i965/vs: Do round-robin register allocation on gen6+ like we do in the FS.

This will free instruction scheduling to make better choices.  No
statistically significant performance difference on GLB2.7 (n=93).

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>

diff --git a/src/mesa/drivers/dri/i965/brw_vec4_reg_allocate.cpp b/src/mesa/drivers/dri/i965/brw_vec4_reg_allocate.cpp
index ac3d401ac3f98a3fd2251e1e2724147d0f8a7b4c..7149d46b47f3546d025527f142ff7b043adba72a 100644 (file)
--- a/src/mesa/drivers/dri/i965/brw_vec4_reg_allocate.cpp
+++ b/src/mesa/drivers/dri/i965/brw_vec4_reg_allocate.cpp
@@ -102,6 +102,8 @@ brw_alloc_reg_set_for_classes(struct brw_context *brw,
                              int class_count,
                              int base_reg_count)
 {
+   struct intel_context *intel = &brw->intel;
+
    /* Compute the total number of registers across all classes. */
    int ra_reg_count = 0;
    for (int i = 0; i < class_count; i++) {
@@ -112,6 +114,8 @@ brw_alloc_reg_set_for_classes(struct brw_context *brw,
    brw->vs.ra_reg_to_grf = ralloc_array(brw, uint8_t, ra_reg_count);
    ralloc_free(brw->vs.regs);
    brw->vs.regs = ra_alloc_reg_set(brw, ra_reg_count);
+   if (intel->gen >= 6)
+      ra_set_allocate_round_robin(brw->vs.regs);
    ralloc_free(brw->vs.classes);
    brw->vs.classes = ralloc_array(brw, int, class_count + 1);