diff --git a/sumpy/p2p.py b/sumpy/p2p.py
index e3b457dd52e885bf66462e475b3b58edb78bd098..c585d0eae7ec73de1231850b3b3be628ba9e3d4d 100644
--- a/sumpy/p2p.py
+++ b/sumpy/p2p.py
@@ -33,7 +33,8 @@ import loopy as lp
 from loopy.version import MOST_RECENT_LANGUAGE_VERSION
 from pymbolic import var
 
-from sumpy.tools import KernelComputation, KernelCacheWrapper
+from sumpy.tools import (
+        KernelComputation, KernelCacheWrapper, is_obj_array_like)
 
 
 __doc__ = """
@@ -214,12 +215,9 @@ class P2P(P2PBase):
         return loopy_knl
 
     def __call__(self, queue, targets, sources, strength, **kwargs):
-        from pytools.obj_array import is_obj_array
         knl = self.get_cached_optimized_kernel(
-                targets_is_obj_array=(
-                    is_obj_array(targets) or isinstance(targets, (tuple, list))),
-                sources_is_obj_array=(
-                    is_obj_array(sources) or isinstance(sources, (tuple, list))))
+                targets_is_obj_array=is_obj_array_like(targets),
+                sources_is_obj_array=is_obj_array_like(sources))
 
         return knl(queue, sources=sources, targets=targets, strength=strength,
                 **kwargs)
@@ -278,12 +276,9 @@ class P2PMatrixGenerator(P2PBase):
         return loopy_knl
 
     def __call__(self, queue, targets, sources, **kwargs):
-        from pytools.obj_array import is_obj_array
         knl = self.get_cached_optimized_kernel(
-                targets_is_obj_array=(
-                    is_obj_array(targets) or isinstance(targets, (tuple, list))),
-                sources_is_obj_array=(
-                    is_obj_array(sources) or isinstance(sources, (tuple, list))))
+                targets_is_obj_array=is_obj_array_like(targets),
+                sources_is_obj_array=is_obj_array_like(sources))
 
         return knl(queue, sources=sources, targets=targets, **kwargs)
 
@@ -390,12 +385,9 @@ class P2PMatrixBlockGenerator(P2PBase):
         :return: a tuple of one-dimensional arrays of kernel evaluations at
             target-source pairs described by `index_set`.
         """
-        from pytools.obj_array import is_obj_array
         knl = self.get_cached_optimized_kernel(
-                targets_is_obj_array=(
-                    is_obj_array(targets) or isinstance(targets, (tuple, list))),
-                sources_is_obj_array=(
-                    is_obj_array(sources) or isinstance(sources, (tuple, list))))
+                targets_is_obj_array=is_obj_array_like(targets),
+                sources_is_obj_array=is_obj_array_like(sources))
 
         return knl(queue,
                    targets=targets,
diff --git a/sumpy/qbx.py b/sumpy/qbx.py
index 9708764c01d0448d7c7e8314efb461989c838acd..dd6a827928832aadf74d6108df0489950836cc59 100644
--- a/sumpy/qbx.py
+++ b/sumpy/qbx.py
@@ -35,7 +35,8 @@ import sumpy.symbolic as sym
 from pytools import memoize_method
 from pymbolic import parse, var
 
-from sumpy.tools import KernelComputation, KernelCacheWrapper
+from sumpy.tools import (
+        KernelComputation, KernelCacheWrapper, is_obj_array_like)
 
 import logging
 logger = logging.getLogger(__name__)
@@ -154,7 +155,7 @@ class LayerPotentialBase(KernelComputation, KernelCacheWrapper):
                 lp.GlobalArg("tgt", None,
                     shape=(self.dim, "ntargets"), order="C"),
                 lp.GlobalArg("center", None,
-                    shape=(self.dim, "ntargets"), dim_tags="sep,C"),
+                    shape=(self.dim, "ntargets"), order="C"),
                 lp.GlobalArg("expansion_radii",
                     None, shape="ntargets"),
                 lp.ValueArg("nsources", None),
@@ -164,10 +165,18 @@ class LayerPotentialBase(KernelComputation, KernelCacheWrapper):
     def get_kernel(self):
         raise NotImplementedError
 
-    def get_optimized_kernel(self):
+    def get_optimized_kernel(self,
+            targets_is_obj_array, sources_is_obj_array, centers_is_obj_array):
         # FIXME specialize/tune for GPU/CPU
         loopy_knl = self.get_kernel()
 
+        if targets_is_obj_array:
+            loopy_knl = lp.tag_array_axes(loopy_knl, "tgt", "sep,C")
+        if sources_is_obj_array:
+            loopy_knl = lp.tag_array_axes(loopy_knl, "src", "sep,C")
+        if centers_is_obj_array:
+            loopy_knl = lp.tag_array_axes(loopy_knl, "center", "sep,C")
+
         import pyopencl as cl
         dev = self.context.devices[0]
         if dev.type & cl.device_type.CPU:
@@ -249,7 +258,10 @@ class LayerPotential(LayerPotentialBase):
             already multiplied in.
         """
 
-        knl = self.get_cached_optimized_kernel()
+        knl = self.get_cached_optimized_kernel(
+                targets_is_obj_array=is_obj_array_like(targets),
+                sources_is_obj_array=is_obj_array_like(sources),
+                centers_is_obj_array=is_obj_array_like(centers))
 
         for i, dens in enumerate(strengths):
             kwargs["strength_%d" % i] = dens
@@ -313,7 +325,10 @@ class LayerPotentialMatrixGenerator(LayerPotentialBase):
         return loopy_knl
 
     def __call__(self, queue, targets, sources, centers, expansion_radii, **kwargs):
-        knl = self.get_cached_optimized_kernel()
+        knl = self.get_cached_optimized_kernel(
+                targets_is_obj_array=is_obj_array_like(targets),
+                sources_is_obj_array=is_obj_array_like(sources),
+                centers_is_obj_array=is_obj_array_like(centers))
 
         return knl(queue, src=sources, tgt=targets, center=centers,
                 expansion_radii=expansion_radii, **kwargs)
@@ -388,9 +403,17 @@ class LayerPotentialMatrixBlockGenerator(LayerPotentialBase):
 
         return loopy_knl
 
-    def get_optimized_kernel(self):
+    def get_optimized_kernel(self,
+            targets_is_obj_array, sources_is_obj_array, centers_is_obj_array):
         loopy_knl = self.get_kernel()
 
+        if targets_is_obj_array:
+            loopy_knl = lp.tag_array_axes(loopy_knl, "tgt", "sep,C")
+        if sources_is_obj_array:
+            loopy_knl = lp.tag_array_axes(loopy_knl, "src", "sep,C")
+        if centers_is_obj_array:
+            loopy_knl = lp.tag_array_axes(loopy_knl, "center", "sep,C")
+
         loopy_knl = lp.split_iname(loopy_knl, "imat", 1024, outer_tag="g.0")
         return loopy_knl
 
@@ -406,8 +429,10 @@ class LayerPotentialMatrixBlockGenerator(LayerPotentialBase):
         :return: a tuple of one-dimensional arrays of kernel evaluations at
             target-source pairs described by `index_set`.
         """
-
-        knl = self.get_cached_optimized_kernel()
+        knl = self.get_cached_optimized_kernel(
+                targets_is_obj_array=is_obj_array_like(targets),
+                sources_is_obj_array=is_obj_array_like(sources),
+                centers_is_obj_array=is_obj_array_like(centers))
 
         return knl(queue,
                    src=sources,
diff --git a/sumpy/tools.py b/sumpy/tools.py
index 4d1098429d1c9db43af4a6e26fbb63509438a375..353ca9dac95673863fa735bd685267257443acb1 100644
--- a/sumpy/tools.py
+++ b/sumpy/tools.py
@@ -717,4 +717,9 @@ def my_syntactic_subs(expr, subst_dict):
         return expr
 
 
+def is_obj_array_like(ary):
+    return (
+            isinstance(ary, (tuple, list))
+            or (isinstance(ary, np.ndarray) and ary.dtype.char == "O"))
+
 # vim: fdm=marker
diff --git a/sumpy/version.py b/sumpy/version.py
index f44cfe20283522f87e8ebb10d41fff268d205080..ced3319c9da32decaffd911398df47c239e034b9 100644
--- a/sumpy/version.py
+++ b/sumpy/version.py
@@ -43,7 +43,7 @@ else:
 # }}}
 
 
-VERSION = (2020, 1)
+VERSION = (2020, 2)
 VERSION_STATUS = "beta1"
 VERSION_TEXT = ".".join(str(x) for x in VERSION) + VERSION_STATUS