asahi: Fix depth/stencil buffers

There are a bunch of bits we need to set right to get depth/stencil
loads/stores working, including with independent settings for each. The
kernel "helps" us here.

Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18380>

src/asahi/lib/cmdbuf.xml

@@ -676,8 +676,8 @@
     <field name="Depth clear value" start="0:0" size="32" type="hex"/>
     <field name="Stencil clear value" start="1:0" size="8" type="uint"/>
     <field name="Unk 1" start="1:8" size="8" type="hex" default="3"/>
-    <field name="Set when reloading Z 1" start="3:8" size="1" type="bool"/>
-    <field name="Set when reloading Z 2" start="4:24" size="1" type="bool"/>
+    <field name="Set when reloading Z or S 1" start="3:8" size="1" type="bool"/>
+    <field name="Set when reloading Z or S 2" start="4:24" size="1" type="bool"/>
     <field name="Z16 Unorm attachment" start="5:8" size="1" type="bool"/>
     <field name="Unk 3" start="6:0" size="32" type="hex" default="0xffffffff"/>
     <field name="Unk 4" start="7:0" size="32" type="hex" default="0xffffffff"/>

src/gallium/drivers/asahi/agx_pipe.c

@@ -542,6 +542,7 @@ agx_flush(struct pipe_context *pctx,
                pipeline_reload,
                pipeline_store,
                clear_pipeline_textures,
+               ctx->batch->clear,
                ctx->batch->clear_depth,
                ctx->batch->clear_stencil);
 

src/gallium/drivers/asahi/magic.c

@@ -167,9 +167,13 @@ demo_cmdbuf(uint64_t *buf, size_t size,
             uint32_t pipeline_load,
             uint32_t pipeline_store,
             bool clear_pipeline_textures,
+            unsigned clear_buffers,
             double clear_depth,
             unsigned clear_stencil)
 {
+   bool should_clear_depth = clear_buffers & PIPE_CLEAR_DEPTH;
+   bool should_clear_stencil = clear_buffers & PIPE_CLEAR_STENCIL;
+
    uint32_t *map = (uint32_t *) buf;
    memset(map, 0, 518 * 4);
 
@@ -197,25 +201,31 @@ demo_cmdbuf(uint64_t *buf, size_t size,
          const struct util_format_description *desc =
             util_format_description(zsbuf->texture->format);
 
-         // note: setting 0x4 bit here breaks partial render with depth 
-         cfg.depth_flags = 0x80000; // no compression, clear
-
          cfg.depth_width = framebuffer->width;
          cfg.depth_height = framebuffer->height;
 
          if (util_format_has_depth(desc)) {
             depth_buffer = agx_map_surface(zsbuf);
+
+            cfg.depth_reload = !should_clear_depth;
+            cfg.depth_flags |= 0x80000;
+            if (!should_clear_depth) cfg.depth_flags |= 0x8000;
          } else {
             stencil_buffer = agx_map_surface(zsbuf);
+            cfg.depth_flags |= 0x40000;
+            if (!should_clear_stencil) cfg.depth_flags |= 0x4000;
          }
 
          if (agx_resource(zsbuf->texture)->separate_stencil) {
             stencil_buffer = agx_map_surface_resource(zsbuf,
                   agx_resource(zsbuf->texture)->separate_stencil);
+
+            cfg.depth_flags |= 0x40000;
+            if (!should_clear_stencil) cfg.depth_flags |= 0x4000;
          }
 
+         cfg.depth_buffer_if_clearing = depth_buffer;
          cfg.stencil_buffer = stencil_buffer;
-         cfg.stencil_buffer_2 = stencil_buffer;
 
          /* It's unclear how tile size is conveyed for depth/stencil targets,
           * which interactions with mipmapping (for example of a 33x33
@@ -225,7 +235,7 @@ demo_cmdbuf(uint64_t *buf, size_t size,
             unreachable("todo: mapping other levels");
 
          cfg.depth_buffer = depth_buffer;
-         cfg.depth_buffer_if_clearing = depth_buffer;
+         cfg.stencil_buffer_2 = stencil_buffer;
       }
    }
 
@@ -236,7 +246,17 @@ demo_cmdbuf(uint64_t *buf, size_t size,
    }
 
    agx_pack(map + 292, IOGPU_CLEAR_Z_S, cfg) {
-      cfg.set_when_reloading_z_1 = clear_pipeline_textures;
+      cfg.set_when_reloading_z_or_s_1 = clear_pipeline_textures;
+
+      if (depth_buffer && !should_clear_depth) {
+         cfg.set_when_reloading_z_or_s_1 = true;
+         cfg.set_when_reloading_z_or_s_2 = true;
+      }
+
+      if (stencil_buffer && !should_clear_stencil) {
+         cfg.set_when_reloading_z_or_s_1 = true;
+         cfg.set_when_reloading_z_or_s_2 = true;
+      }
 
       cfg.depth_clear_value = fui(clear_depth);
       cfg.stencil_clear_value = clear_stencil & 0xff;

src/gallium/drivers/asahi/magic.h

@@ -36,6 +36,7 @@ demo_cmdbuf(uint64_t *buf, size_t size,
             uint32_t pipeline_load,
             uint32_t pipeline_store,
             bool clear_pipeline_textures,
+            unsigned clear_buffers,
             double clear_depth,
             unsigned clear_stencil);