process replay for test/* (#4799)

* add input to unit tests [run_process_replay]

* add setup [run_process_replay]

* run tests [run_process_replay]

* add cuda and amd [run_process_replay]

* run everything but BEAM=2 [run_process_replay]

* skip export_model [run_process_replay]

* fix amd CI

* add concurrency back

test/test_search.py

@@ -8,7 +8,7 @@ from tinygrad.device import Device, Buffer
 from tinygrad.ops import LazyOp, LoadOps, BufferOps, ReduceOps, BinaryOps, MemBuffer, ConstBuffer
 from tinygrad.tensor import Tensor
 from tinygrad.dtype import dtypes
-from tinygrad.helpers import Context, GlobalCounters
+from tinygrad.helpers import Context, GlobalCounters, getenv
 from tinygrad.engine.realize import capturing
 from tinygrad.shape.shapetracker import ShapeTracker
 from tinygrad.shape.view import View
@@ -43,6 +43,7 @@ class TestTimeLinearizer(unittest.TestCase):
     time_linearizer(lin, bufs, allow_test_size=False, cnt=2, disable_cache=True, clear_l2=True)
     assert GlobalCounters.kernel_count == kernel_count, "kernel count was incremented by time_linearizer"
 
+@unittest.skipIf(getenv("RUN_PROCESS_REPLAY"), "TODO: run process replay for BEAM=2")
 class TestBEAM(unittest.TestCase):
   def test_dynamic_beam(self):
     # TODO: make this infra globally usable

test/testextra/test_export_model.py

@@ -1,5 +1,6 @@
 import unittest
 from extra.export_model import export_model, EXPORT_SUPPORTED_DEVICE
+from tinygrad.helpers import getenv
 from tinygrad.tensor import Tensor, Device
 import json
 
@@ -13,6 +14,7 @@ class MockMultiOutputModel:
 
 # TODO: move compile_efficientnet tests here
 @unittest.skipUnless(Device.DEFAULT in EXPORT_SUPPORTED_DEVICE, f"Model export is not supported on {Device.DEFAULT}")
+@unittest.skipIf(getenv("RUN_PROCESS_REPLAY"), "TODO: kernel ordering is non-deterministic")
 class TextModelExport(unittest.TestCase):
   def test_multi_input_model_export(self):
     model = MockMultiInputModel()