enabling fp8 gemm on amd

Summary:
currently we were trying to import cublas which was getting a `RuntimeError` instead of the moduleNotFound error. adding the exception to the `try except` clause with logged statements

This was caught originally via CI additions we are doing in the next diff for AMD integration. see https://www.internalfb.com/intern/testinfra/diagnostics/4785074862901046.844425121438624.1733874779/ for the failure that caused this diff

Reviewed By: nmacchioni

Differential Revision: D67065556

fbshipit-source-id: ede7042bc85a0d5d487f33d281cb8ab09cff135e

tritonbench/operators/fp8_gemm/fp8_gemm.py

@@ -1,14 +1,11 @@
 import argparse
+import logging
 import os
-import statistics
 
 from typing import Any, List, Optional
 
 import torch
 import triton
-import triton.language as tl
-
-from triton.runtime.jit import reinterpret
 
 from tritonbench.utils.triton_op import (
     BenchmarkOperator,
@@ -20,6 +17,7 @@
 
 from .tutorial import matmul as tutorial_matmul
 
+logger = logging.getLogger(__name__)
 try:
     from .persistent import (
         allocate_matmul_tma,
@@ -30,6 +28,10 @@
     HAS_TMA = True
 except ModuleNotFoundError:
     HAS_TMA = False
+    logger.warning("Failed to import TMA due to module not being found")
+except Exception as e:
+    HAS_TMA = False
+    logger.warning(f"Failed to import TMA: {e}")
 
 
 def parse_args(args):