move DEBUG to helpers

accel/triton/ops_triton.py

@@ -10,9 +10,9 @@ import triton # type: ignore # noqa: F401
 import triton.language as tl  # type: ignore # noqa: F401
 
 from typing import Union, Tuple, Optional, Dict
-from tinygrad.ops import UnaryOps, BinaryOps, ReduceOps, LazyOp, Op, ExplicitExecAST, DEBUG, GlobalCounters
+from tinygrad.ops import UnaryOps, BinaryOps, ReduceOps, LazyOp, Op, ExplicitExecAST, GlobalCounters
 from tinygrad.shape import ShapeTracker
-from tinygrad.helpers import prod
+from tinygrad.helpers import prod, DEBUG
 from tinygrad.runtime.cuda import CLBuffer
 from tinygrad.ast import ASTKernel
 

extra/jit.py

@@ -2,7 +2,8 @@ from typing import Callable, List, Tuple
 import itertools
 from tinygrad.lazy import Device
 from tinygrad.tensor import Tensor
-from tinygrad.ops import DEBUG, GlobalCounters
+from tinygrad.helpers import DEBUG
+from tinygrad.ops import GlobalCounters
 
 class TinyJit:
   def __init__(self, fxn):

extra/kernel_search.py

@@ -8,8 +8,7 @@ from tinygrad.ops import LazyOp, ReduceOps, BinaryOps, UnaryOps, MovementOps
 from tinygrad.shape import ShapeTracker, View, ZeroView
 from tinygrad.llops.ops_gpu import GPUBuffer, CLASTKernel
 from tinygrad.runtime.opencl import OSX_TIMING_RATIO
-from tinygrad.ops import DEBUG
-from tinygrad.helpers import getenv
+from tinygrad.helpers import getenv, DEBUG
 from extra.lib_test_ast import test_ast
 
 import pickle, dbm

extra/onnx.py

@@ -1,8 +1,7 @@
 import numpy as np
 from tinygrad.tensor import Tensor
 from tinygrad.helpers import prod
-from tinygrad.ops import DEBUG
-from tinygrad.helpers import getenv
+from tinygrad.helpers import getenv, DEBUG
 from onnx.mapping import TENSOR_TYPE_TO_NP_TYPE
 
 ONNXLIMIT = getenv("ONNXLIMIT", -1)

tinygrad/graph.py

@@ -7,8 +7,8 @@ except ImportError:
   nx = None # graph won't work
 from collections import defaultdict
 from typing import Dict, List, Optional
-from tinygrad.ops import DeviceBuffer, DEBUG, UnaryOps, BinaryOps, ReduceOps, MovementOps, ProcessingOps, LoadOps, Op, OpType, LazyOp, get_buffers, get_lazyops
-from tinygrad.helpers import getenv
+from tinygrad.ops import DeviceBuffer, UnaryOps, BinaryOps, ReduceOps, MovementOps, ProcessingOps, LoadOps, Op, OpType, LazyOp, get_buffers, get_lazyops
+from tinygrad.helpers import getenv, DEBUG
 
 GRAPH, PRUNEGRAPH, GRAPHPATH = getenv("GRAPH", 0), getenv("PRUNEGRAPH", 0), getenv("GRAPHPATH", "/tmp/net")
 

tinygrad/helpers.py

@@ -17,6 +17,7 @@ class Timing(object):
 
 @functools.lru_cache(maxsize=None)
 def getenv(key, default=0): return type(default)(os.getenv(key, default))
+DEBUG = getenv("DEBUG", 0)
 
 def reduce_shape(shape, axis): return tuple(1 if i in axis else shape[i] for i in range(len(shape)))
 def shape_to_axis(old_shape, new_shape):

tinygrad/lazy.py

@@ -2,9 +2,9 @@ from __future__ import annotations
 from typing import Optional, Tuple, Union, List, Dict, Any, ClassVar, Type
 import sys, weakref, os, importlib, inspect
 from weakref import WeakValueDictionary
-from tinygrad.helpers import ConvArgs, prod
+from tinygrad.helpers import ConvArgs, prod, DEBUG
 from tinygrad.shape import ShapeTracker
-from tinygrad.ops import DeviceBuffer, UnaryOps, BinaryOps, ReduceOps, MovementOps, ProcessingOps, LoadOps, OpType, LazyOp, get_buffers, map_buffers, DEBUG, GenericExecAST
+from tinygrad.ops import DeviceBuffer, UnaryOps, BinaryOps, ReduceOps, MovementOps, ProcessingOps, LoadOps, OpType, LazyOp, get_buffers, map_buffers, GenericExecAST
 from tinygrad.graph import log_op
 from tinygrad.helpers import getenv
 

tinygrad/llops/ops_gpu.py

@@ -1,8 +1,8 @@
 from __future__ import annotations
 import numpy as np
 from typing import List, Tuple, Optional, Dict, Union, Set, Final, Callable
-from tinygrad.helpers import prod
-from tinygrad.ops import DEBUG, UnaryOps, BinaryOps, ReduceOps, MovementOps, LazyOp, Op, ExplicitExecAST, GlobalCounters
+from tinygrad.helpers import prod, DEBUG
+from tinygrad.ops import UnaryOps, BinaryOps, ReduceOps, MovementOps, LazyOp, Op, ExplicitExecAST, GlobalCounters
 from tinygrad.ast import ASTKernel, Token, Types
 from tinygrad.lazy import IMAGE
 from tinygrad.shape import ShapeTracker

tinygrad/llops/ops_llvm.py

@@ -2,13 +2,13 @@ from __future__ import annotations
 import math
 import functools
 from typing import Tuple, Union, Dict, Any, List, ClassVar, Optional
-from tinygrad.helpers import prod
+from tinygrad.helpers import prod, DEBUG
 from tinygrad.shape import ShapeTracker
 from tinygrad.ops import LazyOp
 from tinygrad.ast import ASTKernel
 import ctypes
 import numpy as np
-from tinygrad.ops import DEBUG, UnaryOps, BinaryOps, ReduceOps, ExplicitExecAST
+from tinygrad.ops import UnaryOps, BinaryOps, ReduceOps, ExplicitExecAST
 from tinygrad.runtime.llvm import LLVM, ir
 from tinygrad.shape.symbolic import Variable, NumNode, MulNode, DivNode, ModNode, GeNode, LtNode, SumNode, AndNode
 

tinygrad/ops.py

@@ -5,9 +5,6 @@ from typing import Union, Type, NamedTuple, Tuple, Any, List, ClassVar, Optional
 import functools, operator
 from tinygrad.helpers import prod
 from tinygrad.shape import ShapeTracker
-from tinygrad.helpers import getenv
-
-DEBUG = getenv("DEBUG", 0)
 
 # these are the llops your accelerator must implement, along with toCpu
 # the Enum class doesn't work with mypy, this is static. sorry it's ugly

tinygrad/runtime/clang.py

@@ -5,7 +5,7 @@ import hashlib
 import subprocess
 from collections import defaultdict
 from typing import List, Final, Dict
-from tinygrad.ops import DEBUG
+from tinygrad.helpers import DEBUG
 import platform
 OSX = platform.system() == "Darwin"
 

tinygrad/runtime/cuda.py

@@ -3,7 +3,8 @@ import pycuda.autoprimaryctx # type: ignore # pylint: disable=unused-import # no
 import pycuda.driver as cuda # type: ignore
 from pycuda.compiler import compile # type: ignore
 import numpy as np
-from tinygrad.ops import DEBUG, GlobalCounters
+from tinygrad.helpers import DEBUG
+from tinygrad.ops import GlobalCounters
 
 class CLBuffer:
   def __init__(self, size): self._cl = cuda.mem_alloc(size)

tinygrad/runtime/llvm.py

@@ -1,6 +1,6 @@
 from typing import ClassVar
-from tinygrad.helpers import getenv
-from tinygrad.ops import DEBUG, GlobalCounters
+from tinygrad.helpers import getenv, DEBUG
+from tinygrad.ops import GlobalCounters
 import hashlib
 import time
 import ctypes

tinygrad/runtime/metal.py

@@ -2,8 +2,8 @@
 import Metal, Cocoa, libdispatch # type: ignore
 import numpy as np
 from typing import List, Any
-from tinygrad.ops import DEBUG, GlobalCounters
-from tinygrad.helpers import prod, getenv
+from tinygrad.ops import GlobalCounters
+from tinygrad.helpers import prod, getenv, DEBUG
 import subprocess
 
 METAL_XCODE = getenv("METAL_XCODE")

tinygrad/runtime/opencl.py

@@ -3,8 +3,8 @@ import numpy as np
 import pyopencl as cl  # type: ignore
 from typing import Dict, Optional, Tuple, List, ClassVar, Final
 from collections import defaultdict
-from tinygrad.ops import DEBUG, GlobalCounters
-from tinygrad.helpers import getenv
+from tinygrad.ops import GlobalCounters
+from tinygrad.helpers import getenv, DEBUG
 
 OSX = platform.system() == "Darwin"
 OSX_TIMING_RATIO = (125/3) if OSX else 1.0   # see test/external_osx_profiling.py to determine this ratio. it's in like GPU clocks or something

tinygrad/shape/__init__.py

@@ -2,12 +2,9 @@
 from __future__ import annotations
 import functools
 from typing import Tuple, Union, List, Optional
-from tinygrad.helpers import prod, getenv
+from tinygrad.helpers import prod, DEBUG
 from tinygrad.shape.symbolic import Variable
 
-# TODO: fix DEBUG import
-DEBUG = getenv("DEBUG", 0)
-
 @functools.lru_cache(maxsize=None)
 def to_shape_strides(shape:Tuple[int, ...], strides:Tuple[int, ...]) -> List[Tuple[int, int]]:
   assert len(shape) == len(strides)

tinygrad/tensor.py

@@ -5,7 +5,7 @@ import numpy as np
 from tinygrad.helpers import prod, argfix, make_pair, getenv
 from typing import List, Tuple, Callable, Optional, ClassVar, Type, Union
 from tinygrad.lazy import Device, LazyBuffer
-from tinygrad.ops import DEBUG
+from tinygrad.helpers import DEBUG
 
 # An instantiation of the Function is the Context
 class Function: