4063d10 perf(headless): reuse the shows map between flips · neferwl

M Makefile +1 -1


@@ -69,7 +69,7 @@ adapter-imports:
69	69		[ -z "$$bad" ] \|\| { echo "$$bad"; echo 'adapter imports another adapter: use ports or a shared library (Makefile ADAPTER_LIBS)' >&2; exit 1; }
70	70		# Fast, mandatory guards for tiled content publications and SHM copies.
71	71		perf-check:
72		-	CGO_ENABLED=0 go test ./internal/core ./internal/app ./internal/adapters/drm ./internal/adapters/vulkan ./internal/adapters/wayland -run '^(TestAnimateAllocations\|TestPublishAllocations\|TestPublishSubscribersAllocations\|TestPublishStashSlideAllocations\|TestPublishOverviewAllocations\|TestPublishLeavingAllocations\|TestOutputRoutingAllocations\|TestReportSeenAllocations\|TestFrameLifecycleTransitionsAllocations\|TestFullscreenShownAllocations\|TestFrameDecisionAllocations\|TestCommitFrameAllocations\|TestColorPipelineAllocations\|TestShownBySnapshotAllocations\|TestMRUNoteFocusDoesNotAllocateWhenSettled\|TestDueFramesAllocations\|TestRenderSteadyStateAllocations\|TestSceneOpaqueRegionAllocations\|TestSceneDeltaAllocations\|TestAccountFlipAllocations\|TestFlipDoneAllocations\|TestSceneWalkUnchangedTiledSHM\|TestTiledCommitPublishAllocations\|TestCapturedCommitApplyAllocations\|TestCapturedViewportApplyAllocations\|TestEffectiveInputEmptyTreeAllocations\|TestHeadlessTiledSHMCallbackAndCopyBudget\|TestReleaseHeldAllocations\|TestCommitDamageAllocations\|TestEmitContentAllocations\|TestPresentationPushAllocations\|TestAnimatingFrameSendAllocations\|TestContentBatchCycleAllocations)$$' -count=1
	72	+	CGO_ENABLED=0 go test ./internal/core ./internal/app ./internal/adapters/drm ./internal/adapters/headless ./internal/adapters/vulkan ./internal/adapters/wayland -run '^(TestAnimateAllocations\|TestPublishAllocations\|TestPublishSubscribersAllocations\|TestPublishStashSlideAllocations\|TestPublishOverviewAllocations\|TestPublishLeavingAllocations\|TestOutputRoutingAllocations\|TestReportSeenAllocations\|TestFrameLifecycleTransitionsAllocations\|TestFullscreenShownAllocations\|TestFrameDecisionAllocations\|TestCommitFrameAllocations\|TestColorPipelineAllocations\|TestShownBySnapshotAllocations\|TestFlipShowsAllocations\|TestMRUNoteFocusDoesNotAllocateWhenSettled\|TestDueFramesAllocations\|TestRenderSteadyStateAllocations\|TestSceneOpaqueRegionAllocations\|TestSceneDeltaAllocations\|TestAccountFlipAllocations\|TestFlipDoneAllocations\|TestSceneWalkUnchangedTiledSHM\|TestTiledCommitPublishAllocations\|TestCapturedCommitApplyAllocations\|TestCapturedViewportApplyAllocations\|TestEffectiveInputEmptyTreeAllocations\|TestHeadlessTiledSHMCallbackAndCopyBudget\|TestReleaseHeldAllocations\|TestCommitDamageAllocations\|TestEmitContentAllocations\|TestPresentationPushAllocations\|TestAnimatingFrameSendAllocations\|TestContentBatchCycleAllocations)$$' -count=1
73	73		check: vet test arch fakes-check log-check perf-check
74	74
75	75		# On-demand SDR/HDR colour check with examples/testpattern on a headless


@@ -69,7 +69,7 @@ adapter-imports:
69		[ -z "$$bad" ] \|\| { echo "$$bad"; echo 'adapter imports another adapter: use ports or a shared library (Makefile ADAPTER_LIBS)' >&2; exit 1; }	69		[ -z "$$bad" ] \|\| { echo "$$bad"; echo 'adapter imports another adapter: use ports or a shared library (Makefile ADAPTER_LIBS)' >&2; exit 1; }
70		# Fast, mandatory guards for tiled content publications and SHM copies.	70		# Fast, mandatory guards for tiled content publications and SHM copies.
71		perf-check:	71		perf-check:
72	-	CGO_ENABLED=0 go test ./internal/core ./internal/app ./internal/adapters/drm ./internal/adapters/vulkan ./internal/adapters/wayland -run '^(TestAnimateAllocations\|TestPublishAllocations\|TestPublishSubscribersAllocations\|TestPublishStashSlideAllocations\|TestPublishOverviewAllocations\|TestPublishLeavingAllocations\|TestOutputRoutingAllocations\|TestReportSeenAllocations\|TestFrameLifecycleTransitionsAllocations\|TestFullscreenShownAllocations\|TestFrameDecisionAllocations\|TestCommitFrameAllocations\|TestColorPipelineAllocations\|TestShownBySnapshotAllocations\|TestMRUNoteFocusDoesNotAllocateWhenSettled\|TestDueFramesAllocations\|TestRenderSteadyStateAllocations\|TestSceneOpaqueRegionAllocations\|TestSceneDeltaAllocations\|TestAccountFlipAllocations\|TestFlipDoneAllocations\|TestSceneWalkUnchangedTiledSHM\|TestTiledCommitPublishAllocations\|TestCapturedCommitApplyAllocations\|TestCapturedViewportApplyAllocations\|TestEffectiveInputEmptyTreeAllocations\|TestHeadlessTiledSHMCallbackAndCopyBudget\|TestReleaseHeldAllocations\|TestCommitDamageAllocations\|TestEmitContentAllocations\|TestPresentationPushAllocations\|TestAnimatingFrameSendAllocations\|TestContentBatchCycleAllocations)$$' -count=1	72	+	CGO_ENABLED=0 go test ./internal/core ./internal/app ./internal/adapters/drm ./internal/adapters/headless ./internal/adapters/vulkan ./internal/adapters/wayland -run '^(TestAnimateAllocations\|TestPublishAllocations\|TestPublishSubscribersAllocations\|TestPublishStashSlideAllocations\|TestPublishOverviewAllocations\|TestPublishLeavingAllocations\|TestOutputRoutingAllocations\|TestReportSeenAllocations\|TestFrameLifecycleTransitionsAllocations\|TestFullscreenShownAllocations\|TestFrameDecisionAllocations\|TestCommitFrameAllocations\|TestColorPipelineAllocations\|TestShownBySnapshotAllocations\|TestFlipShowsAllocations\|TestMRUNoteFocusDoesNotAllocateWhenSettled\|TestDueFramesAllocations\|TestRenderSteadyStateAllocations\|TestSceneOpaqueRegionAllocations\|TestSceneDeltaAllocations\|TestAccountFlipAllocations\|TestFlipDoneAllocations\|TestSceneWalkUnchangedTiledSHM\|TestTiledCommitPublishAllocations\|TestCapturedCommitApplyAllocations\|TestCapturedViewportApplyAllocations\|TestEffectiveInputEmptyTreeAllocations\|TestHeadlessTiledSHMCallbackAndCopyBudget\|TestReleaseHeldAllocations\|TestCommitDamageAllocations\|TestEmitContentAllocations\|TestPresentationPushAllocations\|TestAnimatingFrameSendAllocations\|TestContentBatchCycleAllocations)$$' -count=1
73		check: vet test arch fakes-check log-check perf-check	73		check: vet test arch fakes-check log-check perf-check
74			74
75		# On-demand SDR/HDR colour check with examples/testpattern on a headless	75		# On-demand SDR/HDR colour check with examples/testpattern on a headless

A internal/adapters/headless/flip_alloc_test.go +28 -0

@@ -0,0 +1,28 @@
	1	+	package headless
	2	+
	3	+	import (
	4	+	"testing"
	5	+
	6	+	"github.com/bnema/neferwl/internal/ports"
	7	+	"github.com/stretchr/testify/require"
	8	+	)
	9	+
	10	+	// Steady-state flip reporting: the Shows snapshot is reused while it does not
	11	+	// change, and a changed one never mutates what an earlier flip handed out.
	12	+	func TestFlipShowsAllocations(t *testing.T) {
	13	+	var f flipShows
	14	+	scene := ports.Scene{OutputWidth: 10, OutputHeight: 10, Windows: []ports.SceneWindow{{ID: 1, Rect: ports.Rect{W: 10, H: 10}}}}
	15	+	contents := map[ports.WindowID]ports.SurfaceContent{1: {ID: 1, Seq: 1}, 2: {ID: 2, Seq: 2}}
	16	+	first := f.flipInfo(scene, contents, nil)
	17	+	require.Equal(t, map[ports.WindowID]uint64{1: 1}, first.Shows)
	18	+
	19	+	require.Zero(t, testing.AllocsPerRun(100, func() { f.of(scene, contents, nil) }), "unchanged shows")
	20	+	// The FlipInfo is the one allocation left: the sink's merge writes into it.
	21	+	require.LessOrEqual(t, testing.AllocsPerRun(100, func() { f.flipInfo(scene, contents, nil) }), 1.0, "flip report")
	22	+
	23	+	contents[1] = ports.SurfaceContent{ID: 1, Seq: 3}
	24	+	second := f.flipInfo(scene, contents, nil)
	25	+	require.Equal(t, map[ports.WindowID]uint64{1: 1}, first.Shows, "an earlier snapshot is immutable")
	26	+	require.Equal(t, map[ports.WindowID]uint64{1: 3}, second.Shows)
	27	+	require.NotSame(t, first, second)
	28	+	}

@@ -0,0 +1,28 @@
	1	+	package headless
	2	+
	3	+	import (
	4	+	"testing"
	5	+
	6	+	"github.com/bnema/neferwl/internal/ports"
	7	+	"github.com/stretchr/testify/require"
	8	+	)
	9	+
	10	+	// Steady-state flip reporting: the Shows snapshot is reused while it does not
	11	+	// change, and a changed one never mutates what an earlier flip handed out.
	12	+	func TestFlipShowsAllocations(t *testing.T) {
	13	+	var f flipShows
	14	+	scene := ports.Scene{OutputWidth: 10, OutputHeight: 10, Windows: []ports.SceneWindow{{ID: 1, Rect: ports.Rect{W: 10, H: 10}}}}
	15	+	contents := map[ports.WindowID]ports.SurfaceContent{1: {ID: 1, Seq: 1}, 2: {ID: 2, Seq: 2}}
	16	+	first := f.flipInfo(scene, contents, nil)
	17	+	require.Equal(t, map[ports.WindowID]uint64{1: 1}, first.Shows)
	18	+
	19	+	require.Zero(t, testing.AllocsPerRun(100, func() { f.of(scene, contents, nil) }), "unchanged shows")
	20	+	// The FlipInfo is the one allocation left: the sink's merge writes into it.
	21	+	require.LessOrEqual(t, testing.AllocsPerRun(100, func() { f.flipInfo(scene, contents, nil) }), 1.0, "flip report")
	22	+
	23	+	contents[1] = ports.SurfaceContent{ID: 1, Seq: 3}
	24	+	second := f.flipInfo(scene, contents, nil)
	25	+	require.Equal(t, map[ports.WindowID]uint64{1: 1}, first.Shows, "an earlier snapshot is immutable")
	26	+	require.Equal(t, map[ports.WindowID]uint64{1: 3}, second.Shows)
	27	+	require.NotSame(t, first, second)
	28	+	}

M internal/adapters/headless/leaving_test.go +3 -2


@@ -10,15 +10,16 @@ import (
10	10		// A window drawn from a content its client withdrew (closed, fading out) is
11	11		// not reported shown: the empty content has no frame to present.
12	12		func TestFlipInfoSkipsKeptContent(t *testing.T) {
	13	+	var shows flipShows
13	14		table := surfaces.New()
14	15		s := ports.Scene{OutputWidth: 2, OutputHeight: 2, Windows: []ports.SceneWindow{{ID: 1, Rect: ports.Rect{W: 2, H: 2}}, {ID: 2, Rect: ports.Rect{W: 1, H: 1}}}}
15	16		table.Update(ports.SurfaceContent{ID: 1, Seq: 3, SHM: &ports.SHMBuffer{Pool: 1}}, s)
16	17		table.Update(ports.SurfaceContent{ID: 2, Seq: 1, SHM: &ports.SHMBuffer{Pool: 2}}, s)
17		-	if f := flipInfo(s, table.Map(), table.Kept); f.Shows[1] != 3 \|\| f.Shows[2] != 1 {
	18	+	if f := shows.flipInfo(s, table.Map(), table.Kept); f.Shows[1] != 3 \|\| f.Shows[2] != 1 {
18	19		t.Fatalf("control: %v", f.Shows)
19	20		}
20	21		table.Update(ports.SurfaceContent{ID: 1, Seq: 4}, s)
21		-	f := flipInfo(s, table.Map(), table.Kept)
	22	+	f := shows.flipInfo(s, table.Map(), table.Kept)
22	23		if _, ok := f.Shows[1]; ok \|\| f.Shows[2] != 1 {
23	24		t.Fatalf("a leaving window is reported shown: %v", f.Shows)
24	25		}

 // A window drawn from a content its client withdrew (closed, fading out) is
 // not reported shown: the empty content has no frame to present.
 func TestFlipInfoSkipsKeptContent(t *testing.T) {
+	var shows flipShows
 	table := surfaces.New()
 	s := ports.Scene{OutputWidth: 2, OutputHeight: 2, Windows: []ports.SceneWindow{{ID: 1, Rect: ports.Rect{W: 2, H: 2}}, {ID: 2, Rect: ports.Rect{W: 1, H: 1}}}}
 	table.Update(ports.SurfaceContent{ID: 1, Seq: 3, SHM: &ports.SHMBuffer{Pool: 1}}, s)
 	table.Update(ports.SurfaceContent{ID: 2, Seq: 1, SHM: &ports.SHMBuffer{Pool: 2}}, s)
-	if f := flipInfo(s, table.Map(), table.Kept); f.Shows[1] != 3 || f.Shows[2] != 1 {
+	if f := shows.flipInfo(s, table.Map(), table.Kept); f.Shows[1] != 3 || f.Shows[2] != 1 {
 		t.Fatalf("control: %v", f.Shows)
 	}
 	table.Update(ports.SurfaceContent{ID: 1, Seq: 4}, s)
-	f := flipInfo(s, table.Map(), table.Kept)
+	f := shows.flipInfo(s, table.Map(), table.Kept)
 	if _, ok := f.Shows[1]; ok || f.Shows[2] != 1 {
 		t.Fatalf("a leaving window is reported shown: %v", f.Shows)
 	}

M internal/adapters/headless/output.go +33 -10


@@ -5,6 +5,7 @@ import (
5	5		"fmt"
6	6		"image"
7	7		"image/png"
	8	+	"maps"
8	9		"os"
9	10		"path/filepath"
10	11		"time"

@@ -117,6 +118,7 @@ func Run(ctx context.Context, opts Options, scenes <-chan ports.Scene, contents
117	118		// reports queues what the output read; every report is pushed to the
118	119		// sink at once (Push never blocks), so it holds one at most.
119	120		var reports presented.Queue
	121	+	var shows flipShows
120	122		var security ports.SecurityState
121	123		securityKnown := false
122	124		failRequests := func(reason error) {

@@ -541,7 +543,7 @@ func Run(ctx context.Context, opts Options, scenes <-chan ports.Scene, contents
541	543		pipeline.EndGate(nil)
542	544		requests.List = capture.Waiting(requests.List, scene)
543	545		frame++
544		-	opts.report(&reports, flipInfo(scene, drawn, table.Kept), seen)
	546	+	opts.report(&reports, shows.flipInfo(scene, drawn, table.Kept), seen)
545	547		if scene.Animating && opts.Frames != nil {
546	548		select {
547	549		case opts.Frames <- ports.OutputFrame{Output: opts.Name}:

@@ -627,19 +629,40 @@ func (opts Options) report(q presented.Queue, flip ports.FlipInfo, seen map[po
627	629		q.Flush(opts.Presented)
628	630		}
629	631
630		-	// flipInfo describes a drawn frame as a flip at the current CLOCK_MONOTONIC
631		-	// time (software clock, refresh unknown), so presentation feedback and
632		-	// frame callbacks follow headless frames.
633		-	func flipInfo(scene ports.Scene, contents map[ports.WindowID]ports.SurfaceContent, kept func(ports.WindowID) bool) *ports.FlipInfo {
634		-	var ts unix.Timespec
635		-	_ = unix.ClockGettime(unix.CLOCK_MONOTONIC, &ts)
636		-	shows := map[ports.WindowID]uint64{}
	632	+	// flipShows builds the Shows of successive flips. Its owner goroutine is the
	633	+	// only user. A flip's Shows map is an immutable snapshot: wayland may still
	634	+	// read the one of an earlier flip on another goroutine, so the scratch map is
	635	+	// never sent and a snapshot is replaced by a clone only when it changed.
	636	+	type flipShows struct {
	637	+	scratch, snapshot map[ports.WindowID]uint64
	638	+	}
	639	+
	640	+	// of returns the content Seq of the windows scene showed to their client. The
	641	+	// result is shared with earlier flips of the same content: never modify it.
	642	+	func (f *flipShows) of(scene ports.Scene, contents map[ports.WindowID]ports.SurfaceContent, kept func(ports.WindowID) bool) map[ports.WindowID]uint64 {
	643	+	if f.scratch == nil {
	644	+	f.scratch = make(map[ports.WindowID]uint64)
	645	+	}
	646	+	clear(f.scratch)
637	647		for id, c := range contents {
638	648		// A closed window drawn from its kept content is not shown to
639	649		// its client.
640	650		if surfaces.Shown(scene, id, kept) {
641		-	shows[id] = c.Seq
	651	+	f.scratch[id] = c.Seq
642	652		}
643	653		}
644		-	return &ports.FlipInfo{When: time.Duration(ts.Nano()), Shows: shows}
	654	+	if f.snapshot == nil \|\| !maps.Equal(f.snapshot, f.scratch) {
	655	+	f.snapshot = maps.Clone(f.scratch)
	656	+	}
	657	+	return f.snapshot
	658	+	}
	659	+
	660	+	// flipInfo describes a drawn frame as a flip at the current CLOCK_MONOTONIC
	661	+	// time (software clock, refresh unknown), so presentation feedback and
	662	+	// frame callbacks follow headless frames. The FlipInfo itself is new per
	663	+	// flip (the sink counts merges on it); only its Shows is shared.
	664	+	func (f flipShows) flipInfo(scene ports.Scene, contents map[ports.WindowID]ports.SurfaceContent, kept func(ports.WindowID) bool) ports.FlipInfo {
	665	+	var ts unix.Timespec
	666	+	_ = unix.ClockGettime(unix.CLOCK_MONOTONIC, &ts)
	667	+	return &ports.FlipInfo{When: time.Duration(ts.Nano()), Shows: f.of(scene, contents, kept)}
645	668		}

 	"fmt"
 	"image"
 	"image/png"
+	"maps"
 	"os"
 	"path/filepath"
 	"time"
 	// reports queues what the output read; every report is pushed to the
 	// sink at once (Push never blocks), so it holds one at most.
 	var reports presented.Queue
+	var shows flipShows
 	var security ports.SecurityState
 	securityKnown := false
 	failRequests := func(reason error) {
 		pipeline.EndGate(nil)
 		requests.List = capture.Waiting(requests.List, scene)
 		frame++
-		opts.report(&reports, flipInfo(scene, drawn, table.Kept), seen)
+		opts.report(&reports, shows.flipInfo(scene, drawn, table.Kept), seen)
 		if scene.Animating && opts.Frames != nil {
 			select {
 			case opts.Frames <- ports.OutputFrame{Output: opts.Name}:
 	q.Flush(opts.Presented)
 }
-// flipInfo describes a drawn frame as a flip at the current CLOCK_MONOTONIC
+// flipShows builds the Shows of successive flips. Its owner goroutine is the
-// time (software clock, refresh unknown), so presentation feedback and
+// only user. A flip's Shows map is an immutable snapshot: wayland may still
-// frame callbacks follow headless frames.
+// read the one of an earlier flip on another goroutine, so the scratch map is
-func flipInfo(scene ports.Scene, contents map[ports.WindowID]ports.SurfaceContent, kept func(ports.WindowID) bool) *ports.FlipInfo {
+// never sent and a snapshot is replaced by a clone only when it changed.
-	var ts unix.Timespec
+type flipShows struct {
-	_ = unix.ClockGettime(unix.CLOCK_MONOTONIC, &ts)
+	scratch, snapshot map[ports.WindowID]uint64
-	shows := map[ports.WindowID]uint64{}
+}
+// of returns the content Seq of the windows scene showed to their client. The
+// result is shared with earlier flips of the same content: never modify it.
+func (f *flipShows) of(scene ports.Scene, contents map[ports.WindowID]ports.SurfaceContent, kept func(ports.WindowID) bool) map[ports.WindowID]uint64 {
+	if f.scratch == nil {
+		f.scratch = make(map[ports.WindowID]uint64)
+	}
+	clear(f.scratch)
 	for id, c := range contents {
 		// A closed window drawn from its kept content is not shown to
 		// its client.
 		if surfaces.Shown(scene, id, kept) {
-			shows[id] = c.Seq
+			f.scratch[id] = c.Seq
 		}
 	}
-	return &ports.FlipInfo{When: time.Duration(ts.Nano()), Shows: shows}
+	if f.snapshot == nil || !maps.Equal(f.snapshot, f.scratch) {
+		f.snapshot = maps.Clone(f.scratch)
+	}
+	return f.snapshot
+}
+// flipInfo describes a drawn frame as a flip at the current CLOCK_MONOTONIC
+// time (software clock, refresh unknown), so presentation feedback and
+// frame callbacks follow headless frames. The FlipInfo itself is new per
+// flip (the sink counts merges on it); only its Shows is shared.
+func (f *flipShows) flipInfo(scene ports.Scene, contents map[ports.WindowID]ports.SurfaceContent, kept func(ports.WindowID) bool) *ports.FlipInfo {
+	var ts unix.Timespec
+	_ = unix.ClockGettime(unix.CLOCK_MONOTONIC, &ts)
+	return &ports.FlipInfo{When: time.Duration(ts.Nano()), Shows: f.of(scene, contents, kept)}
 }