from __future__ import annotations import collections import dataclasses import functools import inspect import itertools import logging import math import operator import os import pprint import textwrap import traceback import typing from collections import Counter, defaultdict from typing import Any, Callable, Generic, Optional, TYPE_CHECKING, TypeVar, Union if TYPE_CHECKING: from collections.abc import Sequence from types import ModuleType import sympy import torch import torch._inductor.async_compile # noqa: F401 required to warm up AsyncCompile pools from torch._dynamo.utils import counters, dynamo_timed from torch._inductor.codecache import LambdaFuture, PyCodeCache from torch._inductor.metrics import get_metric_table, is_metric_table_enabled from torch.fx.experimental.symbolic_shapes import free_unbacked_symbols from torch.utils._ordered_set import OrderedSet from torch.utils._sympy.symbol import free_symbol_is_type, SymT from torch.utils._triton import has_triton from . import comms, config, dependencies, ir, metrics from .analyze_preserves_zero_mask import can_codegen_without_upcasts from .codegen.common import BackendFeature, get_scheduling_for_device, Kernel from .comm_analysis import estimate_nccl_collective_runtime from .dependencies import Dep, MemoryDep, StarDep, WeakDep from .exc import GPUTooOldForTriton, TritonMissing from .ir import ( ComputedBuffer, get_device_type, GraphPartitionSignature, MultiOutput, MultiOutputLayout, ) from .loop_body import LoopBody from .memory import MemoryPlanningInfoForBuffer, MemoryPlanningInfoForNode from .runtime.runtime_utils import green_text, red_text from .sizevars import SimplifyIndexing from .utils import ( cache_on_self, cmp, device_need_guard, get_device_tflops, get_dtype_size, get_gpu_dram_gbps, IndentedBuffer, is_collective, is_gpu, is_multi_outputs_template, is_output_of_multi_outputs_template, is_wait, sympy_product, ) from .virtualized import V log = logging.getLogger(__name__) fusion_log = torch._logging.getArtifactLogger(__name__, "fusion") loop_ordering_log = torch._logging.getArtifactLogger(__name__, "loop_ordering") PartitionType = list["BaseSchedulerNode"] @dataclasses.dataclass class SchedulerBuffer: scheduler: Scheduler node: ir.Buffer defining_op: Optional[BaseSchedulerNode] users: list[NodeUser] = dataclasses.field(default_factory=list) mpi_buffer: MemoryPlanningInfoForBuffer = dataclasses.field( default_factory=MemoryPlanningInfoForBuffer ) def defining_op_name(self) -> str: op = self.defining_op assert op is not None return op.get_name() def __hash__(self) -> int: return hash(self.node.name) def debug_str(self) -> str: result = IndentedBuffer() name = self.get_name() result.writeline(f"{name}: {type(self.node).__name__}") result.writeline(f"{name}.layout = {self.node.layout}") if self.get_aliases(): result.writeline(f"{name}.aliases = {pformat(self.get_aliases())}") if self.get_mutations(): result.writeline(f"{name}.mutations = {pformat(self.get_mutations())}") if len(self.users) <= 1: result.writeline(f"{name}.users = {self.users}") else: result.writeline(f"{name}.users = [") with result.indent(1): for user in self.users: result.writeline(f"{user},") result.writeline("]") return result.getrawvalue() def get_name(self) -> str: return self.node.get_name() def allocate(self) -> None: assert self.node is not None if not self.node.should_allocate(): return if ( self.node.get_inputs_that_alias_output() or self.node.get_mutation_names() or isinstance(self.node.get_output_spec(), ir.CommBufferLayout) ): V.graph.wrapper_code.codegen_allocation(self.node) return # hacky check for if V.kernel is a real kernel or NullHandler if ( hasattr(V.kernel, "args") and self.get_name() in V.kernel.inplace_update_buffers ): input_buffer: Union[ir.DonatedBuffer, ir.Buffer] input_buffer_name = V.kernel.inplace_update_buffers[self.get_name()] if input_buffer_name in self.scheduler.name_to_donated_buffer: input_buffer = self.scheduler.name_to_donated_buffer[ input_buffer_name ].node else: input_buffer = self.scheduler.name_to_buf[input_buffer_name].node V.graph.wrapper_code.codegen_inplace_reuse( input_buffer, self.node, ) else: V.graph.wrapper_code.codegen_allocation(self.node) def can_free(self) -> bool: # There's no real allocated buffer, no need to free it assert self.node is not None if isinstance(self.node.layout, ir.NoneLayout) or is_multi_outputs_template( self.node ): return False for use in self.users: if isinstance(use.node, OutputNode): return False return True def set_users(self, users: list[NodeUser]) -> None: # deduplicate result: dict[int, NodeUser] = {} for use in users: if id(use.node) in result: result[id(use.node)] = use.merge(result[id(use.node)]) else: result[id(use.node)] = use self.users = list(result.values()) def get_aliases(self) -> Sequence[str]: assert self.node is not None return self.node.get_inputs_that_alias_output() def get_mutations(self) -> Sequence[str]: assert self.node is not None return self.node.get_mutation_names() @dataclasses.dataclass class SchedulerDonatedBuffer(SchedulerBuffer): defining_op: Optional[BaseSchedulerNode] = None class BaseSchedulerNode: group: tuple[torch.device, tuple[tuple[sympy.Expr, ...], ...]] read_writes: dependencies.ReadWrites unmet_dependencies: OrderedSet[Dep] # .min_order and .max_order are only relevant for "grouped" nodes such as FusedSchedulerNode. # e.g. if the FusedSchedulerNode includes nodes (op_1, op_2, op_3), and op_X is X-th node # in `self.scheduler.nodes`, then for this FusedSchedulerNode, .min_order is 1 and .max_order is 3. # For non-"grouped" nodes (i.e. regular SchedulerNode), # .min_order = .max_order = X if this node is X-th node in `self.scheduler.nodes`. min_order: int max_order: int mpi_node: MemoryPlanningInfoForNode def __init__(self, scheduler: Scheduler) -> None: self.scheduler: Scheduler = scheduler self.debug_device_str: Callable[[BaseSchedulerNode], list[str]] = ( lambda *args, **kwargs: [] ) def _init_from_node(self, node: ir.Operation) -> None: self.node: Optional[ir.Operation] = node self.ancestors = OrderedSet[str]() self.last_usage = OrderedSet[ str ]() # buffers that won't be used after this kernel self.written = False self.outputs: list[SchedulerBuffer] = [ SchedulerBuffer( scheduler=self.scheduler, node=output, defining_op=self, ) for output in node.get_outputs() ] self.outputs_by_name: dict[str, SchedulerBuffer] = { buf.get_name(): buf for buf in self.outputs } def __repr__(self) -> str: return f"{type(self).__name__}(name={self.get_name()!r})" def debug_str(self) -> str: """Longer form printout for trace logs""" name = self.get_name() buf = IndentedBuffer() buf.splice( f"""\ {name}: {type(self).__name__}({type(getattr(self, "node", None)).__name__}) {name}.writes = {pformat(self.read_writes.writes)} {name}.unmet_dependencies = {pformat(self.unmet_dependencies)} {name}.met_dependencies = {pformat(self.read_writes.reads - self.unmet_dependencies)} {name}.outputs = [ """ ) with buf.indent(): for out in self.get_outputs(): buf.splice(out.debug_str()) buf.writeline("]") try: buf.splice(self.debug_str_extra()) except Exception: log.warning("Ignoring error in debug_str()", exc_info=True) return buf.getrawvalue().rstrip() def debug_str_extra(self) -> str: return "" def _debug_str_for_device(self) -> list[str]: return self.debug_device_str(self) def debug_str_short(self) -> str: maybe_data = getattr(self.node, "data", None) data_str = "" if isinstance(maybe_data, torch._inductor.ir.Pointwise): data_str = ", " + maybe_data.str_helper( [maybe_data.get_size()], shorten=False, multiline=False ) elif isinstance(maybe_data, torch._inductor.ir.Reduction): data_str = ", " + maybe_data.str_helper( [maybe_data.get_reduction_size(), maybe_data.get_reduction_type()], shorten=False, multiline=False, ) return f"{self}{data_str}" def log_details(self) -> None: log.info( "%s: unmet_dependencies = %s, writes = %s", self, self.unmet_dependencies, self.read_writes.writes, ) def reorder_loops_by_dep_pair( self, self_dep: MemoryDep, other_dep: MemoryDep ) -> None: return def update_mutated_names(self, renames: dict[str, str]) -> None: self.set_read_writes(self.read_writes.rename(renames)) def add_fake_dep(self, dep: Dep) -> None: self.set_read_writes(self.read_writes.with_read(dep)) def has_aliasing_or_mutation(self) -> bool: return any( buf.get_aliases() or buf.get_mutations() for buf in self.get_outputs() ) def set_read_writes(self, rw: dependencies.ReadWrites) -> None: self.read_writes = rw self.unmet_dependencies = self.read_writes.reads self.prune_deps() def set_last_usage( self, future_used_buffers: OrderedSet[str], mutation_real_name: dict[str, str] ) -> None: used_buffers = self.used_or_aliased_buffer_names() used_buffers = OrderedSet(mutation_real_name.get(k, k) for k in used_buffers) self.last_usage = used_buffers - future_used_buffers def mark_run(self) -> None: for buf in self.outputs: buf.allocate() def used_buffer_names(self) -> OrderedSet[str]: return OrderedSet( dep.name for dep in itertools.chain(self.read_writes.reads, self.read_writes.writes) ) def used_or_aliased_buffer_names(self) -> OrderedSet[str]: used_names = OrderedSet[str]() deps = [ dep.name for dep in itertools.chain(self.read_writes.reads, self.read_writes.writes) ] while len(deps) > 0: dep = deps.pop() used_names.add(dep) if V.graph.name_to_buffer.get(dep): deps.extend( alias for alias in V.graph.name_to_buffer[ dep ].get_inputs_that_alias_output() if alias not in used_names ) return used_names def prune_deps(self) -> None: self.unmet_dependencies = OrderedSet( dep for dep in self.unmet_dependencies if dep.name not in self.scheduler.available_buffer_names ) def prune_weak_deps(self) -> None: # Prune weak dependencies on operations that have been removed def should_prune(dep: Dep) -> bool: if not isinstance(dep, WeakDep): return False op_name = self.scheduler.name_to_buf[dep.name].defining_op_name() return op_name in V.graph.removed_operations to_remove = OrderedSet( dep for dep in self.read_writes.reads if should_prune(dep) ) self.set_read_writes(self.read_writes.remove_reads(to_remove)) def prune_redundant_deps( self, name_to_fused_node: dict[str, BaseSchedulerNode] ) -> None: _prune_redundant_deps(self, name_to_fused_node, self.scheduler.name_to_buf) def get_name(self) -> str: assert self.node is not None return self.node.get_operation_name() def get_first_name(self) -> str: return self.get_name() @cache_on_self def get_operation_names(self) -> OrderedSet[str]: return OrderedSet(node.get_name() for node in self.get_nodes()) @cache_on_self def get_buffer_names(self) -> OrderedSet[str]: return OrderedSet(out.get_name() for out in self.outputs) @cache_on_self def can_codegen_in_low_precision(self) -> bool: return all( isinstance(n, SchedulerNode) and can_codegen_without_upcasts(n, disallow_fp32_ops=True) for n in self.get_nodes() ) @cache_on_self def can_codegen_without_upcasts(self) -> bool: return all( isinstance(n, SchedulerNode) and can_codegen_without_upcasts(n) for n in self.get_nodes() ) def get_nodes(self) -> Sequence[BaseSchedulerNode]: return [self] def get_outputs(self) -> Sequence[SchedulerBuffer]: return self.outputs def get_output(self, buf_name: str) -> SchedulerBuffer: return self.outputs_by_name[buf_name] def get_device(self) -> Optional[torch.device]: assert self.node is not None return self.node.get_device() def is_cpu(self) -> bool: device = self.get_device() return device is not None and device.type == "cpu" def is_gpu(self) -> bool: device = self.get_device() return device is not None and is_gpu(device.type) def is_reduction(self) -> bool: return False def is_split_scan(self) -> bool: return False def is_template(self) -> bool: return False def is_extern(self) -> bool: return False def is_foreach(self) -> bool: return False def can_inplace(self, read_dep: dependencies.Dep) -> bool: return False def has_side_effects(self) -> bool: return False def decide_inplace_update(self) -> None: """ Decide if there should be inplace updates for the node and record the decision in the active kernel. """ from .codegen.wrapper import can_match_buffer_size if not ( isinstance(self, SchedulerNode) and config.inplace_buffers and V.graph.has_feature(self.get_device(), BackendFeature.INPLACE_BUFFERS) and ( not isinstance(V.kernel, torch._inductor.codegen.simd.SIMDKernel) or getattr(V.kernel, "mutations", None) is not None ) # hacky check for if V.kernel is a real kernel or NullHandler and hasattr(V.kernel, "args") ): return # NOTE remove V.graph.removed_operations once deps issue is fixed inconsequential_nodes = ( self.ancestors | V.graph.removed_operations | self.scheduler.completed_operations ) def single_index_in_fused_node(buf_to_be_inplaced: SchedulerBuffer) -> bool: # Inside of NodeUser, we track that the read and write are equivalent # before deciding if the use can be inplace. # But if that use is fused into a larger kernel, we need to check equivalence # of other accesses in fused scheduler node as well. fused_node = buf_to_be_inplaced.scheduler.get_fused_node(self) buf_name = buf_to_be_inplaced.get_name() # Dedup read/writes with equivalent indices # TODO - would be nice if we could just cache accesses on ReadWrites, # and inforce variant that this class & members are functional.. deps: OrderedSet[Dep] = OrderedSet() for user in buf_to_be_inplaced.users: user_node = user.node if not isinstance(user_node, BaseSchedulerNode): continue if ( buf_to_be_inplaced.scheduler.get_fused_node(user_node) is not fused_node ): continue deps |= ( o for o in user_node.read_writes.reads_and_writes() if o.name == buf_name ) if len(deps) > 1: return False return True for buf in self.get_outputs(): buf_node = buf.node assert buf_node is not None if ( not buf_node.should_allocate() or buf_node.get_inputs_that_alias_output() or buf_node.get_mutation_names() or buf.get_name() in V.graph.removed_buffers ): continue for read in self.read_writes.reads: input_buf: Optional[Union[SchedulerBuffer, SchedulerDonatedBuffer]] if read.name in self.scheduler.name_to_donated_buffer: input_buf = self.scheduler.name_to_donated_buffer[read.name] else: input_buf = self.scheduler.name_to_buf.get(read.name) if ( input_buf and V.graph.wrapper_code.can_reuse(input_buf, self) and not isinstance(input_buf.defining_op, NopKernelSchedulerNode) ): assert input_buf.users is not None remaining_uses = [ x for x in input_buf.users if x.node.get_name() not in inconsequential_nodes ] if ( len(remaining_uses) == 1 and remaining_uses[0].can_inplace and remaining_uses[0].node is self and input_buf.node is not None and not isinstance( input_buf.node.get_output_spec(), ( ir.NoneLayout, ir.MultiOutputLayout, ir.MutationLayoutSHOULDREMOVE, ), ) and not ( input_buf.defining_op and isinstance( input_buf.defining_op.node, (ir.FallbackKernel, ir.MultiOutput), ) and len(input_buf.node.get_inputs_that_alias_output()) > 0 ) and can_match_buffer_size(input_buf.node, buf.node) and single_index_in_fused_node(input_buf) ): # if there isn't a triton kernel, then we don't need to call triton-specific things. # but TODO this might be a convenient place to signal to the Collective kernels to inplace # (and, can we make "kernel" less generic of a name?) V.kernel.args.make_inplace(input_buf.get_name(), buf.get_name()) # mutations not tracked in cpp kernels if isinstance( V.kernel, torch._inductor.codegen.simd.SIMDKernel ): V.kernel.mutations.add(input_buf.get_name()) V.kernel.mutations.add(buf.get_name()) V.kernel.inplace_update_buffers[buf.get_name()] = ( input_buf.get_name() ) break def codegen_originating_info( self, buffer: IndentedBuffer, only_once: bool = True ) -> None: if not config.comment_origin: return if only_once and self.written: return assert self.node is not None origins = self.node.get_origins() out_lines = [] for o in origins: if o.op == "output": # These are boring and samey continue out_lines.append("") # TODO(voz): Should the pragma be constant somewhere? out_lines.append("#pragma CMT ORIGIN:") op_info_str = f"#pragma CMT {o.op} {o.target}" if "seq_nr" in o.meta: op_info_str = op_info_str + f" seq_nr:{o.meta['seq_nr']}" out_lines.append(op_info_str) if "stack_trace" in o.meta: stack_trace = f"{o.meta['stack_trace']}" stack_trace_last_line = stack_trace.split("|")[-1] out_lines.append( "#pragma CMT " + stack_trace_last_line.replace("{", "{{") .replace("}", "}}") .replace("\n", "\\") ) out_lines.append("#pragma CMT END ORIGIN") out_lines.append("") if len(out_lines) == 0: return # TODO(voz): Ostensibly, we should not need this. But there are cases where C++ codegen does # not use BracesBuffer, so we have no good indicator of a C++ buffer atm. buffer.writelines(out_lines) self.written = True @cache_on_self def get_read_write_buffers_sizes(self) -> int: return self.get_read_write_buffers_sizes_impl( include_reads=True, include_writes=True ) @cache_on_self def get_read_buffer_sizes(self) -> int: return self.get_read_write_buffers_sizes_impl( include_reads=True, include_writes=False ) @cache_on_self def get_write_buffer_sizes(self) -> int: return self.get_read_write_buffers_sizes_impl( include_reads=False, include_writes=True ) def get_read_write_buffers_sizes_impl( self, include_reads: bool, include_writes: bool ) -> int: return sum( self.get_read_write_buffer_accesses( include_reads=include_reads, include_writes=include_writes ).values(), start=0, ) def get_read_write_buffer_accesses( self, include_reads: bool, include_writes: bool ) -> dict[str, int]: """ Counting the number of bytes accessed for a kernel is surprisingly tricky. In particular, there is a differentiation between 'theoretical' memory accesses and practical memory accesses. For example, a layernorm kernel may actually access an input 3 times, but in theory, it only needs to access its input once (and may be optimized to do so through say, persistent reductions) Another example is that even though a buffer is passed in, we may not access the entire buffer. This may occur if we are accessing a slice of the buffer. Another tricky case is for indirect indexing, where the amount of bytes accessed depends on the values of the input. What this function aims to compute is the memory accesses for worst-case inputs, best-case optimization. What this means is that for each buffer we compute the amount of potential accesses in two ways and take the minimum. 1. Numel in ranges multiplied by number of deps the buffer has 2. The buffer size Returns memory accesses per buffer. """ if isinstance(self, NopKernelSchedulerNode): return {} if isinstance(self, ExternKernelSchedulerNode) and isinstance( self.node, MultiOutput ): # todo: Calculate this - it's kinda annoying. return {} def try_size_hint(s: sympy.Expr) -> int: return V.graph.sizevars.size_hint(s, fallback=0) if isinstance(self, SchedulerNode): node_numel = try_size_hint( sympy_product(self.get_ranges()[0]) * sympy_product(self.get_ranges()[1]), ) else: node_numel = int(1e9) buf_accesses = collections.defaultdict(list) if include_reads: for dep in self.read_writes.reads: buf_accesses[dep.name].append(dep) if include_writes: for dep in self.read_writes.writes: buf_accesses[dep.name].append(dep) reads = ( OrderedSet(dep.name for dep in self.read_writes.reads) if include_reads else OrderedSet() ) writes = ( OrderedSet(dep.name for dep in self.read_writes.writes) if include_writes else OrderedSet() ) def is_materialized(buf: str, snodes: Sequence[BaseSchedulerNode]) -> bool: users = self.scheduler.name_to_buf[buf].users buf_uses = OrderedSet(user.node for user in users) return len(buf_uses - OrderedSet(snodes)) > 0 if isinstance(self, FusedSchedulerNode): removed_buffers = OrderedSet( dep for dep in writes if not is_materialized(dep, self.snodes) ) writes = writes - removed_buffers reads = reads - removed_buffers buf_byte_accesses: dict[str, int] = {} for buf_name in reads | writes: buf_accessed_elems = sum(node_numel for dep in buf_accesses[buf_name]) buf: Union[ir.Buffer, ir.TensorBox, ir.TorchBindObject] if buf_name in V.graph.name_to_buffer: buf = V.graph.name_to_buffer[buf_name] elif buf_name in V.graph.graph_inputs: buf = V.graph.graph_inputs[buf_name] else: continue def get_buf_bytes( buf: Optional[Union[ir.Buffer, ir.TensorBox, ir.TorchBindObject]], ) -> int: if not buf: return 0 if isinstance(buf, ir.TorchBindObject): return buf.get_buf_bytes() elif isinstance(buf.layout, MultiOutputLayout): # Kind of a lazy way to get the MultiOutput nodes corresponding to # a MultiOutputLayout users = self.scheduler.name_to_buf[buf.get_name()].users tot = 0 for user in users: assert isinstance(user.node, BaseSchedulerNode) if isinstance(user.node.node, MultiOutput): for sched_buf in user.node.get_outputs(): tot += get_buf_bytes(sched_buf.node) else: # Buf is a MultiOutputLayout but not all of its # users are MultiOutputs... # TODO: Figure out what's going on return 0 return tot elif isinstance(buf.layout, ir.NoneLayout): return sum( get_buf_bytes(V.graph.get_buffer(mut_name)) for mut_name in buf.get_mutation_names() ) else: buf_elems = try_size_hint(sympy_product(buf.get_size())) return get_dtype_size(buf.get_dtype()) * min( buf_accessed_elems, buf_elems ) buf_bytes = get_buf_bytes(buf) if buf_name not in buf_byte_accesses: buf_byte_accesses[buf_name] = buf_bytes else: buf_byte_accesses[buf_name] += buf_bytes return buf_byte_accesses @cache_on_self def get_estimated_runtime(self) -> float: """ Returns estimated op runtime in nanoseconds (ns) """ buf = self.get_nodes()[0].get_outputs()[0] layout = buf.node.get_output_spec() if not is_gpu(get_device_type(layout)): # default to no reordering based on runtime return 0 # Collective kernels if is_collective(self.node): assert isinstance(self.node, ir.IRNode) try: return estimate_nccl_collective_runtime(self.node) except ValueError as e: # We don't know how to estimate runtime for this collective, # falling back to 0 log.info(e) return 0 except TypeError as e: # this happens when the collective is not of type ir._CollectiveKernel log.info(e) return 0 elif is_wait(self.node): # ir.Wait is only used for collective ops. # The time needed for the collective op is already estimated and considered # when we are processing the collective op IR node, so ir.Wait takes 0 time # since it doesn't take extra time to get the result after the collective is completed. return 0 dtype = buf.node.maybe_get_dtype() try: gpu_memory_bandwidth = get_gpu_dram_gbps() gpu_flops = get_device_tflops(dtype) * 10**12 except Exception: return 0 if isinstance(self, ExternKernelSchedulerNode): assert isinstance(self.node, ir.ExternKernel), f"{type(self.node)=}" op = kernel_name_to_op.get( getattr(self.node, "python_kernel_name", ""), None ) # if there is a resolved op, dry-run using fake mode and record flop count if op is not None: from torch._subclasses.fake_tensor import FakeTensorMode from torch.utils.flop_counter import FlopCounterMode if any( len(free_unbacked_symbols(n.get_numel())) > 0 for n in self.node.inputs ): # Tensor has unbacked symints, we don't know how to estimate # runtime for that today return 0 with ( FakeTensorMode() as fake_mode, FlopCounterMode(display=False) as flop_counter_mode, V.set_current_node(self.node.fx_node), V.set_fake_mode(fake_mode), ): from .ir import ir_node_to_tensor fake_inputs = [ ir_node_to_tensor(input, guard_shape=False) for input in self.node.inputs ] cls = self.node.__class__ cls.process_kernel(op, *fake_inputs, **self.node.kwargs) # TODO(xmfan): find a better heuristic to model FLOPS/latency relationship factor = 1.0 counted_flops = flop_counter_mode.get_total_flops() counted_bytes = self.get_read_write_buffers_sizes() compute_time = (factor * counted_flops / gpu_flops) * 1e9 transfer_time = counted_bytes / gpu_memory_bandwidth # Return estimated runtime in nanoseconds return max(compute_time, transfer_time) elif isinstance(self, FusedSchedulerNode) or isinstance( self.node, ComputedBuffer ): # Return estimated runtime in nanoseconds (bytes / gbps) return self.get_read_write_buffers_sizes() / gpu_memory_bandwidth return 0 def get_template_node(self) -> Optional[ir.TemplateBuffer]: return None def get_template_node_or_throw(self) -> ir.TemplateBuffer: template = self.get_template_node() assert template is not None return template @staticmethod def get_prologue_template_epilogue( nodes: list[BaseSchedulerNode], ) -> tuple[list[BaseSchedulerNode], BaseSchedulerNode, list[BaseSchedulerNode]]: """ For the list of nodes, get the prologue, template, and epilogue """ template_index = next(i for i, n in enumerate(nodes) if n.is_template()) prologue = nodes[:template_index] template_node = nodes[template_index] epilogue = nodes[template_index + 1 :] return prologue, template_node, epilogue class WhyNoFuse: # TODO when we drop support for Python < 3.10, we can use # @dataclass(slots=True) instead of manually specifying __slots__. __slots__ = ["node1", "node2", "reason", "args"] reason: str args: tuple[Any, ...] def __init__(self, node1: BaseSchedulerNode, node2: BaseSchedulerNode) -> None: self.node1 = node1 self.node2 = node2 def __call__(self, reason: str, *args: Any) -> None: self.reason = reason self.args = args fusion_log.debug(self) def __str__(self) -> str: return f"cannot fuse {self.node1.get_name()} with {self.node2.get_name()}: " + ( self.reason % self.args ) def pformat(obj: Any) -> str: if isinstance(obj, (OrderedSet, set)): # noqa: set_linter # pformat has trouble with sets of sympy exprs obj = sorted(obj, key=str) result = pprint.pformat(obj, indent=4) if "\n" in result: return f"\n{textwrap.indent(result, ' ' * 4)}" return result class OutputNode: def __init__(self, dep: StarDep) -> None: self.unmet_dependencies = OrderedSet([dep]) def is_reduction(self) -> bool: return False def get_inputs_that_alias_output(self) -> Sequence[str]: return () def get_name(self) -> str: return "OUTPUT" __repr__ = get_name def _prune_redundant_deps( node: BaseSchedulerNode, name_to_fused_node: dict[str, BaseSchedulerNode], name_to_buf: dict[str, SchedulerBuffer], ) -> None: """ Prunes weakdeps intended for mutation ordering on an upstream fused node if after fusion there is another dependency on the fused upstream node, making the weakdep redundant In essence this enforces an ordering on fusions. As fusions occur, weakdeps will be incrementally removed, enabling other fusions, ensuring they are fused in order. """ name_to_dep_count: Counter[str] = collections.Counter() for dep in node.unmet_dependencies: if not isinstance(dep, WeakDep): op_name = name_to_buf[dep.name].defining_op_name() name_to_dep_count[name_to_fused_node[op_name].get_name()] += 1 def should_prune(dep: Dep) -> bool: if isinstance(dep, WeakDep): op_name = name_to_buf[dep.name].defining_op_name() is_redundant = name_to_dep_count[name_to_fused_node[op_name].get_name()] > 0 # These can occur because fused nodes always gather deps from their snodes # If B has a weakdep on A # B gets fused with C, then any time BC is fused, the weakdep will reappear is_self_dep = name_to_fused_node[op_name] == node return is_redundant or is_self_dep else: return False deps_to_prune = OrderedSet( dep for dep in node.unmet_dependencies if should_prune(dep) ) if deps_to_prune: node.unmet_dependencies = node.unmet_dependencies - deps_to_prune node.set_read_writes(node.read_writes.remove_reads(deps_to_prune)) # TODO(xmfan): reuse: an existing mapping for this if it exists, or formalize this into ir.py:ExternKernel kernel_name_to_op = { "extern_kernels.convolution": torch.ops.aten.convolution, "extern_kernels.mm": torch.ops.aten.mm, "extern_kernels.bmm": torch.ops.aten.bmm, "extern_kernels.addmm": torch.ops.aten.addmm, "extern_kernels._scaled_mm": torch.ops.aten._scaled_mm, } class ExternKernelSchedulerNode(BaseSchedulerNode): def __init__(self, scheduler: Scheduler, node: ir.Operation) -> None: super().__init__(scheduler) self._init_from_node(node) self.set_read_writes(node.get_read_writes()) def debug_str_extra(self) -> str: return f"{self.get_name()}.node.kernel = {getattr(self.node, 'python_kernel_name', None)}" def is_extern(self) -> bool: return True def has_side_effects(self) -> bool: assert self.node is not None return hasattr(self.node, "has_side_effects") and self.node.has_side_effects() class NopKernelSchedulerNode(BaseSchedulerNode): def __init__(self, scheduler: Scheduler, node: ir.Operation) -> None: super().__init__(scheduler) self._init_from_node(node) self.set_read_writes(node.get_read_writes()) class SchedulerNode(BaseSchedulerNode): _sizes: tuple[Sequence[sympy.Expr], ...] _body: LoopBody def __init__( self, scheduler: Scheduler, node: Union[ir.ComputedBuffer, ir.TemplateBuffer], ) -> None: super().__init__(scheduler) self._init_from_node(node) self._compute_attrs() def _compute_attrs( self, extra_indexing_constraints: Optional[tuple[dict[Any, Any], list[Any]]] = None, recompute_sizes_body_func: Optional[Callable[..., Any]] = None, ) -> None: assert isinstance(self.node, (ir.ComputedBuffer, ir.TemplateBuffer)) self._sizes, self._body = self.node.simplify_and_reorder( extra_indexing_constraints=extra_indexing_constraints, recompute_sizes_body_func=recompute_sizes_body_func, ) device = self.node.get_device_or_error() group_fn = self.scheduler.get_backend(device).group_fn self.group = (device, group_fn(self._sizes)) # Don't normalize since normalization will merge loops which # makes it hard to decide new loop orders. should_normalize = not config.loop_ordering_after_fusion or not is_gpu( device.type ) if isinstance(self.node, ir.TemplateBuffer): self.set_read_writes( self.node.extract_read_writes(normalize=should_normalize) ) else: self.set_read_writes( dependencies.extract_read_writes( self._body, *self._sizes, normalize=should_normalize ) ) def recompute_size_and_body( self, extra_indexing_constraints: Optional[tuple[dict[Any, Any], list[Any]]] = None, recompute_sizes_body_func: Optional[Callable[..., Any]] = None, ) -> None: self._compute_attrs( extra_indexing_constraints=extra_indexing_constraints, recompute_sizes_body_func=recompute_sizes_body_func, ) def refresh_dependencies( self, normalize: bool, need_clear_tiling_cache: bool ) -> None: # Fake dependencies are added manually. They can not be analyzed from # extract_read_writes. Find them out and apply manually. fake_deps: OrderedSet[Dep] = OrderedSet( dep for dep in self.read_writes.reads if isinstance(dep, (WeakDep, StarDep)) ) # don't normalize since the loop order may need to be further changed # later self.set_read_writes( dependencies.extract_read_writes( self._body, *self._sizes, normalize=normalize ).with_read(fake_deps) ) self.pointwise_read_writes.clear_cache(self) if need_clear_tiling_cache: from .codegen.simd import SIMDScheduling # TODO(shunting) if this cause compilation time increase when # enabling LOAF by default, try just clearing the specific cache # entry by using a customized cache implemetation rather than # lru_cache. SIMDScheduling.candidate_tilings.cache_clear() def apply_new_loop_order(self, new_order: Sequence[int]) -> None: self._body = self._body.reorder_iter_loops( new_order, ) self._sizes = self._body.sizes self.refresh_dependencies(normalize=False, need_clear_tiling_cache=True) def merge_loops(self) -> None: self._body = self._body.merge_loops() self._sizes = self._body.sizes # merge_loops is called after loop reordering. # We still need retain fake dependencies since codegen the # estimated amount of memory access rely on them. # # Merge loops does not affect the tiling decision. So we # don't need clear the tiling cache. self.refresh_dependencies(normalize=True, need_clear_tiling_cache=False) def reorder_loops_by_dep_pair( self, self_dep: MemoryDep, other_dep: MemoryDep ) -> None: new_order = None self_sizes = self._sizes[0] if len(self_sizes) == self_dep.num_vars == other_dep.num_vars: new_order = self_dep.decide_loop_order_to_match(other_dep) if new_order: metrics.num_loop_reordering += 1 loop_ordering_log.debug( "Reorder loops for %s with order %s", self.get_name(), new_order ) self.apply_new_loop_order(new_order) else: loop_ordering_log.debug( "Don't reordering %s because we can not decide the suitable loop order", self.get_name(), ) def debug_str_extra(self) -> str: name = self.get_name() lines = [ f"{name}.group.device = {self.group[0]}", f"{name}.group.iteration = {self.group[1]}", f"{name}.sizes = {self._sizes}", ] for dep in self.read_writes.reads_and_writes(): if not isinstance(dep, WeakDep): buf_name = dep.name buf = V.graph.get_buffer(buf_name) if not isinstance(buf, ir.TorchBindObject): lines.append(f"{buf_name}_layout = {pformat(buf.layout)}") if isinstance(self._body, LoopBody): lines.append(f"class {name}_loop_body:") lines.append(textwrap.indent(self._body.debug_str(), " ")) assert self.node is not None lines.extend(self._debug_str_for_device()) return "\n".join(lines) def get_ranges(self) -> Sequence[Sequence[sympy.Expr]]: return self._sizes def is_reduction(self) -> bool: assert isinstance(self.node, (ir.ComputedBuffer, ir.TemplateBuffer)), ( f"{type(self.node)=}" ) return bool(self.node.get_reduction_type()) def is_split_scan(self) -> bool: assert isinstance(self.node, (ir.ComputedBuffer, ir.TemplateBuffer)), ( f"{type(self.node)=}" ) return isinstance(self.node, ir.ComputedBuffer) and isinstance( self.node.data, ir.SplitScan ) def is_template(self) -> bool: return isinstance(self.node, ir.TemplateBuffer) def get_template_node(self) -> Optional[ir.TemplateBuffer]: return self.node if isinstance(self.node, ir.TemplateBuffer) else None def run(self, *index_vars: Sequence[sympy.Expr]) -> None: self.decide_inplace_update() self.mark_run() self.codegen(index_vars) def ranges_from_index_vars( self, index_vars: Sequence[Sequence[sympy.Expr]] ) -> dict[sympy.Expr, sympy.Expr]: sizes = self._sizes assert sum(map(len, sizes)) == sum(map(len, index_vars)) var_ranges = dict( zip( itertools.chain.from_iterable(index_vars), itertools.chain.from_iterable(sizes), ) ) return var_ranges def codegen(self, index_vars: Sequence[Sequence[sympy.Expr]]) -> None: var_ranges = self.ranges_from_index_vars(index_vars) try: with ( V.set_ops_handler(SimplifyIndexing(V.get_ops_handler(), var_ranges)), V.kernel.set_current_node(self), ): self._body(*index_vars) except Exception: log.fatal("Error in codegen for %s", self.node) raise def pointwise_or_reduction_read_writes( self, pointwise: bool = True ) -> dependencies.ReadWrites: """ Get the memory dependencies in either the pointwise or the reduction axes. """ keep_sizes, ignore_sizes = self._sizes if pointwise else reversed(self._sizes) return dependencies.extract_read_writes( self._body, keep_sizes, hidden_args=[[sympy.S.Zero] * len(ignore_sizes)] ) @cache_on_self def pointwise_read_writes(self) -> dependencies.ReadWrites: """ Get the memory dependencies in the non-reduction axes. """ return self.pointwise_or_reduction_read_writes(pointwise=True) @cache_on_self def reduction_read_writes(self) -> dependencies.ReadWrites: """ Get the memory dependencies in the reduction axes. """ return self.pointwise_or_reduction_read_writes(pointwise=False) def can_inplace(self, read_dep: dependencies.Dep) -> bool: if self.is_template(): return False if any(out.get_aliases() for out in self.get_outputs()): return False if len(self.read_writes.writes) == 1 and isinstance( read_dep, dependencies.MemoryDep ): write_dep = next(iter(self.read_writes.writes)) assert isinstance(write_dep, dependencies.MemoryDep), f"{type(write_dep)=}" return read_dep.index == write_dep.index and read_dep.size == write_dep.size return False @cache_on_self def _get_atomic_add_buffers(self) -> OrderedSet[str]: buffers_store_as_atomic_add = OrderedSet[str]() if isinstance(self._body, LoopBody): for node in self._body.get_nodes(): if ( node.op == "call_method" and node.target == "store" and ( ("mode" in node.kwargs and node.kwargs["mode"] == "atomic_add") or (len(node.args) == 5 and node.args[4] == "atomic_add") ) ): buffers_store_as_atomic_add.add( node.kwargs["name"] if "name" in node.kwargs else (node.args[1] if len(node.args) >= 2 else "") ) return buffers_store_as_atomic_add def refresh_group_node_dependencies( group_snode: Union[FusedSchedulerNode, GroupedSchedulerNode], ) -> None: snodes = group_snode.snodes group_snode.set_read_writes( dependencies.ReadWrites.merge_list([x.read_writes for x in snodes]) ) group_snode.unmet_dependencies = ( OrderedSet( dep for dep in OrderedSet.union(*[x.unmet_dependencies for x in snodes]) if dep.name not in group_snode.get_buffer_names() ) - group_snode.read_writes.writes ) def init_group_node( group_snode: Union[FusedSchedulerNode, GroupedSchedulerNode], scheduler: Scheduler, snodes: list[BaseSchedulerNode], ) -> None: assert isinstance(group_snode, (FusedSchedulerNode, GroupedSchedulerNode)) group_snode.snodes = snodes group_snode.scheduler = scheduler group_snode.node = None group_snode.ancestors = OrderedSet.union( *[x.ancestors for x in snodes if x.ancestors is not None] ) refresh_group_node_dependencies(group_snode) group_snode.min_order = min(x.min_order for x in group_snode.snodes) group_snode.max_order = max(x.max_order for x in group_snode.snodes) group_snode.outputs_by_name = { buf.get_name(): buf for buf in group_snode.get_outputs() } class FusedSchedulerNode(BaseSchedulerNode): """ This is a "fake" scheduler node that represents a group of scheduler nodes that are meant to be fused together. The way it does this is by maintaining its unmet dependencies as the union of its constituent nodes. """ snodes: list[BaseSchedulerNode] @classmethod def fuse( cls, node1: BaseSchedulerNode, node2: BaseSchedulerNode ) -> FusedSchedulerNode: assert node1.scheduler is node2.scheduler assert isinstance(node1, (SchedulerNode, FusedSchedulerNode)) if node1.is_template() and isinstance(node2, ExternKernelSchedulerNode): # Fuse multi outputs template and its outputs # * Node1 has memorydep of MultiOutput in reads # * Node2 has StarDep of MultiOutput in writes # Rewrite the Node2' StarDep to MemoryDep, because calculate score_fusion_memory # of the template node and its epilogue requires the same type of dependencies assert isinstance(node2.node, MultiOutput) assert len(node2.read_writes.writes) == 1 assert isinstance(next(iter(node2.read_writes.writes)), StarDep) name = next(iter(node2.read_writes.writes)).name template_nodes = [node for node in node1.get_nodes() if node.is_template()] assert len(template_nodes) == 1 template_node = template_nodes[0] assert len(template_node.read_writes.writes) == 1 write = next(iter(template_node.read_writes.writes)) assert isinstance(write, MemoryDep) node2.read_writes.writes = OrderedSet( [ MemoryDep( name, write.index, write.var_names, write.size, write.mode ), ] ) else: assert isinstance(node2, (SchedulerNode, FusedSchedulerNode)) nodes = list(itertools.chain(node1.get_nodes(), node2.get_nodes())) return cls(node1.scheduler, nodes) def reorder_loops_by_dep_pair( self, self_dep: MemoryDep, other_dep: MemoryDep ) -> None: if self.is_template(): # We can not really reorder loops for a triton template return self_sizes = None for snode in self.snodes: assert isinstance(snode, SchedulerNode) if self_sizes is not None and tuple(self_sizes) != tuple(snode._sizes[0]): loop_ordering_log.debug( "Can not reorder fused node due to different sizes" ) return self_sizes = snode._sizes[0] new_order = None assert self_sizes is not None if len(self_sizes) == self_dep.num_vars == other_dep.num_vars: new_order = self_dep.decide_loop_order_to_match(other_dep) if not new_order: loop_ordering_log.debug( "Dont reordering fused node %s because we can not decide the suitable loop order", self.get_name(), ) return metrics.num_loop_reordering += 1 loop_ordering_log.debug( "Reorder loops for fused node %s with order %s", self.get_name(), new_order ) for snode in self.snodes: assert isinstance(snode, SchedulerNode) snode.apply_new_loop_order(new_order) refresh_group_node_dependencies(self) def __init__(self, scheduler: Scheduler, snodes: list[BaseSchedulerNode]) -> None: super().__init__(scheduler) init_group_node(self, scheduler, snodes) self.users: list[NodeUser] = [] self.group = max(snodes, key=lambda x: int(x.is_reduction())).group @cache_on_self def get_name(self) -> str: return "_".join([x.get_name() for x in self.snodes]) def get_first_name(self) -> str: return self.snodes[0].get_name() @cache_on_self def get_buffer_names(self) -> OrderedSet[str]: return OrderedSet.union(*[x.get_buffer_names() for x in self.snodes]) def get_outputs(self) -> list[SchedulerBuffer]: result: list[SchedulerBuffer] = [] for node in self.snodes: result.extend(node.get_outputs()) return result def debug_str_extra(self) -> str: lines = [ f"{self.get_name()}.snodes[{i}] =\n{node.debug_str()}" for i, node in enumerate(self.snodes) ] node = self.snodes[0].node if node is not None: lines.extend(self._debug_str_for_device()) return textwrap.indent("\n".join(lines).rstrip(), " ") def debug_str_short(self) -> str: snodes_str = [node.debug_str_short() for node in self.snodes] return f"{self}, snodes: {snodes_str}" def set_last_usage( self, future_used_buffers: OrderedSet[str], mutation_real_name: dict[str, str] ) -> None: # Set self.last_usage using the global information # This will be used for inter-kernel optimisations super().set_last_usage(future_used_buffers, mutation_real_name) # Set self.last_usage on the snodes # This will be used for optimisations within the kernel future_used_buffers = OrderedSet[str]() for node in reversed(self.snodes): node.set_last_usage(future_used_buffers, mutation_real_name) future_used_buffers.update(node.last_usage) @cache_on_self def used_buffer_names(self) -> OrderedSet[str]: return OrderedSet.union(*[x.used_buffer_names() for x in self.snodes]) @cache_on_self def used_or_aliased_buffer_names(self) -> OrderedSet[str]: return OrderedSet.union( *[x.used_or_aliased_buffer_names() for x in self.snodes] ) def get_nodes(self) -> Sequence[BaseSchedulerNode]: return self.snodes def __repr__(self) -> str: return f"{type(self).__name__}(nodes={self.get_name()})" @cache_on_self def is_reduction(self) -> bool: return any(x.is_reduction() for x in self.snodes) @cache_on_self def is_split_scan(self) -> bool: return any(x.is_split_scan() for x in self.snodes) @cache_on_self def is_template(self) -> bool: return any(x.is_template() for x in self.snodes) @cache_on_self def get_template_node(self) -> Optional[ir.TemplateBuffer]: for node in self.snodes: if node.is_template(): return node.get_template_node() return None def get_device(self) -> torch.device: return self.group[0] @cache_on_self def has_aliasing_or_mutation(self) -> bool: return any(x.has_aliasing_or_mutation() for x in self.snodes) # None of these need to be implemented, as a FusedSchedulerNode is just an # abstraction for scheduling purposes def update_mutated_names(self, renames: dict[str, str]) -> None: raise NotImplementedError def add_fake_dep(self, name: Dep) -> None: raise NotImplementedError def can_inplace(self, read_dep: dependencies.Dep) -> bool: raise NotImplementedError def debug_str(self) -> str: """Longer form printout for trace logs""" name = self.get_name() node_typestr = ",".join(type(n).__name__ for n in self.snodes) buf = IndentedBuffer() buf.splice( f"""\ {name}: {type(self).__name__}({node_typestr}) {name}.writes = {pformat(self.read_writes.writes)} {name}.unmet_dependencies = {pformat(self.unmet_dependencies)} {name}.met_dependencies = {pformat(self.read_writes.reads - self.unmet_dependencies)} {name}.outputs = [ """ ) with buf.indent(): for out in self.get_outputs(): buf.splice(out.debug_str()) buf.writeline("]") try: buf.splice(self.debug_str_extra()) except Exception: log.warning("Ignoring error in debug_str()", exc_info=True) return buf.getrawvalue().rstrip() class ForeachKernelSchedulerNode(FusedSchedulerNode): """ This is a schedular node that consists of a set of scheduler nodes that has no data dependencies among them and can be executed in parallel. """ def get_consumer_subnode_for( self, producer: BaseSchedulerNode ) -> Optional[BaseSchedulerNode]: for buf in producer.get_outputs(): if buf.get_name() in self.read_to_node: return self.read_to_node[buf.get_name()] return None def get_producer_subnode_for( self, consumer: BaseSchedulerNode ) -> Optional[BaseSchedulerNode]: producers = OrderedSet[BaseSchedulerNode]() for rd in consumer.read_writes.reads: if rd.name not in self.scheduler.name_to_buf: continue node_name = self.scheduler.name_to_buf[rd.name].defining_op_name() if node_name in self.name_to_node: producers.add(self.name_to_node[node_name]) # Don't permit fusion if there are multiple subnodes # that this consumer reads from if len(producers) == 1: return next(iter(producers)) else: return None @classmethod def can_fuse(cls, producer: BaseSchedulerNode, consumer: BaseSchedulerNode) -> bool: why = WhyNoFuse(producer, consumer) if producer.is_foreach() and consumer.is_foreach(): producer = typing.cast(ForeachKernelSchedulerNode, producer) consumer = typing.cast(ForeachKernelSchedulerNode, consumer) foreach_match = len(producer.snodes) == len(consumer.snodes) if not foreach_match: why("foreach do not have same length") return foreach_match and all( producer.scheduler.can_fuse(l, r) for l, r in zip(producer.snodes, consumer.snodes) ) elif consumer.is_foreach(): if producer.is_reduction(): why( "candidate producer is a reduction, foreach ops cannot be fused with reductions currently" ) return False consumer = typing.cast(ForeachKernelSchedulerNode, consumer) consumer_subnode = consumer.get_consumer_subnode_for(producer) if consumer_subnode is not None: return consumer.scheduler.can_fuse(producer, consumer_subnode) why("candidate producer is not dep of any foreach consumer") return False elif producer.is_foreach(): if consumer.is_reduction(): why( "candidate consumer is a reduction, foreach ops cannot be fused with reductions currently" ) return False producer = typing.cast(ForeachKernelSchedulerNode, producer) producer_subnode = producer.get_producer_subnode_for(consumer) if producer_subnode is not None: return producer.scheduler.can_fuse(producer_subnode, consumer) why("candidate consumer has no dep in any foreach producer") return False raise AssertionError( "At least one node passed to ForeachKernelSchedulerNode.can_fuse should be a foreach node" ) @classmethod def fuse( cls, producer: BaseSchedulerNode, consumer: BaseSchedulerNode ) -> ForeachKernelSchedulerNode: assert producer.is_foreach() or consumer.is_foreach() if producer.is_foreach(): producer = typing.cast(ForeachKernelSchedulerNode, producer) use_custom_partition_algo = producer.use_custom_partition_algo enable_autotune = producer.enable_autotune else: consumer = typing.cast(ForeachKernelSchedulerNode, consumer) use_custom_partition_algo = consumer.use_custom_partition_algo enable_autotune = consumer.enable_autotune prev_node_1 = None prev_node_2 = None fused_nodes: list[BaseSchedulerNode] if producer.is_foreach() and consumer.is_foreach(): producer = typing.cast(ForeachKernelSchedulerNode, producer) consumer = typing.cast(ForeachKernelSchedulerNode, consumer) fused_nodes = [ FusedSchedulerNode.fuse(l, r) for l, r in zip(producer.snodes, consumer.snodes) ] elif producer.is_foreach(): producer = typing.cast(ForeachKernelSchedulerNode, producer) producer_subnode = producer.get_producer_subnode_for(consumer) fused_nodes = [] prev_node_1 = producer prev_node_2 = None for node in producer.snodes: if node is producer_subnode: new_node = FusedSchedulerNode.fuse(node, consumer) prev_node_2 = new_node fused_nodes.append(new_node) else: fused_nodes.append(node) elif consumer.is_foreach(): consumer = typing.cast(ForeachKernelSchedulerNode, consumer) consumer_subnode = consumer.get_consumer_subnode_for(producer) fused_nodes = [] prev_node_1 = consumer prev_node_2 = None for node in consumer.snodes: if node is consumer_subnode: new_node = FusedSchedulerNode.fuse(producer, node) prev_node_2 = new_node fused_nodes.append(new_node) else: fused_nodes.append(node) else: raise AssertionError( "At least one node passed to ForeachKernelSchedulerNode.fuse should be a foreach node" ) return cls( producer.scheduler, fused_nodes, use_custom_partition_algo=use_custom_partition_algo, prev_node_1=prev_node_1, prev_node_2=prev_node_2, enable_autotune=enable_autotune, ) def __init__( self, scheduler: Scheduler, snodes: list[BaseSchedulerNode], use_custom_partition_algo: bool, prev_node_1: Optional[BaseSchedulerNode] = None, prev_node_2: Optional[BaseSchedulerNode] = None, enable_autotune: bool = False, ) -> None: self.read_to_node = {} self.name_to_node = {} if prev_node_1 is None or prev_node_2 is None: super().__init__(scheduler, snodes) for node in snodes: for read in node.read_writes.reads: self.read_to_node[read.name] = node for name in node.get_operation_names(): self.name_to_node[name] = node else: self.scheduler = scheduler self.snodes = snodes self.node = None self.users: list[NodeUser] = [] self.set_read_writes( dependencies.ReadWrites.merge_list( [prev_node_1.read_writes, prev_node_2.read_writes] ) ) self.unmet_dependencies = ( OrderedSet( dep for dep in OrderedSet.union( prev_node_1.unmet_dependencies, prev_node_2.unmet_dependencies ) if dep.name not in self.get_buffer_names() ) - self.read_writes.writes ) self.min_order = min([prev_node_1.min_order, prev_node_2.min_order]) self.max_order = max([prev_node_1.max_order, prev_node_2.max_order]) if prev_node_1.is_foreach(): assert isinstance(prev_node_1, ForeachKernelSchedulerNode) foreach_node, other_node = prev_node_1, prev_node_2 else: assert isinstance(prev_node_2, ForeachKernelSchedulerNode) foreach_node, other_node = prev_node_2, prev_node_1 self.ancestors = foreach_node.ancestors self.ancestors.update(other_node.ancestors) self.name_to_node = foreach_node.name_to_node for name in other_node.get_operation_names(): self.name_to_node[name] = other_node self.use_custom_partition_algo = use_custom_partition_algo device = snodes[0].get_device() assert device self.group = (device, ((sympy.Expr("combo_kernel"),),)) self.origins = OrderedSet[torch.fx.Node]() self.enable_autotune = enable_autotune @classmethod def combinable_nodes( cls, nodes: list[BaseSchedulerNode] ) -> list[BaseSchedulerNode]: extern = [x for x in nodes if isinstance(x, ExternKernelSchedulerNode)] if extern: log.debug( "ComboKernels: %d external nodes are filtered %s", len(extern), [node.node.get_origins() for node in extern if node.node is not None], ) filtered_nodes = [ x for x in nodes if not isinstance(x, (NopKernelSchedulerNode, ExternKernelSchedulerNode)) ] foreach_nodes = [ x for x in filtered_nodes if isinstance(x, ForeachKernelSchedulerNode) ] if foreach_nodes: log.debug("ComboKernels: %d foreach nodes are filtered", len(foreach_nodes)) filtered_nodes = [ x for x in filtered_nodes if not isinstance(x, ForeachKernelSchedulerNode) ] template_nodes = [x for x in filtered_nodes if x.is_template()] if template_nodes: log.debug( "ComboKernels: %d template nodes are filtered", OrderedSet([len(template_nodes)]), ) filtered_nodes = [x for x in filtered_nodes if x not in template_nodes] return filtered_nodes @staticmethod def _default_group_nodes_for_combo_kernels( scheduler: Scheduler, ) -> list[list[BaseSchedulerNode]]: """ Returns a list of lists of nodes that are to be grouped together. """ sorted_nodes = scheduler._topological_sort_nodes() grouped_nodes = [] max_num_nodes = 8 for nodes in sorted_nodes: grouped_nodes.extend( [ nodes[i : i + max_num_nodes] for i in range(0, len(nodes), max_num_nodes) ] ) return grouped_nodes group_algorithm_for_combo_kernels: Callable[ [Scheduler], list[list[BaseSchedulerNode]] ] = _default_group_nodes_for_combo_kernels @staticmethod def set_group_algorithm_for_combo_kernels( custom_group_algorithm: Callable[[Scheduler], list[list[BaseSchedulerNode]]], ) -> None: ForeachKernelSchedulerNode.group_algorithm_for_combo_kernels = ( custom_group_algorithm ) @staticmethod def group_nodes_for_combo_kernels( scheduler: Scheduler, ) -> list[list[BaseSchedulerNode]]: return ForeachKernelSchedulerNode.group_algorithm_for_combo_kernels(scheduler) def mark_run(self) -> None: raise NotImplementedError def codegen(self) -> None: raise NotImplementedError def is_foreach(self) -> bool: return True def get_subkernel_nodes(self) -> list[BaseSchedulerNode]: """Returns a list of nodes which comprise the combo kernel. These nodes may be vertically fused.""" return list(self.snodes) def get_nodes(self) -> Sequence[BaseSchedulerNode]: """Returns all nodes contained in this kernel, unpacking fused nodes into their constituent scheduler nodes.""" return list(itertools.chain.from_iterable(x.get_nodes() for x in self.snodes)) def get_first_name(self) -> str: return self.snodes[0].get_first_name() def prune_redundant_deps( self, name_to_fused_node: dict[str, BaseSchedulerNode] ) -> None: _prune_redundant_deps(self, name_to_fused_node, self.scheduler.name_to_buf) for node in self.snodes: node.prune_redundant_deps(name_to_fused_node) class GroupedSchedulerNode(BaseSchedulerNode): """ This is a "fake" scheduler node that represents a group of scheduler nodes that are meant to be *grouped* together (it does not allow another node to be scheduled in between its constituent nodes, nor does it allow another node to fuse into any of its constituent nodes). The way it does this is by maintaining its unmet dependencies as the union of its constituent nodes. Fusion will still happen among the nodes within each GroupedSchedulerNode. At codegen time, this scheduler node will be unpacked and codegen is called on each constituent node. """ snodes: list[BaseSchedulerNode] @classmethod def create(cls, snodes: list[BaseSchedulerNode]) -> GroupedSchedulerNode: scheduler = snodes[0].scheduler assert all(node.scheduler is scheduler for node in snodes) grouped_snode = cls(scheduler, snodes) for snode in snodes: scheduler.name_to_fused_node[snode.get_name()] = grouped_snode scheduler.name_to_fused_node[grouped_snode.get_name()] = grouped_snode return grouped_snode def __init__(self, scheduler: Scheduler, snodes: list[BaseSchedulerNode]) -> None: super().__init__(scheduler) init_group_node(self, scheduler, snodes) def unpack(self) -> list[BaseSchedulerNode]: """ Do fusion among nodes within this GroupedSchedulerNode, and then unpack this GroupedSchedulerNode into regular nodes. """ for snode in self.snodes: self.scheduler.name_to_fused_node[snode.get_name()] = snode del self.scheduler.name_to_fused_node[self.get_name()] return self.scheduler.fuse_nodes(self.snodes) def add_fake_dep(self, fake_dep: Dep) -> None: self.set_read_writes(self.read_writes.with_read(fake_dep)) self.unmet_dependencies.add(fake_dep) @cache_on_self def get_name(self) -> str: return "_".join([x.get_name() for x in self.snodes]) def get_first_name(self) -> str: return self.snodes[0].get_name() @cache_on_self def get_buffer_names(self) -> OrderedSet[str]: return OrderedSet.union(*[x.get_buffer_names() for x in self.snodes]) def get_outputs(self) -> list[SchedulerBuffer]: result: list[SchedulerBuffer] = [] for node in self.snodes: result.extend(node.get_outputs()) return result def get_nodes(self) -> Sequence[BaseSchedulerNode]: return self.snodes @classmethod def can_fuse(cls, producer: BaseSchedulerNode, consumer: BaseSchedulerNode) -> bool: # GroupedSchedulerNode cannot be fused with another node return False def pick_loop_order( stride_lengths: list[list[int]], sizes: Sequence[sympy.Expr], priority_idx: tuple[int, ...] = (), ) -> list[int]: """ A heuristic to decide loop iteration orders. This has not been well tuned and may be something we should autotune. """ @functools.cmp_to_key def index_cmp(a: int, b: int) -> int: if sizes[a] == 1 or sizes[b] == 1: # 1-sizes don't matter, just move them to the end return cmp(sizes[a] == 1, sizes[b] == 1) # Take abs, otherwise flipped dimensions are treated as smaller # strides than contiguous dims stride_len_a = [abs(sl[a]) for sl in stride_lengths] stride_len_b = [abs(sl[b]) for sl in stride_lengths] # equivalent to # np.logical_or(stride_lengths[:, b] == 0, stride_lengths[:, a] < stride_lengths[:, b]).all() a_first = sum( sl_b == 0 or sl_a < sl_b for sl_a, sl_b in zip(stride_len_a, stride_len_b) ) b_first = sum( sl_a == 0 or sl_b < sl_a for sl_a, sl_b in zip(stride_len_a, stride_len_b) ) if a_first > b_first: return -1 if b_first > a_first: return 1 # otherwise contiguous return cmp(b, a) order = list(reversed(range(len(stride_lengths[0])))) if len(priority_idx) > 0: # if we have priority node, only use that node's order stride_lengths = [stride_lengths[pi] for pi in priority_idx] if config.pick_loop_orders: order.sort(key=index_cmp) return order @dataclasses.dataclass class NodeUser: node: Union[BaseSchedulerNode, OutputNode] can_inplace: bool = False # A weak user must be scheduled after a given node, but doesn't actually # use the result is_weak: bool = False def __hash__(self) -> int: return hash((self.node.get_name(), self.can_inplace, self.is_weak)) def __eq__(self, other: object) -> bool: return ( isinstance(other, NodeUser) and self.get_name() == other.get_name() and self.can_inplace == other.can_inplace and self.is_weak == other.is_weak ) def get_name(self) -> str: return self.node.get_name() def merge(self, other: NodeUser) -> NodeUser: assert self.node is other.node return NodeUser( self.node, self.can_inplace and other.can_inplace, self.is_weak and other.is_weak, ) _post_grad_graph_counter = itertools.count() class Scheduler: __dep_size_hint_cache: dict[Dep, int] def __init__(self, nodes: list[ir.Operation]) -> None: with dynamo_timed("Scheduler.__init__"): self._init(nodes) def _init(self, nodes: list[ir.Operation]) -> None: super().__init__() self.__dep_size_hint_cache = {} V.graph.scheduler = self self.backends: dict[torch.device, BaseScheduling] = {} self.post_grad_graph_id = next(_post_grad_graph_counter) self._graph_partition_counter = itertools.count() self.completed_operations = OrderedSet[str]() self.available_buffer_names = OrderedSet( [ *V.graph.graph_inputs.keys(), *V.graph.constants.keys(), *V.graph.torchbind_constants.keys(), ] ) self.nodes = [self.create_scheduler_node(n) for n in nodes] self.update_zero_dim_cpu_tensor() # some new constants could have been created above self.available_buffer_names.update(V.graph.constants.keys()) for node in self.nodes: node.prune_deps() self.name_to_donated_buffer: dict[str, SchedulerDonatedBuffer] = ( self.get_donated_buffers() ) self.name_to_node: dict[str, BaseSchedulerNode] = { n.get_name(): n for n in self.nodes } self.name_to_buf: dict[str, SchedulerBuffer] = { buf.get_name(): buf for node in self.nodes for buf in node.get_outputs() } self.name_to_fused_node: dict[str, BaseSchedulerNode] = self.name_to_node.copy() # mutation_real_name: Maps back to the original name for codegen # Example: # If you mutate buf0 inside of buf1's kernel, then: # mutation_real_name = {"buf0" : "buf1"} # all subsequent uses of buf0 become buf1's usage in dependency graph self.mutation_real_name: dict[str, str] = {} # We handle mutation by renaming modified versions of the same # buffer in the dependency graph to prevent cycles. # mutation_renames: tracks the current name for a given buffer # (changed once per mutation) # Example: # If you mutate buf0 inside of buf1's kernel, then: # mutation_renames = {"buf1" : "buf0"} # in codegen we only use buf0, never buf1 self.mutation_renames: dict[str, str] = {} # Must run first to correctly set dependencies, before all other passes that rely on # reading from .read_writes.reads or .unmet_dependencies self.nodes = comms.decide_global_ordering_of_comms( self.nodes, self.name_to_buf, self.name_to_fused_node, ) self.compute_dependencies() self.nodes = self.topological_sort_schedule(self.nodes) self.dead_node_elimination() self.name_to_fused_node = {n.get_name(): n for n in self.nodes} self.compute_ancestors() metrics.ir_nodes_pre_fusion += len(self.nodes) from torch._inductor.debug import log_ir_post_fusion, log_ir_pre_fusion log_ir_pre_fusion(self.nodes) self.num_orig_nodes = len(self.nodes) self.create_foreach_nodes() self.nodes = self.topological_sort_schedule(self.nodes) self.logged_slow_fusion = OrderedSet[tuple[str, str]]() if config._pre_fusion_custom_pass is not None: self.nodes = config._pre_fusion_custom_pass(self.nodes) self.nodes = self.fuse_nodes(self.nodes) self.merge_loops() self.finalize_multi_template_buffers() if config.combo_kernels: self.create_combo_kernel_nodes(num_ck_nodes=None) # Peak memory pass and overlap pass must run last, otherwise # other reordering passes could undo their effects. if config.reorder_for_peak_memory: from .memory import reorder_for_peak_memory self.nodes = reorder_for_peak_memory( self.nodes, self.name_to_buf, self.name_to_fused_node, OrderedSet(V.graph.graph_inputs.keys()), OrderedSet(V.graph.get_output_names()), ) if config.reorder_for_compute_comm_overlap: self.nodes = comms.reorder_compute_and_comm_for_overlap(self.nodes) self.process_grouped_nodes() self.compute_last_usage() log_ir_post_fusion(self.nodes) V.debug.graph_diagram(self.nodes) self.debug_draw_graph() # used during codegen: self.buffer_names_to_free = OrderedSet[str]() # fx graph node to the position it appears in the graph # for debug attribution self.origin_to_index: dict[torch.fx.Node, int] = {} get_metric_table("graph_stats").add_row( lambda: { "graph_id": self.post_grad_graph_id, "num_nodes_before_fusion": self.num_orig_nodes, "num_nodes_after_fusion": len(self.nodes), } ) def get_donated_buffers(self) -> dict[str, SchedulerDonatedBuffer]: name_to_donated_buf = {} for name in V.graph.graph_inputs_original: if isinstance(V.graph.graph_inputs_original[name], ir.DonatedBuffer): name_to_donated_buf[name] = SchedulerDonatedBuffer( self, V.graph.graph_inputs_original[name], defining_op=None, ) return name_to_donated_buf @property def current_device(self) -> Optional[torch.device]: return V.graph.current_device @current_device.setter def current_device(self, device: Optional[torch.device]) -> None: V.graph.current_device = device def debug_draw_graph(self) -> None: """Generate an image of the graph for debugging""" if os.environ.get("INDUCTOR_WRITE_SCHEDULER_GRAPH", None) == "1": from .debug import draw_buffers draw_buffers(self.nodes, print_graph=True) def debug_print_nodes(self, label: str) -> None: if log.isEnabledFor(logging.INFO): log.info("%s:", label) for node in self.nodes: node.log_details() def create_scheduler_node(self, node: ir.Operation) -> BaseSchedulerNode: assert node.get_origins() is not None, ( "All nodes passed to scheduling must have an origin" ) if node.is_no_op(): return NopKernelSchedulerNode(self, node) elif isinstance(node, (ir.ComputedBuffer, ir.TemplateBuffer)): return SchedulerNode(self, node) elif isinstance(node, ir.ExternKernel): return ExternKernelSchedulerNode(self, node) else: raise NotImplementedError(node) def create_foreach_nodes(self) -> None: removed_node_names = OrderedSet[str]() fe_nodes = [] kept_node_names = self.name_to_fused_node.keys() for names in V.graph.lists.values(): names = [ name for name in names if name in kept_node_names and not isinstance(self.name_to_node[name], NopKernelSchedulerNode) ] if not names: # All nodes eliminated continue removed_node_names.update(names) snodes = [self.name_to_node[name] for name in names] enable_autotune = config.combo_kernels_autotune > 1 fe_node = ForeachKernelSchedulerNode( self, snodes, use_custom_partition_algo=False, enable_autotune=enable_autotune, ) fe_nodes.append(fe_node) for name in names: self.name_to_fused_node[name] = fe_node self.nodes = [ node for node in self.nodes if node.get_name() not in removed_node_names ] + list(fe_nodes) def compute_dependencies(self) -> None: """ Create dependency edges between nodes, handling aliasing and mutation properly. """ T = TypeVar("T") class DedupList(Generic[T]): """ This data structure behaves like a list except it makes sure the elements remain unique. Normally one could use a OrderedSet/dict for this purpose however the list in question gets elements appended as it is being iterated over which means that we need to keep the list semantics. """ def __init__( self, items: Optional[list[T]] = None, membership: Optional[OrderedSet[T]] = None, ) -> None: self.items = items or [] self.membership = membership or OrderedSet() def append(self, node_user: T) -> None: if node_user in self.membership: return self.items.append(node_user) self.membership.add(node_user) def __add__(self, other: DedupList[T]) -> DedupList[T]: new_membership = OrderedSet.union(self.membership, other.membership) new_items = self.items + [ x for x in other.items if x not in self.membership ] return DedupList(new_items, new_membership) name_to_users: defaultdict[str, DedupList[NodeUser]] = collections.defaultdict( DedupList ) # handle aliasing by using python aliasing in name_to_users # if foo aliases bar then we will make name_to_users["foo"] point # to the same python list as name_to_users["bar"] for node in self.nodes: for buf1 in node.get_outputs(): buf1_name = buf1.get_name() for buf2_name in buf1.get_aliases(): if buf1_name in name_to_users and buf2_name in name_to_users: # merge the two list1 = name_to_users[buf1_name] list2 = name_to_users[buf2_name] combined = list1 + list2 for key in name_to_users.keys(): if ( name_to_users[key] is list1 or name_to_users[key] is list2 ): name_to_users[key] = combined elif buf1_name in name_to_users: name_to_users[buf2_name] = name_to_users[buf1_name] else: name_to_users[buf1_name] = name_to_users[buf2_name] def rename(n: str) -> str: if n in self.mutation_renames: return rename(self.mutation_renames[n]) return n def add_user( used_by_name: str, user_node: Union[BaseSchedulerNode, OutputNode], can_inplace: bool = False, is_weak: bool = False, ) -> None: name_to_users[rename(used_by_name)].append( NodeUser(user_node, can_inplace, is_weak) ) unbacked_symbol_to_origin_node: dict[sympy.Symbol, Optional[str]] = {} # NB: None means that the dependency is on an input. Don't actually # generate a dependency because if we do, Inductor will start trying # to free the unbacked int but that's pointless for name, val in V.graph.graph_inputs.items(): if isinstance(val, sympy.Expr): for fs in val.free_symbols: unbacked_symbol_to_origin_node[fs] = None for node in self.nodes: log.debug("scheduling %s", node.node) # unbacked symbols don't follow ordinary buffer dependencies, so # we track their def/uses separately assert node.node is not None unbacked_symbol_defs = sorted( node.node.get_unbacked_symbol_defs(), key=lambda x: x.name ) for s in unbacked_symbol_defs: assert isinstance(s, sympy.Symbol) # Pick the first definer as canonical. There may be multiple # because if a MultiOutputLayout buffer propagates an unbacked # symint to multiple outputs, they will all claim to def it. if s not in unbacked_symbol_to_origin_node: unbacked_symbol_to_origin_node[s] = node.get_name() unbacked_symbol_uses = sorted( node.node.get_unbacked_symbol_uses(), key=lambda x: x.name ) # if a kernel takes unbacked symints, register dependencies for s in unbacked_symbol_uses: assert s in unbacked_symbol_to_origin_node, ( f"{s} not in {unbacked_symbol_to_origin_node}" ) if (r := unbacked_symbol_to_origin_node[s]) is not None: for buf in self.name_to_node[r].get_outputs(): node.add_fake_dep(StarDep(buf.get_name())) if ( len(node.read_writes.writes) == 1 and (dep := next(iter(node.read_writes.writes))) and isinstance(dep, MemoryDep) ): node_mode = dep.mode else: node_mode = None # Handle output mutations for buf in node.get_outputs(): # a node will mutate either 0 or 1 buffers assert len(buf.get_mutations()) <= 1 for alt_name in buf.get_mutations(): alt_name = rename(alt_name) # this node must run after the prior writer add_user(alt_name, node) node.add_fake_dep(StarDep(alt_name, mode=node_mode)) for user in name_to_users[alt_name].items: if user.get_name() == node.get_name(): continue assert isinstance(user.node, BaseSchedulerNode) for other_name in user.node.get_buffer_names(): # this node must run after all prior readers other_name = rename(other_name) node.add_fake_dep( WeakDep(other_name, mutating_buf=buf.get_name()) ) add_user(other_name, node, is_weak=True) # add normal non-mutation dependencies for read in node.read_writes.reads: if not isinstance(read, WeakDep): add_user(read.name, node, node.can_inplace(read)) node.update_mutated_names(self.mutation_renames) # update our renaming scheme for the next iteration for buf in node.get_outputs(): for alt_name in buf.get_mutations(): self.mutation_renames[rename(alt_name)] = buf.get_name() self.mutation_renames[alt_name] = buf.get_name() self.mutation_real_name[buf.get_name()] = ( self.mutation_real_name.get(alt_name, alt_name) ) # make sure outputs aren't dead-code-eliminated for buf_name in V.graph.get_output_names(): log.debug("scheduling output %s", buf_name) add_user(buf_name, OutputNode(StarDep(buf_name))) # make sure unbacked symints aren't dead-code-eliminated for out in V.graph.graph_outputs: for s in out.get_unbacked_symbol_uses(): assert s in unbacked_symbol_to_origin_node, ( f"{s} not in {unbacked_symbol_to_origin_node.keys()}" ) if r := unbacked_symbol_to_origin_node[s]: for buf_name in self.name_to_node[r].get_buffer_names(): log.debug( "scheduling output %s for unbacked symint %s", buf_name, s ) add_user(buf_name, OutputNode(StarDep(buf_name))) # make sure input mutation isn't dead-code-eliminated for name in self.mutation_renames: if name in V.graph.graph_inputs: add_user(name, OutputNode(StarDep(name))) V.graph.mutated_inputs.add(name) elif name in V.graph.constants: # In AOTI, module parameters and buffers are not lifted as graph inputs add_user(name, OutputNode(StarDep(name))) inp_names = { name: index for index, name in enumerate(V.graph.graph_inputs.keys()) } V.graph.mutated_input_idxs = [ inp_names[name] for name in V.graph.mutated_inputs ] # copy users information onto the nodes for node in self.nodes: for buf in node.get_outputs(): buf.set_users(name_to_users[buf.get_name()].items) for name in self.name_to_donated_buffer: self.name_to_donated_buffer[name].set_users(name_to_users[name].items) def dead_node_elimination(self) -> None: """ Remove any nodes without users """ # self.nodes is in topological order, so by iterating in reverse order # we have visited (and potentially removed) all users before visiting a # given node. updated_nodes = [] for node in reversed(self.nodes): def can_eliminate_user(user: NodeUser) -> bool: return user.is_weak or user.get_name() in V.graph.removed_operations active_buffers = False for buf in node.get_outputs(): can_eliminate = all(can_eliminate_user(u) for u in buf.users) if can_eliminate: log.debug("removed dead buffer: %s", buf.get_name()) V.graph.removed_buffers.add(buf.get_name()) else: active_buffers = True can_eliminate = not node.has_side_effects() and not active_buffers if not can_eliminate: updated_nodes.append(node) else: # dead code log.debug("removed dead operation: %s", node.get_name()) V.graph.removed_operations.add(node.get_name()) for read in node.read_writes.reads: if read.name in self.name_to_buf: users = self.name_to_buf[read.name].users self.name_to_buf[read.name].users = [ u for u in users if u.node.get_name() != node.get_name() ] self.nodes = list(reversed(updated_nodes)) # Prune any WeakDeps no longer needed for node in self.nodes: node.prune_weak_deps() def topological_sort_schedule( self, nodes: list[BaseSchedulerNode] ) -> list[BaseSchedulerNode]: """ Ensure nodes is in topologically sorted order """ seen = OrderedSet[BaseSchedulerNode]() name_to_node: dict[str, BaseSchedulerNode] = dict() result: list[BaseSchedulerNode] = [] def visit(n: BaseSchedulerNode) -> None: if n not in seen: seen.add(n) for dep in sorted(n.unmet_dependencies, key=lambda d: d.name): # We only care about doing toposort within `nodes` if dep.name not in name_to_node: continue visit(name_to_node[dep.name]) result.append(n) for node in nodes: for name in node.get_buffer_names(): name_to_node[name] = node for node in nodes: visit(node) return result def _get_unmet_dep_nodes(self, snode: BaseSchedulerNode) -> list[BaseSchedulerNode]: unmet_deps = OrderedSet[str]() if isinstance( snode, ( SchedulerNode, ExternKernelSchedulerNode, NopKernelSchedulerNode, FusedSchedulerNode, ), ): for dep in snode.unmet_dependencies: unmet_deps.add(dep.name) else: raise RuntimeError( f"get_unmet_dep_nodes is not implemented for {type(snode)}." ) unmet_dep_ops = (self.name_to_buf[dep].defining_op_name() for dep in unmet_deps) return list(OrderedSet(self.name_to_fused_node[n] for n in unmet_dep_ops)) def _topological_sort_nodes(self) -> list[list[BaseSchedulerNode]]: """ Sort nodes by their topological order, return a list of node lists. """ order = [] nodes = dict.fromkeys(self.nodes, 0) children: dict[Any, Any] = {} for node in self.nodes: deps = self._get_unmet_dep_nodes(node) nodes[node] = len(deps) for dep in deps: c = children.get(dep, []) c.append(node) children[dep] = c zero_deg_nodes = [n for n, v in nodes.items() if v == 0] while zero_deg_nodes: order.append(zero_deg_nodes) for n in zero_deg_nodes: for user in children.get(n, []): nodes[user] -= 1 nodes.pop(n) zero_deg_nodes = [n for n, v in nodes.items() if v == 0] assert not nodes, "Topological sort failed!" return order def compute_ancestors(self) -> None: """ Populate each node.ancestors """ # note self.nodes is topologically sorted name_to_ancestors: dict[str, OrderedSet[str]] = {} for node in self.nodes: ancestors = OrderedSet[str]() for dep in node.unmet_dependencies: dep_node_name = self.name_to_buf[dep.name].defining_op_name() ancestors.add(dep_node_name) ancestors |= name_to_ancestors[dep_node_name] name_to_ancestors[node.get_name()] = ancestors node.ancestors = ancestors for order, node in enumerate(self.nodes): node.min_order = order node.max_order = order def merge_loops(self) -> None: for node in self.nodes: if not config.loop_ordering_after_fusion: continue # Even for CPU, if we are using the halide backend, we still need # the merge loops steps below if not isinstance(node, (SchedulerNode, FusedSchedulerNode)) or ( not node.is_gpu() and config.cpu_backend != "halide" ): continue for snode in node.get_nodes(): # merge loops for the scheduler node if not isinstance(snode, SchedulerNode) or snode.is_template(): continue snode.merge_loops() # Note that for CPU backend, merging loops will change # snode.group. It's fine for Triton backend. # But if we simplify update snode.group like this: # group_fn = self.get_backend(snode.node.get_device()).group_fn # snode.group = (snode.node.get_device(), group_fn(snode._sizes)) # There is still an issue due to different snode in a # FusedSchedulerNode having different merged loops. # Skip CPU backend for now. def fuse_nodes(self, nodes: list[BaseSchedulerNode]) -> list[BaseSchedulerNode]: """ Combine eligible nodes into FusedSchedulerNodes. """ with dynamo_timed("Scheduler.fused_nodes"): for i in range(10): old_len = len(nodes) fusion_log.debug( "===== attempting fusion (%d/10): %d nodes =====", i + 1, old_len, ) nodes = self.fuse_nodes_once(nodes) new_len = len(nodes) fusion_log.debug( "completed fusion round (%d/10): fused %d nodes into %d nodes\n", i + 1, old_len, new_len, ) if new_len == old_len or new_len == 1: fusion_log.debug( "===== fusion complete (%d iterations) =====", i + 1 ) break return nodes def process_grouped_nodes(self) -> None: """ Unpack GroupedSchedulerNode into regular nodes. """ new_nodes: list[BaseSchedulerNode] = [] for node in self.nodes: new_nodes.extend( node.unpack() if isinstance(node, GroupedSchedulerNode) else [node] ) self.nodes = new_nodes def benchmark_fused_nodes( self, nodes: Sequence[BaseSchedulerNode] ) -> tuple[float, str]: """ Benchmark fused list of nodes and return the execution time in milliseconds on randomly generated inputs. """ assert len(nodes) > 0 device = nodes[0].get_device() self.current_device = device backend = self.get_backend(device) with dynamo_timed( "benchmark_fused_nodes", log_pt2_compile_event=True, dynamo_compile_column_us="compile_time_autotune_time_us", ): return backend.benchmark_fused_nodes(nodes) def generate_kernel_code_from_nodes( self, nodes: Sequence[BaseSchedulerNode], benchmark_kernel: bool ) -> str: """ Benchmark fused list of nodes and return the execution time in milliseconds on randomly generated inputs. """ assert len(nodes) > 0 device = nodes[0].get_device() self.current_device = device backend = self.get_backend(device) with dynamo_timed("benchmark_fused_nodes"): return backend.generate_kernel_code_from_nodes(nodes, benchmark_kernel) def benchmark_codegened_module( self, module: ModuleType, device: torch.device ) -> tuple[float, str]: """ Benchmark fused list of nodes and return the execution time in milliseconds on randomly generated inputs. """ self.current_device = device backend = self.get_backend(device) with dynamo_timed("benchmark_fused_nodes"): return backend.benchmark_codegened_module(module) def finalize_multi_template_buffers(self) -> None: def replace_operation_buffer( orig_node: ir.MultiTemplateBuffer, new_node: ir.OperationBuffer ) -> None: replaced_buf_name = new_node.get_name() orig_buf_name = orig_node.get_name() assert isinstance(orig_buf_name, str) and isinstance(replaced_buf_name, str) replaced_op_name = new_node.get_operation_name() orig_op_name = orig_node.get_operation_name() assert isinstance(orig_op_name, str) and isinstance(replaced_op_name, str) del V.graph.name_to_buffer[replaced_buf_name] new_node.name = orig_buf_name del V.graph.name_to_op[replaced_op_name] new_node.operation_name = orig_op_name orig = V.graph.buffers.index(orig_node) V.graph.buffers.remove(new_node) V.graph.buffers[orig] = new_node V.graph.name_to_buffer[orig_buf_name] = new_node orig = V.graph.operations.index(orig_node) V.graph.operations.remove(new_node) V.graph.operations[orig] = new_node V.graph.name_to_op[orig_op_name] = new_node for i, node in enumerate(self.nodes): if isinstance(node, SchedulerNode) and isinstance( node.node, ir.MultiTemplateBuffer ): multi_node = node.node if not config.test_configs.force_extern_kernel_in_multi_template: min_node_unfused, _ = multi_node.get_min_choice() else: min_node_unfused = next( ( timing for timing in multi_node.choice_timings if isinstance( timing, torch._inductor.select_algorithm.ExternKernelCaller, ) ), ) if isinstance( min_node_unfused, torch._inductor.ir.TritonTemplateCallerBase, ): node.node.finalize_as_triton_caller(min_node_unfused) continue out_tensorbox = min_node_unfused.output_node() out_storage = out_tensorbox.data assert isinstance(out_storage, ir.StorageBox) out_buffer = out_storage.data assert isinstance(out_buffer, ir.OperationBuffer) out_buffer.layout = multi_node.layout replace_operation_buffer(multi_node, out_buffer) new_scheduler_node = self.create_scheduler_node(out_buffer) self.nodes[i] = new_scheduler_node self.name_to_node[node.get_name()] = new_scheduler_node self.name_to_fused_node[node.get_name()] = new_scheduler_node for new_out, old_out in zip( new_scheduler_node.get_outputs(), node.get_outputs() ): self.name_to_buf[old_out.get_name()] = new_out new_out.users = old_out.users new_scheduler_node.min_order = node.min_order new_scheduler_node.max_order = node.max_order new_scheduler_node.last_usage = node.last_usage def _any_atomic_add(self, node_list: Sequence[BaseSchedulerNode]) -> bool: return any( hasattr(n.node, "data") and n.node is not None and hasattr(n.node.data, "scatter_mode") and n.node.data.scatter_mode == "atomic_add" for n in node_list ) def speedup_by_fusion( self, node1: BaseSchedulerNode, node2: BaseSchedulerNode ) -> Union[bool, Callable[[], bool]]: """ If config.benchmark_fusion is False, always return True. Otherwise, return True if fusion can brings speedup. """ is_multi_template = any( n.is_template() and isinstance(n.get_template_node(), ir.MultiTemplateBuffer) for n in (node1, node2) ) if not config.benchmark_fusion and not is_multi_template: return True if ( node1.is_template() and not isinstance(node1.get_template_node(), ir.TritonTemplateBuffer) or node1.is_foreach() or node2.is_foreach() ): # TODO support benchmarking epilogue fusion return True node_list_1 = node1.get_nodes() device = node_list_1[0].get_device() assert device # don't support benchmark fusion for CPU right now. if device.type == "cpu": return True node_list_2 = node2.get_nodes() node_list_fused = list(itertools.chain(node_list_1, node_list_2)) # We can not accurately benchmark kernel using atomic_add # due to how we generate random integer inputs. # Skip benchmarking them by allowing fusion. if self._any_atomic_add(node_list_fused): return True from triton.compiler.errors import CompilationError why = WhyNoFuse(node1, node2) device = node_list_fused[0].get_device() assert device is not None def log_fusion(ms_fused: float, ms1: float, ms2: float) -> None: if fusion_log.isEnabledFor(logging.DEBUG): if ms_fused < ms1 + ms2: fusion_log.debug( "can fuse (benchmark): fusing %s with %s cause %sx speedup", node1.get_buffer_names(), node2.get_buffer_names(), green_text(f"{(ms1 + ms2) / ms_fused:.3f}"), ) else: fusion_log.debug( "cannot fuse (benchmark): fusing %s with %s cause %sx slowdown", node1.get_buffer_names(), node2.get_buffer_names(), red_text(f"{ms_fused / (ms1 + ms2):.3f}"), ) async_compile = torch._inductor.async_compile.AsyncCompile() def compile_kernel( nodes: Sequence[BaseSchedulerNode], ) -> tuple[Optional[LambdaFuture], ModuleType]: src_code = self.generate_kernel_code_from_nodes( nodes, benchmark_kernel=True ) mod = PyCodeCache.load(src_code) if not async_compile.use_process_pool(): fut = None else: fut = async_compile.triton(kernel_name="triton_", source_code=src_code) assert isinstance(fut, LambdaFuture) return (fut, mod) # After the succesful fusion with Template, we finalize its config. # Subsequently we benchmark but dont update. Checking for SchedulerNode, instead of FusedSchedulerNode # accomplishes this. if is_multi_template and any( n.get_template_node() is not None for n in (node1, node2) ): epilogue_fusion = node1.get_template_node() is not None multi_node = ( node1.get_template_node() if epilogue_fusion else node2.get_template_node() ) assert isinstance(multi_node, ir.MultiTemplateBuffer) choice_timings = multi_node.choice_timings _, ms1 = multi_node.get_min_choice() # Eagerly compile and benchmark non-template nodes _, ms1 = multi_node.get_min_choice() ms2, path2 = ( self.benchmark_fused_nodes(node_list_2) if epilogue_fusion else self.benchmark_fused_nodes(node_list_1) ) # Start compiling choices in parallel future_choices: list[tuple[Any, Optional[LambdaFuture], ModuleType]] = [] triton_choices = 0 for choice, unfused_time in sorted( choice_timings.items(), key=lambda x: x[1] ): if not isinstance(choice, torch._inductor.ir.TritonTemplateCallerBase): continue # For prologue fusion we check if the underlying template of the choice # supports all allowed prologue inputs. If not, we skip this choice in # the fusion benchmark. # TODO: Remove this check after all Triton templates support prologue fusion. # Currently, persistent+TMA Triton template does not due to the TMA-based loads. if ( not epilogue_fusion and hasattr(choice, "allowed_prologue_inps") and choice.allowed_prologue_inps != multi_node.allowed_prologue_inps ): continue if unfused_time >= ms1 + ms2: break triton_choices += 1 if triton_choices > config.max_epilogue_benchmarked_choices: break with multi_node.swap_as_triton_caller(choice): future_choices.append((choice, *compile_kernel(node_list_fused))) if len(future_choices) == 0: return False def benchmark_when_ready() -> bool: min_ms_fused = float("inf") ms_fused_choice = None new_timings = {} # Benchmark each choice after compilation completes for choice, future, mod_fused in future_choices: try: if future is not None: future.result() # Ideally we would more narrowly catch Exceptions here but # triton will unpredictably error with valid prologue fusions except Exception as e: if fusion_log.isEnabledFor(logging.DEBUG): fusion_log.debug( "Exception in compiling %s: %s", "prologue" if not epilogue_fusion else "epilogue", str(e), ) continue with multi_node.swap_as_triton_caller(choice): ms_fused, path = self.benchmark_codegened_module( mod_fused, device ) new_timings[choice] = ms_fused if ms_fused < min_ms_fused: min_ms_fused = ms_fused ms_fused_choice = choice log_fusion(min_ms_fused, ms1, ms2) if min_ms_fused < (ms1 + ms2) and ms_fused_choice is not None: multi_node.finalize_as_triton_caller(ms_fused_choice) multi_node._choice_timings = new_timings return True else: return False return benchmark_when_ready else: # Start parallel compilation for all three kernels future_and_mod_l1 = compile_kernel(node_list_1) future_and_mod_l2 = compile_kernel(node_list_2) future_and_mod_l1_fused = compile_kernel(node_list_fused) def benchmark_when_ready() -> bool: from torch._inductor.runtime.triton_heuristics import ( NoTritonConfigsError, ) try: # Wait for all compilations to complete for fut in ( future_and_mod_l1[0], future_and_mod_l2[0], future_and_mod_l1_fused[0], ): if fut is not None: fut.result() ms1, path1 = self.benchmark_codegened_module( future_and_mod_l1[1], device ) if math.isinf(ms1): why("register spilling of the first kernel") return False ms2, path2 = self.benchmark_codegened_module( future_and_mod_l2[1], device ) if math.isinf(ms2): why("register spilling of the second kernel") return False ms_fused, path_fused = self.benchmark_codegened_module( future_and_mod_l1_fused[1], device ) if math.isinf(ms_fused): why("register spilling of the fused kernel") return False log_fusion(ms_fused, ms1, ms2) if ( is_metric_table_enabled("slow_fusion") and ms_fused >= ms1 + ms2 and (path1, path2) not in self.logged_slow_fusion ): self.logged_slow_fusion.add((path1, path2)) get_metric_table("slow_fusion").add_row( lambda: { "kernel1_path": path1, "kernel1_latency": ms1, "kernel2_path": path2, "kernel2_latency": ms2, "fused_kernel_path": path_fused, "fused_kernel_latency": ms_fused, "slow_down_ratio": ms_fused / (ms1 + ms2), } ) return ms_fused < ms1 + ms2 except NoTritonConfigsError: return False except CompilationError as e: if "Loop-carried variable" in str(e): return True raise return benchmark_when_ready def get_fused_node(self, node: BaseSchedulerNode) -> BaseSchedulerNode: "Look up the node in Scheduler name_to_fused_node" return self.name_to_fused_node[node.get_first_name()] def fuse_nodes_once( self, nodes: list[BaseSchedulerNode] ) -> list[BaseSchedulerNode]: """ Combine eligible nodes into FusedSchedulerNodes. This relies on two key functions to control the logic: - self.can_fuse(): checks if a fusion is legal - self.score_fusion(): assigns priority to a given fusion """ fused_nodes = OrderedSet(nodes) if fusion_log.isEnabledFor(logging.DEBUG): fusion_log.debug("fuse_nodes_once, candidates:") for node in fused_nodes: fusion_log.debug(" " + node.debug_str_short()) # noqa: G003 # These are potential fusions which we are async compiling, # and which we will benchmark profitability of. pending_fusions: dict[ BaseSchedulerNode, tuple[Callable[[], bool], BaseSchedulerNode, BaseSchedulerNode], ] = {} def fuse_two_nodes( node1: BaseSchedulerNode, node2: BaseSchedulerNode ) -> BaseSchedulerNode: fusion_log.debug("fusing %s with %s", node1.get_name(), node2.get_name()) device = node1.get_device() assert node2.get_device() == device node3 = self.get_backend(device).fuse(node1, node2) fused_nodes.remove(node1) fused_nodes.remove(node2) fused_nodes.add(node3) self.name_to_fused_node.update( {n.get_name(): node3 for n in node3.get_nodes()} ) return node3 def resolve_pending_fusions( node1: BaseSchedulerNode, node2: BaseSchedulerNode ) -> None: while ( self.get_fused_node(node1) in pending_fusions or self.get_fused_node(node2) in pending_fusions ): pending_fusion = pending_fusions.get( self.get_fused_node(node1), pending_fusions.get(self.get_fused_node(node2), None), ) assert pending_fusion is not None is_speedup, node_key1, node_key2 = pending_fusion pending_fusions.pop(node_key1, None) pending_fusions.pop(node_key2, None) assert self.get_fused_node(node_key1) is node_key1 assert self.get_fused_node(node_key2) is node_key2 if not is_speedup() or self.will_fusion_create_cycle(node1, node2): continue fuse_two_nodes(node_key1, node_key2) for node1, node2 in self.get_possible_fusions(nodes): # if either node is in a pending fusion, resolve it. # since we iterate on potential fusions based on profitability # the first potential fusion should take precedence. resolve_pending_fusions(node1, node2) node1 = self.get_fused_node(node1) node2 = self.get_fused_node(node2) if self.can_fuse(node1, node2) and not self.will_fusion_create_cycle( node1, node2 ): speedup = self.speedup_by_fusion(node1, node2) if callable(speedup): pending_fusions[node1] = (speedup, node1, node2) pending_fusions[node2] = (speedup, node1, node2) continue if not speedup: continue fuse_two_nodes(node1, node2) seen_pair_speedup_fn: OrderedSet[Callable[[], bool]] = OrderedSet() for is_speedup_fn, node_key1, node_key2 in pending_fusions.values(): if is_speedup_fn in seen_pair_speedup_fn: continue seen_pair_speedup_fn.add(is_speedup_fn) assert self.get_fused_node(node_key1) is node_key1 assert self.get_fused_node(node_key2) is node_key2 if is_speedup_fn() and not self.will_fusion_create_cycle( node_key1, node_key2 ): fuse_two_nodes(node_key1, node_key2) nodes = sorted(fused_nodes, key=lambda x: x.min_order) nodes = self.topological_sort_schedule(nodes) self.prune_redundant_deps(nodes) return nodes def create_combo_kernel_nodes(self, num_ck_nodes: Optional[int] = None) -> None: """ Groups parallel nodes """ fused_nodes = OrderedSet(self.nodes) count = 0 num_nodes_orig = len(self.nodes) log.debug("ComboKernels: Generating with num_ck_nodes = %d...", num_ck_nodes) for num, node_list in enumerate( ForeachKernelSchedulerNode.group_nodes_for_combo_kernels(self) ): node_list = ForeachKernelSchedulerNode.combinable_nodes(node_list) if len(node_list) < 2: continue if num_ck_nodes is not None and count > num_ck_nodes: break if not self.speedup_by_combo_kernel(node_list): log.debug("ComboKernels: Not speeding up %d-th group", num) continue count += 1 enable_autotune = config.combo_kernels_autotune > 0 group_snode = ForeachKernelSchedulerNode( node_list[0].scheduler, node_list, use_custom_partition_algo=True, enable_autotune=enable_autotune, ) log.info( "ComboKernels: Combining %d nodes for %d-th group", len(node_list), num, ) for node in node_list: fused_nodes.remove(node) fused_nodes.add(group_snode) self.name_to_fused_node.update( {n.get_name(): group_snode for n in group_snode.get_nodes()} ) self.nodes = sorted(fused_nodes, key=lambda x: x.min_order) self.nodes = self.topological_sort_schedule(self.nodes) log.info( "Generated ComboKernel nodes: %d ComboKernels, totally %d -> %d nodels", count, num_nodes_orig, len(self.nodes), ) self.prune_redundant_deps(self.nodes) def prune_redundant_deps(self, nodes: list[BaseSchedulerNode]) -> None: for node in nodes: node.prune_redundant_deps(self.name_to_fused_node) def get_possible_fusions( self, nodes: list[BaseSchedulerNode] ) -> list[tuple[BaseSchedulerNode, BaseSchedulerNode]]: """ Helper to find all legal fusion opportunities, sorted by self.score_fusion() """ possible_fusions = [] seen = OrderedSet[tuple[BaseSchedulerNode, BaseSchedulerNode]]() def check_all_pairs(nodes: list[BaseSchedulerNode]) -> None: for node1_index, node1 in enumerate(nodes): for node2 in nodes[node1_index + 1 :]: key = (node1, node2) if key in seen: continue seen.add(key) if self.can_fuse(node1, node2): possible_fusions.append(key) elif (node2.is_template() or node2.is_foreach()) and self.can_fuse( node2, node1 ): # foreach fusions and epilogue fusions are order dependent possible_fusions.append((node2, node1)) buffer_names_grouping = collections.defaultdict(list) for node in nodes: if self.unfusable_node(node): continue for buf in node.used_buffer_names(): buffer_names_grouping[buf].append(node) for node_grouping in buffer_names_grouping.values(): check_all_pairs(node_grouping) if config.aggressive_fusion: group_grouping = collections.defaultdict(list) for node in nodes: group = getattr(node, "group", None) if group: group_grouping[group].append(node) for node_grouping in group_grouping.values(): check_all_pairs(node_grouping) possible_fusions = self.get_possible_fusions_with_highest_priority( possible_fusions ) possible_fusions.sort(key=self.score_fusion_key, reverse=True) fusion_log.debug("found %d possible fusions", len(possible_fusions)) return possible_fusions def will_fusion_create_cycle( self, node1: BaseSchedulerNode, node2: BaseSchedulerNode ) -> bool: """ Finds whether there's a path from node1 to node2 (or vice-versa) caused indirectly by other fusions. """ # since we are just returning boolean here, use slightly faster, unordered set visited = OrderedSet[FusedSchedulerNode]() def found_path(node: BaseSchedulerNode) -> bool: # only fused nodes can introduce new ancestors. if isinstance(node, FusedSchedulerNode) and node not in visited: visited.add(node) if node.get_operation_names().issubset(combined_ancestors): # All fusion outputs are in ancestors of node1 and node2, thus # cannot introduce new path: # # 1. if output is neither descendent of node1 or node2, the # output cannot introduce a path # 2. due to [can_fuse]: if WLOG output is descendent of node1, it cannot be # on path(node1->node2), hence it cannot be ancestor of node2 # 3. due to [acyclic]: if WLOG output is descendent of node1, it cannot be # ancestor of node1 return False else: # continue DFS of new ancestors introduced by the fusion return bool(combined_names & node.ancestors) or any( found_path(self.name_to_fused_node[n]) for n in node.ancestors - combined_ancestors ) return False # as above - use slightly faster, unordered set combined_names = ( node1.get_operation_names()._dict.keys() | node2.get_operation_names()._dict.keys() ) combined_ancestors = ( node1.ancestors._dict.keys() | node2.ancestors._dict.keys() ) - combined_names cycle = any(found_path(self.name_to_fused_node[n]) for n in combined_ancestors) if cycle: WhyNoFuse(node1, node2)("will create cycle") return cycle def can_fusion_increase_peak_memory( self, node1: BaseSchedulerNode, node2: BaseSchedulerNode ) -> bool: """ Return true if fusing the two nodes can potentially increasing peak memory. The implementation is more like a heuristic since we don't really know if we are at peak or not when trying to fuse these two ndoes. The order of nodes may change later which makes the peak memory estimation hard. Here is how we decide the LOWER BOUND of extra memory allocation if we fuse these 2 nodes: 1. find all buffers read by each node with a single user. These buffers are supposed to be reused if we don't fuses these 2 nodes 2. find the intersection of these buffers for the two node and sum the total buffer size. If we don't fuse these two nodes, we can at lease avoid this much memory allocation. Note that the extra memory allocation is not necessarily causing peak memory increase. This is just a heuristic. We return true only if the saving for fusion can not trade off the extra memory allocation. """ from .codegen.wrapper import buffer_reuse_key def _find_single_user_inputs( node: BaseSchedulerNode, ) -> list[ir.Buffer]: output = [] for rd in node.read_writes.reads: buf = self.name_to_buf.get(rd.name) if buf and len(buf.users) == 1 and buf.node.has_tensor_output(): output.append(buf.node) return output # Check inputs that can be potentially reused lhs_dep_nodes = _find_single_user_inputs(node1) rhs_dep_nodes = _find_single_user_inputs(node2) lhs_reuse_keys = OrderedSet(buffer_reuse_key(buf) for buf in lhs_dep_nodes) rhs_reuse_keys = OrderedSet(buffer_reuse_key(buf) for buf in rhs_dep_nodes) common_reuse_keys = lhs_reuse_keys.intersection(rhs_reuse_keys) memory_overhead = 0 for key in common_reuse_keys: try: memory_overhead += int(key[2]) except ValueError: # not an interger. Fallback is to fuse return False bw_saving = self.score_fusion_memory(node1, node2) # The factor 32 here is quite arbitrary. if V.graph.sizevars.statically_known_gt(memory_overhead, 32 * bw_saving): return True return False def are_long_distant_nodes( self, node1: BaseSchedulerNode, node2: BaseSchedulerNode ) -> bool: """ This function prevents fusion for nodes that can increase memory footprint. This problem is more common in horizontal fusion, where nodes that are far apart in the original order get fused, lengthening the live intervals of tensors. This is very evident in models with activation checkpointing, where the recomputed nodes from different checkpointed regions get fused and significantly increase the memory footprint. The current attempt is a quick, possibly hacky, heuristic to prevent the fusion of nodes that are far away in the original order. A better but difficult to implement heurisitic would be to use live intervals of the buffers, find region of peak pressure in the original program and prevent fusion that crosses that peak region. We might need special care or good approximation in this implementation, as fusion of node changes live intervals, and re-computing live intervals and peak memory after each fusion can introduce large compilation overhead. """ proximity_score = max( abs(node1.min_order - node2.max_order), abs(node2.min_order - node1.max_order), ) return proximity_score > 64 def decide_fusion_fail_reason( self, node1: BaseSchedulerNode, node2: BaseSchedulerNode, common_buf_names: Union[tuple[str], OrderedSet[str]], ) -> str: """ Try to decide reasons why fusion fail due to no shared memory even though there are common buffers. """ reasons = {} node1_name2dep = {dep.name: dep for dep in node1.read_writes.reads_and_writes()} node2_name2dep = {dep.name: dep for dep in node2.read_writes.reads_and_writes()} for buf_name in common_buf_names: buf = V.graph.get_buffer(buf_name) lhs_dep = node1_name2dep[buf_name] rhs_dep = node2_name2dep[buf_name] if not isinstance(lhs_dep, MemoryDep) or not isinstance(rhs_dep, MemoryDep): reasons[buf_name] = ( f"not MemoryDep: {type(lhs_dep)} v.s. {type(rhs_dep)}" ) continue if lhs_dep.get_numel() != rhs_dep.get_numel(): reasons[buf_name] = ( f"different numel: {lhs_dep.get_numel()} v.s. {rhs_dep.get_numel()}" ) continue # same numel but different MemoryDep.size. Should be broadcasting if sympy_product(lhs_dep.size) != sympy_product(rhs_dep.size): reasons[buf_name] = "broadcast" continue lhs_off = lhs_dep.get_offset() rhs_off = rhs_dep.get_offset() if lhs_off != rhs_off: # One example is in transformer, we use a concatenated linear layer # to project Q/K/V and then split the result. The 3 splits will # point to the same buffer with different offsets. reasons[buf_name] = f"different offset: {lhs_off} v.s. {rhs_off}" continue if ( lhs_dep.normalize_with_stride_order() == rhs_dep.normalize_with_stride_order() ): reasons[buf_name] = f"Mismatch loop orders: {lhs_dep} v.s. {rhs_dep}" continue # Add more rules here layout_str = "" if not isinstance(buf, ir.TorchBindObject): layout_str = f"Layout: {buf.layout}" reasons[buf_name] = ( f"Unknown reason: {lhs_dep} v.s. {rhs_dep}. {layout_str}" ) return str(reasons) def shared_data_after_reordering_loop( self, node1: BaseSchedulerNode, node2: BaseSchedulerNode ) -> int: """ Right now just greedily reorder the loop of node1 to be compatible with node2, but ideally we should have some heuristics to reorder the loop for node2 to be compatibile with node1 if that's more efficient. """ # TODO Don't do loop reordering for CPU for now. # Should debug more why it does not work for CPU codegen if not config.loop_ordering_after_fusion or any( n.is_cpu() for n in [node1, node2] ): return 0 node1_buffer_names = node1.read_writes.buffer_names() node2_buffer_names = node2.read_writes.buffer_names() # Fast path: no common buffers. common_buffer_names = node1_buffer_names & node2_buffer_names if not common_buffer_names: return 0 node1_name2dep = {dep.name: dep for dep in node1.read_writes.reads_and_writes()} node2_name2dep = {dep.name: dep for dep in node2.read_writes.reads_and_writes()} # Find the commons buffers that has different loop orders candidates = [] for buffer_name in common_buffer_names: lhs_dep = node1_name2dep[buffer_name] rhs_dep = node2_name2dep[buffer_name] if ( lhs_dep.normalize_with_stride_order() == rhs_dep.normalize_with_stride_order() ): candidates.append( ( V.graph.sizevars.size_hint(lhs_dep.get_numel(), fallback=0), lhs_dep, rhs_dep, ) ) if len(candidates) == 0: return 0 # Pick the largest buffer to guide the loop reordering _numel, lhs_dep, rhs_dep = max(candidates, key=lambda x: x[0]) if not isinstance(lhs_dep, MemoryDep) or not isinstance(rhs_dep, MemoryDep): return 0 if lhs_dep.num_vars != rhs_dep.num_vars: # this can happen due to we don't merge loops. # We can not do loop reordering in this case right now # Simply returning true if the two Deps are the same after # normalization (merging loops) if lhs_dep.normalize() == rhs_dep.normalize(): return self.dep_size_hint(lhs_dep) return 0 # Only reorder loops for pointwise for now if not node1.is_reduction(): node1.reorder_loops_by_dep_pair(lhs_dep, rhs_dep) elif not node2.is_reduction(): node2.reorder_loops_by_dep_pair(rhs_dep, lhs_dep) else: loop_ordering_log.debug( "Don't reorder loops since both nodes are reductions: %s v.s. %s", node1.get_name(), node2.get_name(), ) return self.score_fusion_memory(node1, node2) def unfusable_node(self, node: BaseSchedulerNode) -> bool: """ Is this node unfusable under any conditions. """ return ( isinstance(node, (ExternKernelSchedulerNode, NopKernelSchedulerNode)) and not node.is_template() and not is_output_of_multi_outputs_template(node.node) ) def check_prologue_fusion_heuristics_fusable( self, prologue_node: BaseSchedulerNode, template_node: BaseSchedulerNode, why: WhyNoFuse, ) -> bool: """ Heuristics to avoid benchmarking predictably slow prologue fusions """ # user opt into more aggressive prologue fusion, dont use heuristics if prologue_node.get_operation_names() <= V.graph.invoke_quant_ops: return True read_bytes = prologue_node.get_read_buffer_sizes() write_bytes = prologue_node.get_write_buffer_sizes() # Initially, only do fusions which will result in fewer memory accesses inside of the template to avoid # potential bad cache behavior and shared memory use. # we also want to avoid benchmarking reliably unprofitable fusions like downcasts from fp32 -> fp16 inside kernel. # allowing gathers by allowing increasing write_bytes by small factor # TODO - make configurable per input, for insance, bias can fuse fp32 -> fp16 profitably BYTES_THRESHOLD_MULTIPLIER = 1.1 if read_bytes > (write_bytes * BYTES_THRESHOLD_MULTIPLIER): why("prologue fusion will not increase amount of bytes read in kernel") return False # we want to avoid attempting to fuse predictably unprofitable prologues # such as increasing the unaligned reads or writes. # TODO - would be nice to generalize this, however, we would need more explicit # knowledge of memory access patterns in the TritonTemplate in order to know # the stride order to check alignment. origins = tuple( e.target for n in prologue_node.get_nodes() if n.node is not None for e in n.node.get_origins() if e.op == "call_function" ) if origins == (torch.ops.aten.constant_pad_nd.default,): why( "prologue fusion will not increase attempt to fuse in padding bc it increases unaligned reads" ) return False def low_prec_fp(dtype: torch.dtype) -> bool: return dtype.itemsize <= 2 and dtype.is_floating_point if ( low_prec_fp(template_node.get_template_node_or_throw().dtype) and not prologue_node.can_codegen_in_low_precision() ): why( "prologue fusion that must be upcast to fp32 not profitable for low precision templates" ) return False return True def can_fuse(self, node1: BaseSchedulerNode, node2: BaseSchedulerNode) -> bool: """ Determine if it is possible to combine node1 and node2 into a single fused node. """ if node1 is node2: return False why = WhyNoFuse(node1, node2) if node1.is_template() and self.get_backend( node1.get_device() ).can_fuse_multi_outputs_template(node1, node2): return True if isinstance(node1, GroupedSchedulerNode) or isinstance( node2, GroupedSchedulerNode ): why("grouped node must not be fused with other nodes") return False if ( isinstance(node1, (ExternKernelSchedulerNode, NopKernelSchedulerNode)) and not node1.is_template() ): why("node1 is extern or nop") return False if ( isinstance(node2, (ExternKernelSchedulerNode, NopKernelSchedulerNode)) and not node2.is_template() ): why("node2 is extern or nop") return False if node2.get_operation_names() & node1.ancestors: why("node1 must go before node2") return False if node2.is_template(): if not config.prologue_fusion: why("prologue fusion turned off") return False if node1.is_reduction() or node1.is_template(): why("prologue fusion only supported for pointwise nodes") return False template = node2.get_template_node_or_throw() if not isinstance(template, ir.TritonTemplateBuffer): why("prologue fusion only supported for TritonTemplates") return False allowed_prologue_inps = template.get_allowed_prologue_inps() unsupported_prologue_args = ( OrderedSet(inp.get_name() for inp in template.inputs) - allowed_prologue_inps ) if node1.get_buffer_names() & unsupported_prologue_args: why("prologue fusion not implemented for kernel for these inputs") return False if node1.has_aliasing_or_mutation() or node1.has_aliasing_or_mutation(): why("template prologue can only fuse functional pointwise nodes") return False prologue_nodes = node1.get_nodes() for node in prologue_nodes[:-1]: node_outs = node.get_outputs() for out in node_outs: if not all(user.node in prologue_nodes for user in out.users): why("template prologue can only fuse nodes with a single use") return False template_snodes = ( [node2] if not isinstance(node2, FusedSchedulerNode) else [n for n in node2.snodes if n.is_template()] ) assert len(template_snodes) == 1 template_snode = template_snodes[0] if not ( len(prologue_nodes[-1].outputs) == 1 and len(prologue_nodes[-1].outputs[0].users) == 1 and prologue_nodes[-1].outputs[0].users[0].node is template_snode ): why( "template prologue can only fuse nodes with a single use into template" ) return False if not self.check_prologue_fusion_heuristics_fusable(node1, node2, why): return False if node1.is_template() and ( node2.has_aliasing_or_mutation() or node2.is_reduction() or not config.epilogue_fusion ): why("template epilogue not satisfied") return False if (node1.get_buffer_names() & V.graph.no_fuse_buffer_names) or ( node2.get_buffer_names() & V.graph.no_fuse_buffer_names ): why("fusion for buffer explicit disabled") return False device = node1.get_device() device2 = node2.get_device() if device != device2: why("device mismatch (%s vs %s)", device, device2) return False del device2 shared_data_score = self.score_fusion_memory(node1, node2) if ( shared_data_score < config.score_fusion_memory_threshold and config.loop_ordering_after_fusion ): shared_data_score = self.shared_data_after_reordering_loop(node1, node2) if loop_ordering_log.isEnabledFor(logging.DEBUG): loop_ordering_log.debug( "%s and %s has %s shared data", node1.get_name(), node2.get_name(), shared_data_score, ) if not V.choices.can_fuse(self, node1, node2, shared_data_score): return False if node1.get_operation_names() & node2.ancestors: # node2 depends on node1 outputs return ( self.can_fuse_vertical(node1, node2) and V.choices.can_fuse_vertical(self, node1, node2, shared_data_score) and self.get_backend(device).can_fuse_vertical(node1, node2) ) else: # nodes don't depend on each other, but may have common reads return V.choices.can_fuse_horizontal( self, node1, node2, shared_data_score ) and self.get_backend(device).can_fuse_horizontal(node1, node2) def can_fuse_vertical( self, node1: BaseSchedulerNode, node2: BaseSchedulerNode ) -> bool: """ Check if it is legal to fuse a consumer (node2) into a producer (node1). We can fuse them if all the reads of node2 either match corresponding writes in node1, or are written by nodes that can be scheduled before the fusion of node1 and node2. """ node1_buf_names = node1.get_buffer_names() why = WhyNoFuse(node1, node2) remaining_deps_by_name: dict[str, list[Dep]] = defaultdict(list) for dep in node2.unmet_dependencies: name = self.mutation_renames.get(dep.name, dep.name) if isinstance(dep, WeakDep) and self.fusable_weak_dep(dep, node1, node2): continue remaining_deps_by_name[name].append(dep) for cd in node1.read_writes.writes: if not isinstance(cd, MemoryDep): continue remaining = remaining_deps_by_name.get( self.mutation_renames.get(cd.name, cd.name) ) if remaining: for rd in remaining: if self.fusable_read_and_write(rd, cd): remaining.remove(rd) remaining_deps = OrderedSet( dep.name for dep in itertools.chain.from_iterable(remaining_deps_by_name.values()) ) if remaining_deps & node1_buf_names: # MemoryDeps didn't match and read different locations of the same buffer. # Examples here include: # - MemoryDep("foo", x) != MemoryDep("foo", x + 1) # - MemoryDep("foo", x) != StarDep("foo") why("memory deps did not match") return False node1_op_names = node1.get_operation_names() for name in remaining_deps: op_name = self.name_to_buf[name].defining_op_name() if node1_op_names & self.name_to_fused_node[op_name].ancestors: why("intermediate nodes between node1 & node2") return False return True def fusable_weak_dep( self, weak_dep: WeakDep, node1: BaseSchedulerNode, node2: BaseSchedulerNode ) -> bool: if weak_dep.name not in node1.get_buffer_names(): return False # A weak dep can be fused if and only if the fused operation acts inplace # on the buffer being mutated. i.e. the same index is being read then mutated mutating_writes = [ write for write in node2.read_writes.writes if write.name == weak_dep.mutating_buf ] if len(mutating_writes) != 1: return False write = mutating_writes[0] assert isinstance(write, MemoryDep) if free_symbol_is_type(write.index, SymT.TMP): return False real_name = self.mutation_real_name[weak_dep.mutating_buf] relevant_reads = [ read for read in node1.read_writes.reads if read.name == real_name ] return all( isinstance(read, MemoryDep) and not free_symbol_is_type(read.index, SymT.TMP) and read.index == write.index and read.size == write.size for read in relevant_reads ) # StarDep doesn't match MemoryDep, different indices don't match # However, broadcasting sometimes strips dimensions, and if that's the case # we still can match unmet dep # if there's indirect indexing, don't match it def fusable_read_and_write(self, read: Dep, write: MemoryDep) -> bool: if isinstance(read, MemoryDep): read_name = self.mutation_renames.get(read.name, read.name) if ( read_name != write.name or free_symbol_is_type(read.index, SymT.TMP) or free_symbol_is_type(write.index, SymT.TMP) ): return False if config.loop_ordering_after_fusion and read.num_vars != write.num_vars: # Need merge loops if we do loop ordering after fusion since # we have not merged the loops yet when creating the scheduler # nodes. read = read.normalize() write = write.normalize() return ( read.index == write.index and len(read.size) >= len(write.size) and read.size[: len(write.size)] == write.size ) elif isinstance(read, StarDep): read_name = self.mutation_renames.get(read.name, read.name) write_name = self.mutation_renames.get(write.name, write.name) if ( read.mode == write.mode and write.mode is not None and read_name == write_name ): return True return False def dep_size_hint(self, dep: Dep) -> int: res = 0 if dep not in self.__dep_size_hint_cache: try: if not dep.has_unbacked_symbols(): res = dep.numbytes_hint() except KeyError: # In at least one test (test/inductor/test_torchbind.py) we # create a StarDep that doesn't exist in the graph and calling # `has_unbacked_symbols()` throws an error. pass self.__dep_size_hint_cache[dep] = res else: res = self.__dep_size_hint_cache[dep] return res def score_fusion_memory( self, node1: BaseSchedulerNode, node2: BaseSchedulerNode ) -> int: """ The first term in our fusion score that estimates number of saved memory operations. """ node1_dep_len = len(node1.read_writes.reads) + len(node1.read_writes.writes) node2_dep_len = len(node1.read_writes.reads) + len(node2.read_writes.writes) # optimization: iter over smaller set if min(node1_dep_len, node2_dep_len) * 4 < max(node1_dep_len, node2_dep_len): if node1_dep_len > node2_dep_len: tmp = node1 node1 = node2 node2 = tmp deps = [ dep for dep in node1.read_writes.reads | node1.read_writes.writes if dep in node2.read_writes.reads or dep in node2.read_writes.writes ] return sum(self.dep_size_hint(dep) for dep in deps) common_memory_deps = (node1.read_writes.reads | node1.read_writes.writes) & ( node2.read_writes.reads | node2.read_writes.writes ) return sum(self.dep_size_hint(dep) for dep in common_memory_deps) def get_possible_fusions_with_highest_priority( self, possible_fusions: list[tuple[BaseSchedulerNode, BaseSchedulerNode]] ) -> list[tuple[BaseSchedulerNode, BaseSchedulerNode]]: # Group the possible fusions based on their priority from the backend. # Only return the group of possible fusions with highest priority. if len(possible_fusions) == 0: return possible_fusions possible_fusions_group_by_priority: dict[ int, list[tuple[BaseSchedulerNode, BaseSchedulerNode]] ] = {} for node1, node2 in possible_fusions: assert node1.get_device() == node2.get_device() device = node1.get_device() fusion_pair_priority = int( self.get_backend(device).get_fusion_pair_priority(node1, node2) ) if fusion_pair_priority not in possible_fusions_group_by_priority: possible_fusions_group_by_priority[fusion_pair_priority] = [ (node1, node2), ] else: possible_fusions_group_by_priority[fusion_pair_priority].append( (node1, node2) ) # return the possible fusions with highest priority possible_fusions_with_highest_priority = min( possible_fusions_group_by_priority.items(), key=operator.itemgetter(0) )[1] assert len(possible_fusions_with_highest_priority) > 0 return possible_fusions_with_highest_priority def score_fusion_key( self, nodes: tuple[BaseSchedulerNode, BaseSchedulerNode] ) -> Any: """ Shim for list.sort(key=...) """ return V.choices.score_fusion(self, *nodes) def compute_last_usage(self) -> None: """ Populate node.last_usage recursively (also for the nodes within a FusedSchedulerNode) """ future_used_buffers = OrderedSet(V.graph.get_output_names()) for node in reversed(self.nodes): node.set_last_usage(future_used_buffers, self.mutation_real_name) future_used_buffers.update(node.last_usage) def free_buffers(self) -> None: """Free any buffers that are no longer needed""" for name in sorted( self.buffer_names_to_free - V.graph.removed_buffers - V.graph.wrapper_code.freed # type: ignore[has-type] ): if name in self.name_to_buf: buf = self.name_to_buf[name] if buf.can_free(): V.graph.wrapper_code.codegen_free(buf.node) elif name in V.graph.graph_inputs: inp = V.graph.graph_inputs[name] if isinstance(inp, ir.TorchBindObject): V.graph.wrapper_code.codegen_free(inp) else: storage = inp.data assert ( isinstance(storage, ir.StorageBox) and storage.is_input_buffer() ) V.graph.wrapper_code.codegen_free(storage.data) self.buffer_names_to_free.clear() def flush(self) -> None: for backend in self.backends.values(): backend.flush() self.free_buffers() def codegen_extern_call(self, scheduler_node: ExternKernelSchedulerNode) -> None: assert isinstance(scheduler_node, ExternKernelSchedulerNode) # 'decide_inplace_update' stores the inplace update decisions in # the current kernel from where 'allocate' retrieve those decisions. # We have to make sure there is a non-NULL kernel handler to store # those inplace update decisions. counters["inductor"]["extern_calls"] += 1 with V.set_kernel_handler(Kernel(increase_kernel_count=False)): scheduler_node.decide_inplace_update() scheduler_node.mark_run() node = scheduler_node.node assert isinstance(node, ir.ExternKernel), f"{type(node)=}" node.codegen(V.graph.wrapper_code) self.free_buffers() def create_backend(self, device: torch.device) -> BaseScheduling: assert not is_gpu(device.type) or device.index is not None, ( f"{device} should have been normalized in lowering" ) V.graph.add_device_info(device) device_scheduling = get_scheduling_for_device(device.type) if device_scheduling is None: raise RuntimeError(f"Unsupported device type: {device.type}") if not has_triton(): if ( device.type == "cuda" and (device_props := torch.cuda.get_device_properties(device)).major < 7 ): raise GPUTooOldForTriton(device_props, inspect.currentframe()) elif is_gpu(device.type) and not device.type == "mps": raise TritonMissing(inspect.currentframe()) return device_scheduling(self) def get_backend(self, device: Optional[torch.device]) -> BaseScheduling: assert device is not None if device not in self.backends: self.backends[device] = self.create_backend(device) return self.backends[device] def enter_context(self, node: BaseSchedulerNode) -> None: def get_order(n: torch.fx.Node) -> int: if n not in self.origin_to_index: self.origin_to_index.update({n: i for i, n in enumerate(n.graph.nodes)}) return self.origin_to_index[n] # Use a dict to have ordering origins = { (get_order(e), e): None for n in node.get_nodes() if n.node is not None for e in n.node.get_origins() } origins = list(origins.keys()) if origins: _, last = max(origins, key=operator.itemgetter(0)) V.graph.wrapper_code.enter_context(last) def can_buffer_be_removed_through_fusion( self, name: str, fused_node_names: OrderedSet[str] ) -> bool: try: users = self.name_to_buf[name].users except KeyError: return False return ( all(user.is_weak or user.get_name() in fused_node_names for user in users) and name not in self.mutation_renames and name not in self.mutation_real_name ) def should_partition(self, node: BaseSchedulerNode) -> bool: """Return True if we should partition the inductor graph on this node""" if not node.is_gpu(): return True if node.node is None: return True if isinstance(node.node, ir.DeviceCopy): return True if isinstance(node.node, ir.Conditional): return True if getattr(node.node, "unbacked_bindings", None): return True if hasattr(node.node, "layout") and any( isinstance(expr, sympy.Expr) and expr.free_symbols for expr in node.node.layout.size ): return True return False def get_name_to_nodes( self, ) -> dict[str, Union[ir.IRNode, ir.TorchBindObject, sympy.Expr]]: """ Return a mapping from name strings to the corresponding graph inputs or base scheduler node outputs. """ name_to_node: dict[str, Union[ir.IRNode, ir.TorchBindObject, sympy.Expr]] = {} name_to_node.update(V.graph.graph_inputs) for node in self.nodes: for name, scheduler_buffer in node.outputs_by_name.items(): name_to_node[name] = scheduler_buffer.node return name_to_node def get_graph_partition_signature( self, partitions: list[PartitionType], skip_cudagraphs: list[bool] ) -> list[GraphPartitionSignature]: """ Gets signature for each graph partition, including input nodes, output nodes, and whether deallocating an input within graph partition. """ signatures = [] unmet_output_names = OrderedSet(V.graph.get_output_names()) name_to_node = self.get_name_to_nodes() for partition, skip_cudagraph in zip( reversed(partitions), reversed(skip_cudagraphs) ): output_names: OrderedSet[str] = OrderedSet() for node in partition: output_names.update(node.outputs_by_name.keys()) returned_output_names = output_names.intersection(unmet_output_names) # all reads/writes are partition inputs except those generated # within the partition read_writes = dependencies.ReadWrites.merge_list( [node.read_writes for node in partition] ) partition_input_names = ( OrderedSet([x.name for x in read_writes.reads | read_writes.writes]) - output_names ) buffer_names_to_free: OrderedSet[str] = OrderedSet() for node in partition: buffer_names_to_free.update(node.last_usage) input_nodes = { name: name_to_node[name] for name in partition_input_names if name in name_to_node } input_deallocation = { name: True if name in buffer_names_to_free else False for name in partition_input_names if name in name_to_node } output_nodes = [name_to_node[name] for name in returned_output_names] signatures.append( GraphPartitionSignature( input_nodes, output_nodes, input_deallocation, skip_cudagraph, ) ) unmet_output_names = partition_input_names.union( unmet_output_names - returned_output_names ) return signatures[::-1] def graph_partition( self, ) -> tuple[list[PartitionType], list[GraphPartitionSignature]]: """ Given a list of BaseSchedulerNodes, split into a list of graph partitions and compute partition input/output signatures. """ partitions: list[PartitionType] = [] skip_cudagraph = True cur_partition: PartitionType = [] skip_cudagraphs = [] for node in self.nodes: should_partition = self.should_partition(node) if cur_partition and skip_cudagraph != should_partition: partitions.append(cur_partition) skip_cudagraphs.append(skip_cudagraph) cur_partition = [] skip_cudagraph = should_partition cur_partition.append(node) if cur_partition: partitions.append(cur_partition) skip_cudagraphs.append(skip_cudagraph) return partitions, self.get_graph_partition_signature( partitions=partitions, skip_cudagraphs=skip_cudagraphs ) def codegen(self) -> None: with dynamo_timed("Scheduler.codegen"): return ( self._codegen_partitions() if torch._inductor.config.graph_partition else self._codegen(self.nodes) ) def _codegen_partition_wrapper( self, partition: PartitionType, signature: GraphPartitionSignature, ) -> None: """Codegen a partition given its inputs/outputs""" parent_wrapper_code = V.graph.wrapper_code graph_partition_id = next(self._graph_partition_counter) with V.graph.set_current_wrapper_code(): V.graph.init_wrapper_code( is_subgraph=True, subgraph_name=f"partition_{graph_partition_id}", parent_wrapper_code=parent_wrapper_code, partition_signatures=signature, ) self._codegen(partition) partition_code, _ = V.graph.wrapper_code.generate(V.graph.is_inference) V.graph.wrapper_code.define_subgraph_launcher_fn(partition_code.value) V.graph.wrapper_code.codegen_partition_call(graph_partition_id, signature) V.graph.wrapper_code.allocated.update( # type: ignore[has-type] [node.get_name() for node in signature.output_nodes] ) def _codegen_partitions(self) -> None: """ Split nodes into partitions and codegen each partition into separate functions. This allows further applying different optimizations (e.g., cudagraph) to each function. """ partitions, signatures = self.graph_partition() for partition, signature in zip(partitions, signatures): assert len(partition) >= 1, ( f"Each partition must have at least one node but found {len(partition)}" ) if signature.skip_cudagraph: self._codegen(partition) else: self._codegen_partition_wrapper(partition, signature) num_partitions = next(self._graph_partition_counter) V.graph.wrapper_code.set_all_partition_names(num_partitions) def _codegen(self, nodes: list[BaseSchedulerNode]) -> None: if config.check_stack_no_cycles_TESTING_ONLY: import torch._dynamo.convert_frame stack = traceback.extract_stack() seen: OrderedSet[tuple[str, int | None]] = OrderedSet() for frame in reversed(stack): # This is where maybe_cprofile is if ( frame.name == "_compile_inner" and frame.filename == torch._dynamo.convert_frame.__file__ ): break key = (frame.filename, frame.lineno) assert key not in seen, ( f"Duplicate stack frame {frame.filename}:{frame.lineno}; " "did you add a decorator to one of the functions in this stack " "trace? If so, try using a context manager instead." ) seen.add(key) self.current_device = None for node in nodes: if log.isEnabledFor(logging.DEBUG): try: log.debug( "Generating code for node %s with estimated runtime %f", node.get_name(), node.get_estimated_runtime(), ) except Exception: log.debug( "Generating code for node %s with estimated runtime 0.0", node.get_name(), ) self.enter_context(node) if device := node.get_device(): if ( device != self.current_device or node.is_extern() or node.is_template() ): self.flush() if device != self.current_device: if self.current_device and device_need_guard( self.current_device.type ): V.graph.wrapper_code.codegen_device_guard_exit() self.current_device = device if device_need_guard(device.type): assert device.index is not None, "device should have an index" V.graph.wrapper_code.codegen_device_guard_enter(device.index) self.buffer_names_to_free.update(node.last_usage) if node.is_template(): prologue, template_node, epilogue = node.get_prologue_template_epilogue( list(node.get_nodes()) ) self.get_backend(device).codegen_template( template_node, epilogue, prologue ) elif node.is_extern(): node = typing.cast(ExternKernelSchedulerNode, node) self.codegen_extern_call(node) elif node.is_foreach(): node = typing.cast(ForeachKernelSchedulerNode, node) backend_ = self.get_backend(device) from .codegen.cuda_combined_scheduling import CUDACombinedScheduling from .codegen.simd import SIMDScheduling if isinstance(backend_, (SIMDScheduling, CUDACombinedScheduling)): backend = backend_ else: raise AssertionError(f"{type(self)=}") backend.codegen_combo_kernel(node) elif isinstance(node, (FusedSchedulerNode, SchedulerNode)): self.get_backend(device).codegen_node(node) else: assert isinstance(node, NopKernelSchedulerNode) node.mark_run() if config.triton.debug_sync_kernel: self.get_backend(device).codegen_sync() self.available_buffer_names.update(node.get_buffer_names()) self.completed_operations.update(node.get_operation_names()) if not isinstance(node, NopKernelSchedulerNode): device = node.get_device() if device is not None and self.get_backend(device).ready_to_flush(): self.flush() if self.current_device and device_need_guard(self.current_device.type): # exit the outermost CUDA device guard. this is # important for nested indentation codegen-ing. V.graph.wrapper_code.codegen_device_guard_exit() self.flush() def benchmark_combo_kernel( self, node_list: Sequence[BaseSchedulerNode] ) -> tuple[float, float, list[Optional[str]]]: """ Benchmark fused list of nodes and return the execution time in milliseconds on randomly generated inputs. """ device = node_list[0].get_device() V.graph.scheduler = self self.current_device = device assert device is not None backend = self.get_backend(device) return backend.benchmark_combo_kernel(node_list) def speedup_by_combo_kernel(self, nodes: list[BaseSchedulerNode]) -> bool: """ If config.benchmark_fusion is False, always return True. Otherwise, return True if fusion can brings speedup. """ if not config.benchmark_combo_kernel: return True subkernel_nodes = nodes device = subkernel_nodes[0].get_device() # don't support benchmark fusion for CPU right now. if device is None or device.type == "cpu": return True from triton.compiler.errors import CompilationError ms1, path1_list = 0.0, [] for i, snode in enumerate(subkernel_nodes): node_list = snode.get_nodes() # We can not accurately benchmark kernel using atomic_add # due to how we generate random integer inputs. if self._any_atomic_add(node_list): fusion_log.debug( "ComboKernel: benchmarking may not accurate due to atomic_add" ) try: ms, path = self.benchmark_fused_nodes(node_list) if math.isinf(ms): fusion_log.debug( "ComboKernel benchmark: register spilling of %d-th subkernel", i, ) return False except CompilationError as e: # workaround triton issue: https://github.com/openai/triton/issues/2151 if "Loop-carried variable" in str(e): fusion_log.debug( "ComboKernel benchmark: return True because of loop-carried variable" ) return True # allow fusion else: raise ms1 += ms path1_list.append(path) try: ms2, ms2_clone, _path2_list = self.benchmark_combo_kernel(subkernel_nodes) except CompilationError as e: # workaround triton issue: https://github.com/openai/triton/issues/2151 if "Loop-carried variable" in str(e): fusion_log.debug( "ComboKernel benchmark: return True because of loop-carried variable" ) return True # allow fusion else: raise # small kernels are very likely to have speedup but hard to benchmark. So we skip benchmarking. small_kernel = ms2 - ms2_clone < 0.3 or ms1 < 0.3 if fusion_log.isEnabledFor(logging.DEBUG): if ms1 > ms2 or small_kernel: fusion_log.debug( "can fuse (benchmark): fusing causes %sx speedup", green_text(f"{ms1 / ms2:.3f}"), ) else: fusion_log.debug( "cannot fuse (benchmark): fusing causes %sx slowdown", red_text(f"{ms1 / ms2:.3f}"), ) # ms1 returned by benchmark_fused_nodes discounted clone time return ms2 - ms2_clone < ms1 or small_kernel def get_buffer_layout(self, buf_name: str) -> ir.Layout: buf = self.name_to_buf[buf_name] assert buf.node is not None return buf.node.get_layout() def update_zero_dim_cpu_tensor(self) -> None: for node in self.nodes: if node.is_gpu(): for read in node.read_writes.reads: buffer = V.graph.name_to_buffer.get(read.name) if ( buffer and get_device_type(buffer) == "cpu" and not isinstance(buffer.layout, MultiOutputLayout) and buffer.get_size() == [] ): V.graph.zero_dim_cpu_tensor_list.add(read.name) class BaseScheduling: def __init__(self, scheduler: Optional[Scheduler]): super().__init__() self.scheduler = scheduler def free_buffers_in_scheduler(self) -> None: if self.scheduler: self.scheduler.free_buffers() def get_backend_features(self, device: torch.device) -> OrderedSet[BackendFeature]: """Return a set of .codegen.common.BackendFeature()""" return OrderedSet() def can_fuse_vertical( self, node1: BaseSchedulerNode, node2: BaseSchedulerNode ) -> bool: """ Check whether node1 and node2 can be vertically fused or not. """ raise NotImplementedError def can_fuse_horizontal( self, node1: BaseSchedulerNode, node2: BaseSchedulerNode ) -> bool: """ Check whether node1 and node2 can be horizontally fused or not. """ raise NotImplementedError def can_fuse_multi_outputs_template( self, node1: BaseSchedulerNode, node2: BaseSchedulerNode ) -> bool: """ A Multi-Output Template (referenced in #144012) is a template node with MultiOutputLayout, and its output buffers are instances of MultiOutput. In this context, we verify whether node1 represents the Multi-Output Template and node2 corresponds to one of its outputs. If so, we further check if backend supports this fusion. """ return False def fuse( self, node1: BaseSchedulerNode, node2: BaseSchedulerNode ) -> FusedSchedulerNode: """ Fuse two nodes """ if node1.is_foreach() or node2.is_foreach(): return ForeachKernelSchedulerNode.fuse(node1, node2) else: return FusedSchedulerNode.fuse(node1, node2) def group_fn( self, sizes: Sequence[Sequence[sympy.Expr]] ) -> tuple[tuple[sympy.Expr, ...], ...]: """ Process the iteration sizes in case a transformation needs to be applied. """ raise NotImplementedError def codegen_template( self, template_node: BaseSchedulerNode, epilogue_nodes: Sequence[BaseSchedulerNode], prologue_nodes: Sequence[BaseSchedulerNode], ) -> Optional[str]: """ Given a template node, generate a kernel. This function is only available for triton now. If the third-party backend behaves as a sub-class of TritonScheduling, it can override it or reuse it. """ raise NotImplementedError def generate_kernel_code_from_nodes( self, nodes: Sequence[BaseSchedulerNode], benchmark_kernel: bool ) -> str: """ Generate a kernel given a list of pre-fused nodes. """ raise NotImplementedError def codegen_node(self, node: Union[FusedSchedulerNode, SchedulerNode]) -> None: """ Generate a kernel given a list of pre-fused nodes. """ raise NotImplementedError def codegen_sync(self) -> None: """ Generate synchronization code for the kernel. This method depends on the hardware characteristics. """ raise NotImplementedError def ready_to_flush(self) -> bool: """ Check whether the backend is requesting the scheduler to flush the generated kernel. If not supported, please return False. """ return False def flush(self) -> None: """ Flush the generated kernel and python wrapper code to the source code file. """ raise NotImplementedError def benchmark_fused_nodes( self, nodes: Sequence[BaseSchedulerNode] ) -> tuple[float, str]: """ Benchmark fused list of nodes and return the execution time in milliseconds on randomly generated inputs. """ raise NotImplementedError def benchmark_codegened_module(self, module: ModuleType) -> tuple[float, str]: """ Benchmark a compiled module and return the execution time in milliseconds on randomly generated inputs. """ raise NotImplementedError def get_fusion_pair_priority( self, node1: BaseSchedulerNode, node2: BaseSchedulerNode ) -> int: """ Return an unsigned integer which represents the priority of this fusion pair. The smaller is with higher priority. """ return 0 def benchmark_combo_kernel( self, node_list: Sequence[BaseSchedulerNode] ) -> tuple[float, float, list[Optional[str]]]: """ Benchmark the list of nodes to combine and return the execution time and memory copy time in milliseconds on randomly generated inputs. """ raise NotImplementedError