# Mesa (master): ra: Take advantage of the adjacency list in finding a node to spill.

Eric Anholt anholt at kemper.freedesktop.org
Tue Jan 18 10:34:22 PST 2011

```Module: Mesa
Branch: master

Author: Eric Anholt <eric at anholt.net>
Date:   Tue Jan 18 01:08:51 2011 -0800

ra: Take advantage of the adjacency list in finding a node to spill.

This revealed a bug in ra_get_spill_benefit where we only considered
the benefit of the first adjacency we were to remove, explaining some
of the ugly spilling I've seen in shaders.  Because of the reduced
spilling, it reduces the runtime of glsl-fs-convolution-1 36.9% +/-
0.9% (n=5).

---

src/mesa/program/register_allocate.c |   12 ++++++------
1 files changed, 6 insertions(+), 6 deletions(-)

diff --git a/src/mesa/program/register_allocate.c b/src/mesa/program/register_allocate.c
index 634a7da..f984e2f 100644
--- a/src/mesa/program/register_allocate.c
+++ b/src/mesa/program/register_allocate.c
@@ -401,17 +401,17 @@ ra_get_spill_benefit(struct ra_graph *g, unsigned int n)
float benefit = 0;
int n_class = g->nodes[n].class;

-   /* Define the benefit of eliminating an interference between n, j
+   /* Define the benefit of eliminating an interference between n, n2
* through spilling as q(C, B) / p(C).  This is similar to the
* "count number of edges" approach of traditional graph coloring,
* but takes classes into account.
*/
-   for (j = 0; j < g->count; j++) {
-      if (j != n && g->nodes[n].adjacency[j]) {
-	 unsigned int j_class = g->nodes[j].class;
-	 benefit += ((float)g->regs->classes[n_class]->q[j_class] /
+   for (j = 0; j < g->nodes[n].adjacency_count; j++) {
+      unsigned int n2 = g->nodes[n].adjacency_list[j];
+      if (n != n2) {
+	 unsigned int n2_class = g->nodes[n2].class;
+	 benefit += ((float)g->regs->classes[n_class]->q[n2_class] /
g->regs->classes[n_class]->p);
-	 break;
}
}

```