freedreno: Add driconf to disable conservative LRZ

The problematic sequence of draws is pretty rare. But there are a small handful of games which do not exhibit the problematic sequence and for which invalidating LRZ on draws with blend plus depthwrite enabled hurts performance slightly. This driconf option enables opting in to the previous behavior. Signed-off-by: Rob Clark <robdclark@chromium.org> Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20708>
2023-01-16 08:35:06 -08:00
parent b392d57042
commit 1c82981bbe
11 changed files with 59 additions and 15 deletions
@@ -436,6 +436,10 @@ static const driOptionDescription tu_dri_options[] = {
      DRI_CONF_VK_WSI_FORCE_BGRA8_UNORM_FIRST(false)
      DRI_CONF_VK_DONT_CARE_AS_LOAD(false)
   DRI_CONF_SECTION_END
+
+   DRI_CONF_SECTION_MISCELLANEOUS
+      DRI_CONF_DISABLE_CONSERVATIVE_LRZ(false)
+   DRI_CONF_SECTION_END
 };

 static void
@@ -449,6 +453,9 @@ tu_init_dri_options(struct tu_instance *instance)

   if (driQueryOptionb(&instance->dri_options, "vk_dont_care_as_load"))
      instance->debug_flags |= TU_DEBUG_DONT_CARE_AS_LOAD;
+
+   instance->conservative_lrz =
+         !driQueryOptionb(&instance->dri_options, "disable_conservative_lrz");
 }

 VKAPI_ATTR VkResult VKAPI_CALL
@@ -149,6 +149,14 @@ struct tu_instance
   struct driOptionCache available_dri_options;

   enum tu_debug_flags debug_flags;
+
+   /* Conservative LRZ (default true) invalidates LRZ on draws with
+    * blend and depth-write enabled, because this can lead to incorrect
+    * rendering.  Driconf can be used to disable conservative LRZ for
+    * games which do not have the problematic sequence of draws *and*
+    * suffer a performance loss with conservative LRZ.
+    */
+   bool conservative_lrz;
 };
 VK_DEFINE_HANDLE_CASTS(tu_instance, vk.base, VkInstance,
                       VK_OBJECT_TYPE_INSTANCE)
@@ -812,7 +812,7 @@ tu6_calculate_lrz_state(struct tu_cmd_buffer *cmd,
    * enable LRZ write.  But this would cause early-z/lrz to discard
    * fragments from draw A which should be visible due to draw B.
    */
-   if (reads_dest && z_write_enable) {
+   if (reads_dest && z_write_enable && cmd->device->instance->conservative_lrz) {
      perf_debug(cmd->device, "Invalidating LRZ due to blend+depthwrite");
      disable_lrz = true;
   }
@@ -267,11 +267,18 @@ pipe_msm_create_screen(int fd, const struct pipe_screen_config *config)
   screen = fd_drm_screen_create(fd, NULL, config);
   return screen ? debug_screen_wrap(screen) : NULL;
 }
-DRM_DRIVER_DESCRIPTOR(msm, NULL, 0)
+
+const driOptionDescription msm_driconf[] = {
+#ifdef GALLIUM_FREEDRENO
+      #include "freedreno/driinfo_freedreno.h"
+#endif
+};
+DRM_DRIVER_DESCRIPTOR(msm, msm_driconf, ARRAY_SIZE(msm_driconf))
+DRM_DRIVER_DESCRIPTOR_ALIAS(msm, kgsl, msm_driconf, ARRAY_SIZE(msm_driconf))
 #else
 DRM_DRIVER_DESCRIPTOR_STUB(msm)
+DRM_DRIVER_DESCRIPTOR_STUB(kgsl)
 #endif
-DRM_DRIVER_DESCRIPTOR_ALIAS(msm, kgsl, NULL, 0)

 #if defined(GALLIUM_VIRGL) || (defined(GALLIUM_FREEDRENO) && !defined(PIPE_LOADER_DYNAMIC))
 #include "virgl/drm/virgl_drm_public.h"
@@ -295,9 +302,7 @@ pipe_virtio_gpu_create_screen(int fd, const struct pipe_screen_config *config)
 }

 const driOptionDescription virgl_driconf[] = {
-#ifdef GALLIUM_VIRGL
      #include "virgl/virgl_driinfo.h.in"
-#endif
 };
 DRM_DRIVER_DESCRIPTOR(virtio_gpu, virgl_driconf, ARRAY_SIZE(virgl_driconf))

@@ -202,17 +202,8 @@ compute_lrz_state(struct fd6_emit *emit) assert_dt
    * Normally looking at the state in draw C, we'd assume we could
    * enable LRZ write.  But this would cause early-z/lrz to discard
    * fragments from draw A which should be visible due to draw B.
-    *
-    * NOTE: So far invalid rendering due to this case has only been
-    * found in a single game (The Walking Dead: Season One -
-    * com.telltalegames.walkingdead100).  But other Telltale Games
-    * use the same Telltale Tool engine, so they might also have the
-    * same issue.  But blend+depthwrite is more common.  Hopefully
-    * it only shows up late in the renderpass where invalidating LRZ
-    * would not be too costly, but if this is found to cause perf
-    * regressions, a driconf allowlist/denylist can be added.
    */
-   if (reads_dest && zsa->writes_z) {
+   if (reads_dest && zsa->writes_z && ctx->screen->conservative_lrz) {
      if (!zsa->perf_warn_blend && rsc->lrz_valid) {
         perf_debug_ctx(ctx, "Invalidating LRZ due to blend+depthwrite");
         zsa->perf_warn_blend = true;
@@ -0,0 +1,5 @@
+// freedreno specific driconf options
+
+DRI_CONF_SECTION_MISCELLANEOUS
+   DRI_CONF_DISABLE_CONSERVATIVE_LRZ(false)
+DRI_CONF_SECTION_END
@@ -1096,6 +1096,9 @@ fd_screen_create(struct fd_device *dev, struct renderonly *ro,
   driParseConfigFiles(config->options, config->options_info, 0, "msm",
                       NULL, fd_dev_name(screen->dev_id), NULL, 0, NULL, 0);

+   screen->conservative_lrz =
+         !driQueryOptionb(config->options, "disable_conservative_lrz");
+
   struct sysinfo si;
   sysinfo(&si);
   screen->ram_size = si.totalram;
@@ -94,6 +94,14 @@ struct fd_screen {
   bool has_robustness;
   bool has_syncobj;

+   /* Conservative LRZ (default true) invalidates LRZ on draws with
+    * blend and depth-write enabled, because this can lead to incorrect
+    * rendering.  Driconf can be used to disable conservative LRZ for
+    * games which do not have the problematic sequence of draws *and*
+    * suffer a performance loss with conservative LRZ.
+    */
+   bool conservative_lrz;
+
   const struct fd_dev_info *info;
   uint32_t ccu_offset_gmem;
   uint32_t ccu_offset_bypass;
@@ -19,6 +19,7 @@
 # SOFTWARE.

 files_libfreedreno = files(
+  'driinfo_freedreno.h',
  'freedreno_autotune.c',
  'freedreno_autotune.h',
  'freedreno_batch.c',
@@ -11,4 +11,12 @@ DRI_CONF_SECTION_MISCELLANEOUS
    DRI_CONF_GLES_EMULATE_BGRA(true)
    DRI_CONF_GLES_APPLY_BGRA_DEST_SWIZZLE(true)
    DRI_CONF_GLES_SAMPLES_PASSED_VALUE(1024, 1, 400000000)
+
+    /*
+     * Native-context drivers also (can) bind to the some virtio_gpu guest
+     * kernel device.  Because drm_helper ties the driconf config to the
+     * drm device name, we also need to include config options for any
+     * possible drm-native-context guest driver:
+     */
+    DRI_CONF_DISABLE_CONSERVATIVE_LRZ(false)
 DRI_CONF_SECTION_END
@@ -515,6 +515,14 @@
   DRI_CONF_OPT_B(format_l8_srgb_enable_readback, def, \
                  "Force-enable reading back L8_SRGB textures")

+/**
+ * \brief freedreno specific configuration options
+ */
+
+#define DRI_CONF_DISABLE_CONSERVATIVE_LRZ(def) \
+   DRI_CONF_OPT_B(disable_conservative_lrz, def, \
+                  "Disable conservative LRZ")
+
 /**
 * \brief venus specific configuration options
 */