From 624cef0619736598116c8ab84be8a91683750beb Mon Sep 17 00:00:00 2001
From: Andreas Kloeckner <inform@tiker.net>
Date: Wed, 12 Oct 2011 21:36:09 -0400
Subject: [PATCH] Restrict condition hoisting to actually used inames.

---
 loopy/codegen/dispatch.py |  8 +++++++-
 loopy/schedule.py         | 23 +++++++++++++++++++++++
 test/test_matmul.py       |  2 +-
 3 files changed, 31 insertions(+), 2 deletions(-)

diff --git a/loopy/codegen/dispatch.py b/loopy/codegen/dispatch.py
index 9d1f0261a..f9c29a08a 100644
--- a/loopy/codegen/dispatch.py
+++ b/loopy/codegen/dispatch.py
@@ -146,8 +146,14 @@ def build_loop_nest(kernel, sched_index, codegen_state):
             # Success: found a big enough group of inames for a conditional.
             # See if there are bounds checks available for that set.
 
+            from loopy.schedule import find_used_inames_within
+            used_inames = set()
+            for subsched_index, _ in sched_indices_and_cond_inames[0:idx]:
+                used_inames |= find_used_inames_within(kernel, subsched_index)
+
             from loopy.codegen.bounds import generate_bounds_checks
-            bounds_checks = generate_bounds_checks(kernel.domain, current_iname_set,
+            bounds_checks = generate_bounds_checks(kernel.domain,
+                    current_iname_set & used_inames,
                     codegen_state.implemented_domain)
         else:
             bounds_checks = []
diff --git a/loopy/schedule.py b/loopy/schedule.py
index 55137d1d8..5cbf88471 100644
--- a/loopy/schedule.py
+++ b/loopy/schedule.py
@@ -759,6 +759,29 @@ def has_barrier_within(kernel, sched_index):
     else:
         return False
 
+
+
+
+def find_used_inames_within(kernel, sched_index):
+    sched_item = kernel.schedule[sched_index]
+
+    if isinstance(sched_item, EnterLoop):
+        loop_contents, _ = gather_schedule_subloop(
+                kernel.schedule, sched_index)
+        run_insns = [subsched_item
+                for subsched_item in loop_contents
+                if isinstance(subsched_item, RunInstruction)]
+    elif isinstance(sched_item, RunInstruction):
+        run_insns = [sched_item]
+    else:
+        return set()
+
+    result = set()
+    for sched_item in run_insns:
+        result.update(kernel.id_to_insn[sched_item.insn_id].all_inames())
+
+    return result
+
 # }}}
 
 
diff --git a/test/test_matmul.py b/test/test_matmul.py
index 42ebf7e9e..71b1d409e 100644
--- a/test/test_matmul.py
+++ b/test/test_matmul.py
@@ -211,7 +211,7 @@ def test_plain_matrix_mul_new_ui(ctx_factory):
                 lp.ArrayArg("c", dtype, shape=(n, n), order=order),
                 lp.ScalarArg("n", np.int32, approximately=n),
                 ],
-            name="matmul", assumptions="n >= 1")
+            name="matmul", assumptions="n >= 16")
 
     knl = lp.split_dimension(knl, "i", 16,
             outer_tag="g.0", inner_tag="l.1", no_slabs=True)
-- 
GitLab