Patrick observed recently that an element of the vector cache could be
arbitrarily large. Let's only cache relatively small vecs.
This has no effect on compiling the libstdc++ stdc++.h, presumably because
nothing in the library requires a vec that large. I figure that this makes it
more likely that a subsequent long list will reuse the same memory when the
later vec gets expanded.
Does this make sense to others?
gcc/c-family/ChangeLog:
* c-common.c (release_tree_vector): Only cache vecs smaller than
16 elements.
---
gcc/c-family/c-common.c | 12 ++++++++++--
1 file changed, 10 insertions(+), 2 deletions(-)
diff --git a/gcc/c-family/c-common.c b/gcc/c-family/c-common.c
index 436df45df68..90e8ec87b6b 100644
--- a/gcc/c-family/c-common.c
+++ b/gcc/c-family/c-common.c
@@ -8213,8 +8213,16 @@ release_tree_vector (vec<tree, va_gc> *vec)
{
if (vec != NULL)
{
- vec->truncate (0);
- vec_safe_push (tree_vector_cache, vec);
+ if (vec->allocated () >= 16)
+ /* Don't cache vecs that have expanded more than once. On a p64
+ target, vecs double in alloc size with each power of 2 elements, e.g
+ at 16 elements the alloc increases from 128 to 256 bytes. */
+ vec_free (vec);
+ else
+ {
+ vec->truncate (0);
+ vec_safe_push (tree_vector_cache, vec);
+ }
}
}
base-commit: 132f1c27770fa6dafdf14591878d301aedd5ae16
--
2.27.0
