Mesa (main): nine: track bound sampler count to optimize unbinds

[GitLab Mirror]

Module: Mesa
Branch: main
Commit: 740752d2327be00a8a073e32366be4950ef32e0b
URL:    http://cgit.freedesktop.org/mesa/mesa/commit/?id=740752d2327be00a8a073e32366be4950ef32e0b

Author: Mike Blumenkrantz <michael.blumenkrantz at gmail.com>
Date:   Tue Jul 13 19:56:42 2021 -0400

nine: track bound sampler count to optimize unbinds

Acked-by: Marek Olšák <marek.olsak at amd.com>

Reviewed-by: <Axel Davy davyaxel0 at gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11865>

---

 src/gallium/frontends/nine/nine_state.c | 14 ++++++++------
 src/gallium/frontends/nine/nine_state.h |  2 ++
 2 files changed, 10 insertions(+), 6 deletions(-)

diff --git a/src/gallium/frontends/nine/nine_state.c b/src/gallium/frontends/nine/nine_state.c
index 94604bd9815..d7f12954e52 100644
--- a/src/gallium/frontends/nine/nine_state.c
+++ b/src/gallium/frontends/nine/nine_state.c
@@ -1001,7 +1001,6 @@ update_textures_and_samplers(struct NineDevice9 *device)
                             device->ff.ps->sampler_mask;
 
     commit_samplers = FALSE;
-    uint16_t prev_mask = context->bound_samplers_mask_ps;
     const uint16_t ps_mask = sampler_mask | context->enabled_samplers_mask_ps;
     context->bound_samplers_mask_ps = ps_mask;
     num_textures = util_last_bit(ps_mask) + 1;
@@ -1040,15 +1039,14 @@ update_textures_and_samplers(struct NineDevice9 *device)
        view[i] = NULL;
 
     pipe->set_sampler_views(pipe, PIPE_SHADER_FRAGMENT, 0, num_textures,
-                            /* unbind trailing based on bitcount of shifted mask */
-                            util_bitcount(prev_mask >> num_textures), view);
+                            num_textures < context->enabled_sampler_count_ps ? context->enabled_sampler_count_ps - num_textures : 0, view);
+    context->enabled_sampler_count_ps = num_textures;
 
     if (commit_samplers)
         cso_single_sampler_done(context->cso, PIPE_SHADER_FRAGMENT);
 
     commit_samplers = FALSE;
     sampler_mask = context->programmable_vs ? context->vs->sampler_mask : 0;
-    prev_mask = context->bound_samplers_mask_vs;
     const uint16_t vs_mask = sampler_mask | context->enabled_samplers_mask_vs;
     context->bound_samplers_mask_vs = vs_mask;
     num_textures = util_last_bit(vs_mask) + 1;
@@ -1085,8 +1083,8 @@ update_textures_and_samplers(struct NineDevice9 *device)
        view[i] = NULL;
 
     pipe->set_sampler_views(pipe, PIPE_SHADER_VERTEX, 0, num_textures,
-                            /* unbind trailing based on bitcount of shifted mask */
-                            util_bitcount(prev_mask >> num_textures), view);
+                            num_textures < context->enabled_sampler_count_vs ? context->enabled_sampler_count_vs - num_textures : 0, view);
+    context->enabled_sampler_count_vs = num_textures;
 
     if (commit_samplers)
         cso_single_sampler_done(context->cso, PIPE_SHADER_VERTEX);
@@ -2838,6 +2836,8 @@ void nine_state_restore_non_cso(struct NineDevice9 *device)
     context->changed.vtxbuf = (1ULL << device->caps.MaxStreams) - 1;
     context->changed.ucp = TRUE;
     context->commit |= NINE_STATE_COMMIT_CONST_VS | NINE_STATE_COMMIT_CONST_PS;
+    context->enabled_sampler_count_vs = 0;
+    context->enabled_sampler_count_ps = 0;
 }
 
 void
@@ -2962,6 +2962,8 @@ nine_context_clear(struct NineDevice9 *device)
 
     cso_set_samplers(cso, PIPE_SHADER_VERTEX, 0, NULL);
     cso_set_samplers(cso, PIPE_SHADER_FRAGMENT, 0, NULL);
+    context->enabled_sampler_count_vs = 0;
+    context->enabled_sampler_count_ps = 0;
 
     pipe->set_sampler_views(pipe, PIPE_SHADER_VERTEX, 0, 0, NINE_MAX_SAMPLERS_VS, NULL);
     pipe->set_sampler_views(pipe, PIPE_SHADER_FRAGMENT, 0, 0, NINE_MAX_SAMPLERS_PS, NULL);
diff --git a/src/gallium/frontends/nine/nine_state.h b/src/gallium/frontends/nine/nine_state.h
index 0da60dbaf8a..dad9e60de94 100644
--- a/src/gallium/frontends/nine/nine_state.h
+++ b/src/gallium/frontends/nine/nine_state.h
@@ -303,6 +303,8 @@ struct nine_context {
 
     uint8_t bound_samplers_mask_vs;
     uint8_t enabled_samplers_mask_vs;
+    uint8_t enabled_sampler_count_vs;
+    uint8_t enabled_sampler_count_ps;
     uint16_t bound_samplers_mask_ps;
     uint16_t enabled_samplers_mask_ps;