reproed failure in emulation

.github/workflows/test.yml

@@ -272,6 +272,8 @@ jobs:
     #   run: NULL=1 DEBUG=1 python3 examples/sdxl.py --seed 0 --noshow --timing --fakeweights
     - name: Run Clip tests for SD MLPerf on NULL backend
       run: NULL=1 python -m pytest -n=auto test/external/mlperf_stable_diffusion/external_test_models.py::TestOpenClip --durations=20
+    - name: Run AMD emulated BERT training on NULL backend
+      run: EMULATE=AMD_RDNA4 NULL=1 CAPTURE_PROCESS_REPLAY=0 DEFAULT_FLOAT=HALF BENCHMARK=10 BS=66 GPUS=1 BERT_LAYERS=2 MODEL=bert python3 examples/mlperf/model_train.py
     # TODO: support fake weights
     #- name: Run LLaMA 7B on 4 fake devices
     #  run: NULL=1 python3 examples/llama.py --gen 1 --size 7B --shard 4 --prompt "Hello." --count 3 --temperature 0 --timing

tinygrad/runtime/ops_null.py

@@ -2,7 +2,8 @@ import functools
 from typing import cast
 from tinygrad.device import Compiled, Compiler, Allocator
 from tinygrad.engine.jit import MultiGraphRunner
-from tinygrad.renderer.cstyle import Renderer, CStyleLanguage, AMDRenderer
+from tinygrad.renderer.cstyle import Renderer, CStyleLanguage
+from tinygrad.renderer.llvmir import AMDLLVMRenderer
 from tinygrad.uop.ops import Ops
 from tinygrad.helpers import cpu_profile, EMULATE
 
@@ -33,7 +34,8 @@ class NullDevice(Compiled):
   def __init__(self, device:str):
     renderer:functools.partial|type[Renderer]
     match cast(str, EMULATE.value):
-      case "AMD": renderer = functools.partial(AMDRenderer, "gfx1100")
+      case "AMD": renderer = functools.partial(AMDLLVMRenderer, "gfx1100")
+      case "AMD_RDNA4": renderer = functools.partial(AMDLLVMRenderer, "gfx1201")
       case "": renderer = NullRenderer
       case _: raise RuntimeError(f"can't EMULATE device: {EMULATE.value}")
     super().__init__(device, NullAllocator(self), [(renderer, Compiler)], functools.partial(NullProgram, device), NullGraph)