Mesa (staging/20.2): radeonsi: don't count unusable vertices to the NGG LDS size

[GitLab Mirror]

Module: Mesa
Branch: staging/20.2
Commit: 3bf0368f9ea2e7764a22c75f1944cbbd07076681
URL:    http://cgit.freedesktop.org/mesa/mesa/commit/?id=3bf0368f9ea2e7764a22c75f1944cbbd07076681

Author: Marek Olšák <marek.olsak at amd.com>
Date:   Thu Jul 30 08:19:48 2020 -0400

radeonsi: don't count unusable vertices to the NGG LDS size

Now we get optimal LDS usage.

Fixes: a23802bcb9a - ac,radeonsi: start adding support for gfx10.3

Acked-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer at amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/6137>
(cherry picked from commit 68b3e92fef09330ac880e713a744dc7a57e78f05)

---

 .pick_status.json                               |  2 +-
 src/gallium/drivers/radeonsi/gfx10_shader_ngg.c | 14 +++++++++++---
 2 files changed, 12 insertions(+), 4 deletions(-)

diff --git a/.pick_status.json b/.pick_status.json
index e8f58d05a6d..169e3aef2f5 100644
--- a/.pick_status.json
+++ b/.pick_status.json
@@ -40,7 +40,7 @@
         "description": "radeonsi: don't count unusable vertices to the NGG LDS size",
         "nominated": true,
         "nomination_type": 1,
-        "resolution": 0,
+        "resolution": 1,
         "master_sha": null,
         "because_sha": "a23802bcb9a42a02d34a5a36d6e66d6532813a0d"
     },
diff --git a/src/gallium/drivers/radeonsi/gfx10_shader_ngg.c b/src/gallium/drivers/radeonsi/gfx10_shader_ngg.c
index 0797f9cdb3a..efeb9e8838c 100644
--- a/src/gallium/drivers/radeonsi/gfx10_shader_ngg.c
+++ b/src/gallium/drivers/radeonsi/gfx10_shader_ngg.c
@@ -2027,9 +2027,15 @@ retry_select_mode:
 
          max_gsprims = align(max_gsprims, wavesize);
          max_gsprims = MIN2(max_gsprims, max_gsprims_base);
-         if (gsprim_lds_size)
+         if (gsprim_lds_size) {
+            /* Don't count unusable vertices to the LDS size. Those are vertices above
+             * the maximum number of vertices that can occur in the workgroup,
+             * which is e.g. max_gsprims * 3 for triangles.
+             */
+            unsigned usable_esverts = MIN2(max_esverts, max_gsprims * max_verts_per_prim);
             max_gsprims =
-               MIN2(max_gsprims, (max_lds_size - max_esverts * esvert_lds_size) / gsprim_lds_size);
+               MIN2(max_gsprims, (max_lds_size - usable_esverts * esvert_lds_size) / gsprim_lds_size);
+         }
          clamp_gsprims_to_esverts(&max_gsprims, max_esverts, min_verts_per_prim, use_adjacency);
          assert(max_esverts >= max_verts_per_prim && max_gsprims >= 1);
       } while (orig_max_esverts != max_esverts || orig_max_gsprims != max_gsprims);
@@ -2067,7 +2073,9 @@ retry_select_mode:
    shader->ngg.prim_amp_factor = prim_amp_factor;
    shader->ngg.max_vert_out_per_gs_instance = max_vert_out_per_gs_instance;
 
-   shader->gs_info.esgs_ring_size = max_esverts * esvert_lds_size;
+   /* Don't count unusable vertices. */
+   shader->gs_info.esgs_ring_size = MIN2(max_esverts, max_gsprims * max_verts_per_prim) *
+                                    esvert_lds_size;
    shader->ngg.ngg_emit_size = max_gsprims * gsprim_lds_size;
 
    assert(shader->ngg.hw_max_esverts >= min_esverts); /* HW limitation */