From f07ebbf8f2c23a3472f232abd78edba29000e83b Mon Sep 17 00:00:00 2001
From: arghdos <arghdos@gmail.com>
Date: Wed, 4 Oct 2017 14:05:23 -0400
Subject: [PATCH 1/3] project out local-id-mapped inames before domain
 comparison for barriers; solves #94 on gitlab

---
 loopy/check.py | 15 +++++++++++++++
 1 file changed, 15 insertions(+)

diff --git a/loopy/check.py b/loopy/check.py
index a8ec1ad35..4b2af1b13 100644
--- a/loopy/check.py
+++ b/loopy/check.py
@@ -708,6 +708,16 @@ def check_implemented_domains(kernel, implemented_domains, code=None):
                 (insn_impl_domain & assumptions)
                 .project_out_except(insn_inames, [dim_type.set]))
 
+        from loopy.kernel.instruction import BarrierInstruction
+        from loopy.kernel.data import LocalIndexTag
+        if isinstance(insn, BarrierInstruction):
+            # project out local-id-mapped inames, solves #94 on gitlab
+            non_lid_inames = frozenset(
+                [iname for iname in insn_inames if not isinstance(
+                    kernel.iname_to_tag.get(iname), LocalIndexTag)])
+            insn_impl_domain = insn_impl_domain.project_out_except(
+                non_lid_inames, [dim_type.set])
+
         insn_domain = kernel.get_inames_domain(insn_inames)
         insn_parameters = frozenset(insn_domain.get_var_names(dim_type.param))
         assumptions, insn_domain = align_two(assumption_non_param, insn_domain)
@@ -715,6 +725,11 @@ def check_implemented_domains(kernel, implemented_domains, code=None):
             .project_out_except(insn_inames, [dim_type.set])
             .project_out_except(insn_parameters, [dim_type.param]))
 
+        if isinstance(insn, BarrierInstruction):
+            # project out local-id-mapped inames, solves #94 on gitlab
+            desired_domain = desired_domain.project_out_except(
+                non_lid_inames, [dim_type.set])
+
         insn_impl_domain = (insn_impl_domain
                 .project_out_except(insn_parameters, [dim_type.param]))
         insn_impl_domain, desired_domain = align_two(
-- 
GitLab


From 0dd58286356f8d7c9b5f21f102865c0d69946388 Mon Sep 17 00:00:00 2001
From: arghdos <arghdos@gmail.com>
Date: Wed, 4 Oct 2017 14:13:45 -0400
Subject: [PATCH 2/3] add test

---
 test/test_loopy.py | 36 ++++++++++++++++++++++++++++++++++++
 1 file changed, 36 insertions(+)

diff --git a/test/test_loopy.py b/test/test_loopy.py
index 563964cf0..3888de0f3 100644
--- a/test/test_loopy.py
+++ b/test/test_loopy.py
@@ -2265,6 +2265,42 @@ def test_barrier_insertion_near_bottom_of_loop():
     assert_barrier_between(knl, "ainit", "aupdate", ignore_barriers_in_levels=[1])
 
 
+def test_barrier_in_overridden_get_grid_size_expanded_kernel():
+    from loopy.kernel.data import temp_var_scope as scopes
+
+    # make simple barrier'd kernel
+    knl = lp.make_kernel('{[i]: 0 <= i < 10}',
+                   """
+              for i
+                    a[i] = i {id=a}
+                    ... lbarrier {id=barrier}
+                    b[i + 1] = a[i] {nosync=a}
+              end
+                   """,
+                   [lp.TemporaryVariable("a", np.float32, shape=(10,), order='C',
+                                         scope=scopes.LOCAL),
+                    lp.GlobalArg("b", np.float32, shape=(11,), order='C')],
+               seq_dependencies=True)
+
+    # split into kernel w/ vesize larger than iname domain
+    vecsize = 16
+    knl = lp.split_iname(knl, 'i', vecsize, inner_tag='l.0')
+
+    # artifically expand via overridden_get_grid_sizes_for_insn_ids
+    class ggs(object):
+        def __init__(self, clean, vecsize=vecsize):
+            self.clean = clean
+            self.vecsize = vecsize
+
+        def __call__(self, insn_ids, ignore_auto=True):
+            gsize, _ = self.clean.get_grid_sizes_for_insn_ids(insn_ids, ignore_auto)
+            return gsize, (self.vecsize,)
+
+    knl = knl.copy(overridden_get_grid_sizes_for_insn_ids=ggs(knl.copy(), vecsize))
+    # make sure we can generate the code
+    lp.generate_code_v2(knl)
+
+
 def test_multi_argument_reduction_type_inference():
     from loopy.type_inference import TypeInferenceMapper
     from loopy.library.reduction import SegmentedSumReductionOperation
-- 
GitLab


From b8cdf5f2246b48653b01c1144c7e09d7d26cf0cf Mon Sep 17 00:00:00 2001
From: arghdos <arghdos@gmail.com>
Date: Wed, 4 Oct 2017 14:22:16 -0400
Subject: [PATCH 3/3] flake8 naming

---
 test/test_loopy.py | 5 +++--
 1 file changed, 3 insertions(+), 2 deletions(-)

diff --git a/test/test_loopy.py b/test/test_loopy.py
index 3888de0f3..97e3a0806 100644
--- a/test/test_loopy.py
+++ b/test/test_loopy.py
@@ -2287,7 +2287,7 @@ def test_barrier_in_overridden_get_grid_size_expanded_kernel():
     knl = lp.split_iname(knl, 'i', vecsize, inner_tag='l.0')
 
     # artifically expand via overridden_get_grid_sizes_for_insn_ids
-    class ggs(object):
+    class GridOverride(object):
         def __init__(self, clean, vecsize=vecsize):
             self.clean = clean
             self.vecsize = vecsize
@@ -2296,7 +2296,8 @@ def test_barrier_in_overridden_get_grid_size_expanded_kernel():
             gsize, _ = self.clean.get_grid_sizes_for_insn_ids(insn_ids, ignore_auto)
             return gsize, (self.vecsize,)
 
-    knl = knl.copy(overridden_get_grid_sizes_for_insn_ids=ggs(knl.copy(), vecsize))
+    knl = knl.copy(overridden_get_grid_sizes_for_insn_ids=GridOverride(
+        knl.copy(), vecsize))
     # make sure we can generate the code
     lp.generate_code_v2(knl)
 
-- 
GitLab