gallium/u_threaded: offload begin/end_intel_perf_query

Fixes: 206495cac4 ("iris: Enable u_threaded_context")
Signed-off-by: Marcin Ślusarz <marcin.slusarz@intel.com>
Acked-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9922>

src/gallium/auxiliary/util/u_threaded_context.c

@@ -2997,6 +2997,7 @@ tc_get_intel_perf_query_info(struct pipe_context *_pipe,
    struct threaded_context *tc = threaded_context(_pipe);
    struct pipe_context *pipe = tc->pipe;
 
+   tc_sync(tc); /* n_active vs begin/end_intel_perf_query */
    pipe->get_intel_perf_query_info(pipe, query_index, name, data_size,
          n_counters, n_active);
 }
@@ -3029,24 +3030,35 @@ tc_new_intel_perf_query_obj(struct pipe_context *_pipe, unsigned query_index)
    return pipe->new_intel_perf_query_obj(pipe, query_index);
 }
 
+static void
+tc_call_begin_intel_perf_query(struct pipe_context *pipe, union tc_payload *payload)
+{
+   (void)pipe->begin_intel_perf_query(pipe, payload->query);
+}
+
 static bool
 tc_begin_intel_perf_query(struct pipe_context *_pipe, struct pipe_query *q)
 {
    struct threaded_context *tc = threaded_context(_pipe);
-   struct pipe_context *pipe = tc->pipe;
 
-   tc_sync(tc);
-   return pipe->begin_intel_perf_query(pipe, q);
+   tc_add_small_call(tc, TC_CALL_begin_intel_perf_query)->query = q;
+
+   /* assume success, begin failure can be signaled from get_intel_perf_query_data */
+   return true;
+}
+
+static void
+tc_call_end_intel_perf_query(struct pipe_context *pipe, union tc_payload *payload)
+{
+   pipe->end_intel_perf_query(pipe, payload->query);
 }
 
 static void
 tc_end_intel_perf_query(struct pipe_context *_pipe, struct pipe_query *q)
 {
    struct threaded_context *tc = threaded_context(_pipe);
-   struct pipe_context *pipe = tc->pipe;
 
-   tc_sync(tc);
-   pipe->end_intel_perf_query(pipe, q);
+   tc_add_small_call(tc, TC_CALL_end_intel_perf_query)->query = q;
 }
 
 static void
@@ -3055,7 +3067,7 @@ tc_delete_intel_perf_query(struct pipe_context *_pipe, struct pipe_query *q)
    struct threaded_context *tc = threaded_context(_pipe);
    struct pipe_context *pipe = tc->pipe;
 
-   tc_sync(tc);
+   tc_sync(tc); /* flush potentially pending begin/end_intel_perf_queries */
    pipe->delete_intel_perf_query(pipe, q);
 }
 
@@ -3065,7 +3077,7 @@ tc_wait_intel_perf_query(struct pipe_context *_pipe, struct pipe_query *q)
    struct threaded_context *tc = threaded_context(_pipe);
    struct pipe_context *pipe = tc->pipe;
 
-   tc_sync(tc);
+   tc_sync(tc); /* flush potentially pending begin/end_intel_perf_queries */
    pipe->wait_intel_perf_query(pipe, q);
 }
 
@@ -3075,11 +3087,11 @@ tc_is_intel_perf_query_ready(struct pipe_context *_pipe, struct pipe_query *q)
    struct threaded_context *tc = threaded_context(_pipe);
    struct pipe_context *pipe = tc->pipe;
 
-   tc_sync(tc);
+   tc_sync(tc); /* flush potentially pending begin/end_intel_perf_queries */
    return pipe->is_intel_perf_query_ready(pipe, q);
 }
 
-static void
+static bool
 tc_get_intel_perf_query_data(struct pipe_context *_pipe,
                              struct pipe_query *q,
                              size_t data_size,
@@ -3089,8 +3101,8 @@ tc_get_intel_perf_query_data(struct pipe_context *_pipe,
    struct threaded_context *tc = threaded_context(_pipe);
    struct pipe_context *pipe = tc->pipe;
 
-   tc_sync(tc);
-   pipe->get_intel_perf_query_data(pipe, q, data_size, data, bytes_written);
+   tc_sync(tc); /* flush potentially pending begin/end_intel_perf_queries */
+   return pipe->get_intel_perf_query_data(pipe, q, data_size, data, bytes_written);
 }
 
 /********************************************************************

src/gallium/auxiliary/util/u_threaded_context_calls.h

@@ -78,3 +78,6 @@ CALL(delete_tcs_state)
 CALL(delete_tes_state)
 CALL(delete_vertex_elements_state)
 CALL(delete_sampler_state)
+
+CALL(begin_intel_perf_query)
+CALL(end_intel_perf_query)

src/gallium/drivers/iris/iris_performance_query.c

@@ -28,6 +28,7 @@
 struct iris_perf_query {
    struct gl_perf_query_object base;
    struct intel_perf_query_object *query;
+   bool begin_succeeded;
 };
 
 static unsigned
@@ -107,7 +108,7 @@ iris_begin_perf_query(struct pipe_context *pipe, struct pipe_query *q)
    struct intel_perf_query_object *obj = perf_query->query;
    struct intel_perf_context *perf_ctx = ice->perf_ctx;
 
-   return intel_perf_begin_query(perf_ctx, obj);
+   return (perf_query->begin_succeeded = intel_perf_begin_query(perf_ctx, obj));
 }
 
 static void
@@ -118,7 +119,8 @@ iris_end_perf_query(struct pipe_context *pipe, struct pipe_query *q)
    struct intel_perf_query_object *obj = perf_query->query;
    struct intel_perf_context *perf_ctx = ice->perf_ctx;
 
-   intel_perf_end_query(perf_ctx, obj);
+   if (perf_query->begin_succeeded)
+      intel_perf_end_query(perf_ctx, obj);
 }
 
 static void
@@ -188,7 +190,8 @@ iris_wait_perf_query(struct pipe_context *pipe, struct pipe_query *q)
    struct intel_perf_query_object *obj = perf_query->query;
    struct intel_perf_context *perf_ctx = ice->perf_ctx;
 
-   intel_perf_wait_query(perf_ctx, obj, &ice->batches[IRIS_BATCH_RENDER]);
+   if (perf_query->begin_succeeded)
+      intel_perf_wait_query(perf_ctx, obj, &ice->batches[IRIS_BATCH_RENDER]);
 }
 
 static bool
@@ -201,12 +204,14 @@ iris_is_perf_query_ready(struct pipe_context *pipe, struct pipe_query *q)
 
    if (perf_query->base.Ready)
       return true;
+   if (!perf_query->begin_succeeded)
+      return true;
 
    return intel_perf_is_query_ready(perf_ctx, obj,
                                     &ice->batches[IRIS_BATCH_RENDER]);
 }
 
-static void
+static bool
 iris_get_perf_query_data(struct pipe_context *pipe,
                          struct pipe_query *q,
                          size_t data_size,
@@ -218,8 +223,12 @@ iris_get_perf_query_data(struct pipe_context *pipe,
    struct intel_perf_query_object *obj = perf_query->query;
    struct intel_perf_context *perf_ctx = ice->perf_ctx;
 
-   intel_perf_get_query_data(perf_ctx, obj, &ice->batches[IRIS_BATCH_RENDER],
-         data_size, data, bytes_written);
+   if (perf_query->begin_succeeded) {
+      intel_perf_get_query_data(perf_ctx, obj, &ice->batches[IRIS_BATCH_RENDER],
+            data_size, data, bytes_written);
+   }
+
+   return perf_query->begin_succeeded;
 }
 
 void

src/gallium/include/pipe/p_context.h

@@ -271,7 +271,7 @@ struct pipe_context {
 
    bool (*is_intel_perf_query_ready)(struct pipe_context *pipe, struct pipe_query *q);
 
-   void (*get_intel_perf_query_data)(struct pipe_context *pipe,
+   bool (*get_intel_perf_query_data)(struct pipe_context *pipe,
                                      struct pipe_query *q,
                                      size_t data_size,
                                      uint32_t *data,

src/mesa/drivers/dri/i965/brw_performance_query.c

@@ -300,7 +300,7 @@ brw_is_perf_query_ready(struct gl_context *ctx,
 /**
  * Driver hook for glGetPerfQueryDataINTEL().
  */
-static void
+static bool
 brw_get_perf_query_data(struct gl_context *ctx,
                         struct gl_perf_query_object *o,
                         GLsizei data_size,
@@ -325,6 +325,8 @@ brw_get_perf_query_data(struct gl_context *ctx,
 
    intel_perf_get_query_data(brw->perf_ctx, obj, &brw->batch,
                            data_size, data, bytes_written);
+
+   return true;
 }
 
 static struct gl_perf_query_object *

src/mesa/main/dd.h

@@ -951,7 +951,7 @@ struct dd_function_table {
                          struct gl_perf_query_object *obj);
    bool (*IsPerfQueryReady)(struct gl_context *ctx,
                             struct gl_perf_query_object *obj);
-   void (*GetPerfQueryData)(struct gl_context *ctx,
+   bool (*GetPerfQueryData)(struct gl_context *ctx,
                             struct gl_perf_query_object *obj,
                             GLsizei dataSize,
                             GLuint *data,

src/mesa/main/performance_query.c

@@ -648,6 +648,13 @@ _mesa_GetPerfQueryDataINTEL(GLuint queryHandle, GLuint flags,
       }
    }
 
-   if (obj->Ready)
-      ctx->Driver.GetPerfQueryData(ctx, obj, dataSize, data, bytesWritten);
+   if (obj->Ready) {
+      if (!ctx->Driver.GetPerfQueryData(ctx, obj, dataSize, data, bytesWritten)) {
+         memset(data, 0, dataSize);
+         *bytesWritten = 0;
+
+         _mesa_error(ctx, GL_INVALID_OPERATION,
+                     "glGetPerfQueryDataINTEL(deferred begin query failure)");
+      }
+   }
 }

src/mesa/state_tracker/st_cb_perfquery.c

@@ -184,7 +184,7 @@ st_IsPerfQueryReady(struct gl_context *ctx, struct gl_perf_query_object *o)
    return pipe->is_intel_perf_query_ready(pipe, (struct pipe_query *)o);
 }
 
-static void
+static bool
 st_GetPerfQueryData(struct gl_context *ctx,
                     struct gl_perf_query_object *o,
                     GLsizei data_size,
@@ -200,8 +200,8 @@ st_GetPerfQueryData(struct gl_context *ctx,
     */
    assert(o->Ready);
 
-   pipe->get_intel_perf_query_data(pipe, (struct pipe_query *)o, data_size, data,
-                                   bytes_written);
+   return pipe->get_intel_perf_query_data(pipe, (struct pipe_query *)o,
+                                          data_size, data, bytes_written);
 }
 
 static struct gl_perf_query_object *