From 141d92d244aadd9db229307cfdbdeaaba543b2cf Mon Sep 17 00:00:00 2001
From: zachjweiner <zweiner2@illinois.edu>
Date: Fri, 30 Nov 2018 17:33:47 -0600
Subject: [PATCH 1/5] andreas's start

---
 loopy/transform/iname.py | 34 ++++++++++++++++++++++++++++++++++
 1 file changed, 34 insertions(+)

diff --git a/loopy/transform/iname.py b/loopy/transform/iname.py
index 83598dcc2..0c0cd215a 100644
--- a/loopy/transform/iname.py
+++ b/loopy/transform/iname.py
@@ -1750,4 +1750,38 @@ def add_inames_to_insn(knl, inames, insn_match):
 # }}}
 
 
+# {{{ nest_and_predicate_instructions
+
+def nest_and_predicate_instructions(kernel, iname, within):
+    from loopy.match import parse_insn_match
+    within = parse_insn_match(within)
+
+    from pymbolic import var
+    from pymbolic.primitives import Comparison
+
+    def rewrite_insn(insn):
+        if within(insn):
+            if iname in insn.within_inames:
+                raise LoopyError("instruction '%s' is already nested "
+                    "within iname '%s'" % (insn.id, iname))
+            return insn.copy(
+                within_inames=insn.within_inames | frozenset([iname]),
+                predicates=(
+                    # FIXME Do a better job finding the lower bound
+                    insn.predicates 
+                    | frozenset(Comparison(var(iname), "==", 0)))
+                )
+        else:
+            return insn
+
+
+    return kernel.copy(
+        instructions=[rewrite_insn(insn) for insn in kernel.instructions]
+        )
+
+
+
+
+# }}}
+
 # vim: foldmethod=marker
-- 
GitLab


From 78428af635509a2b52e4d1cc16d95267c51995d2 Mon Sep 17 00:00:00 2001
From: zachjweiner <zweiner2@illinois.edu>
Date: Wed, 5 Dec 2018 14:00:03 -0600
Subject: [PATCH 2/5] parse_insn_match -> parse_match

---
 loopy/transform/iname.py | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/loopy/transform/iname.py b/loopy/transform/iname.py
index 0c0cd215a..b8fb89d15 100644
--- a/loopy/transform/iname.py
+++ b/loopy/transform/iname.py
@@ -1753,8 +1753,8 @@ def add_inames_to_insn(knl, inames, insn_match):
 # {{{ nest_and_predicate_instructions
 
 def nest_and_predicate_instructions(kernel, iname, within):
-    from loopy.match import parse_insn_match
-    within = parse_insn_match(within)
+    from loopy.match import parse_match
+    within = parse_match(within)
 
     from pymbolic import var
     from pymbolic.primitives import Comparison
-- 
GitLab


From 5aa1ced225e25f4f55fab57cc7156b0ea459fc95 Mon Sep 17 00:00:00 2001
From: zachjweiner <zweiner2@illinois.edu>
Date: Thu, 6 Dec 2018 13:28:18 -0600
Subject: [PATCH 3/5] fixes bugs so that function runs without error

---
 loopy/transform/iname.py | 5 ++---
 1 file changed, 2 insertions(+), 3 deletions(-)

diff --git a/loopy/transform/iname.py b/loopy/transform/iname.py
index b8fb89d15..42deb981d 100644
--- a/loopy/transform/iname.py
+++ b/loopy/transform/iname.py
@@ -1758,9 +1758,8 @@ def nest_and_predicate_instructions(kernel, iname, within):
 
     from pymbolic import var
     from pymbolic.primitives import Comparison
-
     def rewrite_insn(insn):
-        if within(insn):
+        if within(kernel, insn):
             if iname in insn.within_inames:
                 raise LoopyError("instruction '%s' is already nested "
                     "within iname '%s'" % (insn.id, iname))
@@ -1769,7 +1768,7 @@ def nest_and_predicate_instructions(kernel, iname, within):
                 predicates=(
                     # FIXME Do a better job finding the lower bound
                     insn.predicates 
-                    | frozenset(Comparison(var(iname), "==", 0)))
+                    | frozenset([Comparison(var(iname), "==", 0)]))
                 )
         else:
             return insn
-- 
GitLab


From 8fd7ea487366112c6c331c1022e15423870d94fd Mon Sep 17 00:00:00 2001
From: zachjweiner <zweiner2@illinois.edu>
Date: Thu, 6 Dec 2018 13:28:27 -0600
Subject: [PATCH 4/5] add docstring

---
 loopy/transform/iname.py | 6 ++++++
 1 file changed, 6 insertions(+)

diff --git a/loopy/transform/iname.py b/loopy/transform/iname.py
index 42deb981d..dbdde7924 100644
--- a/loopy/transform/iname.py
+++ b/loopy/transform/iname.py
@@ -1753,6 +1753,12 @@ def add_inames_to_insn(knl, inames, insn_match):
 # {{{ nest_and_predicate_instructions
 
 def nest_and_predicate_instructions(kernel, iname, within):
+    """
+    Nests an instruction inside of a loop while only executing on one iteration.
+    Useful for instructions which don't require all local hardware axes.
+    :arg iname: loop iname to nest instruction inside
+    "arg within: instruction to nest and predicate
+    """
     from loopy.match import parse_match
     within = parse_match(within)
 
-- 
GitLab


From 05f3bbeae0cc49a194b930d09020842af553b648 Mon Sep 17 00:00:00 2001
From: zachjweiner <zweiner2@illinois.edu>
Date: Fri, 7 Dec 2018 17:24:07 -0600
Subject: [PATCH 5/5] appease flake8

---
 loopy/transform/iname.py | 6 ++----
 1 file changed, 2 insertions(+), 4 deletions(-)

diff --git a/loopy/transform/iname.py b/loopy/transform/iname.py
index dbdde7924..ddc412d94 100644
--- a/loopy/transform/iname.py
+++ b/loopy/transform/iname.py
@@ -1764,6 +1764,7 @@ def nest_and_predicate_instructions(kernel, iname, within):
 
     from pymbolic import var
     from pymbolic.primitives import Comparison
+
     def rewrite_insn(insn):
         if within(kernel, insn):
             if iname in insn.within_inames:
@@ -1773,20 +1774,17 @@ def nest_and_predicate_instructions(kernel, iname, within):
                 within_inames=insn.within_inames | frozenset([iname]),
                 predicates=(
                     # FIXME Do a better job finding the lower bound
-                    insn.predicates 
+                    insn.predicates
                     | frozenset([Comparison(var(iname), "==", 0)]))
                 )
         else:
             return insn
 
-
     return kernel.copy(
         instructions=[rewrite_insn(insn) for insn in kernel.instructions]
         )
 
 
-
-
 # }}}
 
 # vim: foldmethod=marker
-- 
GitLab