tinyrender5

tinyrender4/main.go

@@ -19,7 +19,7 @@ const (
 	NearClip    = 0.1
 	FarClip     = 5 // Because the head is so small and close
 	ObjectFile  = "head.obj"
-	TextureFile = "head.jpg"
+	TextureFile = "../head.jpg"
 )
 
 func must(err error) {

tinyrender5/.gitignore

@@ -0,0 +1,3 @@
+tinyrender1
+tinyrender4
+render

tinyrender5/animation.sh

@@ -0,0 +1,24 @@
+#!/bin/sh
+
+if [ $# -ne 1 ]; then
+  echo "You must pass the basename for the animation folder"
+  exit 1
+fi
+
+echo "Building"
+go build -o render
+mkdir -p $1
+
+echo "Running"
+
+frame=0
+for x in $(seq -3 0.1 3); do
+  ff=$(printf "%03d" $frame)
+  ./render "-repeat=1" "-xofs=$x" "-zofs=-1.8" "-fov=70" "-p6file=$1/$ff.ppm"
+  frame=$((frame + 1))
+done
+
+echo "Converting animation"
+
+cd $1
+convert -delay 5 -loop 0 *.ppm -resize 256x256 anim.gif

tinyrender5/go.mod

@@ -0,0 +1,3 @@
+module tinyrender4
+
+go 1.22.5

tinyrender5/image.go

@@ -0,0 +1,143 @@
+package main
+
+import (
+	"bytes"
+	"fmt"
+	"image"
+	"image/color"
+	"log"
+	//"math"
+	"strings"
+)
+
+// Convert rgb to uint
+func Col2Uint(r, g, b byte) uint {
+	return (uint(r) << 16) | (uint(g) << 8) | uint(b)
+}
+
+func Color2Uint(col color.Color) uint {
+	r, g, b, _ := col.RGBA()
+	//log.Print(r, g, b)
+	return uint(((r & 0xff00) << 8) | (g & 0xff00) | ((b & 0xff00) >> 8))
+}
+
+// Convert uint to rgb (in that order)
+func Uint2Col(col uint) (byte, byte, byte) {
+	return byte((col >> 16) & 0xFF), byte((col >> 8) & 0xFF), byte(col & 0xFF)
+}
+
+// Color is in ARGB (alpha not used right now)
+type Framebuffer struct {
+	Data    []uint
+	ZBuffer []uint16 //[]float32 //uint16 // Apparently 16 bit z-buffers are used
+	Width   uint
+	Height  uint
+}
+
+// Create a new framebuffer for the given width and height.
+func NewFramebuffer(width uint, height uint) Framebuffer {
+	return Framebuffer{
+		Data:    make([]uint, width*height),
+		ZBuffer: make([]uint16, width*height),
+		Width:   width,
+		Height:  height,
+	}
+}
+
+func NewTexture(texture image.Image, skip int) Framebuffer {
+	bounds := texture.Bounds()
+	width := bounds.Dx() / skip
+	height := bounds.Dy() / skip
+	result := Framebuffer{
+		Data:   make([]uint, width*height),
+		Width:  uint(width),
+		Height: uint(height),
+	}
+	for y := bounds.Min.Y; y < bounds.Max.Y; y += skip {
+		for x := bounds.Min.X; x < bounds.Max.X; x += skip {
+			col := texture.At(x, y)
+			result.Set(uint(x/skip), uint(y/skip), Color2Uint(col))
+		}
+	}
+	return result
+}
+
+// Fill zbuffer with pixels that are max distance away
+func (fb *Framebuffer) ResetZBuffer() {
+	for i := range fb.ZBuffer {
+		fb.ZBuffer[i] = 65535 //math.MaxFloat32
+	}
+}
+
+func (fb *Framebuffer) GetUv(u float32, v float32) uint {
+	// log.Print(u, v)
+	x := uint(float32(fb.Width) * u)
+	y := uint(float32(fb.Height) * (1 - v))
+	return fb.Data[x+y*fb.Width]
+}
+
+// Sure hope this gets inlined...
+func (fb *Framebuffer) Set(x uint, y uint, color uint) {
+	fb.Data[x+y*fb.Width] = color
+}
+
+func (fb *Framebuffer) SetSafe(x uint, y uint, color uint) {
+	if x >= fb.Width || y >= fb.Height {
+		return
+	}
+	fb.Data[x+y*fb.Width] = color
+}
+
+// Given some image data, return a string that is the ppm of it
+func (fb *Framebuffer) ExportPPM() string {
+	log.Printf("ExportPPM called for framebuffer %dx%d", fb.Width, fb.Height)
+	var result strings.Builder
+	result.WriteString(fmt.Sprintf("P3\n%d %d\n255\n", fb.Width, fb.Height))
+	for y := range fb.Height {
+		for x := range fb.Width {
+			r, g, b := Uint2Col(fb.Data[x+y*fb.Width])
+			result.WriteString(fmt.Sprintf("%d %d %d\t", r, g, b))
+		}
+		result.WriteRune('\n')
+	}
+	return result.String()
+}
+
+func (fb *Framebuffer) ExportPPMP6() []byte {
+	log.Printf("ExportPPM6 called for framebuffer %dx%d", fb.Width, fb.Height)
+	var result bytes.Buffer
+	result.WriteString(fmt.Sprintf("P6\n%d %d\n255\n", fb.Width, fb.Height))
+	for i := range fb.Data {
+		r, g, b := Uint2Col(fb.Data[i])
+		result.Write([]byte{r, g, b})
+		//result.WriteString(fmt.Sprintf("%d %d %d\t", r, g, b))
+	}
+	//result.WriteRune('\n')
+	return result.Bytes()
+}
+
+/*func (fb *Framebuffer) ZBuffer_ExportPPM() string {
+	var result strings.Builder
+	mini := float32(math.MaxFloat32)
+	maxi := float32(-math.MaxFloat32)
+	for _, f := range fb.ZBuffer {
+		if f == math.MaxFloat32 {
+			continue
+		}
+		mini = min(f, mini)
+		maxi = max(f, maxi)
+	}
+	result.WriteString(fmt.Sprintf("P2\n%d %d\n255\n", fb.Width, fb.Height))
+	for y := range fb.Height {
+		for x := range fb.Width {
+			if fb.ZBuffer[x+y*fb.Width] == math.MaxFloat32 {
+				result.WriteString("0 ")
+			} else {
+				zp := byte(math.Abs(float64(255 * fb.ZBuffer[x+y*fb.Width] / (maxi - mini))))
+				result.WriteString(fmt.Sprintf("%d ", zp))
+			}
+		}
+		result.WriteRune('\n')
+	}
+	return result.String()
+}*/

tinyrender5/main.go

@@ -0,0 +1,138 @@
+package main
+
+import (
+	"flag"
+	"fmt"
+	"image"
+	"log"
+	//"math"
+	//"math/rand"
+	"os"
+	"runtime/pprof" // For performance profiling (unnecessary)
+
+	_ "image/jpeg"
+)
+
+const (
+	Width       = 512
+	Height      = 512
+	NearClip    = 0.1
+	FarClip     = 100
+	ObjectFile  = "head.obj"
+	TextureFile = "../head.jpg"
+)
+
+func must(err error) {
+	if err != nil {
+		panic(err)
+	}
+}
+
+// However flag works... idk
+var cpuprofile = flag.String("cpuprofile", "", "write cpu profile to file")
+var dozbuf = flag.Bool("zbuffer", false, "Write zbuffer instead of image")
+var p6file = flag.String("p6file", "", "Output binary ppm to given file instead")
+var fov = flag.Float64("fov", 90, "Horizontal FOV in degrees")
+var xofs = flag.Float64("xofs", 0, "Offset image by x")
+var zofs = flag.Float64("zofs", -1.5, "Offset image by z (should be negative)")
+var repeat = flag.Int("repeat", 60, "Amount of times to repeat render")
+
+// var zcuthigh = flag.Float64("zcuthigh", math.MaxFloat32, "High cutoff for z (values above this will be removed)")
+// var zcutlow = flag.Float64("zcutlow", -math.MaxFloat32, "Low cutoff for z (values below are removed)")
+
+func main() {
+	log.Printf("Program start")
+
+	// Little section for doing cpu profiling. I guess that's all you have to do?
+	flag.Parse()
+	if *cpuprofile != "" {
+		log.Printf("CPU profiling requested, write to %s", *cpuprofile)
+		f, err := os.Create(*cpuprofile)
+		must(err)
+		defer f.Close()
+		err = pprof.StartCPUProfile(f)
+		must(err)
+		defer pprof.StopCPUProfile()
+	}
+
+	fb := NewFramebuffer(Width, Height)
+
+	log.Printf("Loading obj %s, texture %s", ObjectFile, TextureFile)
+
+	of, err := os.Open(ObjectFile)
+	must(err)
+	defer of.Close()
+	o, err := ParseObj(of)
+	must(err)
+
+	jf, err := os.Open(TextureFile)
+	must(err)
+	defer jf.Close()
+	timg, _, err := image.Decode(jf)
+	must(err)
+	texture := NewTexture(timg, 4)
+
+	log.Printf("Running render")
+
+	light := Vec3f{0, 0, -1}
+
+	var projection Mat44f
+	var worldToCamera Mat44f
+	var viewport Mat44f
+	projection.SetProjection(float32(*fov), NearClip, FarClip)
+	worldToCamera.SetTranslation(float32(*xofs), 0, float32(*zofs))
+	viewport.SetViewportSimple(int(fb.Width), int(fb.Height), 65535)
+
+	// Premultiply all the translation/etc matrices. Why do we do world to camera THEN
+	// projection? I guess that makes sense actually, oops... projection is the last step.
+	screenmat := worldToCamera.Multiply(&projection)
+	screenmat = screenmat.Multiply(&viewport)
+	// light = worldToCamera.MultiplyPoint3(light)
+	// light = light.Normalize()
+
+	var sc [3]Vertex
+	// var hi = float32(fb.Height - 1)
+	for range *repeat {
+		fb.ResetZBuffer()
+		for _, f := range o.Faces {
+			// Precompute perspective for vertices to save time. Notice Z
+			// is not considered: is this orthographic projection? Yeah probably...
+			var fpt [3]Vec3f
+			for i := range 3 { // Triangles, bro
+				//fp := screenmat.MultiplyPoint3(f[i].Pos)
+				sc[i] = f[i]
+				sc[i].Pos = screenmat.MultiplyPoint3(f[i].Pos)
+				fpt[i] = worldToCamera.MultiplyPoint3(f[i].Pos)
+				// sc[i].Pos.X = (fp.X + 1) * halfwidth
+				// sc[i].Pos.Y = hi - (fp.Y+1)*halfheight
+				// sc[i].Pos.Z = fp.Z
+			}
+
+			l1 := fpt[2].Sub(fpt[0])
+			n := l1.CrossProduct(fpt[1].Sub(fpt[0]))
+			n = n.Normalize()
+			intensity := n.MultSimp(&light)
+			// intensity := float32(1)
+			if intensity > 0 {
+				Triangle3t(&fb, &texture, intensity, sc[0], sc[1], sc[2])
+			}
+		}
+	}
+
+	// log.Print(fb.ZBuffer)
+
+	if *dozbuf {
+		log.Printf("Exporting zbuffer ppm to stdout")
+		//fmt.Print(fb.ZBuffer_ExportPPM())
+	} else {
+		if *p6file != "" {
+			err := os.WriteFile(*p6file, fb.ExportPPMP6(), 0660)
+			must(err)
+		} else {
+			log.Printf("Exporting ppm to stdout")
+			fmt.Print(fb.ExportPPM())
+		}
+	}
+
+	log.Printf("Program end")
+}

tinyrender5/obj.go

@@ -0,0 +1,260 @@
+package main
+
+// This reads obj files?
+import (
+	"bufio"
+	"fmt"
+	"io"
+	"log"
+	"math"
+	"strings"
+)
+
+type Vec3f struct {
+	X, Y, Z float32
+}
+
+type Vec2i struct {
+	X, Y int
+}
+
+type Vec2f struct {
+	X, Y float32
+}
+
+// A ROW MAJOR matrix
+type Mat44f [16]float32
+
+func (m *Mat44f) Set(x int, y int, val float32) {
+	m[x+y*4] = val
+}
+func (m *Mat44f) Get(x int, y int) float32 {
+	return m[x+y*4]
+}
+func (m *Mat44f) ZeroFill() {
+	for i := range m {
+		m[i] = 0
+	}
+}
+func (m *Mat44f) SetIdentity() {
+	m.ZeroFill()
+	for i := range 4 {
+		m.Set(i, i, 1)
+	}
+}
+
+// NOTE: we use "Set" instead of "Create" for all these so we reuse the matrix
+// instead of creating a new one all the time (garbage collection)
+
+// Compute the projection matrix, filling the given matrix. FOV is in degrees
+func (m *Mat44f) SetProjection(fov float32, near float32, far float32) {
+	// Projection matrix is (ROW MAJOR!)
+	// S    0    0    0
+	// 0    S    0    0
+	// 0    0 -f/(f-n) -1
+	// 0    0 -fn/(f-n) 0
+	// where S (scale) is 1 / tan(fov / 2) (assuming fov is radians)
+	// NOTE: -1 there is actually -1/c, where c is distance from viewer to
+	// projection plane. We fix it at 1 for now but...
+	m.ZeroFill()
+	scale := float32(1 / math.Tan(float64(fov)*0.5*math.Pi/180))
+	m.Set(0, 0, scale)
+	m.Set(1, 1, scale)
+	m.Set(2, 2, -far/(far-near))
+	m.Set(3, 2, -1)
+	m.Set(2, 3, -far*near/(far-near))
+}
+
+func (m *Mat44f) SetTranslation(x, y, z float32) {
+	m.SetIdentity()
+	m.Set(0, 3, x) // Let user decide how to offset x
+	m.Set(1, 3, y) // Let user decide how to offset x
+	m.Set(2, 3, z) // Get farther away from the face (user)
+}
+
+func (m *Mat44f) SetViewport(tl Vec3f, br Vec3f) { //width, height, depth int) {
+	m.ZeroFill()
+	m.Set(0, 0, (br.X-tl.X)/2)
+	m.Set(1, 1, (tl.Y-br.Y)/2) // Inverted because screen funny
+	m.Set(2, 2, (br.Z-tl.Z)/2)
+	m.Set(3, 3, 1)
+	m.Set(0, 3, (br.X+tl.X)/2)
+	m.Set(1, 3, (br.Y+tl.Y)/2)
+	m.Set(2, 3, (br.Z+tl.Z)/2)
+}
+
+func (m *Mat44f) SetViewportSimple(width, height, depth int) {
+	var tl Vec3f // All zero
+	br := Vec3f{
+		X: float32(width),
+		Y: float32(height),
+		Z: float32(depth),
+	}
+	m.SetViewport(tl, br)
+}
+
+// Multiply the given point by our vector. Remember this is row-major order
+func (m *Mat44f) MultiplyPoint3(p Vec3f) Vec3f {
+	var out Vec3f
+	// We hope very much that Go will optimize the function calls for us,
+	// along with computing the constants.
+	out.X = p.X*m.Get(0, 0) + p.Y*m.Get(0, 1) + p.Z*m.Get(0, 2) + m.Get(0, 3)
+	out.Y = p.X*m.Get(1, 0) + p.Y*m.Get(1, 1) + p.Z*m.Get(1, 2) + m.Get(1, 3)
+	out.Z = p.X*m.Get(2, 0) + p.Y*m.Get(2, 1) + p.Z*m.Get(2, 2) + m.Get(2, 3)
+	w := p.X*m.Get(3, 0) + p.Y*m.Get(3, 1) + p.Z*m.Get(3, 2) + m.Get(3, 3)
+	if w != 1 {
+		out.X /= w
+		out.Y /= w
+		out.Z /= w
+	}
+	return out
+}
+
+// Multiply two 4x4 matrices together (not optimized)
+func (m *Mat44f) Multiply(m2 *Mat44f) Mat44f {
+	var result Mat44f
+	// This is the x and y of our resulting matrix
+	for y := 0; y < 4; y++ {
+		for x := 0; x < 4; x++ {
+			for i := 0; i < 4; i++ {
+				result[x+y*4] += m[i+y*4] * m2[x+i*4]
+			}
+		}
+	}
+	return result
+}
+
+// A single vertex generally has multiple items associated with it
+// when it's part of a face.
+type Vertex struct {
+	Pos Vec3f
+	Tex Vec3f
+}
+
+type Facef [3]Vertex
+
+// struct {
+// 	Vertices      [3]Vec3f
+// 	TextureCoords [3]Vec2i
+// }
+
+type ObjModel struct {
+	Vertices []Vec3f
+	VTexture []Vec3f
+	Faces    []Facef
+}
+
+func (vi *Vec2i) ToF() Vec2f {
+	return Vec2f{float32(vi.X), float32(vi.Y)}
+}
+
+func (vi *Vec3f) ToVec2i() Vec2i {
+	return Vec2i{int(vi.X), int(vi.Y)}
+}
+
+func (v0 *Vec3f) Sub(v1 Vec3f) Vec3f {
+	return Vec3f{
+		X: v0.X - v1.X,
+		Y: v0.Y - v1.Y,
+		Z: v0.Z - v1.Z,
+	}
+}
+
+func (v0 *Vec3f) CrossProduct(v1 Vec3f) Vec3f {
+	return Vec3f{
+		X: v0.Y*v1.Z - v0.Z*v1.Y,
+		Y: v0.Z*v1.X - v0.X*v1.Z,
+		Z: v0.X*v1.Y - v0.Y*v1.X,
+	}
+}
+
+//func (v
+
+func (v *Vec3f) Normalize() Vec3f {
+	l := float32(math.Sqrt(float64(v.MultSimp(v))))
+	return Vec3f{
+		X: v.X / l,
+		Y: v.Y / l,
+		Z: v.Z / l,
+	}
+}
+
+func (v0 *Vec3f) MultSimp(v1 *Vec3f) float32 {
+	return v0.X*v1.X + v0.Y*v1.Y + v0.Z*v1.Z
+}
+
+// Parse an obj file at the given reader. Only handles v and f right now
+func ParseObj(reader io.Reader) (*ObjModel, error) {
+	result := ObjModel{
+		Vertices: make([]Vec3f, 0),
+		Faces:    make([]Facef, 0),
+	}
+	breader := bufio.NewReader(reader)
+	done := false
+	for !done {
+		// Scan a line
+		line, err := breader.ReadString('\n')
+		if err != nil {
+			if err == io.EOF {
+				done = true
+			} else {
+				log.Printf("NOT EOF ERR?")
+				return nil, err
+			}
+		}
+		line = strings.Trim(line, " \t\n\r")
+		if len(line) == 0 {
+			continue
+		}
+		// Find the first "item", whatever that is. This also gets rid of comments
+		// since we just don't use lines that start with # (no handler
+		var t string
+		_, err = fmt.Sscan(line, &t)
+		if err != nil {
+			log.Printf("SSCANF ERR")
+			return nil, err
+		}
+		line = line[len(t):]
+		if t == "v" {
+			// Read a vertex, should be just three floats
+			var v Vec3f
+			_, err := fmt.Sscan(line, &v.X, &v.Y, &v.Z)
+			if err != nil {
+				return nil, err
+			}
+			result.Vertices = append(result.Vertices, v)
+		} else if t == "vt" {
+			// Read a vertex tex coord, should be just three floats too
+			var vt Vec3f
+			_, err := fmt.Sscan(line, &vt.X, &vt.Y, &vt.Z)
+			if err != nil {
+				return nil, err
+			}
+			result.VTexture = append(result.VTexture, vt)
+		} else if t == "f" {
+			// Read a face; in our example, it's always three sets.
+			// For THIS example, we throw away those other values
+			var face Facef
+			var vi [3]int
+			var vti [3]int
+			var ti int
+			_, err := fmt.Sscanf(line, "%d/%d/%d %d/%d/%d %d/%d/%d",
+				&vi[0], &vti[0], &ti, &vi[1], &vti[1], &ti, &vi[2], &vti[2], &ti)
+			if err != nil {
+				return nil, err
+			}
+			for i := range 3 {
+				if vi[i] > len(result.Vertices) || vi[i] < 1 {
+					return nil, fmt.Errorf("Face vertex index out of bounds: %d", vi[i])
+				}
+				face[i].Pos = result.Vertices[vi[i]-1]
+				if vti[i] > len(result.VTexture) || vti[i] < 1 {
+					return nil, fmt.Errorf("Face vertex texture index out of bounds: %d", vti[i])
+				}
+				face[i].Tex = result.VTexture[vti[i]-1]
+			}
+			result.Faces = append(result.Faces, face)
+		}
+	}
+	return &result, nil
+}

tinyrender5/render.go

@@ -0,0 +1,169 @@
+package main
+
+import (
+// "log"
+)
+
+// Figure out the minimum bounding box for a triangle defined by
+// these vertices. Returns the top left and bottom right points,
+// inclusive
+func ComputeBoundingBox(v0, v1, v2 Vec2i) (Vec2i, Vec2i) {
+	return Vec2i{min(v0.X, v1.X, v2.X), min(v0.Y, v1.Y, v2.Y)},
+		Vec2i{max(v0.X, v1.X, v2.X), max(v0.Y, v1.Y, v2.Y)}
+}
+
+// The generic edge function, returning positive if P is on the right side of
+// the line drawn between v1 and v2. This is counter clockwise
+func EdgeFunction(v1, v2, p Vec2f) float32 {
+	return (p.X-v1.X)*(v2.Y-v1.Y) - (p.Y-v1.Y)*(v2.X-v1.X)
+}
+
+// This computes the x and y per-pixel increment for the line going
+// between v1 and v2 (also counter clockwise)
+func EdgeIncrement(v1, v2 Vec2f) (float32, float32) {
+	return (v2.Y - v1.Y), -(v2.X - v1.X)
+}
+
+// The generic edge function, returning positive if P is on the right side of
+// the line drawn between v1 and v2. This is counter clockwise
+func EdgeFunctioni(v1, v2, p Vec2i) int {
+	return (p.X-v1.X)*(v2.Y-v1.Y) - (p.Y-v1.Y)*(v2.X-v1.X)
+}
+
+// This computes the x and y per-pixel increment for the line going
+// between v1 and v2 (also counter clockwise)
+func EdgeIncrementi(v1, v2 Vec2i) (int, int) {
+	return (v2.Y - v1.Y), -(v2.X - v1.X)
+}
+
+func Triangle3(fb *Framebuffer, color uint, v0f Vec3f, v1f Vec3f, v2f Vec3f) {
+	v0 := v0f.ToVec2i()
+	v1 := v1f.ToVec2i()
+	v2 := v2f.ToVec2i()
+	boundsTL, boundsBR := ComputeBoundingBox(v0, v1, v2)
+	if boundsTL.Y < 0 {
+		boundsTL.Y = 0
+	}
+	if boundsTL.X < 0 {
+		boundsTL.X = 0
+	}
+	if boundsBR.Y >= int(fb.Height) {
+		boundsBR.Y = int(fb.Height - 1)
+	}
+	if boundsBR.X >= int(fb.Width) {
+		boundsBR.X = int(fb.Width - 1)
+	}
+	// Where to start our scanning
+	pstart := Vec2i{boundsTL.X, boundsTL.Y}
+	parea := EdgeFunctioni(v0, v1, v2)
+	// if parea < 0 {
+	// 	v1, v2 = v2, v1
+	// 	v1f, v2f = v2f, v1f
+	// 	parea = EdgeFunctioni(v0, v1, v2)
+	// }
+	invarea := 1 / float32(parea)
+	w0_y := EdgeFunctioni(v1, v2, pstart)
+	w1_y := EdgeFunctioni(v2, v0, pstart)
+	w2_y := EdgeFunctioni(v0, v1, pstart)
+	w0_xi, w0_yi := EdgeIncrementi(v1, v2)
+	w1_xi, w1_yi := EdgeIncrementi(v2, v0)
+	w2_xi, w2_yi := EdgeIncrementi(v0, v1)
+
+	for y := uint(boundsTL.Y); y <= uint(boundsBR.Y); y++ {
+		w0 := w0_y
+		w1 := w1_y
+		w2 := w2_y
+		for x := uint(boundsTL.X); x <= uint(boundsBR.X); x++ {
+			if (w0 | w1 | w2) >= 0 {
+				//fb.Data[di] = color
+				//done = true
+				w0a := float32(w0) * invarea
+				w1a := float32(w1) * invarea
+				w2a := float32(w2) * invarea
+				pz := uint16(w0a*v0f.Z + w1a*v1f.Z + w2a*v2f.Z)
+				if pz < fb.ZBuffer[x+y*fb.Width] {
+					//log.Print(pz)
+					fb.ZBuffer[x+y*fb.Width] = pz
+					fb.Set(x, y, color)
+				}
+				// fb.Set(x, y, Col2Uint(byte(255*w0a), byte(255*w1a), byte(255*w2a)))
+			}
+			w0 += w0_xi
+			w1 += w1_xi
+			w2 += w2_xi
+		}
+		w0_y += w0_yi
+		w1_y += w1_yi
+		w2_y += w2_yi
+	}
+}
+
+func Triangle3t(fb *Framebuffer, texture *Framebuffer, intensity float32, v0v Vertex, v1v Vertex, v2v Vertex) {
+	v0 := v0v.Pos.ToVec2i()
+	v1 := v1v.Pos.ToVec2i()
+	v2 := v2v.Pos.ToVec2i()
+	boundsTL, boundsBR := ComputeBoundingBox(v0, v1, v2)
+	if boundsBR.Y < 0 || boundsBR.X < 0 || boundsTL.X >= int(fb.Width) || boundsTL.Y >= int(fb.Height) {
+		return
+	}
+	parea := EdgeFunctioni(v0, v1, v2)
+	if parea == 0 {
+		return
+	}
+	if boundsTL.Y < 0 {
+		boundsTL.Y = 0
+	}
+	if boundsTL.X < 0 {
+		boundsTL.X = 0
+	}
+	if boundsBR.Y >= int(fb.Height) {
+		boundsBR.Y = int(fb.Height - 1)
+	}
+	if boundsBR.X >= int(fb.Width) {
+		boundsBR.X = int(fb.Width - 1)
+	}
+	// Where to start our scanning
+	pstart := Vec2i{boundsTL.X, boundsTL.Y}
+	// if parea < 0 {
+	// 	v1, v2 = v2, v1
+	// 	v1f, v2f = v2f, v1f
+	// 	parea = EdgeFunctioni(v0, v1, v2)
+	// }
+	invarea := 1 / float32(parea)
+	w0_y := EdgeFunctioni(v1, v2, pstart)
+	w1_y := EdgeFunctioni(v2, v0, pstart)
+	w2_y := EdgeFunctioni(v0, v1, pstart)
+	w0_xi, w0_yi := EdgeIncrementi(v1, v2)
+	w1_xi, w1_yi := EdgeIncrementi(v2, v0)
+	w2_xi, w2_yi := EdgeIncrementi(v0, v1)
+
+	for y := uint(boundsTL.Y); y <= uint(boundsBR.Y); y++ {
+		w0 := w0_y
+		w1 := w1_y
+		w2 := w2_y
+		for x := uint(boundsTL.X); x <= uint(boundsBR.X); x++ {
+			if (w0 | w1 | w2) >= 0 {
+				w0a := float32(w0) * invarea
+				w1a := float32(w1) * invarea
+				w2a := float32(w2) * invarea
+				pz := uint16(w0a*v0v.Pos.Z + w1a*v1v.Pos.Z + w2a*v2v.Pos.Z)
+				if pz < fb.ZBuffer[x+y*fb.Width] {
+					fb.ZBuffer[x+y*fb.Width] = pz
+					// if math.IsNaN(v0v.Tex.X) || math.IsNaN(v1v.Tex.X) || math.IsNaN
+					col := texture.GetUv(
+						(w0a*v0v.Tex.X + w1a*v1v.Tex.X + w2a*v2v.Tex.X),
+						(w0a*v0v.Tex.Y + w1a*v1v.Tex.Y + w2a*v2v.Tex.Y),
+					)
+					r, g, b := Uint2Col(col)
+					fb.Set(x, y, Col2Uint(byte(float32(r)*intensity), byte(float32(g)*intensity), byte(float32(b)*intensity))) //uint(texture.Bounds().Dx())
+				}
+			}
+			w0 += w0_xi
+			w1 += w1_xi
+			w2 += w2_xi
+		}
+		w0_y += w0_yi
+		w1_y += w1_yi
+		w2_y += w2_yi
+	}
+}

tinyrender5/render_less.go

@@ -0,0 +1,177 @@
+package main
+
+import (
+	//"log"
+	"math"
+)
+
+func Bresenham2(fb *Framebuffer, color uint, x0 int, y0 int, x1 int, y1 int) {
+	dx := int(math.Abs(float64(x1 - x0)))
+	sx := -1
+	if x0 < x1 {
+		sx = 1
+	}
+	dy := -int(math.Abs(float64(y1 - y0)))
+	sy := -1
+	if y0 < y1 {
+		sy = 1
+	}
+	err := dx + dy
+
+	for {
+		fb.SetSafe(uint(x0), uint(y0), color)
+		if x0 == x1 && y0 == y1 {
+			break
+		}
+		e2 := 2 * err
+		if e2 >= dy {
+			if x0 == x1 {
+				break
+			}
+			err += dy
+			x0 += sx
+		}
+		if e2 <= dx {
+			if y0 == y1 {
+				break
+			}
+			err += dx
+			y0 += sy
+		}
+	}
+}
+
+func line(fb *Framebuffer, color uint, v0 Vec2i, v1 Vec2i) {
+	Bresenham2(fb, color, v0.X, v0.Y, v1.X, v1.Y)
+}
+
+/*func LineSweep(fb *Framebuffer, color uint, v0 Vec2i, v1 Vec2i, v2 Vec2i) {
+
+}*/
+
+func Triangle1(fb *Framebuffer, color uint, v0 Vec2i, v1 Vec2i, v2 Vec2i) {
+	// The dude gets rid of "degenerate" triangles so... we do too?
+	if v2.Y == v1.Y && v1.Y == v0.Y {
+		return
+	}
+	// Very silly manual sorting by Y
+	if v2.Y < v0.Y {
+		v0, v2 = v2, v0
+	}
+	if v1.Y < v0.Y {
+		v0, v1 = v1, v0
+	}
+	if v2.Y < v1.Y {
+		v1, v2 = v2, v1
+	}
+
+	var v02step, v01step, v12step, xlong, xshort float32
+
+	xlong = float32(v0.X)
+	xshort = xlong
+
+	// The first and last Y CAN'T be equal because sorting!!
+	if v1.Y == v0.Y {
+		xshort = float32(v1.X)
+	}
+
+	// We can check just for greater than because we sorted the vertices
+	// Assume 02 is on the right(?)  and 01 on the left
+	v02step = (float32(v2.X - v0.X)) / (float32(v2.Y-v0.Y) + 0.001) // long side always
+	v01step = (float32(v1.X - v0.X)) / (float32(v1.Y-v0.Y) + 0.001) // first short side
+	v12step = (float32(v2.X - v1.X)) / (float32(v2.Y-v1.Y) + 0.001) // second short side
+
+	for y := v0.Y; y <= v2.Y; y++ {
+		xleft := int(xshort)
+		xright := int(xlong)
+		if xleft > xright {
+			xleft, xright = xright, xleft
+		}
+		if xleft < 0 || xright >= int(fb.Width) {
+			continue
+		}
+		// Draw a horizontal line from left to right
+		for x := xleft; x <= xright; x++ {
+			fb.SetSafe(uint(x), uint(y), color)
+		}
+		xlong += v02step
+		if y < v1.Y {
+			xshort += v01step
+		} else {
+			xshort += v12step
+		}
+	}
+}
+
+// How does this work? Compare with your
+// other barycentric function (in a different repo). In the original
+// cpp code, they used an overloaded operator ^ to mean cross product
+func Barycentric(v0, v1, v2, p Vec2i) Vec3f {
+	// WARN: Just not doing this one
+	u := Vec3f{}
+	if math.Abs(float64(u.Z)) < 1 {
+		return Vec3f{-1, 1, 1}
+	}
+	return Vec3f{1 - (u.X+u.Y)/u.Z, u.Y / u.Z, u.X / u.Z}
+}
+
+func Triangle2(fb *Framebuffer, color uint, v0 Vec2i, v1 Vec2i, v2 Vec2i) {
+	boundsTL, boundsBR := ComputeBoundingBox(v0, v1, v2)
+	if boundsTL.Y < 0 {
+		boundsTL.Y = 0
+	}
+	if boundsTL.X < 0 {
+		boundsTL.X = 0
+	}
+	if boundsBR.Y >= int(fb.Height) {
+		boundsBR.Y = int(fb.Height - 1)
+	}
+	if boundsBR.X >= int(fb.Width) {
+		boundsBR.X = int(fb.Width - 1)
+	}
+	// Where to start our scanning
+	pstart := Vec2i{boundsTL.X, boundsTL.Y}
+	//log.Print(boundsTL, boundsBR)
+	// v0f := v0.ToF()
+	// v1f := v1.ToF()
+	// v2f := v2.ToF()
+	// parea := EdgeFunction(v0f, v1f, v2f)
+	// invarea := 1 / parea
+	w0_y := EdgeFunctioni(v1, v2, pstart)
+	w1_y := EdgeFunctioni(v2, v0, pstart)
+	w2_y := EdgeFunctioni(v0, v1, pstart)
+	w0_xi, w0_yi := EdgeIncrementi(v1, v2)
+	w1_xi, w1_yi := EdgeIncrementi(v2, v0)
+	w2_xi, w2_yi := EdgeIncrementi(v0, v1)
+	//dyi := int(fb.Width)
+	//dy := boundsTL.X + dyi*boundsTL.Y
+
+	for y := uint(boundsTL.Y); y <= uint(boundsBR.Y); y++ {
+		w0 := w0_y
+		w1 := w1_y
+		w2 := w2_y
+		//di := dy
+		//done := false
+		for x := uint(boundsTL.X); x <= uint(boundsBR.X); x++ {
+			if (w0 | w1 | w2) >= 0 {
+				//fb.Data[di] = color
+				fb.Set(x, y, color)
+				//done = true
+				// w0a := w0 * invarea
+				// w1a := w1 * invarea
+				// w2a := w2 * invarea
+				// fb.Set(x, y, Col2Uint(byte(255*w0a), byte(255*w1a), byte(255*w2a)))
+			} /*else if done {
+				break
+			}*/
+			//di += 1
+			w0 += w0_xi
+			w1 += w1_xi
+			w2 += w2_xi
+		}
+		//dy += dyi
+		w0_y += w0_yi
+		w1_y += w1_yi
+		w2_y += w2_yi
+	}
+}

tinyrender5/run.sh

@@ -0,0 +1,14 @@
+#!/bin/sh
+
+set -e
+
+if [ $# -ne 1 ]; then
+  echo "You must pass the basename for the .prof and .ppm"
+  exit 1
+fi
+
+echo "Building"
+go build -o render
+echo "Running"
+./render "-cpuprofile=$1.prof" "-p6file=$1.ppm" "-repeat=500"
+# ./render "-zbuffer" >"$1_zbuffer.ppm"

tinyrender5/slices.sh

@@ -0,0 +1,22 @@
+#!/bin/sh
+
+if [ $# -ne 1 ]; then
+  echo "You must pass the basename for the slices"
+  exit 1
+fi
+
+echo "Building"
+go build -o render
+mkdir -p $1
+
+frame=0
+for x in $(seq -0.7 0.05 0.7); do
+  ff=$(printf "%03d" $frame)
+  ./render "-p6file=$1/$ff.ppm" "-zcuthigh=$x"
+  frame=$((frame + 1))
+done
+
+echo "Converting animation"
+
+cd $1
+convert -delay 10 -loop 0 *.ppm -resize 256x256 anim.gif