diff options
author | Rémi Verschelde <rverschelde@gmail.com> | 2022-10-31 10:43:13 +0100 |
---|---|---|
committer | Rémi Verschelde <rverschelde@gmail.com> | 2022-10-31 10:43:13 +0100 |
commit | 5a00568afbb8423a9952d57afe71711be275f2e1 (patch) | |
tree | aeea61629b1da18023adfdda44a8c112b98f5def | |
parent | 40258bbbe89b0372700e056e8c8af07b0fd2179f (diff) | |
parent | aca964de4fcf0bd30ec1a0530c6fd08cbb53587b (diff) |
Merge pull request #67032 from clayjohn/GLES3-gpu-profiling
Add OpenGL timer queries to OpenGL3 backend
-rw-r--r-- | drivers/gles3/rasterizer_gles3.cpp | 10 | ||||
-rw-r--r-- | drivers/gles3/storage/utilities.cpp | 163 | ||||
-rw-r--r-- | drivers/gles3/storage/utilities.h | 85 |
3 files changed, 119 insertions, 139 deletions
diff --git a/drivers/gles3/rasterizer_gles3.cpp b/drivers/gles3/rasterizer_gles3.cpp index ea97dff522..b2d01b02fb 100644 --- a/drivers/gles3/rasterizer_gles3.cpp +++ b/drivers/gles3/rasterizer_gles3.cpp @@ -101,8 +101,7 @@ void RasterizerGLES3::begin_frame(double frame_step) { scene->set_time(time_total, frame_step); GLES3::Utilities *utils = GLES3::Utilities::get_singleton(); - utils->info.render_final = utils->info.render; - utils->info.render.reset(); + utils->_capture_timestamps_begin(); //scene->iteration(); } @@ -272,6 +271,13 @@ RasterizerGLES3::~RasterizerGLES3() { } void RasterizerGLES3::prepare_for_blitting_render_targets() { + // This is a hack, but this function is called one time after all viewports have been updated. + // So it marks the end of the frame for all viewports + // In the OpenGL renderer we have to call end_frame for each viewport so we can swap the + // buffers for each window before proceeding to the next. + // This allows us to only increment the frame after all viewports are done. + GLES3::Utilities *utils = GLES3::Utilities::get_singleton(); + utils->capture_timestamps_end(); } void RasterizerGLES3::_blit_render_target_to_screen(RID p_render_target, DisplayServer::WindowID p_screen, const Rect2 &p_screen_rect, uint32_t p_layer) { diff --git a/drivers/gles3/storage/utilities.cpp b/drivers/gles3/storage/utilities.cpp index 0a04f0d4d0..a2d1ba376a 100644 --- a/drivers/gles3/storage/utilities.cpp +++ b/drivers/gles3/storage/utilities.cpp @@ -38,16 +38,35 @@ #include "particles_storage.h" #include "texture_storage.h" +#include "servers/rendering/rendering_server_globals.h" + using namespace GLES3; Utilities *Utilities::singleton = nullptr; Utilities::Utilities() { singleton = this; + frame = 0; + for (int i = 0; i < FRAME_COUNT; i++) { + frames[i].index = 0; + glGenQueries(max_timestamp_query_elements, frames[i].queries); + + frames[i].timestamp_names.resize(max_timestamp_query_elements); + frames[i].timestamp_cpu_values.resize(max_timestamp_query_elements); + frames[i].timestamp_count = 0; + + frames[i].timestamp_result_names.resize(max_timestamp_query_elements); + frames[i].timestamp_cpu_result_values.resize(max_timestamp_query_elements); + frames[i].timestamp_result_values.resize(max_timestamp_query_elements); + frames[i].timestamp_result_count = 0; + } } Utilities::~Utilities() { singleton = nullptr; + for (int i = 0; i < FRAME_COUNT; i++) { + glDeleteQueries(max_timestamp_query_elements, frames[i].queries); + } } Vector<uint8_t> Utilities::buffer_get_data(GLenum p_target, GLuint p_buffer, uint32_t p_buffer_size) { @@ -213,87 +232,69 @@ void Utilities::visibility_notifier_call(RID p_notifier, bool p_enter, bool p_de /* TIMING */ -//void Utilities::render_info_begin_capture() { -// info.snap = info.render; -//} - -//void Utilities::render_info_end_capture() { -// info.snap.object_count = info.render.object_count - info.snap.object_count; -// info.snap.draw_call_count = info.render.draw_call_count - info.snap.draw_call_count; -// info.snap.material_switch_count = info.render.material_switch_count - info.snap.material_switch_count; -// info.snap.surface_switch_count = info.render.surface_switch_count - info.snap.surface_switch_count; -// info.snap.shader_rebind_count = info.render.shader_rebind_count - info.snap.shader_rebind_count; -// info.snap.vertices_count = info.render.vertices_count - info.snap.vertices_count; -// info.snap._2d_item_count = info.render._2d_item_count - info.snap._2d_item_count; -// info.snap._2d_draw_call_count = info.render._2d_draw_call_count - info.snap._2d_draw_call_count; -//} - -//int Utilities::get_captured_render_info(RS::RenderInfo p_info) { -// switch (p_info) { -// case RS::INFO_OBJECTS_IN_FRAME: { -// return info.snap.object_count; -// } break; -// case RS::INFO_VERTICES_IN_FRAME: { -// return info.snap.vertices_count; -// } break; -// case RS::INFO_MATERIAL_CHANGES_IN_FRAME: { -// return info.snap.material_switch_count; -// } break; -// case RS::INFO_SHADER_CHANGES_IN_FRAME: { -// return info.snap.shader_rebind_count; -// } break; -// case RS::INFO_SURFACE_CHANGES_IN_FRAME: { -// return info.snap.surface_switch_count; -// } break; -// case RS::INFO_DRAW_CALLS_IN_FRAME: { -// return info.snap.draw_call_count; -// } break; -// /* -// case RS::INFO_2D_ITEMS_IN_FRAME: { -// return info.snap._2d_item_count; -// } break; -// case RS::INFO_2D_DRAW_CALLS_IN_FRAME: { -// return info.snap._2d_draw_call_count; -// } break; -// */ -// default: { -// return get_render_info(p_info); -// } -// } -//} - -//int Utilities::get_render_info(RS::RenderInfo p_info) { -// switch (p_info) { -// case RS::INFO_OBJECTS_IN_FRAME: -// return info.render_final.object_count; -// case RS::INFO_VERTICES_IN_FRAME: -// return info.render_final.vertices_count; -// case RS::INFO_MATERIAL_CHANGES_IN_FRAME: -// return info.render_final.material_switch_count; -// case RS::INFO_SHADER_CHANGES_IN_FRAME: -// return info.render_final.shader_rebind_count; -// case RS::INFO_SURFACE_CHANGES_IN_FRAME: -// return info.render_final.surface_switch_count; -// case RS::INFO_DRAW_CALLS_IN_FRAME: -// return info.render_final.draw_call_count; -// /* -// case RS::INFO_2D_ITEMS_IN_FRAME: -// return info.render_final._2d_item_count; -// case RS::INFO_2D_DRAW_CALLS_IN_FRAME: -// return info.render_final._2d_draw_call_count; -//*/ -// case RS::INFO_USAGE_VIDEO_MEM_TOTAL: -// return 0; //no idea -// case RS::INFO_VIDEO_MEM_USED: -// return info.vertex_mem + info.texture_mem; -// case RS::INFO_TEXTURE_MEM_USED: -// return info.texture_mem; -// case RS::INFO_VERTEX_MEM_USED: -// return info.vertex_mem; -// default: -// return 0; //no idea either -// } -//} +void Utilities::capture_timestamps_begin() { + capture_timestamp("Frame Begin"); +} + +void Utilities::capture_timestamp(const String &p_name) { + ERR_FAIL_COND(frames[frame].timestamp_count >= max_timestamp_query_elements); + +#ifdef GLES_OVER_GL + glQueryCounter(frames[frame].queries[frames[frame].timestamp_count], GL_TIMESTAMP); +#endif + + frames[frame].timestamp_names[frames[frame].timestamp_count] = p_name; + frames[frame].timestamp_cpu_values[frames[frame].timestamp_count] = OS::get_singleton()->get_ticks_usec(); + frames[frame].timestamp_count++; +} + +void Utilities::_capture_timestamps_begin() { + // frame is incremented at the end of the frame so this gives us the queries for frame - 2. By then they should be ready. + if (frames[frame].timestamp_count) { +#ifdef GLES_OVER_GL + for (uint32_t i = 0; i < frames[frame].timestamp_count; i++) { + uint64_t temp = 0; + glGetQueryObjectui64v(frames[frame].queries[i], GL_QUERY_RESULT, &temp); + frames[frame].timestamp_result_values[i] = temp; + } +#endif + SWAP(frames[frame].timestamp_names, frames[frame].timestamp_result_names); + SWAP(frames[frame].timestamp_cpu_values, frames[frame].timestamp_cpu_result_values); + } + + frames[frame].timestamp_result_count = frames[frame].timestamp_count; + frames[frame].timestamp_count = 0; + frames[frame].index = Engine::get_singleton()->get_frames_drawn(); + capture_timestamp("Internal Begin"); +} + +void Utilities::capture_timestamps_end() { + capture_timestamp("Internal End"); + frame = (frame + 1) % FRAME_COUNT; +} + +uint32_t Utilities::get_captured_timestamps_count() const { + return frames[frame].timestamp_result_count; +} + +uint64_t Utilities::get_captured_timestamps_frame() const { + return frames[frame].index; +} + +uint64_t Utilities::get_captured_timestamp_gpu_time(uint32_t p_index) const { + ERR_FAIL_UNSIGNED_INDEX_V(p_index, frames[frame].timestamp_result_count, 0); + return frames[frame].timestamp_result_values[p_index]; +} + +uint64_t Utilities::get_captured_timestamp_cpu_time(uint32_t p_index) const { + ERR_FAIL_UNSIGNED_INDEX_V(p_index, frames[frame].timestamp_result_count, 0); + return frames[frame].timestamp_cpu_result_values[p_index]; +} + +String Utilities::get_captured_timestamp_name(uint32_t p_index) const { + ERR_FAIL_UNSIGNED_INDEX_V(p_index, frames[frame].timestamp_result_count, String()); + return frames[frame].timestamp_result_names[p_index]; +} /* MISC */ diff --git a/drivers/gles3/storage/utilities.h b/drivers/gles3/storage/utilities.h index 936ac35cc9..55a875958e 100644 --- a/drivers/gles3/storage/utilities.h +++ b/drivers/gles3/storage/utilities.h @@ -79,62 +79,35 @@ public: /* TIMING */ - struct Info { - uint64_t texture_mem = 0; - uint64_t vertex_mem = 0; - - struct Render { - uint32_t object_count; - uint32_t draw_call_count; - uint32_t material_switch_count; - uint32_t surface_switch_count; - uint32_t shader_rebind_count; - uint32_t vertices_count; - uint32_t _2d_item_count; - uint32_t _2d_draw_call_count; - - void reset() { - object_count = 0; - draw_call_count = 0; - material_switch_count = 0; - surface_switch_count = 0; - shader_rebind_count = 0; - vertices_count = 0; - _2d_item_count = 0; - _2d_draw_call_count = 0; - } - } render, render_final, snap; - - Info() { - render.reset(); - render_final.reset(); - } - - } info; - - virtual void capture_timestamps_begin() override {} - virtual void capture_timestamp(const String &p_name) override {} - virtual uint32_t get_captured_timestamps_count() const override { - return 0; - } - virtual uint64_t get_captured_timestamps_frame() const override { - return 0; - } - virtual uint64_t get_captured_timestamp_gpu_time(uint32_t p_index) const override { - return 0; - } - virtual uint64_t get_captured_timestamp_cpu_time(uint32_t p_index) const override { - return 0; - } - virtual String get_captured_timestamp_name(uint32_t p_index) const override { - return String(); - } - - // void render_info_begin_capture() override; - // void render_info_end_capture() override; - // int get_captured_render_info(RS::RenderInfo p_info) override; - - // int get_render_info(RS::RenderInfo p_info) override; +#define MAX_QUERIES 256 +#define FRAME_COUNT 3 + + struct Frame { + GLuint queries[MAX_QUERIES]; + TightLocalVector<String> timestamp_names; + TightLocalVector<uint64_t> timestamp_cpu_values; + uint32_t timestamp_count = 0; + TightLocalVector<String> timestamp_result_names; + TightLocalVector<uint64_t> timestamp_cpu_result_values; + TightLocalVector<uint64_t> timestamp_result_values; + uint32_t timestamp_result_count = 0; + uint64_t index = 0; + }; + + const uint32_t max_timestamp_query_elements = MAX_QUERIES; + + Frame frames[FRAME_COUNT]; // Frames for capturing timestamps. We use 3 so we don't need to wait for commands to complete + uint32_t frame = 0; + + virtual void capture_timestamps_begin() override; + virtual void capture_timestamp(const String &p_name) override; + virtual uint32_t get_captured_timestamps_count() const override; + virtual uint64_t get_captured_timestamps_frame() const override; + virtual uint64_t get_captured_timestamp_gpu_time(uint32_t p_index) const override; + virtual uint64_t get_captured_timestamp_cpu_time(uint32_t p_index) const override; + virtual String get_captured_timestamp_name(uint32_t p_index) const override; + void _capture_timestamps_begin(); + void capture_timestamps_end(); /* MISC */ |