replace if with range

tinygrad/codegen/__init__.py

@@ -1,10 +1,8 @@
-from typing import cast
+import itertools
 from tinygrad.helpers import QUANTIZE, DEVECTORIZE, TRANSCENDENTAL, SPEC
-from tinygrad.uop.ops import PatternMatcher, graph_rewrite, UOp, pm_lower_index_dtype, Ops, UPat
+from tinygrad.uop.ops import PatternMatcher, graph_rewrite, UOp, pm_lower_index_dtype
 from tinygrad.uop.spec import type_verify, program_spec, kernel_spec
 from tinygrad.renderer import Renderer
-from tinygrad.dtype import dtypes
-from tinygrad.helpers import panic
 
 # import all pattern matchers here
 from tinygrad.codegen.quantize import pm_quant
@@ -17,7 +15,7 @@ from tinygrad.codegen.late.devectorizer import load_store_folding, load_store_in
 from tinygrad.codegen.opt.postrange import apply_opts
 from tinygrad.codegen.simplify import pm_simplify_ranges, pm_flatten_range, pm_split_ranges, pm_load_collapse, pm_split_store
 from tinygrad.schedule.rangeify import pm_add_buffers_local, rangeify_codegen
-from tinygrad.codegen.late.linearizer import CFGContext, pm_split_ends, pm_add_control_flow, linearize
+from tinygrad.codegen.late.linearizer import CFGContext, pm_prepare_control_flow, pm_add_control_flow, linearize
 
 def full_rewrite_to_sink(sink:UOp, ren:Renderer|None=None, optimize:bool=True) -> UOp:
   if ren is None: ren = Renderer()
@@ -85,35 +83,18 @@ def full_rewrite_to_sink(sink:UOp, ren:Renderer|None=None, optimize:bool=True) -
 
   # final rules for the renderer (without sym)
   extra_matcher = ren.extra_matcher if ren.extra_matcher is not None else PatternMatcher([])
-  pm_final_rewrite = pm_decomp+pm_render+extra_matcher+pm_split_ends
+  pm_final_rewrite = pm_decomp+pm_render+extra_matcher
   sink = graph_rewrite(sink, pm_final_rewrite, ctx=ren.device, name="final rewrite")
 
+  # prepare for control flow
+  sink = graph_rewrite(sink, pm_prepare_control_flow, ctx=itertools.count(10000), name="split ends + add if ranges")
+
   # this was the linearizer
   sink = graph_rewrite(sink, pm_add_control_flow, ctx=CFGContext(sink), name="add control flow", bottom_up=True)
 
   # return the rewritten sink
   return sink
 
-# inject IF/ENDIF. only needed if device doesn't support gated stores
-pm_linearize_cleanups = PatternMatcher([
-  # if statements are not allowed in the graph
-  (UPat((Ops.IF, Ops.ENDIF)), lambda: panic(RuntimeError("if not allowed in graph"))),
-  # gated INDEX becomes IF-STORE-ENDIF. this is the only use of IF-ENDIF
-  (UPat(Ops.STORE, name="u", src=(UPat(Ops.INDEX, src=(UPat(), UPat(), UPat(name="gate", dtype=dtypes.bool))).or_casted(), UPat()),
-        allow_any_len=True), lambda u, gate: (u, [mif:=UOp(Ops.IF, src=(gate, u.src[0])), u, UOp(Ops.ENDIF, src=(mif,))]))
-])
-
-# requires lst be toposorted. like graph rewrite, but for lines
-def line_rewrite(lst:list[UOp], pm:PatternMatcher) -> list[UOp]:
-  newlst = []
-  replaced: dict[UOp, UOp] = {}
-  for u in lst:
-    nu = u.replace(src=tuple([replaced[x] for x in u.src]))
-    ret: tuple[UOp, list[UOp]] = cast(tuple[UOp, list[UOp]]|None, pm.rewrite(nu)) or (nu, [nu])
-    replaced[u] = ret[0]
-    newlst.extend(ret[1])
-  return newlst
-
 def full_rewrite(sink:UOp, ren:Renderer|None=None) -> list[UOp]:
   """
   Function to transform the Kernel UOp graph into a linearized program.
@@ -128,6 +109,6 @@ def full_rewrite(sink:UOp, ren:Renderer|None=None) -> list[UOp]:
 
   full_sink = full_rewrite_to_sink(sink, ren, optimize=sink.tag is None)
   assert len(full_sink.ranges) == 0, "all ranges must end by the sink"
-  lst = line_rewrite(linearize(full_sink), pm_linearize_cleanups)
+  lst = linearize(full_sink)
   if SPEC: type_verify(lst, program_spec)
   return lst

tinygrad/codegen/late/linearizer.py

@@ -1,6 +1,7 @@
 import heapq
 from collections import defaultdict
-from tinygrad.uop.ops import PatternMatcher, UOp, Ops, UPat
+from tinygrad.dtype import dtypes
+from tinygrad.uop.ops import PatternMatcher, UOp, Ops, UPat, AxisType
 
 def linearize(u:UOp) -> list[UOp]:
   # this is a toposort with priority
@@ -75,7 +76,10 @@ def do_split_ends(e:UOp):
   for r in list(UOp.sink(*e.src[1:]).ranges)[::-1]: ret = ret.end(r)
   return ret
 
-pm_split_ends = PatternMatcher([
+pm_prepare_control_flow = PatternMatcher([
   # split the ends
   (UPat(Ops.END, name="e"), do_split_ends),
+  # add if ranges
+  (UPat(Ops.STORE, src=(UPat(Ops.INDEX, src=(UPat.var("buf"), UPat.var("idx"), UPat.var("gate"))), UPat.var("val"))), lambda ctx,buf,idx,gate,val:
+   buf.after(r:=UOp.range(gate.cast(dtypes.int), next(ctx), AxisType.IF, dtype=dtypes.int)).index(idx).store(val).end(r)),
 ])

tinygrad/helpers.py

@@ -85,7 +85,9 @@ def word_wrap(x, wrap=80):
   while len(ansistrip(x[:i])) < wrap and i < len(x): i += 1
   return x[:i] + "\n" + word_wrap(x[i:], wrap)
 def pad_bytes(b:bytes, align:int) -> bytes: return b + b'\x00' * ((align - (len(b) % align)) % align)
-def panic(e:Exception): raise e
+def panic(e:Exception|None=None):
+  if e is None: raise RuntimeError("PANIC!")
+  raise e
 
 @functools.cache
 def canonicalize_strides(shape:tuple[T, ...], strides:tuple[T, ...]) -> tuple[T, ...]:

tinygrad/uop/ops.py

@@ -15,11 +15,12 @@ if TYPE_CHECKING:
 class AxisType(Enum):
   def __repr__(self): return str(self)
   GLOBAL = auto(); WARP = auto(); LOCAL = auto(); LOOP = auto(); GROUP_REDUCE = auto(); REDUCE = auto(); UPCAST = auto(); UNROLL = auto() # noqa: E702
-  THREAD = auto()
+  THREAD = auto(); IF = auto() # noqa: E702
 axis_letters = {AxisType.GLOBAL: "g", AxisType.THREAD: "t", AxisType.LOCAL: "l", AxisType.WARP: "w", AxisType.LOOP: "L", AxisType.UPCAST: "u",
-                AxisType.GROUP_REDUCE: "G", AxisType.REDUCE: "R", AxisType.UNROLL: "r"}
+                AxisType.GROUP_REDUCE: "G", AxisType.REDUCE: "R", AxisType.UNROLL: "r", AxisType.IF: "I"}
 axis_colors = {AxisType.GLOBAL: "blue", AxisType.THREAD: "BLUE", AxisType.LOCAL: "cyan", AxisType.WARP: "CYAN", AxisType.LOOP: "WHITE",
-               AxisType.UPCAST: "yellow", AxisType.GROUP_REDUCE: "RED", AxisType.REDUCE: "red", AxisType.UNROLL: "magenta"}
+               AxisType.UPCAST: "yellow", AxisType.GROUP_REDUCE: "RED", AxisType.REDUCE: "red", AxisType.UNROLL: "magenta",
+               AxisType.IF: "green"}
 
 range_start = {Ops.BUFFERIZE: 1, Ops.REDUCE: 1, Ops.STORE: 2, Ops.WMMA: 3, Ops.END: 1}