From 7ac0cace7d1eed574de31e0a8b28517e759034a7 Mon Sep 17 00:00:00 2001
From: Magnus Norddahl <dpjudas@users.noreply.github.com>
Date: Sun, 20 Nov 2016 02:07:55 +0100
Subject: [PATCH] Stop copying the matrix and remove r_triangle

---
 src/CMakeLists.txt           |    1 -
 src/r_compiler/llvmdrawers.h |    2 -
 src/r_plane.cpp              |  266 -------
 src/r_poly_decal.cpp         |    2 +-
 src/r_poly_decal.h           |    2 +-
 src/r_poly_intersection.h    |    2 +-
 src/r_poly_particle.cpp      |    2 +-
 src/r_poly_plane.cpp         |    4 +-
 src/r_poly_sky.cpp           |    3 +-
 src/r_poly_sky.h             |    1 +
 src/r_poly_sprite.cpp        |    2 +-
 src/r_poly_triangle.cpp      |  165 ++++-
 src/r_poly_triangle.h        |   16 +-
 src/r_poly_wall.cpp          |    2 +-
 src/r_poly_wallsprite.cpp    |    2 +-
 src/r_poly_wallsprite.h      |    2 +-
 src/r_triangle.cpp           | 1338 ----------------------------------
 src/r_triangle.h             |  116 ---
 18 files changed, 183 insertions(+), 1745 deletions(-)
 delete mode 100644 src/r_triangle.cpp
 delete mode 100644 src/r_triangle.h

diff --git a/src/CMakeLists.txt b/src/CMakeLists.txt
index 8c2e36220..987f23f2f 100644
--- a/src/CMakeLists.txt
+++ b/src/CMakeLists.txt
@@ -1092,7 +1092,6 @@ set( FASTMATH_PCH_SOURCES
 	r_segs.cpp
 	r_sky.cpp
 	r_things.cpp
-	r_triangle.cpp
 	s_advsound.cpp
 	s_environment.cpp
 	s_playlist.cpp
diff --git a/src/r_compiler/llvmdrawers.h b/src/r_compiler/llvmdrawers.h
index e42153743..1db2f5b2a 100644
--- a/src/r_compiler/llvmdrawers.h
+++ b/src/r_compiler/llvmdrawers.h
@@ -234,8 +234,6 @@ struct TriUniforms
 		nearest_filter = 2,
 		fixed_light = 4
 	};
-
-	TriMatrix objectToClip;
 };
 
 struct TriDrawTriangleArgs
diff --git a/src/r_plane.cpp b/src/r_plane.cpp
index 17728e80c..ca8547003 100644
--- a/src/r_plane.cpp
+++ b/src/r_plane.cpp
@@ -59,7 +59,6 @@
 #include "v_palette.h"
 #include "r_data/colormaps.h"
 #include "r_draw_rgba.h"
-#include "r_triangle.h"
 
 #ifdef _MSC_VER
 #pragma warning(disable:4244)
@@ -1015,261 +1014,6 @@ static const BYTE *R_GetTwoSkyColumns (FTexture *fronttex, int x)
 	}
 }
 
-static void R_DrawCubeSky(visplane_t *pl)
-{
-	int x1 = pl->left;
-	int x2 = pl->right;
-	short *uwal = (short *)pl->top;
-	short *dwal = (short *)pl->bottom;
-
-	static TriVertex cube[6 * 6] =
-	{
-		// Top
-		{ -1.0f,  1.0f,  0.6f, 1.0f, 1.0f, 0.0f },
-		{ 1.0f,  1.0f,  0.6f, 1.0f, 0.0f, 0.0f },
-		{ 1.0f, -1.0f,  0.6f, 1.0f, 0.0f, 0.1f },
-
-		{ 1.0f, -1.0f,  0.6f, 1.0f, 0.0f, 0.1f },
-		{ -1.0f, -1.0f,  0.6f, 1.0f, 1.0f, 0.1f },
-		{ -1.0f,  1.0f,  0.6f, 1.0f, 1.0f, 0.0f },
-
-		// Bottom
-		{ 1.0f, -1.0f, -1.0f, 1.0f, 1.0f, 0.9f },
-		{ 1.0f,  1.0f, -1.0f, 1.0f, 1.0f, 1.0f },
-		{ -1.0f,  1.0f, -1.0f, 1.0f, 0.0f, 1.0f },
-
-		{ -1.0f,  1.0f, -1.0f, 1.0f, 0.0f, 1.0f },
-		{ -1.0f, -1.0f, -1.0f, 1.0f, 0.0f, 0.9f },
-		{ 1.0f, -1.0f, -1.0f, 1.0f, 1.0f, 0.9f },
-
-		// Front
-		{ 1.0f,  1.0f, -1.0f, 1.0f, 1.0f, 2.0f },
-		{ 1.0f,  1.0f,  0.6f, 1.0f, 1.0f, 0.0f },
-		{ -1.0f,  1.0f,  0.6f, 1.0f, 0.0f, 0.0f },
-
-		{ -1.0f,  1.0f,  0.6f, 1.0f, 0.0f, 0.0f },
-		{ -1.0f,  1.0f, -1.0f, 1.0f, 0.0f, 2.0f },
-		{ 1.0f,  1.0f, -1.0f, 1.0f, 1.0f, 2.0f },
-
-		// Back
-		{ -1.0f, -1.0f,  0.6f, 1.0f, 1.0f, 0.0f },
-		{ 1.0f, -1.0f,  0.6f, 1.0f, 0.0f, 0.0f },
-		{ 1.0f, -1.0f, -1.0f, 1.0f, 0.0f, 2.0f },
-
-		{ 1.0f, -1.0f, -1.0f, 1.0f, 0.0f, 2.0f },
-		{ -1.0f, -1.0f, -1.0f, 1.0f, 1.0f, 2.0f },
-		{ -1.0f, -1.0f,  0.6f, 1.0f, 1.0f, 0.0f },
-
-		// Right
-		{ 1.0f, -1.0f,  0.6f, 1.0f, 1.0f, 0.0f },
-		{ 1.0f,  1.0f,  0.6f, 1.0f, 0.0f, 0.0f },
-		{ 1.0f,  1.0f, -1.0f, 1.0f, 0.0f, 2.0f },
-
-		{ 1.0f,  1.0f, -1.0f, 1.0f, 0.0f, 2.0f },
-		{ 1.0f, -1.0f, -1.0f, 1.0f, 1.0f, 2.0f },
-		{ 1.0f, -1.0f,  0.6f, 1.0f, 1.0f, 0.0f },
-
-		// Left
-		{ -1.0f,  1.0f, -1.0f, 1.0f, 1.0f, 2.0f },
-		{ -1.0f,  1.0f,  0.6f, 1.0f, 1.0f, 0.0f },
-		{ -1.0f, -1.0f,  0.6f, 1.0f, 0.0f, 0.0f },
-
-		{ -1.0f, -1.0f,  0.6f, 1.0f, 0.0f, 0.0f },
-		{ -1.0f, -1.0f, -1.0f, 1.0f, 0.0f, 2.0f },
-		{ -1.0f,  1.0f, -1.0f, 1.0f, 1.0f, 2.0f }
-	};
-
-	TriMatrix objectToWorld = TriMatrix::translate((float)ViewPos.X, (float)ViewPos.Y, (float)ViewPos.Z) * TriMatrix::scale(1000.0f, 1000.0f, 1000.0f);
-	TriMatrix objectToClip = TriMatrix::viewToClip() * TriMatrix::worldToView() * objectToWorld;
-	//TriMatrix objectToWorld = TriMatrix::scale(1000.0f, 1000.0f, 1000.0f);
-	//TriMatrix objectToClip = TriMatrix::viewToClip() * objectToWorld;
-
-	uint32_t solid_top = frontskytex->GetSkyCapColor(false);
-	uint32_t solid_bottom = frontskytex->GetSkyCapColor(true);
-
-	solid_top = RGB32k.RGB[(RPART(solid_top) >> 3)][(GPART(solid_top) >> 3)][(BPART(solid_top) >> 3)];
-	solid_bottom = RGB32k.RGB[(RPART(solid_bottom) >> 3)][(GPART(solid_bottom) >> 3)][(BPART(solid_bottom) >> 3)];
-
-	TriUniforms uniforms;
-	uniforms.objectToClip = objectToClip;
-	uniforms.light = 256;
-	uniforms.flags = 0;
-
-	TriangleDrawer::fill(uniforms, cube, 6, TriangleDrawMode::Normal, false, x1, x2 - 1, uwal, dwal, solid_top);
-	TriangleDrawer::fill(uniforms, cube + 6, 6, TriangleDrawMode::Normal, false, x1, x2 - 1, uwal, dwal, solid_bottom);
-	TriangleDrawer::draw(uniforms, cube + 2 * 6, 4 * 6, TriangleDrawMode::Normal, false, x1, x2 - 1, uwal, dwal, frontskytex);
-}
-
-namespace
-{
-	class SkyDome
-	{
-	public:
-		SkyDome() { CreateDome(); }
-		void Render(visplane_t *pl);
-
-	private:
-		TArray<TriVertex> mVertices;
-		TArray<unsigned int> mPrimStart;
-		int mRows, mColumns;
-
-		void SkyVertex(int r, int c, bool yflip);
-		void CreateSkyHemisphere(bool zflip);
-		void CreateDome();
-		void RenderRow(int row, visplane_t *pl);
-		void RenderCapColorRow(int row, bool bottomCap, visplane_t *pl);
-
-		TriVertex SetVertex(float xx, float yy, float zz, float uu = 0, float vv = 0);
-		TriVertex SetVertexXYZ(float xx, float yy, float zz, float uu = 0, float vv = 0);
-	};
-
-	TriVertex SkyDome::SetVertex(float xx, float yy, float zz, float uu, float vv)
-	{
-		TriVertex v;
-		v.x = xx;
-		v.y = yy;
-		v.z = zz;
-		v.w = 1.0f;
-		v.varying[0] = uu;
-		v.varying[1] = vv;
-		return v;
-	}
-
-	TriVertex SkyDome::SetVertexXYZ(float xx, float yy, float zz, float uu, float vv)
-	{
-		TriVertex v;
-		v.x = xx;
-		v.y = zz;
-		v.z = yy;
-		v.w = 1.0f;
-		v.varying[0] = uu;
-		v.varying[1] = vv;
-		return v;
-	}
-
-	void SkyDome::SkyVertex(int r, int c, bool zflip)
-	{
-		static const FAngle maxSideAngle = 60.f;
-		static const float scale = 10000.;
-
-		FAngle topAngle = (c / (float)mColumns * 360.f);
-		FAngle sideAngle = maxSideAngle * (mRows - r) / mRows;
-		float height = sideAngle.Sin();
-		float realRadius = scale * sideAngle.Cos();
-		FVector2 pos = topAngle.ToVector(realRadius);
-		float z = (!zflip) ? scale * height : -scale * height;
-
-		float u, v;
-		//uint32_t color = r == 0 ? 0xffffff : 0xffffffff;
-
-		// And the texture coordinates.
-		if (!zflip)	// Flipped Y is for the lower hemisphere.
-		{
-			u = (-c / (float)mColumns);
-			v = (r / (float)mRows);
-		}
-		else
-		{
-			u = (-c / (float)mColumns);
-			v = 1.0f + ((mRows - r) / (float)mRows);
-		}
-
-		if (r != 4) z += 300;
-
-		// And finally the vertex.
-		TriVertex vert;
-		vert = SetVertexXYZ(-pos.X, z - 1.f, pos.Y, u * 4.0f, v + 0.5f/*, color*/);
-		mVertices.Push(vert);
-	}
-
-	void SkyDome::CreateSkyHemisphere(bool zflip)
-	{
-		int r, c;
-
-		mPrimStart.Push(mVertices.Size());
-
-		for (c = 0; c < mColumns; c++)
-		{
-			SkyVertex(1, c, zflip);
-		}
-
-		// The total number of triangles per hemisphere can be calculated
-		// as follows: rows * columns * 2 + 2 (for the top cap).
-		for (r = 0; r < mRows; r++)
-		{
-			mPrimStart.Push(mVertices.Size());
-			for (c = 0; c <= mColumns; c++)
-			{
-				SkyVertex(r + zflip, c, zflip);
-				SkyVertex(r + 1 - zflip, c, zflip);
-			}
-		}
-	}
-
-	void SkyDome::CreateDome()
-	{
-		mColumns = 128;
-		mRows = 4;
-		CreateSkyHemisphere(false);
-		CreateSkyHemisphere(true);
-		mPrimStart.Push(mVertices.Size());
-	}
-
-	void SkyDome::RenderRow(int row, visplane_t *pl)
-	{
-		int x1 = pl->left;
-		int x2 = pl->right;
-		short *uwal = (short *)pl->top;
-		short *dwal = (short *)pl->bottom;
-		TriMatrix objectToWorld = TriMatrix::translate((float)ViewPos.X, (float)ViewPos.Y, (float)ViewPos.Z);
-		TriMatrix objectToClip = TriMatrix::viewToClip() * TriMatrix::worldToView() * objectToWorld;
-		TriUniforms uniforms;
-		uniforms.objectToClip = objectToClip;
-		uniforms.light = 256;
-		uniforms.flags = 0;
-		TriangleDrawer::draw(uniforms, &mVertices[mPrimStart[row]], mPrimStart[row + 1] - mPrimStart[row], TriangleDrawMode::Strip, false, x1, x2 - 1, uwal, dwal, frontskytex);
-	}
-
-	void SkyDome::RenderCapColorRow(int row, bool bottomCap, visplane_t *pl)
-	{
-		uint32_t solid = frontskytex->GetSkyCapColor(bottomCap);
-		solid = RGB32k.RGB[(RPART(solid) >> 3)][(GPART(solid) >> 3)][(BPART(solid) >> 3)];
-
-		int x1 = pl->left;
-		int x2 = pl->right;
-		short *uwal = (short *)pl->top;
-		short *dwal = (short *)pl->bottom;
-		TriMatrix objectToWorld = TriMatrix::translate((float)ViewPos.X, (float)ViewPos.Y, (float)ViewPos.Z);
-		TriMatrix objectToClip = TriMatrix::viewToClip() * TriMatrix::worldToView() * objectToWorld;
-		TriUniforms uniforms;
-		uniforms.objectToClip = objectToClip;
-		uniforms.light = 256;
-		uniforms.flags = 0;
-		TriangleDrawer::fill(uniforms, &mVertices[mPrimStart[row]], mPrimStart[row + 1] - mPrimStart[row], TriangleDrawMode::Fan, bottomCap, x1, x2 - 1, uwal, dwal, solid);
-	}
-
-	void SkyDome::Render(visplane_t *pl)
-	{
-		int rc = mRows + 1;
-
-		// No need to draw this as the software renderer can't look that high anyway
-		//RenderCapColorRow(0, false, pl);
-		//RenderCapColorRow(rc, true, pl);
-
-		for (int i = 1; i <= mRows; i++)
-		{
-			RenderRow(i, pl);
-			RenderRow(rc + i, pl);
-		}
-	}
-}
-
-static void R_DrawDomeSky(visplane_t *pl)
-{
-	static SkyDome skydome;
-	skydome.Render(pl);
-}
-
 static void R_DrawSkyColumnStripe(int start_x, int y1, int y2, int columns, double scale, double texturemid, double yrepeat)
 {
 	uint32_t height = frontskytex->GetHeight();
@@ -1467,16 +1211,6 @@ static void R_DrawSky (visplane_t *pl)
 		R_DrawCapSky(pl);
 		return;
 	}
-	else if (r_skymode == 3)
-	{
-		R_DrawCubeSky(pl);
-		return;
-	}
-	else if (r_skymode == 4)
-	{
-		R_DrawDomeSky(pl);
-		return;
-	}
 
 	int x;
 	float swal;
diff --git a/src/r_poly_decal.cpp b/src/r_poly_decal.cpp
index 5641bfcd1..52caae7e2 100644
--- a/src/r_poly_decal.cpp
+++ b/src/r_poly_decal.cpp
@@ -129,7 +129,6 @@ void RenderPolyDecal::Render(const TriMatrix &worldToClip, DBaseDecal *decal, co
 	bool fullbrightSprite = (decal->RenderFlags & RF_FULLBRIGHT) == RF_FULLBRIGHT;
 
 	TriUniforms uniforms;
-	uniforms.objectToClip = worldToClip;
 	if (fullbrightSprite || fixedlightlev >= 0 || fixedcolormap)
 	{
 		uniforms.light = 256;
@@ -145,6 +144,7 @@ void RenderPolyDecal::Render(const TriMatrix &worldToClip, DBaseDecal *decal, co
 
 	PolyDrawArgs args;
 	args.uniforms = uniforms;
+	args.objectToClip = &worldToClip;
 	args.vinput = vertices;
 	args.vcount = 4;
 	args.mode = TriangleDrawMode::Fan;
diff --git a/src/r_poly_decal.h b/src/r_poly_decal.h
index 39b250abb..61087a773 100644
--- a/src/r_poly_decal.h
+++ b/src/r_poly_decal.h
@@ -22,7 +22,7 @@
 
 #pragma once
 
-#include "r_triangle.h"
+#include "r_poly_triangle.h"
 
 class RenderPolyDecal
 {
diff --git a/src/r_poly_intersection.h b/src/r_poly_intersection.h
index 48ebef031..b06bc1200 100644
--- a/src/r_poly_intersection.h
+++ b/src/r_poly_intersection.h
@@ -22,7 +22,7 @@
 
 #pragma once
 
-#include "r_triangle.h"
+#include "r_poly_triangle.h"
 #include <algorithm>
 #include <cmath>
 
diff --git a/src/r_poly_particle.cpp b/src/r_poly_particle.cpp
index e5797a108..8448c8f32 100644
--- a/src/r_poly_particle.cpp
+++ b/src/r_poly_particle.cpp
@@ -71,7 +71,6 @@ void RenderPolyParticle::Render(const TriMatrix &worldToClip, particle_t *partic
 	bool fullbrightSprite = particle->bright != 0;
 
 	TriUniforms uniforms;
-	uniforms.objectToClip = worldToClip;
 	if (fullbrightSprite || fixedlightlev >= 0 || fixedcolormap)
 	{
 		uniforms.light = 256;
@@ -88,6 +87,7 @@ void RenderPolyParticle::Render(const TriMatrix &worldToClip, particle_t *partic
 
 	PolyDrawArgs args;
 	args.uniforms = uniforms;
+	args.objectToClip = &worldToClip;
 	args.vinput = vertices;
 	args.vcount = 4;
 	args.mode = TriangleDrawMode::Fan;
diff --git a/src/r_poly_plane.cpp b/src/r_poly_plane.cpp
index 9005e32f1..a68a4b568 100644
--- a/src/r_poly_plane.cpp
+++ b/src/r_poly_plane.cpp
@@ -103,7 +103,6 @@ void RenderPolyPlane::Render3DFloor(const TriMatrix &worldToClip, subsector_t *s
 	}
 
 	TriUniforms uniforms;
-	uniforms.objectToClip = worldToClip;
 	uniforms.light = (uint32_t)(lightlevel / 255.0f * 256.0f);
 	if (fixedlightlev >= 0 || fixedcolormap)
 		uniforms.light = 256;
@@ -133,6 +132,7 @@ void RenderPolyPlane::Render3DFloor(const TriMatrix &worldToClip, subsector_t *s
 
 	PolyDrawArgs args;
 	args.uniforms = uniforms;
+	args.objectToClip = &worldToClip;
 	args.vinput = vertices;
 	args.vcount = sub->numlines;
 	args.mode = TriangleDrawMode::Fan;
@@ -195,7 +195,6 @@ void RenderPolyPlane::Render(const TriMatrix &worldToClip, subsector_t *sub, uin
 	bool isSky = picnum == skyflatnum;
 
 	TriUniforms uniforms;
-	uniforms.objectToClip = worldToClip;
 	uniforms.light = (uint32_t)(frontsector->lightlevel / 255.0f * 256.0f);
 	if (fixedlightlev >= 0 || fixedcolormap)
 		uniforms.light = 256;
@@ -225,6 +224,7 @@ void RenderPolyPlane::Render(const TriMatrix &worldToClip, subsector_t *sub, uin
 
 	PolyDrawArgs args;
 	args.uniforms = uniforms;
+	args.objectToClip = &worldToClip;
 	args.vinput = vertices;
 	args.vcount = sub->numlines;
 	args.mode = TriangleDrawMode::Fan;
diff --git a/src/r_poly_sky.cpp b/src/r_poly_sky.cpp
index b0b06f262..a60b3f460 100644
--- a/src/r_poly_sky.cpp
+++ b/src/r_poly_sky.cpp
@@ -48,9 +48,9 @@ void PolySkyDome::Render(const TriMatrix &worldToClip)
 		backskytex = TexMan(sky2tex, true);
 
 	TriMatrix objectToWorld = TriMatrix::translate((float)ViewPos.X, (float)ViewPos.Y, (float)ViewPos.Z);
+	objectToClip = worldToClip * objectToWorld;
 
 	TriUniforms uniforms;
-	uniforms.objectToClip = worldToClip * objectToWorld;
 	uniforms.light = 256;
 	uniforms.flags = 0;
 	uniforms.subsectorDepth = RenderPolyScene::SkySubsectorDepth;
@@ -59,6 +59,7 @@ void PolySkyDome::Render(const TriMatrix &worldToClip)
 
 	PolyDrawArgs args;
 	args.uniforms = uniforms;
+	args.objectToClip = &objectToClip;
 	args.stenciltestvalue = 255;
 	args.stencilwritevalue = 1;
 	args.SetTexture(frontskytex);
diff --git a/src/r_poly_sky.h b/src/r_poly_sky.h
index dd4bd29ca..95ec27ed6 100644
--- a/src/r_poly_sky.h
+++ b/src/r_poly_sky.h
@@ -33,6 +33,7 @@ private:
 	TArray<TriVertex> mVertices;
 	TArray<unsigned int> mPrimStart;
 	int mRows, mColumns;
+	TriMatrix objectToClip;
 
 	void SkyVertex(int r, int c, bool yflip);
 	void CreateSkyHemisphere(bool zflip);
diff --git a/src/r_poly_sprite.cpp b/src/r_poly_sprite.cpp
index 8c74d0d1a..84ff8e221 100644
--- a/src/r_poly_sprite.cpp
+++ b/src/r_poly_sprite.cpp
@@ -114,7 +114,6 @@ void RenderPolySprite::Render(const TriMatrix &worldToClip, AActor *thing, subse
 	bool fullbrightSprite = ((thing->renderflags & RF_FULLBRIGHT) || (thing->flags5 & MF5_BRIGHT));
 
 	TriUniforms uniforms;
-	uniforms.objectToClip = worldToClip;
 	if (fullbrightSprite || fixedlightlev >= 0 || fixedcolormap)
 	{
 		uniforms.light = 256;
@@ -129,6 +128,7 @@ void RenderPolySprite::Render(const TriMatrix &worldToClip, AActor *thing, subse
 
 	PolyDrawArgs args;
 	args.uniforms = uniforms;
+	args.objectToClip = &worldToClip;
 	args.vinput = vertices;
 	args.vcount = 4;
 	args.mode = TriangleDrawMode::Fan;
diff --git a/src/r_poly_triangle.cpp b/src/r_poly_triangle.cpp
index ad9f7ff95..5574251e6 100644
--- a/src/r_poly_triangle.cpp
+++ b/src/r_poly_triangle.cpp
@@ -124,28 +124,28 @@ void PolyTriangleDrawer::draw_arrays(const PolyDrawArgs &drawargs, TriDrawVarian
 		for (int i = 0; i < vcount / 3; i++)
 		{
 			for (int j = 0; j < 3; j++)
-				vert[j] = shade_vertex(drawargs.uniforms, *(vinput++));
+				vert[j] = shade_vertex(*drawargs.objectToClip, *(vinput++));
 			draw_shaded_triangle(vert, ccw, &args, thread, drawfunc);
 		}
 	}
 	else if (drawargs.mode == TriangleDrawMode::Fan)
 	{
-		vert[0] = shade_vertex(drawargs.uniforms, *(vinput++));
-		vert[1] = shade_vertex(drawargs.uniforms, *(vinput++));
+		vert[0] = shade_vertex(*drawargs.objectToClip, *(vinput++));
+		vert[1] = shade_vertex(*drawargs.objectToClip, *(vinput++));
 		for (int i = 2; i < vcount; i++)
 		{
-			vert[2] = shade_vertex(drawargs.uniforms, *(vinput++));
+			vert[2] = shade_vertex(*drawargs.objectToClip, *(vinput++));
 			draw_shaded_triangle(vert, ccw, &args, thread, drawfunc);
 			vert[1] = vert[2];
 		}
 	}
 	else // TriangleDrawMode::Strip
 	{
-		vert[0] = shade_vertex(drawargs.uniforms, *(vinput++));
-		vert[1] = shade_vertex(drawargs.uniforms, *(vinput++));
+		vert[0] = shade_vertex(*drawargs.objectToClip, *(vinput++));
+		vert[1] = shade_vertex(*drawargs.objectToClip, *(vinput++));
 		for (int i = 2; i < vcount; i++)
 		{
-			vert[2] = shade_vertex(drawargs.uniforms, *(vinput++));
+			vert[2] = shade_vertex(*drawargs.objectToClip, *(vinput++));
 			draw_shaded_triangle(vert, ccw, &args, thread, drawfunc);
 			vert[0] = vert[1];
 			vert[1] = vert[2];
@@ -154,10 +154,10 @@ void PolyTriangleDrawer::draw_arrays(const PolyDrawArgs &drawargs, TriDrawVarian
 	}
 }
 
-TriVertex PolyTriangleDrawer::shade_vertex(const TriUniforms &uniforms, TriVertex v)
+TriVertex PolyTriangleDrawer::shade_vertex(const TriMatrix &objectToClip, TriVertex v)
 {
 	// Apply transform to get clip coordinates:
-	return uniforms.objectToClip * v;
+	return objectToClip * v;
 }
 
 void PolyTriangleDrawer::draw_shaded_triangle(const TriVertex *vert, bool ccw, TriDrawTriangleArgs *args, WorkerThreadData *thread, void(*drawfunc)(const TriDrawTriangleArgs *, WorkerThreadData *))
@@ -341,3 +341,150 @@ FString DrawPolyTrianglesCommand::DebugInfo()
 {
 	return "DrawPolyTriangles";
 }
+
+/////////////////////////////////////////////////////////////////////////////
+
+TriMatrix TriMatrix::null()
+{
+	TriMatrix m;
+	memset(m.matrix, 0, sizeof(m.matrix));
+	return m;
+}
+
+TriMatrix TriMatrix::identity()
+{
+	TriMatrix m = null();
+	m.matrix[0] = 1.0f;
+	m.matrix[5] = 1.0f;
+	m.matrix[10] = 1.0f;
+	m.matrix[15] = 1.0f;
+	return m;
+}
+
+TriMatrix TriMatrix::translate(float x, float y, float z)
+{
+	TriMatrix m = identity();
+	m.matrix[0 + 3 * 4] = x;
+	m.matrix[1 + 3 * 4] = y;
+	m.matrix[2 + 3 * 4] = z;
+	return m;
+}
+
+TriMatrix TriMatrix::scale(float x, float y, float z)
+{
+	TriMatrix m = null();
+	m.matrix[0 + 0 * 4] = x;
+	m.matrix[1 + 1 * 4] = y;
+	m.matrix[2 + 2 * 4] = z;
+	m.matrix[3 + 3 * 4] = 1;
+	return m;
+}
+
+TriMatrix TriMatrix::rotate(float angle, float x, float y, float z)
+{
+	float c = cosf(angle);
+	float s = sinf(angle);
+	TriMatrix m = null();
+	m.matrix[0 + 0 * 4] = (x*x*(1.0f - c) + c);
+	m.matrix[0 + 1 * 4] = (x*y*(1.0f - c) - z*s);
+	m.matrix[0 + 2 * 4] = (x*z*(1.0f - c) + y*s);
+	m.matrix[1 + 0 * 4] = (y*x*(1.0f - c) + z*s);
+	m.matrix[1 + 1 * 4] = (y*y*(1.0f - c) + c);
+	m.matrix[1 + 2 * 4] = (y*z*(1.0f - c) - x*s);
+	m.matrix[2 + 0 * 4] = (x*z*(1.0f - c) - y*s);
+	m.matrix[2 + 1 * 4] = (y*z*(1.0f - c) + x*s);
+	m.matrix[2 + 2 * 4] = (z*z*(1.0f - c) + c);
+	m.matrix[3 + 3 * 4] = 1.0f;
+	return m;
+}
+
+TriMatrix TriMatrix::swapYZ()
+{
+	TriMatrix m = null();
+	m.matrix[0 + 0 * 4] = 1.0f;
+	m.matrix[1 + 2 * 4] = 1.0f;
+	m.matrix[2 + 1 * 4] = -1.0f;
+	m.matrix[3 + 3 * 4] = 1.0f;
+	return m;
+}
+
+TriMatrix TriMatrix::perspective(float fovy, float aspect, float z_near, float z_far)
+{
+	float f = (float)(1.0 / tan(fovy * M_PI / 360.0));
+	TriMatrix m = null();
+	m.matrix[0 + 0 * 4] = f / aspect;
+	m.matrix[1 + 1 * 4] = f;
+	m.matrix[2 + 2 * 4] = (z_far + z_near) / (z_near - z_far);
+	m.matrix[2 + 3 * 4] = (2.0f * z_far * z_near) / (z_near - z_far);
+	m.matrix[3 + 2 * 4] = -1.0f;
+	return m;
+}
+
+TriMatrix TriMatrix::frustum(float left, float right, float bottom, float top, float near, float far)
+{
+	float a = (right + left) / (right - left);
+	float b = (top + bottom) / (top - bottom);
+	float c = -(far + near) / (far - near);
+	float d = -(2.0f * far) / (far - near);
+	TriMatrix m = null();
+	m.matrix[0 + 0 * 4] = 2.0f * near / (right - left);
+	m.matrix[1 + 1 * 4] = 2.0f * near / (top - bottom);
+	m.matrix[0 + 2 * 4] = a;
+	m.matrix[1 + 2 * 4] = b;
+	m.matrix[2 + 2 * 4] = c;
+	m.matrix[2 + 3 * 4] = d;
+	m.matrix[3 + 2 * 4] = -1;
+	return m;
+}
+
+TriMatrix TriMatrix::worldToView()
+{
+	TriMatrix m = null();
+	m.matrix[0 + 0 * 4] = (float)ViewSin;
+	m.matrix[0 + 1 * 4] = (float)-ViewCos;
+	m.matrix[1 + 2 * 4] = 1.0f;
+	m.matrix[2 + 0 * 4] = (float)-ViewCos;
+	m.matrix[2 + 1 * 4] = (float)-ViewSin;
+	m.matrix[3 + 3 * 4] = 1.0f;
+	return m * translate((float)-ViewPos.X, (float)-ViewPos.Y, (float)-ViewPos.Z);
+}
+
+TriMatrix TriMatrix::viewToClip()
+{
+	float near = 5.0f;
+	float far = 65536.0f;
+	float width = (float)(FocalTangent * near);
+	float top = (float)(CenterY / InvZtoScale * near);
+	float bottom = (float)(top - viewheight / InvZtoScale * near);
+	return frustum(-width, width, bottom, top, near, far);
+}
+
+TriMatrix TriMatrix::operator*(const TriMatrix &mult) const
+{
+	TriMatrix result;
+	for (int x = 0; x < 4; x++)
+	{
+		for (int y = 0; y < 4; y++)
+		{
+			result.matrix[x + y * 4] =
+				matrix[0 * 4 + x] * mult.matrix[y * 4 + 0] +
+				matrix[1 * 4 + x] * mult.matrix[y * 4 + 1] +
+				matrix[2 * 4 + x] * mult.matrix[y * 4 + 2] +
+				matrix[3 * 4 + x] * mult.matrix[y * 4 + 3];
+		}
+	}
+	return result;
+}
+
+TriVertex TriMatrix::operator*(TriVertex v) const
+{
+	float vx = matrix[0 * 4 + 0] * v.x + matrix[1 * 4 + 0] * v.y + matrix[2 * 4 + 0] * v.z + matrix[3 * 4 + 0] * v.w;
+	float vy = matrix[0 * 4 + 1] * v.x + matrix[1 * 4 + 1] * v.y + matrix[2 * 4 + 1] * v.z + matrix[3 * 4 + 1] * v.w;
+	float vz = matrix[0 * 4 + 2] * v.x + matrix[1 * 4 + 2] * v.y + matrix[2 * 4 + 2] * v.z + matrix[3 * 4 + 2] * v.w;
+	float vw = matrix[0 * 4 + 3] * v.x + matrix[1 * 4 + 3] * v.y + matrix[2 * 4 + 3] * v.z + matrix[3 * 4 + 3] * v.w;
+	v.x = vx;
+	v.y = vy;
+	v.z = vz;
+	v.w = vw;
+	return v;
+}
diff --git a/src/r_poly_triangle.h b/src/r_poly_triangle.h
index f0797460f..db6719de9 100644
--- a/src/r_poly_triangle.h
+++ b/src/r_poly_triangle.h
@@ -22,15 +22,27 @@
 
 #pragma once
 
-#include "r_triangle.h"
+#include "r_draw.h"
+#include "r_thread.h"
+#include "r_compiler/llvmdrawers.h"
 #include "r_data/r_translate.h"
 
+class FTexture;
+
+enum class TriangleDrawMode
+{
+	Normal,
+	Fan,
+	Strip
+};
+
 struct TriDrawTriangleArgs;
 
 class PolyDrawArgs
 {
 public:
 	TriUniforms uniforms;
+	const TriMatrix *objectToClip = nullptr;
 	const TriVertex *vinput = nullptr;
 	int vcount = 0;
 	TriangleDrawMode mode = TriangleDrawMode::Normal;
@@ -83,7 +95,7 @@ public:
 	static void draw(const PolyDrawArgs &args, TriDrawVariant variant, TriBlendMode blendmode);
 
 private:
-	static TriVertex shade_vertex(const TriUniforms &uniforms, TriVertex v);
+	static TriVertex shade_vertex(const TriMatrix &objectToClip, TriVertex v);
 	static void draw_arrays(const PolyDrawArgs &args, TriDrawVariant variant, TriBlendMode blendmode, WorkerThreadData *thread);
 	static void draw_shaded_triangle(const TriVertex *vertices, bool ccw, TriDrawTriangleArgs *args, WorkerThreadData *thread, void(*drawfunc)(const TriDrawTriangleArgs *, WorkerThreadData *));
 	static bool cullhalfspace(float clipdistance1, float clipdistance2, float &t1, float &t2);
diff --git a/src/r_poly_wall.cpp b/src/r_poly_wall.cpp
index e8bad52a7..f0eb33c32 100644
--- a/src/r_poly_wall.cpp
+++ b/src/r_poly_wall.cpp
@@ -176,13 +176,13 @@ void RenderPolyWall::Render(const TriMatrix &worldToClip)
 	}
 
 	TriUniforms uniforms;
-	uniforms.objectToClip = worldToClip;
 	uniforms.light = (uint32_t)(GetLightLevel() / 255.0f * 256.0f);
 	uniforms.flags = 0;
 	uniforms.subsectorDepth = SubsectorDepth;
 
 	PolyDrawArgs args;
 	args.uniforms = uniforms;
+	args.objectToClip = &worldToClip;
 	args.vinput = vertices;
 	args.vcount = 4;
 	args.mode = TriangleDrawMode::Fan;
diff --git a/src/r_poly_wallsprite.cpp b/src/r_poly_wallsprite.cpp
index ccebe5f0e..cd331dcb5 100644
--- a/src/r_poly_wallsprite.cpp
+++ b/src/r_poly_wallsprite.cpp
@@ -99,7 +99,6 @@ void RenderPolyWallSprite::Render(const TriMatrix &worldToClip, AActor *thing, s
 	bool fullbrightSprite = ((thing->renderflags & RF_FULLBRIGHT) || (thing->flags5 & MF5_BRIGHT));
 
 	TriUniforms uniforms;
-	uniforms.objectToClip = worldToClip;
 	if (fullbrightSprite || fixedlightlev >= 0 || fixedcolormap)
 	{
 		uniforms.light = 256;
@@ -114,6 +113,7 @@ void RenderPolyWallSprite::Render(const TriMatrix &worldToClip, AActor *thing, s
 
 	PolyDrawArgs args;
 	args.uniforms = uniforms;
+	args.objectToClip = &worldToClip;
 	args.vinput = vertices;
 	args.vcount = 4;
 	args.mode = TriangleDrawMode::Fan;
diff --git a/src/r_poly_wallsprite.h b/src/r_poly_wallsprite.h
index dd4ba28da..66a92b033 100644
--- a/src/r_poly_wallsprite.h
+++ b/src/r_poly_wallsprite.h
@@ -22,7 +22,7 @@
 
 #pragma once
 
-#include "r_triangle.h"
+#include "r_poly_triangle.h"
 
 class RenderPolyWallSprite
 {
diff --git a/src/r_triangle.cpp b/src/r_triangle.cpp
deleted file mode 100644
index 13a8ca292..000000000
--- a/src/r_triangle.cpp
+++ /dev/null
@@ -1,1338 +0,0 @@
-/*
-**  Triangle drawers
-**  Copyright (c) 2016 Magnus Norddahl
-**
-**  This software is provided 'as-is', without any express or implied
-**  warranty.  In no event will the authors be held liable for any damages
-**  arising from the use of this software.
-**
-**  Permission is granted to anyone to use this software for any purpose,
-**  including commercial applications, and to alter it and redistribute it
-**  freely, subject to the following restrictions:
-**
-**  1. The origin of this software must not be misrepresented; you must not
-**     claim that you wrote the original software. If you use this software
-**     in a product, an acknowledgment in the product documentation would be
-**     appreciated but is not required.
-**  2. Altered source versions must be plainly marked as such, and must not be
-**     misrepresented as being the original software.
-**  3. This notice may not be removed or altered from any source distribution.
-**
-*/
-
-#include <stddef.h>
-#include "templates.h"
-#include "doomdef.h"
-#include "i_system.h"
-#include "w_wad.h"
-#include "r_local.h"
-#include "v_video.h"
-#include "doomstat.h"
-#include "st_stuff.h"
-#include "g_game.h"
-#include "g_level.h"
-#include "r_data/r_translate.h"
-#include "v_palette.h"
-#include "r_data/colormaps.h"
-#include "r_triangle.h"
-
-void TriangleDrawer::draw(const TriUniforms &uniforms, const TriVertex *vinput, int vcount, TriangleDrawMode mode, bool ccw, int clipleft, int clipright, const short *cliptop, const short *clipbottom, FTexture *texture)
-{
-	if (r_swtruecolor)
-		queue_arrays(uniforms, vinput, vcount, mode, ccw, clipleft, clipright, cliptop, clipbottom, (const uint8_t*)texture->GetPixelsBgra(), texture->GetWidth(), texture->GetHeight(), 0);
-	else
-		draw_arrays(uniforms, vinput, vcount, mode, ccw, clipleft, clipright, cliptop, clipbottom, texture->GetPixels(), texture->GetWidth(), texture->GetHeight(), 0, nullptr, &ScreenTriangleDrawer::draw);
-}
-
-void TriangleDrawer::fill(const TriUniforms &uniforms, const TriVertex *vinput, int vcount, TriangleDrawMode mode, bool ccw, int clipleft, int clipright, const short *cliptop, const short *clipbottom, int solidcolor)
-{
-	if (r_swtruecolor)
-		queue_arrays(uniforms, vinput, vcount, mode, ccw, clipleft, clipright, cliptop, clipbottom, nullptr, 0, 0, solidcolor);
-	else
-		draw_arrays(uniforms, vinput, vcount, mode, ccw, clipleft, clipright, cliptop, clipbottom, nullptr, 0, 0, solidcolor, nullptr, &ScreenTriangleDrawer::fill);
-}
-
-void TriangleDrawer::queue_arrays(const TriUniforms &uniforms, const TriVertex *vinput, int vcount, TriangleDrawMode mode, bool ccw, int clipleft, int clipright, const short *cliptop, const short *clipbottom, const uint8_t *texturePixels, int textureWidth, int textureHeight, int solidcolor)
-{
-	if (clipright < clipleft || clipleft < 0 || clipright > MAXWIDTH)
-		return;
-
-	int cliplength = clipright - clipleft + 1;
-	short *clipdata = (short*)DrawerCommandQueue::AllocMemory(cliplength * 2 * sizeof(short));
-	if (!clipdata)
-	{
-		DrawerCommandQueue::WaitForWorkers();
-		clipdata = (short*)DrawerCommandQueue::AllocMemory(cliplength * 2 * sizeof(short));
-		if (!clipdata)
-			return;
-	}
-
-	for (int i = 0; i < cliplength; i++)
-		clipdata[i] = cliptop[clipleft + i];
-	for (int i = 0; i < cliplength; i++)
-		clipdata[cliplength + i] = clipbottom[clipleft + i];
-
-	DrawerCommandQueue::QueueCommand<DrawTrianglesCommand>(uniforms, vinput, vcount, mode, ccw, clipleft, clipright, clipdata, texturePixels, textureWidth, textureHeight, solidcolor);
-}
-
-void TriangleDrawer::draw_arrays(const TriUniforms &uniforms, const TriVertex *vinput, int vcount, TriangleDrawMode mode, bool ccw, int clipleft, int clipright, const short *cliptop, const short *clipbottom, const uint8_t *texturePixels, int textureWidth, int textureHeight, int solidcolor, WorkerThreadData *thread, void(*drawfunc)(const ScreenTriangleDrawerArgs *, WorkerThreadData *))
-{
-	if (vcount < 3)
-		return;
-
-	ScreenTriangleDrawerArgs args;
-	args.dest = dc_destorg;
-	args.pitch = dc_pitch;
-	args.clipleft = clipleft;
-	args.clipright = clipright;
-	args.cliptop = cliptop;
-	args.clipbottom = clipbottom;
-	args.texturePixels = texturePixels;
-	args.textureWidth = textureWidth;
-	args.textureHeight = textureHeight;
-	args.solidcolor = solidcolor;
-	args.uniforms = &uniforms;
-
-	TriVertex vert[3];
-	if (mode == TriangleDrawMode::Normal)
-	{
-		for (int i = 0; i < vcount / 3; i++)
-		{
-			for (int j = 0; j < 3; j++)
-				vert[j] = shade_vertex(uniforms, *(vinput++));
-			draw_shaded_triangle(vert, ccw, &args, thread, drawfunc);
-		}
-	}
-	else if (mode == TriangleDrawMode::Fan)
-	{
-		vert[0] = shade_vertex(uniforms, *(vinput++));
-		vert[1] = shade_vertex(uniforms, *(vinput++));
-		for (int i = 2; i < vcount; i++)
-		{
-			vert[2] = shade_vertex(uniforms, *(vinput++));
-			draw_shaded_triangle(vert, ccw, &args, thread, drawfunc);
-			vert[1] = vert[2];
-		}
-	}
-	else // TriangleDrawMode::Strip
-	{
-		vert[0] = shade_vertex(uniforms, *(vinput++));
-		vert[1] = shade_vertex(uniforms, *(vinput++));
-		for (int i = 2; i < vcount; i++)
-		{
-			vert[2] = shade_vertex(uniforms, *(vinput++));
-			draw_shaded_triangle(vert, ccw, &args, thread, drawfunc);
-			vert[0] = vert[1];
-			vert[1] = vert[2];
-			ccw = !ccw;
-		}
-	}
-}
-
-TriVertex TriangleDrawer::shade_vertex(const TriUniforms &uniforms, TriVertex v)
-{
-	// Apply transform to get clip coordinates:
-	return uniforms.objectToClip * v;
-}
-
-void TriangleDrawer::draw_shaded_triangle(const TriVertex *vert, bool ccw, ScreenTriangleDrawerArgs *args, WorkerThreadData *thread, void(*drawfunc)(const ScreenTriangleDrawerArgs *, WorkerThreadData *))
-{
-	// Cull, clip and generate additional vertices as needed
-	TriVertex clippedvert[max_additional_vertices];
-	int numclipvert;
-	clipedge(vert, clippedvert, numclipvert);
-
-	// Map to 2D viewport:
-	for (int j = 0; j < numclipvert; j++)
-	{
-		auto &v = clippedvert[j];
-
-		// Calculate normalized device coordinates:
-		v.w = 1.0f / v.w;
-		v.x *= v.w;
-		v.y *= v.w;
-		v.z *= v.w;
-
-		// Apply viewport scale to get screen coordinates:
-		v.x = viewwidth * (1.0f + v.x) * 0.5f;
-		v.y = viewheight * (1.0f - v.y) * 0.5f;
-	}
-
-	// Draw screen triangles
-	if (ccw)
-	{
-		for (int i = numclipvert; i > 1; i--)
-		{
-			args->v1 = &clippedvert[numclipvert - 1];
-			args->v2 = &clippedvert[i - 1];
-			args->v3 = &clippedvert[i - 2];
-			drawfunc(args, thread);
-		}
-	}
-	else
-	{
-		for (int i = 2; i < numclipvert; i++)
-		{
-			args->v1 = &clippedvert[0];
-			args->v2 = &clippedvert[i - 1];
-			args->v3 = &clippedvert[i];
-			drawfunc(args, thread);
-		}
-	}
-}
-
-bool TriangleDrawer::cullhalfspace(float clipdistance1, float clipdistance2, float &t1, float &t2)
-{
-	if (clipdistance1 < 0.0f && clipdistance2 < 0.0f)
-		return true;
-
-	if (clipdistance1 < 0.0f)
-		t1 = MAX(-clipdistance1 / (clipdistance2 - clipdistance1), 0.0f);
-	else
-		t1 = 0.0f;
-
-	if (clipdistance2 < 0.0f)
-		t2 = MIN(1.0f + clipdistance2 / (clipdistance1 - clipdistance2), 1.0f);
-	else
-		t2 = 1.0f;
-
-	return false;
-}
-
-void TriangleDrawer::clipedge(const TriVertex *verts, TriVertex *clippedvert, int &numclipvert)
-{
-	// Clip and cull so that the following is true for all vertices:
-	// -v.w <= v.x <= v.w
-	// -v.w <= v.y <= v.w
-	// -v.w <= v.z <= v.w
-	
-	// use barycentric weights while clipping vertices
-	float weights[max_additional_vertices * 3 * 2];
-	for (int i = 0; i < 3; i++)
-	{
-		weights[i * 3 + 0] = 0.0f;
-		weights[i * 3 + 1] = 0.0f;
-		weights[i * 3 + 2] = 0.0f;
-		weights[i * 3 + i] = 1.0f;
-	}
-	
-	// halfspace clip distances
-	float clipdistance[6 * 3];
-	for (int i = 0; i < 3; i++)
-	{
-		const auto &v = verts[i];
-		clipdistance[i * 6 + 0] = v.x + v.w;
-		clipdistance[i * 6 + 1] = v.w - v.x;
-		clipdistance[i * 6 + 2] = v.y + v.w;
-		clipdistance[i * 6 + 3] = v.w - v.y;
-		clipdistance[i * 6 + 4] = v.z + v.w;
-		clipdistance[i * 6 + 5] = v.w - v.z;
-	}
-	
-	// Clip against each halfspace
-	float *input = weights;
-	float *output = weights + max_additional_vertices * 3;
-	int inputverts = 3;
-	int outputverts = 0;
-	for (int p = 0; p < 6; p++)
-	{
-		// Clip each edge
-		outputverts = 0;
-		for (int i = 0; i < inputverts; i++)
-		{
-			int j = (i + 1) % inputverts;
-			float clipdistance1 =
-				clipdistance[0 * 6 + p] * input[i * 3 + 0] +
-				clipdistance[1 * 6 + p] * input[i * 3 + 1] +
-				clipdistance[2 * 6 + p] * input[i * 3 + 2];
-
-			float clipdistance2 =
-				clipdistance[0 * 6 + p] * input[j * 3 + 0] +
-				clipdistance[1 * 6 + p] * input[j * 3 + 1] +
-				clipdistance[2 * 6 + p] * input[j * 3 + 2];
-				
-			float t1, t2;
-			if (!cullhalfspace(clipdistance1, clipdistance2, t1, t2) && outputverts + 1 < max_additional_vertices)
-			{
-				// add t1 vertex
-				for (int k = 0; k < 3; k++)
-					output[outputverts * 3 + k] = input[i * 3 + k] * (1.0f - t1) + input[j * 3 + k] * t1;
-				outputverts++;
-			
-				if (t2 != 1.0f && t2 > t1)
-				{
-					// add t2 vertex
-					for (int k = 0; k < 3; k++)
-						output[outputverts * 3 + k] = input[i * 3 + k] * (1.0f - t2) + input[j * 3 + k] * t2;
-					outputverts++;
-				}
-			}
-		}
-		std::swap(input, output);
-		std::swap(inputverts, outputverts);
-		if (inputverts == 0)
-			break;
-	}
-	
-	// Convert barycentric weights to actual vertices
-	numclipvert = inputverts;
-	for (int i = 0; i < numclipvert; i++)
-	{
-		auto &v = clippedvert[i];
-		memset(&v, 0, sizeof(TriVertex));
-		for (int w = 0; w < 3; w++)
-		{
-			float weight = input[i * 3 + w];
-			v.x += verts[w].x * weight;
-			v.y += verts[w].y * weight;
-			v.z += verts[w].z * weight;
-			v.w += verts[w].w * weight;
-			for (int iv = 0; iv < TriVertex::NumVarying; iv++)
-				v.varying[iv] += verts[w].varying[iv] * weight;
-		}
-	}
-}
-
-/////////////////////////////////////////////////////////////////////////////
-
-void ScreenTriangleDrawer::draw(const ScreenTriangleDrawerArgs *args, WorkerThreadData *thread)
-{
-	uint8_t *dest = args->dest;
-	int pitch = args->pitch;
-	const TriVertex &v1 = *args->v1;
-	const TriVertex &v2 = *args->v2;
-	const TriVertex &v3 = *args->v3;
-	int clipleft = args->clipleft;
-	int clipright = args->clipright;
-	const short *cliptop = args->cliptop;
-	const short *clipbottom = args->clipbottom;
-	const uint8_t *texturePixels = args->texturePixels;
-	int textureWidth = args->textureWidth;
-	int textureHeight = args->textureHeight;
-
-	// 28.4 fixed-point coordinates
-	const int Y1 = (int)round(16.0f * v1.y);
-	const int Y2 = (int)round(16.0f * v2.y);
-	const int Y3 = (int)round(16.0f * v3.y);
-
-	const int X1 = (int)round(16.0f * v1.x);
-	const int X2 = (int)round(16.0f * v2.x);
-	const int X3 = (int)round(16.0f * v3.x);
-
-	// Deltas
-	const int DX12 = X1 - X2;
-	const int DX23 = X2 - X3;
-	const int DX31 = X3 - X1;
-
-	const int DY12 = Y1 - Y2;
-	const int DY23 = Y2 - Y3;
-	const int DY31 = Y3 - Y1;
-
-	// Fixed-point deltas
-	const int FDX12 = DX12 << 4;
-	const int FDX23 = DX23 << 4;
-	const int FDX31 = DX31 << 4;
-
-	const int FDY12 = DY12 << 4;
-	const int FDY23 = DY23 << 4;
-	const int FDY31 = DY31 << 4;
-
-	// Bounding rectangle
-	int clipymin = cliptop[clipleft];
-	int clipymax = clipbottom[clipleft];
-	for (int i = clipleft + 1; i <= clipright; i++)
-	{
-		clipymin = MIN(clipymin, (int)cliptop[i]);
-		clipymax = MAX(clipymax, (int)clipbottom[i]);
-	}
-	int minx = MAX((MIN(MIN(X1, X2), X3) + 0xF) >> 4, clipleft);
-	int maxx = MIN((MAX(MAX(X1, X2), X3) + 0xF) >> 4, clipright);
-	int miny = MAX((MIN(MIN(Y1, Y2), Y3) + 0xF) >> 4, clipymin);
-	int maxy = MIN((MAX(MAX(Y1, Y2), Y3) + 0xF) >> 4, clipymax - 1);
-	if (minx >= maxx || miny >= maxy)
-		return;
-
-	// Block size, standard 8x8 (must be power of two)
-	const int q = 8;
-
-	// Start in corner of 8x8 block
-	minx &= ~(q - 1);
-	miny &= ~(q - 1);
-
-	dest += miny * pitch;
-
-	// Half-edge constants
-	int C1 = DY12 * X1 - DX12 * Y1;
-	int C2 = DY23 * X2 - DX23 * Y2;
-	int C3 = DY31 * X3 - DX31 * Y3;
-
-	// Correct for fill convention
-	if (DY12 < 0 || (DY12 == 0 && DX12 > 0)) C1++;
-	if (DY23 < 0 || (DY23 == 0 && DX23 > 0)) C2++;
-	if (DY31 < 0 || (DY31 == 0 && DX31 > 0)) C3++;
-
-	// Gradients
-	float gradWX = gradx(v1.x, v1.y, v2.x, v2.y, v3.x, v3.y, v1.w, v2.w, v3.w);
-	float gradWY = grady(v1.x, v1.y, v2.x, v2.y, v3.x, v3.y, v1.w, v2.w, v3.w);
-	float startW = v1.w + gradWX * (minx - v1.x) + gradWY * (miny - v1.y);
-	float gradVaryingX[TriVertex::NumVarying], gradVaryingY[TriVertex::NumVarying], startVarying[TriVertex::NumVarying];
-	for (int i = 0; i < TriVertex::NumVarying; i++)
-	{
-		gradVaryingX[i] = gradx(v1.x, v1.y, v2.x, v2.y, v3.x, v3.y, v1.varying[i] * v1.w, v2.varying[i] * v2.w, v3.varying[i] * v3.w);
-		gradVaryingY[i] = grady(v1.x, v1.y, v2.x, v2.y, v3.x, v3.y, v1.varying[i] * v1.w, v2.varying[i] * v2.w, v3.varying[i] * v3.w);
-		startVarying[i] = v1.varying[i] * v1.w + gradVaryingX[i] * (minx - v1.x) + gradVaryingY[i] * (miny - v1.y);
-	}
-
-	// Loop through blocks
-	for (int y = miny; y < maxy; y += q)
-	{
-		for (int x = minx; x < maxx; x += q)
-		{
-			// Corners of block
-			int x0 = x << 4;
-			int x1 = (x + q - 1) << 4;
-			int y0 = y << 4;
-			int y1 = (y + q - 1) << 4;
-
-			// Evaluate half-space functions
-			bool a00 = C1 + DX12 * y0 - DY12 * x0 > 0;
-			bool a10 = C1 + DX12 * y0 - DY12 * x1 > 0;
-			bool a01 = C1 + DX12 * y1 - DY12 * x0 > 0;
-			bool a11 = C1 + DX12 * y1 - DY12 * x1 > 0;
-			int a = (a00 << 0) | (a10 << 1) | (a01 << 2) | (a11 << 3);
-
-			bool b00 = C2 + DX23 * y0 - DY23 * x0 > 0;
-			bool b10 = C2 + DX23 * y0 - DY23 * x1 > 0;
-			bool b01 = C2 + DX23 * y1 - DY23 * x0 > 0;
-			bool b11 = C2 + DX23 * y1 - DY23 * x1 > 0;
-			int b = (b00 << 0) | (b10 << 1) | (b01 << 2) | (b11 << 3);
-
-			bool c00 = C3 + DX31 * y0 - DY31 * x0 > 0;
-			bool c10 = C3 + DX31 * y0 - DY31 * x1 > 0;
-			bool c01 = C3 + DX31 * y1 - DY31 * x0 > 0;
-			bool c11 = C3 + DX31 * y1 - DY31 * x1 > 0;
-			int c = (c00 << 0) | (c10 << 1) | (c01 << 2) | (c11 << 3);
-
-			// Skip block when outside an edge
-			if (a == 0x0 || b == 0x0 || c == 0x0) continue;
-
-			// Check if block needs clipping
-			int clipcount = 0;
-			for (int ix = x; ix < x + q; ix++)
-			{
-				clipcount += (clipleft > ix) || (clipright < ix) || (cliptop[ix] > y) || (clipbottom[ix] <= y + q - 1);
-			}
-
-			// Calculate varying variables for affine block
-			float offx0 = (x - minx) + 0.5f;
-			float offy0 = (y - miny) + 0.5f;
-			float offx1 = offx0 + q;
-			float offy1 = offy0 + q;
-			float rcpWTL = 1.0f / (startW + offx0 * gradWX + offy0 * gradWY);
-			float rcpWTR = 1.0f / (startW + offx1 * gradWX + offy0 * gradWY);
-			float rcpWBL = 1.0f / (startW + offx0 * gradWX + offy1 * gradWY);
-			float rcpWBR = 1.0f / (startW + offx1 * gradWX + offy1 * gradWY);
-			float varyingTL[TriVertex::NumVarying];
-			float varyingTR[TriVertex::NumVarying];
-			float varyingBL[TriVertex::NumVarying];
-			float varyingBR[TriVertex::NumVarying];
-			for (int i = 0; i < TriVertex::NumVarying; i++)
-			{
-				varyingTL[i] = (startVarying[i] + offx0 * gradVaryingX[i] + offy0 * gradVaryingY[i]) * rcpWTL;
-				varyingTR[i] = (startVarying[i] + offx1 * gradVaryingX[i] + offy0 * gradVaryingY[i]) * rcpWTR;
-				varyingBL[i] = ((startVarying[i] + offx0 * gradVaryingX[i] + offy1 * gradVaryingY[i]) * rcpWBL - varyingTL[i]) * (1.0f / q);
-				varyingBR[i] = ((startVarying[i] + offx1 * gradVaryingX[i] + offy1 * gradVaryingY[i]) * rcpWBR - varyingTR[i]) * (1.0f / q);
-			}
-
-			uint8_t *buffer = dest;
-
-			// Accept whole block when totally covered
-			if (a == 0xF && b == 0xF && c == 0xF && clipcount == 0)
-			{
-				for (int iy = 0; iy < q; iy++)
-				{
-					uint32_t varying[TriVertex::NumVarying], varyingStep[TriVertex::NumVarying];
-					for (int i = 0; i < TriVertex::NumVarying; i++)
-					{
-						float pos = varyingTL[i] + varyingBL[i] * iy;
-						float step = (varyingTR[i] + varyingBR[i] * iy - pos) * (1.0f / q);
-
-						varying[i] = (uint32_t)((pos - floor(pos)) * 0x100000000LL);
-						varyingStep[i] = (uint32_t)(step * 0x100000000LL);
-					}
-
-					for (int ix = x; ix < x + q; ix++)
-					{
-						uint32_t ufrac = varying[0];
-						uint32_t vfrac = varying[1];
-
-						uint32_t upos = ((ufrac >> 16) * textureWidth) >> 16;
-						uint32_t vpos = ((vfrac >> 16) * textureHeight) >> 16;
-						uint32_t uvoffset = upos * textureHeight + vpos;
-
-						buffer[ix] = texturePixels[uvoffset];
-
-						for (int i = 0; i < TriVertex::NumVarying; i++)
-							varying[i] += varyingStep[i];
-					}
-
-					buffer += pitch;
-				}
-			}
-			else // Partially covered block
-			{
-				int CY1 = C1 + DX12 * y0 - DY12 * x0;
-				int CY2 = C2 + DX23 * y0 - DY23 * x0;
-				int CY3 = C3 + DX31 * y0 - DY31 * x0;
-
-				for (int iy = 0; iy < q; iy++)
-				{
-					int CX1 = CY1;
-					int CX2 = CY2;
-					int CX3 = CY3;
-
-					uint32_t varying[TriVertex::NumVarying], varyingStep[TriVertex::NumVarying];
-					for (int i = 0; i < TriVertex::NumVarying; i++)
-					{
-						float pos = varyingTL[i] + varyingBL[i] * iy;
-						float step = (varyingTR[i] + varyingBR[i] * iy - pos) * (1.0f / q);
-
-						varying[i] = (uint32_t)((pos - floor(pos)) * 0x100000000LL);
-						varyingStep[i] = (uint32_t)(step * 0x100000000LL);
-					}
-
-					for (int ix = x; ix < x + q; ix++)
-					{
-						bool visible = ix >= clipleft && ix <= clipright && (cliptop[ix] <= y + iy) && (clipbottom[ix] > y + iy);
-
-						if (CX1 > 0 && CX2 > 0 && CX3 > 0 && visible)
-						{
-							uint32_t ufrac = varying[0];
-							uint32_t vfrac = varying[1];
-
-							uint32_t upos = ((ufrac >> 16) * textureWidth) >> 16;
-							uint32_t vpos = ((vfrac >> 16) * textureHeight) >> 16;
-							uint32_t uvoffset = upos * textureHeight + vpos;
-
-							buffer[ix] = texturePixels[uvoffset];
-						}
-
-						for (int i = 0; i < TriVertex::NumVarying; i++)
-							varying[i] += varyingStep[i];
-
-						CX1 -= FDY12;
-						CX2 -= FDY23;
-						CX3 -= FDY31;
-					}
-
-					CY1 += FDX12;
-					CY2 += FDX23;
-					CY3 += FDX31;
-
-					buffer += pitch;
-				}
-			}
-		}
-
-		dest += q * pitch;
-	}
-}
-
-void ScreenTriangleDrawer::fill(const ScreenTriangleDrawerArgs *args, WorkerThreadData *thread)
-{
-	uint8_t *dest = args->dest;
-	int pitch = args->pitch;
-	const TriVertex &v1 = *args->v1;
-	const TriVertex &v2 = *args->v2;
-	const TriVertex &v3 = *args->v3;
-	int clipleft = args->clipleft;
-	int clipright = args->clipright;
-	const short *cliptop = args->cliptop;
-	const short *clipbottom = args->clipbottom;
-	int solidcolor = args->solidcolor;
-
-	// 28.4 fixed-point coordinates
-	const int Y1 = (int)round(16.0f * v1.y);
-	const int Y2 = (int)round(16.0f * v2.y);
-	const int Y3 = (int)round(16.0f * v3.y);
-
-	const int X1 = (int)round(16.0f * v1.x);
-	const int X2 = (int)round(16.0f * v2.x);
-	const int X3 = (int)round(16.0f * v3.x);
-
-	// Deltas
-	const int DX12 = X1 - X2;
-	const int DX23 = X2 - X3;
-	const int DX31 = X3 - X1;
-
-	const int DY12 = Y1 - Y2;
-	const int DY23 = Y2 - Y3;
-	const int DY31 = Y3 - Y1;
-
-	// Fixed-point deltas
-	const int FDX12 = DX12 << 4;
-	const int FDX23 = DX23 << 4;
-	const int FDX31 = DX31 << 4;
-
-	const int FDY12 = DY12 << 4;
-	const int FDY23 = DY23 << 4;
-	const int FDY31 = DY31 << 4;
-
-	// Bounding rectangle
-	int clipymin = cliptop[clipleft];
-	int clipymax = clipbottom[clipleft];
-	for (int i = clipleft + 1; i <= clipright; i++)
-	{
-		clipymin = MIN(clipymin, (int)cliptop[i]);
-		clipymax = MAX(clipymax, (int)clipbottom[i]);
-	}
-	int minx = MAX((MIN(MIN(X1, X2), X3) + 0xF) >> 4, clipleft);
-	int maxx = MIN((MAX(MAX(X1, X2), X3) + 0xF) >> 4, clipright);
-	int miny = MAX((MIN(MIN(Y1, Y2), Y3) + 0xF) >> 4, clipymin);
-	int maxy = MIN((MAX(MAX(Y1, Y2), Y3) + 0xF) >> 4, clipymax - 1);
-	if (minx >= maxx || miny >= maxy)
-		return;
-
-	// Block size, standard 8x8 (must be power of two)
-	const int q = 8;
-
-	// Start in corner of 8x8 block
-	minx &= ~(q - 1);
-	miny &= ~(q - 1);
-
-	dest += miny * pitch;
-
-	// Half-edge constants
-	int C1 = DY12 * X1 - DX12 * Y1;
-	int C2 = DY23 * X2 - DX23 * Y2;
-	int C3 = DY31 * X3 - DX31 * Y3;
-
-	// Correct for fill convention
-	if (DY12 < 0 || (DY12 == 0 && DX12 > 0)) C1++;
-	if (DY23 < 0 || (DY23 == 0 && DX23 > 0)) C2++;
-	if (DY31 < 0 || (DY31 == 0 && DX31 > 0)) C3++;
-
-	// Loop through blocks
-	for (int y = miny; y < maxy; y += q)
-	{
-		for (int x = minx; x < maxx; x += q)
-		{
-			// Corners of block
-			int x0 = x << 4;
-			int x1 = (x + q - 1) << 4;
-			int y0 = y << 4;
-			int y1 = (y + q - 1) << 4;
-
-			// Evaluate half-space functions
-			bool a00 = C1 + DX12 * y0 - DY12 * x0 > 0;
-			bool a10 = C1 + DX12 * y0 - DY12 * x1 > 0;
-			bool a01 = C1 + DX12 * y1 - DY12 * x0 > 0;
-			bool a11 = C1 + DX12 * y1 - DY12 * x1 > 0;
-			int a = (a00 << 0) | (a10 << 1) | (a01 << 2) | (a11 << 3);
-
-			bool b00 = C2 + DX23 * y0 - DY23 * x0 > 0;
-			bool b10 = C2 + DX23 * y0 - DY23 * x1 > 0;
-			bool b01 = C2 + DX23 * y1 - DY23 * x0 > 0;
-			bool b11 = C2 + DX23 * y1 - DY23 * x1 > 0;
-			int b = (b00 << 0) | (b10 << 1) | (b01 << 2) | (b11 << 3);
-
-			bool c00 = C3 + DX31 * y0 - DY31 * x0 > 0;
-			bool c10 = C3 + DX31 * y0 - DY31 * x1 > 0;
-			bool c01 = C3 + DX31 * y1 - DY31 * x0 > 0;
-			bool c11 = C3 + DX31 * y1 - DY31 * x1 > 0;
-			int c = (c00 << 0) | (c10 << 1) | (c01 << 2) | (c11 << 3);
-
-			// Skip block when outside an edge
-			if (a == 0x0 || b == 0x0 || c == 0x0) continue;
-
-			// Check if block needs clipping
-			int clipcount = 0;
-			for (int ix = x; ix < x + q; ix++)
-			{
-				clipcount += (clipleft > ix) || (clipright < ix) || (cliptop[ix] > y) || (clipbottom[ix] <= y + q - 1);
-			}
-
-			uint8_t *buffer = dest;
-
-			// Accept whole block when totally covered
-			if (a == 0xF && b == 0xF && c == 0xF && clipcount == 0)
-			{
-				for (int iy = 0; iy < q; iy++)
-				{
-					for (int ix = x; ix < x + q; ix++)
-					{
-						buffer[ix] = solidcolor;
-					}
-
-					buffer += pitch;
-				}
-			}
-			else // Partially covered block
-			{
-				int CY1 = C1 + DX12 * y0 - DY12 * x0;
-				int CY2 = C2 + DX23 * y0 - DY23 * x0;
-				int CY3 = C3 + DX31 * y0 - DY31 * x0;
-
-				for (int iy = 0; iy < q; iy++)
-				{
-					int CX1 = CY1;
-					int CX2 = CY2;
-					int CX3 = CY3;
-
-					for (int ix = x; ix < x + q; ix++)
-					{
-						bool visible = ix >= clipleft && ix <= clipright && (cliptop[ix] <= y + iy) && (clipbottom[ix] > y + iy);
-
-						if (CX1 > 0 && CX2 > 0 && CX3 > 0 && visible)
-						{
-							buffer[ix] = solidcolor;
-						}
-
-						CX1 -= FDY12;
-						CX2 -= FDY23;
-						CX3 -= FDY31;
-					}
-
-					CY1 += FDX12;
-					CY2 += FDX23;
-					CY3 += FDX31;
-
-					buffer += pitch;
-				}
-			}
-		}
-
-		dest += q * pitch;
-	}
-}
-
-void ScreenTriangleDrawer::draw32(const ScreenTriangleDrawerArgs *args, WorkerThreadData *thread)
-{
-	uint32_t *dest = (uint32_t *)args->dest;
-	int pitch = args->pitch;
-	const TriVertex &v1 = *args->v1;
-	const TriVertex &v2 = *args->v2;
-	const TriVertex &v3 = *args->v3;
-	int clipleft = args->clipleft;
-	int clipright = args->clipright;
-	const short *cliptop = args->cliptop;
-	const short *clipbottom = args->clipbottom;
-	const uint32_t *texturePixels = (const uint32_t *)args->texturePixels;
-	int textureWidth = args->textureWidth;
-	int textureHeight = args->textureHeight;
-	uint32_t light = args->uniforms->light;
-
-	// 28.4 fixed-point coordinates
-	const int Y1 = (int)round(16.0f * v1.y);
-	const int Y2 = (int)round(16.0f * v2.y);
-	const int Y3 = (int)round(16.0f * v3.y);
-
-	const int X1 = (int)round(16.0f * v1.x);
-	const int X2 = (int)round(16.0f * v2.x);
-	const int X3 = (int)round(16.0f * v3.x);
-
-	// Deltas
-	const int DX12 = X1 - X2;
-	const int DX23 = X2 - X3;
-	const int DX31 = X3 - X1;
-
-	const int DY12 = Y1 - Y2;
-	const int DY23 = Y2 - Y3;
-	const int DY31 = Y3 - Y1;
-
-	// Fixed-point deltas
-	const int FDX12 = DX12 << 4;
-	const int FDX23 = DX23 << 4;
-	const int FDX31 = DX31 << 4;
-
-	const int FDY12 = DY12 << 4;
-	const int FDY23 = DY23 << 4;
-	const int FDY31 = DY31 << 4;
-
-	// Bounding rectangle
-	int clipymin = cliptop[clipleft];
-	int clipymax = clipbottom[clipleft];
-	for (int i = clipleft + 1; i <= clipright; i++)
-	{
-		clipymin = MIN(clipymin, (int)cliptop[i]);
-		clipymax = MAX(clipymax, (int)clipbottom[i]);
-	}
-	int minx = MAX((MIN(MIN(X1, X2), X3) + 0xF) >> 4, clipleft);
-	int maxx = MIN((MAX(MAX(X1, X2), X3) + 0xF) >> 4, clipright);
-	int miny = MAX((MIN(MIN(Y1, Y2), Y3) + 0xF) >> 4, clipymin);
-	int maxy = MIN((MAX(MAX(Y1, Y2), Y3) + 0xF) >> 4, clipymax - 1);
-	if (minx >= maxx || miny >= maxy)
-		return;
-
-	// Block size, standard 8x8 (must be power of two)
-	const int q = 8;
-
-	// Start in corner of 8x8 block
-	minx &= ~(q - 1);
-	miny &= ~(q - 1);
-
-	dest += miny * pitch;
-
-	// Half-edge constants
-	int C1 = DY12 * X1 - DX12 * Y1;
-	int C2 = DY23 * X2 - DX23 * Y2;
-	int C3 = DY31 * X3 - DX31 * Y3;
-
-	// Correct for fill convention
-	if (DY12 < 0 || (DY12 == 0 && DX12 > 0)) C1++;
-	if (DY23 < 0 || (DY23 == 0 && DX23 > 0)) C2++;
-	if (DY31 < 0 || (DY31 == 0 && DX31 > 0)) C3++;
-
-	// Gradients
-	float gradWX = gradx(v1.x, v1.y, v2.x, v2.y, v3.x, v3.y, v1.w, v2.w, v3.w);
-	float gradWY = grady(v1.x, v1.y, v2.x, v2.y, v3.x, v3.y, v1.w, v2.w, v3.w);
-	float startW = v1.w + gradWX * (minx - v1.x) + gradWY * (miny - v1.y);
-	float gradVaryingX[TriVertex::NumVarying], gradVaryingY[TriVertex::NumVarying], startVarying[TriVertex::NumVarying];
-	for (int i = 0; i < TriVertex::NumVarying; i++)
-	{
-		gradVaryingX[i] = gradx(v1.x, v1.y, v2.x, v2.y, v3.x, v3.y, v1.varying[i] * v1.w, v2.varying[i] * v2.w, v3.varying[i] * v3.w);
-		gradVaryingY[i] = grady(v1.x, v1.y, v2.x, v2.y, v3.x, v3.y, v1.varying[i] * v1.w, v2.varying[i] * v2.w, v3.varying[i] * v3.w);
-		startVarying[i] = v1.varying[i] * v1.w + gradVaryingX[i] * (minx - v1.x) + gradVaryingY[i] * (miny - v1.y);
-	}
-
-	// Loop through blocks
-	for (int y = miny; y < maxy; y += q)
-	{
-		for (int x = minx; x < maxx; x += q)
-		{
-			// Corners of block
-			int x0 = x << 4;
-			int x1 = (x + q - 1) << 4;
-			int y0 = y << 4;
-			int y1 = (y + q - 1) << 4;
-
-			// Evaluate half-space functions
-			bool a00 = C1 + DX12 * y0 - DY12 * x0 > 0;
-			bool a10 = C1 + DX12 * y0 - DY12 * x1 > 0;
-			bool a01 = C1 + DX12 * y1 - DY12 * x0 > 0;
-			bool a11 = C1 + DX12 * y1 - DY12 * x1 > 0;
-			int a = (a00 << 0) | (a10 << 1) | (a01 << 2) | (a11 << 3);
-
-			bool b00 = C2 + DX23 * y0 - DY23 * x0 > 0;
-			bool b10 = C2 + DX23 * y0 - DY23 * x1 > 0;
-			bool b01 = C2 + DX23 * y1 - DY23 * x0 > 0;
-			bool b11 = C2 + DX23 * y1 - DY23 * x1 > 0;
-			int b = (b00 << 0) | (b10 << 1) | (b01 << 2) | (b11 << 3);
-
-			bool c00 = C3 + DX31 * y0 - DY31 * x0 > 0;
-			bool c10 = C3 + DX31 * y0 - DY31 * x1 > 0;
-			bool c01 = C3 + DX31 * y1 - DY31 * x0 > 0;
-			bool c11 = C3 + DX31 * y1 - DY31 * x1 > 0;
-			int c = (c00 << 0) | (c10 << 1) | (c01 << 2) | (c11 << 3);
-
-			// Skip block when outside an edge
-			if (a == 0x0 || b == 0x0 || c == 0x0) continue;
-
-			// Check if block needs clipping
-			int clipcount = 0;
-			for (int ix = x; ix < x + q; ix++)
-			{
-				clipcount += (clipleft > ix) || (clipright < ix) || (cliptop[ix] > y) || (clipbottom[ix] <= y + q - 1);
-			}
-
-			// Calculate varying variables for affine block
-			float offx0 = (x - minx) + 0.5f;
-			float offy0 = (y - miny) + 0.5f;
-			float offx1 = offx0 + q;
-			float offy1 = offy0 + q;
-			float rcpWTL = 1.0f / (startW + offx0 * gradWX + offy0 * gradWY);
-			float rcpWTR = 1.0f / (startW + offx1 * gradWX + offy0 * gradWY);
-			float rcpWBL = 1.0f / (startW + offx0 * gradWX + offy1 * gradWY);
-			float rcpWBR = 1.0f / (startW + offx1 * gradWX + offy1 * gradWY);
-			float varyingTL[TriVertex::NumVarying];
-			float varyingTR[TriVertex::NumVarying];
-			float varyingBL[TriVertex::NumVarying];
-			float varyingBR[TriVertex::NumVarying];
-			for (int i = 0; i < TriVertex::NumVarying; i++)
-			{
-				varyingTL[i] = (startVarying[i] + offx0 * gradVaryingX[i] + offy0 * gradVaryingY[i]) * rcpWTL;
-				varyingTR[i] = (startVarying[i] + offx1 * gradVaryingX[i] + offy0 * gradVaryingY[i]) * rcpWTR;
-				varyingBL[i] = ((startVarying[i] + offx0 * gradVaryingX[i] + offy1 * gradVaryingY[i]) * rcpWBL - varyingTL[i]) * (1.0f / q);
-				varyingBR[i] = ((startVarying[i] + offx1 * gradVaryingX[i] + offy1 * gradVaryingY[i]) * rcpWBR - varyingTR[i]) * (1.0f / q);
-			}
-
-			uint32_t *buffer = dest;
-
-			// Accept whole block when totally covered
-			if (a == 0xF && b == 0xF && c == 0xF && clipcount == 0)
-			{
-				for (int iy = 0; iy < q; iy++)
-				{
-					uint32_t varying[TriVertex::NumVarying], varyingStep[TriVertex::NumVarying];
-					for (int i = 0; i < TriVertex::NumVarying; i++)
-					{
-						float pos = varyingTL[i] + varyingBL[i] * iy;
-						float step = (varyingTR[i] + varyingBR[i] * iy - pos) * (1.0f / q);
-
-						varying[i] = (uint32_t)((pos - floor(pos)) * 0x100000000LL);
-						varyingStep[i] = (uint32_t)(step * 0x100000000LL);
-					}
-
-					if ((y + iy) % thread->num_cores == thread->core)
-					{
-						for (int ix = x; ix < x + q; ix++)
-						{
-							uint32_t ufrac = varying[0];
-							uint32_t vfrac = varying[1];
-
-							uint32_t upos = ((ufrac >> 16) * textureWidth) >> 16;
-							uint32_t vpos = ((vfrac >> 16) * textureHeight) >> 16;
-							uint32_t uvoffset = upos * textureHeight + vpos;
-
-							uint32_t fg = texturePixels[uvoffset];
-							uint32_t fg_red = (RPART(fg) * light) >> 8;
-							uint32_t fg_green = (GPART(fg) * light) >> 8;
-							uint32_t fg_blue = (BPART(fg) * light) >> 8;
-							uint32_t fg_alpha = APART(fg);
-
-							if (fg_alpha > 127)
-								buffer[ix] = 0xff000000 | (fg_red << 16) | (fg_green << 8) | fg_blue;
-
-							for (int i = 0; i < TriVertex::NumVarying; i++)
-								varying[i] += varyingStep[i];
-						}
-					}
-
-					buffer += pitch;
-				}
-			}
-			else // Partially covered block
-			{
-				int CY1 = C1 + DX12 * y0 - DY12 * x0;
-				int CY2 = C2 + DX23 * y0 - DY23 * x0;
-				int CY3 = C3 + DX31 * y0 - DY31 * x0;
-
-				for (int iy = 0; iy < q; iy++)
-				{
-					int CX1 = CY1;
-					int CX2 = CY2;
-					int CX3 = CY3;
-
-					float varying[TriVertex::NumVarying], varyingStep[TriVertex::NumVarying];
-					for (int i = 0; i < TriVertex::NumVarying; i++)
-					{
-						varying[i] = varyingTL[i] + varyingBL[i] * iy;
-						varyingStep[i] = (varyingTR[i] + varyingBR[i] * iy - varying[i]) * (1.0f / q);
-					}
-
-					if ((y + iy) % thread->num_cores == thread->core)
-					{
-						for (int ix = x; ix < x + q; ix++)
-						{
-							bool visible = ix >= clipleft && ix <= clipright && (cliptop[ix] <= y + iy) && (clipbottom[ix] > y + iy);
-
-							if (CX1 > 0 && CX2 > 0 && CX3 > 0 && visible)
-							{
-								uint32_t ufrac = (uint32_t)((varying[0] - floor(varying[0])) * 0x100000000LL);
-								uint32_t vfrac = (uint32_t)((varying[1] - floor(varying[1])) * 0x100000000LL);
-
-								uint32_t upos = ((ufrac >> 16) * textureWidth) >> 16;
-								uint32_t vpos = ((vfrac >> 16) * textureHeight) >> 16;
-								uint32_t uvoffset = upos * textureHeight + vpos;
-
-								uint32_t fg = texturePixels[uvoffset];
-								uint32_t fg_red = (RPART(fg) * light) >> 8;
-								uint32_t fg_green = (GPART(fg) * light) >> 8;
-								uint32_t fg_blue = (BPART(fg) * light) >> 8;
-								uint32_t fg_alpha = APART(fg);
-
-								if (fg_alpha > 127)
-									buffer[ix] = 0xff000000 | (fg_red << 16) | (fg_green << 8) | fg_blue;
-							}
-
-							for (int i = 0; i < TriVertex::NumVarying; i++)
-								varying[i] += varyingStep[i];
-
-							CX1 -= FDY12;
-							CX2 -= FDY23;
-							CX3 -= FDY31;
-						}
-					}
-
-					CY1 += FDX12;
-					CY2 += FDX23;
-					CY3 += FDX31;
-
-					buffer += pitch;
-				}
-			}
-		}
-
-		dest += q * pitch;
-	}
-}
-
-void ScreenTriangleDrawer::fill32(const ScreenTriangleDrawerArgs *args, WorkerThreadData *thread)
-{
-	uint32_t *dest = (uint32_t *)args->dest;
-	int pitch = args->pitch;
-	const TriVertex &v1 = *args->v1;
-	const TriVertex &v2 = *args->v2;
-	const TriVertex &v3 = *args->v3;
-	int clipleft = args->clipleft;
-	int clipright = args->clipright;
-	const short *cliptop = args->cliptop;
-	const short *clipbottom = args->clipbottom;
-	int solidcolor = args->solidcolor;
-
-	// 28.4 fixed-point coordinates
-	const int Y1 = (int)round(16.0f * v1.y);
-	const int Y2 = (int)round(16.0f * v2.y);
-	const int Y3 = (int)round(16.0f * v3.y);
-
-	const int X1 = (int)round(16.0f * v1.x);
-	const int X2 = (int)round(16.0f * v2.x);
-	const int X3 = (int)round(16.0f * v3.x);
-
-	// Deltas
-	const int DX12 = X1 - X2;
-	const int DX23 = X2 - X3;
-	const int DX31 = X3 - X1;
-
-	const int DY12 = Y1 - Y2;
-	const int DY23 = Y2 - Y3;
-	const int DY31 = Y3 - Y1;
-
-	// Fixed-point deltas
-	const int FDX12 = DX12 << 4;
-	const int FDX23 = DX23 << 4;
-	const int FDX31 = DX31 << 4;
-
-	const int FDY12 = DY12 << 4;
-	const int FDY23 = DY23 << 4;
-	const int FDY31 = DY31 << 4;
-
-	// Bounding rectangle
-	int clipymin = cliptop[clipleft];
-	int clipymax = clipbottom[clipleft];
-	for (int i = clipleft + 1; i <= clipright; i++)
-	{
-		clipymin = MIN(clipymin, (int)cliptop[i]);
-		clipymax = MAX(clipymax, (int)clipbottom[i]);
-	}
-	int minx = MAX((MIN(MIN(X1, X2), X3) + 0xF) >> 4, clipleft);
-	int maxx = MIN((MAX(MAX(X1, X2), X3) + 0xF) >> 4, clipright);
-	int miny = MAX((MIN(MIN(Y1, Y2), Y3) + 0xF) >> 4, clipymin);
-	int maxy = MIN((MAX(MAX(Y1, Y2), Y3) + 0xF) >> 4, clipymax - 1);
-	if (minx >= maxx || miny >= maxy)
-		return;
-
-	// Block size, standard 8x8 (must be power of two)
-	const int q = 8;
-
-	// Start in corner of 8x8 block
-	minx &= ~(q - 1);
-	miny &= ~(q - 1);
-
-	dest += miny * pitch;
-
-	// Half-edge constants
-	int C1 = DY12 * X1 - DX12 * Y1;
-	int C2 = DY23 * X2 - DX23 * Y2;
-	int C3 = DY31 * X3 - DX31 * Y3;
-
-	// Correct for fill convention
-	if (DY12 < 0 || (DY12 == 0 && DX12 > 0)) C1++;
-	if (DY23 < 0 || (DY23 == 0 && DX23 > 0)) C2++;
-	if (DY31 < 0 || (DY31 == 0 && DX31 > 0)) C3++;
-
-	// Loop through blocks
-	for (int y = miny; y < maxy; y += q)
-	{
-		for (int x = minx; x < maxx; x += q)
-		{
-			// Corners of block
-			int x0 = x << 4;
-			int x1 = (x + q - 1) << 4;
-			int y0 = y << 4;
-			int y1 = (y + q - 1) << 4;
-
-			// Evaluate half-space functions
-			bool a00 = C1 + DX12 * y0 - DY12 * x0 > 0;
-			bool a10 = C1 + DX12 * y0 - DY12 * x1 > 0;
-			bool a01 = C1 + DX12 * y1 - DY12 * x0 > 0;
-			bool a11 = C1 + DX12 * y1 - DY12 * x1 > 0;
-			int a = (a00 << 0) | (a10 << 1) | (a01 << 2) | (a11 << 3);
-
-			bool b00 = C2 + DX23 * y0 - DY23 * x0 > 0;
-			bool b10 = C2 + DX23 * y0 - DY23 * x1 > 0;
-			bool b01 = C2 + DX23 * y1 - DY23 * x0 > 0;
-			bool b11 = C2 + DX23 * y1 - DY23 * x1 > 0;
-			int b = (b00 << 0) | (b10 << 1) | (b01 << 2) | (b11 << 3);
-
-			bool c00 = C3 + DX31 * y0 - DY31 * x0 > 0;
-			bool c10 = C3 + DX31 * y0 - DY31 * x1 > 0;
-			bool c01 = C3 + DX31 * y1 - DY31 * x0 > 0;
-			bool c11 = C3 + DX31 * y1 - DY31 * x1 > 0;
-			int c = (c00 << 0) | (c10 << 1) | (c01 << 2) | (c11 << 3);
-
-			// Skip block when outside an edge
-			if (a == 0x0 || b == 0x0 || c == 0x0) continue;
-
-			// Check if block needs clipping
-			int clipcount = 0;
-			for (int ix = x; ix < x + q; ix++)
-			{
-				clipcount += (clipleft > ix) || (clipright < ix) || (cliptop[ix] > y) || (clipbottom[ix] <= y + q - 1);
-			}
-
-			uint32_t *buffer = dest;
-
-			// Accept whole block when totally covered
-			if (a == 0xF && b == 0xF && c == 0xF && clipcount == 0)
-			{
-				for (int iy = 0; iy < q; iy++)
-				{
-					if ((y + iy) % thread->num_cores == thread->core)
-					{
-						for (int ix = x; ix < x + q; ix++)
-						{
-							buffer[ix] = solidcolor;
-						}
-					}
-
-					buffer += pitch;
-				}
-			}
-			else // Partially covered block
-			{
-				int CY1 = C1 + DX12 * y0 - DY12 * x0;
-				int CY2 = C2 + DX23 * y0 - DY23 * x0;
-				int CY3 = C3 + DX31 * y0 - DY31 * x0;
-
-				for (int iy = 0; iy < q; iy++)
-				{
-					int CX1 = CY1;
-					int CX2 = CY2;
-					int CX3 = CY3;
-
-					if ((y + iy) % thread->num_cores == thread->core)
-					{
-						for (int ix = x; ix < x + q; ix++)
-						{
-							bool visible = ix >= clipleft && ix <= clipright && (cliptop[ix] <= y + iy) && (clipbottom[ix] > y + iy);
-
-							if (CX1 > 0 && CX2 > 0 && CX3 > 0 && visible)
-							{
-								buffer[ix] = solidcolor;
-							}
-
-							CX1 -= FDY12;
-							CX2 -= FDY23;
-							CX3 -= FDY31;
-						}
-					}
-
-					CY1 += FDX12;
-					CY2 += FDX23;
-					CY3 += FDX31;
-
-					buffer += pitch;
-				}
-			}
-		}
-
-		dest += q * pitch;
-	}
-}
-
-float ScreenTriangleDrawer::gradx(float x0, float y0, float x1, float y1, float x2, float y2, float c0, float c1, float c2)
-{
-	float top = (c1 - c2) * (y0 - y2) - (c0 - c2) * (y1 - y2);
-	float bottom = (x1 - x2) * (y0 - y2) - (x0 - x2) * (y1 - y2);
-	return top / bottom;
-}
-
-float ScreenTriangleDrawer::grady(float x0, float y0, float x1, float y1, float x2, float y2, float c0, float c1, float c2)
-{
-	float top = (c1 - c2) * (x0 - x2) - (c0 - c2) * (x1 - x2);
-	float bottom = -((x1 - x2) * (y0 - y2) - (x0 - x2) * (y1 - y2));
-	return top / bottom;
-}
-
-/////////////////////////////////////////////////////////////////////////////
-
-TriMatrix TriMatrix::null()
-{
-	TriMatrix m;
-	memset(m.matrix, 0, sizeof(m.matrix));
-	return m;
-}
-
-TriMatrix TriMatrix::identity()
-{
-	TriMatrix m = null();
-	m.matrix[0] = 1.0f;
-	m.matrix[5] = 1.0f;
-	m.matrix[10] = 1.0f;
-	m.matrix[15] = 1.0f;
-	return m;
-}
-
-TriMatrix TriMatrix::translate(float x, float y, float z)
-{
-	TriMatrix m = identity();
-	m.matrix[0 + 3 * 4] = x;
-	m.matrix[1 + 3 * 4] = y;
-	m.matrix[2 + 3 * 4] = z;
-	return m;
-}
-
-TriMatrix TriMatrix::scale(float x, float y, float z)
-{
-	TriMatrix m = null();
-	m.matrix[0 + 0 * 4] = x;
-	m.matrix[1 + 1 * 4] = y;
-	m.matrix[2 + 2 * 4] = z;
-	m.matrix[3 + 3 * 4] = 1;
-	return m;
-}
-
-TriMatrix TriMatrix::rotate(float angle, float x, float y, float z)
-{
-	float c = cosf(angle);
-	float s = sinf(angle);
-	TriMatrix m = null();
-	m.matrix[0 + 0 * 4] = (x*x*(1.0f - c) + c);
-	m.matrix[0 + 1 * 4] = (x*y*(1.0f - c) - z*s);
-	m.matrix[0 + 2 * 4] = (x*z*(1.0f - c) + y*s);
-	m.matrix[1 + 0 * 4] = (y*x*(1.0f - c) + z*s);
-	m.matrix[1 + 1 * 4] = (y*y*(1.0f - c) + c);
-	m.matrix[1 + 2 * 4] = (y*z*(1.0f - c) - x*s);
-	m.matrix[2 + 0 * 4] = (x*z*(1.0f - c) - y*s);
-	m.matrix[2 + 1 * 4] = (y*z*(1.0f - c) + x*s);
-	m.matrix[2 + 2 * 4] = (z*z*(1.0f - c) + c);
-	m.matrix[3 + 3 * 4] = 1.0f;
-	return m;
-}
-
-TriMatrix TriMatrix::swapYZ()
-{
-	TriMatrix m = null();
-	m.matrix[0 + 0 * 4] = 1.0f;
-	m.matrix[1 + 2 * 4] = 1.0f;
-	m.matrix[2 + 1 * 4] = -1.0f;
-	m.matrix[3 + 3 * 4] = 1.0f;
-	return m;
-}
-
-TriMatrix TriMatrix::perspective(float fovy, float aspect, float z_near, float z_far)
-{
-	float f = (float)(1.0 / tan(fovy * M_PI / 360.0));
-	TriMatrix m = null();
-	m.matrix[0 + 0 * 4] = f / aspect;
-	m.matrix[1 + 1 * 4] = f;
-	m.matrix[2 + 2 * 4] = (z_far + z_near) / (z_near - z_far);
-	m.matrix[2 + 3 * 4] = (2.0f * z_far * z_near) / (z_near - z_far);
-	m.matrix[3 + 2 * 4] = -1.0f;
-	return m;
-}
-
-TriMatrix TriMatrix::frustum(float left, float right, float bottom, float top, float near, float far)
-{
-	float a = (right + left) / (right - left);
-	float b = (top + bottom) / (top - bottom);
-	float c = -(far + near) / (far - near);
-	float d = -(2.0f * far) / (far - near);
-	TriMatrix m = null();
-	m.matrix[0 + 0 * 4] = 2.0f * near / (right - left);
-	m.matrix[1 + 1 * 4] = 2.0f * near / (top - bottom);
-	m.matrix[0 + 2 * 4] = a;
-	m.matrix[1 + 2 * 4] = b;
-	m.matrix[2 + 2 * 4] = c;
-	m.matrix[2 + 3 * 4] = d;
-	m.matrix[3 + 2 * 4] = -1;
-	return m;
-}
-
-TriMatrix TriMatrix::worldToView()
-{
-	TriMatrix m = null();
-	m.matrix[0 + 0 * 4] = (float)ViewSin;
-	m.matrix[0 + 1 * 4] = (float)-ViewCos;
-	m.matrix[1 + 2 * 4] = 1.0f;
-	m.matrix[2 + 0 * 4] = (float)-ViewCos;
-	m.matrix[2 + 1 * 4] = (float)-ViewSin;
-	m.matrix[3 + 3 * 4] = 1.0f;
-	return m * translate((float)-ViewPos.X, (float)-ViewPos.Y, (float)-ViewPos.Z);
-}
-
-TriMatrix TriMatrix::viewToClip()
-{
-	float near = 5.0f;
-	float far = 65536.0f;
-	float width = (float)(FocalTangent * near);
-	float top = (float)(CenterY / InvZtoScale * near);
-	float bottom = (float)(top - viewheight / InvZtoScale * near);
-	return frustum(-width, width, bottom, top, near, far);
-}
-
-TriMatrix TriMatrix::operator*(const TriMatrix &mult) const
-{
-	TriMatrix result;
-	for (int x = 0; x < 4; x++)
-	{
-		for (int y = 0; y < 4; y++)
-		{
-			result.matrix[x + y * 4] =
-				matrix[0 * 4 + x] * mult.matrix[y * 4 + 0] +
-				matrix[1 * 4 + x] * mult.matrix[y * 4 + 1] +
-				matrix[2 * 4 + x] * mult.matrix[y * 4 + 2] +
-				matrix[3 * 4 + x] * mult.matrix[y * 4 + 3];
-		}
-	}
-	return result;
-}
-
-TriVertex TriMatrix::operator*(TriVertex v) const
-{
-	float vx = matrix[0 * 4 + 0] * v.x + matrix[1 * 4 + 0] * v.y + matrix[2 * 4 + 0] * v.z + matrix[3 * 4 + 0] * v.w;
-	float vy = matrix[0 * 4 + 1] * v.x + matrix[1 * 4 + 1] * v.y + matrix[2 * 4 + 1] * v.z + matrix[3 * 4 + 1] * v.w;
-	float vz = matrix[0 * 4 + 2] * v.x + matrix[1 * 4 + 2] * v.y + matrix[2 * 4 + 2] * v.z + matrix[3 * 4 + 2] * v.w;
-	float vw = matrix[0 * 4 + 3] * v.x + matrix[1 * 4 + 3] * v.y + matrix[2 * 4 + 3] * v.z + matrix[3 * 4 + 3] * v.w;
-	v.x = vx;
-	v.y = vy;
-	v.z = vz;
-	v.w = vw;
-	return v;
-}
-
-/////////////////////////////////////////////////////////////////////////////
-
-DrawTrianglesCommand::DrawTrianglesCommand(const TriUniforms &uniforms, const TriVertex *vinput, int vcount, TriangleDrawMode mode, bool ccw, int clipleft, int clipright, const short *clipdata, const uint8_t *texturePixels, int textureWidth, int textureHeight, int solidcolor)
-	: uniforms(uniforms), vinput(vinput), vcount(vcount), mode(mode), ccw(ccw), clipleft(clipleft), clipright(clipright), clipdata(clipdata), texturePixels(texturePixels), textureWidth(textureWidth), textureHeight(textureHeight), solidcolor(solidcolor)
-{
-}
-
-void DrawTrianglesCommand::Execute(DrawerThread *thread)
-{
-	int cliplength = clipright - clipleft + 1;
-	for (int i = 0; i < cliplength; i++)
-	{
-		thread->triangle_clip_top[clipleft + i] = clipdata[i];
-		thread->triangle_clip_bottom[clipleft + i] = clipdata[cliplength + i];
-	}
-
-	WorkerThreadData thread_data;
-	thread_data.core = thread->core;
-	thread_data.num_cores = thread->num_cores;
-	thread_data.pass_start_y = thread->pass_start_y;
-	thread_data.pass_end_y = thread->pass_end_y;
-	thread_data.temp = thread->dc_temp_rgba;
-
-	TriangleDrawer::draw_arrays(
-		uniforms, vinput, vcount, mode, ccw,
-		clipleft, clipright, thread->triangle_clip_top, thread->triangle_clip_bottom,
-		texturePixels, textureWidth, textureHeight, solidcolor,
-		&thread_data, texturePixels ? ScreenTriangleDrawer::draw32 : ScreenTriangleDrawer::fill32);
-}
-
-FString DrawTrianglesCommand::DebugInfo()
-{
-	return "DrawTriangles";
-}
diff --git a/src/r_triangle.h b/src/r_triangle.h
deleted file mode 100644
index 6ec413d3f..000000000
--- a/src/r_triangle.h
+++ /dev/null
@@ -1,116 +0,0 @@
-/*
-**  Triangle drawers
-**  Copyright (c) 2016 Magnus Norddahl
-**
-**  This software is provided 'as-is', without any express or implied
-**  warranty.  In no event will the authors be held liable for any damages
-**  arising from the use of this software.
-**
-**  Permission is granted to anyone to use this software for any purpose,
-**  including commercial applications, and to alter it and redistribute it
-**  freely, subject to the following restrictions:
-**
-**  1. The origin of this software must not be misrepresented; you must not
-**     claim that you wrote the original software. If you use this software
-**     in a product, an acknowledgment in the product documentation would be
-**     appreciated but is not required.
-**  2. Altered source versions must be plainly marked as such, and must not be
-**     misrepresented as being the original software.
-**  3. This notice may not be removed or altered from any source distribution.
-**
-*/
-
-
-#ifndef __R_TRIANGLE__
-#define __R_TRIANGLE__
-
-#include "r_draw.h"
-#include "r_thread.h"
-#include "r_compiler/llvmdrawers.h"
-
-class FTexture;
-struct ScreenTriangleDrawerArgs;
-
-enum class TriangleDrawMode
-{
-	Normal,
-	Fan,
-	Strip
-};
-
-class TriangleDrawer
-{
-public:
-	static void draw(const TriUniforms &uniforms, const TriVertex *vinput, int vcount, TriangleDrawMode mode, bool ccw, int clipleft, int clipright, const short *cliptop, const short *clipbottom, FTexture *texture);
-	static void fill(const TriUniforms &uniforms, const TriVertex *vinput, int vcount, TriangleDrawMode mode, bool ccw, int clipleft, int clipright, const short *cliptop, const short *clipbottom, int solidcolor);
-
-private:
-	static TriVertex shade_vertex(const TriUniforms &uniforms, TriVertex v);
-	static void draw_arrays(const TriUniforms &uniforms, const TriVertex *vinput, int vcount, TriangleDrawMode mode, bool ccw, int clipleft, int clipright, const short *cliptop, const short *clipbottom, const uint8_t *texturePixels, int textureWidth, int textureHeight, int solidcolor, WorkerThreadData *thread, void(*drawfunc)(const ScreenTriangleDrawerArgs *, WorkerThreadData *));
-	static void draw_shaded_triangle(const TriVertex *vertices, bool ccw, ScreenTriangleDrawerArgs *args, WorkerThreadData *thread, void(*drawfunc)(const ScreenTriangleDrawerArgs *, WorkerThreadData *));
-	static bool cullhalfspace(float clipdistance1, float clipdistance2, float &t1, float &t2);
-	static void clipedge(const TriVertex *verts, TriVertex *clippedvert, int &numclipvert);
-
-	static void queue_arrays(const TriUniforms &uniforms, const TriVertex *vinput, int vcount, TriangleDrawMode mode, bool ccw, int clipleft, int clipright, const short *cliptop, const short *clipbottom, const uint8_t *texturePixels, int textureWidth, int textureHeight, int solidcolor);
-
-	enum { max_additional_vertices = 16 };
-
-	friend class DrawTrianglesCommand;
-};
-
-struct ScreenTriangleDrawerArgs
-{
-	uint8_t *dest;
-	int pitch;
-	TriVertex *v1;
-	TriVertex *v2;
-	TriVertex *v3;
-	int clipleft;
-	int clipright;
-	const short *cliptop;
-	const short *clipbottom;
-	const uint8_t *texturePixels;
-	int textureWidth;
-	int textureHeight;
-	int solidcolor;
-	const TriUniforms *uniforms;
-};
-
-class ScreenTriangleDrawer
-{
-public:
-	static void draw(const ScreenTriangleDrawerArgs *args, WorkerThreadData *thread);
-	static void fill(const ScreenTriangleDrawerArgs *args, WorkerThreadData *thread);
-
-	static void draw32(const ScreenTriangleDrawerArgs *args, WorkerThreadData *thread);
-	static void fill32(const ScreenTriangleDrawerArgs *args, WorkerThreadData *thread);
-
-private:
-	static float gradx(float x0, float y0, float x1, float y1, float x2, float y2, float c0, float c1, float c2);
-	static float grady(float x0, float y0, float x1, float y1, float x2, float y2, float c0, float c1, float c2);
-};
-
-class DrawTrianglesCommand : public DrawerCommand
-{
-public:
-	DrawTrianglesCommand(const TriUniforms &uniforms, const TriVertex *vinput, int vcount, TriangleDrawMode mode, bool ccw, int clipleft, int clipright, const short *clipdata, const uint8_t *texturePixels, int textureWidth, int textureHeight, int solidcolor);
-
-	void Execute(DrawerThread *thread) override;
-	FString DebugInfo() override;
-
-private:
-	const TriUniforms uniforms;
-	const TriVertex *vinput;
-	int vcount;
-	TriangleDrawMode mode;
-	bool ccw;
-	int clipleft;
-	int clipright;
-	const short *clipdata;
-	const uint8_t *texturePixels;
-	int textureWidth;
-	int textureHeight;
-	int solidcolor;
-};
-
-#endif