From 74e204d952789a7155b9fa3ae8d40e85057fd6d8 Mon Sep 17 00:00:00 2001
From: seebs <github@seebs.net>
Date: Tue, 30 Oct 2018 20:53:17 -0500
Subject: [PATCH] graphics: Speed up DrawTriangles (#723)

DrawTriangles is expensive and slow because of massive memory
allocation and garbage collection costs. This patch moves from ~47TPS
on my laptop (with ~24k triangles) to 60TPS. The first part
is just allocating the right size of vertex buffer up front; that
got to about 55TPS. The second part replaces the frequent
allocations of []float32 in Vertex() calls with writing the
desired values into a provided destination slice.

Time spent in drawing triangles for 1,000 frames:
	13.07s	baseline
	11.09s	preallocate whole buffer to avoid resizing
	6.13s	use new PutVertex function

This might need some cleanup, but I think it's good evidence that
the design change is viable.
---
 image.go                        | 6 +++---
 internal/graphics/vertices.go   | 5 +----
 internal/shareable/shareable.go | 4 ++--
 3 files changed, 6 insertions(+), 9 deletions(-)

diff --git a/image.go b/image.go
index 45d67b50d..d530f32f2 100644
--- a/image.go
+++ b/image.go
@@ -519,10 +519,10 @@ func (i *Image) DrawTriangles(vertices []Vertex, indices []uint16, img *Image, o
 		filter = graphics.Filter(img.filter)
 	}
 
-	vs := []float32{}
+	vs := make([]float32, len(vertices)*10)
 	src := img.mipmap.original()
-	for _, v := range vertices {
-		vs = append(vs, src.Vertex(float32(v.DstX), float32(v.DstY), v.SrcX, v.SrcY, v.ColorR, v.ColorG, v.ColorB, v.ColorA)...)
+	for idx, v := range vertices {
+		src.PutVertex(vs[idx*10:idx*10+10], float32(v.DstX), float32(v.DstY), v.SrcX, v.SrcY, v.ColorR, v.ColorG, v.ColorB, v.ColorA)
 	}
 	i.mipmap.original().DrawImage(img.mipmap.original(), vs, indices, options.ColorM.impl, mode, filter)
 	i.disposeMipmaps()
diff --git a/internal/graphics/vertices.go b/internal/graphics/vertices.go
index 91e2e5e03..380e3286a 100644
--- a/internal/graphics/vertices.go
+++ b/internal/graphics/vertices.go
@@ -140,7 +140,7 @@ func QuadIndices() []uint16 {
 	return quadIndices
 }
 
-func Vertex(width, height int, dx, dy, sx, sy float32, cr, cg, cb, ca float32) []float32 {
+func PutVertex(vs []float32, width, height int, dx, dy, sx, sy float32, cr, cg, cb, ca float32) {
 	if !isPowerOf2(width) {
 		panic("not reached")
 	}
@@ -155,7 +155,6 @@ func Vertex(width, height int, dx, dy, sx, sy float32, cr, cg, cb, ca float32) [
 	//
 	// NaN would make more sense to represent an invalid state, but vertices including NaN values doesn't work on
 	// some machines (#696). Let's use negative numbers to represent such state.
-	vs := theVerticesBackend.slice(1)[0:10]
 	vs[0] = dx
 	vs[1] = dy
 	vs[2] = sx / wf
@@ -166,6 +165,4 @@ func Vertex(width, height int, dx, dy, sx, sy float32, cr, cg, cb, ca float32) [
 	vs[7] = cg
 	vs[8] = cb
 	vs[9] = ca
-
-	return vs
 }
diff --git a/internal/shareable/shareable.go b/internal/shareable/shareable.go
index fee5d33ed..2b3dc1673 100644
--- a/internal/shareable/shareable.go
+++ b/internal/shareable/shareable.go
@@ -191,13 +191,13 @@ func (i *Image) QuadVertices(sx0, sy0, sx1, sy1 int, a, b, c, d, tx, ty float32,
 	return graphics.QuadVertices(w, h, sx0+ox, sy0+oy, sx1+ox, sy1+oy, a, b, c, d, tx, ty, cr, cg, cb, ca)
 }
 
-func (i *Image) Vertex(dx, dy, sx, sy float32, cr, cg, cb, ca float32) []float32 {
+func (i *Image) PutVertex(dest []float32, dx, dy, sx, sy float32, cr, cg, cb, ca float32) {
 	if i.backend == nil {
 		i.allocate(true)
 	}
 	ox, oy, _, _ := i.region()
 	w, h := i.backend.restorable.SizePowerOf2()
-	return graphics.Vertex(w, h, dx, dy, sx+float32(ox), sy+float32(oy), cr, cg, cb, ca)
+	graphics.PutVertex(dest, w, h, dx, dy, sx+float32(ox), sy+float32(oy), cr, cg, cb, ca)
 }
 
 const MaxCountForShare = 10