Merge pull request #605 from TomHarte/DisplayMetrics

UNREADY! Introduces dynamic output quality selection.

OSBindings/Mac/Clock Signal.xcodeproj/project.pbxproj

@@ -215,6 +215,7 @@
 		4B595FAE2086DFBA0083CAA8 /* AudioToggle.cpp in Sources */ = {isa = PBXBuildFile; fileRef = 4B595FAC2086DFBA0083CAA8 /* AudioToggle.cpp */; };
 		4B5FADBA1DE3151600AEC565 /* FileHolder.cpp in Sources */ = {isa = PBXBuildFile; fileRef = 4B5FADB81DE3151600AEC565 /* FileHolder.cpp */; };
 		4B5FADC01DE3BF2B00AEC565 /* Microdisc.cpp in Sources */ = {isa = PBXBuildFile; fileRef = 4B5FADBE1DE3BF2B00AEC565 /* Microdisc.cpp */; };
+		4B622AE5222E0AD5008B59F2 /* DisplayMetrics.cpp in Sources */ = {isa = PBXBuildFile; fileRef = 4B622AE3222E0AD5008B59F2 /* DisplayMetrics.cpp */; };
 		4B643F3A1D77AD1900D431D6 /* CSStaticAnalyser.mm in Sources */ = {isa = PBXBuildFile; fileRef = 4B643F391D77AD1900D431D6 /* CSStaticAnalyser.mm */; };
 		4B643F3F1D77B88000D431D6 /* DocumentController.swift in Sources */ = {isa = PBXBuildFile; fileRef = 4B643F3E1D77B88000D431D6 /* DocumentController.swift */; };
 		4B69FB3D1C4D908A00B5F0AA /* Tape.cpp in Sources */ = {isa = PBXBuildFile; fileRef = 4B69FB3B1C4D908A00B5F0AA /* Tape.cpp */; };
@@ -897,6 +898,8 @@
 		4B5FADB91DE3151600AEC565 /* FileHolder.hpp */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.cpp.h; path = FileHolder.hpp; sourceTree = "<group>"; };
 		4B5FADBE1DE3BF2B00AEC565 /* Microdisc.cpp */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.cpp.cpp; name = Microdisc.cpp; path = Oric/Microdisc.cpp; sourceTree = "<group>"; };
 		4B5FADBF1DE3BF2B00AEC565 /* Microdisc.hpp */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.cpp.h; name = Microdisc.hpp; path = Oric/Microdisc.hpp; sourceTree = "<group>"; };
+		4B622AE3222E0AD5008B59F2 /* DisplayMetrics.cpp */ = {isa = PBXFileReference; lastKnownFileType = sourcecode.cpp.cpp; name = DisplayMetrics.cpp; path = ../../Outputs/DisplayMetrics.cpp; sourceTree = "<group>"; };
+		4B622AE4222E0AD5008B59F2 /* DisplayMetrics.hpp */ = {isa = PBXFileReference; lastKnownFileType = sourcecode.cpp.h; name = DisplayMetrics.hpp; path = ../../Outputs/DisplayMetrics.hpp; sourceTree = "<group>"; };
 		4B643F381D77AD1900D431D6 /* CSStaticAnalyser.h */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.c.h; name = CSStaticAnalyser.h; path = StaticAnalyser/CSStaticAnalyser.h; sourceTree = "<group>"; };
 		4B643F391D77AD1900D431D6 /* CSStaticAnalyser.mm */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.cpp.objcpp; name = CSStaticAnalyser.mm; path = StaticAnalyser/CSStaticAnalyser.mm; sourceTree = "<group>"; };
 		4B643F3C1D77AE5C00D431D6 /* CSMachine+Target.h */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.c.h; path = "CSMachine+Target.h"; sourceTree = "<group>"; };
@@ -1780,7 +1783,9 @@
 		4B366DFD1B5C165F0026627B /* Outputs */ = {
 			isa = PBXGroup;
 			children = (
+				4B622AE3222E0AD5008B59F2 /* DisplayMetrics.cpp */,
 				4B05401D219D1618001BF69C /* ScanTarget.cpp */,
+				4B622AE4222E0AD5008B59F2 /* DisplayMetrics.hpp */,
 				4BD601A920D89F2A00CBCE57 /* Log.hpp */,
 				4BF52672218E752E00313227 /* ScanTarget.hpp */,
 				4B0CCC411C62D0B3001CAC5F /* CRT */,
@@ -3025,8 +3030,8 @@
 				4BD5D2672199148100DDF17D /* ScanTargetGLSLFragments.cpp */,
 				4BD191D9219113B80042E144 /* OpenGL.hpp */,
 				4BD191F32191180E0042E144 /* ScanTarget.hpp */,
-				4BD424DC2193B5340097291A /* Primitives */,
 				4B961408222760E0001A7BF2 /* Screenshot.hpp */,
+				4BD424DC2193B5340097291A /* Primitives */,
 			);
 			name = OpenGL;
 			path = ../../Outputs/OpenGL;
@@ -3895,6 +3900,7 @@
 				4BD3A30B1EE755C800B5B501 /* Video.cpp in Sources */,
 				4B5FADBA1DE3151600AEC565 /* FileHolder.cpp in Sources */,
 				4B643F3A1D77AD1900D431D6 /* CSStaticAnalyser.mm in Sources */,
+				4B622AE5222E0AD5008B59F2 /* DisplayMetrics.cpp in Sources */,
 				4B1497881EE4A1DA00CE2596 /* ZX80O81P.cpp in Sources */,
 				4B894520201967B4007DE474 /* StaticAnalyser.cpp in Sources */,
 				4B2B3A4B1F9B8FA70062DABF /* Typer.cpp in Sources */,

Outputs/DisplayMetrics.cpp

@@ -0,0 +1,88 @@
+//
+//  DisplayMetrics.cpp
+//  Clock Signal
+//
+//  Created by Thomas Harte on 04/03/2019.
+//  Copyright © 2019 Thomas Harte. All rights reserved.
+//
+
+#include "DisplayMetrics.hpp"
+
+#include <numeric>
+
+using namespace Outputs::Display;
+
+// MARK: Frame size estimation.
+
+void Metrics::announce_event(ScanTarget::Event event) {
+	switch(event) {
+		case ScanTarget::Event::EndHorizontalRetrace:
+			++lines_this_frame_;
+		break;
+		case ScanTarget::Event::BeginVerticalRetrace:
+			add_line_total(lines_this_frame_);
+		break;
+		case ScanTarget::Event::EndVerticalRetrace:
+			lines_this_frame_ = 0;
+		break;
+		default: break;
+	}
+}
+
+void Metrics::add_line_total(int total) {
+	line_total_history_[line_total_history_pointer_] = total;
+	line_total_history_pointer_ = (line_total_history_pointer_ + 1) % line_total_history_.size();
+}
+
+float Metrics::visible_lines_per_frame_estimate() {
+	// Just average the number of records contained in line_total_history_ to provide this estimate;
+	// that array should be an even number, to allow for potential interlaced sources.
+	return float(std::accumulate(line_total_history_.begin(), line_total_history_.end(), 0)) / float(line_total_history_.size());
+}
+
+int Metrics::current_line() {
+	return lines_this_frame_;
+}
+
+// MARK: GPU processing speed decisions.
+
+void Metrics::announce_did_resize() {
+	frames_missed_ = frames_hit_ = 0;
+}
+
+void Metrics::announce_draw_status(size_t lines, std::chrono::high_resolution_clock::duration duration, bool complete) {
+	if(!complete) {
+		++frames_missed_;
+	} else {
+		++frames_hit_;
+	}
+
+	// Don't allow the record of history to extend too far into the past.
+	if(frames_hit_ + frames_missed_ > 200) {
+		// Subtract from whichever wasn't just incremented, to ensure the
+		// most recent information is more important than the historic stuff.
+		if(!complete) {
+			--frames_hit_;
+		} else {
+			--frames_missed_;
+		}
+
+		// Rebalance if either thing has gone negative.
+		if(frames_hit_ < 0) {
+			frames_missed_ += frames_hit_;
+			frames_hit_ = 0;
+		}
+		if(frames_missed_ < 0) {
+			frames_hit_ += frames_missed_;
+			frames_missed_ = 0;
+		}
+	}
+}
+
+bool Metrics::should_lower_resolution() {
+	// If less than 100 frames are on record, return no opinion; otherwise
+	// suggest a lower resolution if more than 10 frames in the last 100-200
+	// took too long to produce.
+	if(frames_hit_ + frames_missed_ < 100) return false;
+	return frames_missed_ > 10;
+}

Outputs/DisplayMetrics.hpp

@@ -0,0 +1,57 @@
+//
+//  DisplayMetrics.hpp
+//  Clock Signal
+//
+//  Created by Thomas Harte on 04/03/2019.
+//  Copyright © 2019 Thomas Harte. All rights reserved.
+//
+
+#ifndef DisplayMetrics_hpp
+#define DisplayMetrics_hpp
+
+#include "ScanTarget.hpp"
+
+#include <array>
+#include <chrono>
+
+namespace Outputs {
+namespace Display {
+
+/*!
+	A class to derive various metrics about the input to a ScanTarget,
+	based purely on empirical observation. In particular it is intended
+	to allow for host-client frame synchronisation.
+*/
+class Metrics {
+	public:
+		/// Notifies Metrics of a beam event.
+		void announce_event(ScanTarget::Event event);
+
+		/// Notifies Metrics that the size of the output buffer has changed.
+		void announce_did_resize();
+		/// Provides Metrics with a new data point for output speed estimation.
+		void announce_draw_status(size_t lines, std::chrono::high_resolution_clock::duration duration, bool complete);
+
+		/// @returns @c true if Metrics thinks a lower output buffer resolution is desirable in the abstract; @c false otherwise.
+		bool should_lower_resolution();
+
+		/// @returns An estimate of the number of lines being produced per frame, excluding vertical sync.
+		float visible_lines_per_frame_estimate();
+
+		/// @returns The number of lines since vertical retrace ended.
+		int current_line();
+
+	private:
+		int lines_this_frame_ = 0;
+		std::array<int, 20> line_total_history_;
+		size_t line_total_history_pointer_ = 0;
+		void add_line_total(int);
+
+		int frames_hit_ = 0;
+		int frames_missed_ = 0;
+};
+
+}
+}
+
+#endif /* DisplayMetrics_hpp */

Outputs/OpenGL/Primitives/TextureTarget.cpp

@@ -33,7 +33,7 @@ TextureTarget::TextureTarget(GLsizei width, GLsizei height, GLenum texture_unit,
 	bind_texture();
 
 	// Set dimensions and set the user-supplied magnification filter.
-	test_gl(glTexImage2D, GL_TEXTURE_2D, 0, GL_RGBA, GLsizei(expanded_width_), GLsizei(expanded_height_), 0, GL_RGBA, GL_UNSIGNED_BYTE, nullptr);
+	test_gl(glTexImage2D, GL_TEXTURE_2D, 0, GL_RGB, GLsizei(expanded_width_), GLsizei(expanded_height_), 0, GL_RGB, GL_UNSIGNED_BYTE, nullptr);
 	test_gl(glTexParameteri, GL_TEXTURE_2D, GL_TEXTURE_MAG_FILTER, mag_filter);
 	test_gl(glTexParameteri, GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, GL_NEAREST);
 

Outputs/OpenGL/ScanTarget.cpp

@@ -13,6 +13,7 @@
 
 #include <cassert>
 #include <cstring>
+#include <limits>
 
 using namespace Outputs::Display::OpenGL;
 
@@ -254,6 +255,9 @@ void ScanTarget::submit() {
 }
 
 void ScanTarget::announce(Event event, bool is_visible, const Outputs::Display::ScanTarget::Scan::EndPoint &location, uint8_t composite_amplitude) {
+	// Forward the event to the display metrics tracker.
+	display_metrics_.announce_event(event);
+
 	if(event == ScanTarget::Event::EndVerticalRetrace) {
 		// The previous-frame-is-complete flag is subject to a two-slot queue because
 		// measurement for *this* frame needs to begin now, meaning that the previous
@@ -383,14 +387,26 @@ void ScanTarget::setup_pipeline() {
 	input_shader_->set_uniform("textureName", GLint(SourceDataTextureUnit - GL_TEXTURE0));
 }
 
+Outputs::Display::Metrics &ScanTarget::display_metrics() {
+	return display_metrics_;
+}
+
 void ScanTarget::update(int output_width, int output_height) {
 	if(fence_ != nullptr) {
 		// if the GPU is still busy, don't wait; we'll catch it next time
 		if(glClientWaitSync(fence_, GL_SYNC_FLUSH_COMMANDS_BIT, 0) == GL_TIMEOUT_EXPIRED) {
+			display_metrics_.announce_draw_status(
+				lines_submitted_,
+				std::chrono::high_resolution_clock::now() - line_submission_begin_time_,
+				false);
 			return;
 		}
 		fence_ = nullptr;
 	}
+	display_metrics_.announce_draw_status(
+		lines_submitted_,
+		std::chrono::high_resolution_clock::now() - line_submission_begin_time_,
+		true);
 
 	// Spin until the is-drawing flag is reset; the wait sync above will deal
 	// with instances where waiting is inappropriate.
@@ -403,10 +419,16 @@ void ScanTarget::update(int output_width, int output_height) {
 		modals_are_dirty_ = false;
 	}
 
+	// Determine the start time of this submission group.
+	line_submission_begin_time_ = std::chrono::high_resolution_clock::now();
+
 	// Grab the current read and submit pointers.
 	const auto submit_pointers = submit_pointers_.load();
 	const auto read_pointers = read_pointers_.load();
 
+	// Determine how many lines are about to be submitted.
+	lines_submitted_ = (read_pointers.line + line_buffer_.size() - submit_pointers.line) % line_buffer_.size();
+
 	// Submit scans; only the new ones need to be communicated.
 	size_t new_scans = (submit_pointers.scan_buffer + scan_buffer_.size() - read_pointers.scan_buffer) % scan_buffer_.size();
 	if(new_scans) {
@@ -529,13 +551,26 @@ void ScanTarget::update(int output_width, int output_height) {
 		test_gl(glDrawArraysInstanced, GL_TRIANGLE_STRIP, 0, 4, GLsizei(new_scans));
 	}
 
-	// Ensure the accumulation buffer is properly sized.
-	// TODO: based on a decision about host speed, potentially switch to the std::min fragment as shown below,
-	// which would limit total output buffer size to 1440x1080.
-	const int framebuffer_height = output_height;//std::min(output_height, 1080);
+	// Logic for reducing resolution: start doing so if the metrics object reports that
+	// it's a good idea. Go up to a quarter of the requested resolution, subject to
+	// clamping at each stage. If the output resolution changes, or anything else about
+	// the output pipeline, just start trying the highest size again.
+	if(display_metrics_.should_lower_resolution()) {
+		resolution_reduction_level_ = std::min(resolution_reduction_level_+1, 4);
+	}
+	if(output_height_ != output_height || did_setup_pipeline) {
+		resolution_reduction_level_ = 1;
+		output_height_ = output_height;
+	}
+
+	// Ensure the accumulation buffer is properly sized, allowing for the metrics object's
+	// feelings about whether too high a resolution is being used.
+	const int framebuffer_height = std::max(output_height / resolution_reduction_level_, std::min(540, output_height));
 	const int proportional_width = (framebuffer_height * 4) / 3;
-	const bool did_create_accumulation_texture = !accumulation_texture_ || (	/* !synchronous && */ (accumulation_texture_->get_width() != proportional_width || accumulation_texture_->get_height() != framebuffer_height));
+	const bool did_create_accumulation_texture = !accumulation_texture_ || ( (accumulation_texture_->get_width() != proportional_width || accumulation_texture_->get_height() != framebuffer_height));
 	if(did_create_accumulation_texture) {
+		LOG("Changed output resolution to " << proportional_width << " by " << framebuffer_height);
+		display_metrics_.announce_did_resize();
 		std::unique_ptr<OpenGL::TextureTarget> new_framebuffer(
 			new TextureTarget(
 				GLsizei(proportional_width),
@@ -549,7 +584,7 @@ void ScanTarget::update(int output_width, int output_height) {
 
 			test_gl(glActiveTexture, AccumulationTextureUnit);
 			accumulation_texture_->bind_texture();
-			accumulation_texture_->draw(float(output_width) / float(output_height));
+			accumulation_texture_->draw(4.0f / 3.0f);
 
 			test_gl(glClear, GL_STENCIL_BUFFER_BIT);
 

Outputs/OpenGL/ScanTarget.hpp

@@ -9,8 +9,9 @@
 #ifndef ScanTarget_hpp
 #define ScanTarget_hpp
 
-#include "../ScanTarget.hpp"
 #include "../Log.hpp"
+#include "../DisplayMetrics.hpp"
+#include "../ScanTarget.hpp"
 
 #include "OpenGL.hpp"
 #include "Primitives/TextureTarget.hpp"
@@ -21,6 +22,7 @@
 #include <array>
 #include <atomic>
 #include <cstdint>
+#include <chrono>
 #include <list>
 #include <memory>
 #include <string>
@@ -47,6 +49,9 @@ class ScanTarget: public Outputs::Display::ScanTarget {
 		/*! Processes all the latest input, at a resolution suitable for later output to a framebuffer of the specified size. */
 		void update(int output_width, int output_height);
 
+		/*! @returns The DisplayMetrics object that this ScanTarget has been providing with announcements and draw overages. */
+		Metrics &display_metrics();
+
 	private:
 #ifndef NDEBUG
 		struct OpenGLVersionDumper {
@@ -78,6 +83,13 @@ class ScanTarget: public Outputs::Display::ScanTarget {
 
 		bool output_is_visible_ = false;
 
+		Metrics display_metrics_;
+		int resolution_reduction_level_ = 1;
+		int output_height_ = 0;
+
+		size_t lines_submitted_ = 0;
+		std::chrono::high_resolution_clock::time_point line_submission_begin_time_;
+
 		// Extends the definition of a Scan to include two extra fields,
 		// relevant to the way that this scan target processes video.
 		struct Scan {