Avoid heap allocating in GetTextRuns+Perf notes+pprof by default

.gitignore

@@ -14,3 +14,4 @@
 # Dependency directories (remove the comment below to include it)
 # vendor/
 *.png
+*.cpu

glyphs/glyphs.go

@@ -31,6 +31,7 @@ type GlyphRend struct {
 	GlyphMesh    *meshes.Mesh
 	InstancedBuf buffers.Buffer
 	GlyphMat     *materials.Material
+	TextRunsBuf  []TextRun
 
 	GlyphCount uint32
 	//NOTE: Because of the sad realities (bugs?) of CGO, passing an array in a struct
@@ -60,7 +61,8 @@ func (gr *GlyphRend) DrawTextOpenGL01(text string, screenPos *gglm.Vec3, color *
 //Color is RGBA in the range [0,1].
 func (gr *GlyphRend) DrawTextOpenGLAbs(text string, screenPos *gglm.Vec3, color *gglm.Vec4) {
 
-	runs := gr.GetTextRuns(text)
+	runs := gr.TextRunsBuf[:]
+	gr.GetTextRuns(text, &runs)
 	if runs == nil {
 		return
 	}
@@ -180,19 +182,20 @@ type TextRun struct {
 	IsLtr bool
 }
 
-func (gr *GlyphRend) GetTextRuns(t string) []TextRun {
+func (gr *GlyphRend) GetTextRuns(t string, textRunsBuf *[]TextRun) {
 
-	//PERF: Might be better to pass a []TextRun buffer to avoid allocating on the heap
+	if len(t) == 0 {
-	rs := []rune(t)
+		return
-
-	if len(rs) == 0 {
-		return nil
 	}
 
-	runs := make([]TextRun, 0, 10)
+	rs := []rune(t)
+
+	runs := textRunsBuf
 	currRunScript := RuneInfos[rs[0]].ScriptTable
 
 	//TODO: We need to detect neutral characters through BiDi category, not being in common
+	//TODO: Diacritics go into things like 'Category_Mn' and don't necessairly follow the parent script (e.g. Arabic diacritics are NOT in unicode.Arabic).
+	//They should be part of the same run but right now we split them into their own run.
 	runStartIndex := 0
 	for i := 1; i < len(rs); i++ {
 
@@ -216,9 +219,9 @@ func (gr *GlyphRend) GetTextRuns(t string) []TextRun {
 		//If we have a run without trailing neutrals or had a run of just neutrals (e.g. starting sentence with spaces)
 		//then the full run is added, otherwise we slice the run to put neturals in a separate run
 		if trailingCommonsCount == 0 || len(newRunRunes) == trailingCommonsCount {
-			runs = append(runs, TextRun{Runes: newRunRunes})
+			*runs = append(*runs, TextRun{Runes: newRunRunes})
 		} else {
-			runs = append(runs,
+			*runs = append(*runs,
 				TextRun{Runes: newRunRunes[:len(newRunRunes)-trailingCommonsCount]}, TextRun{Runes: newRunRunes[len(newRunRunes)-trailingCommonsCount:]})
 		}
 
@@ -227,12 +230,12 @@ func (gr *GlyphRend) GetTextRuns(t string) []TextRun {
 		currRunScript = ri.ScriptTable
 	}
 
-	runs = append(runs, TextRun{Runes: rs[runStartIndex:]})
+	*runs = append(*runs, TextRun{Runes: rs[runStartIndex:]})
 
 	//Detect directionality of each run
-	for i := 0; i < len(runs); i++ {
+	for i := 0; i < len(*runs); i++ {
 
-		run := &runs[i]
+		run := &(*runs)[i]
 		bidiCat := BidiCategory_L
 		for _, r := range run.Runes {
 			if !unicode.Is(unicode.Common, r) {
@@ -242,12 +245,12 @@ func (gr *GlyphRend) GetTextRuns(t string) []TextRun {
 		}
 		run.IsLtr = !(bidiCat == BidiCategory_R || bidiCat == BidiCategory_AL || bidiCat == BidiCategory_RLE || bidiCat == BidiCategory_RLO || bidiCat == BidiCategory_RLI || bidiCat == BidiCategory_RLM)
 	}
-
-	return runs
 }
 
 func (gr *GlyphRend) glyphFromRunes(curr, prev, next rune) FontAtlasGlyph {
 
+	//PERF: Map access times are absolute garbage to the point that ~85%+ of the runtime of this func
+	//is spent reading from maps :). Using nSet or fMap or similar would be a lot better.
 	type PosCtx int
 	const (
 		PosCtx_start PosCtx = iota
@@ -438,6 +441,7 @@ func NewGlyphRend(fontFile string, fontOptions *truetype.Options, screenWidth, s
 	gr := &GlyphRend{
 		GlyphCount:   0,
 		GlyphVBO:     make([]float32, floatsPerGlyph*MaxGlyphsPerBatch),
+		TextRunsBuf:  make([]TextRun, 0, 20),
 		SpacesPerTab: 4,
 	}
 

imgui.ini

@@ -1,5 +1,5 @@
 [Window][Debug##Default]
 Pos=814,53
-Size=396,114
+Size=399,134
 Collapsed=0
 

main.go

@@ -2,7 +2,10 @@ package main
 
 import (
 	"fmt"
+	"math"
 	"math/rand"
+	"os"
+	"runtime/pprof"
 
 	"github.com/bloeys/gglm/gglm"
 	"github.com/bloeys/nmage/engine"
@@ -10,6 +13,7 @@ import (
 	"github.com/bloeys/nmage/materials"
 	"github.com/bloeys/nmage/meshes"
 	"github.com/bloeys/nmage/renderer/rend3dgl"
+	"github.com/bloeys/nmage/timing"
 	nmageimgui "github.com/bloeys/nmage/ui/imgui"
 	"github.com/bloeys/nterm/glyphs"
 	"github.com/golang/freetype/truetype"
@@ -74,12 +78,11 @@ func main() {
 	//Don't flash white
 	p.win.SDLWin.GLSwap()
 
-	// var pf, _ = os.Create("pprof.cpu")
+	var pf, _ = os.Create("pprof.cpu")
-	// defer pf.Close()
+	defer pf.Close()
-
+	pprof.StartCPUProfile(pf)
-	// pprof.StartCPUProfile(pf)
 	engine.Run(p, p.win, p.imguiInfo)
-	// pprof.StopCPUProfile()
+	pprof.StopCPUProfile()
 }
 
 func (p *program) Init() {
@@ -164,7 +167,8 @@ func (p *program) Update() {
 	imgui.InputText("", &textToShow)
 
 	if imgui.Button("Print Runs") {
-		runs := p.GlyphRend.GetTextRuns(textToShow)
+		runs := make([]glyphs.TextRun, 0, 20)
+		p.GlyphRend.GetTextRuns(textToShow, &runs)
 		for _, run := range runs {
 			fmt.Printf("%s; runes: %#x\n\n", string(run.Runes), run.Runes)
 		}
@@ -179,9 +183,12 @@ func (p *program) Update() {
 			p.GlyphRend.GlyphMat.SetUnifInt32("drawBounds", 0)
 		}
 	}
+
+	imgui.Checkbox("Draw many", &drawManyLines)
 }
 
 var isDrawingBounds = false
+var drawManyLines = false
 var textToShow = " Hello there يا friend. اسمي عمر wow!"
 
 var xOff float32 = 0
@@ -215,23 +222,22 @@ func (p *program) Render() {
 		b = 0
 	}
 
-	textColor := gglm.NewVec4(r, g, b, 1)
 	str := textToShow
-	// str := "مرحبا بك my friend"
+	charCount := len([]rune(str))
-	// str := "my, friend"
+	fps := int(timing.GetAvgFPS())
-	// str := " hello there يا friend. أسمي عمر wow"
+	textColor := gglm.NewVec4(r, g, b, 1)
-	// str := " ijojo\n\n Hello there, friend|. pq?\n ABCDEFG\tHIJKLMNOPQRSTUVWXYZ\nمرحبا بك"
+	if drawManyLines {
-	// str := " ijojo\n\n Hello there, friend|. pq?\n ABCDEFG\tHIJKLMNOPQRSTUVWXYZ"
 
+		const charsPerFrame = 500_000
+		for i := 0; i < charsPerFrame/charCount; i++ {
 			p.GlyphRend.DrawTextOpenGLAbs(str, gglm.NewVec3(xOff, float32(p.GlyphRend.Atlas.LineHeight)*5+yOff, 0), textColor)
-
+		}
-	// strLen := len(str)
+		p.win.SDLWin.SetTitle(fmt.Sprint("FPS: ", fps, " Draws/f: ", math.Ceil(charsPerFrame/glyphs.MaxGlyphsPerBatch), " chars/f: ", charsPerFrame, " chars/s: ", fps*charsPerFrame))
-	// const charsPerFrame = 100_000
+	} else {
-	// for i := 0; i < charsPerFrame/strLen; i++ {
+		charsPerFrame := float64(charCount)
-	// 	p.GlyphRend.DrawTextOpenGLAbs(str, gglm.NewVec3(xOff, float32(p.GlyphRend.Atlas.LineHeight)*8+yOff, 0), textColor)
+		p.GlyphRend.DrawTextOpenGLAbs(str, gglm.NewVec3(xOff, float32(p.GlyphRend.Atlas.LineHeight)*5+yOff, 0), textColor)
-	// }
+		p.win.SDLWin.SetTitle(fmt.Sprint("FPS: ", fps, " Draws/f: ", math.Ceil(charsPerFrame/glyphs.MaxGlyphsPerBatch), " chars/f: ", charsPerFrame, " chars/s: ", fps*int(charsPerFrame)))
-	// fps := int(timing.GetAvgFPS())
+	}
-	// p.win.SDLWin.SetTitle(fmt.Sprint("FPS: ", fps, " Draws/f: ", math.Ceil(charsPerFrame/glyphs.MaxGlyphsPerBatch), " chars/f: ", charsPerFrame, " chars/s: ", fps*charsPerFrame))
 
 }