rename is_b200 to is_blackwell (#646)

Summary:

NA

Reviewed By: q10, sryap

Differential Revision: D87106508

Author: henrylhtsang

tritonbench/operators/blackwell_attentions/operator.py

@@ -120,7 +120,9 @@
     torch.cuda.is_available() and torch.version.cuda and torch.version.cuda >= "12.4"
 )
 
-IS_B200 = is_cuda() and "B200" in get_nvidia_gpu_model()
+IS_BLACKWELL = is_cuda() and (
+    "B200" in get_nvidia_gpu_model() or "B300" in get_nvidia_gpu_model()
+)
 
 
 def parse_op_args(args: List[str]):
@@ -385,7 +387,7 @@ def xformers_splitk(
         )
 
     @register_benchmark(
-        enabled=IS_B200 and _is_sdpa_cudnn_attention_available(),
+        enabled=IS_BLACKWELL and _is_sdpa_cudnn_attention_available(),
         label=f"cudnn-sdpa-{torch.backends.cudnn.version()}",
     )
     def cudnn_sdpa(self, q, k, v):
@@ -398,7 +400,7 @@ def cudnn_sdpa(self, q, k, v):
         )
 
     @register_benchmark(
-        enabled=(IS_B200 and HAS_FLASH_CUTE), label="FAv4", fwd_only=True
+        enabled=(IS_BLACKWELL and HAS_FLASH_CUTE), label="FAv4", fwd_only=True
     )
     def cutedsl_blackwell(
         self, q: torch.Tensor, k: torch.Tensor, v: torch.Tensor