Add texture batching with TDD coverage at enqueue and prepare seams.

Opt-in batch_group regrouping cuts texture binds; prepare_draw_batches and draw_list tests lock the contract so end_frame cannot silently drop regroup.

engine/app.odin

@@ -16,8 +16,15 @@ SPRITE_VERTS_SIZE :: SPRITE_VERT_COUNT * size_of(Vertex)
 VERTEX_BUFFER_SIZE :: MAX_SPRITES * SPRITE_VERTS_SIZE
 
 Queued_Sprite :: struct {
+	texture:     ^sdl.GPUTexture,
+	verts:       [SPRITE_VERT_COUNT]Vertex,
+	batch_group: u32, // 0 = strict order; nonzero = caller permits regrouping
+}
+
+Draw_Batch :: struct {
+	start:   int,
+	count:   int,
 	texture: ^sdl.GPUTexture,
-	verts:   [SPRITE_VERT_COUNT]Vertex,
 }
 
 App :: struct {
@@ -65,7 +72,7 @@ init :: proc(app: ^App, title: cstring, width, height: i32) -> bool {
 	}
 
 	requested: sdl.GPUShaderFormat = {.SPIRV, .DXIL, .MSL}
-	app.device = sdl.CreateGPUDevice(requested, true, nil)
+	app.device = sdl.CreateGPUDevice(requested, false, nil)
 	if app.device == nil {
 		fmt.eprintfln("CreateGPUDevice failed: %s", sdl.GetError())
 		return false
@@ -76,6 +83,10 @@ init :: proc(app: ^App, title: cstring, width, height: i32) -> bool {
 		return false
 	}
 
+	if !sdl.SetGPUAllowedFramesInFlight(app.device, 3) {
+		fmt.eprintfln("SetGPUAllowedFramesInFlight failed: %s", sdl.GetError())
+	}
+
 	// Prefer uncapped present for profiling; fall back if unsupported.
 	present := sdl.GPUPresentMode.VSYNC
 	if sdl.WindowSupportsGPUPresentMode(app.device, app.window, .IMMEDIATE) {
@@ -242,7 +253,11 @@ end_frame :: proc(app: ^App) {
 	}
 
 	n := len(app.draw_list)
+	batches: [dynamic]Draw_Batch
+	defer delete(batches)
 	if n > 0 {
+		prepare_draw_batches(app.draw_list[:], &batches)
+
 		map_ptr := sdl.MapGPUTransferBuffer(app.device, app.transfer_buffer, false)
 		if map_ptr == nil {
 			fmt.eprintfln("MapGPUTransferBuffer failed: %s", sdl.GetError())
@@ -286,26 +301,20 @@ end_frame :: proc(app: ^App) {
 	app.render_pass = sdl.BeginGPURenderPass(cmd, &color_info, 1, nil)
 	sdl.BindGPUGraphicsPipeline(app.render_pass, app.pipeline)
 
-	i := 0
-	for i < n {
-		run := texture_run_len(app.draw_list[:], i)
-		q0 := app.draw_list[i]
-
+	for batch in batches {
 		sampler_binding := sdl.GPUTextureSamplerBinding {
-			texture = q0.texture,
+			texture = batch.texture,
 			sampler = app.sampler,
 		}
 		sdl.BindGPUFragmentSamplers(app.render_pass, 0, &sampler_binding, 1)
 
 		vb_binding := sdl.GPUBufferBinding {
 			buffer = app.vertex_buffer,
-			offset = u32(i * SPRITE_VERTS_SIZE),
+			offset = u32(batch.start * SPRITE_VERTS_SIZE),
 		}
 		sdl.BindGPUVertexBuffers(app.render_pass, 0, &vb_binding, 1)
 
-		sdl.DrawGPUPrimitives(app.render_pass, u32(run * SPRITE_VERT_COUNT), 1, 0, 0)
-
-		i += run
+		sdl.DrawGPUPrimitives(app.render_pass, u32(batch.count * SPRITE_VERT_COUNT), 1, 0, 0)
 	}
 	sdl.EndGPURenderPass(app.render_pass)
 	app.render_pass = nil
@@ -454,3 +463,60 @@ texture_run_len :: proc(list: []Queued_Sprite, start: int) -> int {
 
 	return n
 }
+
+texture_run_count :: proc(list: []Queued_Sprite) -> int {
+	if len(list) == 0 do return 0
+	count := 0
+	i := 0
+	for i < len(list) {
+		run := texture_run_len(list, i)
+		count += 1
+		i += run
+	}
+	return count
+}
+
+prepare_draw_batches :: proc(list: []Queued_Sprite, batches: ^[dynamic]Draw_Batch) {
+	clear(batches)
+	group_texture_runs(list)
+	i := 0
+	for i < len(list) {
+		run := texture_run_len(list, i)
+		append(batches, Draw_Batch{start = i, count = run, texture = list[i].texture})
+		i += run
+	}
+}
+
+// Within each contiguous nonzero batch_group, stably sort by texture so
+// consecutive same-texture sprites become one draw. Group 0 and group
+// boundaries are never crossed.
+group_texture_runs :: proc(list: []Queued_Sprite) {
+	start := 0
+	for start < len(list) {
+		group := list[start].batch_group
+		if group == 0 {
+			start += 1
+			continue
+		}
+
+		end := start + 1
+		for end < len(list) && list[end].batch_group == group {
+			end += 1
+		}
+
+		// Stable insertion sort is sufficient while MAX_SPRITES is 128.
+		for i in start + 1 ..< end {
+			item := list[i]
+			j := i
+			for j > start {
+				if uintptr(list[j - 1].texture) <= uintptr(item.texture) {
+					break
+				}
+				list[j] = list[j - 1]
+				j -= 1
+			}
+			list[j] = item
+		}
+		start = end
+	}
+}

engine/batching_test.odin

@@ -7,6 +7,18 @@ fake_tex :: proc(id: uintptr) -> ^sdl.GPUTexture {
 	return cast(^sdl.GPUTexture)id
 }
 
+queued :: proc(tex_id: uintptr, group: u32, marker: f32) -> Queued_Sprite {
+	q: Queued_Sprite
+	q.texture = fake_tex(tex_id)
+	q.batch_group = group
+	q.verts[0].pos = {marker, 0}
+	return q
+}
+
+marker_of :: proc(q: Queued_Sprite) -> f32 {
+	return q.verts[0].pos.x
+}
+
 @(test)
 texture_run_len_empty_or_oob :: proc(t: ^testing.T) {
 	testing.expect_value(t, texture_run_len(nil, 0), 0)
@@ -62,3 +74,293 @@ texture_run_len_all_different :: proc(t: ^testing.T) {
 	testing.expect_value(t, texture_run_len(list, 1), 1)
 	testing.expect_value(t, texture_run_len(list, 2), 1)
 }
+
+@(test)
+group_texture_runs_strict_order_unchanged :: proc(t: ^testing.T) {
+	// Group 0 alternates textures; sorting must not reorder (alpha order).
+	list := []Queued_Sprite {
+		queued(2, 0, 1),
+		queued(1, 0, 2),
+		queued(2, 0, 3),
+		queued(1, 0, 4),
+	}
+	before_runs := texture_run_count(list)
+	group_texture_runs(list)
+	testing.expect_value(t, texture_run_count(list), before_runs)
+	testing.expect_value(t, marker_of(list[0]), f32(1))
+	testing.expect_value(t, marker_of(list[1]), f32(2))
+	testing.expect_value(t, marker_of(list[2]), f32(3))
+	testing.expect_value(t, marker_of(list[3]), f32(4))
+}
+
+@(test)
+group_texture_runs_reduces_runs_inside_group :: proc(t: ^testing.T) {
+	list := []Queued_Sprite {
+		queued(2, 1, 1),
+		queued(1, 1, 2),
+		queued(2, 1, 3),
+		queued(1, 1, 4),
+	}
+	testing.expect_value(t, texture_run_count(list), 4)
+	group_texture_runs(list)
+	testing.expect_value(t, texture_run_count(list), 2)
+	testing.expect(t, list[0].texture == fake_tex(1), "lower texture pointer first")
+	testing.expect(t, list[1].texture == fake_tex(1), "same texture run")
+	testing.expect(t, list[2].texture == fake_tex(2), "second texture run")
+	testing.expect(t, list[3].texture == fake_tex(2), "second texture run cont")
+}
+
+@(test)
+group_texture_runs_stable_same_texture :: proc(t: ^testing.T) {
+	list := []Queued_Sprite {
+		queued(2, 1, 10),
+		queued(1, 1, 20),
+		queued(2, 1, 30),
+		queued(1, 1, 40),
+	}
+	group_texture_runs(list)
+	// Same-texture relative order preserved (stable sort).
+	testing.expect_value(t, marker_of(list[0]), f32(20))
+	testing.expect_value(t, marker_of(list[1]), f32(40))
+	testing.expect_value(t, marker_of(list[2]), f32(10))
+	testing.expect_value(t, marker_of(list[3]), f32(30))
+}
+
+@(test)
+group_texture_runs_respects_group_boundaries :: proc(t: ^testing.T) {
+	list := []Queued_Sprite {
+		queued(2, 1, 1),
+		queued(1, 1, 2),
+		queued(2, 0, 3), // strict barrier
+		queued(1, 2, 4),
+		queued(2, 2, 5),
+	}
+	group_texture_runs(list)
+	testing.expect_value(t, marker_of(list[2]), f32(3)) // barrier stays put
+	testing.expect(t, list[0].texture == fake_tex(1))
+	testing.expect(t, list[1].texture == fake_tex(2))
+	testing.expect(t, list[3].texture == fake_tex(1))
+	testing.expect(t, list[4].texture == fake_tex(2))
+	testing.expect_value(t, list[0].batch_group, u32(1))
+	testing.expect_value(t, list[1].batch_group, u32(1))
+	testing.expect_value(t, list[2].batch_group, u32(0))
+	testing.expect_value(t, list[3].batch_group, u32(2))
+	testing.expect_value(t, list[4].batch_group, u32(2))
+}
+
+@(test)
+group_texture_runs_does_not_merge_across_different_groups :: proc(t: ^testing.T) {
+	// Adjacent nonzero groups with different ids must not merge runs across.
+	list := []Queued_Sprite {
+		queued(1, 1, 1),
+		queued(2, 1, 2),
+		queued(1, 2, 3),
+		queued(2, 2, 4),
+	}
+	group_texture_runs(list)
+	testing.expect_value(t, texture_run_count(list), 4)
+	testing.expect_value(t, list[0].batch_group, u32(1))
+	testing.expect_value(t, list[1].batch_group, u32(1))
+	testing.expect_value(t, list[2].batch_group, u32(2))
+	testing.expect_value(t, list[3].batch_group, u32(2))
+}
+
+@(test)
+group_texture_runs_empty :: proc(t: ^testing.T) {
+	list := []Queued_Sprite{}
+	testing.expect_value(t, texture_run_count(list), 0)
+	group_texture_runs(list)
+	testing.expect_value(t, texture_run_count(list), 0)
+}
+
+@(test)
+group_texture_runs_already_optimal :: proc(t: ^testing.T) {
+	list := []Queued_Sprite {
+		queued(1, 1, 10),
+		queued(1, 1, 20),
+		queued(2, 1, 30),
+		queued(2, 1, 40),
+	}
+	testing.expect_value(t, texture_run_count(list), 2)
+	group_texture_runs(list)
+	testing.expect_value(t, texture_run_count(list), 2)
+	testing.expect_value(t, marker_of(list[0]), f32(10))
+	testing.expect_value(t, marker_of(list[1]), f32(20))
+	testing.expect_value(t, marker_of(list[2]), f32(30))
+	testing.expect_value(t, marker_of(list[3]), f32(40))
+}
+
+@(test)
+group_texture_runs_noncontiguous_same_group_id :: proc(t: ^testing.T) {
+	// Same nonzero id split by group 0: each window regroups alone.
+	list := []Queued_Sprite {
+		queued(2, 1, 1),
+		queued(1, 1, 2),
+		queued(2, 0, 3),
+		queued(2, 1, 4),
+		queued(1, 1, 5),
+	}
+	group_texture_runs(list)
+	testing.expect_value(t, marker_of(list[2]), f32(3))
+	testing.expect(t, list[0].texture == fake_tex(1))
+	testing.expect(t, list[1].texture == fake_tex(2))
+	testing.expect_value(t, list[2].batch_group, u32(0))
+	testing.expect(t, list[3].texture == fake_tex(1))
+	testing.expect(t, list[4].texture == fake_tex(2))
+	testing.expect_value(t, list[0].batch_group, u32(1))
+	testing.expect_value(t, list[1].batch_group, u32(1))
+	testing.expect_value(t, list[3].batch_group, u32(1))
+	testing.expect_value(t, list[4].batch_group, u32(1))
+}
+
+make_test_draw_app :: proc() -> App {
+	app: App
+	app.cmd = cast(^sdl.GPUCommandBuffer)uintptr(1)
+	app.swapchain_texture = fake_tex(99)
+	app.swapchain_w = 800
+	app.swapchain_h = 600
+	app.camera = camera_default()
+	app.draw_list = make([dynamic]Queued_Sprite)
+	return app
+}
+
+destroy_test_draw_app :: proc(app: ^App) {
+	if app == nil do return
+	delete(app.draw_list)
+	app^ = {}
+}
+
+make_test_draw_character :: proc() -> Character_Data {
+	data: Character_Data
+	data.texture = fake_tex(42)
+	data.width = 100
+	data.height = 100
+	data.def.pivot = {0.5, 1.0}
+	data.def.clips = make(map[string]Clip_Def)
+	frames := make([]Frame_Def, 1)
+	frames[0] = Frame_Def {
+		rect        = {0, 0, 10, 10},
+		source_size = {10, 10},
+		trim_offset = {0, 0},
+	}
+	data.def.clips["idle"] = Clip_Def {
+		loop   = true,
+		fps    = 10,
+		frames = frames,
+	}
+	return data
+}
+
+destroy_test_draw_character :: proc(data: ^Character_Data) {
+	if data == nil do return
+	keys := make([dynamic]string, context.temp_allocator)
+	for key, clip in data.def.clips {
+		delete(clip.frames)
+		append(&keys, key)
+	}
+	for key in keys {
+		delete_key(&data.def.clips, key)
+	}
+	delete(data.def.clips)
+	data^ = {}
+}
+
+@(test)
+draw_sprite_stamps_batch_group_zero :: proc(t: ^testing.T) {
+	app := make_test_draw_app()
+	defer destroy_test_draw_app(&app)
+	data := make_test_draw_character()
+	defer destroy_test_draw_character(&data)
+
+	sprite := spawn_sprite(&data, {100, 200}, "idle", 0)
+	draw_sprite(&app, &sprite)
+
+	testing.expect_value(t, len(app.draw_list), 1)
+	testing.expect_value(t, app.draw_list[0].batch_group, u32(0))
+	testing.expect(t, app.draw_list[0].texture == data.texture)
+}
+
+@(test)
+draw_sprite_batched_stamps_batch_group :: proc(t: ^testing.T) {
+	app := make_test_draw_app()
+	defer destroy_test_draw_app(&app)
+	data := make_test_draw_character()
+	defer destroy_test_draw_character(&data)
+
+	sprite := spawn_sprite(&data, {100, 200}, "idle", 0)
+	draw_sprite_batched(&app, &sprite, 7)
+
+	testing.expect_value(t, len(app.draw_list), 1)
+	testing.expect_value(t, app.draw_list[0].batch_group, u32(7))
+	testing.expect(t, app.draw_list[0].texture == data.texture)
+}
+
+@(test)
+draw_sprite_batched_guards_leave_list_unchanged :: proc(t: ^testing.T) {
+	app := make_test_draw_app()
+	defer destroy_test_draw_app(&app)
+	data := make_test_draw_character()
+	defer destroy_test_draw_character(&data)
+	sprite := spawn_sprite(&data, {100, 200}, "idle", 0)
+
+	app.cmd = nil
+	draw_sprite_batched(&app, &sprite, 1)
+	testing.expect_value(t, len(app.draw_list), 0)
+	app.cmd = cast(^sdl.GPUCommandBuffer)uintptr(1)
+
+	app.swapchain_texture = nil
+	draw_sprite_batched(&app, &sprite, 1)
+	testing.expect_value(t, len(app.draw_list), 0)
+	app.swapchain_texture = fake_tex(99)
+
+	draw_sprite_batched(&app, nil, 1)
+	testing.expect_value(t, len(app.draw_list), 0)
+
+	no_data := sprite
+	no_data.data = nil
+	draw_sprite_batched(&app, &no_data, 1)
+	testing.expect_value(t, len(app.draw_list), 0)
+
+	no_tex := sprite
+	tex_data := data
+	tex_data.texture = nil
+	no_tex.data = &tex_data
+	draw_sprite_batched(&app, &no_tex, 1)
+	testing.expect_value(t, len(app.draw_list), 0)
+}
+
+@(test)
+prepare_draw_batches_regroups_then_plans_runs :: proc(t: ^testing.T) {
+	list := []Queued_Sprite {
+		queued(2, 1, 1),
+		queued(1, 1, 2),
+		queued(2, 1, 3),
+		queued(1, 1, 4),
+		queued(3, 0, 5),
+		queued(1, 0, 6),
+	}
+	batches := make([dynamic]Draw_Batch)
+	defer delete(batches)
+
+	prepare_draw_batches(list, &batches)
+
+	testing.expect_value(t, len(batches), 4)
+	testing.expect_value(t, batches[0].start, 0)
+	testing.expect_value(t, batches[0].count, 2)
+	testing.expect(t, batches[0].texture == fake_tex(1))
+	testing.expect_value(t, batches[1].start, 2)
+	testing.expect_value(t, batches[1].count, 2)
+	testing.expect(t, batches[1].texture == fake_tex(2))
+	testing.expect_value(t, batches[2].start, 4)
+	testing.expect_value(t, batches[2].count, 1)
+	testing.expect(t, batches[2].texture == fake_tex(3))
+	testing.expect_value(t, batches[3].start, 5)
+	testing.expect_value(t, batches[3].count, 1)
+	testing.expect(t, batches[3].texture == fake_tex(1))
+
+	// Group 0 submission order preserved.
+	testing.expect_value(t, marker_of(list[4]), f32(5))
+	testing.expect_value(t, marker_of(list[5]), f32(6))
+	testing.expect_value(t, list[4].batch_group, u32(0))
+	testing.expect_value(t, list[5].batch_group, u32(0))
+}

engine/sprite.odin

@@ -82,7 +82,31 @@ to_clip :: proc(px, py, sw, sh: f32) -> [2]f32 {
 	}
 }
 
+// Axis-aligned quad: two unique x and y values, so scale once and reuse.
+sprite_quad_to_clip :: proc(x0, y0, x1, y1, sw, sh: f32) -> [4]Vec2 {
+	sx := 2.0 / sw
+	sy := 2.0 / sh
+
+	left := x0 * sx - 1
+	right := x1 * sx - 1
+	top := 1 - y0 * sy
+	bottom := 1 - y1 * sy
+
+	return {
+		{left, top},
+		{right, top},
+		{right, bottom},
+		{left, bottom},
+	}
+}
+
 draw_sprite :: proc(app: ^App, sprite: ^Sprite) {
+	draw_sprite_batched(app, sprite, 0)
+}
+
+// Nonzero batch_group lets end_frame regroup consecutive same-group sprites by
+// texture. Group 0 keeps exact submission order for correct alpha overlap.
+draw_sprite_batched :: proc(app: ^App, sprite: ^Sprite, batch_group: u32) {
 	if app.cmd == nil || app.swapchain_texture == nil {
 		return
 	}
@@ -124,10 +148,8 @@ draw_sprite :: proc(app: ^App, sprite: ^Sprite) {
 
 	sw := f32(app.swapchain_w)
 	sh := f32(app.swapchain_h)
-	p0 := to_clip(x0_px, y0_px, sw, sh)
-	p1 := to_clip(x1_px, y0_px, sw, sh)
-	p2 := to_clip(x1_px, y1_px, sw, sh)
-	p3 := to_clip(x0_px, y1_px, sw, sh)
+	points := sprite_quad_to_clip(x0_px, y0_px, x1_px, y1_px, sw, sh)
+	p0, p1, p2, p3 := points[0], points[1], points[2], points[3]
 
 	tex_w := f32(sprite.data.width)
 	tex_h := f32(sprite.data.height)
@@ -142,7 +164,14 @@ draw_sprite :: proc(app: ^App, sprite: ^Sprite) {
 		{pos = p3, uv = {u0, v1}},
 	}
 
-	append(&app.draw_list, Queued_Sprite{texture = sprite.data.texture, verts = verts})
+	append(
+		&app.draw_list,
+		Queued_Sprite {
+			texture = sprite.data.texture,
+			verts = verts,
+			batch_group = batch_group,
+		},
+	)
 }
 
 set_sprite_clip :: proc(sprite: ^Sprite, clip: string) {