add "tinygrad" to be part of cache_dir [pr] (#8188)

instead of having sqlite / http download / metal compile to add "tinygrad" separately. also make it non-private since it's used in metal

tinygrad/runtime/ops_metal.py

@@ -1,7 +1,7 @@
 from __future__ import annotations
 import os, pathlib, struct, ctypes, tempfile, functools
 from typing import List, Any, Union, Tuple, cast
-from tinygrad.helpers import prod, to_mv, getenv, round_up, _cache_dir, T, init_c_struct_t
+from tinygrad.helpers import prod, to_mv, getenv, round_up, cache_dir, T, init_c_struct_t
 from tinygrad.device import Compiled, Compiler, CompileError, LRUAllocator
 from tinygrad.renderer.cstyle import MetalRenderer
 
@@ -84,7 +84,7 @@ class MetalCompiler(Compiler):
         ret = CompileError(errorMessage.decode())
     # llvm will create modules.timestamp in cache path and cache compilation of metal stdlib (250ms => 8ms compilation time)
     # note that llvm won't necessarily create anything else here as apple has prebuilt versions of many standard libraries
-    params = f'-fno-fast-math -std=metal3.1 --driver-mode=metal -x metal -fmodules-cache-path="{os.path.join(_cache_dir, "tinygrad")}"'
+    params = f'-fno-fast-math -std=metal3.1 --driver-mode=metal -x metal -fmodules-cache-path="{cache_dir}"'
     # source blob has to be padded to multiple of 4 but at least one 'b\x00' should be added, params blob just has to be null terminated
     src_padded, params_padded = src.encode() + b'\x00'*(round_up(len(src) + 1, 4) - len(src)), params.encode() + b'\x00'
     request = struct.pack('<QQ', len(src_padded), len(params_padded)) + src_padded + params_padded