kernel op cleanups + use ScheduleItem [pr] (#9009)

tinygrad/engine/schedule.py

@@ -249,7 +249,7 @@ break_sched = PatternMatcher([
   (UPat(Ops.VIEW, name="st", src=(UPat(Ops.BUFFER, name="b"), UPat.var("x"))), store_or_fuse),
 ])
 
-# **** convert Kernel to a ScheduleItem (for legacy reasons)
+# **** ScheduleItem creation (TODO: replace ScheduleItem with the KERNEL UOp)
 
 @dataclass(frozen=True)
 class ScheduleItem:
@@ -267,18 +267,6 @@ class ScheduleItem:
   @functools.cached_property
   def output_idxs(self) -> tuple[int, ...]: return tuple(x.src[0].arg for x in self.ast.src) if self.ast.op is Ops.SINK else (0,)
 
-def kernel_to_si(k:UOp) -> ScheduleItem:
-  assert k.op is Ops.KERNEL and isinstance(k.metadata, tuple), f"must be KERNEL {k}"
-  return ScheduleItem(k.arg.ast, tuple(u.buf_uop.buffer for u in k.src), k.metadata)
-
-# **** Kernel creation
-
-@dataclass(frozen=True)
-class Kernel:
-  ast: UOp
-  metadata: tuple[Metadata, ...]
-  def __repr__(self): return f"<Kernel {len(list(self.ast.toposort))} {self.ast.op} {self.metadata}>"
-
 @dataclass(frozen=True)
 class KernelContext:
   var_vals: dict[Variable, int]
@@ -380,14 +368,14 @@ def unbind_variable(ctx:dict[Variable, int], bind:UOp, var:UOp, val:UOp):
   return var
 unbind_vars = PatternMatcher([(UPat(Ops.BIND, name="bind", src=(UPat.var("var"), UPat.cvar("val"))), unbind_variable),])
 
-def schedule_uop(pre:UOp, ctx:ScheduleContext) -> UOp:
+def schedule_uop(pre:UOp, ctx:ScheduleContext) -> ScheduleItem:
   # unbind_vars + push views to edges
   sink = graph_rewrite(graph_rewrite(pre, unbind_vars+view_left, ctx=ctx.var_vals), view_right)
   # remove extra uops from SINK + substitue BUFFER with DEFINE_GLOBAL
   ast = graph_rewrite(sink, to_si, si_ctx:=KernelContext(ctx.var_vals))
   # NOTE: we only add the metadata for fused tensors
   metadata = tuple(dedup(m for x in pre.toposort if x.op is not Ops.BUFFER and (m:=ctx.ops_metadata.get(x)) is not None))
-  return UOp(Ops.KERNEL, src=tuple(si_ctx.bufs), arg=Kernel(ast, metadata))
+  return ScheduleItem(ast, tuple(u.buffer for u in si_ctx.bufs), metadata)
 
 PROCESS_REPLAY_CAPTURE:dict[str, bytes] = {}
 if CAPTURE_PROCESS_REPLAY:
@@ -395,8 +383,13 @@ if CAPTURE_PROCESS_REPLAY:
   def save_process_replay():
     for k,v in PROCESS_REPLAY_CAPTURE.items(): diskcache_put("schedule_process_replay", k, v, prepickled=True)
 
+@dataclass(frozen=True)
+class Kernel:
+  ast: UOp
+  def __repr__(self): return f"<Kernel {len(list(self.ast.toposort))} {self.ast.op}>"
+
 # NOTE: realizes become ASSIGN(BUFFER, KERNEL) in the schedule graph
-def init_kernel(ctx:dict[UOp, UOp], u:UOp): return u.buf_uop.assign(UOp(Ops.KERNEL, src=u.src, arg=Kernel(ctx[u.buf_uop].sink(), ())))
+def init_kernel(ctx:dict[UOp, UOp], u:UOp): return u.buf_uop.assign(UOp(Ops.KERNEL, src=u.src, arg=Kernel(ctx[u.buf_uop].sink())))
 def is_kernel(u:UOp) -> bool: return u.op is Ops.ASSIGN and u.src[1].op is Ops.KERNEL
 
 DONT_PLACE_IN_KERNEL = {Ops.KERNEL, Ops.BUFFER}
@@ -502,7 +495,7 @@ def create_schedule_with_vars(big_sink:UOp) -> tuple[list[ScheduleItem], dict[Va
   schedule: list[ScheduleItem] = []
   while queue:
     u = queue.popleft()
-    schedule.append(kernel_to_si(schedule_uop(u.src[1].arg.ast, ctx)))
+    schedule.append(schedule_uop(u.src[1].arg.ast, ctx))
     for x in children.get(u, []):
       in_degree[x] -= 1
       if in_degree[x] == 0: queue.append(x)

tinygrad/ops.py

@@ -500,7 +500,7 @@ class UOp(MathTrait, metaclass=UOpMetaClass):
     return ret
   def clone(self) -> UOp: return self.copy_to_device(self.device, clone=True)
   @property
-  def metadata(self) -> tuple[Metadata, ...]|Metadata|None: return self.arg.metadata if self.op is Ops.KERNEL else all_metadata.get(self, None)
+  def metadata(self) -> Metadata|None: return all_metadata.get(self, None)
 
   # *** uop movement ops ***