fix memory corruption in NIR, reenable process replay (#14204)

.github/workflows/test.yml

@@ -759,7 +759,6 @@ jobs:
       - name: Run TRANSCENDENTAL math
         run: TRANSCENDENTAL=2 python -m pytest -n=auto test/test_ops.py::TestOps::test_sin test/test_ops.py::TestOps::test_cos test/test_ops.py::TestOps::test_tan test/test_ops.py::TestOps::test_exp test/test_ops.py::TestOps::test_log --durations=20
       - name: Run process replay tests
-        if: matrix.backend != 'lvp'
         uses: ./.github/actions/process-replay
 
 # ****** OSX Tests ******
@@ -889,7 +888,6 @@ jobs:
       - name: Run pytest (${{ matrix.backend }})
         run: python3 -m pytest -n=auto test/ --ignore=test/models --ignore=test/unit --durations=20
       - name: Run process replay tests
-        if: matrix.backend != 'lvp'
         uses: ./.github/actions/process-replay
       - name: Run macOS-specific unit test
         if: matrix.backend == 'cpu'

tinygrad/helpers.py

@@ -469,7 +469,6 @@ def to_mv(ptr:int, sz:int) -> memoryview: return memoryview((ctypes.c_uint8 * sz
 def mv_address(mv): return ctypes.addressof(ctypes.c_char.from_buffer(mv))
 def to_char_p_p(options: list[bytes], to_type=ctypes.c_char):
   return (ctypes.POINTER(to_type) * len(options))(*[ctypes.cast(ctypes.create_string_buffer(o), ctypes.POINTER(to_type)) for o in options])
-def charptr(s:str|bytes): return ctypes.cast(ctypes.c_char_p(s if isinstance(s, bytes) else s.encode()), ctypes.POINTER(ctypes.c_char))
 def flat_mv(mv:memoryview): return mv if len(mv) == 0 else mv.cast("B", shape=(mv.nbytes,))
 
 # *** tqdm

tinygrad/renderer/nir.py

@@ -1,10 +1,11 @@
 from typing import Callable, cast, Any
 from tinygrad.dtype import AddrSpace, DType, PtrDType, ImageDType, dtypes
-from tinygrad.helpers import DEBUG, OSX, unwrap, charptr, fromimport
+from tinygrad.helpers import DEBUG, OSX, unwrap, fromimport
 from tinygrad.renderer import Renderer
 from tinygrad.renderer.cstyle import CUDARenderer
 from tinygrad.uop.ops import GroupOp, Ops, UOp, PatternMatcher, UPat, range_str
 from tinygrad.runtime.autogen import mesa
+from tinygrad.runtime.support.c import POINTER
 import base64, ctypes, ctypes.util, struct, functools, inspect, contextlib, itertools
 
 def g(s:str): return getattr(mesa, s)
@@ -187,7 +188,8 @@ class NIRRenderer(Renderer):
       elif u.op is Ops.AFTER:
         self.r[u] = self.r[u.src[0]]
       elif u.op == Ops.SINK:
-        if u.arg is not None: self.b.shader.contents.info.name = charptr(u.arg.function_name.encode())
+        if u.arg is not None:
+          self.b.shader.contents.info.name = ctypes.cast(ctypes.create_string_buffer(u.arg.function_name.encode()), POINTER[ctypes.c_char])
       elif u.op == Ops.DEFINE_LOCAL:
         self.r[u] = nimm(self.b, self.b.shader.contents.info.shared_size, dtypes.long)
         self.b.shader.contents.info.shared_size += u.dtype.nbytes()