asahi: Add non-occlusion query tracking

For other GPU queries, handled similarly.

Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24847>

src/gallium/drivers/asahi/agx_batch.c

@@ -108,6 +108,7 @@ agx_batch_init(struct agx_context *ctx,
    util_dynarray_init(&batch->scissor, ctx);
    util_dynarray_init(&batch->depth_bias, ctx);
    util_dynarray_init(&batch->occlusion_queries, ctx);
+   util_dynarray_init(&batch->nonocclusion_queries, ctx);
 
    batch->clear = 0;
    batch->draw = 0;
@@ -146,7 +147,7 @@ agx_batch_cleanup(struct agx_context *ctx, struct agx_batch *batch, bool reset)
 
    assert(ctx->batch != batch);
 
-   agx_finish_batch_occlusion_queries(batch);
+   agx_finish_batch_queries(batch);
    batch->occlusion_buffer.cpu = NULL;
    batch->occlusion_buffer.gpu = 0;
 
@@ -182,9 +183,7 @@ agx_batch_cleanup(struct agx_context *ctx, struct agx_batch *batch, bool reset)
    util_dynarray_fini(&batch->scissor);
    util_dynarray_fini(&batch->depth_bias);
    util_dynarray_fini(&batch->occlusion_queries);
-   util_unreference_framebuffer_state(&batch->key);
-
-   agx_batch_mark_complete(batch);
+   util_dynarray_fini(&batch->nonocclusion_queries);
 
    if (!(dev->debug & (AGX_DBG_TRACE | AGX_DBG_SYNC))) {
       agx_batch_print_stats(dev, batch);

src/gallium/drivers/asahi/agx_query.c

@@ -196,9 +196,9 @@ agx_get_oq_index(struct agx_batch *batch, struct agx_query *query)
 }
 
 void
-agx_finish_batch_occlusion_queries(struct agx_batch *batch)
+agx_finish_batch_queries(struct agx_batch *batch)
 {
-   uint64_t *results = (uint64_t *)batch->occlusion_buffer.cpu;
+   uint64_t *occlusion = (uint64_t *)batch->occlusion_buffer.cpu;
 
    util_dynarray_foreach(&batch->occlusion_queries, struct agx_query *, it) {
       struct agx_query *query = *it;
@@ -209,11 +209,11 @@ agx_finish_batch_occlusion_queries(struct agx_batch *batch)
 
       assert(query->writer == batch);
 
-      /* Get the result for this batch. If results is NULL, it means that no
+      /* Get the result for this batch. If occlusion is NULL, it means that no
        * draws actually enabled any occlusion queries, so there's no change.
        */
-      if (results != NULL) {
-         uint64_t result = *(results++);
+      if (occlusion != NULL) {
+         uint64_t result = *(occlusion++);
 
          /* Accumulate with the previous result (e.g. in case we split a frame
           * into multiple batches so an API-level query spans multiple batches).
@@ -227,6 +227,23 @@ agx_finish_batch_occlusion_queries(struct agx_batch *batch)
       query->writer = NULL;
       query->writer_index = 0;
    }
+
+   /* Now handle non-occlusion queries in a similar way */
+   util_dynarray_foreach(&batch->nonocclusion_queries, struct agx_query *, it) {
+      struct agx_query *query = *it;
+      if (query == NULL)
+         continue;
+
+      assert(query->writer == batch);
+
+      /* Accumulate */
+      uint64_t *value = query->ptr.cpu;
+      query->value += (*value);
+      query->writer = NULL;
+      query->writer_index = 0;
+      query->ptr.cpu = NULL;
+      query->ptr.gpu = 0;
+   }
 }
 
 static void

src/gallium/drivers/asahi/agx_state.h

@@ -268,6 +268,9 @@ struct agx_batch {
    struct util_dynarray occlusion_queries;
    struct agx_ptr occlusion_buffer;
 
+   /* Non-occlusion queries */
+   struct util_dynarray nonocclusion_queries;
+
    /* Result buffer where the kernel places command execution information */
    union agx_batch_result *result;
    size_t result_off;
@@ -529,6 +532,12 @@ struct agx_query {
    struct agx_batch *writer;
    unsigned writer_index;
 
+   /* For GPU queries other than occlusion queries, the value of the query as
+    * written by the `writer` if a writer is non-NULL, and irrelevant otherwise.
+    * When flushing the query, this value is read and added to agx_query::value.
+    */
+   struct agx_ptr ptr;
+
    /* Accumulator flushed to the CPU */
    uint64_t value;
 };
@@ -797,7 +806,7 @@ uint64_t agx_build_meta(struct agx_batch *batch, bool store,
 /* Query management */
 uint16_t agx_get_oq_index(struct agx_batch *batch, struct agx_query *query);
 
-void agx_finish_batch_occlusion_queries(struct agx_batch *batch);
+void agx_finish_batch_queries(struct agx_batch *batch);
 
 bool agx_render_condition_check_inner(struct agx_context *ctx);