Merge pull request #142 from TomHarte/BufferOverflow

Corrects a source of potential desynchronisation between the texture and geometry builders

Components/6560/6560.hpp

@@ -259,7 +259,7 @@ template <class T> class MOS6560 {
 				if(this_state_ != output_state_) {
 					switch(output_state_) {
 						case State::Sync:			crt_->output_sync(cycles_in_state_ * 4);														break;
-						case State::ColourBurst:	crt_->output_colour_burst(cycles_in_state_ * 4, (is_odd_frame_ || is_odd_line_) ? 128 : 0, 0);	break;
+						case State::ColourBurst:	crt_->output_colour_burst(cycles_in_state_ * 4, (is_odd_frame_ || is_odd_line_) ? 128 : 0);		break;
 						case State::Border:			output_border(cycles_in_state_ * 4);															break;
 						case State::Pixels:			crt_->output_data(cycles_in_state_ * 4, 1);														break;
 					}

Outputs/CRT/CRT.cpp

@@ -8,9 +8,10 @@
 
 #include "CRT.hpp"
 #include "CRTOpenGL.hpp"
-#include <stdarg.h>
-#include <math.h>
+#include <cstdarg>
+#include <cmath>
 #include <algorithm>
+#include <cassert>
 
 using namespace Outputs::CRT;
 
@@ -144,6 +145,7 @@ void CRT::advance_cycles(unsigned int number_of_cycles, bool hsync_requested, bo
 
 		if(next_run) {
 			// output_y and texture locations will be written later; we won't necessarily know what it is outside of the locked region
+			openGL_output_builder_.texture_builder.retain_latest();
 			source_output_position_x1() = (uint16_t)horizontal_flywheel_->get_current_output_position();
 			source_phase() = colour_burst_phase_;
 			source_amplitude() = colour_burst_amplitude_;
@@ -192,9 +194,10 @@ void CRT::advance_cycles(unsigned int number_of_cycles, bool hsync_requested, bo
 						output_x2() = (uint16_t)horizontal_flywheel_->get_current_output_position();
 					}
 					openGL_output_builder_.array_builder.flush(
-						[output_y, this] (uint8_t *input_buffer, size_t input_size, uint8_t *output_buffer, size_t output_size) {
+						[=] (uint8_t *input_buffer, size_t input_size, uint8_t *output_buffer, size_t output_size) {
 							openGL_output_builder_.texture_builder.flush(
-								[output_y, input_buffer] (const std::vector<TextureBuilder::WriteArea> &write_areas, size_t number_of_write_areas) {
+								[=] (const std::vector<TextureBuilder::WriteArea> &write_areas, size_t number_of_write_areas) {
+									assert(number_of_write_areas * SourceVertexSize == input_size);
 									for(size_t run = 0; run < number_of_write_areas; run++) {
 										*(uint16_t *)&input_buffer[run * SourceVertexSize + SourceVertexOffsetOfInputStart + 0] = write_areas[run].x;
 										*(uint16_t *)&input_buffer[run * SourceVertexSize + SourceVertexOffsetOfInputStart + 2] = write_areas[run].y;
@@ -311,6 +314,7 @@ void CRT::output_blank(unsigned int number_of_cycles) {
 }
 
 void CRT::output_level(unsigned int number_of_cycles) {
+	openGL_output_builder_.texture_builder.reduce_previous_allocation_to(1);
 	Scan scan {
 		.type = Scan::Type::Level,
 		.number_of_cycles = number_of_cycles,
@@ -329,13 +333,7 @@ void CRT::output_colour_burst(unsigned int number_of_cycles, uint8_t phase, uint
 }
 
 void CRT::output_default_colour_burst(unsigned int number_of_cycles) {
-	Scan scan {
-		.type = Scan::Type::ColourBurst,
-		.number_of_cycles = number_of_cycles,
-		.phase = (uint8_t)((phase_numerator_ * 256) / phase_denominator_ + (is_alernate_line_ ? 128 : 0)),
-		.amplitude = 32
-	};
-	output_scan(&scan);
+	output_colour_burst(number_of_cycles, (uint8_t)((phase_numerator_ * 256) / phase_denominator_ + (is_alernate_line_ ? 128 : 0)));
 }
 
 void CRT::output_data(unsigned int number_of_cycles, unsigned int source_divider) {

Outputs/CRT/CRT.hpp

@@ -187,7 +187,7 @@ class CRT {
 			@param amplitude The amplitude of the colour burst in 1/256ths of the amplitude of the
 			positive portion of the wave.
 		*/
-		void output_colour_burst(unsigned int number_of_cycles, uint8_t phase, uint8_t amplitude);
+		void output_colour_burst(unsigned int number_of_cycles, uint8_t phase, uint8_t amplitude = 102);
 
 		/*! Outputs a colour burst exactly in phase with CRT expectations using the idiomatic amplitude.
 

Outputs/CRT/Internals/TextureBuilder.cpp

@@ -39,7 +39,6 @@ TextureBuilder::TextureBuilder(size_t bytes_per_pixel, GLenum texture_unit) :
 		write_areas_start_y_(0),
 		is_full_(false),
 		did_submit_(false),
-		has_write_area_(false),
 		number_of_write_areas_(0) {
 	image_.resize(bytes_per_pixel * InputBufferBuilderWidth * InputBufferBuilderHeight);
 	glGenTextures(1, &texture_name_);
@@ -74,7 +73,6 @@ uint8_t *TextureBuilder::allocate_write_area(size_t required_length) {
 		starting_y = write_areas_[number_of_write_areas_ - 1].y;
 	}
 
-	WriteArea next_write_area;
 	if(starting_x + required_length + 2 > InputBufferBuilderWidth) {
 		starting_x = 0;
 		starting_y++;
@@ -85,33 +83,34 @@ uint8_t *TextureBuilder::allocate_write_area(size_t required_length) {
 		}
 	}
 
-	next_write_area.x = starting_x + 1;
-	next_write_area.y = starting_y;
-	next_write_area.length = (uint16_t)required_length;
-	if(number_of_write_areas_ < write_areas_.size())
-		write_areas_[number_of_write_areas_] = next_write_area;
-	else
-		write_areas_.push_back(next_write_area);
-	number_of_write_areas_++;
-	has_write_area_ = true;
+	write_area_.x = starting_x + 1;
+	write_area_.y = starting_y;
+	write_area_.length = (uint16_t)required_length;
 
-	return pointer_to_location(next_write_area.x, next_write_area.y);
+	return pointer_to_location(write_area_.x, write_area_.y);
 }
 
 bool TextureBuilder::is_full() {
 	return is_full_;
 }
 
-void TextureBuilder::reduce_previous_allocation_to(size_t actual_length) {
-	if(is_full_ || !has_write_area_) return;
+void TextureBuilder::retain_latest() {
+	if(is_full_) return;
+	if(number_of_write_areas_ < write_areas_.size())
+		write_areas_[number_of_write_areas_] = write_area_;
+	else
+		write_areas_.push_back(write_area_);
+	number_of_write_areas_++;
+}
 
-	has_write_area_ = false;
-	WriteArea &write_area = write_areas_[number_of_write_areas_-1];
-	write_area.length = (uint16_t)actual_length;
+void TextureBuilder::reduce_previous_allocation_to(size_t actual_length) {
+	if(is_full_) return;
+
+	write_area_.length = (uint16_t)actual_length;
 
 	// book end the allocation with duplicates of the first and last pixel, to protect
 	// against rounding errors when this run is drawn
-	uint8_t *start_pointer = pointer_to_location(write_area.x, write_area.y);
+	uint8_t *start_pointer = pointer_to_location(write_area_.x, write_area_.y);
 	memcpy(	&start_pointer[-bytes_per_pixel_],
 			start_pointer,
 			bytes_per_pixel_);
@@ -172,6 +171,5 @@ void TextureBuilder::flush(const std::function<void(const std::vector<WriteArea>
 	}
 
 	did_submit_ = false;
-	has_write_area_ = false;
 	number_of_write_areas_ = 0;
 }

Outputs/CRT/Internals/TextureBuilder.hpp

@@ -24,6 +24,32 @@ namespace CRT {
 	Owns an OpenGL texture resource and provides mechanisms to fill it from bottom left to top right
 	with runs of data, ensuring each run is neighboured immediately to the left and right by copies of its
 	first and last pixels.
+
+	Intended usage:
+
+		(i)		allocate a write area with allocate_write_area, supplying a maximum size.
+		(ii)	call reduce_previous_allocation_to to announce the actual size written.
+		
+	This will cause you to have added source data to the target texture. You can then either use that data
+	or allow it to expire.
+	
+		(iii)	call retain_latest to add the most recently written write area to the flush queue.
+
+	The flush queue contains provisional data, that can sit in the CPU's memory space indefinitely. This facility
+	is provided because it is expected that a texture will be built alontside some other collection of data —
+	that data in the flush queue is expected to become useful in coordination with something else but should
+	be retained at least until then.
+	
+		(iv)	call flush to move data to the submit queue.
+
+	When you flush, you'll receive a record of the bounds of all newly-flushed areas of source data. That gives
+	an opportunity to correlate the data with whatever else it is being tied to. It will continue to sit in
+	the CPU's memory space but has now passed beyond any further modification or reporting.
+
+		(v)		call submit to move data to the GPU and free up its CPU-side resources.
+
+	The data is now on the GPU, for whatever use the caller desires.
+
 */
 class TextureBuilder {
 	public:
@@ -41,6 +67,9 @@ class TextureBuilder {
 		/// and indicates that its actual final size was @c actual_length.
 		void reduce_previous_allocation_to(size_t actual_length);
 
+		/// Allocated runs are provisional; they will not appear in the next flush queue unless retained.
+		void retain_latest();
+
 		/// @returns @c true if all future calls to @c allocate_write_area will fail on account of the input texture
 		/// being full; @c false if calls may succeed.
 		bool is_full();
@@ -64,12 +93,14 @@ class TextureBuilder {
 		std::vector<uint8_t> image_;
 		GLuint texture_name_;
 
-		// the current list of write areas
+		// the current write area
+		WriteArea write_area_;
+
+		// the list of write areas that have ascended to the flush queue
 		std::vector<WriteArea> write_areas_;
 		size_t number_of_write_areas_;
 		bool is_full_;
 		bool did_submit_;
-		bool has_write_area_;
 		inline uint8_t *pointer_to_location(uint16_t x, uint16_t y);
 
 		// Usually: the start position for the current batch of write areas.