Mesa (main): v3d: Use the ra_alloc_contig_reg_class() function to speed up RA.

[GitLab Mirror]

Module: Mesa
Branch: main
Commit: ec3bc5da74ede1e93a16beb8ed876edb991a9bef
URL:    http://cgit.freedesktop.org/mesa/mesa/commit/?id=ec3bc5da74ede1e93a16beb8ed876edb991a9bef

Author: Eric Anholt <eric at anholt.net>
Date:   Thu Mar  4 16:52:20 2021 -0800

v3d: Use the ra_alloc_contig_reg_class() function to speed up RA.

It means we don't need to do the n^2 loop over the regs to set up the pq
values, nor do we need the register conflicts lists.

Acked-by: Erico Nunes <nunes.erico at gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9437>

---

 src/broadcom/compiler/vir_register_allocate.c | 10 +++++-----
 1 file changed, 5 insertions(+), 5 deletions(-)

diff --git a/src/broadcom/compiler/vir_register_allocate.c b/src/broadcom/compiler/vir_register_allocate.c
index 6a6bfd8a3d0..c55d8dc937d 100644
--- a/src/broadcom/compiler/vir_register_allocate.c
+++ b/src/broadcom/compiler/vir_register_allocate.c
@@ -501,19 +501,19 @@ vir_init_reg_sets(struct v3d_compiler *compiler)
         int max_thread_index = (compiler->devinfo->ver >= 40 ? 2 : 3);
 
         compiler->regs = ra_alloc_reg_set(compiler, PHYS_INDEX + PHYS_COUNT,
-                                          true);
+                                          false);
         if (!compiler->regs)
                 return false;
 
         for (int threads = 0; threads < max_thread_index; threads++) {
                 compiler->reg_class_any[threads] =
-                        ra_alloc_reg_class(compiler->regs);
+                        ra_alloc_contig_reg_class(compiler->regs, 1);
                 compiler->reg_class_r5[threads] =
-                        ra_alloc_reg_class(compiler->regs);
+                        ra_alloc_contig_reg_class(compiler->regs, 1);
                 compiler->reg_class_phys_or_acc[threads] =
-                        ra_alloc_reg_class(compiler->regs);
+                        ra_alloc_contig_reg_class(compiler->regs, 1);
                 compiler->reg_class_phys[threads] =
-                        ra_alloc_reg_class(compiler->regs);
+                        ra_alloc_contig_reg_class(compiler->regs, 1);
 
                 for (int i = PHYS_INDEX;
                      i < PHYS_INDEX + (PHYS_COUNT >> threads); i++) {