freedreno/perfetto: Add non-draw stage

Make it easier to see the scope of non-draw passes, rather than only
showing individual compute/blit jobs.

Signed-off-by: Rob Clark <rob.clark@oss.qualcomm.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/42013>

src/gallium/drivers/freedreno/freedreno_gmem.c

@@ -708,14 +708,11 @@ fd_gmem_render_tiles(struct fd_batch *batch)
    if (!(batch->cleared || batch->num_draws))
       sysmem = true;
 
-   if (!batch->nondraw) {
 #if HAVE_PERFETTO
-      /* For non-draw batches, we don't really have a good place to
-       * match up the api event submit-id to the on-gpu rendering,
-       * so skip this for non-draw batches.
-       */
-      fd_perfetto_submit(ctx);
+   fd_perfetto_submit(ctx);
 #endif
+
+   if (!batch->nondraw) {
       trace_flush_batch(&batch->trace, batch->gmem, batch, batch->cleared,
                         batch->gmem_reason, batch->num_draws);
       trace_framebuffer_state(&batch->trace, batch->gmem, pfb);
@@ -760,8 +757,11 @@ fd_gmem_render_tiles(struct fd_batch *batch)
 
    if (batch->nondraw) {
       DBG("%p: rendering non-draw", batch);
-      if (!fd_ringbuffer_empty(batch->draw))
+      if (!fd_ringbuffer_empty(batch->draw)) {
+         trace_start_nondraw(&batch->trace, batch->gmem, ctx->submit_count);
          render_sysmem(batch);
+         trace_end_nondraw(&batch->trace, batch->gmem);
+      }
       ctx->stats.batch_nondraw++;
    } else if (sysmem) {
       trace_render_sysmem(&batch->trace, batch->gmem);

src/gallium/drivers/freedreno/freedreno_perfetto.cc

@@ -187,6 +187,8 @@ stage_end(struct pipe_context *pctx, uint64_t ts_ns, enum fd_stage_id stage)
             data->set_name("binHeight");
             data->set_value(std::to_string(p->binh));
          }
+      } else if (stage == NONDRAW_STAGE_ID) {
+         event->set_submission_id(p->submit_id);
       } else if (stage == COMPUTE_STAGE_ID) {
          {
             auto data = event->add_extra_data();
@@ -368,6 +370,28 @@ fd_end_render_pass(struct pipe_context *pctx, uint64_t ts_ns,
    stage_end(pctx, ts_ns, SURFACE_STAGE_ID);
 }
 
+void
+fd_start_nondraw(struct pipe_context *pctx, uint64_t ts_ns,
+                 uint16_t tp_idx, const void *flush_data,
+                 const struct trace_start_nondraw *payload,
+                 const void *indirect_data)
+{
+   stage_start(pctx, ts_ns, NONDRAW_STAGE_ID);
+
+   struct fd_perfetto_state *p = &fd_context(pctx)->perfetto;
+
+   p->submit_id = payload->submit_id;
+}
+
+void
+fd_end_nondraw(struct pipe_context *pctx, uint64_t ts_ns,
+               uint16_t tp_idx, const void *flush_data,
+               const struct trace_end_nondraw *payload,
+               const void *indirect_data)
+{
+   stage_end(pctx, ts_ns, NONDRAW_STAGE_ID);
+}
+
 void
 fd_start_binning_ib(struct pipe_context *pctx, uint64_t ts_ns,
                     uint16_t tp_idx, const void *flush_data,

src/gallium/drivers/freedreno/freedreno_perfetto.h

@@ -17,6 +17,7 @@ extern "C" {
  */
 enum fd_stage_id {
    SURFACE_STAGE_ID, /* Surface is a sort of meta-stage for render-target info */
+   NONDRAW_STAGE_ID,
    BINNING_STAGE_ID,
    GMEM_STAGE_ID,
    BYPASS_STAGE_ID,
@@ -37,6 +38,7 @@ static const struct {
    const char *desc;
 } stages[] = {
    [SURFACE_STAGE_ID] = {"Surface"},
+   [NONDRAW_STAGE_ID] = {"Non-draw"},
    [BINNING_STAGE_ID] = {"Binning", "Perform Visibility pass and determine target bins"},
    [GMEM_STAGE_ID]    = {"Render", "Rendering to GMEM"},
    [BYPASS_STAGE_ID]  = {"Render", "Rendering to system memory"},

src/gallium/drivers/freedreno/freedreno_tracepoints.py

@@ -102,6 +102,10 @@ begin_end_tp('render_pass',
           TracepointArg(type='uint16_t',         var='binh',          c_format='%u')],
 )
 
+begin_end_tp('nondraw',
+    args=[TracepointArg(type='uint32_t',         var='submit_id',     c_format='%u')],
+)
+
 begin_end_tp('binning_ib')
 begin_end_tp('vsc_overflow_test')
 begin_end_tp('prologue')