godotengine/godot
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Add frame diagnostics for GLES2 Batch renderer

Browse Source 
Added project setting to enable / disable print frame diagnostics every 10 seconds. This prints out a list of batches and info, which is useful to optimize games and identify performance problems.
This commit is contained in:
lawnjelly 2020-04-17 08:44:12 +01:00
parent b6d652367b
commit 72adefa5cf
7 changed files with 139 additions and 20 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

30
doc/classes/Performance.xml
View File

@ -79,40 +79,46 @@
<constant name="RENDER_DRAW_CALLS_IN_FRAME" value="17" enum="Monitor"> <constant name="RENDER_DRAW_CALLS_IN_FRAME" value="17" enum="Monitor">
Draw calls per frame. 3D only. Draw calls per frame. 3D only.
</constant> </constant>
<constant name="RENDER_VIDEO_MEM_USED" value="18" enum="Monitor"> <constant name="RENDER_2D_ITEMS_IN_FRAME" value="18" enum="Monitor">
Items or joined items drawn per frame.
</constant>
<constant name="RENDER_2D_DRAW_CALLS_IN_FRAME" value="19" enum="Monitor">
Draw calls per frame.
</constant>
<constant name="RENDER_VIDEO_MEM_USED" value="20" enum="Monitor">
The amount of video memory used, i.e. texture and vertex memory combined. The amount of video memory used, i.e. texture and vertex memory combined.
</constant> </constant>
<constant name="RENDER_TEXTURE_MEM_USED" value="19" enum="Monitor"> <constant name="RENDER_TEXTURE_MEM_USED" value="21" enum="Monitor">
The amount of texture memory used. The amount of texture memory used.
</constant> </constant>
<constant name="RENDER_VERTEX_MEM_USED" value="20" enum="Monitor"> <constant name="RENDER_VERTEX_MEM_USED" value="22" enum="Monitor">
The amount of vertex memory used. The amount of vertex memory used.
</constant> </constant>
<constant name="RENDER_USAGE_VIDEO_MEM_TOTAL" value="21" enum="Monitor"> <constant name="RENDER_USAGE_VIDEO_MEM_TOTAL" value="23" enum="Monitor">
Unimplemented in the GLES2 and GLES3 rendering backends, always returns 0. Unimplemented in the GLES2 and GLES3 rendering backends, always returns 0.
</constant> </constant>
<constant name="PHYSICS_2D_ACTIVE_OBJECTS" value="22" enum="Monitor"> <constant name="PHYSICS_2D_ACTIVE_OBJECTS" value="24" enum="Monitor">
Number of active [RigidBody2D] nodes in the game. Number of active [RigidBody2D] nodes in the game.
</constant> </constant>
<constant name="PHYSICS_2D_COLLISION_PAIRS" value="23" enum="Monitor"> <constant name="PHYSICS_2D_COLLISION_PAIRS" value="25" enum="Monitor">
Number of collision pairs in the 2D physics engine. Number of collision pairs in the 2D physics engine.
</constant> </constant>
<constant name="PHYSICS_2D_ISLAND_COUNT" value="24" enum="Monitor"> <constant name="PHYSICS_2D_ISLAND_COUNT" value="26" enum="Monitor">
Number of islands in the 2D physics engine. Number of islands in the 2D physics engine.
</constant> </constant>
<constant name="PHYSICS_3D_ACTIVE_OBJECTS" value="25" enum="Monitor"> <constant name="PHYSICS_3D_ACTIVE_OBJECTS" value="27" enum="Monitor">
Number of active [RigidBody] and [VehicleBody] nodes in the game. Number of active [RigidBody] and [VehicleBody] nodes in the game.
</constant> </constant>
<constant name="PHYSICS_3D_COLLISION_PAIRS" value="26" enum="Monitor"> <constant name="PHYSICS_3D_COLLISION_PAIRS" value="28" enum="Monitor">
Number of collision pairs in the 3D physics engine. Number of collision pairs in the 3D physics engine.
</constant> </constant>
<constant name="PHYSICS_3D_ISLAND_COUNT" value="27" enum="Monitor"> <constant name="PHYSICS_3D_ISLAND_COUNT" value="29" enum="Monitor">
Number of islands in the 3D physics engine. Number of islands in the 3D physics engine.
</constant> </constant>
<constant name="AUDIO_OUTPUT_LATENCY" value="28" enum="Monitor"> <constant name="AUDIO_OUTPUT_LATENCY" value="30" enum="Monitor">
Output latency of the [AudioServer]. Output latency of the [AudioServer].
</constant> </constant>
<constant name="MONITOR_MAX" value="29" enum="Monitor"> <constant name="MONITOR_MAX" value="31" enum="Monitor">
Represents the size of the [enum Monitor] enum. Represents the size of the [enum Monitor] enum.
</constant> </constant>
</constants> </constants>

5
doc/classes/ProjectSettings.xml
View File

@ -984,10 +984,13 @@
<member name="rendering/gles2/batching/use_batching" type="bool" setter="" getter="" default="true"> <member name="rendering/gles2/batching/use_batching" type="bool" setter="" getter="" default="true">
Turns batching on and off. Batching increases performance by reducing the amount of graphics API drawcalls. Turns batching on and off. Batching increases performance by reducing the amount of graphics API drawcalls.
</member> </member>
<member name="rendering/gles2/debug/diagnose_frame" type="bool" setter="" getter="" default="false">
When batching is on, this regularly prints a frame diagnosis log. Note that this will degrade performance.
</member>
<member name="rendering/gles2/debug/flash_batching" type="bool" setter="" getter="" default="false"> <member name="rendering/gles2/debug/flash_batching" type="bool" setter="" getter="" default="false">
[b]Experimental[/b] For regression testing against the old renderer. If this is switched on, and [code]use_batching[/code] is set, the renderer will swap alternately between using the old renderer, and the batched renderer, on each frame. This makes it easy to identify visual differences. Performance will be degraded. [b]Experimental[/b] For regression testing against the old renderer. If this is switched on, and [code]use_batching[/code] is set, the renderer will swap alternately between using the old renderer, and the batched renderer, on each frame. This makes it easy to identify visual differences. Performance will be degraded.
</member> </member>
<member name="rendering/gles2/debug/use_batching_in_editor" type="bool" setter="" getter="" default="false"> <member name="rendering/gles2/debug/use_batching_in_editor" type="bool" setter="" getter="" default="true">
[b]Experimental[/b] Switches on batching within the editor. Use with caution - note that if your editor does not render correctly you may need to edit your [code]project.godot[/code] and remove the use_batching_in_editor setting manually. [b]Experimental[/b] Switches on batching within the editor. Use with caution - note that if your editor does not render correctly you may need to edit your [code]project.godot[/code] and remove the use_batching_in_editor setting manually.
</member> </member>
<member name="rendering/limits/buffers/blend_shape_max_buffer_size_kb" type="int" setter="" getter="" default="4096"> <member name="rendering/limits/buffers/blend_shape_max_buffer_size_kb" type="int" setter="" getter="" default="4096">

8
doc/classes/Viewport.xml
View File

@ -377,7 +377,13 @@
<constant name="RENDER_INFO_DRAW_CALLS_IN_FRAME" value="5" enum="RenderInfo"> <constant name="RENDER_INFO_DRAW_CALLS_IN_FRAME" value="5" enum="RenderInfo">
Amount of draw calls in frame. Amount of draw calls in frame.
</constant> </constant>
<constant name="RENDER_INFO_MAX" value="6" enum="RenderInfo"> <constant name="RENDER_INFO_2D_ITEMS_IN_FRAME" value="6" enum="RenderInfo">
Amount of items or joined items in frame.
</constant>
<constant name="RENDER_INFO_2D_DRAW_CALLS_IN_FRAME" value="7" enum="RenderInfo">
Amount of draw calls in frame.
</constant>
<constant name="RENDER_INFO_MAX" value="8" enum="RenderInfo">
Represents the size of the [enum RenderInfo] enum. Represents the size of the [enum RenderInfo] enum.
</constant> </constant>
<constant name="DEBUG_DRAW_DISABLED" value="0" enum="DebugDraw"> <constant name="DEBUG_DRAW_DISABLED" value="0" enum="DebugDraw">

22
doc/classes/VisualServer.xml
View File

@ -4601,7 +4601,13 @@
<constant name="VIEWPORT_RENDER_INFO_DRAW_CALLS_IN_FRAME" value="5" enum="ViewportRenderInfo"> <constant name="VIEWPORT_RENDER_INFO_DRAW_CALLS_IN_FRAME" value="5" enum="ViewportRenderInfo">
Number of draw calls during this frame. Number of draw calls during this frame.
</constant> </constant>
<constant name="VIEWPORT_RENDER_INFO_MAX" value="6" enum="ViewportRenderInfo"> <constant name="VIEWPORT_RENDER_INFO_2D_ITEMS_IN_FRAME" value="6" enum="ViewportRenderInfo">
Number of 2d items drawn this frame.
</constant>
<constant name="VIEWPORT_RENDER_INFO_2D_DRAW_CALLS_IN_FRAME" value="7" enum="ViewportRenderInfo">
Number of 2d draw calls during this frame.
</constant>
<constant name="VIEWPORT_RENDER_INFO_MAX" value="8" enum="ViewportRenderInfo">
Represents the size of the [enum ViewportRenderInfo] enum. Represents the size of the [enum ViewportRenderInfo] enum.
</constant> </constant>
<constant name="VIEWPORT_DEBUG_DRAW_DISABLED" value="0" enum="ViewportDebugDraw"> <constant name="VIEWPORT_DEBUG_DRAW_DISABLED" value="0" enum="ViewportDebugDraw">
@ -4748,16 +4754,22 @@
<constant name="INFO_DRAW_CALLS_IN_FRAME" value="5" enum="RenderInfo"> <constant name="INFO_DRAW_CALLS_IN_FRAME" value="5" enum="RenderInfo">
The amount of draw calls in frame. The amount of draw calls in frame.
</constant> </constant>
<constant name="INFO_USAGE_VIDEO_MEM_TOTAL" value="6" enum="RenderInfo"> <constant name="INFO_2D_ITEMS_IN_FRAME" value="6" enum="RenderInfo">
The amount of 2d items in the frame.
</constant>
<constant name="INFO_2D_DRAW_CALLS_IN_FRAME" value="7" enum="RenderInfo">
The amount of 2d draw calls in frame.
</constant>
<constant name="INFO_USAGE_VIDEO_MEM_TOTAL" value="8" enum="RenderInfo">
Unimplemented in the GLES2 and GLES3 rendering backends, always returns 0. Unimplemented in the GLES2 and GLES3 rendering backends, always returns 0.
</constant> </constant>
<constant name="INFO_VIDEO_MEM_USED" value="7" enum="RenderInfo"> <constant name="INFO_VIDEO_MEM_USED" value="9" enum="RenderInfo">
The amount of video memory used, i.e. texture and vertex memory combined. The amount of video memory used, i.e. texture and vertex memory combined.
</constant> </constant>
<constant name="INFO_TEXTURE_MEM_USED" value="8" enum="RenderInfo"> <constant name="INFO_TEXTURE_MEM_USED" value="10" enum="RenderInfo">
The amount of texture memory used. The amount of texture memory used.
</constant> </constant>
<constant name="INFO_VERTEX_MEM_USED" value="9" enum="RenderInfo"> <constant name="INFO_VERTEX_MEM_USED" value="11" enum="RenderInfo">
The amount of vertex memory used. The amount of vertex memory used.
</constant> </constant>
<constant name="FEATURE_SHADERS" value="0" enum="Features"> <constant name="FEATURE_SHADERS" value="0" enum="Features">

82
drivers/gles2/rasterizer_canvas_gles2.cpp
View File

@ -60,9 +60,13 @@ RasterizerCanvasGLES2::BatchData::BatchData() {
settings_colored_vertex_format_threshold = 0.0f; settings_colored_vertex_format_threshold = 0.0f;
settings_batch_buffer_num_verts = 0; settings_batch_buffer_num_verts = 0;
scissor_threshold_area = 0.0f; scissor_threshold_area = 0.0f;
diagnose_frame = false;
next_diagnose_tick = 10000;
diagnose_frame_number = 9999999999; // some high number
settings_use_batching_original_choice = false; settings_use_batching_original_choice = false;
settings_flash_batching = false; settings_flash_batching = false;
settings_diagnose_frame = false;
settings_scissor_lights = false; settings_scissor_lights = false;
settings_scissor_threshold = -1.0f; settings_scissor_threshold = -1.0f;
} }
@ -642,6 +646,32 @@ void RasterizerCanvasGLES2::_batch_render_rects(const Batch &p_batch, Rasterizer
glBindBuffer(GL_ELEMENT_ARRAY_BUFFER, 0); glBindBuffer(GL_ELEMENT_ARRAY_BUFFER, 0);
} }
void RasterizerCanvasGLES2::diagnose_batches(Item::Command *const *p_commands) {
int num_batches = bdata.batches.size();
for (int batch_num = 0; batch_num < num_batches; batch_num++) {
const Batch &batch = bdata.batches[batch_num];
bdata.frame_string += "\t\tbatch ";
switch (batch.type) {
case Batch::BT_RECT: {
bdata.frame_string += "R ";
bdata.frame_string += itos(batch.num_commands);
bdata.frame_string += " [" + itos(batch.batch_texture_id) + "]";
if (batch.num_commands > 1) {
bdata.frame_string += " MULTI\n";
} else {
bdata.frame_string += "\n";
}
} break;
default: {
bdata.frame_string += "D ";
bdata.frame_string += itos(batch.num_commands) + "\n";
} break;
}
}
}
void RasterizerCanvasGLES2::render_batches(Item::Command *const *p_commands, Item *p_current_clip, bool &r_reclip, RasterizerStorageGLES2::Material *p_material) { void RasterizerCanvasGLES2::render_batches(Item::Command *const *p_commands, Item *p_current_clip, bool &r_reclip, RasterizerStorageGLES2::Material *p_material) {
int num_batches = bdata.batches.size(); int num_batches = bdata.batches.size();
@ -1562,6 +1592,10 @@ void RasterizerCanvasGLES2::flush_render_batches(Item *p_first_item, Item *p_cur
Item::Command *const *commands = p_first_item->commands.ptr(); Item::Command *const *commands = p_first_item->commands.ptr();
if (bdata.diagnose_frame) {
diagnose_batches(commands);
}
render_batches(commands, p_current_clip, r_reclip, p_material); render_batches(commands, p_current_clip, r_reclip, p_material);
} }
@ -1637,6 +1671,33 @@ void RasterizerCanvasGLES2::join_items(Item *p_item_list, int p_z) {
} }
} }
void RasterizerCanvasGLES2::canvas_begin() {
// diagnose_frame?
if (bdata.settings_diagnose_frame) {
bdata.diagnose_frame = false;
uint32_t tick = OS::get_singleton()->get_ticks_msec();
uint64_t frame = Engine::get_singleton()->get_frames_drawn();
if (tick >= bdata.next_diagnose_tick) {
bdata.next_diagnose_tick = tick + 10000;
// the plus one is prevent starting diagnosis half way through frame
bdata.diagnose_frame_number = frame + 1;
}
if (frame == bdata.diagnose_frame_number) {
bdata.diagnose_frame = true;
}
if (bdata.diagnose_frame) {
bdata.frame_string = "canvas_begin FRAME " + itos(frame) + "\n";
}
}
RasterizerCanvasBaseGLES2::canvas_begin();
}
void RasterizerCanvasGLES2::canvas_render_items_begin(const Color &p_modulate, Light *p_light, const Transform2D &p_base_transform) { void RasterizerCanvasGLES2::canvas_render_items_begin(const Color &p_modulate, Light *p_light, const Transform2D &p_base_transform) {
// if we are debugging, flash each frame between batching renderer and old version to compare for regressions // if we are debugging, flash each frame between batching renderer and old version to compare for regressions
if (bdata.settings_flash_batching) { if (bdata.settings_flash_batching) {
@ -1666,6 +1727,10 @@ void RasterizerCanvasGLES2::canvas_render_items_end() {
return; return;
} }
if (bdata.diagnose_frame) {
bdata.frame_string += "items\n";
}
// batching render is deferred until after going through all the z_indices, joining all the items // batching render is deferred until after going through all the z_indices, joining all the items
canvas_render_items_implementation(0, 0, _render_item_state.item_group_modulate, canvas_render_items_implementation(0, 0, _render_item_state.item_group_modulate,
_render_item_state.item_group_light, _render_item_state.item_group_light,
@ -1673,6 +1738,10 @@ void RasterizerCanvasGLES2::canvas_render_items_end() {
bdata.items_joined.reset(); bdata.items_joined.reset();
bdata.item_refs.reset(); bdata.item_refs.reset();
if (bdata.diagnose_frame) {
print_line(bdata.frame_string);
}
} }
void RasterizerCanvasGLES2::canvas_render_items(Item *p_item_list, int p_z, const Color &p_modulate, Light *p_light, const Transform2D &p_base_transform) { void RasterizerCanvasGLES2::canvas_render_items(Item *p_item_list, int p_z, const Color &p_modulate, Light *p_light, const Transform2D &p_base_transform) {
@ -2284,6 +2353,10 @@ void RasterizerCanvasGLES2::render_joined_item(const BItemJoined &p_bij, RenderI
storage->info.render._2d_item_count++; storage->info.render._2d_item_count++;
if (bdata.diagnose_frame) {
bdata.frame_string += "\tjoined_item " + itos(p_bij.num_item_refs) + " refs\n";
}
// all the joined items will share the same state with the first item // all the joined items will share the same state with the first item
Item *ci = bdata.item_refs[p_bij.first_item_ref].item; Item *ci = bdata.item_refs[p_bij.first_item_ref].item;
@ -2798,6 +2871,12 @@ void RasterizerCanvasGLES2::initialize() {
bdata.settings_flash_batching = false; bdata.settings_flash_batching = false;
} }
// frame diagnosis. print out the batches every nth frame
bdata.settings_diagnose_frame = false;
if (!Engine::get_singleton()->is_editor_hint() && bdata.settings_use_batching) {
bdata.settings_diagnose_frame = GLOBAL_GET("rendering/gles2/debug/diagnose_frame");
}
// the maximum num quads in a batch is limited by GLES2. We can have only 16 bit indices, // the maximum num quads in a batch is limited by GLES2. We can have only 16 bit indices,
// which means we can address a vertex buffer of max size 65535. 4 vertices are needed per quad. // which means we can address a vertex buffer of max size 65535. 4 vertices are needed per quad.
@ -2823,7 +2902,8 @@ void RasterizerCanvasGLES2::initialize() {
batching_options_string += "\tcolored_vertex_format_threshold " + String(Variant(bdata.settings_colored_vertex_format_threshold)) + "\n"; batching_options_string += "\tcolored_vertex_format_threshold " + String(Variant(bdata.settings_colored_vertex_format_threshold)) + "\n";
batching_options_string += "\tbatch_buffer_size " + itos(bdata.settings_batch_buffer_num_verts) + "\n"; batching_options_string += "\tbatch_buffer_size " + itos(bdata.settings_batch_buffer_num_verts) + "\n";
batching_options_string += "\tlight_scissor_area_threshold " + String(Variant(bdata.settings_scissor_threshold)) + "\n"; batching_options_string += "\tlight_scissor_area_threshold " + String(Variant(bdata.settings_scissor_threshold)) + "\n";
batching_options_string += "\tdebug_flash " + String(Variant(bdata.settings_flash_batching)); batching_options_string += "\tdebug_flash " + String(Variant(bdata.settings_flash_batching)) + "\n";
batching_options_string += "\tdiagnose_frame " + String(Variant(bdata.settings_diagnose_frame));
} else { } else {
batching_options_string += "OFF"; batching_options_string += "OFF";
} }

11
drivers/gles2/rasterizer_canvas_gles2.h
View File

@ -169,10 +169,17 @@ class RasterizerCanvasGLES2 : public RasterizerCanvasBaseGLES2 {
// measured in pixels, recalculated each frame // measured in pixels, recalculated each frame
float scissor_threshold_area; float scissor_threshold_area;
// diagnose this frame, every nTh frame when settings_diagnose_frame is on
bool diagnose_frame;
String frame_string;
uint32_t next_diagnose_tick;
uint64_t diagnose_frame_number;
// global settings // global settings
bool settings_use_batching; // the current use_batching (affected by flash) bool settings_use_batching; // the current use_batching (affected by flash)
bool settings_use_batching_original_choice; // the choice entered in project settings bool settings_use_batching_original_choice; // the choice entered in project settings
bool settings_flash_batching; // for regression testing, flash between non-batched and batched renderer bool settings_flash_batching; // for regression testing, flash between non-batched and batched renderer
bool settings_diagnose_frame; // print out batches to help optimize / regression test
int settings_max_join_item_commands; int settings_max_join_item_commands;
float settings_colored_vertex_format_threshold; float settings_colored_vertex_format_threshold;
int settings_batch_buffer_num_verts; int settings_batch_buffer_num_verts;
@ -227,6 +234,7 @@ public:
virtual void canvas_render_items_begin(const Color &p_modulate, Light *p_light, const Transform2D &p_base_transform); virtual void canvas_render_items_begin(const Color &p_modulate, Light *p_light, const Transform2D &p_base_transform);
virtual void canvas_render_items_end(); virtual void canvas_render_items_end();
virtual void canvas_render_items(Item *p_item_list, int p_z, const Color &p_modulate, Light *p_light, const Transform2D &p_base_transform); virtual void canvas_render_items(Item *p_item_list, int p_z, const Color &p_modulate, Light *p_light, const Transform2D &p_base_transform);
virtual void canvas_begin();
private: private:
// legacy codepath .. to remove after testing // legacy codepath .. to remove after testing
@ -263,6 +271,9 @@ private:
bool _light_scissor_begin(const Rect2 &p_item_rect, const Transform2D &p_light_xform, const Rect2 &p_light_rect) const; bool _light_scissor_begin(const Rect2 &p_item_rect, const Transform2D &p_light_xform, const Rect2 &p_light_rect) const;
void _calculate_scissor_threshold_area(); void _calculate_scissor_threshold_area();
// debug
void diagnose_batches(Item::Command *const *p_commands);
public: public:
void initialize(); void initialize();
RasterizerCanvasGLES2(); RasterizerCanvasGLES2();

1
servers/visual_server.cpp
View File

@ -2422,6 +2422,7 @@ VisualServer::VisualServer() {
GLOBAL_DEF("rendering/gles2/batching/light_scissor_area_threshold", 1.0f); GLOBAL_DEF("rendering/gles2/batching/light_scissor_area_threshold", 1.0f);
GLOBAL_DEF("rendering/gles2/batching/batch_buffer_size", 16384); GLOBAL_DEF("rendering/gles2/batching/batch_buffer_size", 16384);
GLOBAL_DEF("rendering/gles2/debug/flash_batching", false); GLOBAL_DEF("rendering/gles2/debug/flash_batching", false);
GLOBAL_DEF("rendering/gles2/debug/diagnose_frame", false);
GLOBAL_DEF_RST("rendering/gles2/debug/use_batching_in_editor", true); GLOBAL_DEF_RST("rendering/gles2/debug/use_batching_in_editor", true);
ProjectSettings::get_singleton()->set_custom_property_info("rendering/gles2/batching/max_join_item_commands", PropertyInfo(Variant::INT, "rendering/gles2/batching/max_join_item_commands", PROPERTY_HINT_RANGE, "0,65535")); ProjectSettings::get_singleton()->set_custom_property_info("rendering/gles2/batching/max_join_item_commands", PropertyInfo(Variant::INT, "rendering/gles2/batching/max_join_item_commands", PROPERTY_HINT_RANGE, "0,65535"));