tu/perfetto: Add a performance warning track to perfetto

The idea is to emit single tracepoints with warning that would
stick until the relevant render stage ends.

Signed-off-by: Danylo Piliaiev <dpiliaiev@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/40935>

src/freedreno/vulkan/tu_cs.cc

@@ -630,6 +630,15 @@ tu_cs_trace_end(struct u_trace_context *utctx, void *cs, const char *fmt, ...)
    va_end(args);
 }
 
+void
+tu_cs_trace_singular(struct u_trace_context *utctx, void *cs, const char *fmt, ...)
+{
+   va_list args;
+   va_start(args, fmt);
+   tu_cs_emit_debug_magic_strv((struct tu_cs *) cs, CP_NOP_MESG, fmt, args);
+   va_end(args);
+}
+
 tu_crb
 tu_cs::crb(uint32_t nregs)
 {

src/freedreno/vulkan/tu_cs.h

@@ -450,6 +450,9 @@ tu_cs_trace_start(struct u_trace_context *utctx,
 __attribute__((format(printf, 3, 4))) void
 tu_cs_trace_end(struct u_trace_context *utctx, void *cs, const char *fmt, ...);
 
+__attribute__((format(printf, 3, 4))) void
+tu_cs_trace_singular(struct u_trace_context *utctx, void *cs, const char *fmt, ...);
+
 struct tu_cs_patchable_state {
    uint32_t *nop_header;
    uint32_t dwords;

src/freedreno/vulkan/tu_perfetto.cc

@@ -38,6 +38,7 @@ static const struct {
    [ANNOTATIONS_QUEUE_ID] = {"Annotations", "Annotations Queue"},
    [BR_HW_QUEUE_ID] = {"GPU Queue 0", "Default Adreno Hardware Queue"},
    [BV_HW_QUEUE_ID] = {"GPU Queue 1", "Adreno Bin Visibility Queue"},
+   [PERF_WARNINGS_QUEUE_ID] = {"Performance Warnings", "Performance Warnings Queue"},
 };
 
 static const struct {
@@ -330,6 +331,23 @@ stage_end(struct tu_device *dev, uint64_t ts_ns, enum tu_stage_id stage_id,
    }
 
    uint32_t queue_id = BR_HW_QUEUE_ID;
+
+   /* Warnings should last until the end of the corresponding stage.
+    * This is a roundabout way to achive that, since we don't track
+    * which stage warning corresponds to.
+    */
+   if (stage_id != CLEAR_SYSMEM_STAGE_ID &&
+       stage_id != CLEAR_GMEM_STAGE_ID &&
+       stage_id != GENERIC_CLEAR_STAGE_ID &&
+       stage_id != GMEM_LOAD_STAGE_ID &&
+       stage_id != GMEM_STORE_STAGE_ID) {
+      while (dev->perfetto.sticky_warnings_stack.stage_depth > 0) {
+         struct tu_perfetto_stage *stage =
+            &dev->perfetto.sticky_warnings_stack.stages[dev->perfetto.sticky_warnings_stack.stage_depth - 1];
+         stage_end(dev, ts_ns, (tu_stage_id) stage->stage_id, nullptr, flush_data);
+      }
+   }
+
    switch (stage->stage_id) {
       case CMD_BUFFER_ANNOTATION_STAGE_ID:
       case CMD_BUFFER_ANNOTATION_RENDER_PASS_STAGE_ID:
@@ -587,6 +605,14 @@ tu_perfetto_end_submit(struct tu_queue *queue,
          (trace_payload_as_extra_func) &trace_payload_as_extra_end_##event_name);   \
    }
 
+#define CREATE_STICKY_WARNING_EVENT_CALLBACK(event_name, stage_id)                                                     \
+   void tu_perfetto_##event_name(struct tu_device *dev, uint64_t ts_ns, uint16_t tp_idx, const void *flush_data,       \
+                                 const struct trace_##event_name *payload, const void *indirect_data)                  \
+   {                                                                                                                   \
+      stage_start(dev, ts_ns, stage_id, payload, sizeof(*payload), indirect_data,                                      \
+                  (trace_payload_as_extra_func) & trace_payload_as_extra_##event_name);                                \
+   }
+
 CREATE_EVENT_CALLBACK(cmd_buffer, CMD_BUFFER_STAGE_ID)
 CREATE_EVENT_CALLBACK(secondary_cmd_buffer, SECONDARY_CMD_BUFFER_STAGE_ID)
 CREATE_EVENT_CALLBACK(render_pass, RENDER_PASS_STAGE_ID)

src/freedreno/vulkan/tu_perfetto.h

@@ -41,6 +41,7 @@ enum tu_queue_id {
    ANNOTATIONS_QUEUE_ID,
    BR_HW_QUEUE_ID,
    BV_HW_QUEUE_ID,
+   PERF_WARNINGS_QUEUE_ID,
 
    TU_QUEUE_ID_COUNT,
 };
@@ -111,6 +112,7 @@ struct tu_perfetto_clocks
 struct tu_perfetto_state {
    struct tu_perfetto_stage_stack annotations_stack;
    struct tu_perfetto_stage_stack render_stack;
+   struct tu_perfetto_stage_stack sticky_warnings_stack;
 
    uint64_t context_iid;
    uint64_t queue_iids[TU_QUEUE_ID_COUNT];

src/freedreno/vulkan/tu_tracepoints.py

@@ -83,6 +83,29 @@ def begin_end_tp(name, args=[], tp_struct=None, tp_print=None,
                tp_print=end_tp_print if queue_tp else None,
                tp_markers='tu_cs_trace_end' if marker_tp else None)
 
+def singular_tp(name, args=[], tp_struct=None, tp_print=None,
+                tp_default_enabled=True, marker_tp=True,
+                queue_tp=True, toggle_name=None):
+    global tu_default_tps
+
+    if not toggle_name:
+        toggle_name = name
+
+    if tp_default_enabled and toggle_name not in tu_default_tps:
+        tu_default_tps.append(toggle_name)
+
+    tp_struct = [command_buffer_struct] + (tp_struct if tp_struct else [])
+    args = [command_buffer_arg] + (args if args else [])
+
+    Tracepoint('{0}'.format(name),
+               toggle_name=toggle_name,
+               args=args,
+               tp_struct=tp_struct,
+               tp_perfetto='tu_perfetto_{0}'.format(name) if queue_tp else None,
+               tp_print=tp_print if queue_tp else None,
+               tp_markers='tu_cs_trace_singular' if marker_tp else None)
+
+
 begin_end_tp('cmd_buffer',
     args=[Arg(type='str',                       var='TUdebugFlags', c_format='%s', length_arg='96', copy_func='strncpy'),
           Arg(type='str',                       var='IR3debugFlags', c_format='%s', length_arg='96', copy_func='strncpy')],