make that a single rewrite

even faster
recursive substitute
2025-10-03 18:18:43 +08:00 · 2025-10-03 18:14:41 +08:00 · 2025-10-03 18:03:55 +08:00
1 changed files with 18 additions and 4 deletions
@@ -715,8 +715,23 @@ replace_contiguous = PatternMatcher([
  (UPat(GroupOp.ALU, name="alu"), lambda ctx,alu: alu.replace(src=new_src) if (new_src:=tuple(ctx.get(s, s) for s in alu.src)) != alu.src else None),
 ])

-def do_sub(s:UOp): return s.src[0].substitute(dict(zip(s.src[1].src, s.src[2].src)))
-pm_substitute = PatternMatcher([(UPat(Ops.SUBSTITUTE, name="s"), do_sub)])
+def do_sub_recurse(s:UOp):
+  x,keys,values = s.src[0], s.src[1].src, s.src[2].src
+  # SUBSTITUTE applied to SUBSTITUTE runs the child SUB on the parents. though this is probably wrong in the generic case
+  if x.op is Ops.SUBSTITUTE:
+    sub_k = UOp(Ops.SUBSTITUTE, src=(x.src[1],)+s.src[1:])
+    sub_v = UOp(Ops.SUBSTITUTE, src=(x.src[2],)+s.src[1:])
+    return UOp(Ops.SUBSTITUTE, src=(x.src[0], sub_k, sub_v))
+  # here we actually do the SUBSTITUTE
+  if x in keys: return values[keys.index(x)]
+  # we filter any keys that aren't in parents. this keeps the algorithm O(output graph size)
+  new_kv = {k:v for k,v in zip(keys,values) if k in x.sparents}
+  # if there's no SUBSTITUTEs left, we can just return x
+  if len(new_kv) == 0: return x
+  # then we add SUBSTITUTE to all parents
+  uop_keys, uop_values = UOp(Ops.NOOP, src=tuple(new_kv.keys())), UOp(Ops.NOOP, src=tuple(new_kv.values()))
+  return x.replace(src=tuple([UOp(Ops.SUBSTITUTE, src=(y,uop_keys,uop_values)) for y in x.src]))
+pm_substitute_recurse = PatternMatcher([(UPat(Ops.SUBSTITUTE, src=(UPat(), UPat(Ops.NOOP), UPat(Ops.NOOP)), name="s"), do_sub_recurse)])

@track_rewrites(lambda _,ret: f"Schedule {pluralize('Kernel', len([u for u in UOp.sink(*ret.values()).toposort() if u.op is Ops.KERNEL]))}", True)
 def get_rangeify_map(sink:UOp) -> dict[UOp, UOp]:
@@ -735,8 +750,7 @@ def get_rangeify_map(sink:UOp) -> dict[UOp, UOp]:
  tsink = graph_rewrite(tsink, pm_rangeify, ctx=(rangeify_ctx:=RangeifyContext()), bottom_up=True, name="rangeify")
  # NOTE: sym (vs symbolic_simple) breaks things here because ranges with len 1 aren't handled right
  tsink = graph_rewrite(tsink, symbolic_simple+pm_reduce_unparented, name="symbolic")  # this supports const folding
-  tsink = graph_rewrite(tsink, pm_cleanups, bottom_up=True, name="remove costly buffers")
-  tsink = graph_rewrite(tsink, pm_substitute, name="do substitute of ranges")
+  tsink = graph_rewrite(tsink, pm_cleanups+pm_substitute_recurse, bottom_up=True, name="remove costly buffers")
  tsink = graph_rewrite(tsink, pm_limit_bufs, ctx=rangeify_ctx, name="limit buffers")

  # rebuild the sink with all the BUFFERIZEs with tags, this is what's ending up in the tensor graph
Author	SHA1	Message	Date
geohot	40f0acbc23	make that a single rewrite	2025-10-03 18:18:43 +08:00
geohot	f3ac529438	even faster	2025-10-03 18:14:41 +08:00
geohot	a3991948e9	recursive substitute	2025-10-03 18:03:55 +08:00