[Triton-MLIR] Minor fixes related with scf/swizzling support (#791)

1, Disable static loop unrolling in the frontend by default;
2, A minor fix in axisAnalysis in order to support scf;
3, A minor fix in TritonGPUToLLVM to support swizzling.

python/triton/compiler.py

@@ -558,7 +558,10 @@ class CodeGenerator(ast.NodeVisitor):
             raise RuntimeError('Only `range` iterator currently supported')
         # static for loops: all iterator arguments are constexpr
         iter_args = [self.visit(arg) for arg in node.iter.args]
-        is_static = all([isinstance(x, triton.language.constexpr) for x in iter_args])
+        static_unrolling = os.environ.get('TRITON_STATIC_LOOP_UNROLLING', False)
+        is_static = False
+        if static_unrolling:
+            is_static = all([isinstance(x, triton.language.constexpr) for x in iter_args])
         if is_static:
             iter_args = [arg.value for arg in iter_args]
             range = iterator(*iter_args)