Tweak loop bound generation

loopy/isl_helpers.py

@@ -204,10 +204,11 @@ def static_extremum_of_pw_aff(pw_aff, constants_only, set_method, what, context)
                     % (what, pw_aff))
         return result
 
-    # put constant bounds first
+    # Put constant bounds last, they're usually too restrictive.
     pieces = (
-            [(set, aff) for set, aff in pieces if aff.is_cst()]
-            + [(set, aff) for set, aff in pieces if not aff.is_cst()])
+            [(set, aff) for set, aff in pieces if not aff.is_cst()]
+            + [(set, aff) for set, aff in pieces if aff.is_cst()]
+            )
 
     reference = pw_aff.get_aggregate_domain()
     if context is not None:

test/test_loopy.py

@@ -1620,7 +1620,8 @@ def test_slab_decomposition_does_not_double_execute(ctx_factory):
 
     knl = lp.make_kernel(ctx.devices[0],
         "{ [i]: 0<=i<n }",
-        "a[i] = 2*a[i]")
+        "a[i] = 2*a[i]",
+        assumptions="n>=1")
 
     ref_knl = knl