[SPEC] Add Triton 3.7 comments

Author: zhzhcookie

third_party/mthreads/backend/spec/triton/__init__.py

@@ -6,5 +6,5 @@
 
 def triton_extend_globals(globals_dict):
     # NOTE: Must use absolute path import.
-    from triton.compiler.compiler import max_shared_mem
-    globals_dict["max_shared_mem"] = max_shared_mem
+    from triton.compiler.compiler import max_shared_mem  # Triton 3.7
+    globals_dict["max_shared_mem"] = max_shared_mem  # Triton 3.7

third_party/mthreads/backend/spec/triton/experimental/gluon/language/__init__.py

@@ -81,7 +81,7 @@
     static_range,
     store,
     sub,
-    barrier,
+    barrier,  # Triton 3.7: thread_barrier() => barrier(*, cluster=False)
     to_linear_layout,
     to_tensor,
     warp_specialize,

third_party/mthreads/backend/spec/triton/language/__init__.py

@@ -1,10 +1,10 @@
 def language_extend_globals(globals_dict):
     # NOTE: Must use absolute path import.
-    from triton.language.standard import squeeze, unsqueeze
+    from triton.language.standard import squeeze, unsqueeze  # Triton 3.7
+    from triton.language.core import to_tensor  # Triton 3.7
     from triton.language.core import _experimental_descriptor_load, _experimental_descriptor_store
-    from triton.language.core import to_tensor
-    globals_dict["squeeze"] = squeeze
-    globals_dict["unsqueeze"] = unsqueeze
+    globals_dict["squeeze"] = squeeze  # Triton 3.7
+    globals_dict["unsqueeze"] = unsqueeze  # Triton 3.7
+    globals_dict["to_tensor"] = to_tensor  # Triton 3.7
     globals_dict["_experimental_descriptor_load"] = _experimental_descriptor_load
     globals_dict["_experimental_descriptor_store"] = _experimental_descriptor_store
-    globals_dict["to_tensor"] = to_tensor