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)) + pv_done_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) + + 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) + 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 read view (for MMA only, NOT for softmax store) --- + tP = cute.make_tensor(tStS.iterator, p_tmem_s.outer) + tOrP_base = pv_thr.make_fragment_A(tP) + tOrP = tOrP_base[(None,None,None,0)] + tOrP0 = cute.make_tensor( + tOrP.iterator + self.qk_acc_dtype.width // self.q_dtype.width * self.tmem_p0_offset, + tOrP.layout) + + tCtS_fake = qk_mma.make_fragment_C(cute.append(qk_as, self.num_acc_stage)) + 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 + if warp_idx == self.tma_warp_id: + qp.reset(); qh = qp.acquire_and_advance() + cute.copy(tma_q,tAgQ[(None,qh.count)],tAsQ[(None,qh.index)],tma_bar_ptr=qh.barrier) + qp.tail() + kvp.reset(); pk = kvp.try_acquire() + for kt in cutlass.range(n_kv_tiles,unroll=1): + kh = kvp.acquire_and_advance(pk) + cute.copy(tma_k,tBgK[(None,kh.count)],tBsK[(None,kh.index)],tma_bar_ptr=kh.barrier) + pk = cutlass.Boolean(1) + vh = kvp.acquire_and_advance(pk) + cute.copy(tma_v,tVgV[(None,vh.count)],tVsV[(None,vh.index)],tma_bar_ptr=vh.barrier) + pk = cutlass.Boolean(1) + kvp.tail() + + # MMA + 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(n_kv_tiles): + kh = 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,kh.index)], tStS0) + qk_mma.set(tcgen05.Field.ACCUMULATE, True) + cute.arch.fence_view_async_tmem_store() + sh.commit(); kh.release() + softmax_done_bar.arrive_and_wait() + vh = kvc.wait_and_advance(pk); pk = cutlass.Boolean(1) + pv_mma.set(tcgen05.Field.ACCUMULATE, kt != 0) + 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,vh.index)], tOtO0) + pv_mma.set(tcgen05.Field.ACCUMULATE, True) + cute.arch.fence_view_async_tmem_store() + vh.release() + pv_done_bar.arrive() + acc_pipe.producer_commit(acc_st); acc_st.advance() + acc_pipe.producer_tail(acc_st) + + # ===================== EPILOGUE WARPS (STAGE C: ONLINE SOFTMAX) ===================== + 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 (QK C-fragment) --- + 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 (QK C-fragment composition, FMHA pattern) --- + 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))) + tStP0 = cute.make_tensor(tStS.iterator + self.tmem_p0_offset, 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) + + # --- Vector TMEM (per-row row_sum storage, FMHA pattern) --- + # composition(tStS.layout, (128, 2)) = 2 FP32 columns per logical row + # vec[0] = row_sum (final, after loop), vec[1] = unused + # Reuses S TMEM region (offset 0), free after softmax loop writes + + tStS_vec_layout = cute.composition(tStS.layout, cute.make_layout((128, 2))) + tStS_vec = cute.make_tensor(tStS.iterator + self.tmem_vec_offset, tStS_vec_layout) + tScS_vec_layout = cute.composition(tScS.layout, cute.make_layout((128, 2))) + tScS_vec = cute.make_tensor(tScS.iterator, tScS_vec_layout) + tmem_store_vec_atom = cute.make_copy_atom(tcgen05.copy.St32x32bOp(tcgen05.copy.Repetition(2)), self.qk_acc_dtype) + tiled_tmem_store_vec = tcgen05.make_tmem_copy(tmem_store_vec_atom, tStS_vec) + thr_tmem_store_vec = tiled_tmem_store_vec.get_slice(sfw_idx) + tTMEM_STORE_VECtS = thr_tmem_store_vec.partition_D(tStS_vec) + tTMEM_STORE_VECcS = thr_tmem_store_vec.partition_S(tScS_vec) + tmem_load_vec_atom = cute.make_copy_atom(tcgen05.copy.Ld32x32bOp(tcgen05.copy.Repetition(2)), self.qk_acc_dtype) + tiled_tmem_load_vec = tcgen05.make_tmem_copy(tmem_load_vec_atom, tStS_vec) + thr_tmem_load_vec = tiled_tmem_load_vec.get_slice(sfw_idx) + tTMEM_LOAD_VECtS = thr_tmem_load_vec.partition_S(tStS_vec) + tTMEM_LOAD_VECcS = thr_tmem_load_vec.partition_D(tScS_vec) + + # --- C6: O TMEM load/store for rescale (correction_rescale pattern) --- + corr_tile_size = 16 + cO = cute.make_identity_tensor((self.pv_mma_tiler[0], self.pv_mma_tiler[1])) + tOcO = pv_thr.partition_C(cO) + o_tmem_load_atom = cute.make_copy_atom(tcgen05.copy.Ld32x32bOp(tcgen05.copy.Repetition(corr_tile_size)), self.qk_acc_dtype) + o_tmem_store_atom = cute.make_copy_atom(tcgen05.copy.St32x32bOp(tcgen05.copy.Repetition(corr_tile_size)), self.qk_acc_dtype) + 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) + o_tiled_tmem_load = tcgen05.make_tmem_copy(o_tmem_load_atom, tOtO_i) + o_tiled_tmem_store = tcgen05.make_tmem_copy(o_tmem_store_atom, tOtO_i) + o_thr_load = o_tiled_tmem_load.get_slice(sfw_idx) + o_thr_store = o_tiled_tmem_store.get_slice(sfw_idx) + tTMEM_LOADtO = o_thr_load.partition_S(tOtO_i) + tTMEM_LOADcO = o_thr_load.partition_D(tOcO_i) + tTMEM_STOREtO = o_thr_store.partition_D(tOtO_i) + o_col_tiles = self.pv_mma_tiler[1] // corr_tile_size + + # --- C2: Per-thread row state (persist across KV tiles) --- + row_max = -cutlass.Float32.inf + row_sum = cutlass.Float32(0.0) + + # --- C3: QK scale = 1/sqrt(HEAD_DIM) * log2(e) for exp2 --- + scale = self.scale_softmax_log2 + + # ============================================================= + # Per-KV-tile online softmax loop + # ============================================================= + for kt in range(n_kv_tiles): + si_handle = s_cons.wait_and_advance() + + # Load S from TMEM (FP32, QK C-fragment layout) + tTMEM_LOADrS = cute.make_rmem_tensor(tTMEM_LOADcS.shape, self.qk_acc_dtype) + cute.copy(tiled_tmem_load, tTMEM_LOADtS, tTMEM_LOADrS) + + # --- C4: Compute tile_max via .reduce(MAX) --- + old_row_max = row_max + row_max = tTMEM_LOADrS.load().reduce(cute.ReductionOp.MAX, row_max, 0) + row_max_safe = row_max + if row_max == -cutlass.Float32.inf: + row_max_safe = cutlass.Float32(0.0) + + # --- C5: Compute rescale factor --- + acc_scale = cute.math.exp2(scale * (old_row_max - row_max_safe), fastmath=True) + + # --- C6: Rescale O in TMEM (load O, multiply by acc_scale, store O) --- + # acc_scale belongs to QK row (N//4), but O rows are in PV partition (N). + # Store acc_scale to vector by QK row, read by PV row. + if kt > 0: + pv_done_bar.arrive_and_wait() + + # Store acc_scale to vector indexed by QK logical row + qk_row_c6 = tTMEM_LOADcS[0][0] + thr_vs_c6 = tiled_tmem_store_vec.get_slice(qk_row_c6) + tVStore_c6 = thr_vs_c6.partition_D(tStS_vec) + tVStoreSrc_c6 = thr_vs_c6.partition_S(tScS_vec) + tVStoreRmem_c6 = cute.make_rmem_tensor(tVStoreSrc_c6.shape, self.qk_acc_dtype) + tVStoreRmem_c6[0] = acc_scale + cute.copy(tiled_tmem_store_vec, tVStoreRmem_c6, tVStore_c6) + cute.arch.fence_view_async_tmem_store() + + # Read acc_scale from vector indexed by PV logical row + pv_row_c6 = tTMEM_LOADcO[0][0] + thr_vl_c6 = tiled_tmem_load_vec.get_slice(pv_row_c6) + tVLoad_c6 = thr_vl_c6.partition_S(tStS_vec) + tVLoadDst_c6 = thr_vl_c6.partition_D(tScS_vec) + tVLoadRmem_c6 = cute.make_rmem_tensor(tVLoadDst_c6.shape, self.qk_acc_dtype) + cute.copy(tiled_tmem_load_vec, tVLoad_c6, tVLoadRmem_c6) + cute.arch.fence_view_async_tmem_load() + acc_scale_pv = tVLoadRmem_c6[0] + + tTMrO = cute.make_rmem_tensor((tTMEM_LOADcO.shape, o_col_tiles), self.qk_acc_dtype) + for i in range(o_col_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(o_tiled_tmem_load, tTMEM_LOADtO_i, tTMrO_i) + for j in cutlass.range(cute.size(tTMrO_i), vectorize=True): + tTMrO_i[j] = tTMrO_i[j] * acc_scale_pv + cute.copy(o_tiled_tmem_store, tTMrO_i, tTMEM_STOREtO_i) + cute.arch.fence_view_async_tmem_store() + + # Rescale row_sum + row_sum = row_sum * acc_scale + + # --- C7: Compute P = exp2((S - row_max_safe) * scale) --- + minus_row_max_scale = (cutlass.Float32(0.0) - row_max_safe) * scale + + # Register bridge (FMHA pattern: FP32 backing + BF16 view) + 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) + + frg_cnt = 4 + frg_tile = cute.size(tTMEM_LOADrS) // frg_cnt + tTMEM_LOADrS_frg = cute.logical_divide(tTMEM_LOADrS, cute.make_layout(frg_tile)) + rP_bf16_frg = cute.logical_divide(rP_bf16, cute.make_layout(frg_tile)) + + # Scale S, compute exp2, store through register bridge + for j in range(frg_cnt): + for k in cutlass.range(cute.size(tTMEM_LOADrS_frg, mode=[0]), vectorize=True): + tTMEM_LOADrS_frg[k, j] = tTMEM_LOADrS_frg[k, j] * scale + minus_row_max_scale + tTMEM_LOADrS_frg[k, j] = cute.math.exp2(tTMEM_LOADrS_frg[k, j], fastmath=True) + s_vec = tTMEM_LOADrS_frg[None, j].load() + rP_bf16_frg[None, j].store(s_vec.to(self.q_dtype)) + + # Store P to TMEM + cute.copy(tiled_tmem_store, rP_words, tTMEM_STOREtP) + cute.arch.fence_view_async_tmem_store() + si_handle.release() + softmax_done_bar.arrive() + + # --- C8: Row sum accumulation (CUTLASS FMHA packed f32x2 pattern) --- + # P values still in tTMEM_LOADrS registers. + # 4 accumulators for 4 reduction_unroll columns. + local_row_sum_0 = (cutlass.Float32(0.0), cutlass.Float32(0.0)) + local_row_sum_1 = (cutlass.Float32(0.0), cutlass.Float32(0.0)) + local_row_sum_2 = (cutlass.Float32(0.0), cutlass.Float32(0.0)) + local_row_sum_3 = (cutlass.Float32(0.0), cutlass.Float32(0.0)) + + reduction_unroll = 4 + rfrg_tile = cute.size(tTMEM_LOADrS) // reduction_unroll + tTMEM_LOADrS_rfrg = cute.logical_divide(tTMEM_LOADrS, cute.make_layout(rfrg_tile)) + + for j in cutlass.range_constexpr(0, cute.size(tTMEM_LOADrS_rfrg, mode=[0]), 2): + local_row_sum_0 = cute.arch.add_packed_f32x2( + local_row_sum_0, (tTMEM_LOADrS_rfrg[j, 0], tTMEM_LOADrS_rfrg[j + 1, 0])) + local_row_sum_1 = cute.arch.add_packed_f32x2( + local_row_sum_1, (tTMEM_LOADrS_rfrg[j, 1], tTMEM_LOADrS_rfrg[j + 1, 1])) + local_row_sum_2 = cute.arch.add_packed_f32x2( + local_row_sum_2, (tTMEM_LOADrS_rfrg[j, 2], tTMEM_LOADrS_rfrg[j + 1, 2])) + local_row_sum_3 = cute.arch.add_packed_f32x2( + local_row_sum_3, (tTMEM_LOADrS_rfrg[j, 3], tTMEM_LOADrS_rfrg[j + 1, 3])) + + local_row_sum_0 = cute.arch.add_packed_f32x2(local_row_sum_0, local_row_sum_1) + local_row_sum_2 = cute.arch.add_packed_f32x2(local_row_sum_2, local_row_sum_3) + local_row_sum_0 = cute.arch.add_packed_f32x2(local_row_sum_0, local_row_sum_2) + tile_sum = local_row_sum_0[0] + local_row_sum_0[1] + + row_sum = row_sum + tile_sum + + # --- C9: Final normalization via O TMEM rescale --- + pv_done_bar.arrive_and_wait() + + # Write row_sum to TMEM vector using QK partition (correct row mapping) + qk_row_c9 = tTMEM_LOADcS[0][0] + thr_vs_c9 = tiled_tmem_store_vec.get_slice(qk_row_c9) + tVStore_c9 = thr_vs_c9.partition_D(tStS_vec) + tVStoreSrc_c9 = thr_vs_c9.partition_S(tScS_vec) + tVStoreRmem_c9 = cute.make_rmem_tensor(tVStoreSrc_c9.shape, self.qk_acc_dtype) + tVStoreRmem_c9[0] = row_sum + cute.copy(tiled_tmem_store_vec, tVStoreRmem_c9, tVStore_c9) + cute.arch.fence_view_async_tmem_store() + + # Read row_sum from TMEM vector using PV partition (correct for O rows) + pv_row_c9 = tTMEM_LOADcO[0][0] + thr_vl_c9 = tiled_tmem_load_vec.get_slice(pv_row_c9) + tVLoad_c9 = thr_vl_c9.partition_S(tStS_vec) + tVLoadDst_c9 = thr_vl_c9.partition_D(tScS_vec) + tVLoadRmem_c9 = cute.make_rmem_tensor(tVLoadDst_c9.shape, self.qk_acc_dtype) + cute.copy(tiled_tmem_load_vec, tVLoad_c9, tVLoadRmem_c9) + cute.arch.fence_view_async_tmem_load() + pv_row_sum = tVLoadRmem_c9[0] + + inv_row_sum = cutlass.Float32(1.0) / pv_row_sum + + # Normalize O in TMEM using PV-correct inv_row_sum + tTMrO_final = cute.make_rmem_tensor((tTMEM_LOADcO.shape, o_col_tiles), self.qk_acc_dtype) + for i in range(o_col_tiles): + tTMrO_i_ = tTMrO_final[None, i] + tTMrO_i_layout = cute.composition(tTMrO_i_.layout, cute.make_layout(tTMrO_final.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(o_tiled_tmem_load, 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(o_tiled_tmem_store, tTMrO_i, tTMEM_STOREtO_i) + cute.arch.fence_view_async_tmem_store() + + # Now O in TMEM is normalized. Use standard epilogue_tma_store with identity. + 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) + + +def test(): + import math + torch.manual_seed(42) + for n in [128, 256, 384]: + m, hd = 128, HEAD_DIM + q = torch.randn(m, hd, 1, dtype=torch.bfloat16, device="cuda") + k = torch.randn(n, hd, 1, dtype=torch.bfloat16, device="cuda") + v = torch.randn(n, hd, dtype=torch.bfloat16, device="cuda") + v_kernel = v.unsqueeze(-1) + c = torch.zeros(m, hd, 1, dtype=torch.bfloat16, device="cuda") + qf = q[:,:,0].float(); kf = k[:,:,0].float() + attn = qf @ kf.T / math.sqrt(hd) + ref = torch.softmax(attn, dim=-1) @ v.float() + mQ = ct.from_dlpack(q).mark_layout_dynamic(leading_dim=ct.get_leading_dim(q)) + mK = ct.from_dlpack(k).mark_layout_dynamic(leading_dim=ct.get_leading_dim(k)) + mV = ct.from_dlpack(v_kernel).mark_layout_dynamic(leading_dim=ct.get_leading_dim(v_kernel)) + mC = ct.from_dlpack(c).mark_layout_dynamic(leading_dim=ct.get_leading_dim(c)) + stream = cuda.CUstream(torch.cuda.current_stream().cuda_stream) + kernel = FmhaV3Softmax() + print(f"n={n}: Compiling...", flush=True) + compiled = cute.compile(kernel, mQ, mK, mV, mC, stream) + print(f"n={n}: tmem: s0={kernel.tmem_s0_offset} p0={kernel.tmem_p0_offset} o0={kernel.tmem_o0_offset} vec={kernel.tmem_vec_offset} alloc={kernel.num_tmem_alloc_cols}", flush=True) + print(f"n={n}: Running...", flush=True) + compiled(mQ, mK, mV, mC, stream) + torch.cuda.synchronize() + out = c[:,:,0].float() + cos = torch.nn.functional.cosine_similarity(out.flatten().unsqueeze(0), ref.flatten().unsqueeze(0)).item() + max_err = (out - ref).abs().max().item() + print(f"FMHA softmax n={n}: cosine {cos:.6f} max_err {max_err:.6f} {'PASS' if cos >= 0.999 else 'FAIL'}", flush=True) + +if __name__ == "__main__": + test() + +