Mesa (main): radeonsi: allow and finish TC-compatible MSAA HTILE

[GitLab Mirror]

Module: Mesa
Branch: main
Commit: c0f723ce2b80a27e8d813396a5b7c9cbca7006a7
URL:    http://cgit.freedesktop.org/mesa/mesa/commit/?id=c0f723ce2b80a27e8d813396a5b7c9cbca7006a7

Author: Marek Olšák <marek.olsak at amd.com>
Date:   Fri Oct 29 15:36:05 2021 -0400

radeonsi: allow and finish TC-compatible MSAA HTILE

This improves perf for Catia by 4%.

Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer at amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13603>

---

 src/gallium/drivers/radeonsi/si_descriptors.c |  4 ++++
 src/gallium/drivers/radeonsi/si_state.c       | 17 +++++++++++++----
 src/gallium/drivers/radeonsi/si_texture.c     |  1 -
 3 files changed, 17 insertions(+), 5 deletions(-)

diff --git a/src/gallium/drivers/radeonsi/si_descriptors.c b/src/gallium/drivers/radeonsi/si_descriptors.c
index 7c3ae2eb8ce..3fad21ca9ca 100644
--- a/src/gallium/drivers/radeonsi/si_descriptors.c
+++ b/src/gallium/drivers/radeonsi/si_descriptors.c
@@ -371,6 +371,10 @@ void si_set_mutable_tex_desc_fields(struct si_screen *sscreen, struct si_texture
                       */
                      S_00A018_WRITE_COMPRESS_ENABLE(ac_surface_supports_dcc_image_stores(sscreen->info.chip_class, &tex->surface) &&
                                                     (access & SI_IMAGE_ACCESS_ALLOW_DCC_STORE));
+
+         /* TC-compatible MSAA HTILE requires ITERATE_256. */
+         if (tex->is_depth && tex->buffer.b.b.nr_samples >= 2)
+            state[6] |= S_00A018_ITERATE_256(1);
       }
 
       state[7] = meta_va >> 16;
diff --git a/src/gallium/drivers/radeonsi/si_state.c b/src/gallium/drivers/radeonsi/si_state.c
index 1ad6988a58e..53e1f7226c2 100644
--- a/src/gallium/drivers/radeonsi/si_state.c
+++ b/src/gallium/drivers/radeonsi/si_state.c
@@ -3293,10 +3293,10 @@ static void si_emit_framebuffer_state(struct si_context *sctx)
       radeon_add_to_buffer_list(sctx, &sctx->gfx_cs, &tex->buffer, RADEON_USAGE_READWRITE |
                                 (zb->base.texture->nr_samples > 1 ? RADEON_PRIO_DEPTH_BUFFER_MSAA
                                                                   : RADEON_PRIO_DEPTH_BUFFER));
+      bool tc_compat_htile = vi_tc_compat_htile_enabled(tex, zb->base.u.tex.level, PIPE_MASK_ZS);
 
       /* Set fields dependent on tc_compatile_htile. */
-      if (sctx->chip_class >= GFX9 &&
-          vi_tc_compat_htile_enabled(tex, zb->base.u.tex.level, PIPE_MASK_ZS)) {
+      if (sctx->chip_class >= GFX9 && tc_compat_htile) {
          unsigned max_zplanes = 4;
 
          if (tex->db_render_format == PIPE_FORMAT_Z16_UNORM && tex->buffer.b.b.nr_samples > 1)
@@ -3305,8 +3305,17 @@ static void si_emit_framebuffer_state(struct si_context *sctx)
          db_z_info |= S_028038_DECOMPRESS_ON_N_ZPLANES(max_zplanes + 1);
 
          if (sctx->chip_class >= GFX10) {
-            db_z_info |= S_028040_ITERATE_FLUSH(1);
-            db_stencil_info |= S_028044_ITERATE_FLUSH(!tex->htile_stencil_disabled);
+            bool iterate256 = tex->buffer.b.b.nr_samples >= 2;
+            db_z_info |= S_028040_ITERATE_FLUSH(1) |
+                         S_028040_ITERATE_256(iterate256);
+            db_stencil_info |= S_028044_ITERATE_FLUSH(!tex->htile_stencil_disabled) |
+                               S_028044_ITERATE_256(iterate256);
+
+            /* Workaround for a DB hang when ITERATE_256 is set to 1. Only affects 4X MSAA D/S images. */
+            if (sctx->screen->info.has_two_planes_iterate256_bug && iterate256 &&
+                !tex->htile_stencil_disabled && tex->buffer.b.b.nr_samples == 4) {
+               max_zplanes = 1;
+            }
          } else {
             db_z_info |= S_028038_ITERATE_FLUSH(1);
             db_stencil_info |= S_02803C_ITERATE_FLUSH(1);
diff --git a/src/gallium/drivers/radeonsi/si_texture.c b/src/gallium/drivers/radeonsi/si_texture.c
index 813cc4d438a..ce0f113b6cb 100644
--- a/src/gallium/drivers/radeonsi/si_texture.c
+++ b/src/gallium/drivers/radeonsi/si_texture.c
@@ -1232,7 +1232,6 @@ si_texture_create_with_modifier(struct pipe_screen *screen,
       sscreen->info.family != CHIP_TONGA && sscreen->info.family != CHIP_ICELAND &&
       (templ->flags & PIPE_RESOURCE_FLAG_TEXTURING_MORE_LIKELY) &&
       !(sscreen->debug_flags & DBG(NO_HYPERZ)) && !is_flushed_depth &&
-      templ->nr_samples <= 1 && /* TC-compat HTILE is less efficient with MSAA */
       is_zs;
    enum radeon_surf_mode tile_mode = si_choose_tiling(sscreen, templ, tc_compatible_htile);