disable process replay for AMD emulator renderer [pr] (#14766)

* disable process replay for AMD emulator renderer [pr]

* line

* skip

.github/workflows/test.yml

@@ -837,6 +837,8 @@ jobs:
         NV_PTX: 1
         NV: 1
         FORWARD_ONLY: 1
+        # TODO: failing due to library loading error
+        CAPTURE_PROCESS_REPLAY: 0
       run: |
         python3 -m pytest -n=auto test/device/test_hcq.py test/test_tiny.py --durations=20
     - name: Run process replay tests

test/external/process_replay/process_replay.py

@@ -88,7 +88,8 @@ def diff(offset:int, fxns:dict[str, Callable[..., tuple|None]]) -> None:
     name, loc = "", ""
     try:
       name, args, kwargs, ctx_vals, loc, ret = pickle.loads(row[0])
-      ctx_vars = {k:v.value for k,v in ctx_vals.items() if k != "DEBUG" and (var:=ContextVar._cache.get(k)) is not None and var.value != v.value}
+      ctx_vars = {k:v.value for k,v in ctx_vals.items() if k not in ("DEBUG", "CAPTURE_PROCESS_REPLAY")
+                  and (var:=ContextVar._cache.get(k)) is not None and var.value != v.value}
       if (replayer:=fxns.get(name)) is None: continue
       with Context(**ctx_vars):
         if (ret:=replayer(ret, *args, **kwargs)) is None: continue

test/mockgpu/amd/emu.py

@@ -1178,7 +1178,8 @@ def _get_runner(inst_bytes: bytes, arch: str = "rdna3"):
   canonical_name = f"{_op_name(inst).lower()}_{base.to_bytes(size, 'little').hex()}"
   sink = sink.replace(arg=KernelInfo(name=canonical_name)).rtag(1)
 
-  with Context(NOOPT=1, CHECK_OOB=0, TUPLE_ORDER=0, EMULATED_DTYPES=""):
+  # NOTE: renderer output is not reproducible because of _MXCSRContext
+  with Context(NOOPT=1, CHECK_OOB=0, TUPLE_ORDER=0, EMULATED_DTYPES="", CAPTURE_PROCESS_REPLAY=0):
     runner = get_runner('CPU', sink)
   _canonical_runner_cache.append((base, mask, size, runner))
   return runner

tinygrad/helpers.py

@@ -184,6 +184,7 @@ CORRECT_DIVMOD_FOLDING, FUSE_OPTIM = ContextVar("CORRECT_DIVMOD_FOLDING", 0), Co
 ALLOW_DEVICE_USAGE, MAX_BUFFER_SIZE = ContextVar("ALLOW_DEVICE_USAGE", 1), ContextVar("MAX_BUFFER_SIZE", 0)
 MAX_KERNEL_BUFFERS = ContextVar("MAX_KERNEL_BUFFERS", 0)
 EMULATE, EMULATED_DTYPES = ContextVar("EMULATE", ""), ContextVar("EMULATED_DTYPES", "")
+CAPTURE_PROCESS_REPLAY = ContextVar("CAPTURE_PROCESS_REPLAY", 0)
 CPU_COUNT = ContextVar("CPU_COUNT", max(1, len(os.sched_getaffinity(0)) if hasattr(os, "sched_getaffinity") else (os.cpu_count() or 1)))
 # Compilers
 CPU_CC, CPU_LLVM, CPU_LVP = ContextVar("CPU_CC", ""), ContextVar("CPU_LLVM", 0), ContextVar("CPU_LVP", 0)

tinygrad/uop/ops.py

@@ -7,7 +7,7 @@ from tinygrad.uop import Ops, GroupOp
 from tinygrad.dtype import ConstType, ImageDType, dtypes, DType, truncate, PtrDType, least_upper_dtype, Invalid, AddrSpace, ConstFloat, PyConst
 from tinygrad.dtype import storage_fmt_for_dtype, to_storage_scalar, from_storage_scalar
 from tinygrad.helpers import ContextVar, all_int, prod, getenv, all_same, Context, partition, temp, unwrap, T, argfix, Metadata, flatten, TRACEMETA
-from tinygrad.helpers import PROFILE, dedup, cdiv, cmod, diskcache_put, to_function_name, cpu_profile, TracingKey, VIZ, SPEC
+from tinygrad.helpers import PROFILE, dedup, cdiv, cmod, diskcache_put, to_function_name, cpu_profile, TracingKey, VIZ, SPEC, CAPTURE_PROCESS_REPLAY
 from tinygrad.helpers import strip_parens, colored, ansilen, printable, panic
 if TYPE_CHECKING:
   from tinygrad.device import Buffer, MultiBuffer
@@ -1071,7 +1071,7 @@ tracked_keys:list[TracingKey] = []
 tracked_ctxs:list[list[TrackedGraphRewrite]] = []
 _name_cnt:dict[str, itertools.count] = {}
 
-if getenv("CAPTURE_PROCESS_REPLAY"):
+if CAPTURE_PROCESS_REPLAY:
   replay_capture: list[bytes] = []
   import atexit, uuid
   @atexit.register
@@ -1095,7 +1095,7 @@ def track_rewrites(name:Callable[..., str|TracingKey]|bool=True, replay:bool=Fal
         assert isinstance(name_ret, (TracingKey, str)), f"name function returned {type(name_ret)}"
         tracked_keys[-1] = k = TracingKey(n:=tracked_keys[-1].display_name.replace(fn, name_ret), (n,)) if isinstance(name_ret, str) else name_ret
         e.name = TracingKey(k.display_name if isinstance(name_ret, str) else f"{fn} for {k.display_name}", k.keys)
-      if getenv("CAPTURE_PROCESS_REPLAY") and replay:
+      if CAPTURE_PROCESS_REPLAY and replay:
         # find the unittest frame we're capturing in
         frm = sys._getframe(1)
         while (f_back:=frm.f_back) is not None and "unittest" not in f_back.f_code.co_filename: frm = f_back