From: Iago Toral Quiroga <itoral@igalia.com>
Date: Mon, 7 Mar 2022 13:03:03 +0000 (+0100)
Subject: broadcom/compiler: add comment on why we don't use r5 with ldunifa
X-Git-Tag: upstream/22.3.5~11894
X-Git-Url: http://review.tizen.org/git/?a=commitdiff_plain;h=77f58b46d9f20af89e74d1301e4220b5e48fa8be;p=platform%2Fupstream%2Fmesa.git

broadcom/compiler: add comment on why we don't use r5 with ldunifa

Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15276>
---

diff --git a/src/broadcom/compiler/vir_register_allocate.c b/src/broadcom/compiler/vir_register_allocate.c
index 4862f42..46c7e52 100644
--- a/src/broadcom/compiler/vir_register_allocate.c
+++ b/src/broadcom/compiler/vir_register_allocate.c
@@ -916,6 +916,11 @@ update_graph_and_reg_classes_for_inst(struct v3d_compile *c, int *acc_nodes,
         if (inst->dst.file == QFILE_TEMP) {
                 /* Only a ldunif gets to write to R5, which only has a
                  * single 32-bit channel of storage.
+                 *
+                 * NOTE: ldunifa is subject to the same, however, going by
+                 * shader-db it is best to keep r5 exclusive to ldunif, probably
+                 * because ldunif has usually a shorter lifespan, allowing for
+                 * more accumulator reuse and QPU merges.
                  */
                 if (!inst->qpu.sig.ldunif) {
                         uint8_t class_bits =