[mutter/gbsneto/software-picking: 4/6] cogl/journal: Track viewport

[Georges Basile Stavracas Neto <gbsneto src gnome org>]

commit 0556138b9f731048f2c7c31c9e3a00d431ea7d9a
Author: Georges Basile Stavracas Neto <georges stavracas gmail com>
Date:   Tue Feb 5 10:11:37 2019 -0200

    cogl/journal: Track viewport
    
    CoglJournal tracks a few OpenGL states so that they can
    be batch-applied if necessary. It also has a nice property
    of allowing purely CPU-based glReadPixels() when the scene
    is composed of simple rectangles.
    
    However, the current journal implementation leaves various
    other GL states out, such as dithering and the viewport.
    In Clutter, that causes the journal to be flushed when
    picking, touching the GPU when we didn't really need to.
    
    Track the viewport of the framebuffer in the journal so that
    we can avoid flushing the journal so often.
    
    https://gitlab.gnome.org/GNOME/mutter/merge_requests/402

 cogl/cogl/cogl-framebuffer-private.h |  5 +++
 cogl/cogl/cogl-framebuffer.c         | 25 +++++++++----
 cogl/cogl/cogl-journal-private.h     |  1 +
 cogl/cogl/cogl-journal.c             | 72 ++++++++++++++++++++++++++++++------
 4 files changed, 84 insertions(+), 19 deletions(-)
---
diff --git a/cogl/cogl/cogl-framebuffer-private.h b/cogl/cogl/cogl-framebuffer-private.h
index eada73577..1f5b31e51 100644
--- a/cogl/cogl/cogl-framebuffer-private.h
+++ b/cogl/cogl/cogl-framebuffer-private.h
@@ -414,6 +414,11 @@ _cogl_framebuffer_try_creating_gl_fbo (CoglContext *ctx,
                                        CoglOffscreenAllocateFlags flags,
                                        CoglGLFramebuffer *gl_framebuffer);
 
+
+void
+cogl_framebuffer_set_viewport4fv (CoglFramebuffer *framebuffer,
+                                  float *viewport);
+
 unsigned long
 _cogl_framebuffer_compare (CoglFramebuffer *a,
                            CoglFramebuffer *b,
diff --git a/cogl/cogl/cogl-framebuffer.c b/cogl/cogl/cogl-framebuffer.c
index fb46b569b..f4d131315 100644
--- a/cogl/cogl/cogl-framebuffer.c
+++ b/cogl/cogl/cogl-framebuffer.c
@@ -481,6 +481,23 @@ _cogl_framebuffer_set_clip_stack (CoglFramebuffer *framebuffer,
   framebuffer->clip_stack = stack;
 }
 
+void
+cogl_framebuffer_set_viewport4fv (CoglFramebuffer *framebuffer,
+                                  float *viewport)
+{
+  if (framebuffer->viewport_x == viewport[0] &&
+      framebuffer->viewport_y == viewport[1] &&
+      framebuffer->viewport_width == viewport[2] &&
+      framebuffer->viewport_height == viewport[3])
+    return;
+
+  framebuffer->viewport_x = viewport[0];
+  framebuffer->viewport_y = viewport[1];
+  framebuffer->viewport_width = viewport[2];
+  framebuffer->viewport_height = viewport[3];
+  framebuffer->viewport_age++;
+}
+
 void
 cogl_framebuffer_set_viewport (CoglFramebuffer *framebuffer,
                                float x,
@@ -488,8 +505,6 @@ cogl_framebuffer_set_viewport (CoglFramebuffer *framebuffer,
                                float width,
                                float height)
 {
-  CoglContext *context = framebuffer->context;
-
   g_return_if_fail (width > 0 && height > 0);
 
   if (framebuffer->viewport_x == x &&
@@ -498,16 +513,10 @@ cogl_framebuffer_set_viewport (CoglFramebuffer *framebuffer,
       framebuffer->viewport_height == height)
     return;
 
-  _cogl_framebuffer_flush_journal (framebuffer);
-
   framebuffer->viewport_x = x;
   framebuffer->viewport_y = y;
   framebuffer->viewport_width = width;
   framebuffer->viewport_height = height;
-  framebuffer->viewport_age++;
-
-  if (context->current_draw_buffer == framebuffer)
-    context->current_draw_buffer_changes |= COGL_FRAMEBUFFER_STATE_VIEWPORT;
 }
 
 float
diff --git a/cogl/cogl/cogl-journal-private.h b/cogl/cogl/cogl-journal-private.h
index 5d38877d4..4b83e3bf2 100644
--- a/cogl/cogl/cogl-journal-private.h
+++ b/cogl/cogl/cogl-journal-private.h
@@ -78,6 +78,7 @@ typedef struct _CoglJournalEntry
   CoglPipeline            *pipeline;
   CoglMatrixEntry         *modelview_entry;
   CoglClipStack           *clip_stack;
+  float                    viewport[4];
   /* Offset into ctx->logged_vertices */
   size_t                   array_offset;
   int                      n_layers;
diff --git a/cogl/cogl/cogl-journal.c b/cogl/cogl/cogl-journal.c
index 083f4a3d4..4f7cb0f05 100644
--- a/cogl/cogl/cogl-journal.c
+++ b/cogl/cogl/cogl-journal.c
@@ -1044,6 +1044,55 @@ compare_entry_clip_stacks (CoglJournalEntry *entry0, CoglJournalEntry *entry1)
   return entry0->clip_stack == entry1->clip_stack;
 }
 
+static void
+_cogl_journal_flush_viewport_and_entries (CoglJournalEntry *batch_start,
+                                          int               batch_len,
+                                          void             *data)
+{
+  CoglJournalFlushState *state = data;
+  CoglFramebuffer *framebuffer = state->journal->framebuffer;
+  CoglContext *ctx = framebuffer->context;
+  float current_viewport[4];
+
+  COGL_STATIC_TIMER (time_flush_viewport_and_entries,
+                     "Journal Flush", /* parent */
+                     "flush: viewport+clip+vbo+texcoords+pipeline+entries",
+                     "The time spent flushing viewport + clip + vbo + texcoord offsets + "
+                     "pipeline + entries",
+                     0 /* no application private data */);
+
+  COGL_TIMER_START (_cogl_uprof_context, time_flush_viewport_and_entries);
+
+  if (G_UNLIKELY (COGL_DEBUG_ENABLED (COGL_DEBUG_BATCHING)))
+    g_print ("BATCHING:  viewport batch len = %d\n", batch_len);
+
+  ctx->current_draw_buffer_changes |= COGL_FRAMEBUFFER_STATE_VIEWPORT;
+
+  cogl_framebuffer_get_viewport4fv (framebuffer, current_viewport);
+  cogl_framebuffer_set_viewport4fv (framebuffer, batch_start->viewport);
+
+  _cogl_framebuffer_flush_state (framebuffer,
+                                 framebuffer,
+                                 COGL_FRAMEBUFFER_STATE_VIEWPORT);
+
+  batch_and_call (batch_start,
+                  batch_len,
+                  compare_entry_clip_stacks,
+                  _cogl_journal_flush_clip_stacks_and_entries,
+                  state);
+
+  if (memcmp (batch_start->viewport, current_viewport, sizeof (float) * 4) != 0)
+    cogl_framebuffer_set_viewport4fv (framebuffer, current_viewport);
+
+  COGL_TIMER_STOP (_cogl_uprof_context, time_flush_viewport_and_entries);
+}
+
+static gboolean
+compare_entry_viewports (CoglJournalEntry *entry0, CoglJournalEntry *entry1)
+{
+  return memcmp (entry0->viewport, entry1->viewport, sizeof (float) * 4) == 0;
+}
+
 /* Gets a new vertex array from the pool. A reference is taken on the
    array so it can be treated as if it was just newly allocated */
 static CoglAttributeBuffer *
@@ -1331,12 +1380,13 @@ _cogl_journal_flush (CoglJournal *journal)
   if (G_UNLIKELY (COGL_DEBUG_ENABLED (COGL_DEBUG_BATCHING)))
     g_print ("BATCHING: journal len = %d\n", journal->entries->len);
 
-  /* NB: the journal deals with flushing the modelview stack and clip
-     state manually */
+  /* NB: the journal deals with flushing the viewport, the modelview
+   * stack and clip state manually */
   _cogl_framebuffer_flush_state (framebuffer,
                                  framebuffer,
                                  COGL_FRAMEBUFFER_STATE_ALL &
-                                 ~(COGL_FRAMEBUFFER_STATE_MODELVIEW |
+                                 ~(COGL_FRAMEBUFFER_STATE_VIEWPORT |
+                                   COGL_FRAMEBUFFER_STATE_MODELVIEW |
                                    COGL_FRAMEBUFFER_STATE_CLIP));
 
   /* We need to mark the current modelview state of the framebuffer as
@@ -1395,11 +1445,11 @@ _cogl_journal_flush (CoglJournal *journal)
    *      Note: Splitting by modelview changes is skipped when are doing the
    *      vertex transformation in software at log time.
    */
-  batch_and_call ((CoglJournalEntry *)journal->entries->data, /* first entry */
-                  journal->entries->len, /* max number of entries to consider */
-                  compare_entry_clip_stacks,
-                  _cogl_journal_flush_clip_stacks_and_entries, /* callback */
-                  &state); /* data */
+  batch_and_call ((CoglJournalEntry *)journal->entries->data,
+                  journal->entries->len,
+                  compare_entry_viewports,
+                  _cogl_journal_flush_viewport_and_entries,
+                  &state);
 
   for (i = 0; i < state.attributes->len; i++)
     cogl_object_unref (g_array_index (state.attributes, CoglAttribute *, i));
@@ -1553,6 +1603,8 @@ _cogl_journal_log_quad (CoglJournal  *journal,
   clip_stack = _cogl_framebuffer_get_clip_stack (framebuffer);
   entry->clip_stack = _cogl_clip_stack_ref (clip_stack);
 
+  cogl_framebuffer_get_viewport4fv (framebuffer, entry->viewport);
+
   if (G_UNLIKELY (final_pipeline != pipeline))
     cogl_object_unref (final_pipeline);
 
@@ -1582,7 +1634,7 @@ entry_to_screen_polygon (CoglFramebuffer *framebuffer,
   CoglMatrix projection;
   CoglMatrix modelview;
   int i;
-  float viewport[4];
+  const float *viewport = entry->viewport;
 
   poly[0] = vertices[0];
   poly[1] = vertices[1];
@@ -1631,8 +1683,6 @@ entry_to_screen_polygon (CoglFramebuffer *framebuffer,
                               poly, /* points_out */
                               4 /* n_points */);
 
-  cogl_framebuffer_get_viewport4fv (framebuffer, viewport);
-
 /* Scale from OpenGL normalized device coordinates (ranging from -1 to 1)
  * to Cogl window/framebuffer coordinates (ranging from 0 to buffer-size) with
  * (0,0) being top left. */