zink: always start/stop/resume queries inside renderpasses

this avoids potentially splitting renderpasses by ensuring that
all (non-cs) query operations always occur inside renderpasses

zink_query_update_gs_states() now has to be called inside renderpass
to catch the active queries

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21534>

src/gallium/drivers/zink/zink_batch.c

@@ -473,9 +473,6 @@ zink_start_batch(struct zink_context *ctx, struct zink_batch *batch)
    }
 #endif
 
-   if (!ctx->queries_disabled)
-      zink_resume_queries(ctx, batch);
-
    /* descriptor buffers must always be bound at the start of a batch */
    if (zink_descriptor_mode == ZINK_DESCRIPTOR_MODE_DB && !(ctx->flags & ZINK_CONTEXT_COPY_ONLY))
       zink_batch_bind_db(ctx);

src/gallium/drivers/zink/zink_context.c

@@ -2747,6 +2747,10 @@ zink_batch_rp(struct zink_context *ctx)
    else
       clear_buffers = begin_rendering(ctx);
    assert(!ctx->rp_changed);
+   if (!ctx->queries_disabled) {
+      zink_resume_queries(ctx, &ctx->batch);
+      zink_query_update_gs_states(ctx);
+   }
    if (in_rp || !ctx->batch.in_rp)
       return; //dead swapchain or continued renderpass
    if (ctx->render_condition.query)
@@ -2761,6 +2765,11 @@ zink_batch_no_rp(struct zink_context *ctx)
       return;
    if (ctx->render_condition.query)
       zink_stop_conditional_render(ctx);
+   /* suspend all queries that were started in a renderpass
+    * they can then be resumed upon beginning a new renderpass
+    */
+   if (!ctx->queries_disabled)
+      zink_query_renderpass_suspend(ctx);
    if (ctx->gfx_pipeline_state.render_pass)
       zink_end_render_pass(ctx);
    else {

src/gallium/drivers/zink/zink_query.c

@@ -953,16 +953,13 @@ zink_begin_query(struct pipe_context *pctx,
    util_dynarray_clear(&query->starts);
    query->start_offset = 0;
 
-   /* A query must either begin and end inside the same subpass of a render pass
+   if (batch->in_rp) {
-      instance, or must both begin and end outside of a render pass instance
+      begin_query(ctx, batch, query);
-      (i.e. contain entire render pass instances).
+   } else {
-      - 18.2. Query Operation
+      /* never directly start queries out of renderpass, always defer */
-
+      list_addtail(&query->active_list, &ctx->suspended_queries);
-    * tilers prefer out-of-renderpass queries for perf reasons, so force all queries
+      query->suspended = true;
-    * out of renderpasses
+   }
-    */
-   zink_batch_no_rp(ctx);
-   begin_query(ctx, batch, query);
 
    return true;
 }
@@ -1041,7 +1038,6 @@ zink_end_query(struct pipe_context *pctx,
 
    /* FIXME: this can be called from a thread, but it needs to write to the cmdbuf */
    threaded_context_unwrap_sync(pctx);
-   zink_batch_no_rp(ctx);
 
    if (list_is_linked(&query->stats_list))
       list_delinit(&query->stats_list);
@@ -1113,11 +1109,11 @@ suspend_query(struct zink_context *ctx, struct zink_query *query)
 }
 
 static void
-suspend_queries(struct zink_context *ctx)
+suspend_queries(struct zink_context *ctx, bool rp_only)
 {
    set_foreach(&ctx->batch.state->active_queries, entry) {
       struct zink_query *query = (void*)entry->key;
-      if (query->suspended)
+      if (query->suspended || (rp_only && !query->started_in_rp))
          continue;
       if (query->active && !is_time_query(query)) {
          /* the fence is going to steal the set off the batch, so we have to copy
@@ -1133,7 +1129,7 @@ suspend_queries(struct zink_context *ctx)
 void
 zink_suspend_queries(struct zink_context *ctx, struct zink_batch *batch)
 {
-   suspend_queries(ctx);
+   suspend_queries(ctx, false);
 }
 
 void
@@ -1162,6 +1158,12 @@ zink_resume_cs_query(struct zink_context *ctx)
    }
 }
 
+void
+zink_query_renderpass_suspend(struct zink_context *ctx)
+{
+   suspend_queries(ctx, true);
+}
+
 void
 zink_query_update_gs_states(struct zink_context *ctx)
 {

src/gallium/drivers/zink/zink_query.h

@@ -38,6 +38,9 @@ zink_suspend_queries(struct zink_context *ctx, struct zink_batch *batch);
 void
 zink_resume_queries(struct zink_context *ctx, struct zink_batch *batch);
 
+void
+zink_query_renderpass_suspend(struct zink_context *ctx);
+
 void
 zink_resume_cs_query(struct zink_context *ctx);