r_main.cpp

// SimpleGraphic Engine
// (c) David Gowor, 2014
//
// Module: Render Main
//

#define GLAD_GLES2_IMPLEMENTATION
#define IMGUI_DEFINE_MATH_OPERATORS
#include "r_local.h"

#include "common/base64.h"

#include <algorithm>
#include <array>
#include <filesystem>
#include <fmt/chrono.h>
#include <future>
#include <map>
#include <numeric>
#include <random>
#include <sstream>
#include <vector>

#include <imgui_impl_glfw.h>
#include <imgui_impl_opengl3.h>
#include <imgui_stdlib.h>

static uint64_t MurmurHash64A(void const* data, int len, uint64_t seed);

// =======
// Classes
// =======

enum r_takeScreenshot_e {
	R_SSNONE,
	R_SSTGA,
	R_SSJPEG,
	R_SSPNG
};

// ============
// Shader Class
// ============

class r_shader_c {
public:
	r_renderer_c* renderer;
	std::string name;
	dword		nameHash;
	int			refCount;
	r_tex_c* tex;

	r_shader_c(r_renderer_c* renderer, std::string_view shname, int flags);
	r_shader_c(r_renderer_c* renderer, std::string_view shname, int flags, std::unique_ptr<image_c> img);
	~r_shader_c();
};

r_shader_c::r_shader_c(r_renderer_c* renderer, std::string_view shname, int flags)
	: renderer(renderer)
{
	name = shname;
	nameHash = StringHash(name.c_str(), 0xFFFF);
	refCount = 0;
	tex = new r_tex_c(renderer->texMan, name, flags);
	if (tex->error) {
		renderer->sys->con->Warning("couldn't load texture '%s'", name.c_str());
	}
}

r_shader_c::r_shader_c(r_renderer_c* renderer, std::string_view shname, int flags, std::unique_ptr<image_c> img)
	: renderer(renderer)
{
	name = shname;
	nameHash = StringHash(name.c_str(), 0xFFFF);
	refCount = 0;
	tex = new r_tex_c(renderer->texMan, std::move(img), flags);
}

r_shader_c::~r_shader_c()
{
	delete tex;
}

// ===================
// Shader Handle Class
// ===================

r_shaderHnd_c::r_shaderHnd_c(r_shader_c* sh)
	: sh(sh)
{
	sh->refCount++;
}

r_shaderHnd_c::~r_shaderHnd_c()
{
	sh->refCount--;
	if (sh->refCount == 0) {
		sh->tex->AbortLoad();
	}
}

struct Mat4 {
	float m[16];

	float const* data() const {
		return m;
	}
};

Mat4 OrthoMatrix(double left, double right, double bottom, double top, double nearVal, double farVal)
{
	Mat4 ret;
	std::fill_n(ret.m, 16, 0.0f);
	ret.m[0] = (float)(2.0f / (right - left));
	ret.m[5] = (float)(2.0f / (top - bottom));
	ret.m[10] = (float)(-2.0f / (farVal - nearVal));
	ret.m[12] = (float)-((right + left) / (right - left));
	ret.m[13] = (float)-((top + bottom) / (top - bottom));
	ret.m[14] = (float)-((farVal + nearVal) / (farVal - nearVal));
	ret.m[15] = 1.0f;
	return ret;
}

// =================
// Layer queue class
// =================

struct r_layerCmd_s {
	enum Command {
		VIEWPORT,
		BLEND,
		BIND,
		COLOR,
		QUAD,
	} cmd;
};

struct r_layerCmdViewport_s {
	r_layerCmd_s::Command cmd;
	r_viewport_s viewport;
};

struct r_layerCmdBlend_s {
	r_layerCmd_s::Command cmd;
	int blendMode;
};

struct r_layerCmdBind_s {
	r_layerCmd_s::Command cmd;
	r_tex_c* tex;
};

struct r_layerCmdColor_s {
	r_layerCmd_s::Command cmd;
	col4_t col;
};

struct r_layerCmdQuad_s {
	r_layerCmd_s::Command cmd;
	struct {
		float s[4];
		float t[4];
		float x[4];
		float y[4];
		int stackLayer, maskLayer;
	} quad;
};

r_layer_c::r_layer_c(r_renderer_c* renderer, int layer, int subLayer)
	: renderer(renderer), layer(layer), subLayer(subLayer)
{
	cmdStorage.resize(1ull << 23);
	cmdCursor = 0;
	numCmd = 0;
}

r_layer_c::~r_layer_c()
{
}

static size_t CommandSize(r_layerCmd_s::Command cmd, size_t extraSize = 0) {
	using Tag = r_layerCmd_s::Command;
	switch (cmd) {
	case Tag::VIEWPORT: return sizeof(r_layerCmdViewport_s);
	case Tag::BLEND: return sizeof(r_layerCmdBlend_s);
	case Tag::BIND: return sizeof(r_layerCmdBind_s);
	case Tag::COLOR: return sizeof(r_layerCmdColor_s);
	case Tag::QUAD: return sizeof(r_layerCmdQuad_s);
	default:
		abort();
	}
}

r_layer_c::CmdHandle r_layer_c::GetFirstCommand()
{
	CmdHandle ret{};
	ret.offset = 0;
	if (cmdCursor > 0) {
		ret.cmd = (r_layerCmd_s*)cmdStorage.data();
	}
	return ret;
}

bool r_layer_c::GetNextCommand(r_layer_c::CmdHandle& handle)
{
	if (handle.cmd == nullptr) {
		return false;
	}
	handle.offset += (uint32_t)CommandSize(handle.cmd->cmd);
	if (handle.offset >= cmdCursor) {
		handle.cmd = nullptr;
		return false;
	}
	handle.cmd = (r_layerCmd_s*)(cmdStorage.data() + handle.offset);
	return true;
}

r_layerCmd_s* r_layer_c::NewCommand(size_t size)
{
	size_t const cmdEnd = cmdCursor + size;
	if (cmdEnd >= cmdStorage.size()) {
		return nullptr;
	}
	auto *ret = (r_layerCmd_s*)(cmdStorage.data() + cmdCursor);
	cmdCursor = cmdEnd;
	++numCmd;
	return ret;
}

void r_layer_c::SetViewport(r_viewport_s* viewport)
{
	if (auto* cmd = (r_layerCmdViewport_s*)NewCommand(CommandSize(r_layerCmd_s::VIEWPORT))) {
		cmd->cmd = r_layerCmd_s::VIEWPORT;
		cmd->viewport.x = viewport->x;
		cmd->viewport.y = viewport->y;
		cmd->viewport.width = viewport->width;
		cmd->viewport.height = viewport->height;
	}
}

void r_layer_c::SetBlendMode(int mode)
{
	if (auto* cmd = (r_layerCmdBlend_s*)NewCommand(CommandSize(r_layerCmd_s::BLEND))) {
		cmd->cmd = r_layerCmd_s::BLEND;
		cmd->blendMode = mode;
	}
}

void r_layer_c::Bind(r_tex_c* tex)
{
	if (auto* cmd = (r_layerCmdBind_s*)NewCommand(CommandSize(r_layerCmd_s::BIND))) {
		cmd->cmd = r_layerCmd_s::BIND;
		cmd->tex = tex;
	}
}

void r_layer_c::Color(col4_t col)
{
	if (auto* cmd = (r_layerCmdColor_s*)NewCommand(CommandSize(r_layerCmd_s::COLOR))) {
		cmd->cmd = r_layerCmd_s::COLOR;
		Vector4Copy(col, cmd->col);
	}
}

void r_layer_c::Quad(float s0, float t0, float x0, float y0, float s1, float t1, float x1, float y1, float s2, float t2, float x2, float y2, float s3, float t3, float x3, float y3, int stackLayer, int maskLayer)
{
	if (auto* cmd = (r_layerCmdQuad_s*)NewCommand(CommandSize(r_layerCmd_s::QUAD))) {
		cmd->cmd = r_layerCmd_s::QUAD;
		cmd->quad.s[0] = s0; cmd->quad.s[1] = s1; cmd->quad.s[2] = s2; cmd->quad.s[3] = s3;
		cmd->quad.t[0] = t0; cmd->quad.t[1] = t1; cmd->quad.t[2] = t2; cmd->quad.t[3] = t3;
		cmd->quad.x[0] = x0; cmd->quad.x[1] = x1; cmd->quad.x[2] = x2; cmd->quad.x[3] = x3;
		cmd->quad.y[0] = y0; cmd->quad.y[1] = y1; cmd->quad.y[2] = y2; cmd->quad.y[3] = y3;
		cmd->quad.stackLayer = stackLayer;
		cmd->quad.maskLayer = maskLayer;
	}
}

// =================
// Geometric queries
// =================

struct r_aabb_s {
	float lo[2];
	float hi[2];
};

r_aabb_s AabbFromCmdQuad(decltype(r_layerCmdQuad_s::quad)& q, r_viewport_s& vp)
{
	r_aabb_s r{
		{+FLT_MAX, +FLT_MAX},
		{-FLT_MAX, -FLT_MAX},
	};
	for (size_t i = 0; i < 4; ++i) {
		r.lo[0] = (std::min)(r.lo[0], (float)q.x[i]);
		r.lo[1] = (std::min)(r.lo[1], (float)q.y[i]);
		r.hi[0] = (std::max)(r.hi[0], (float)q.x[i]);
		r.hi[1] = (std::max)(r.hi[1], (float)q.y[i]);
	}
	r.lo[0] += vp.x;
	r.lo[1] += vp.y;
	r.hi[0] += vp.x;
	r.hi[1] += vp.y;
	return r;
}

r_aabb_s AabbFromViewport(r_viewport_s& vp)
{
	r_aabb_s r{
		{(float)vp.x, (float)vp.y },
		{(float)(vp.x + vp.width), (float)(vp.y + vp.height) },
	};
	return r;
}

bool AabbAabbIntersects(r_aabb_s& a, r_aabb_s& b)
{
	// A.lo <= B.hi && A.hi >= B.lo
	return a.lo[0] <= b.hi[0] && a.lo[1] <= b.hi[1] && a.hi[0] >= b.lo[0] && a.hi[1] >= b.lo[1];
}

struct Vertex {
	float x, y;
	float u, v;
	float r, g, b, a;
	float viewX, viewY, viewW, viewH;
	float texId, stackIdx, maskIdx;
};

struct Batch {
	explicit Batch(GLuint prog);
	Batch(Batch&& rhs);
	Batch& operator = (Batch&& rhs);
	Batch(Batch const&) = delete;
	Batch& operator = (Batch const&) = delete;
	~Batch();

	GLuint prog;
	GLint xyAttr;
	GLint uvAttr;
	GLint tintAttr;
	GLint viewportAttr;
	GLint texIdAttr;

	std::vector<Vertex> vertices;

	void Execute(GLuint sharedVbo, size_t vertexBase);
};

Batch::Batch(GLuint prog)
	: prog(prog)
{
	xyAttr = glGetAttribLocation(prog, "a_vertex");
	uvAttr = glGetAttribLocation(prog, "a_texcoord");
	tintAttr = glGetAttribLocation(prog, "a_tint");
	viewportAttr = glGetAttribLocation(prog, "a_viewport");
	texIdAttr = glGetAttribLocation(prog, "a_texId");
}

Batch::Batch(Batch&& rhs)
	: prog(rhs.prog)
	, xyAttr(rhs.xyAttr)
	, uvAttr(rhs.uvAttr)
	, tintAttr(rhs.tintAttr)
	, viewportAttr(rhs.viewportAttr)
	, texIdAttr(rhs.texIdAttr)
	, vertices(std::move(rhs.vertices))
{
}

Batch& Batch::operator = (Batch&& rhs) {
	prog = rhs.prog;
	xyAttr = rhs.xyAttr;
	uvAttr = rhs.uvAttr;
	tintAttr = rhs.tintAttr;
	viewportAttr = rhs.viewportAttr;
	texIdAttr = rhs.texIdAttr;
	vertices = std::move(rhs.vertices);

	return *this;
}

Batch::~Batch() {}

void Batch::Execute(GLuint sharedVbo, size_t vertexBase)
{
	if (vertices.empty()) {
		return;
	}

	glBindBuffer(GL_ARRAY_BUFFER, sharedVbo);
	auto dataPtr = (uint8_t const*)vertices.data();
	auto dataOff = vertexBase * sizeof(Vertex);
	auto dataSize = vertices.size() * sizeof(Vertex);
	glBufferSubData(GL_ARRAY_BUFFER, dataOff, dataSize, dataPtr);
	glVertexAttribPointer(xyAttr, 2, GL_FLOAT, GL_FALSE, sizeof(Vertex), (void const*)offsetof(Vertex, x));
	glVertexAttribPointer(uvAttr, 2, GL_FLOAT, GL_FALSE, sizeof(Vertex), (void const*)offsetof(Vertex, u));
	glVertexAttribPointer(tintAttr, 4, GL_FLOAT, GL_FALSE, sizeof(Vertex), (void const*)offsetof(Vertex, r));
	glVertexAttribPointer(viewportAttr, 4, GL_FLOAT, GL_FALSE, sizeof(Vertex), (void const*)offsetof(Vertex, viewX));
	glVertexAttribPointer(texIdAttr, 3, GL_FLOAT, GL_FALSE, sizeof(Vertex), (void const*)offsetof(Vertex, texId));
	glEnableVertexAttribArray(xyAttr);
	glEnableVertexAttribArray(uvAttr);
	glEnableVertexAttribArray(tintAttr);
	glEnableVertexAttribArray(viewportAttr);
	glEnableVertexAttribArray(texIdAttr);
	glDrawArrays(GL_TRIANGLES, 0, (GLsizei)vertices.size());
	glDisableVertexAttribArray(xyAttr);
	glDisableVertexAttribArray(uvAttr);
	glDisableVertexAttribArray(tintAttr);
	glDisableVertexAttribArray(viewportAttr);
	glDisableVertexAttribArray(texIdAttr);
	glBindBuffer(GL_ARRAY_BUFFER, 0);
	vertices.clear();
}

struct RenderStrategy {
	virtual ~RenderStrategy() = default;

	virtual void ProcessCommand(r_layerCmd_s* cmd) = 0;
	virtual void Flush() = 0;
	virtual void SetShowStats(bool showStats) { showStats_ = showStats; }

protected:
	bool showStats_{};
};

static std::map<r_blendMode_e, char const*> const s_blendModeString{
	{RB_ALPHA, "RB_ALPHA"},
	{RB_PRE_ALPHA, "RB_PRE_ALPHA"},
	{RB_ADDITIVE, "RB_ADDITIVE"},
};

struct AdjacentMergeStrategy : RenderStrategy {
	AdjacentMergeStrategy(r_layer_c* layer, r_renderer_c* renderer, GLuint prog)
		: layer_(layer), renderer_(renderer), prog_(prog), batch_(prog)
	{
		for (size_t i = 0;; ++i) {
			GLint loc = glGetUniformLocation(prog, fmt::format("s_tex[{}]", i).c_str());
			if (loc == -1) {
				break;
			}
			texLocs_.push_back(loc);
		}
		mvpMatrixLoc_ = glGetUniformLocation(prog_, "mvp_matrix");
		batchTextureCap_ = texLocs_.size();
		glGenBuffers(1, &vbo_);
	}

	~AdjacentMergeStrategy() {
		glDeleteBuffers(1, &vbo_);
	}

	struct BatchKey {
		int blendMode = -1;

		bool operator < (BatchKey const& rhs) const {
			return blendMode < rhs.blendMode;
		}

		bool operator == (BatchKey const& rhs) const {
			return !(*this < rhs) && !(rhs < *this);
		}

		bool operator != (BatchKey const& rhs) const {
			return !(*this == rhs);
		}
	};

	void ProcessCommand(r_layerCmd_s* cmd) override {
		switch (cmd->cmd) {
		case r_layerCmd_s::VIEWPORT: {
			auto* c = (r_layerCmdViewport_s*)cmd;
			nextViewport_ = c->viewport;
			if (showStats_) {
				// ImGui::Text("VIEWPORT: %dx%d @ %d,%d", c->viewport.width, c->viewport.height, c->viewport.x, c->viewport.y);
			}
		} break;
		case r_layerCmd_s::BLEND: {
			auto* c = (r_layerCmdBlend_s*)cmd;
			latchKey_.blendMode = c->blendMode;
			if (showStats_) {
				// ImGui::Text("BLEND: %s", s_blendModeString.at((r_blendMode_e)c->blendMode));
			}
		} break;
		case r_layerCmd_s::BIND: {
			auto* c = (r_layerCmdBind_s*)cmd;
			nextTex_ = c->tex;
			if (showStats_) {
				// ImGui::Text("TEX: %s", c->tex->fileName.c_str());
			}
		} break;
		case r_layerCmd_s::COLOR: {
			auto* c = (r_layerCmdColor_s*)cmd;
			std::copy_n(c->col, 4, tint_.data());
		} break;
		case r_layerCmd_s::QUAD: {
			auto* c = (r_layerCmdQuad_s*)cmd;
			if (showStats_) {
				// ImGui::Text("QUAD");
			}

			// Cull the quad first before it influences any boundary cuts.
			if (!!renderer_->r_drawCull->intVal) {
				auto a = AabbFromCmdQuad(c->quad, nextViewport_);
				auto b = AabbFromViewport(nextViewport_);
				bool intersects = AabbAabbIntersects(a, b);
				if (!intersects) {
					break;
				}
			}

			// If the current batch is incompatible key-wise, dispatch it to get a fresh
			// batch to grow in.
			if (!batch_.batch.vertices.empty() && batch_.key != latchKey_) {
				Dispatch();
			}
			batch_.key = latchKey_;

			// Check current (and only) batch if the texture set has the latched texture.
			// If it's there, use its index as vertex attribute.
			// If it's not, insert it if room, otherwise dispatch batch and prepare a fresh one.
			size_t texSlot{};
			{
				auto& textures = batch_.textures;
				auto texI = std::find(textures.begin(), textures.end(), nextTex_);
				if (texI == textures.end()) {
					if (textures.size() == batchTextureCap_) {
						Dispatch();
					}
					texI = textures.insert(textures.end(), nextTex_);
				}
				texSlot = std::distance(textures.begin(), texI);
			}

			Vertex quad[4]{};
			for (int v = 0; v < 4; v++) {
				auto& q = quad[v];
				auto& vp = nextViewport_;
				q.u = c->quad.s[v];
				q.v = c->quad.t[v];
				q.x = c->quad.x[v];
				q.y = c->quad.y[v];
				q.r = tint_[0];
				q.g = tint_[1];
				q.b = tint_[2];
				q.a = tint_[3];
				q.viewX = (float)vp.x;
				q.viewY = (float)vp.y;
				q.viewW = (float)vp.width;
				q.viewH = (float)vp.height;
				q.texId = (float)texSlot;
				q.stackIdx = (float)c->quad.stackLayer;
				q.maskIdx = (float)c->quad.maskLayer;
			}
			// 3-2
			// |/|
			// 0-1
			size_t indices[] = { 0, 1, 2, 0, 2, 3 };
			for (auto idx : indices) {
				batch_.batch.vertices.push_back(quad[idx]);
			}
			totalVertexCount_ += std::size(indices);
		} break;
		}
	}

	void Flush() {
		if (!batch_.batch.vertices.empty()) {
			Dispatch();
		}
		if (showStats_) {
			ImGui::BulletText("Layer %d:%d - %d batches", layer_->layer, layer_->subLayer, batchIndex);
		}
	}

private:
	void Dispatch() {
		glBindBuffer(GL_ARRAY_BUFFER, vbo_);
		auto& batch = batch_.batch;
		auto& textures = batch_.textures;
		size_t vertexCount = batch.vertices.size();
		glBufferData(GL_ARRAY_BUFFER, vertexCount * sizeof(Vertex), nullptr, GL_STREAM_DRAW);
		glUseProgram(prog_);

		auto& key = batch_.key;
		auto& lastKey = lastDispatchKey_;

		if (showStats_) {
			ImGui::Text("Batch %d", batchIndex);
			ImGui::Text("%d verts", batch.vertices.size());
		}

		{
			auto& vid = renderer_->sys->video->vid;
			float fbScaleX = vid.fbSize[0] / (float)vid.size[0];
			float fbScaleY = vid.fbSize[1] / (float)vid.size[1];
			int virtualW = renderer_->VirtualScreenWidth();
			int virtualH = renderer_->VirtualScreenHeight();
			glViewport(0, 0, virtualW, virtualH);
			Mat4 mvpMatrix = OrthoMatrix(0, virtualW, virtualH, 0, -9999, 9999);
			glUniformMatrix4fv(mvpMatrixLoc_, 1, GL_FALSE, mvpMatrix.data());
		}
		if (!lastKey || lastKey->blendMode != key.blendMode) {
			if (showStats_) {
				ImGui::Text("New blend mode %s", s_blendModeString.at((r_blendMode_e)key.blendMode));
			}
			switch (key.blendMode) {
			case RB_ALPHA:
				glBlendFunc(GL_SRC_ALPHA, GL_ONE_MINUS_SRC_ALPHA);
				break;
			case RB_PRE_ALPHA:
				glBlendFunc(GL_ONE, GL_ONE_MINUS_SRC_ALPHA);
				break;
			case RB_ADDITIVE:
				glBlendFunc(GL_ONE, GL_ONE);
				break;
			}
		}
		{
			for (size_t i = 0, numTex = texLocs_.size(); i < numTex; ++i) {
				glUniform1i(texLocs_[i], (GLint)i);
				glActiveTexture((GLenum)(GL_TEXTURE0 + i));
				if (i < textures.size()) {
					auto tex = textures[i];
					tex->Bind();
					if (showStats_) {
						ImGui::Text("New tex %d (%s)", tex->texId, tex->fileName.c_str());
					}
				}
				else {
					glBindTexture(GL_TEXTURE_2D_ARRAY, 0);
				}
			}
			glActiveTexture(GL_TEXTURE0);
		}

		batch.Execute(vbo_, 0);

		lastDispatchKey_ = key;
		batch_.batch.vertices.clear();
		batch_.textures.clear();

		glUseProgram(0);

		batchIndex += 1;
	}

	r_layer_c* layer_{};
	r_renderer_c* renderer_{};
	GLuint prog_{};
	std::vector<GLint> texLocs_;
	GLint mvpMatrixLoc_{};

	size_t batchTextureCap_{};
	GLuint vbo_{};

	struct TexturedBatch {
		explicit TexturedBatch(GLuint prog) : batch(prog) {
			textures.reserve(1ull << 20);
		}

		BatchKey key{};
		Batch batch;
		std::vector<r_tex_c*> textures;
	};

	BatchKey latchKey_{};
	r_viewport_s nextViewport_{};
	r_tex_c* nextTex_{};
	std::optional<BatchKey> lastDispatchKey_;
	TexturedBatch batch_;

	std::array<float, 4> tint_{ 1.0f, 1.0f, 1.0f, 1.0f };

	size_t totalVertexCount_ = 0;
	size_t batchIndex = 0;
};

void r_layer_c::Render()
{
	int const optLevel = renderer->r_layerOptimize->intVal;
	bool const shuffle = renderer->r_layerShuffle->intVal == 1;

	std::unique_ptr<RenderStrategy> strat(new AdjacentMergeStrategy(this, renderer, renderer->tintedTextureProgram));

	if (renderer->glPushGroupMarkerEXT)
	{
		std::ostringstream oss;
		oss << "Layer " << layer << ", sub-layer " << subLayer;
		renderer->glPushGroupMarkerEXT(0, oss.str().c_str());
	}

	if (strat) {
		bool showStats{};
		if (renderer->debugLayers) {
			if (ImGui::Begin("Layers", &renderer->debugLayers)) {
				std::string heading = fmt::format("Layer {}:{}", layer, subLayer);
				showStats = ImGui::CollapsingHeader(heading.c_str(), ImGuiTreeNodeFlags_DefaultOpen);
			}
		}
		strat->SetShowStats(showStats);

		for (CmdHandle cmdH = GetFirstCommand(); cmdH.cmd != nullptr; GetNextCommand(cmdH)) {
			strat->ProcessCommand(cmdH.cmd);
		}

		strat->Flush();

		if (renderer->debugLayers) {
			ImGui::End();
		}
	}

	if (renderer->glPopGroupMarkerEXT) {
		renderer->glPopGroupMarkerEXT();
	}
}

void r_layer_c::Discard()
{
	cmdCursor = 0;
	numCmd = 0;
}

// =====================
// r_IRenderer Interface
// =====================

r_IRenderer* r_IRenderer::GetHandle(sys_IMain* sysHnd)
{
	return new r_renderer_c(sysHnd);
}

void r_IRenderer::FreeHandle(r_IRenderer* hnd)
{
	delete (r_renderer_c*)hnd;
}

r_renderer_c::r_renderer_c(sys_IMain* sysHnd)
	: conCmdHandler_c(sysHnd->con), sys(sysHnd)
{
	r_compress = sys->con->Cvar_Add("r_compress", CV_ARCHIVE, "0");
	r_screenshotFormat = sys->con->Cvar_Add("r_screenshotFormat", CV_ARCHIVE, "jpg");
	r_layerDebug = sys->con->Cvar_Add("r_layerDebug", CV_ARCHIVE, "0");
	r_layerOptimize = sys->con->Cvar_Add("r_layerOptimize", CV_ARCHIVE | CV_CLAMP, "1", 0, 1);
	r_layerShuffle = sys->con->Cvar_Add("r_layerShuffle", CV_ARCHIVE | CV_CLAMP, "0", 0, 1);
	r_elideFrames = sys->con->Cvar_Add("r_elideFrames", CV_ARCHIVE | CV_CLAMP, "1", 0, 1);
	r_drawCull = sys->con->Cvar_Add("r_drawCull", CV_ARCHIVE | CV_CLAMP, "1", 0, 1);

	Cmd_Add("screenshot", 0, "[<format>]", this, &r_renderer_c::C_Screenshot);
}

static bool GetShaderCompileSuccess(GLuint id)
{
	GLint success{};
	glGetShaderiv(id, GL_COMPILE_STATUS, &success);
	return success == GL_TRUE;
}

static std::string GetShaderInfoLog(GLuint id)
{
	GLint len{};
	glGetShaderiv(id, GL_INFO_LOG_LENGTH, &len);
	std::vector<char> msg(len);
	glGetShaderInfoLog(id, (GLsizei)msg.size(), &len, msg.data());
	return std::string(msg.data(), msg.data() + len);
}

static bool GetProgramLinkSuccess(GLuint id)
{
	GLint success{};
	glGetProgramiv(id, GL_LINK_STATUS, &success);
	return success == GL_TRUE;
}

static std::string GetProgramInfoLog(GLuint id)
{
	GLint len{};
	glGetProgramiv(id, GL_INFO_LOG_LENGTH, &len);
	std::vector<char> msg(len);
	glGetProgramInfoLog(id, (GLsizei)msg.size(), &len, msg.data());
	return std::string(msg.data(), msg.data() + len);
}

static char const* s_tintedTextureVertexSource = R"(#version 300 es

uniform mat4 mvp_matrix;

in vec2 a_vertex;
in vec2 a_texcoord;
in vec4 a_tint;
in vec4 a_viewport;
in vec3 a_texId;

out vec2 v_screenPos;
out vec2 v_texcoord;
out vec4 v_tint;
out vec4 v_viewport;
out vec3 v_texId;

void main(void)
{
	v_texcoord = a_texcoord;
	v_tint = a_tint;
	v_texId = a_texId;
	vec2 vp0 = a_viewport.xy + vec2(0.0, a_viewport.w);
	vec2 vp1 = a_viewport.xy + vec2(a_viewport.z, 0.0);
	v_viewport = vec4(
		(mvp_matrix * vec4(vp0, 0.0, 1.0)).xy,
		(mvp_matrix * vec4(vp1, 0.0, 1.0)).xy);
	vec4 pos = mvp_matrix * vec4(a_vertex + a_viewport.xy, 0.0, 1.0);
	v_screenPos = pos.xy;
	gl_Position = pos;
}
)";

static char const* s_tintedTextureFragmentTemplate = R"(#version 300 es
precision mediump float;

uniform highp sampler2DArray s_tex[{SG_TEXTURE_COUNT}];
uniform vec4 i_tint;

in vec2 v_screenPos;
in vec2 v_texcoord;
in vec4 v_tint;
in vec4 v_viewport; // x0, y0, x1, y1
in vec3 v_texId;

out vec4 f_fragColor;

void main(void)
{{
	float x = v_screenPos[0], y = v_screenPos[1];
	if (x < v_viewport[0] ||
	    y < v_viewport[1] ||
	    x >= v_viewport[2] ||
	    y >= v_viewport[3]) {{
		discard;
	}}
	vec4 color;
	{SG_TEXTURE_SWITCH}
	f_fragColor = color * v_tint;
}}
)";

std::string const s_scaleVsSource = R"(#version 300 es
in vec4 a_position;
in vec2 a_texcoord;

out vec2 v_texcoord;

void main(void) {
	gl_Position = a_position;
	v_texcoord = a_texcoord;
}
)";

std::string const s_scaleFsSource = R"(#version 300 es
precision mediump float;

uniform highp sampler2D s_tex;

in vec2 v_texcoord;

out vec4 f_fragColor;

void main(void) {
	vec3 color = texture(s_tex, v_texcoord).rgb;
	f_fragColor = vec4(color, 1.0);
}
)";

// =============
// Init/Shutdown
// =============

void r_renderer_c::Init(r_featureFlag_e features)
{
	sys->con->PrintFunc("Render Init");

	apiDpiAware = !!(features & F_DPI_AWARE);

	timer_c timer;
	timer.Start();

	// Initialise OpenGL
	openGL = sys_IOpenGL::GetHandle(sys);
	sys_glSet_s set;
	set.bColor = 32;
	set.bDepth = 24;
	set.bStencil = 0;
	set.vsync = true;
	if (openGL->Init(&set)) {
		sys->Error("OpenGL initialisation failed");
	}

	// Get strings
	st_vendor = (const char*)glGetString(GL_VENDOR);
	st_renderer = (const char*)glGetString(GL_RENDERER);
	st_ver = (const char*)glGetString(GL_VERSION);
	st_ext = (const char*)glGetString(GL_EXTENSIONS);

	glGetIntegerv(GL_MAX_TEXTURE_SIZE, (int*)&texMaxDim);
	sys->con->Printf("GL_MAX_TEXTURE_SIZE: %d\n", texMaxDim);

	// Set default state
	glClearColor(0.0, 0.0, 0.0, 1.0);
	glEnable(GL_TEXTURE_2D);
	glDisable(GL_DEPTH_TEST);
	glEnable(GL_BLEND);

	// Load extensions
	sys->con->Printf("Loading OpenGL extensions...\n");

	if (strstr(st_ext, "GL_EXT_texture_compression_s3tc")) {
		sys->con->Printf("using GL_EXT_texture_compression_s3tc\n");
		glCompressedTexImage2D = (PFNGLCOMPRESSEDTEXIMAGE2DPROC)openGL->GetProc("glCompressedTexImage2D");
	}
	else {
		sys->con->Printf("GL_EXT_texture_compression_s3tc not supported\n");
		glCompressedTexImage2D = NULL;
	}

	if (strstr(st_ext, "GL_EXT_texture_compression_bptc")) {
		sys->con->Printf("using GL_EXT_texture_compression_bptc\n");
		texBC7 = true;
	}
	else {
		sys->con->Printf("GL_EXT_texture_compression_bptc not supported\n");
		texBC7 = false;
	}

	if (strstr(st_ext, "GL_EXT_debug_marker")) {
		sys->con->Printf("using GL_EXT_debug_marker\n");
		glInsertEventMarkerEXT = (PFNGLINSERTEVENTMARKEREXTPROC)openGL->GetProc("glInsertEventMarkerEXT");
		glPushGroupMarkerEXT = (PFNGLPUSHGROUPMARKEREXTPROC)openGL->GetProc("glPushGroupMarkerEXT");
		glPopGroupMarkerEXT = (PFNGLPOPGROUPMARKEREXTPROC)openGL->GetProc("glPopGroupMarkerEXT");
	}
	else {
		sys->con->Printf("GL_EXT_debug_marker not supported\n");
		glInsertEventMarkerEXT = NULL;
		glPushGroupMarkerEXT = NULL;
		glPopGroupMarkerEXT = NULL;
	}

	texNonPOT = true;

	// Initialise texture manager
	texMan = r_ITexManager::GetHandle(this);

	// Initialise shader array
	numShader = 0;
	memset(shaderList, 0, sizeof(shaderList));

	GLint maxTextureImageUnits{};
	glGetIntegerv(GL_MAX_TEXTURE_IMAGE_UNITS, &maxTextureImageUnits);

	// Initialise vertex programs
	{
		GLint success = GL_FALSE;
		GLuint prog = glCreateProgram();
		GLuint vs = glCreateShader(GL_VERTEX_SHADER);
		glShaderSource(vs, 1, &s_tintedTextureVertexSource, nullptr);
		glCompileShader(vs);
		if (!GetShaderCompileSuccess(vs)) {
			std::string log = GetShaderInfoLog(vs);
			sys->Error("Failed to compile vertex shader:\n%s", log.c_str());
		}
		GLuint fs = glCreateShader(GL_FRAGMENT_SHADER);
		std::string textureSwitch;
		{
			fmt::memory_buffer buf;
			for (size_t i = 0; i < maxTextureImageUnits; ++i) {
				if (i == 0) {
					fmt::format_to(fmt::appender(buf), "if (v_texId.x < {}.5) ", i);
				}
				else if (i == maxTextureImageUnits - 1) {
					fmt::format_to(fmt::appender(buf), "else ");
				}
				else {
					fmt::format_to(fmt::appender(buf), "else if (v_texId.x < {}.5)", i);
				}
				fmt::format_to(fmt::appender(buf), R"( {{
	color = texture(s_tex[{}], vec3(v_texcoord, v_texId.y));
	if (v_texId.z > -0.5)
		color *= texture(s_tex[{}], vec3(v_texcoord, v_texId.z));
}}
)", i, i);
			}
			textureSwitch = to_string(buf);
		}
		std::string fragSource = fmt::format(s_tintedTextureFragmentTemplate,
			fmt::arg("SG_TEXTURE_COUNT", maxTextureImageUnits),
			fmt::arg("SG_TEXTURE_SWITCH", textureSwitch));
		char const* fragSourcePtr = fragSource.c_str();
		glShaderSource(fs, 1, &fragSourcePtr, nullptr);
		glCompileShader(fs);
		if (!GetShaderCompileSuccess(fs)) {
			std::string log = GetShaderInfoLog(fs);
			sys->Error("Failed to compile fragment shader:\n%s", log.c_str());
		}

		glAttachShader(prog, vs);
		glAttachShader(prog, fs);
		glLinkProgram(prog);
		if (!GetProgramLinkSuccess(prog)) {
			std::string log = GetProgramInfoLog(prog);
			sys->Error("Failed to link program:\n%s", log.c_str());
		}
		glDeleteShader(vs);
		glDeleteShader(fs);
		tintedTextureProgram = prog;
	}

	// Initialise layer array
	numLayer = 1;
	layerListSize = 16;
	layerList = new r_layer_c * [layerListSize];
	layerList[0] = new r_layer_c(this, 0, 0);

	// Initialise layer command bin
	layerCmdBinCount = 0;
	layerCmdBinSize = 1024;
	layerCmdBin = new r_layerCmd_s * [layerCmdBinSize];

	takeScreenshot = R_SSNONE;

	// Set up DPI-scaling render target
	for (int i = 0; i < 2; ++i) {
		auto& rtt = rttMain[i];
		if (i > 0) {
			rtt = rttMain[0]; // Reuse shared parts like dimensions and program/locations.
		}
		glGenFramebuffers(1, &rtt.framebuffer);
		glGenTextures(1, &rtt.colorTexture);
		
		if (i == 0) {
			auto compileShader = [](std::string_view src, GLenum type) -> GLuint {
				GLuint id = glCreateShader(type);
				auto sourcePtr = src.data();
				glShaderSource(id, 1, &sourcePtr, nullptr);
				glCompileShader(id);
				return id;
				};

			auto vsId = compileShader(s_scaleVsSource, GL_VERTEX_SHADER);
			if (!GetShaderCompileSuccess(vsId)) {
				auto log = GetShaderInfoLog(vsId);
				sys->con->Printf("Scaling VS compile failure: %s\n", log.c_str());
			}
			auto fsId = compileShader(s_scaleFsSource, GL_FRAGMENT_SHADER);
			if (!GetShaderCompileSuccess(fsId)) {
				auto log = GetShaderInfoLog(fsId);
				sys->con->Printf("Scaling FS compile failure: %s\n", log.c_str());
			}

			GLuint prog = rtt.blitProg = glCreateProgram();
			glAttachShader(prog, vsId);
			glAttachShader(prog, fsId);
			glLinkProgram(prog);
			if (!GetProgramLinkSuccess(prog)) {
				auto log = GetProgramInfoLog(prog);
				sys->con->Printf("Scaling program link failure: %s\n", log.c_str());
			}

			GLint linked = GL_FALSE;
			glGetProgramiv(prog, GL_LINK_STATUS, &linked);

			glDeleteShader(vsId);
			glDeleteShader(fsId);

			rtt.blitAttribLocPos = glGetAttribLocation(prog, "a_position");
			rtt.blitAttribLocTC = glGetAttribLocation(prog, "a_texcoord");
			rtt.blitSampleLocColour = glGetUniformLocation(prog, "s_tex");
		}
	}

	// Load render resources
	sys->con->Printf("Loading resources...\n");

	whiteImage = RegisterShader("@white", 0);
	blackImage = RegisterShader("@black", 0);

	imguiCtx = ImGui::CreateContext();
	ImGui::SetCurrentContext(imguiCtx);

	ImGui_ImplGlfw_InitForOpenGL((GLFWwindow*)sys->video->GetWindowHandle(), true);
	ImGui_ImplOpenGL3_Init("#version 100");

	fonts[F_FIXED] = new r_font_c(this, "Bitstream Vera Sans Mono");
	fonts[F_VAR] = new r_font_c(this, "Liberation Sans");
	fonts[F_VAR_BOLD] = new r_font_c(this, "Liberation Sans Bold");
	fonts[F_FONTIN_SC] = new r_font_c(this, "Fontin SmallCaps");
	fonts[F_FONTIN_SC_ITALIC] = new r_font_c(this, "Fontin SmallCaps Italic");
	fonts[F_FONTIN] = new r_font_c(this, "Fontin");
	fonts[F_FONTIN_ITALIC] = new r_font_c(this, "Fontin Italic");

	sys->con->Printf("Renderer initialised in %d msec.\n", timer.Get());
}

void r_renderer_c::Shutdown()
{
	sys->con->PrintFunc("Render Shutdown");

	sys->con->Printf("Unloading resources...\n");

	ImGui_ImplOpenGL3_Shutdown();
	ImGui_ImplGlfw_Shutdown();
	ImGui::DestroyContext(imguiCtx);

	delete whiteImage;

	for (int f = 0; f < F_NUMFONTS; f++) {
		delete fonts[f];
	}

	for (int s = 0; s < numShader; s++) {
		delete shaderList[s];
	}

	for (int l = 0; l < numLayer; l++) {
		delete layerList[l];
	}
	delete layerList;
	for (int c = 0; c < layerCmdBinCount; c++) {
		delete layerCmdBin[c];
	}
	delete layerCmdBin;

	for (int i = 0; i < 2; ++i) {
		auto& rtt = rttMain[i];
		glDeleteTextures(1, &rtt.colorTexture);
		glDeleteFramebuffers(1, &rtt.framebuffer);
	}
	glDeleteProgram(rttMain[0].blitProg);

	// Shutdown texture manager
	r_ITexManager::FreeHandle(texMan);

	// Shutdown OpenGL
	openGL->Shutdown();
	sys_IOpenGL::FreeHandle(openGL);

	sys->con->Printf("Renderer shutdown complete.\n");
}

// =================
// Render Management
// =================

void r_renderer_c::PumpShaders()
{
	texMan->ProcessPendingTextureUploads();
	for (size_t idx = 0; idx < numShader; ++idx)
		if (auto* sh = shaderList[idx])
			if (auto tex = sh->tex; tex && tex->status != r_tex_c::DONE) {
				inhibitElision = true;
				break;
			}
}

void r_renderer_c::BeginFrame()
{
	ImGui_ImplOpenGL3_NewFrame();
	ImGui_ImplGlfw_NewFrame();
	ImGui::NewFrame();
	{
		auto& vid = sys->video->vid;
		int wNew = VirtualScreenWidth();
		int hNew = VirtualScreenHeight();
		bool const wantIntegerScaling = fmodf(vid.dpiScale, 1.0f) < 0.0005f;
		for (int i = 0; i < 2; ++i) {
			auto& rtt = rttMain[i];
			if (rtt.width != wNew || rtt.height != hNew) {
				GLint prevTex2D, prevFB;
				glGetIntegerv(GL_TEXTURE_BINDING_2D, &prevTex2D);
				glGetIntegerv(GL_FRAMEBUFFER_BINDING, &prevFB);
				glBindTexture(GL_TEXTURE_2D, rtt.colorTexture);
				glTexImage2D(GL_TEXTURE_2D, 0, GL_RGB, wNew, hNew, 0, GL_RGB, GL_UNSIGNED_BYTE, nullptr);
				glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_S, GL_CLAMP_TO_EDGE);
				glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_T, GL_CLAMP_TO_EDGE);
				GLint const filterMode = wantIntegerScaling ? GL_NEAREST : GL_LINEAR;
				glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MAG_FILTER, filterMode);
				glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, filterMode);

				rtt.width = wNew;
				rtt.height = hNew;

				glBindFramebuffer(GL_FRAMEBUFFER, rtt.framebuffer);
				glFramebufferTexture2D(GL_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, GL_TEXTURE_2D, rtt.colorTexture, 0);

				glCheckFramebufferStatus(GL_FRAMEBUFFER);

				glBindFramebuffer(GL_FRAMEBUFFER, prevFB);
				glBindTexture(GL_TEXTURE_2D, prevTex2D);
			}
		}
	}

	curLayer = layerList[0];

	SetViewport();
	SetBlendMode(RB_ALPHA);
	DrawColor();

	beginFrameToc = std::chrono::steady_clock::now();
}

static int layerCompFunc(const void* va, const void* vb)
{
	r_layer_c* a = *(r_layer_c**)va;
	r_layer_c* b = *(r_layer_c**)vb;
	if (a->layer < b->layer) {
		return -1;
	}
	else if (a->layer > b->layer) {
		return 1;
	}
	else if (a->subLayer < b->subLayer) {
		return -1;
	}
	else {
		return 1;
	}
}

void CVarSliderInt(char const* label, conVar_c* cvar) {
	int curOpt = cvar->intVal;
	if (ImGui::SliderInt(label, &curOpt, cvar->min, cvar->max, "%d", ImGuiSliderFlags_AlwaysClamp | ImGuiSliderFlags_NoInput)) {
		if (curOpt != cvar->intVal) {
			cvar->Set(curOpt);
		}
	}
}

void CVarCheckbox(char const* label, conVar_c* cvar) {
	bool checked = cvar->intVal == 1;
	if (ImGui::Checkbox(label, &checked)) {
		cvar->intVal = +checked;
	}
}

static std::string BinaryUnitPrefix(uint64_t quantity) {
	if (quantity < 1ull<<10) {
		return fmt::format("{} ", quantity);
	}
	if (quantity < 1ull << 20) {
		return fmt::format("{:0.2f} Ki", quantity / 1024.0);
	}
	if (quantity < 1ull << 30) {
		return fmt::format("{:0.2f} Mi", quantity / 1024.0 / 1024.0);
	}
	if (quantity < 1ull << 40) {
		return fmt::format("{:0.2f} Gi", quantity / 1024.0 / 1024.0 / 1024.0);
	}
	if (quantity < 1ull << 50) {
		return fmt::format("{:0.2f} Ti", quantity / 1024.0 / 1024.0 / 1024.0 / 1024.0);
	}
	if (quantity < 1ull << 60) {
		return fmt::format("{:0.2f} Ti", quantity / 1024.0 / 1024.0 / 1024.0 / 1024.0 / 1024.0);
	}
	return fmt::format("{:0.2f} Pi", quantity / 1024.0 / 1024.0 / 1024.0 / 1024.0 / 1024.0);
}

void r_renderer_c::EndFrame()
{
	inhibitElision = false;
	PumpShaders();

	std::chrono::time_point endFrameTic = std::chrono::steady_clock::now();
	frameStats.AppendDuration(&FrameStats::midFrameStepDurations, endFrameTic - beginFrameToc);

	static bool showDemo = false;
	static bool showMetrics = false;
	static bool showHash = false;
	static bool showTiming = false;
	if (debugImGui) {
		if (ImGui::Begin("Debug Hub", &debugImGui)) {
			if (ImGui::Button("ImGui Demo")) {
				showDemo = true;
			}
			if (ImGui::Button("Metrics")) {
				showMetrics = true;
			}
			if (ImGui::Button("Layers")) {
				debugLayers = true;
			}
		}
		ImGui::End();
	}

	if (showDemo) {
		ImGui::ShowDemoWindow(&showDemo);
	}
	if (showMetrics) {
		ImGui::ShowMetricsWindow(&showMetrics);
	}

	r_layer_c** layerSort = new r_layer_c * [numLayer];
	for (int l = 0; l < numLayer; l++) {
		layerSort[l] = layerList[l];
	}
	qsort(layerSort, numLayer, sizeof(r_layer_c*), layerCompFunc);
	if (r_layerDebug->intVal) {
		size_t totalCmd = 0;
		for (int l = 0; l < numLayer; l++) {
			totalCmd += layerSort[l]->numCmd;
			char str[1024];
			sprintf(str, "%zu (%4d,%4d) [%2d]", layerSort[l]->numCmd, layerSort[l]->layer, layerSort[l]->subLayer, l);
			float w = (float)DrawStringWidth(16, F_FIXED, str);
			DrawColor(0x7F000000);
			DrawImage(NULL, { (float)VirtualScreenWidth() - w, VirtualScreenHeight() - (l + 2) * 16.0f }, { w, 16 });
			DrawStringFormat(0, VirtualScreenHeight() - (l + 2) * 16.0f, F_RIGHT, 16, colorWhite, F_FIXED, str);
		}
		char str[1024];
		sprintf(str, "%zu", totalCmd);
		float w = (float)DrawStringWidth(16, F_FIXED, str);
		DrawColor(0xAF000000);
		DrawImage(NULL, { (float)VirtualScreenWidth() - w, VirtualScreenHeight() - 16.0f }, { w, 16 });
		DrawStringFormat(0, VirtualScreenHeight() - 16.0f, F_RIGHT, 16, colorWhite, F_FIXED, str);
	}

	std::optional<std::pair<int, int>> layerBreak;
	if (debugLayers) {
		if (ImGui::Begin("Layers", &debugLayers)) {
			ImGui::Text("Layers: %d", numLayer);
			ImGui::Text("%d out of %d frames drawn.", drawnFrames, totalFrames);
			CVarSliderInt("Optimization", r_layerOptimize);
			CVarCheckbox("Elide identical frames", r_elideFrames);
			ImGui::BeginDisabled(true);
			ImGui::Checkbox("Elision inhibited", &inhibitElision);
			ImGui::EndDisabled();
			CVarCheckbox("Draw command culling", r_drawCull);

			size_t totalFootprint{}, totalDenseFootprint{};
			for (int l = 0; l < numLayer; ++l) {
				size_t byteAcc{};
				auto layer = layerSort[l];
				size_t const numCmd = layer->numCmd;
				totalFootprint += numCmd * sizeof(r_layerCmdQuad_s); // legacy footprint
				totalDenseFootprint += layer->cmdCursor;
			}

			ImGui::Text("Total payload footprint: %sB", BinaryUnitPrefix(totalFootprint).c_str());
			ImGui::Text("Total dense footprint: %sB", BinaryUnitPrefix(totalDenseFootprint).c_str());

			size_t totalCmd{};
			if (ImGui::BeginTable("Layer stats", 7, ImGuiTableFlags_Borders | ImGuiTableFlags_SizingFixedFit)) {
				ImGui::TableSetupColumn("Index");
				ImGui::TableSetupColumn("Layer");
				ImGui::TableSetupColumn("Sublayer");
				ImGui::TableSetupColumn("Command count");
				ImGui::TableSetupColumn("Dense");
				ImGui::TableSetupColumn("Debug");
				ImGui::TableHeadersRow();
				for (int l = 0; l < numLayer; ++l) {
					auto layer = layerSort[l];
					ImGui::PushID(layer->layer);
					ImGui::PushID(layer->subLayer);
					totalCmd += layer->numCmd;
					ImGui::TableNextRow();
					ImGui::TableNextColumn();
					ImGui::Text("%d", l);
					ImGui::TableNextColumn();
					ImGui::Text("%d", layer->layer);
					ImGui::TableNextColumn();
					ImGui::Text("%d", layer->subLayer);
					ImGui::TableNextColumn();
					ImGui::Text("%d", layer->numCmd);
					ImGui::TableNextColumn();
					ImGui::Text("%sB", BinaryUnitPrefix(layer->cmdCursor).c_str());
					ImGui::TableNextColumn();
					if (ImGui::Button("Debug")) {
						layerBreak = { layer->layer, layer->subLayer };
					}
					ImGui::PopID();
					ImGui::PopID();
				}
				ImGui::EndTable();
			}
		}
		ImGui::End();
	}

	if (inhibitElision || elideFrames != !!r_elideFrames->intVal) {
		elideFrames = !!r_elideFrames->intVal;
		lastFrameHash.clear();
	}

	std::future<std::optional<std::vector<uint8_t>>> elidedFrameHashFut;
	if (elideFrames) {
		elidedFrameHashFut = std::async([&]() -> std::optional<std::vector<uint8_t>> {
			std::vector<uint8_t> commandDigest;

			for (auto lIdx = 0; lIdx < numLayer; ++lIdx) {
				auto layer = layerSort[lIdx];
				uint64_t subHash = MurmurHash64A(layer->cmdStorage.data(), (int)layer->cmdCursor, 0ull);
				uint8_t const* p = (uint8_t const*)&subHash;
				commandDigest.insert(commandDigest.end(), p, p + sizeof(subHash));
			}

			return commandDigest;
		});
	}
	else {
		std::promise<std::optional<std::vector<uint8_t>>> p;
		elidedFrameHashFut = p.get_future();
		p.set_value({});
	}

	elidedFrameHashFut.wait();

	++totalFrames;
	bool decideDraw = false;
	bool elideDraw = false;
	{
		glBindFramebuffer(GL_FRAMEBUFFER, GetDrawRenderTarget().framebuffer);
		glClear(GL_DEPTH_BUFFER_BIT | GL_COLOR_BUFFER_BIT | GL_STENCIL_BUFFER_BIT);
		int l{};
		for (l = 0; l < numLayer; l++) {
			if (!decideDraw && elidedFrameHashFut.wait_for(std::chrono::milliseconds(0)) == std::future_status::ready) {
				decideDraw = true;
				auto commandDigest = elidedFrameHashFut.get();
				if (commandDigest) {
					if (*commandDigest == lastFrameHash) {
						elideDraw = true;
						break;
					}
					else {
						lastFrameHash = *commandDigest;
					}
				}
				else {
					lastFrameHash.clear();
				}
			}
			auto& layer = layerSort[l];
			if (layerBreak && layerBreak->first == layer->layer && layerBreak->second == layer->subLayer) {
#ifdef _WIN32
				DebugBreak();
#endif
			}
			layer->Render();
		}
		if (!elideDraw) {
			presentRtt = 1 - presentRtt;
			++drawnFrames;
		}
	}

	if (!decideDraw) {
		if (auto commandDigest = elidedFrameHashFut.get()) {
			lastFrameHash = *commandDigest;
		}
		else {
			lastFrameHash.clear();
		}
	}

	if (inhibitElision) {
		// If we explicitly inhibited elision due to things like incomplete textures, make sure that the next frame is drawn.
		lastFrameHash.clear();
	}

	for (int l = 0; l < numLayer; ++l) {
		layerSort[l]->Discard();
	}
	delete[] layerSort;

	{
		auto& rtt = GetPresentRenderTarget();
		glBindFramebuffer(GL_FRAMEBUFFER, 0);
		glClearColor(0.0f, 0.0f, 0.0f, 1.0f);
		glClear(GL_DEPTH_BUFFER_BIT | GL_COLOR_BUFFER_BIT | GL_STENCIL_BUFFER_BIT);

		float blitTriPos[] = {
			-1.0f, -1.0f, //
			3.0f, -1.0f, //
			-1.0f, 3.0f, //
		};
		float blitTriUV[] = {
			0.0f, 0.0f, //
			2.0f, 0.0f, //
			0.0f, 2.0f, //
		};

		glViewport(0, 0, sys->video->vid.fbSize[0], sys->video->vid.fbSize[1]);
		glUseProgram(rtt.blitProg);
		glVertexAttribPointer(0, 2, GL_FLOAT, GL_FALSE, 0, std::data(blitTriPos));
		glVertexAttribPointer(1, 2, GL_FLOAT, GL_FALSE, 0, std::data(blitTriUV));
		glEnableVertexAttribArray(0);
		glEnableVertexAttribArray(1);
		glBindTexture(GL_TEXTURE_2D, rtt.colorTexture);
		glUniform1i(rtt.blitSampleLocColour, 0);
		glDrawArrays(GL_TRIANGLES, 0, 3);
		glBindTexture(GL_TEXTURE_2D, 0);
		glUseProgram(0);
	}

	if (showHash) {
		if (ImGui::Begin("Hash")) {
			char* b64{};
			size_t b64Len{};
			Base64UrlEncode((char const*)lastFrameHash.data(), lastFrameHash.size(), &b64, &b64Len);
			ImGui::Text("%s", b64);
			free(b64);
		}
		ImGui::End();
	}

	std::chrono::time_point endFrameToc = std::chrono::steady_clock::now();
	frameStats.AppendDuration(&FrameStats::endFrameStepDurations, endFrameToc - endFrameTic);
	
	if (showTiming) {
		if (ImGui::Begin("Timing")) {
			auto stepStatsUi = [&](std::string label, auto& seq) {
				auto [I, J] = std::minmax_element(seq.begin(), seq.end());
				ImGui::LabelText(fmt::format("{} min", label).c_str(), "%2.2f ms", *I * 1'000.0f);
				ImGui::LabelText(fmt::format("{} cur", label).c_str(), "%2.2f ms", seq.back() * 1'000.0f);
				ImGui::LabelText(fmt::format("{} max", label).c_str(), "%2.2f ms", *J * 1'000.0f);
				ImGui::PlotLines(label.c_str(),
					[](void* data, int idx) -> float { auto& dq = *(std::deque<float>*)data; return dq[idx]; },
					&seq, (int)seq.size(), 0, nullptr, 0.0f, 30.0f / 1000.0f);
				};
			stepStatsUi("MidFrame", frameStats.midFrameStepDurations);
			ImGui::Separator();
			stepStatsUi("EndFrame", frameStats.endFrameStepDurations);
		}
		ImGui::End();
	}

	ImGui::Render();
	ImGui_ImplOpenGL3_RenderDrawData(ImGui::GetDrawData());

	// Swap output buffers
	openGL->Swap();

	// Take screenshot
	switch (takeScreenshot) {
	case R_SSTGA:
	{
		targa_c i(sys->con);
		DoScreenshot(&i, IMGTYPE_RGB, "tga");
	}
	break;
	case R_SSJPEG:
	{
		jpeg_c i(sys->con);
		DoScreenshot(&i, IMGTYPE_RGB, "jpg");
	}
	break;
	case R_SSPNG:
	{
		png_c i(sys->con);
		DoScreenshot(&i, IMGTYPE_RGB, "png");
	}
	break;
	}
	takeScreenshot = R_SSNONE;

	PurgeShaders();
}

// =================
// Shader Management
// =================

std::optional<int> r_shaderHnd_c::StackCount() const
{
	if (!sh || sh->tex->status != r_tex_c::Status::DONE)
		return {};
	return (int)sh->tex->stackLayers;
}

void r_renderer_c::PurgeShaders()
{
	// Delete released shaders
	for (int s = 0; s < numShader; s++) {
		if (shaderList[s] && shaderList[s]->refCount == 0 && shaderList[s]->tex->status == r_tex_c::DONE) {
			delete shaderList[s];
			shaderList[s] = NULL;
		}
	}
}

r_shaderHnd_c* r_renderer_c::RegisterShader(std::string_view shname, int flags)
{
	if (shname.empty()) {
		return NULL;
	}

	std::string name(shname);
	dword nameHash = StringHash(name, 0xFFFF);
	int newId = -1;
	for (int s = 0; s < numShader; s++) {
		if (!shaderList[s]) {
			newId = s;
		}
		else if (shaderList[s]->nameHash == nameHash && _stricmp(name.c_str(), shaderList[s]->name.c_str()) == 0 && shaderList[s]->tex->flags == flags) {
			// Shader already exists, return a new handle for it
			// Ensure texture is loaded as soon as possible
			shaderList[s]->tex->ForceLoad();
			return new r_shaderHnd_c(shaderList[s]);
		}
	}
	if (newId == -1) {
		if (numShader == R_MAXSHADERS) {
			sys->con->Warning("shader limit reached");
			return NULL;
		}
		newId = numShader++;
	}
	shaderList[newId] = new r_shader_c(this, shname, flags);
	return new r_shaderHnd_c(shaderList[newId]);
}

r_shaderHnd_c* r_renderer_c::RegisterShaderFromImage(std::unique_ptr<image_c> img, int flags)
{
	int newId = -1;
	for (int s = 0; s < numShader; s++) {
		if (!shaderList[s]) {
			newId = s;
			break;
		}
	}
	if (newId == -1) {
		if (numShader == R_MAXSHADERS) {
			sys->con->Warning("shader limit reached");
			return NULL;
		}
		newId = numShader++;
	}
	char shname[32];
	sprintf(shname, "data:%d", newId);
	shaderList[newId] = new r_shader_c(this, shname, flags, std::move(img));
	return new r_shaderHnd_c(shaderList[newId]);
}

void r_renderer_c::GetShaderImageSize(r_shaderHnd_c* hnd, int& width, int& height)
{
	if (hnd)
	{
		while (hnd->sh->tex->status < r_tex_c::SIZE_KNOWN) {
			Sleep(1);
		}
		width = hnd->sh->tex->fileWidth;
		height = hnd->sh->tex->fileHeight;
	}
	else {
		width = 0;
		height = 0;
	}
}

void r_renderer_c::SetShaderLoadingPriority(r_shaderHnd_c* hnd, int pri)
{
	if (hnd && hnd->sh->tex->status != r_tex_c::DONE) {
		hnd->sh->tex->loadPri = pri;
	}
}

int r_renderer_c::GetTexAsyncCount()
{
	return texMan->GetAsyncCount();
}

// ==========
// 2D Drawing
// ==========

void r_renderer_c::SetClearColor(const col4_t col)
{
	glClearColor(col[0], col[1], col[2], col[3]);
}

void r_renderer_c::SetDrawLayer(int layer, int subLayer)
{
	if (layer == curLayer->layer && subLayer == curLayer->subLayer) {
		return;
	}
	r_layer_c* newCurLayer = NULL;
	for (int l = 0; l < numLayer; l++) {
		if (layerList[l]->layer == layer && layerList[l]->subLayer == subLayer) {
			newCurLayer = layerList[l];
			break;
		}
	}
	if (!newCurLayer) {
		if (numLayer == layerListSize) {
			layerListSize <<= 1;
			trealloc(layerList, layerListSize);
		}
		layerList[numLayer] = new r_layer_c(this, layer, subLayer);
		newCurLayer = layerList[numLayer];
		numLayer++;
	}
	curLayer = newCurLayer;
	curLayer->SetViewport(&curViewport);
	curLayer->SetBlendMode(curBlendMode);
}

void r_renderer_c::SetDrawSubLayer(int subLayer)
{
	SetDrawLayer(curLayer->layer, subLayer);
}

int r_renderer_c::GetDrawLayer()
{
	return curLayer->subLayer;
}

void r_renderer_c::SetViewport(int x, int y, int width, int height)
{
	if (height == 0) {
		auto& vid = sys->video->vid;
		width = VirtualScreenWidth();
		height = VirtualScreenHeight();
	}
	curViewport.x = x;
	curViewport.y = y;
	curViewport.width = width;
	curViewport.height = height;
	curLayer->SetViewport(&curViewport);
}

void r_renderer_c::SetBlendMode(int mode)
{
	curBlendMode = mode;
	curLayer->SetBlendMode(mode);
}

void r_renderer_c::DrawColor(const col4_t col)
{
	if (col) {
		Vector4Copy(col, drawColor);
	}
	else {
		drawColor[0] = 1.0f;
		drawColor[1] = 1.0f;
		drawColor[2] = 1.0f;
		drawColor[3] = 1.0f;
	}
}

void r_renderer_c::DrawColor(dword col)
{
	drawColor[0] = ((col >> 16) & 0xFF) / 255.0f;
	drawColor[1] = ((col >> 8) & 0xFF) / 255.0f;
	drawColor[2] = (col & 0xFF) / 255.0f;
	drawColor[3] = (col >> 24) / 255.0f;
}

void r_renderer_c::GetDrawColor(col4_t color)
{
	color[0] = drawColor[0];
	color[1] = drawColor[1];
	color[2] = drawColor[2];
	color[3] = drawColor[3];
}

void r_renderer_c::DrawImage(r_shaderHnd_c* hnd, glm::vec2 pos, glm::vec2 extent, glm::vec2 uv1, glm::vec2 uv2, int stackLayer, std::optional<int> maskLayer)
{
	DrawImageQuad(hnd,
		pos,
		pos + glm::vec2{ extent.x, 0 },
		pos + extent,
		pos + glm::vec2{ 0, extent.y },
		uv1,
		{ uv2.s, uv1.t },
		uv2,
		{ uv1.s, uv2.t },
		stackLayer, maskLayer);
}

void r_renderer_c::DrawImageQuad(r_shaderHnd_c* hnd, glm::vec2 p0, glm::vec2 p1, glm::vec2 p2, glm::vec2 p3, glm::vec2 uv0, glm::vec2 uv1, glm::vec2 uv2, glm::vec2 uv3, int stackLayer, std::optional<int> maskLayer)
{
	if (hnd) {
		curLayer->Bind(hnd->sh->tex);
		stackLayer = clamp(stackLayer, 0, (int)hnd->sh->tex->stackLayers - 1);
	}
	else {
		curLayer->Bind(whiteImage->sh->tex);
		stackLayer = 0;
	}
	curLayer->Color(drawColor);
	curLayer->Quad(
		uv0.s, uv0.t, p0.x, p0.y,
		uv1.s, uv1.t, p1.x, p1.y,
		uv2.s, uv2.t, p2.x, p2.y,
		uv3.s, uv3.t, p3.x, p3.y,
		stackLayer, maskLayer.value_or(-1));
}

void r_renderer_c::DrawString(float x, float y, int align, int height, const col4_t col, int font, const char* str)
{
	auto idxStr = IndexUTF8ToUTF32(str);
	if (font < 0 || font >= F_NUMFONTS) {
		font = F_FIXED;
	}

	scp_t pos = { x, y };
	if (col) {
		col4_t tcol;
		Vector4Copy(col, tcol);
		fonts[font]->Draw(pos, align, height, tcol, idxStr.text);
	}
	else {
		fonts[font]->Draw(pos, align, height, drawColor, idxStr.text);
	}
}

void r_renderer_c::DrawStringFormat(float x, float y, int align, int height, const col4_t col, int font, const char* fmt, ...)
{
	if (font < 0 || font >= F_NUMFONTS) {
		font = F_FIXED;
	}

	va_list va;
	va_start(va, fmt);

	scp_t pos = { x, y };
	if (col) {
		col4_t tcol;
		Vector4Copy(col, tcol);
		fonts[font]->VDraw(pos, align, height, tcol, fmt, va);
	}
	else {
		fonts[font]->VDraw(pos, align, height, drawColor, fmt, va);
	}

	va_end(va);
}

int	r_renderer_c::DrawStringWidth(int height, int font, const char* str)
{
	if (!*str) {
		return 0;
	}
	auto idxStr = IndexUTF8ToUTF32(str);
	if (font < 0 || font >= F_NUMFONTS) {
		font = F_FIXED;
	}
	return fonts[font]->StringWidth(height, idxStr.text);
}

int r_renderer_c::DrawStringCursorIndex(int height, int font, const char* str, int curX, int curY)
{
	if (!*str) {
		return 0;
	}
	std::string_view narrowView(str);
	auto idxStr = IndexUTF8ToUTF32(narrowView);
	if (font < 0 || font >= F_NUMFONTS) {
		font = F_FIXED;
	}
	size_t index = fonts[font]->StringCursorIndex(height, idxStr.text, curX, curY);
	if (index < idxStr.sourceCodeUnitOffsets.size()) {
		return (int)idxStr.sourceCodeUnitOffsets[index];
	}
	return (int)narrowView.size();
}

// ==============
// Virtual screen
// ==============

int r_renderer_c::VirtualScreenWidth() {
	return sys->video->vid.fbSize[0];
}

int r_renderer_c::VirtualScreenHeight() {
	return sys->video->vid.fbSize[1];
}

float r_renderer_c::VirtualScreenScaleFactor() {
	if (dpiScaleOverridePercent > 0) {
		return dpiScaleOverridePercent / 100.0f;
	}
	return sys->video->vid.dpiScale;
}

void r_renderer_c::SetDpiScaleOverridePercent(int percent) {
	dpiScaleOverridePercent = percent;
}

int r_renderer_c::DpiScaleOverridePercent() const {
	return dpiScaleOverridePercent;
}

int r_renderer_c::VirtualMap(int properValue) {
	if (apiDpiAware) {
		return properValue;
	}
	return static_cast<int>(properValue / VirtualScreenScaleFactor());
}

int r_renderer_c::VirtualUnmap(int mappedValue) {
	if (apiDpiAware) {
		return mappedValue;
	}
	return static_cast<int>(mappedValue * VirtualScreenScaleFactor());
}

float r_renderer_c::APICoordScale() {
	if (apiDpiAware) {
		return 1.0f;
	}
	return VirtualScreenScaleFactor();
}

// =====
// Debug
// =====

void r_renderer_c::ToggleDebugImGui() {
	debugImGui = !debugImGui;
}

// ===========
// Screenshots
// ===========

void r_renderer_c::C_Screenshot(IConsole* conHnd, args_c& args)
{
	const char* fmtName = args.argc >= 2 ? args.argv[1] : r_screenshotFormat->strVal.c_str();
	takeScreenshot = R_SSNONE;
	if (!_stricmp(fmtName, "tga")) {
		takeScreenshot = R_SSTGA;
	}
	else if (!_stricmp(fmtName, "jpg") || !_stricmp(fmtName, "jpeg")) {
		takeScreenshot = R_SSJPEG;
	}
	else if (!_stricmp(fmtName, "png")) {
		takeScreenshot = R_SSPNG;
	}
	else {
		conHnd->Warning("Unknown screenshot format '%s', valid formats: jpg, tga, png", fmtName);
	}
}

void r_renderer_c::DoScreenshot(image_c* i, int type, const char* ext)
{
	if (type != IMGTYPE_RGB) {
		return;
	}
	auto& rt = GetPresentRenderTarget();
	int const xs = rt.width;
	int const ys = rt.height;

	// Pixel reading only supports RGBA and an implementation-specific format.
	// Use RGBA for convenience as that's close enough to what we want to save in the end.
	int const readSize = xs * ys * 4;
	int const writeSize = xs * ys * 3;
	std::vector<byte> sbuf(readSize);

	// Read the front buffer
	GLint oldFb{};
	GLenum oglErr = glGetError();
	GLenum implColorReadFormat{}, implColorReadType{};
	glGetIntegerv(GL_FRAMEBUFFER_BINDING, &oldFb);
	glBindFramebuffer(GL_FRAMEBUFFER, rt.framebuffer);
	glPixelStorei(GL_PACK_ALIGNMENT, 1);
	glReadPixels(0, 0, xs, ys, GL_RGBA, GL_UNSIGNED_BYTE, sbuf.data());
	oglErr = glGetError();
	glBindFramebuffer(GL_FRAMEBUFFER, oldFb);

	// Flip and convert the image to RGB
	int const readSpan = xs * 4;
	int	const writeSpan = xs * 3;
	std::vector<byte> ss(writeSize);
	byte* p1 = sbuf.data();
	byte* p2 = ss.data() + writeSize - writeSpan;
	for (int y = 0; y < ys; ++y, p2 -= writeSpan * 2) {
		for (int x = 0; x < xs; ++x) {
			*p2++ = *p1++; // R
			*p2++ = *p1++; // G
			*p2++ = *p1++; // B
			p1++; // A
		}
	}
	sbuf.clear();

	// Set image info
	i->CopyRaw(IMGTYPE_RGB, xs, ys, ss.data());
	ss.clear();

	time_t curTime;
	time(&curTime);
	auto ssPath = std::filesystem::u8path(fmt::format(CFG_DATAPATH "Screenshots/{:%m%d%y_%H%M%S}.{}",
		*std::localtime(&curTime), ext));

	// Make folder if it doesn't exist
	std::error_code ec;
	std::filesystem::create_directories(ssPath.parent_path(), ec);
	if (ec) {
		sys->con->Print("Couldn't create screenshot folder!\n");
		return;
	}

	if (i->Save(ssPath)) {
		sys->con->Print("Couldn't write screenshot!\n");
		return;
	}
	sys->con->Print(fmt::format("Wrote screenshot to {}\n", ssPath.generic_u8string()).c_str());
}

r_renderer_c::RenderTarget& r_renderer_c::GetDrawRenderTarget()
{
	return rttMain[1 - presentRtt];
}

r_renderer_c::RenderTarget& r_renderer_c::GetPresentRenderTarget()
{
	return rttMain[presentRtt];
}

// ===========================================================
// MurmurHash implementation from public domain, obtained from
// https://github.com/explosion/murmurhash/blob/9281c4825c24e64476457db89fb1d39bf09b3d23/murmurhash/MurmurHash2.cpp
// ===========================================================

#if _WIN32
#define BIG_CONSTANT(x) (x)
#else
#define BIG_CONSTANT(x) (x##LLU)
#endif

static inline uint64_t MurmurHashGetBlock(const uint64_t* p)
{
#if defined(__BYTE_ORDER__) && (__BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__)
	return *p;
#else
	const uint8_t* c = (const uint8_t*)p;
	return (uint64_t)c[0] |
		(uint64_t)c[1] << 8 |
		(uint64_t)c[2] << 16 |
		(uint64_t)c[3] << 24 |
		(uint64_t)c[4] << 32 |
		(uint64_t)c[5] << 40 |
		(uint64_t)c[6] << 48 |
		(uint64_t)c[7] << 56;
#endif
}

uint64_t MurmurHash64A(const void* key, int len, uint64_t seed)
{
	const uint64_t m = BIG_CONSTANT(0xc6a4a7935bd1e995);
	const int r = 47;

	uint64_t h = seed ^ (len * m);

	const uint64_t* data = (const uint64_t*)key;
	const uint64_t* end = data + (len / 8);

	while (data != end)
	{
		uint64_t k = MurmurHashGetBlock(data++);

		k *= m;
		k ^= k >> r;
		k *= m;

		h ^= k;
		h *= m;
	}

	const unsigned char* data2 = (const unsigned char*)data;

	switch (len & 7)
	{
	case 7: h ^= uint64_t(data2[6]) << 48;
	case 6: h ^= uint64_t(data2[5]) << 40;
	case 5: h ^= uint64_t(data2[4]) << 32;
	case 4: h ^= uint64_t(data2[3]) << 24;
	case 3: h ^= uint64_t(data2[2]) << 16;
	case 2: h ^= uint64_t(data2[1]) << 8;
	case 1: h ^= uint64_t(data2[0]);
		h *= m;
	};

	h ^= h >> r;
	h *= m;
	h ^= h >> r;

	return h;
}