[CI/Build][AMD] Skip marlin, machete, and hadacore tests since these require...

[CI/Build][AMD] Skip marlin, machete, and hadacore tests since these require _C functions not defined for ROCm (#30109)
Signed-off-by: Randall Smith <ransmith@amd.com>
Co-authored-by: Randall Smith <ransmith@amd.com>

tests/kernels/quantization/test_hadacore.py

@@ -8,6 +8,13 @@ import torch
 from compressed_tensors.transform import deterministic_hadamard_matrix
 
 from vllm import _custom_ops as ops
+from vllm.platforms import current_platform
+
+if current_platform.is_rocm():
+    pytest.skip(
+        "These tests require hadacore_transform, not supported on ROCm.",
+        allow_module_level=True,
+    )
 
 
 @pytest.mark.parametrize("batch_size", [1, 32])

tests/kernels/quantization/test_machete_mm.py

@@ -23,6 +23,12 @@ from vllm.model_executor.layers.quantization.utils.quant_utils import (
 from vllm.platforms import current_platform
 from vllm.scalar_type import ScalarType, scalar_types
 
+if current_platform.is_rocm():
+    pytest.skip(
+        "These tests require machete_prepack_B, not supported on ROCm.",
+        allow_module_level=True,
+    )
+
 CUDA_DEVICES = [f"cuda:{i}" for i in range(1 if torch.cuda.device_count() == 1 else 2)]
 
 # TODO: in future PR refactor this and `is_quant_method_supported` in the kernel

tests/kernels/quantization/test_marlin_gemm.py

@@ -56,6 +56,14 @@ from vllm.model_executor.layers.quantization.utils.quant_utils import (
 from vllm.platforms import current_platform
 from vllm.scalar_type import scalar_types
 
+if current_platform.is_rocm():
+    pytest.skip(
+        "These tests require gptq_marlin_repack,"
+        "marlin_int4_fp8_preprocess, gptq_marlin_24_gemm,"
+        "or gptq_marlin_gemm which are not supported on ROCm.",
+        allow_module_level=True,
+    )
+
 ACT_ORDER_OPTS = [False, True]
 K_FULL_OPTS = [False, True]
 USE_ATOMIC_ADD_OPTS = [False, True]