fix st_vars

tinygrad/codegen/__init__.py

@@ -17,6 +17,7 @@ from tinygrad.codegen.devectorizer import load_store_folding, load_store_indexin
 from tinygrad.codegen.optional import get_late_rewrite_patterns
 from tinygrad.codegen.linearize import block_create, pm_blockend_merge, block_merge, pm_finalize, BlockContext
 from tinygrad.opt import pm_optimize
+from tinygrad.opt.swizzler import view_left, view_right, fix_kernel_ops
 
 @dataclass
 class RewriteStep:
@@ -44,6 +45,11 @@ def _get_rewrites_for_renderer(opts:Renderer, linearizer:bool, _QUANTIZE, _DEVEC
   # ** lowerer (rewrite_shapetracker_with_index) **
   ret: list[RewriteStep] = []
 
+  # TODO: move these to codegen
+  ret.append(RewriteStep(view_left, name="Main View Left"))
+  ret.append(RewriteStep(view_right, name="Main View Right"))
+  ret.append(RewriteStep(view_left+fix_kernel_ops, bottom_up=True, name="replace buffer"))
+
   # this is kernel.py
   ret.append(RewriteStep(pm_optimize, ctx=lambda _: opts, name="optimize ast"))
 

tinygrad/schedule/kernelize.py

@@ -7,7 +7,7 @@ from tinygrad.helpers import Metadata, all_int, all_same, prod, dedup, unwrap, g
 from tinygrad.dtype import ImageDType
 from tinygrad.schedule.multi import multi_pm
 from tinygrad.schedule.grouper import group_realizes, ALWAYS_CONTIGUOUS
-from tinygrad.opt.swizzler import merge_views, view_left, view_right, fix_kernel_ops, apply_swizzle, swizzle_reduceop
+from tinygrad.opt.swizzler import merge_views, apply_swizzle, swizzle_reduceop
 
 # creation can recurse a lot
 import sys
@@ -177,10 +177,6 @@ def fix_kernel_ast(k:UOp) -> UOp|None:
   ast = graph_rewrite(ast, early_buffer_ops, bufs, bottom_up=True, name="replace buffer early")
   if ast.op is Ops.SINK and not all_same([x.device for x in k.src]):
     raise RuntimeError(f"all buffers must be on the same device: {tuple(b.buf_uop.buffer for b in k.src)}")
-  # TODO: move these to codegen
-  ast = graph_rewrite(ast, view_left, name="Main View Left")
-  ast = graph_rewrite(ast, view_right, name="Main View Right")
-  ast = graph_rewrite(ast, view_left+fix_kernel_ops, bottom_up=True, name="replace buffer")
   return k.replace(arg=Kernel(ast, k.arg.metadata))
 
 create_ast = PatternMatcher([(UPat(Ops.KERNEL, name="k"), fix_kernel_ast),])